Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

We thank the reviewer for his valuable input and careful assessment, which have significantly improved the clarity and rigor of our manuscript.

Summary:

Mazer & Yovel 2025 dissect the inverse problem of how echolocators in groups manage to navigate their surroundings despite intense jamming using computational simulations.

The authors show that despite the 'noisy' sensory environments that echolocating groups present, agents can still access some amount of echo-related information and use it to navigate their local environment. It is known that echolocating bats have strong small and large-scale spatial memory that plays an important role for individuals. The results from this paper also point to the potential importance of an even lower-level, short-term role of memory in the form of echo 'integration' across multiple calls, despite the unpredictability of echo detection in groups. The paper generates a useful basis to think about the mechanisms in echolocating groups for experimental investigations too.

Strengths:

(1) The paper builds on biologically well-motivated and parametrised 2D acoustics and sensory simulation setup to investigate the various key parameters of interest

(2) The 'null-model' of echolocators not being able to tell apart objects & conspecifics while echolocating still shows agents successfully emerge from groups - even though the probability of emergence drops severely in comparison to cognitively more 'capable' agents. This is nonetheless an important result showing the directionof-arrival of a sound itself is the 'minimum' set of ingredients needed for echolocators navigating their environment.

(3) The results generate an important basis in unraveling how agents may navigate in sensorially noisy environments with a lot of irrelevant and very few relevant cues.

(4) The 2D simulation framework is simple and computationally tractable enough to perform multiple runs to investigate many variables - while also remaining true to the aim of the investigation.

Weaknesses:

There are a few places in the paper that can be misunderstood or don't provide complete details. Here is a selection:

(1) Line 61: '... studies have focused on movement algorithms while overlooking the sensory challenges involved' : This statement does not match the recent state of the literature. While the previous models may have had the assumption that all neighbours can be detected, there are models that specifically study the role of limited interaction arising from a potential inability to track all neighbours due to occlusion, and the effect of responding to only one/few neighbours at a time e.g. Bode et al. 2011 R. Soc. Interface, Rosenthal et al. 2015 PNAS, Jhawar et al. 2020 Nature Physics.

We appreciate the reviewer's comment and the relevant references. We have revised the manuscript accordingly to clarify the distinction between studies that incorporate limited interactions and those that explicitly analyze sensory constraints and interference. We have refined our statement to acknowledge these contributions while maintaining our focus on sensory challenges beyond limited neighbor detection, such as signal degradation, occlusion effects, and multimodal sensory integration (see lines 58-64):

(2) The word 'interference' is used loosely places (Line 89: '...took all interference signals...', Line 319: 'spatial interference') - this is confusing as it is not clear whether the authors refer to interference in the physics/acoustics sense, or broadly speaking as a synonym for reflections and/or jamming.

To improve clarity, we have revised the manuscript to distinguish between different types of interference:

• Acoustic interference (jamming): Overlapping calls that completely obscure echo detection, preventing bats from perceiving necessary environmental cues.

• Acoustic interference (masking): Partial reduction in signal clarity due to competing calls.

• Spatial interference: Physical obstruction by conspecifics affecting movement and navigation.

We have updated the manuscript to use these terms consistently and explicitly define them in relevant sections (see lines 84-85, 119-120). This distinction ensures that the reader can differentiate between interference as an acoustic phenomenon and its broader implications in navigation.

(3) The paper discusses original results without reference to how they were obtained or what was done. The lack of detail here must be considered while interpreting the Discussion e.g. Line 302 ('our model suggests...increasing the call-rate..' - no clear mention of how/where call-rate was varied) & Line 323 '..no benefit beyond a certain level..' - also no clear mention of how/where call-level was manipulated in the simulations.

All tested parameters, including call rate dynamics and call intensity variations, are detailed in the Methods section and Tables 1 and 2. Specifically:

• Call Rate Variation: The Inter-Pulse Interval (IPI) was modeled based on documented echolocation behavior, decreasing from 100 msec during the search phase to 35 msec (~28 calls per second) at the end of the approach phase, and to 5 msec (200 calls per second) during the final buzz (see Table 2). This natural variation in call rate was not manually manipulated in the model but emerged from the simulated bat behavior.

• Call Intensity Variation: The tested call intensity levels (100, 110, 120, 130 dB SPL) are presented in Table 1 under the “Call Level” parameter. The effect of increasing call intensity was analyzed in relation to exit probability, jamming probability, and collision rate. This is now explicitly referenced in the Discussion. We have revised the manuscript to explicitly reference these aspects in the Results and Discussion sections – see lines 346-349, 372-375.

Reviewer #2 (Public review):

We are grateful for the reviewer’s insightful feedback, which has helped us clarify key aspects of our research and strengthen our conclusions.

This manuscript describes a detailed model of bats flying together through a fixed geometry. The model considers elements that are faithful to both bat biosonar production and reception and the acoustics governing how sound moves in the air and interacts with obstacles. The model also incorporates behavioral patterns observed in bats, like one-dimensional feature following and temporal integration of cognitive maps. From a simulation study of the model and comparison of the results with the literature, the authors gain insight into how often bats may experience destructive interference of their acoustic signals and those of their peers, and how much such interference may actually negatively affect the groups' ability to navigate effectively. The authors use generalized linear models to test the significance of the effects they observe.

In terms of its strengths, the work relies on a thoughtful and detailed model that faithfully incorporates salient features, such as acoustic elements like the filter for a biological receiver and temporal aggregation as a kind of memory in the system. At the same time, the authors' abstract features are complicating without being expected to give additional insights, as can be seen in the choice of a twodimensional rather than three-dimensional system. I thought that the level of abstraction in the model was perfect, enough to demonstrate their results without needless details. The results are compelling and interesting, and the authors do a great job discussing them in the context of the biological literature. 

The most notable weakness I found in this work was that some aspects of the model were not entirely clear to me. 

For example, the directionality of the bat's sonar call in relation to its velocity. Are these the same?

For simplicity, in our model, the head is aligned with the body, therefore the direction of the echolocation beam is the same as the direction of the flight. 

Moreover, call directionality (directivity) is not directly influenced by velocity. Instead, directionality is estimated using the piston model, as described in the Methods section. The directionality is based on the emission frequency and is thus primarily linked to the behavioral phases of the bat, with frequency shifts occurring as the bat transitions from search to approach to buzz phases. During the approach phase, the bat emits calls with higher frequencies, resulting in increased directionality. This is supported by the literature (Jakobsen and Surlykke, 2010; Jakobsen, Brinkløv and Surlykke, 2013). This phase is also associated with a natural reduction in flight speed, which is a well-documented behavioral adaptation in echolocating bats(Jakobsen et al., 2024).

To clarify this in the manuscript, we have updated the text to explicitly state that directionality follows phase-dependent frequency changes rather than being a direct function of velocity, see lines 543-545. 

If so, what is the difference between phi_target and phi_tx in the model equations? 

𝝓_{𝒕𝒂𝒓𝒈𝒆𝒕} represents the angle between the bat and the reflected object (target).

𝝓_𝑻𝒙 the angle [rad], between the masking bat and target (from the transmitter’s perspective)

𝝓_{𝑻𝒙𝑹𝒙} refers to the angle between the transmitting conspecific and the receiving focal bat, from the transmitter’s point of view.

𝝓_{𝑹𝒙𝑻𝒙} represents the angle between the receiving bat and the transmitting bat, from the receiver’s point of view.

These definitions have been explicitly stated in the revised manuscript to prevent any ambiguity (lines 525-530). Additionally, a Supplementary figure demonstrating the geometrical relations has been added to the manuscript.

What is a bat's response to colliding with a conspecific (rather than a wall)? 

In nature, minor collisions between bats are common and typically do not result in significant disruptions to flight (Boerma et al., 2019; Roy et al., 2019; Goldshtein et al., 2025). Given this, our model does not explicitly simulate the physical impact of a collision event. Instead, during the collision event the bat keeps decreasing its velocity and changing its flight direction until the distance between bats is above the threshold (0.4 m). We assume that the primary cost of such interactions arises from the effort required to avoid collisions, rather than from the collision itself. This assumption aligns with observations of bat behavior in dense flight environments, where individuals prioritize collision avoidance rather than modeling post-collision dynamics. See lines 479-484.

From the statistical side, it was not clear if replicate simulations were performed. If they were, which I believe is the right way due to stochasticity in the model, how many replicates were used, and are the standard errors referred to throughout the paper between individuals in the same simulation or between independent simulations, or both? 

The number of repetitions for each scenario is detailed in Table 1, but we included it in a more prominent location in the text for clarity. Specifically, we now state (Lines 110-111):

"The number of repetitions for each scenario was as follows: 1 bat: 240; 2 bats: 120; 5 bats: 48; 10 bats: 24; 20 bats: 12; 40 bats: 12; 100 bats: 6."

Regarding the reported standard errors, they are calculated across all individuals within each scenario, without distinguishing between different simulation trials. 

We clarified in the revised text (Lines 627-628 in Statistical Analysis) 

Overall, I found these weaknesses to be superficial and easily remedied by the authors. The authors presented well-reasoned arguments that were supported by their results, and which were used to demonstrate how call interference impacts the collective's roost exit as measured by several variables. As the authors highlight, I think this work is valuable to individuals interested in bat biology and behavior, as well as to applications in engineered multi-agent systems like robotic swarms.

Reviewer #3 (Public review):

We sincerely appreciate the reviewer’s thoughtful comments and the time invested in evaluating our work, which have greatly contributed to refining our study.

We would like to note that in general, our model often simplifies some of the bats’ abilities, under the assumption that if the simulated bats manage to perform this difficult task with simpler mechanisms, real better adapted bats will probably perform even better. This thought strategy will be repeated in several of the s below.

Summary:

The authors describe a model to mimic bat echolocation behavior and flight under high-density conditions and conclude that the problem of acoustic jamming is less severe than previously thought, conflating the success of their simulations (as described in the manuscript) with hard evidence for what real bats are actually doing. The authors base their model on two species of bats that fly at "high densities" (defined by the authors as colony sizes from tens to tens of thousands of individuals and densities of up to 33.3 bats/m2), Pipistrellus kuhli and Rhinopoma microphyllum. This work fits into the broader discussion of bat sensorimotor strategies during collective flight, and simulations are important to try to understand bat behavior, especially given a lack of empirical data. However, I have major concerns about the assumptions of the parameters used for the simulation, which significantly impact both the results of the simulation and the conclusions that can be made from the data. These details are elaborated upon below, along with key recommendations the authors should consider to guide the refinement of the model.

Strengths:

This paper carries out a simulation of bat behavior in dense swarms as a way to explain how jamming does not pose a problem in dense groups. Simulations are important when we lack empirical data. The simulation aims to model two different species with different echolocation signals, which is very important when trying to model echolocation behavior. The analyses are fairly systematic in testing all ranges of parameters used and discussing the differential results.

Weaknesses:

The justification for how the different foraging phase call types were chosen for different object detection distances in the simulation is unclear. Do these distances match those recorded from empirical studies, and if so, are they identical for both species used in the simulation? 

The distances at which bats transition between echolocation phases are identical for both species in our model (see Table 2). These distances are based on welldocumented empirical studies of bat hunting and obstacle avoidance behavior (Griffin, Webster and Michael, 1958; Simmons and Kick, 1983; Schnitzler et al., 1987; Kalko, 1995; Hiryu et al., 2008; Vanderelst and Peremans, 2018). These references provide extensive evidence that insectivorous bats systematically adjust their echolocation calls in response to object proximity, following the characteristic phases of search, approach, and buzz.

To improve clarity, we have updated the text to explicitly state that the phase transition distances are empirically grounded and apply equally to both modeled species (lines 499-508).

What reasoning do the authors have for a bat using the same call characteristics to detect a cave wall as they would for detecting a small insect? 

In echolocating bats, call parameters are primarily shaped by the target distance and echo strength. Accordingly, there is little difference in call structure between prey capture and obstacles-related maneuvers, aside from intensity adjustments based on target strength (Hagino et al., 2007; Hiryu et al., 2008; Surlykke, Ghose and Moss, 2009; Kothari et al., 2014). In our study, due to the dense cave environment, the bats are found to operate in the approach phase most of the time, which is consistent with natural cave emergence, where they are navigating through a cluttered environment rather than engaging in open-space search. For one of the species (Rhinopoma), we also have empirical recordings of individuals flying under similar conditions (Goldshtein et al., 2025). Our model was designed to remain as simple as possible while relying on conservative assumptions that may underestimate bat performance. If, in reality, bats fine-tune their echolocation calls even earlier or more precisely during navigation than assumed, our model would still conservatively reflect their actual capabilities. See lines 500-508.

The two species modeled have different calls. In particular, the bandwidth varies by a factor of 10, meaning the species' sonars will have different spatial resolutions. Range resolution is about 10x better for PK compared to RM, but the authors appear to use the same thresholds for "correct detection" for both, which doesn't seem appropriate.

The detection process in our model is based on Saillant’s method using a filterbank, as detailed in the paper (Saillant et al., 1993; Neretti et al., 2003; Sanderson et al., 2003). This approach inherently incorporates the advantages of a wider bandwidth, meaning that the differences in range resolution between the species are already accounted for within the signal-processing framework. Thus, there is no need to explicitly adjust the model parameters for bandwidth variations, as these effects emerge from the applied method.

Also, the authors did not mention incorporating/correcting for/exploiting Doppler, which leads me to assume they did not model it.

The reviewer is correct. To maintain model simplicity, we did not incorporate the Doppler effect or its impact on echolocation. The exclusion of Doppler effects was based on the assumption that while Doppler shifts can influence frequency perception, their impact on jamming and overall navigation performance is minor within the modelled context.

The maximal Doppler shifts expected for the bats in this scenario are of ~ 1kHz. These shifts would be applied variably across signals due to the semi-random relative velocities between bats, leading to a mixed effect on frequency changes. This variability would likely result in an overall reduction in jamming rather than exacerbating it, aligning with our previous statement that our model may overestimate the severity of acoustic interference. Such Doppler shifts would result in errors of 2-4 cm in localization (i.e., 200-400 micro-seconds) (Boonman, Parsons and Jones, 2003).

We have now explicitly highlighted this in the revised version (see 548-581).

The success of the simulation may very well be due to variation in the calls of the bats, which ironically enough demonstrates the importance of a jamming avoidance response in dense flight. This explains why the performance of the simulation falls when bats are not able to distinguish their own echoes from other signals. For example, in Figure C2, there are calls that are labeled as conspecific calls and have markedly shorter durations and wider bandwidths than others. These three phases for call types used by the authors may be responsible for some (or most) of the performance of the model since the correlation between different call types is unlikely to exceed the detection threshold. But it turns out this variation in and of itself is what a jamming avoidance response may consist of. So, in essence, the authors are incorporating a jamming avoidance response into their simulation. 

We fully agree that the natural variations in call design between the phases contribute significantly to interference reduction (see our discussion in a previous paper in Mazar & Yovel, 2020). However, we emphasize that this cannot be classified as a Jamming Avoidance Response (JAR). In our model, bats respond only to the physical presence of objects and not to the acoustic environment or interference itself. There is no active or adaptive adjustment of call design to minimize jamming beyond the natural phase-dependent variations in call structure. Therefore, while variation in call types does inherently reduce interference, this effect emerges passively from the modeled behavior rather than as an intentional strategy to avoid jamming. 

The authors claim that integration over multiple pings (though I was not able to determine the specifics of this integration algorithm) reduces the masking problem. Indeed, it should: if you have two chances at detection, you've effectively increased your SNR by 3dB.  

The reviewer is correct. Indeed, integration over multiple calls improves signal-tonoise ratio (SNR), effectively increasing it by approximately 3 dB per doubling of observations. The specifics of the integration algorithm are detailed in the Methods section, where we describe how sensory information is aggregated across multiple time steps to enhance detection reliability.

They also claim - although it is almost an afterthought - that integration dramatically reduces the degradation caused by false echoes. This also makes sense: from one ping to the next, the bat's own echo delays will correlate extremely well with the bat's flight path. Echo delays due to conspecifics will jump around kind of randomly. However, the main concern is regarding the time interval and number of pings of the integration, especially in the context of the bat's flight speed. The authors say that a 1s integration interval (5-10 pings) dramatically reduces jamming probability and echo confusion. This number of pings isn't very high, and it occurs over a time interval during which the bat has moved 5-10m. This distance is large compared to the 0.4m distance-to-obstacle that triggers an evasive maneuver from the bat, so integration should produce a latency in navigation that significantly hinders the ability to avoid obstacles. Can the authors provide statistics that describe this latency, and discussion about why it doesn't seem to be a problem? 

As described in the Methods section, the bat’s collision avoidance response does not solely rely on the integration process. Instead, the model incorporates real-time echoes from the last calls, which are used independently of the integration process for immediate obstacle avoidance maneuvers. This ensures that bats can react to nearby obstacles without being hindered by the integration latency. The slower integration on the other hand is used for clustering, outlier removal and estimation wall directions to support the pathfinding process, as illustrated in Supplementary Figure 1.

Additionally, our model assumes that bats store the physical positions of echoes in an allocentric coordinate system (x-y). The integration occurs after transforming these detections from a local relative reference frame to a global spatial representation. This allows for stable environmental mapping while maintaining responsiveness to immediate changes in the bat’s surroundings.

See lines 600-616 in the revised version.

The authors are using a 2D simulation, but this very much simplifies the challenge of a 3D navigation task, and there is an explanation as to why this is appropriate. Bat densities and bat behavior are discussed per unit area when realistically it should be per unit volume. In fact, the authors reference studies to justify the densities used in the simulation, but these studies were done in a 3D world. If the authors have justification for why it is realistic to model a 3D world in a 2D simulation, I encourage them to provide references justifying this approach. 

We acknowledge that this is a simplification; however, from an echolocation perspective, a 2D framework represents a worst-case scenario in terms of bat densities and maneuverability:

• Higher Effective Density: A 2D model forces all bats into a single plane rather than distributing them through a 3D volume, increasing the likelihood of overlap in calls and echoes and making jamming more severe. As described in the text: the average distance to the nearest bat in our simulation is 0.27m (with 100 bats), whereas reported distances in very dense colonies are 0.5m (Fujioka et al., 2021), as observed in Myotis grisescens (Sabol and Hudson, 1995) and Tadarida brasiliensis (Theriault et al., no date; Betke et al., 2008; Gillam et al., 2010)

• Reduced Maneuverability: In 3D space, bats can use vertical movement to avoid obstacles and conspecifics. A 2D constraint eliminates this degree of freedom, increasing collision risk and limiting escape options.

Thus, our 2D model provides a conservative difficult test case, ensuring that our findings are valid under conditions where jamming and collision risks are maximized. Additionally, the 2D framework is computationally efficient, allowing us to perform multiple simulation runs to explore a broad parameter space and systematically test the impact of different variables.

To address the reviewer’s concern, we have clarified this justification in the revised text and will provide supporting references where applicable (see Methods lines 450455).

The focus on "masking" (which appears to be just in-band noise), especially relative to the problem of misassigned echoes, is concerning. If the bat calls are all the same waveform (downsweep linear FM of some duration, I assume - it's not clear from the text), false echoes would be a major problem. Masking, as the authors define it, just reduces SNR. This reduction is something like sqrt(N), where N is the number of conspecifics whose echoes are audible to the bat, so this allows the detection threshold to be set lower, increasing the probability that a bat's echo will exceed a detection threshold. False echoes present a very different problem. They do not reduce SNR per se, but rather they cause spurious threshold excursions (N of them!) that the bat cannot help but interpret as obstacle detection. I would argue that in dense groups the mis-assignment problem is much more important than the SNR problem. 

There is substantial literature supporting the assumption that bats can recognize their own echoes and distinguish them from conspecific signals (Schnitzler, Bioscience and 2001, no date; Kazial, Burnett and Masters, 2001; Burnett and Masters, 2002; Kazial, Kenny and Burnett, 2008; Chili, Xian and Moss, 2009; Yovel et al., 2009; Beetz and Hechavarría, 2022)). However, we acknowledge that false echoes may present a major challenge in dense groups. To address this, we explicitly tested the impact of the self-echo identification assumption in our study see Results Figure 1: The impact of confusion on performance, and lines 399-404 in the Discussion.

Furthermore, we examined a full confusion scenario, where all reflected echoes from conspecifics were misinterpreted as obstacle reflections (i.e., 100% confusion). Our results show that this significantly degrades navigation performance, supporting the argument that echo misassignment is a critical issue. However, we also explored a simple mitigation strategy based on temporal integration with outlier rejection, which provided some improvement in performance. This suggests that real bats may possess additional mechanisms to enhance self-echo identification and reduce false detections. See lines 411-420 in the manuscript for further discussion. 

We actually used logarithmically frequency modulated (FM) chirps, generated using the MATLAB built-in function chirp(t, f0, t1, f1, 'logarithmic'). This method aligns with the nonlinear FM characteristics of Pipistrellus kuhlii (PK) and Rhinopoma microphyllum (RM) and provides a realistic approximation of their echolocation signals. We acknowledge that this was not sufficiently emphasized in the original text, and we have now explicitly highlighted this in the revised version to ensure clarity (see Lines 509-512 in Methods).

The criteria set for flight behavior (lines 393-406) are not justified with any empirical evidence of the flight behavior of wild bats in collective flight. How did the authors determine the avoidance distances? Also, what is the justification for the time limit of 15 seconds to emerge from the opening? Instead of an exit probability, why not instead use a time criterion, similar to "How long does it take X% of bats to exit?"  :

While we acknowledge that wild bats may employ more complex behaviors for collision avoidance, we chose to implement a simplified decision-making rule in our model to maintain computational tractability.

The avoidance distances (1.5 m from walls and 0.4 m from other bats) were selected as internal parameters to support stable and realistic flight trajectories while maintaining a reasonable collision rate. These values reflect a trade-off between maneuverability and behavioral coherence under crowding. To address this point, we added a sensitivity analysis to the revised manuscript. Specifically, we tested the effect of varying the conspecific avoidance distance from 0.2 to 1.6 meters at bat densities of 2 to 40 bats/3m². The only statistically significant impact was at the highest density (40 bats/3m²), where exit probability increased slightly from 82% to 88% (p = 0.024, t = 2.25, DF = 958). No significant changes were observed in exit time, collision rate, or jamming probability across other densities or conditions (GLM, see revised Methods). These results suggest that the selected avoidance distances are robust and not a major driver of model performance, see lines 469-47.

The 15-second exit limit was determined as described in the text (Lines 489-491): “A 15-second window was chosen because it is approximately twice the average exit time for 40 bats and allows for a second corrective maneuver if needed.” In other words, it allowed each bat to circle the ‘cave’ twice to exit even in the most crowded environment. This threshold was set to keep simulation time reasonable while allowing sufficient time for most bats to exit successfully.

We acknowledge that the alternative approach suggested by the reviewer— measuring the time taken for a certain percentage of bats to exit—is also valid. However, in our model, some outlier bats fail to exit and continue flying for many minutes, such simulations would lead to excessive simulation times making it difficult to generate repetitions and not teaching us much – they usually resulted from the bat slightly missing the opening (see video S1. Our chosen approach ensures practical runtime constraints while still capturing relevant performance metrics.

What is the empirical justification for the 1-10 calls used for integration?  

The "average exit time for 40 bats" is also confusing and not well explained. Was this determined empirically? From the simulation? If the latter, what are the conditions?

Does it include masking, no masking, or which species? 

Previous studies have demonstrated that bats integrate acoustic information received sequentially over several echolocation calls (2-15), effectively constructing an auditory scene in complex environments (Ulanovsky and Moss, 2008; Chili, Xian and Moss, 2009; Moss and Surlykke, 2010; Yovel and Ulanovsky, 2017; Salles, Diebold and Moss, 2020). Additionally, bats are known to produce echolocation sound groups when spatiotemporal localization demands are high (Kothari et al., 2014). Studies have documented call sequences ranging from 2 to 15 grouped calls (Moss and Surlykke, 2010), and it has been hypothesized that grouping facilitates echo segregation.

We did not use a single integration window - we tested integration sizes between 1 and 10 calls and presented the results in Figure 3A. This range was chosen based on prior empirical findings and to explore how different levels of temporal aggregation impact navigation performance. Indeed, the results showed that the performance levels between 5-10 calls integration window (Figure 3A)

Regarding the average exit time for 40 bats, this value was determined from our simulations, where it represents the mean time for successful exits under standard conditions with masking. We have revised the text to clarify these details see, lines 489-491.

Reviewer #1 (Recommendations for the authors):

(1) Data Availability:

As it stands now, this reviewer cannot vouch for the uploaded code as it wasn't accessible according to F.A.I.R principles. The link to the code/data points to a private company's file-hosting account that requires logging in or account creation to see its contents, and thus cannot be accessed.

This reviewer urges the authors to consider uploading the code onto an academic data repository from the many on offer (e.g. Dryad, Zenodo, OSF). Some repositories offer an option to share a private link (e.g. Zenodo) to the folder that can then be shared only with reviewers so it is not completely public.

This is a computational paper, and the credibility of the results is based on the code used to generate them.

The code is available at GitHub as required:

https://github.com/omermazar/Colony-Exit-Bat-Simulation

(2) Abstract:

Line 22: 'To explore whether..' - replace 'whether' with 'how'?

The sentence was rephrased as suggested by the reviewer.

(2) Main text:

Line 43: '...which may share...' - correct to '...which share...', as elegantly framed in the authors' previous work - jamming avoidance is unavoidable because all FM bats of a species still share >90% of spectral bandwidth despite a few kHz shift here and there.

The sentence was rephrased as suggested by the reviewer.

Line 49: The authors may wish to additionally cite the work of Fawcett et al. 2015 (J. Comp. Phys A & Biology Open)

Thank you for the suggestion. We have included a citation to the work of Fawcett et al. (2015) in the revised manuscript.

Line 61: This statement does not match the recent state of the literature. While the previous models may have assumed that all neighbours can be detected, there are models that specifically study the role of limited interaction arising from the potential inability to track all neighbours, and the effect of responding to only one/few neighbours at a time e.g. Bode et al. 2011 R. Soc. Interface, Jhawar et al. 2020 Nature Physics.

We have added citations to the important studies suggested by the reviewer, as detailed in the Public Review above.

Line 89: '..took all interference signals into account...' - what is meant by 'interference signals' - are the authors referring to reflections, unclear.

We have revised the sentence and detailed the acoustic signals involved in the process: self-generated echoes, calls from conspecifics, and echoes from cave walls and other bats evoked by those calls, see lines 99-106.

Figure 1A: The colour scheme with overlapping points makes the figure very hard to understand what is happening. The legend has colours from subfigures B-D, adding to the confusion.

What does the yellow colour represent? This is not clear. Also, in general, the color schemes in the simulation trajectories and the legend are not the same, creating some amount of confusion for the reader. It would be good to make the colour schemes consistent and visually separable (e.g. consp. call direct is very similar to consp. echo from consp. call), and perhaps also if possible add a higher resolution simulation visualisation. Maybe it is best to separate out the colour legends for each sub-figure.

The updated figure now includes clearer, more visually separable colors, and consistent color coding across all sub-panels. The yellow trajectory representing the focal bat’s flight path is now explicitly labeled, and we adjusted the color mapping of acoustic signals (e.g., conspecific calls vs. echoes) to improve distinction. We also revised the figure caption accordingly and ensured that the legend is aligned with the updated visuals. These modifications aim to enhance interpretability and reduce ambiguity for the reader.

Figure C3: What is 'FB Channel', this is not explained in the legend.

FB Channel’ stands for ‘Filter Bank Channel’. This clarification has been added to the caption of Figure 1. 

Figure 3: Visually noticing that the colour legend is placed only on sub-figure A is tricky and readers may be left searching for the colour legend. Maybe lay out the legend horizontally on top of the entire figure, so it stands out?

We have adjusted the placement of the color legend in Figure 3 to improve visibility and consistency.

Line 141: '..the probability of exiting..' - how is this probability calculated - not clear.

We have clarified in the revised text that the probability of exiting the cave within 15 seconds is defined as the number of bats that exited the cave within that time divided by the total number of bats in each scenario, see lines 159160.

Line 142: What are the sample sizes here - i.e. how many simulation replicates were performed?

We have clarified the number of repetitions in each scenario the revised text, as detailed in the Public Review above.

Line 151: 'The jamming probability,...number of jammed echoes divided by the total number of reflected echoes' - it seems like these are referring to 'own' echoes or first-order reflections, it is important to clarify this.

The reviewer is right. We have clarified it in the revised text, see lines 173175.

Line 153: '..with a maximum difference of ...' - how is this difference calculated? What two quantities are being compared - not clear.

We have revised the text to clarify that the 14.3% value reflects the maximum difference in jamming probability between the RM and PK models, which occurred at a density of 10 bats. The values at each density are shown in Figure 2D, see lines 175-177.

Line 221: '..temporal aggregation helps..' - I'm assuming the authors meant temporal integration? However, I would caution against using the exact term 'temporal integration' as it is used in the field of audition to mean something different. Perhaps something like 'sensory integration' , or 'multi-call integration'

To avoid ambiguity and better reflect the process modeled in our work, we have replaced the term "temporal aggregation" with "multi-call integration" throughout the revised manuscript. This term more accurately conveys the idea of combining information from multiple echolocation calls without conflicting with existing terminology.

(4) Discussion

Lines 302: 'Our model suggests...increasing the call-rate..' - not clear where this is explicitly tested or referred to in this manuscript. Can't see what was done to measure/quantify the effect of this variable in the Methods or anywhere else.

We have rephrased this paragraph as detailed in the Public Review above, see lines 346-349.

Line 319: 'spatial interference' - unclear what this means. This reviewer would strongly caution against creating new terms unless there is an absolute need for it. What is meant by 'interference' in this paper is hard to assess given that the word seems to be used as a synonym for jamming and also for actual physical wave-based interference.

We have rephrased this paragraph as detailed in the Public Review above, see line 119-120, 366-367.

Line 323: '..no benefit beyond a certain level...' - also not clear where this is explicitly tested. It seems like there was a set of simulations run for a variety of parameters but this is not written anywhere explicitly. What type of parameter search was done, was it all possible parameter combinations - or only a subset? This is not clear.

We have rephrased this paragraph as detailed in the Public Review above, see lines 372-375.

Line 324: '..ca. 110 dB-SPL.' - what reference distance?

All call levels were simulated and reported in dB-SPL, referenced at 0.1 meters from the emitting bat. We have clarified it in the revised text in the relevant contexts and specifically in line 529.

(5) Methods

Line 389 : '...over a 2 x 1.5 m2 area..' It took a while to understand this statement and put it in context. Since there is no previous description of the entire L-arena, the reviewer took it to mean the simulations happened over the space of a 2 x 1.5 m2 area. Include a top-down description of the simulation's spatial setup and rephrase this sentence.

To address the confusion, we revised the text to clarify that the full simulation environment represents a corridor-shaped cave measuring 14.5 × 2.5 meters, with a right-angle turn located 5.5 meters before the exit, as shown in Figure 1A. The 2 × 1.5 m area refers specifically to the small zone at the far end of the cave where bats begin their flight. The revised description now includes a clearer spatial overview to prevent ambiguity, see lines 456-460.

Line 398: Replace 'High proximity' with 'Close proximity'

Replaced.

Line 427: 'uniform target strength of -23 dB' - at what distance is this target strength defined? Given the reference distance can vary by echolocation convention (0.1 or 1 m), one can't assess if this is a reasonable value or not.

The reference distance for the reported target strength is 1 meter, in line with standard acoustic conventions. We have revised the text to clarify this explicitly (line 531).

Also, independent of the reference distance, particularly with reference to bats, the target strength is geometry-dependent, based on whether the wings are open or not. Using the entire wingspan of a bat to parametrise the target strength is an overestimate of the available reflective area. The effective reflective area is likely to be somewhere closer to the surface area of the body and a fraction of the wingspan together. This is important to note and/or mention explicitly since the value is not experimentally parametrised.

For comparison, experimentally based measurements used in Goetze et al. 2016 are -40 dB (presumably at 1 m since the source level is also defined at 1 m?), and Beleyur & Goerlitz 2019 show a range between -43 to -34 dB at 1 m.

We agree with the reviewer that target strength in bats is strongly influenced by their geometry, particularly wing posture during flight. In our model, we simplified this aspect by using a constant target strength, as the detailed temporal variation in body and wing geometry is pseudo-random and not explicitly modeled. We acknowledge that this is a simplification, and have now stated this limitation clearly in the revised manuscript. We chose a fixed value of –23 dB at 1 meter to reflect a plausible mid-range estimate, informed by anatomical data and consistent with values reported for similarly sized species (Beleyur and Goerlitz, 2019). To support this, we directly measured the target strength of a 3D-printed RM bat model, obtaining –32dB. 

Moreover, a sensitivity analysis across a wide range (–49 to –23 dB) confirmed that performance metrics remain largely stable, indicating that our conclusions are not sensitive to this parameter, and suggesting that our results hold for different-sized bats. See lines 384-390, 533-538, and Supplementary Figures 3 and 4 in the revised article. 

Line 434: 'To model the bat's cochlea...'. Bats have two cochleas. This model only describes one, while the agents are also endowed with the ability to detect sound direction - which requires two ears/cochleas.... There is missing information about the steps in between that needs to be provided.

We appreciate the reviewer’s observation. Indeed, our model is monaural, and simulates detection using a single cochlear-like filter bank receiver. We have clarified this in the revised text to avoid confusion. This paragraph specifically describes the detection stage of the auditory processing pipeline. The localization process, which builds on detection and includes directional estimation, is described in the following paragraph (see line 583 onward), as discussed in the next comment and response.

Line 457: 'After detection, the bat estimates the range and Direction of Arrival...' This paragraph describes the overall idea, but not the implementation. What were the inputs and outputs for the range and DOA calculation performed by the agent? Or was this information 'fed' in by the simulation framework? If there was no explicit DOA step that the agent performed, but it was assumed that agents can detect DOA, then this needs to be stated.

In the current simulation, the Direction of Arrival (DOA) was not modeled via an explicit binaural processing mechanism. Instead, based on experimental studies (Simmons et al., 1983; Popper and Fay, 1995).  we assumed that bats can estimate the direction of an echo with an angular error that depends on the signal-to-noise ratio (SNR). Accordingly, the inputs to the DOA estimation were the peak level of the desired echo, noise level, and the level of acoustic interference. The output was an estimated direction of arrival that included a random angular error, drawn from a normal distribution whose standard deviation varied with the SNR. We have revised the relevant paragraph (Lines 583-592) to clarify this implementation.

Line 464: 'To evaluate the impact of the assumption...' - the 'self' and 'non-self' echoes can be distinguished perhaps using pragmatic time-delay cues, but also using spectro-temporal differences in individual calls/echoes. Do the agents have individual call structures, or do all the agents have the same call 'shape'? The echolocation parameters for the two modelled species are given, but whether there is call parameter variation implemented in the agents is not mentioned.

In our relatively simple model, all individuals emit the same type of chirp call, with parameters adapted only based on the distance to the nearest detected object. However, individual variation is introduced by assigning each bat a terminal frequency drawn from a normal distribution with a standard deviation of 1 kHz, as described in the revised version -lines 519-520. This small variation is not used explicitly as a spectro-temporal cue for echo discrimination.

In our model, all spectro-temporal variations—whether due to call structure or variations resulting from overlapping echoes from nearby reflectors—are processed through the filter bank, which compares the received echoes to the transmitted call during the detection stage. As such, the detection process itself can act as a discriminative filter, to some extent, based on similarity to the emitted call.

We acknowledge that real bats likely rely on a variety of spectro-temporal features for distinguishing self from non-self-echoes—such as call duration, received level, multi-harmonic structure, or amplitude modulation. In our simulation, we focus on comparing two limiting conditions: full recognition of self-generated echoes versus full confusion. Implementing a more nuanced self-recognition mechanism based on temporal or spectral cues would be a valuable extension for future work.

(6) References

Reference 22: Formatting error - and extra '4' in the reference.

The error has been fixed.

(7) Thoughts/comments

Even without 'recogntion' of walls & conspecifics, bats may be able to avoid obstacles - this is a neat result. Also, using their framework the authors show that successful 'blind' object-agnostic obstacle avoidance can occur only when supported by some sort of memory. In some sense, this is a nice intermediate step showing the role of memory in bat navigation. We know that bats have good long-term and long-spatial scale memory, and here the authors show that short-term spatial memory is important in situations where immediate sensory information is unreliable or unavailable.

We appreciate the reviewer’s thoughtful summary. Indeed, one of the main takeaways of our study is that successful obstacle avoidance can occur even without explicit recognition of walls or conspecifics—provided that a clustered multi-call integration is in place. Our model shows that when immediate sensory information is unreliable, integrating detections over time becomes essential for effective navigation. This supports the broader view that memory, even on short timescales, plays an important role in bat behavior.

(8) Reporting GLM results

The p-value, t-statistic, and degrees of freedom are reported consistently across multiple GLM results. However, the most important part which is the effect size is not consistently reported - and this needs to be included in all results, and even in the table. The effect size provides an indicator of the parameter's magnitude, and thus scientific context.

We agree that the effect size provides essential scientific context. In fact, we already include the effect size explicitly in Table 1, as shown in the “Effect Size” column for each tested parameter. These values describe the magnitude of each parameter’s effect on exit probability, jamming probability, and collision rate. In the main text, effect sizes are presented as concrete changes in performance metrics (e.g., “exit probability increased from 20% to 87%,” or “with a decrease of 3.5%±8% to 5.5%±5% (mean ± s.e.)”), which we believe improves interpretability and scientific relevance.  

To further clarify this in the main text, we have reviewed the reported results and ensured that effect sizes are mentioned more consistently wherever GLM outcomes are discussed. Additionally, we have added a brief note in the table caption to emphasize that effect sizes are provided for all tested parameters.

The 'tStat' appears multiple times and seems to be the output of the MATLAB GLM function. This acronym is specific to the MATLAB implementation and needs to be replaced with a conventionally used acronym such as 't', or the full form 't-statistic' too. This step is to keep the results independent of the programming language used.

We have replaced all instances of tStat with the more conventional term ‘t’ throughout the manuscript to maintain consistency with standard reporting practices.

Reviewer #2 (Recommendations for the authors):

In addition to my public review, I had a few minor points that the authors may want to consider when revising their paper.

(1) Figures 2, 3, and 4 may benefit from using different marker styles, in addition to different colors, to show the different cases.

Thank you for the suggestion. In Figures 2–4, the markers represent means with standard error bars. To maintain clarity and consistency across all conditions, we have chosen to keep a standardized marker style – and we clarify this in the legend. We found that varying only the colors is sufficient for distinguishing between conditions without introducing visual clutter.

(2) The text "PK" in the inset for Figure 2A is very difficult to read. I would suggest using grey as with "RM" in the other inset.

We have updated the insert in Figure 2A to improve legibility.

(3) Are the error bars in Figure 3 very small? I wasn't able to see them. If that is the case, the authors may want to mention this in the caption.

You are correct—the error bars are present in all plots but appear very small due to the large number of simulation repetitions and low variability. We have revised the caption to explicitly mention this.

(4) The species name of PK is spelled inconsistently (kuhli, khulli, and kuhlii).

We have corrected the species name throughout the manuscript.

(5) Table 1 is a great condensation of all the results, but the time to exit is missing. It may be helpful if summary statistics on that were here as well.

We have added time-to-exit to the effect size column in Table 1, alongside the other performance metrics, to provide a more complete summary of the simulation results.

(6) I may have missed it, but why are there two values for the exit probability when nominal flight speed is varied?

The exit probability was not monotonic with flight speed, but rather showed a parabolic trend with a clear optimum. Therefore, we reported two values representing the effect before and after the peak. We have clarified this in the revised table and updated the caption accordingly.

(7) Table 2 has an extra header after the page break on page 18.

The extra header in Table 2 after the page break has been removed in the revised manuscript.

(8) The G functions have 2 arguments in their definitions and Equation 1, but only one argument in Equations 2 and 3. I wasn't able to see why.

Thank you for pointing this out. You are correct—this was a typographical error. We have corrected the argument notation in Equations 2 and 3 and explicitly included the frequency dependence of the gain (G) functions in both equations.

(9) D_txrx was not defined but it was used in Equation 2.

The variable D_txrx is defined in the equation notation section as: D_₍ₜₓr_ₓ – the distance [m] between the transmitting conspecific and the receiving focal bat, from the transmitter’s perspective. We have now ensured that this definition is clearly linked to Equation 2 in the revised text. Moreover, we have added a supplementary figure that illustrates the geometric configuration defined by the equations to further support clarity, as described in the Public Review above.

(10) It was hard for me to understand what was meant by phi_rx and phi_tx. These were described as angles between the rx or tx bats and the target, but I couldn't tell what the point defining the angle was. Perhaps a diagram would help, or more precise definitions.

We have revised the caption to provide clearer and more precise definitions Additionally, we have included a geometric diagram as a supplementary figure, as noted in the Public Review above, to visually clarify the spatial relationships and angle definitions used in the equations, see lines 498-499.

(11) Was the hearing threshold the same for both species?

Yes. We have clarified it in the revised version.

(12) Collision avoidance is described as turning to the "opposite direction" in the supplemental figure explaining the model. Is this 90 degrees or 180 degrees? If 90 degrees, how do these turns decide between right and left?

In our model, the bat does not perform a fixed 90° or 180° turn. Instead, the avoidance behavior is implemented by setting the maximum angular velocity in the direction opposite to the detected echo. For example, if the obstacle or conspecific is detected on the bat’s right side, the bat begins turning left, and vice versa.

This turning direction is re-evaluated at each decision step, which occurs after every echolocation pulse. The bat continues turning in the same direction if the obstacle remains in front, otherwise it resumes regular pathfinding. We have clarified this behavior in the updated figure caption and model description, see lines 478-493.

Reviewer #3 (Recommendations for the authors):

(1) Lines 27-31: These sentences mischaracterize the results. This claim appears to equate "the model works" with "this is what bats actually do." Also, the model does not indicate that bats' echolocation strategies are robust enough to mitigate the effects of jamming - this is self-evident from the fact that bats navigate successfully via echolocation in dense groups.

Thank you for the comment. Our aim was not to claim that the model confirms actual bat behavior, but rather to demonstrate that simple and biologically plausible strategies—such as signal redundancy and basic pathfinding—are sufficient to explain how bats might cope with acoustic interference in dense settings. We have revised the wording to better reflect this goal and to avoid overinterpreting the model's implications.

See abstract in the revised version.  

(2) Line 37: This number underestimates the number of bats that form some of the largest aggregations of individuals worldwide - the free-tailed bats can form aggregations exceeding several million bats.

We have revised the text to reflect that some bat species, such as free-tailed bats, are known to form colonies of several million individuals, which exceed the typical range. The updated sentence accounts for these extreme cases, see lines 36-37.

(3) The flight densities explained in the introduction and chosen references are not representative of the literature - without providing additional justification for the chosen species, it can be interpreted that the selection of the species for the simulation is somewhat arbitrary. If the goal is to model dense emergence flight, why not use a species that has been studied in terms of acoustic and flight behavior during dense emergence flights---such as Tadarida brasiliensis?

Our goal was to develop a general model applicable to a broad class of FMecholocating bat species. The two species we selected—Pipistrellus kuhlii (PK) and Rhinopoma microphyllum (RM)—span a wide range of signal characteristics: from wideband (PK) to narrowband (RM), providing a representative contrast in call structure. 

Although we did not include Tadarida brasiliensis (TB) specifically, its echolocation calls are acoustically similar to RM in terminal frequency and fall between PK and RM in bandwidth. Therefore, we believe our findings are likely to generalize to TB and other FM-bats.

Moreover, as noted in a previous response, the average inter-bat distance in our highest-density simulations (0.27 m) is still smaller than those reported for Tadarida brasiliensis during dense emergences—further supporting the relevance of our model to such scenarios.

To support broader applicability, we also provide a supplementary graphical user interface (GUI) that allows users to modify key echolocation parameters and explore their impact on behavior—making the framework adaptable to additional species, including TB.

(4) Line 78: It is not clear how (or even if) the simulated bats estimate the direction of obstacles. The explanation given in lines 457-463 is quite confusing. What is the acoustic/neurological mechanism that enables this direction estimation? If there is some mechanism (such as binaural processing), how does this extrapolate to 3D?

This comment echoes a similar concern raised by a previous reviewer. As explained earlier, in the current simulation, the Direction of Arrival (DOA) was not modeled via an explicit binaural processing mechanism. The complete  is detailed in  to Reviewer #1, Line 457. This implementation is now clarified in the revised text, and a detailed description of the localization process is also provided in the Methods section (lines 583-592).

(5) The authors propose they are modeling the dynamic echolocation of bats in the simulation (line 79), but it appears (whether this is due to a lack of information in the manuscript or true lack in the simulation) that the authors only modeled a flight response. How did the authors account for bats dynamically changing their echolocation? This is unclear and from what I can tell may just mean that the bats can switch between foraging phase call types depending on the distance to a detected obstacle. Can the authors elaborate more on this?

The echolocation behavior of the bats—including dynamic call adjustments— was implemented in the simulation and is described in detail in the Methods section (lines 498-520 and Table 2). To avoid redundancy, the Results chapter originally referred to this section, but we have now added a brief explanation in the Results to clarify that the bats’ call parameters (IPI, duration, and frequency range) adapt based on the distance to detected objects, following empirically documented echolocation phases ("search," "approach," "buzz"). These dynamics are consistent with established bat behavior during navigation in cluttered environments such as caves.

(6) Figure 1 C3: "Detection threshold": what is this and how was it derived?

The caption also mentions yellow arrows, but they are absent from the figure. C4: Each threshold excursion is marked with an asterisk, but there are many more excursions than asterisks. Why are only some marked? Unclear.

C3: The detection threshold is determined dynamically. It is set to the greater of either 7 dB above the noise level (0 dB-SPL)(Kick, 1982; Saillant et al., 1993; Sanderson et al., 2003; Boonman et al., 2013) or the maximal received level minus 70 dB, effectively applying a dynamic range of 70 dB. This clarification has been added to the Methods section. The yellow arrow has been added.

C4: Thank you for this important observation. Only peaks marked with asterisks represent successful detections—those that were identified in both the interference-free and full detection conditions, as explained in the Methods. Other visible peaks result from masking signals or overlapping echoes from nearby reflectors, but they do not meet the detection criteria. To keep the figure caption concise, we have elaborated on this process more clearly in the revised Methods section. We added this information to the legend

(7) Figure 2: A line indicating RM, No Masking is absent

Thank you for pointing this out. The missing line for RM, No Masking has now been added in the revised version of Figure 2.

(8) Line 121: "reflected off conspecifics". Does this mean echoes due to conspecifics?

The phrase "reflected off conspecifics" refers to echoes originating from the bat’s own call and reflected off the bodies of nearby conspecifics. We have clarified the wording in the revised text to avoid confusion

(9) Line 125: Why are low-frequency channels stimulated by higher frequencies? This needs further clarification.

The cochlear filter bank in our model is implemented using gammatone filters, each modeled as an 8th-order Butterworth filter. Due to the non-ideal filter response and relatively broad bandwidths—especially in the lower-frequency channels—strong energy from the beginning of the downward FM chirp (at higher frequencies) can still produce residual activation in lower-frequency channels. While these stimulations are usually below the detection threshold, they may still be visible as early sub-threshold responses. Given the technical nature of this explanation (a property of the filter implementation) and it does not influence the detection outcomes, we have chosen not to elaborate on it in the figure caption or Methods.

(10) Lines 146-150: This is an interesting finding. Is there a theoretical justification for it?

This outcome arises directly from the simulation results. As noted in the Discussion (lines 359-365), although Pipistrellus kuhlii (PK) shows a modest advantage in jamming resistance due to its broader bandwidth, the redundancy in sensory information across calls—enabled by frequent echolocation—appears to compensate for these signal differences. As a result, the small variations in echo quality between species do not translate into significant differences in performance. We speculate that if the difference in jamming probability had been larger, performance disparities would likely have emerged.

(11) Line 151: The authors define a jammed echo as an echo entirely missed due to masking. Is this appropriate? Doesn't echo mis-assignment also constitute jamming?

We agree that echo mis-assignment can also degrade performance; however, in our model, we distinguish between two outcomes: (1) complete masking (echo not detected), and (2) detection with a localization error. As explained in the Methods (lines 500–507), we run the detection analysis twice—once with only desired echoes (“interference-free detection”) and once including masking signals (“full detection”). If a previously detected echo is no longer detected, it is classified as a jammed echo. If the echo is still detected but the delay shifts by more than 100 µs compared to the interference-free condition, it is also considered jammed. If the delay shift is smaller, it is treated as a detection with localization error rather than full jamming. We have clarified this distinction in the revised Methods section.

(12) Figure 2-E: Detection probability statistics are of limited usefulness without accompanying false alarm rate (FAR) statistics. Do the authors have FAR numbers?

We understand FAR to refer to instances where masking signals or other acoustic phenomena are mistakenly interpreted as real echoes from physical objects. As explained in the manuscript, we implemented two model versions: one without confusion, and one with full confusion.

Figure 2E reports detection performance under the non-confusion model, in which only echoes from actual physical reflectors are used, and no false detections occur—hence, the false alarm rate is effectively zero in this condition. In the full-confusion model, all detected echoes—including those originating from masking signals or conspecific calls—are treated as valid detections, which may include false alarms. However, we did not explicitly quantify the false alarm rate as a separate metric in this simulation.

We agree that tracking FAR could be informative and will consider incorporating it into future versions of the model.

(13) Line 161: RM bats suffered from a significantly higher probability of the "desired conspecific's echoes" being jammed. What does "desired conspecific's echoes" mean? This is unclear.

The term “desired conspecific's echoes” refers to echoes originating from the bat’s own call, reflected off nearby conspecifics, which are treated as relevant reflectors for collision avoidance. We have revised the wording in the text for clarity.

(14) Line 188: Why didn't the size of the integration window affect jamming probability? I couldn't find this explained in the discussion.

The jamming probability in our analysis is computed at the individual-echo level, prior to any temporal integration. Since the integration window is applied after the detection step, it does not influence whether a specific echo is masked (i.e., jammed) or not. Therefore, as expected, we did not observe a significant effect of integration window size on jamming probability.

(15) Line 217-218: Why do the authors think this would be?

Thank you for the thoughtful question. We agree that, in theory, increasing call intensity should raise the levels of both desired echoes and masking signals proportionally. However, in our model, the environmental noise floor and detection threshold remain constant, meaning that higher call intensities increase the signal-to-noise ratio (SNR) more effectively for weaker echoes, especially those at longer distances or with low reflectivity. This could lead to a higher likelihood of those echoes crossing the detection threshold, resulting in a small but measurable reduction in jamming probability.

Additionally, the non-linear behavior of the filter-bank receiver—including such as thresholding at multiple stages—can introduce asymmetries in how increased signal levels affect the detection of target versus masking signals.

That said, the effect size was small, and the improvement in jamming probability did not translate into any significant gain in behavioral performance (e.g., exit probability or collision rate), as shown in Figure 3C.

(16) Line 233: I'm not sure I understand how a slightly improved aggregation model that clustered detected reflectors over one-second periods is different. Doesn't this just lead to on average more calls integrated into memory?

While increasing the memory duration does lead to more detections being available, the enhanced aggregation model (we now refer to as multi-call clustering) differs fundamentally from the simpler one. As detailed in the Methods, it includes additional processing steps: clustering spatially close detections, removing outliers, and estimating wall directions based on the spatial structure of clustered echoes. In contrast, the simpler model treats each detection as an isolated point without estimating obstacle orientation. These additional steps allow for more robust environmental interpretation and significantly improve performance under high-confusion conditions. We have clarified it in revised text (lines 606-616) and added a Supplementary Figure 2B.

(17) Table 1: What about conspecific target strength?

We have now added the conspecific target strength as a tested parameter in Table 1, along with its tested range, default value, and measured effect sizes. A detailed sensitivity analysis is also presented in Supplementary Figure 4, demonstrating that variations in conspecific target strength had relatively minor effects on performance metrics.  

(18) Figure 3-A: The x-axis is the number of calls in the integration window. But the leftmost sample on each curve is at 0 calls. Shouldn't this be 1?

“0 calls” refers to the case where only the most recent call is used for pathfinding—without integrating any information from prior calls. The x-axis reflects the number of previous calls stored in memory, so a value of 0 still includes the current call. We’ve clarified this terminology in the figure caption.

(19) Lines 282-283: This statement needs to be clarified that it is with the constraints of using a 2D simulation with at most 33 bats/m^2. It also should be clarified that it is assumed the bat can reliably distinguish between its own echoes and conspecific echoes, which is a very important caveat.

We have revised the text to clarify that the results are based on a 2D simulation with a maximum tested density of 33 bats/m². We also now explicitly state that the model assumes bats can distinguish between their own echoes and those generated by conspecifics—an assumption we recognize as a simplification. These clarifications help place the results within the scope and constraints of the simulation. Moreover, as described in the text (and noted in previous response): the average distance to the nearest bat in our simulation is 0.27m (with 100 bats), whereas reported distances in very dense colonies are 0.5m

(20) Line 294: What is this sentence referring to?

The sentence refers to the finding that, even under high bat densities, a substantial portion of the echoes—particularly those reflected from nearby obstacles (e.g., 1 m away)—were jammed due to masking. Nevertheless, the bats in the simulation were still able to navigate successfully using partial sensory input. We have clarified the sentence in the revised text to make this point more explicit, see line 333-336.

(21) Line 302: Was jamming less likely when IPI was higher or lower? I could not find this demonstrated anywhere in the manuscript.

We agree that the original text was not sufficiently clear on this point. While we did not explicitly test fixed IPI values as a parameter, the model does simulate the natural behavior of decreasing IPI as bats approach obstacles. This behavior is supported by empirical observations and is incorporated into the echolocation dynamics of the simulation. We have clarified this point in the revised text (see Lines 346-351) and explained that while lower IPI introduces more acoustic overlap, it also increases redundancy and improves detection through temporal integration.

(22) Lines 313-314: This is an interesting assumption, but it is not evident that is substantiated by the references.

The claim is based on well-established principles in signal processing and bioacoustics. Wideband signals—such as those emitted by PK bats— distribute their energy over a broader frequency range, which makes them inherently more resistant to narrowband interference and masking. This concept is commonly applied in both biological and artificial sonar systems and is supported by empirical studies in bats and theory in acoustic sensing.

For example, Beleyur & Goerlitz (2019) demonstrate that broader bandwidth calls improve detection in cluttered and jamming-prone environments. Similarly, Ulanovsky et al. (2004) and Schnitzler & Kalko (200) discuss how FM bats' wideband calls enhance temporal and spatial resolution, helping to reduce the impact of overlapping signals from conspecifics. These findings align with communication theory where spread-spectrum techniques improve robustness in noisy environments.

We agree with the reviewer that this is an important point and we have updated the manuscript to clarify this rationale and cite the relevant literature accordingly – lines 631-363,

(23) Lines 318-319: What is the justification for "probably"? Isn't this just a supposition?

We agree with the reviewer’s point and have rephrased the sentence

(24) Line 320: How does this 63% performance match the sentence in line 295?

The sentence in Line 295 refers to the overall ability of the bats to navigate successfully despite high jamming levels, highlighting the robustness of the strategy under challenging conditions. The figure in Line 320 (63%) quantifies this performance under the most extreme simulated scenario (100 bats / 3 m²), where both spatial and acoustic interferences are maximal. We have rephrased the text in the revised version (lines 324-327).

(25) Lines 341-345: It seems like this is more likely to be the main takeaway of the paper.

As noted in the Public Review above, there is substantial literature supporting the assumption that bats can recognize their own echoes and distinguish them from those of conspecifics (e.g., Schnitzler, Bioscience, 2001; Kazial et al., 2001, 2008; Burnett & Masters, 2002; Chiu et al., 2009; Yovel et al., 2009; Beetz & Hechavarría, 2022). Therefore, we consider our assumption of selfrecognition to be well-supported, at least under typical conditions. That said, we agree that the impact of echo confusion on performance is significant and highlights a critical challenge in dense environments.

To our knowledge, this is the first computational model to explicitly simulate both self-recognition and full echo confusion under high-density conditions. We believe that the combination of modeled constraints and the demonstrated robustness of simple sensorimotor strategies, even under worst-case assumptions, is what makes this contribution both novel and meaningful.

(26) Lines 349-350: What is the aggregation model? What is meant by "integration"?

We have revised the text to clarify that the “aggregation model” refers to a multi-call clustering process that includes clustering of detections, removal of outliers, and estimation of wall orientation, as described in detail in the revised Methods and Results sections.

(27) Line 354: Again, why isn't this the assumption we're working under?

As addressed in our response to Comment 25, our primary model assumes that bats can recognize their own echoes—an assumption supported by substantial empirical evidence. The alternative "full confusion" model was included to explore a worst-case scenario and highlight the behavioral consequences of failing to distinguish self from conspecific echoes. We assume that real bats may experience some degree of echo misidentification; however, our assumption of full confusion represents a worst-case scenario.

(28) Line 382: "Under the assumption that..." I agree that bats probably can, but if we assume they can differentiate them all, where's the jamming problem?

The assumption that bats can theoretically distinguish between different signal sources applies after successful detection. However, the jamming problem arises during the detection and localization stages, where acoustic interference can prevent echoes from crossing the detection threshold or distort their timing.

(29) Lines 386-387: The paper referenced focused on JAR in the context of foraging. What changes were made to the simulation to switch to obstacle avoidance?

While the simulation framework in Mazar & Yovel (2020) was developed to study jamming avoidance during foraging, the core components—such as the acoustic calculations, receiver model, and echolocation behavior—remain applicable. For the current study, we adapted the simulation extensively to address colony-exit behavior. These modifications include modeling cave walls as acoustic reflectors, implementing a pathfinding algorithm, integrating obstacle-avoidance maneuvers, and adapting the integration window and integration processes. These updates are detailed throughout the Methods section.

(30) Line 400-402: Something doesn't add up with the statement: each decision relies on an integration window that records estimated locations of detected reflectors from the last five echolocation calls, with the parameter being tested between 1 and 10 calls. Can the authors reword this to make it less confusing?

We have reworded the sentence to clarify that the default integration window includes five calls, while we systematically tested the effect of using 1 to 10 calls, see lines 486-487.

(31) Line 393: "30 deg/sec" why was this value chosen?

The turning rate of 30 deg/sec was manually selected to approximate the curvature of natural foraging flight paths observed in Rhinopoma microphyllum using on-board tags. Moreover, in Mazar & Yovel (2020), we showed that the flight dynamics of simulated bats in a closed room closely matched those of Pipistrellus kuhlii flying in a room of similar dimensions. However, in the current simulation, bats rarely follow a random-walk trajectory due to the structured environment and frequent obstacle detection. As a result, this parameter has no meaningful impact on the simulation outcomes.

(32) Line 412: "Harmony" --- do you mean harmonic? And what is the empirical evidence that RM bats use the 2nd harmonic compared to the 1st?

Perhaps showing a spectrogram of a real RM signal would be helpful.

The typo-error was corrected. For reference See (Goldshtein et al., 2025)

(33) Table 2: Something is incorrect with the table. The first row on the next page is the wrong species name. Also, where are the citations for these parameter values?

The table header has been corrected in the revised version. The parameter values for flight and echolocation behavior were derived from existing literature and empirical data: Pipistrellus kuhlii parameters were based on Kalko (1995), and Rhinopoma microphyllum parameters were extracted from our own recordings using on-board tags, as described in Goldstein et al. (2025). We have added the appropriate citations to Table 2.

(34) Line 442: How was the threshold level chosen?

The detection threshold in each level is set to the greater of either 7 dB above the noise level (0 dB-SPL) or the maximal received level minus 70 dB, effectively applying a dynamic range of 70 dB.

(35) Line 445: 100 micros: This is about 3cm. The resolution of PK is about 1cm. For RM it's about 10cm. So, this window is generous for PK, but too strict for RM.

To keep the model simple and avoid introducing species-specific detection thresholds, we selected a biologically plausible compromise that could reasonably apply to both species. This simplification ensures consistency across simulations while remaining within the known behavioral range.

(36) Line 448: What is the spectrum of the Gaussian noise, and did it change between PK and RM?

We used the same white Gaussian noise with a flat spectrum across the relevant frequency range (10–80 kHz) for both species. We have clarified this in the revised text in lines 570-572.

(37) Line 451: 4 milliseconds is 1.3m. Is this appropriate?

The 4 milliseconds window was selected based on established auditory masking thresholds described in Mazar & Yovel (2020), and supported by (Popper and Fay, 1995) ch. 2.4.5, ((Blauert, 1997),  ch. 3.1 and (Mohl and Surlykke, 1989). These values provide conservative lower bounds on bats’ ability to cope with masking (Beleyur and Goerlitz, 2019). For simplicity, we used constant thresholds within each window, see lines 574-576.  

(38) Line 452: Citation for the forward and backward masking durations?

See the  to the previous comment.

(39) Lines 460-461: This is unclear. How does the bat get directional information? The authors claim to be able to measure direction-of-arrival for each detection, but it is not clear how this is done

As noted in our response to Reviewer 1 (Comment on Line 457), directional information is not computed via an explicit binaural model. Instead, we assume the bat estimates the direction of arrival with an angular error that depends on the SNR, based on established studies (e.g., Simmons et al., 1983; Popper & Fay, 1995). We have clarified this in the revised text in lines 583-592.

(40) Line 467: It seems like the authors are modeling pulse-echo ambiguity, at least in this one alternative model, which is good! However the alternative model doesn't get much attention in the paper. Is there a reason for this?

We would like to clarify that we did not model pulse-echo. In our confusion model, all echoes received within the IPI are attributed to the bat’s most recent call. This includes echoes that may in fact originate from conspecific calls, but the model does not assign self-echoes to earlier pulses or span multiple IPIs. Therefore, while the model captures echo confusion, it does not include true pulse-echo ambiguity. We have clarified this point in the revised text in lines 551-553.

(41) Line 41: "continuous" is more appropriate than "constant".

Thank you, we have rephrased the text accordingly.

(42) Line 69: "band width" should be one word.

Thank you, we have corrected it to “bandwidth”.

(43) Line 79: "bats" should be in the possessive.

Thank you, the text has been rephrased.

(44) Line 128: "convoluted" don't you mean "convolved"?

We have replaced “convoluted” with the correct term “convolved” in the revised text.

(45) Please check your references, as there are some incomplete citations and typos.

Thank you, we have reviewed and corrected all references for completeness and consistency.

References

Beetz, M.J. and Hechavarría, J.C. (2022) ‘Neural Processing of Naturalistic Echolocation Signals in Bats’, Frontiers in Neural Circuits, 16, p. 899370. Available at: https://doi.org/10.3389/FNCIR.2022.899370/BIBTEX.

Beleyur, T. and Goerlitz, H.R. (2019) ‘Modeling active sensing reveals echo detection even in large groups of bats’, Proceedings of the National Academy of Sciences of the United States of America, 116(52), pp. 26662–26668. Available at: https://doi.org/10.1073/pnas.1821722116.

Betke, M. et al. (2008) ‘Thermal Imaging Reveals Significantly Smaller Brazilian Free-Tailed Bat Colonies Than Previously Estimated’, Journal of Mammalogy, 89(1), pp. 18–24. Available at: https://doi.org/10.1644/07-MAMM-A-011.1.

Blauert, J. (1997) ‘Spatial Hearing: The Psychophysics of Human Sound Localization (rev. ed.)’.

Boerma, D.B. et al. (2019) ‘Wings as inertial appendages: How bats recover from aerial stumbles’, Journal of Experimental Biology, 222(20). Available at: https://doi.org/10.1242/JEB.204255/VIDEO-3.

Boonman, A. et al. (2013) ‘It’s not black or white-on the range of vision and echolocation in echolocating bats’, Frontiers in Physiology, 4 SEP(September), pp. 1–12. Available at: https://doi.org/10.3389/fphys.2013.00248.

Boonman, A.M., Parsons, S. and Jones, G. (2003) ‘The influence of flight speed on the ranging performance of bats using frequency modulated echolocation pulses’, The Journal of the Acoustical Society of America, 113(1), p. 617. Available at: https://doi.org/10.1121/1.1528175.

Burnett, S.C. and Masters, W.M. (2002) ‘Identifying Bats Using Computerized Analysis and Artificial Neural Networks’, North American Symposium on Bat Research, 9.

Chili, C., Xian, W. and Moss, C.F. (2009) ‘Adaptive echolocation behavior in bats for the analysis of auditory scenes’, Journal of Experimental Biology, 212(9), pp. 1392–1404. Available at: https://doi.org/10.1242/jeb.027045.

Fujioka, E. et al. (2021) ‘Three-Dimensional Trajectory Construction and Observation of Group Behavior of Wild Bats During Cave Emergence’, Journal of Robotics and Mechatronics, 33(3), pp. 556–563. Available at: https://doi.org/10.20965/jrm.2021.p0556.

Gillam, E.H. et al. (2010) ‘Echolocation behavior of Brazilian free-tailed bats during dense emergence flights’, Journal of Mammalogy, 91(4), pp. 967–975. Available at: https://doi.org/10.1644/09-MAMM-A-302.1.

Goldshtein, A. et al. (2025) ‘Onboard recordings reveal how bats maneuver under severe acoustic interference’, Proceedings of the National Academy of Sciences, 122(14), p. e2407810122. Available at: https://doi.org/10.1073/PNAS.2407810122.

Griffin, D.R., Webster, F.A. and Michael, C.R. (1958) ‘THE ECHOLOCATION OF FLYING INSECTS BY BATS ANIMAL BEHAVIOUR , Viii , 3-4’.

Hagino, T. et al. (2007) ‘Adaptive SONAR sounds by echolocating bats’, International Symposium on Underwater Technology, UT 2007 - International Workshop on Scientific Use of Submarine Cables and Related Technologies 2007, pp. 647–651. Available at: https://doi.org/10.1109/UT.2007.370829.

Hiryu, S. et al. (2008) ‘Adaptive echolocation sounds of insectivorous bats, Pipistrellus abramus, during foraging flights in the field’, The Journal of the Acoustical Society of America, 124(2), pp. EL51–EL56. Available at: https://doi.org/10.1121/1.2947629.

Jakobsen, L. et al. (2024) ‘Velocity as an overlooked driver in the echolocation behavior of aerial hawking vespertilionid bats’. Available at: https://doi.org/10.1016/j.cub.2024.12.042. Jakobsen, L., Brinkløv, S. and Surlykke, A. (2013) ‘Intensity and directionality of bat echolocation signals’, Frontiers in Physiology, 4 APR(April), pp. 1–9. Available at: https://doi.org/10.3389/fphys.2013.00089.

Jakobsen, L. and Surlykke, A. (2010) ‘Vespertilionid bats control the width of their biosonar sound beam dynamically during prey pursuit’, 107(31). Available at:

https://doi.org/10.1073/pnas.1006630107.

Kalko, E.K. V. (1995) ‘Insect pursuit, prey capture and echolocation in pipistrelle bats (Microchirptera)’, Animal Behaviour, 50(4), pp. 861–880.

Kazial, K.A., Burnett, S.C. and Masters, W.M. (2001) ‘ Individual and Group Variation in Echolocation Calls of Big Brown Bats, Eptesicus Fuscus (Chiroptera: Vespertilionidae) ’, Journal of Mammalogy, 82(2), pp. 339–351. Available at: https://doi.org/10.1644/15451542(2001)082<0339:iagvie>2.0.co;2.

Kazial, K.A., Kenny, T.L. and Burnett, S.C. (2008) ‘Little brown bats (Myotis lucifugus) recognize individual identity of conspecifics using sonar calls’, Ethology, 114(5), pp. 469– 478. Available at: https://doi.org/10.1111/j.1439-0310.2008.01483.x.

Kick, S.A. (1982) ‘Target-detection by the echolocating bat, Eptesicus fuscus’, Journal of Comparative Physiology □ A, 145(4), pp. 431–435. Available at: https://doi.org/10.1007/BF00612808/METRICS.

Kothari, N.B. et al. (2014) ‘Timing matters: Sonar call groups facilitate target localization in bats’, Frontiers in Physiology, 5 MAY. Available at: https://doi.org/10.3389/fphys.2014.00168.

Mohl, B. and Surlykke, A. (1989) ‘Detection of sonar signals in the presence of pulses of masking noise by the echolocating bat , Eptesicus fuscus’, pp. 119–124.

Moss, C.F. and Surlykke, A. (2010) ‘Probing the natural scene by echolocation in bats’, Frontiers in Behavioral Neuroscience. Available at: https://doi.org/10.3389/fnbeh.2010.00033.

Neretti, N. et al. (2003) ‘Time-frequency model for echo-delay resolution in wideband biosonar’, The Journal of the Acoustical Society of America, 113(4), pp. 2137–2145. Available at: https://doi.org/10.1121/1.1554693.

Popper, A.N. and Fay, R.R. (1995) Hearing by Bats. Springer-Verlag.

Roy, S. et al. (2019) ‘Extracting interactions between flying bat pairs using model-free methods’, Entropy, 21(1). Available at: https://doi.org/10.3390/e21010042.

Sabol, B.M. and Hudson, M.K. (1995) ‘Technique using thermal infrared-imaging for estimating populations of gray bats’, Journal of Mammalogy, 76(4). Available at: https://doi.org/10.2307/1382618.

Saillant, P.A. et al. (1993) ‘A computational model of echo processing and acoustic imaging in frequency- modulated echolocating bats: The spectrogram correlation and transformation receiver’, The Journal of the Acoustical Society of America, 94(5). Available at: https://doi.org/10.1121/1.407353.

Salles, A., Diebold, C.A. and Moss, C.F. (2020) ‘Echolocating bats accumulate information from acoustic snapshots to predict auditory object motion’, Proceedings of the National Academy of Sciences of the United States of America, 117(46), pp. 29229–29238. Available at: https://doi.org/10.1073/PNAS.2011719117/SUPPL_FILE/PNAS.2011719117.SAPP.PDF.

Sanderson, M.I. et al. (2003) ‘Evaluation of an auditory model for echo delay accuracy in wideband biosonar’, The Journal of the Acoustical Society of America, 114(3), pp. 1648– 1659. Available at: https://doi.org/10.1121/1.1598195.

Schnitzler, H., Bioscience, E.K.- and 2001, undefined (no date) ‘Echolocation by insecteating bats: we define four distinct functional groups of bats and find differences in signal structure that correlate with the typical echolocation ’, academic.oup.comHU Schnitzler, EKV KalkoBioscience, 2001•academic.oup.com [Preprint]. Available at: https://academic.oup.com/bioscience/article-abstract/51/7/557/268230 (Accessed: 17 March 2025).

Schnitzler, H.-U. et al. (1987) ‘The echolocation and hunting behavior of the bat,Pipistrellus kuhli’, Journal of Comparative Physiology A, 161(2), pp. 267–274. Available at: https://doi.org/10.1007/BF00615246.

Simmons, J.A. et al. (1983) ‘Acuity of horizontal angle discrimination by the echolocating bat , Eptesicus fuscus’. Simmons, J.A. and Kick, S.A. (1983) ‘Interception of Flying Insects by Bats’, Neuroethology and Behavioral Physiology, pp. 267–279. Available at: https://doi.org/10.1007/978-3-64269271-0_20.

Surlykke, A., Ghose, K. and Moss, C.F. (2009) ‘Acoustic scanning of natural scenes by echolocation in the big brown bat, Eptesicus fuscus’, Journal of Experimental Biology, 212(7), pp. 1011–1020. Available at: https://doi.org/10.1242/JEB.024620.

Theriault, D.H. et al. (no date) ‘Reconstruction and analysis of 3D trajectories of Brazilian free-tailed bats in flight’, cs-web.bu.edu [Preprint]. Available at: https://csweb.bu.edu/faculty/betke/papers/2010-027-3d-bat-trajectories.pdf (Accessed: 4 May 2023).

Ulanovsky, N. and Moss, C.F. (2008) ‘What the bat’s voice tells the bat’s brain’, Proceedings of the National Academy of Sciences of the United States of America, 105(25), pp. 8491– 8498. Available at: https://doi.org/10.1073/pnas.0703550105. Vanderelst, D. and Peremans, H. (2018) ‘Modeling bat prey capture in echolocating bats : The feasibility of reactive pursuit’, Journal of theoretical biology, 456, pp. 305–314.

Yovel, Y. et al. (2009) ‘The voice of bats: How greater mouse-eared bats recognize individuals based on their echolocation calls’, PLoS Computational Biology, 5(6). Available at: https://doi.org/10.1371/journal.pcbi.1000400.

Yovel, Y. and Ulanovsky, N. (2017) ‘Bat Navigation’, The Curated Reference Collection in Neuroscience and Biobehavioral Psychology, pp. 333–345. Available at: https://doi.org/10.1016/B978-0-12-809324-5.21031-6.

Author Response:

We thank the reviewers and editors for their thoughtful and constructive feedback on our manuscript, “Morphology and ultrastructure of pharyngeal sense organs of Drosophila larvae.” We are pleased that both reviewers found our ultrastructural analysis and 3D reconstructions of the larval pharyngeal sensory system to be of high quality, and we appreciate the recognition of the study’s significance and potential impact on the Drosophila neurobiology field.

We want to address the concern raised regarding the limited referencing and comparison with previous work on pharyngeal sensory organs, particularly in adult Drosophila and other insect species.

As noted by the reviewers, our manuscript is concise and focused. We want to clarify that we initially prepared and submitted this study with the intention of it being considered as a Short Report, which comes with limitations on the number of characters and figures that can be included. During the submission process, we were asked by the editors if we would like to submit our work as a full-length Research Advance, which we agreed to.

That said, we are now happy to expand the discussion in the broader context of related studies — including prior EM and anatomical work — which would enrich the manuscript and provide readers with a deeper comparative perspective.

We are grateful for the positive assessment of our manuscript and for the opportunity to clarify this point.

Sincerely,

Vincent Richter and Andreas S. Thum

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public review)

Summary:

This manuscript addresses the question of whether spontaneous activity contributes to the clustering of retinogeniculate synapses before eye opening. The authors re-analyze a previously published dataset to answer the question. The authors conclude that synaptic clustering is eye-specific and activity dependent during the first postnatal week. While there is useful information in this manuscript, I don't see how the data meaningfully supports the claims made about clustering.

In adult retinogeniculate connections, functionally specificity is supported by select pairings of retinal ganglion cells and thalamocortical cells forming dozens of synaptic connections in subcellular microcircuits called glomeruli. In this manuscript, the authors measure whether the frequency of nearby synapses is higher in the observed data than in a model where synapses are randomly distributed throughout the volume. Any real anatomical data will deviate from such a model. The interesting biological question is not whether a developmental state deviates from random. The interesting question is how much of the adult clustering occurs before eye opening. In trying to decode the analysis in this manuscript, I can't tell if the answer is 99% or 0.001%.

We thank the reviewer for their helpful critique through both rounds of review. We have refocused the manuscript on paired eye-specific measurements of active zone addition and spatial relationships among active zones at each age. All effect sizes and power values for each comparison are now reported in Table S2. These measures allow readers to gauge biological significance more transparently.

Strengths:

The source dataset is high resolution data showing the colocalization of multiple synaptic proteins across development. Added to this data is labeling that distinguishes axons from the right eye from axons from the left eye. The first order analysis of this data showing changes in synapse density and in the occurrence of multi-active zone synapses is useful information about the development of an important model system.

Weaknesses:

I don't think the analysis of clustering within this dataset improves our understanding of how the system works. It is possible that the result is clear to the authors based on looking at the images. As a reader trying to interpret the analysis, I ran into the following problems:

• It is not possible to estimate biologically meaningful effect sizes from the data provided. Spontaneous activity in the post natal week could be responsible for 99% or 0.001% of RGC synapse clustering.

• The sample size is too small for the kinds of comparisons being made. The authors point out that many STORM studies use an n of 1 while the authors have n = 3 for each of their six experimental groups. However, the critical bit is what kinds of questions you are trying to answer with a given sample size. This study depends on determining whether the differences between groups are due to age, genotype, or individual variation. This study also makes multiple comparisons of many different noisy parameters that test the same or similar hypothesis. In this context, it is unlikely that n = 3 sufficiently controls for individual variation.

We have revised the manuscript to focus on eye-specific differences, which are paired measurements collected at each age. We have measured effect sizes and performed power tests for all comparisons presented in the manuscript. These measurements are shown for every figure in a new supplemental table S2.

• There is no clear biological interpretation of the core measure of the publication, the normalized clustering index. The normalized clustering index starts with counting the fraction of single active zone synapses within various distances to the edge of synapses. This frequency is compared to a randomization model in which the positions of synapses are randomized throughout a volume. The authors found that the biggest deviation between the observed and randomized proximity frequency using a distance threshold of 1.5 um. They consider the deviation from the random model to be a sign of clustering. However, two RGC synapses 1.5 um apart have a good chance of coming from the same RGC axon. At this scale, real observations will, therefore, always look more clustered than a model where synapses are randomly placed in a volume. If you randomly place synapses on an axon, they will be much closer together than if you randomly place synapses within a volume. The authors normalize their clustering measure by dividing by the frequency of clustering in the normalized model. That makes the measure of clustering an ambiguous mix of synapse clustering, axon morphology, and synaptic density.

We have removed the “normalized clustering index”. “Clustered” inputs are now defined strictly as those that have a neighboring single active-zone (sAZ) synapse within 1.5 mm. For each type of input (sAZ and mAZ) we show 1) the ratio of clustered to isolated inputs for both eyes, and 2) the number of neighboring sAZs (Figure 4).

We agree with the reviewer that many synapses are likely made nearby along the same axon from an individual RGC. In this scenario, sAZ synapses that are nearby a neighboring mAZ input may be part of the same nascent bouton. And, sAZ synapses nearby other sAZ neighbors may ultimately mature into a mAZ input. At the same time, inputs from one RGC may form nearby other inputs from neighboring RGCs. We discuss these motifs and potential mechanisms of cell-autonomous and non-autonomous development (Lines 300-308).

• Other measures are also very derived. For instance, one argument is based on determining that the cumulative distribution of the distance of dominant-eye multi-active zone synapses with nearby single-active zone synapses from dominant-eye multi-active zone synapses is statistically different from the cumulative distribution of the distance of dominant-eye multi-active zones without nearby single-active zone synapses from dominant-eye multi-active zones. Multiple permutations of this measure are compared.

We have simplified the presentation to show all measured path lengths for every input. This allows the reader to see each of the inputs and their relative distances. We present these data for like-eye type interactions at P4 and P8 (Figures 5 and S5).   

• There are major biological differences between groups that are difficult to control for. Between P2, P4, and P8, there are changes in cell morphology and synaptic density. There are also large differences in synapse density between wild type and KO mice. It is difficult to be confident that these differences are not responsible for the relatively subtle changes in clustering indices.

• Many claims are based on complicated comparisons between groups rather than the predominating effects within the data. It is noted that: "In KO mice, dominant eye projections showed increased clustering around mAZ synapses compared to sAC synapses suggesting partial maintenance of synaptic clustering despite retinal wave defects". In contrast, I did not notice any discussion of the fact that the most striking trend in those measures is that the clustering index decreases from P2 to P8.

Related to the points above, we have revised the manuscript to focus on eye-specific release site addition and spatial relationships. For clarity, we have removed the clustering index and instead present ratios of clustered and isolated inputs, the number of sAZ synapses near each input type, and distance between like-eye mAZ inputs (Figure 4).      

• Statistics are improperly applied. In my first review I tried to push the authors to calculate confidence intervals for two reasons. First, I believed the reader should be able to answer questions such as whether 99% or 0.01% of RGC synaptic clustering occurred in the first postnatal week. Second, I wanted the authors to deal with the fact that n=3 is underpowered for many of the questions they were asking. While many confidence intervals can now be found leading up to a claim, it is difficult to find claims that are directly supported by the correct confidence interval. Many claims are still incorrectly based on which combinations of comparisons produced statistically significant differences and which combinations did not.

We have substantially revised the manuscript to focus on within-group paired effects between eye-of-origin. We performed power tests for all statistical presentations and effect sizes and powers are presented for every figure in a new supplemental table S2. To simplify the manuscript and make it easier to read, we report confidence interval measurements in a separate supplemental table S3.

Reviewer #2 (Public review):

Summary:

This study provides a valuable data set showing changes in the spatial organization of synaptic proteins at the retinogeniculate connection during a developmental period of active axonal and synaptic remodeling. The data collected by STORM microscopy is state-of-the-art in terms of the high-resolution view of the presynaptic components of a plastic synapse. The revision has addressed many, but not all, of the initial concerns about the authors interpretation of their data. However, with the revisions, the manuscript has become very dense and difficult to follow.

We greatly appreciate the reviewer’s thoughtful comments through two rounds of review. To improve the clarity of the manuscript, we have substantially revised the work to streamline the narrative, clearly define terminology, and simplify data presentations, allowing readers to more directly interpret results and their implications.

Strengths:

The data presented is of good quality and provides an unprecedented view at high resolution of the presynaptic components of the retinogeniculate synapse during active developmental remodeling. This approach offers an advance to the previous mouse EM studies of this synapse because the CTB label allows identification of the eye from which the presynaptic terminal arises.

Weaknesses:

From these data the authors conclude that eye-specific increase in mAZ synapse density occur over retinogeniculate refinement, that sAZ synapses cluster close to mAZ synapses over age, and that this process depends on spontaneous activity and proximity to eye-specific mAZ synapses. While the interpretation of this data set is much more grounded in this revised submission, some of the authors' conclusions/statements still lack convincing supporting evidence.

This includes:

(1) The conclusion that multi-active zone synapses are loci for synaptic clustering. This statement, or similar ones (e.g., line 407) suggest that mAZ synapses actively or through some indirect way influence the clustering of sAZ synapses. There is no evidence for this. Clustering of retinal synapses are in part due to the fact that retinal inputs synapse on the proximal dendrites. With increased synaptogenesis, there will be increased density of retinal terminals that are closely localized. And with development, perhaps sAZ synapses mature into mAZ synapses. This scenario could also explain a large part of this data set.

We thank the reviewer for their comment. We have removed the ambiguous phrasing and clarified the manuscript to explicitly discuss alternative interpretations consistent with the results (Lines 300-308). This includes a discussion of sAZ synapse maturation into mAZ inputs (Lines 294-296).

(2) The conclusion that, "clustering depends on spontaneous retinal activity" could be misleading to the reader given that the authors acknowledge that their data is most consistent with a failure of synaptogenesis in the mutant mice (in the rebuttal). Additionally clustering does occur in CTB+ projections around mAZ synapses.

We have removed the highlighted phrase and revised the manuscript to focus on differences in release site addition between eye-of-origin. We clarified our discussion of activity-dependent changes to state that synapses fail to form in the mutant and synaptic clustering was reduced (Lines 324-330).

(3) Line 403: "Since mAZ synapses are expected to have a higher release probability, they likely play an important role in driving plasticity mechanisms reliant on neurotransmission.":What evidence do the authors have that mAZ are expected to have higher release probability?

We thank the reviewer for their careful reading. Because they have several active zones, mAZ synapses are expected to have a higher number of release sites (N), which could be independent of release probability at any individual active zone (Pr). We have removed the reference to release probability. Instead, we maintain focus on active zone number.

Reviewer #3 (Public review):

This study is a follow-up to a recent study of synaptic development based on a powerful data set that combines anterograde labeling, immunofluorescence labeling of synaptic proteins, and STORM imaging (Cell Reports, 2023). Specifically, they use anti-Vglut2 label to determine the size of the presynaptic structure (which they describe as the vesicle pool size), anti-Bassoon to label active zones with the resolution to count them, and anti-Homer to identify postsynaptic densities. Their previous study compared the detailed synaptic structure across the development of synapses made with contra-projecting vs. ipsi-projecting RGCs and compared this developmental profile with a mouse model with reduced retinal waves. In this study, they produce a new detailed analysis on the same data set in which they classify synapses into "multi-active zone" vs. "single-active zone" synapses and assess the number and spacing of these synapses. The authors use measurements to make conclusions about the role of retinal waves in the generation of same-eye synaptic clusters, providing key insight into how neural activity drives synapse maturation.

Strengths:

This is a fantastic data set for describing the structural details of synapse development in a part of the brain undergoing activity-dependent synaptic rearrangements. The fact that they can differentiate eye of origin is what makes this data set unique over previous structural work. The addition of example images from EM data set provides confidence in their categorization scheme.

Weaknesses:

Though the descriptions of synaptic clusters are important and represent a significant advance, the authors conclusions regarding the biological processes driving these clusters are not testable by such a small sample. This limitation is expected given the massive effort that goes into generating this data set. Of course the authors are free to speculate, but many of the conclusions of the paper are not statistically supported.

We thank the reviewer for their helpful comments throughout the revision process. We have substantially modified the manuscript to reframe the work around release site addition during eye-specific competition. Power tests and effect size measurements are presented for every figure in a new supplemental table S2.

Reviewer #2 (Recommendations for the authors):

(1) Authors should discuss that it is not clear what the relationship is between sAZ and mAZ, and sAZ could turn into a mAZ. This is not unreasonable that the number of AZ/bouton increases with development given that in the adult rodent retinogeniculate bouton, there is an average of 27 active zones (Budisantoso et al, 2012).

We thank the reviewer for their helpful suggestion. We have added a discussion of the relationship between sAZ and mAZ inputs and the point that sAZ synapses may mature into mAZ synapses (Lines 294-296). We now reference the work of Budisantoso et al., J. Neurosci. 2012.   

(2) The authors should clarify how the statistics are calculated for the normalized clustering index (figure 3B, C). For ratios of values each with variance, the variance is summed when calculating SEM.

For clarity, we have removed the normalized clustering index analysis. We have simplified the work to present a clear definition of clustered and unclustered inputs, where clustering is defined by the presence of a nearby neighboring synapse within 1.5mm. We present the ratio of clustered and unclustered inputs for each input type and eye-of-origin. We also show the number of sAZ synapses nearby each clustered input (Figure 4).

(3) The authors have significantly clarified the terminology that they use in the text. This is much appreciated. However, it would be helpful to the naïve reader if they could define their use of the word "synapse" as referring to individual active zones/release sites or to terminals/boutons. For example:

Line 378: "Prior electron microscopy studies in the mouse found limited evidence of convergent synaptic clustering from neighboring RGCs at postnatal day 8 (10, 13), suggesting that the mAZ synapses seen in STORM images are single retinogeniculate terminals. The lack of synaptic convergence in prior EM reconstructions at P8 implies that early clustering around mAZ synapses may result from local output clustering within individual RGC arbors.":

What do the authors mean by "convergent synaptic clustering": do they mean clustering of release sites from different RGC inputs? And what does "local output clustering" mean?

We thank the reviewer for their suggestion to use clear terminology. We have revised the manuscript to define our use of the term “synapse” as a single active zone/release site (Lines 134-136). We refer to mAZ boutons in STORM data as “inputs”. We have revised the discussion of prior EM studies (Lines 130-132) and clarified all discussions of synaptic clustering throughout the work.

(4) While the authors argue that the retina-specific β2-nAChR mice exhibit disrupted retinal waves and defects in eye specific segregation, the authors are studying issues of active zone density which may depend on mechanisms depending on the postsynaptic neuron. This should be acknowledged.

We have updated the text to discuss the fact that postsynaptic mechanisms are also critical for the refinement of eye-specific synapses (Lines 332-340). We have added several additional references to the manuscript accordingly.

Reviewer #3 (Recommendations for the authors):

The authors have addressed many of my original concerns. The additional description of criteria for categorizing synapses, showing all the data points, gives the reader a stronger sense of where the numbers in the quantification come from. Replacing the "complex/simple" distinction with the "multi/single active zone" and the other clarifying text was effective. The addition of the EM data was also a very nice example to help interpret STORM images. It does appear there was no quantification on this EM data set and perhaps just a few example images were taken as "proof of principle". If, by chance, the authors have more EM images to make a data set of them that allows for some quantification, that would be great to add.

We thank the reviewer for their helpful comments on the manuscript through both rounds of review. The EM data we collected were 2D images of a subset of physical sections at postnatal day 8. Most dAPEX2(+) profiles had a single active zone, but a definitive identification would require 3D imaging so that each terminal can be assessed in its entirety for release sites that might be missed in a single cross section. Similarly, multi-active zone boutons are positively identified in 2D images, but definitive measurements of AZ number would require 3D information. We analyzed our 2D EM images and present a plot of dAPEX2(+) profile size versus active zone number below. These measures are positively correlated (r = 0.74), with larger profiles containing more active zones.

Author response image 1.<br />

Unfortunately, we are not currently equipped to perform volumetric EM imaging at our home institution and are concerned that analysis of 2D data may be inconclusive. For these reasons, we are opting to maintain a qualitative presentation of our current EM results and we look forward to collaborating with other experts to achieve volumetric EM reconstructions in the future

Author response:

We thank the editors and the reviewers for their constructive comments, which have greatly helped us identify key areas to strengthen the manuscript. We acknowledge the validity of the major points raised, and we plan the following revisions:

Criticality

As suggested by Reviewer #1, we will carefully examine whether the dynamics we observe are indeed poised near criticality. We will perform additional analyses to assess how structural and dynamic features change when parameters are tuned away from the coil–globule transition, and we will revise the title and text to ensure that our claims are appropriately moderated.

Role of the homie element

We agree with Reviewer #2 that the presence of homie elements introduces major modifications to chromosome structure and dynamics. We initially considered that this factor might even explain the paradox described in Gregor’s work. In the first phase of our study, we carried out simulations including homie elements and found that the potential confounding effects are largely resolved if we restrict the analysis to trajectories prior to encounters between the two homie copies. We will include these simulations and expand the discussion accordingly in the revised version.

Comparison to Hi-C data

Both reviewers noted a visual discrepancy between experimental and simulated Hi-C maps. We will address this by testing alternative similarity measures (e.g., Pearson correlation, as suggested) and by exploring parameter ranges that may improve the agreement.<br /> Together, these modifications will strengthen the manuscript, clarify the scope of our conclusions, and directly address the reviewers’ central concerns.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public Review):

Summary

In this investigation Kapustin et al. demonstrate that vascular smooth muscle cells (VSMCs) exposed to the extracellular matrix fibronectin stimulates the release of small extracellular vesicles (sEVs). The authors provide experimental evidence that stimulation of the actin cytoskeleton boosts sEV secretion and posit that sEVs harbor both fibronectin and collagen IV protein themselves which also, in turn, alter cell migration parameters. It is well established that fibronectin is associated with increased cell migration and adherence; therefore, this association with VSMCs is not novel.

The reviewer is correct that FN has been associated with migration and adherence in previous studies.  However we have extended these observations to show that the extracellular fibronectin matrix stimulates small extracellular vesicle (sEVs) secretion by modulating the actin cytoskeleton. We also showed that sEVs are trapped in the extracellular matrix and that by presenting collagen VI induce early focal adhesion formation, reduce excessive cellular spreading and guide cell invasion directionality though a 3D matrix. Hence, sEVs mediate cell-matrix cross talk and change cell behaviour in the context of fibronectin matrix. This is critically important for vasculature where regulated VSMC invasion is essential for repair with its deregulation leading to pathology.

The authors purport that sEV are largely born of filopodia origin; however, this data is not well executed and seems generally at odds with the presented data.

Our experimental data showed that CD63 MVs are associated with filopodia in fixed and live cells (Fig 2E, 2F and Video S1) and that inhibition of filopodia formation using the formin inhibitor, SMIFH2 reduced sEV secretion on FN (Fig 2B). However, we agree with the reviewer that further studies are required to connect sEV secretion to filopodia.  To address this we have provided further data analysis but also toned down our conclusions regarding this point: . Changes include:

(1) Title: Matrix-associated extracellular vesicles modulate smooth muscle cell adhesion and directionality by presenting collagen VI.

(2) Results, section title: 2. FN-induced sEV secretion is modulated by Arp2/3 and formin-dependent actin cytoskeleton remodelling

(3) Results, page 6 Line 27-44 and conclusion page 7, Ln 3 “Interestingly, CD63+ MVBs can be observed in filopodia-like structures suggesting that sEV secretion can also occur spatially via cellular protrusion-like filopodia but more studies are needed to confirm this hypothesis.”

(4) Discussion, page 12, line 19. “Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells.”

Similarly, the effect of sEVs on parameters of cell migration has almost no magnitude of effect, making mechanism exploration somewhat nebulous.

VSMC are mesenchymal-type cells with a low migration rate and we agree that the changes in the motility are not of great magnitude even for the positive controls suggesting that this is a complex, multifactorial process for VSMCs. In our experiments we collected data from >5000 individual cells to measure the average speed and found that fibronectin matrix on its own increased VSMC speed from ~0.61 um/min to ~0.68 μm/min (~12% raise) which was statistically significant (Fig 5A). Addition of a sEV inhibitor caused a modest but significant decrease in cellular speed. Interestingly, addition of ECM-associated sEVs did not influence cell speed in 2D or 3D assays. However in a 3D model we observed a 22% change in cell directionality (Fig 5G) and  a 235% change in cell alignment index (FMI, Fig 5H) which we believe is very strong evidence that VSMC-derived sEVs are involved in a regulation of VSMC invasion directionality.  These data are also in agreement with sEV effects in tumour cells (Sung et al., 2015) though this previous study did not identify the factor driving the directionality and we think our Collagen VI data extends significantly these previous observations. 

Results, page 9: “Hence, ECM-associated sEVs have modest influence on VSMC speed but influence VSMC invasion directionality.”.  

Lastly, the proposed mechanism of VSMCs responding to, and depositing, ECM proteins via sEVs was not rigorously executed; again, making the conclusions challenging for the reader to interpret.

We appreciate the reviewer’s comment regarding the mechanistic aspects of VSMCs responding to and depositing ECM proteins via sEVs. In our revised manuscript, we have expanded the data demonstrating that sEVs can be retained within the extracellular matrix (see Figs 3A, 3B, S3A, S3B). Additionally, we show that collagen VI is present on the surface of sEVs, where it may modulate cell adhesion and influence the directionality of cell invasion (Fig 7E). Our results further indicate that both fibronectin (FN) and collagen VI can be recycled through multivesicular bodies (see Figs S3C, S3D, S3E–S3G). However, we acknowledge that the precise mechanisms governing the selective loading of ECM proteins onto sEVs, as well as the specific contributions of sEVs to overall ECM organization, remain to be fully elucidated and warrant further investigation. Based on our current evidence, we propose that collagen VI–loaded sEVs act primarily in a signaling capacity by modulating focal adhesion formation but are not directly involved in ECM structural remodeling.

Results, page 7: To quantify ECM-trapped sEVs we applied a modified protocol for the sequential extraction of extracellular proteins using salt buffer (0.5M NaCl) to release sEVs which are loosely-attached to ECM via ionic interactions, followed by 4M guanidine HCl buffer (GuHCl) treatment to solubilize strongly-bound sEVs (Fig S3A) [42]. We quantified total sEV and characterised the sEV tetraspanin profile in conditioned media, and the 0.5M NaCl and GuHCl fractions using ExoView. The total particle count showed that EVs are both loosely bound and strongly trapped within the ECM. sEV tetraspanin profiling showed differences between these 3 EV populations.  While there was close similarity between the conditioned media and the 0.5M NaCl fraction with high abundance of CD63+/CD81+ sEVs as well as CD63+/CD81+/CD9+ in both fractions (Fig S3A). In contrast, the GuHCl fraction was particularly enriched with CD63+ and CD63+/CD81+ sEVs with very low abundance of CD9+ EVs (Fig S3A). The abundance of CD63+/CD81+ sEVs was confirmed independently by a CD63+ bead capture assay in the media and loosely bound fractions (Fig S3B).

Results, page 7: We previously found that the serum protein prothrombin binds to the sEV surface both in the media and MVB lumen showing it is recycled in sEVs and catalyses thrombogenesis being on the sEV surface43. So we investigated whether FN can also be associated with sEV surface where it can be directly involved in sEV-cell cross-talk43.   We treated serum-deprived primary human aortic VSMCs with FN-Alexa568 and found that it was endocytosed and subsequently delivered to early and late endosomes together with fetuin A, another abundant serum protein that is a recycled sEV cargo and elevated in plaques (Figs S3C and S3D). CD63 visualisation with a different fluorophore (Alexa488) confirmed FN colocalization with CD63+ MVBs (Fig S3E). Next, we stained non-serum deprived VSMC cultured in normal growth media (RPMI supplemented with 20% FBS) with an anti-FN antibody and observed colocalization of CD63 and serum-derived FN.  Co-localisation was reducd likely due to competitive bulk protein uptake by non-deprived cells (Fig S3F). Notably, when we compared FN distribution in sparsely growing VSMCs versus confluent cells we found that FN intracellular spots, as well as colocalization with CD63, completely disappeared in the confluent state (Fig S3F and S3G). This correlated with nearly complete loss of CD63+/CD81+ sEV secretion by the confluent cells indicating that confluence abrogates intracellular FN trafficking as well as sEV secretion by VSMCs (Fig S3H). Finally, FN could be co-purified with sEVs from VSMC conditioned media (Fig S3I) and detected on the surface of sEVs by flow cytometry confirming its loading and secretion via sEVs (Fig 3C).

Results: page 10  Collagen VI was the most abundant protein in VSMC-derived sEVs (Fig 7B, Table S7) and  was previously implicated in the interaction with the proteoglycan NG2[53] and suppression of cell spreading on FN[54]. To confirm the presence of collagen VI in ECM-associated sEVs we analysed sEVs extracted from the 3D matrix using 0.5M NaCl treatment and showed that both collagen VI and FN are present (Fig 7D). Next, we analysed the distribution of collagen VI using dot-blot. Alix staining was bright only upon permeabilization of sEV indicating that it is preferentially a luminal protein (Fig 7E). On the contrary, CD63 staining was similar in both conditions showing that it is surface protein (Fig 7E). Interestingly, collagen VI staining revealed that 40% of the protein is located on the outside surface with 60% in the sEV lumen (Fig 7E). 

Discussion page 12. “In fact, we observed that an extensive secretion of sEVs effectively ceased protrusion activity; also VSMCs acquired a rounded morphology when “hovering” over the FN matrix decorated with sEVs (data not shown). Hence, it will be interesting in future studies to investigate whether sEVs can stimulate Rho activity by presenting adhesion modulators—particularly collagen VI—on their surface, thereby guiding cell directionality during invasion..”

Discussion, page 14 “In summary, cooperative activation of integrin signalling and F-actin cytoskeleton pathways results in the secretion of sEVs which associate with the ECM and play a signalling role by controling FA formation and cell-ECM crosstalk. Further studies are needed to test these mechanisms across various cell types and ECM matrices.     

Strengths

The authors provide a comprehensive battery of cytoskeletal experiments to test how fibronectin and sEVs impact both sEV release and vascular smooth muscle cell migratory activation.

We appreciate this comment reflecting our efforts to apply a range of orthogonal methods to show the role of the integrin/actin cytoskeleton in ECM-stimulated sEV secretion.

Weaknesses

Unfortunately, this article suffers from many weaknesses. First, the rigor of the experimental approach is low, which calls into question the merit of the conclusions. In this vein, there is a lack of proper controls or inclusion of experiments addressing alternative explanations for the phenotype or lack thereof.

We acknowledge this comment and agree that there was not sufficient evidence to conclude that sEV secretion occurs via filopodia despite the microscopy/inhibitory data so this claim has now been excluded from the study. However we believe that our experimental data does clearly show that FN stimulates the secretion of collagenVI-loaded sEVs which are trapped by the ECM and have the capacity to modulate VSMC adhesion and invasion directionality. To support this, we have now extended the dataset in the revised version:

(1) In addition to the use of inhibitors and live cell analysis we have added quantitative data confirming that a large proportion of CD63+ endosomes are associated with F-actin/cortactin tails and this colocalization is increased upon the inhibition of sEV secretion with 3-OMS (Fig  2D, Fig S2B).

(2) We developed a method to extract ECM-associated sEVs and quantified/characterized these using ExoView Assays further confirming significant sEV entrapment by the ECM (Figs 3B, S3A, S3B).    

(3) We extended the controls to confirm FN delivery to CD63+ endosomes and showed that FN recycling is stopped upon reaching cell confluence (Figs S3F, S3G and Fig S3H).

(4) We included more intensive characterisation of human atherosclerotic plaque morphology (H&E, Masson’s trichrome staining, Orcein, elastin fibers staining) to confirm predominant accumulation of sEV in the neointima (Figs S4A, S4B and S4C). We also excluded an endothelial origin for the  CD81+ sEVs (Fig 4G).

(5) We included individual cellular tracks to the 2D migration analysis to confirm the statistical significance and concluded that ECM-associated sEVs regulate cell invasion directionality but not the cell speed (Figs 5A and 5B).

(6) We showed surface localisation of collagen VI on sEVs confirming that it can activate signalling pathways leading to early FA formation on the FN matrix  (Figs 7D and 7E).

(7) We included alternative explanations for some of our data in the discussion.      

Reviewer #2 (Public Review):

Extracellular vesicles have recently gained significant attention across a wide variety of fields, and they have therefore been implicated in numerous physiological and pathophysiological processes. When such a discovery and an explosion of interest occur in science, there is often much excitement and hope for answers to mechanisms that have remained elusive and poorly understood. Unfortunately, there is an equal amount of hype and overstatement that may also be put forth in the name of "impact", but this temptation must be avoided so that scientists and the broader public are not misled by overreaching interpretations and statements that lack rigorous and fully convincing evidence.

Thank you for your comment and we agree that investigating sEVs is particularly challenging due to the their heterogeneity and nano-size, as well as complex biogenesis mechanisms. ECM-associated sEVs is a very new direction for the EV field but one that is particularly relevant to the vasculature where cells must invade through a thick ECM and where the accumulation of ECM-bound EVs is a unique and documented phenomenon.  To further strengthen out conclusions we have included new data to support our statements but also excluded statements re: filopodia as the origin of sEVs, that are out of scope of our study and need to be investigated further.

The study presented by Kapustin et al. is certainly intriguing and timely, and it offers an interesting working hypothesis for the fields of extracellular vesicles and vascular biology to consider. The authors do a reasonable job at detecting these small extracellular vesicles, though some aspects of data presentation are missing such as full Western blots with accompanying size markers for the viewer to more fully appreciate that data and comparisons being made (see Figures 1 and 7).

We agree with the reviewer and have now included molecular weight markers (Fig 1F, 7C, 7D, S3I, S4E) and provided all original western blot scans (uncropped and unedited) to the eLife editor. 

Much of the imaging data from cell-based experiments is strong and conducted with many cutting-edge tools and approaches. That said, the static images and the dynamic imaging fall short of being fully convincing that the small extracellular vesicles found in the neighboring extracellular matrix are indeed being deposited there via the smooth muscle cell filopodia. Many of the lines of evidence presented suggest that this could occur, but alternative hypotheses also exist that were not fully ruled out, such as the ECM-deposited vesicles were secreted more from the soma and/or the lamellipodia that are also emitted and retracted from the cells. In particular, the authors show very nice dynamic imaging (Supplementary Figure S2A and Supplemental Video S1) that is interpreted as "extracellular vesicles being released from the cell" and these are seen as "bursts" of fluorescent signal; however, none of these appear to occur in filopodia as they appear within the cell proper (a "burst" of signal vs. a more intense "streak" of signal), which would be a stronger and more consistent observation predicted by the working model proposed by the authors.

Our live and fixed cell microscope data as well as inhibitor analysis showed that sEV secretion can be associated with the filopodia. However we agree with the reviewer that the data generated using pHluoron GFP marker clearly indicate that the majority of sEVs are secreted from the cell soma toward the ECM:

To reflect this, we have added further changes:

(1) Title: Matrix-associated extracellular vesicles modulate smooth muscle cell adhesion and directionality by presenting collagen VI.

(2) Results, section title: 2. FN-induced sEV secretion is modulated by Arp2/3 and formin-dependent actin cytoskeleton remodelling

(3)  Results, page 6 Line 27-36 “Formins and the Arp2/3 complex play a crucial role in the formation of filopodia, a cellular protrusion required for sensing the extracellular environment and cell-ECM interactions36. To test whether MVBs can be delivered to filopodia, we stained VSMCs for Myosin-10 (Myo10)37. We observed no difference between total filopodia number per cell on plastic or FN matrices (n=18±8 and n=14±3, respectively) however the presence of endogenous CD63+ MVBs along the Myo10-positive filopodia were observed in both conditions (Fig 2E, arrows). Filopodia have been implicated in sEV capture and delivery to endocytosis “hot-spots”38, so next we examined the directionality of CD63+ MVB movement in filopodia by overexpressing Myo10-GFP and CD63-RFP in live VSMCs. Importantly, we observed anterograde MVB transport toward the filopodia tip (Fig 2F and Supplementary Video S2) indicative of MVB secretion”.

(4) Results, page 6, Ln 37-44 “We also attempted to visualise sEV release in filopodia using CD63-pHluorin where fluorescence is only observed upon the fusion of MVBs with the plasma membrane39. Using total internal reflection fluorescence microscopy (TIRF) we observed the typical “burst”-like appearance of sEV secretion at the cell-ECM interface in full agreement with an earlier report showing MVB recruitment to invadopodia-like structures in tumor cells18 (Fig S2B and Supplementary Video S1). Although we also observed an intense CD63-pHluorin staining along filopodia-like structures we were not able to detect typical “burst”-like events to confirm sEV secretion in filopodia. (Fig S2C and Supplemental Video S1)”.

(5) Results, page 7 Ln 3 “Interestingly, CD63+ MVBs can be observed in filopodia-like structures suggesting that sEV secretion can also occur spatially via cellular protrusion-like filopodia but more studies are needed to confirm this hypothesis.”

(6) Discussion, page 12, line 19. “Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells.”

Imaging of related human samples is certainly a strength of the paper, and the authors are commended for attempting to connect the findings from their cell culture experiments to an important clinical scenario. However, the marker selected for marking extracellular vesicles is CD81, which has been described as present on the endothelium of atherosclerotic plaques with a proposed role in the recruitment of monocytes into diseased arteries (Rohlena et al. Cardiovasc Res 2009). More data should address this potentially confounding interpretation of the signals presented in images within Figure 4.

We thank the reviewer for this insightful comment that the  sEV marker CD81 can originate from endothelial cells in agreement with Rohlena et al., 2009.   To address this we investigated the spatial overlap between CD81 and the endothelial marker, CD31. We observed very strong CD81 staining in the intact endothelial cell (intima) layer and occasional CD31 positive cells in the neointima. Importantly, quantification of colocalization confirmed that 80% of CD81 in the neointima does not overlap with CD31 excluding an endothelial origin of these sEVs. (Fig 4G).  Moreover, we included complete morphological characterisation of the atherosclerotic plaques confirming that CD81 sEVs were primarily observed in the neointima where VSMCs constitute the cellular majority (Fig S4A, S4B, S4C and S4D).

On a conceptual level, the idea that the small extracellular vesicles contain Type VI Collagen, and this element of their cargo is modulating smooth muscle cell migration, is an intriguing aspect of the authors' working model. Nevertheless, the evidence supporting this potential mechanism does not quite fit together as presented. It is not entirely clear how the collagen VI within the vesicles is somehow accessed by the smooth muscle cell filopodia during migration. Are the vesicles lysed open once on the extracellular matrix? If so, what is the proposed mechanism for that to occur? If not, how are the adhesion molecules on the smooth muscle cell surface engaging the collagen VI fibers that are contained within the vesicles? This aspect of the model does not quite fit together with the proposed mechanism and may be an interesting speculative interpretation, warranting further investigation, but it should not be considered a strong conclusion with sufficient convincing data supporting this idea.

We thank the reviewer for their insightful comments regarding the mechanism by which collagen VI associated with sEVs could modulate smooth muscle cell adhesion and migration. To clarify, our new data suggest that collagen VI is predominantly present on the surface of the sEVs, as evidenced by Fig 7E. This surface localization strongly implies that collagen VI can be directly accessed by cell surface adhesion receptors, without the need for vesicle lysis or opening. While we cannot entirely rule out all alternative mechanisms, we consider vesicle rupture or lysis within the extracellular matrix to be a highly unlikely route for collagen VI exposure, given the known stability of sEVs under physiological conditions. We have added these points to clarify:

(1) Results, page 10, Ln 45 “To confirm the presence of collagen VI in ECM-associated sEVs we analysed sEVs extracted from the 3D matrix using 0.5M NaCl treatment and showed that both collagen VI and FN are present (Fig 7D). Next, we analysed the distribution of collagen VI using dot-blot. Alix staining was bright only upon permeabilization of sEV indicating that it is preferentially a luminal protein (Fig 7E). On the contrary, CD63 staining was similar in both conditions showing that it is surface protein (Fig 7E). Interestingly, collagen VI staining revealed that 40% of the protein is located on the outside surface with 60% in the sEV lumen (Fig 7E).”

(2) Discussion, page 13, Ln 2 “Hence, it will be interesting in future studies to investigate whether sEVs can stimulate Rho activity by presenting adhesion modulators—particularly collagen VI—on their surface, thereby guiding cell directionality during invasion..”

(3) Discussion, page 14, Ln 30: In addition to collagen VI the unique adhesion cluster in VSMC-derived sEVS also includes EGF-like repeat and discoidin I-like domain-containing protein (EDIL3), transforming growth factor-beta-induced protein ig-h3 (TGFBI) and the lectin galactoside-binding soluble 3 binding protein (LGALS3BP) and these proteins are also directly implicated in activation of integrin signalling and cellular invasiveness85-87. Although we found that collagen VI plays the key role in sEV-induced early formation of FAs in VSMCs, it is tempting to speculate that the high sEV efficacy in stimulating FA formation is driven by cooperative action of this unique adhesion complex on the sEVs surface and targeting this novel sEV-dependent mechanism of VSMC invasion may open-up new therapeutic opportunities to modulate atherosclerotic plaque development or even to prevent undesired VSMC motility in restenosis.    .   

(4) Abstract Figure

On a technical level, some of the statistical analysis is not readily understood from the data presented. It is very much appreciated that the authors show many of the graphs with technical and biological replicate values in addition to the means and standard deviations (though this is not clearly stated in all figure legends). However, in figures such as Figure 5, there are bars shown and indicated to be different by statistical comparison (see panel B in Figure 5). It is not clear how the values for Group 1 (no FN, no 3-OMS, no sEV) are statistically different (denoted by three asterisks but no p value provided in the legend) than Group 3 (no FN, 3-OMS added, no sEV), when their means and standard deviations appear almost identical. If this is an oversight, this needs to be corrected. If this is truly the outcome, further explanation is warranted. A higher level of transparency in such instances would certainly go a long way in helping address the current crisis of mistrust within the scientific community and at the interface with society at-large.

We thank the reviewer for their careful reading and important comments on the statistical analysis. We acknowledge that the technical and biological replicate data were not clearly reported in all figure legends and that the statistical approach for Figures 5A and 5B required clarification. In response, we have made several changes for greater transparency and rigor:

First, we have now explicitly included the numbers of biological replicates (N) and technical replicates (n) in all relevant figure legends for Figures 1–7. In addition, the number of individual cell tracks is now annotated for the migration/invasion analyses, along with the mean values for each dataset.

Upon review, we found that the original statistical analyses for Figures 5A and 5B were conducted using pooled averaged data. To address this, we have repeated the statistical tests using pooled individual cell track data, applying the Kruskal–Wallis test with Dunn’s multiple comparison correction. This more stringent approach revealed revised p-values, which are now indicated in Figures 5A and 5B.

With these corrections, we reconfirm our major findings: In the 2D model, fibronectin (FN) coating promotes VSMC velocity, while inhibition of sEV secretion with 3-OMS leads to reduced cell speed (Fig. 5A). Addition of sEVs to the ECM had no effect on VSMC speed at baseline but did rescue cell speed and distance in the presence of 3-OMS, consistent with EVs acting primarily on invasion directionality rather than speed in both 2D and 3D models (Fig. 5A, 5D). Furthermore, sEVs continue to significantly impact VSMC invasion directionality (Figs. 5G, 5H), in agreement with previous reports in tumor cells (Sung et al., 2015).

In summary, we have implemented the following revisions:

(1) Figures 5A and 5B: Individual cell track data are now shown, and statistical analyses have been repeated using the Kruskal–Wallis test with Dunn’s multiple comparisons.

(2) Figure legends and results sections: Numbers of biological and technical replicates, as well as individual data points, are now clearly stated.

Results, page 9, line 14: The text has been updated to clarify the statistical approach and major findings as described above.

We hope that these changes address the reviewer’s concerns and improve the transparency and reproducibility of our data presentation

Reviewer #1 (Recommendations For The Authors):

We are very thankful for the comprehensive review and comments which helped to improve our data.

Figure 1.<br /> The authors clearly show that FN stimulation (immobilized or cell-derived) promotes sEV secretion via canonical integrin pathways. FN is a promigratory substrate, hence its extensive use as a cell adhesion aid; thus one could assume that simply plating on FN induces a pro-migratory phenotype (later data supports this notion). Does the addition of growth factors also increase sEV release? An endogenous function of FN is siloing of various GFs during clot formation. Also, FAK and SRC networks intersect with canonical RTK signaling in terms of promoting Rac1, CDC42 and other migration mediators. The reason I believe this is important is because the data could be interpreted in two ways: 1) FN induces pro-migration signaling and then sEVs are released, or visa versa, FN induces sEV release and migration is initiated. GF supplementation in the absence of FN would clarify this relationship.

We thank the reviewer for this insightful comment regarding the possible role of growth factors (GFs) and the mechanistic relationship between FN stimulation, sEV secretion, and cell migration. We agree that FN is a well-established promoter of cell migration, and it is important to distinguish whether FN directly induces a pro-migratory phenotype or does so via sEV-mediated signaling.

Our data show that FN stimulation markedly increases VSMC motility, as reflected by enhanced cell speed (Fig. 5A), an increased number of focal adhesions (Fig. 6E), and facilitated centripetal movement of FAs (Fig. 6F). Interestingly, ECM-associated sEVs appear to play a complementary but distinct role: they do not significantly affect cell migration speed (Fig. 5A) but instead guide cell invasion directionality (Figs. 5G, 5H), reduce the number of FAs per cell (Fig. 6E), and promote early peripheral FA formation (Fig. 6F). In light of these findings, we have updated our graphical abstract to reflect the unique cross-talk mediated by sEVs between VSMCs and the ECM.

Regarding the influence of growth factors, we acknowledge that FN can bind and present different GFs, which could also contribute to changes in sEV secretion. Although our inhibition studies and integrin-blocking antibody results support a primary role for β1 integrin activation and actin assembly in triggering sEV secretion, we cannot entirely exclude the possibility that FN-bound growth factors play a role in this process. We have now incorporated this point into the discussion to address the reviewer’s suggestion.

Discussion, page 14 , Ln 7 “Although our small inhibitors and integrin modulating antibody data clearly indicate that β1 activation triggers sEV secretion via activation of actin assembly we cannot fully rule out that FN may also be modulating growth factor activity which in turn contributes to sEV secretion by VSMCs²³.  Excessive collagen and elastin matrix breakdown in atheroma has been tightly linked to acute coronary events hence it will be interesting to study the possible link between sEV secretion and plaque stability as sEV-dependent invasion is also likely to influence the necessary ECM degradation induced by invading cells⁹⁶

Figure 2.<br /> • The authors provide no evidence (or references) that SMIFH2 or CK666 halts filopodia extensions.

Thank you for this important note. We have included the corresponding references:

Results, page 5: “So next we tested the contribution of Arp2/3 and formins by using the small molecule inhibitors, CK666 and SMIFH2, respectively31, 32”.  

• Is there an increase in filopodia density when plated on FN vs plastic? Similarly, if there are more filopodia present is that associated with more sEV? Please provide evidence in this regard.

We agree that connecting the number of filopodia with the secretion level of sEVs may be an important clue if sEV secretion can be driven by FN-induced filopodia formation. However, Myosin10 staining to quantify filopodia (Fig 2E) showed no difference between VSMCs plated on plastic versus FN matrix. Therefore, we agree with the reviewer that the filopodia contribution to sEV secretion needs to be investigated further.  This idea is reflected in the following comments:

(1) Results, page 6, Ln 29 “We observed no difference between total filopodia number per cell on plastic or FN matrices (n=18±8 and n=14±3, respectively) however the presence of endogenous CD63+ MVBs along the Myo10-positive filopodia were observed in both conditions (Fig 2E, arrows).

(2) Results, page 6, Ln 37 “We also attempted to visualise sEV release in filopodia using CD63-pHluorin where fluorescence is only observed upon the fusion of MVBs with the plasma membrane39. Using total internal reflection fluorescence microscopy (TIRF) we observed the typical “burst”-like appearance of sEV secretion at the cell-ECM interface in full agreement with an earlier report showing MVB recruitment to invadopodia-like structures in tumor cells18 (Fig S2B and Supplementary Video S1). Although we also observed an intense CD63-pHluorin staining along filopodia-like structures we were not able to detect typical “burst”-like events to confirm sEV secretion in filopodia. (Fig S2C and Supplemental Video S1)..”

(3) Discussion, page 12, Ln 15 : “Focal complexes either disassemble or mature into the elongated centripetally located FAs48. In turn, these mature FAs anchor the ECM to actin stress fibres and the traction force generated by actomyosin-mediated contractility pulls the FAs rearward and the cell body forward12, 13. Here we report that β1 integrin activation triggers sEV release followed by sEV entrapment by the ECM. Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells..”

As hinted above, this data could be interpreted in the light of generally inhibiting cell migration to blunt sEV shedding. Does cell confluence affect sEV release? If cells are cultured to 100% confluency this would limit filopodia formation regardless of ECM type. If sEV secretion remains elevated on FN in this culture condition it would suggest a lack of dependency on filopodia.

We thank the reviewer for this thoughtful suggestion regarding the influence of cell confluence on sEV release and filopodia formation. To directly address this hypothesis, we performed additional experiments comparing VSMCs cultured at low and high confluency. As described in the revised Results (page 7, line 39), we found that high cellular confluency reduced FN recycling, as indicated by the marked decrease in intracellular FN-positive spots and loss of colocalization with CD63 (Figs S3F, S3G). Importantly, this was accompanied by a significant reduction in CD63+/CD81+ sEV secretion by confluent cells (Fig S3H). These results suggest that VSMC confluence, which suppresses filopodia formation, nearly abolishes both intracellular FN trafficking and sEV secretion, even in the presence of FN. Thus, under our experimental conditions, sEV secretion by VSMCs appears to be closely linked to dynamic cell–matrix interactions and is dramatically reduced when these processes are limited by confluence:

(1) Results, page 7, Ln 39 : “Notably, when we compared FN distribution in sparsely growing VSMCs versus confluent cells we found that FN intracellular spots, as well as colocalization with CD63, completely disappeared in the confluent state (Fig S3F and S3G). This correlated with nearly complete loss of CD63+/CD81+ sEV secretion by the confluent cells indicating that confluence abrogates intracellular FN trafficking as well as sEV secretion by VSMCs (Fig S3H)..  

• Inhibition of branched actin polymerization has been shown to reduce both exocytic and endocytic activity. Thus, it is hard to interpret the results of Fig. 2B than anything more than a generalized effect of losing actin.

We thank the reviewer for this important point regarding the broad cellular functions of branched actin polymerization, and agree that generalized actin loss can influence both exocytic and endocytic pathways. To address this, we performed additional experiments and analyses to better define the relationship between branched actin structures and sEV-related processes in VSMCs.

As described in the revised Results (page 6), we overexpressed ARPC2-GFP (an Arp2/3 subunit) together with F-tractin-RFP in VSMCs and carried out live-cell imaging. This approach revealed that Arp2/3 and F-actin organize into lamellipodial scaffolds at the cell cortex, as expected (Fig. S2A; Supplementary Video S2). Additionally, and more unexpectedly, we observed numerous Arp2/3– and F-actin–positive dynamic spots within the VSMC cytoplasm. These structures resemble actin comet tails seen in other systems, previously implicated in endosomal propulsion (Fig. S2A, arrow; Supplementary Video S2).

Quantitative analysis confirmed that a substantial fraction of these dynamic F-actin/cortactin spots colocalized with CD63+ endosomes (Fig. 2D), and that these structures are indeed branched actin tails based on cortactin immunostaining. Furthermore, inhibition of SMPD3 (with 3-OMS) induced enlarged cortactin/F-actin/CD63+ complexes, morphologically similar to invadopodia (Fig. 2D, arrowheads), supporting a functional link between actin branching and MVB dynamics.

To quantify the association, we calculated Manders’ colocalization coefficients for F-actin tails and CD63+ endosomal structures in fixed VSMCs, observing that ~50% of F-actin tails were associated with ~13% of endosomes. Upon 3-OMS treatment, this overlap increased further (Fig. S2B).

Finally, using live-cell imaging (Fig 2C; Supplementary Video S4), we directly observed CD63+ MVBs being propelled through the cytoplasm by Arp2/3-driven actin tails, suggesting a mechanistic role for branched actin assembly in MVB intracellular transport, rather than a generalized effect of actin disruption alone.

We believe these combined data reinforce a more specific mechanistic role for Arp2/3-mediated branched actin in MVB/endosome transport and, consequently, in sEV secretion in VSMCs—over and above an indirect effect of global actin loss. We hope these additional experiments and quantitative analyses address the reviewer’s concern and clarify the functional relevance of branched actin structures to sEV trafficking:

(1) Results, page 6, Ln 3 “As regulators of branched actin assembly, the Arp2/3 complex and cortactin are thought to contribute to sEV secretion in tumour cells by mediating MVB intracellular transport and plasma membrane docking[28, 33]. Therefore, we overexpressed the Arp2/3 subunit, ARPC2-GFP and the F-actin marker, F-tractin-RFP in VSMCs and performed live-cell imaging. As expected, Arp2/3 and F-actin bundles formed a distinct lamellipodia scaffold in the cellular cortex (Fig S2A and Supplementary Video S2). Unexpectedly, we also observed numerous  Arp2/3/F-actin positive spots moving  through the VSMC cytoplasm that resembled previously described endosome actin tails observed in Xenopus eggs[33] and parasite infected cells where actin comet tails propel parasites via filopodia to neighbouring cells[34, 35] (Fig S2A, arrow, and Supplementary Video S2). Analysis of the intracellular distribution of Arp2/3 and CD63-positive endosomes in VSMCs showed CD63-MVB propulsion by the F-actin tail in live cells (Fig 2C and Supplementary Video S4).”

(2) Results, New data Fig 2D, page 6, Ln 14. “we observed numerous F-actin spots in fixed VSMCs that were positive both for F-actin and cortactin indicating that these are branched-actin tails (Fig 2D). Moreover, cortactin/F-actin spots colocalised with CD63+ endosomes and addition of the SMPD3 inhibitor, 3-OMS, induced the appearance of enlarged doughnut-like cortactin/F-actin/CD63 complexes resembling invadopodia-like structures similar to those observed in tumour cells (Fig 2D, arrowheads)[18].”

(3) Results, New data Fig S2B, page 6, Ln 19 “To quantify CD63 overlap with the actin tail-like structures, we extracted round-shaped actin structures and calculated the thresholded Manders colocalization coefficient (Fig S2B).  We observed overlap between F-actin tails and CD63 as well as close proximity of these markers in fixed VSMCs (Fig S2B). Approximately 50% of the F-actin tails were associated with 13% of all endosomes (tM1=0.44±0.23 and tM2= 0.13±0.06, respectively, N=3). Addition of 3-OMS enhanced this overlap further (tM1=0.75±0.18 and tM2=0.25±0.09) suggesting that Arp2/3-driven branched F-actin tails are involved in CD63+ MVB intracellular transport in VSMCs”

• In video 1 the author states (lines 8-9; pg6) "intense CD63 staining along filopodia" Although, there is some fluorescence (not strong) in these structures, there was no visible exocytic activity. This data is more suggestive that sEVs (marked by CD63) are not associated with filopodia. The following conclusion statement the authors make is overreaching given this result.

We thank the reviewer for this careful observation and agree that the previous conclusion regarding sEV release from filopodia was overstated. In response, we have revised both the Results and Discussion sections to more accurately reflect the data..

(1) Results, page 6, Ln37 “We also attempted to visualise sEV release in filopodia using CD63-pHluorin where fluorescence is only observed upon the fusion of MVBs with the plasma membrane39. Using total internal reflection fluorescence microscopy (TIRF) we observed the typical “burst”-like appearance of sEV secretion at the cell-ECM interface in full agreement with an earlier report showing MVB recruitment to invadopodia-like structures in tumor cells18 (Fig S2B and Supplementary Video S1). Although we also observed an intense CD63-pHluorin staining along filopodia-like structures we were not able to detect typical “burst”-like events to confirm sEV secretion in filopodia. (Fig S2C and Supplemental Video S1)..”

(2) Discussion, page 12, Ln19 “Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells.”. 

• Fig 2D and video 2 are wholly unconvincing with regard to sEV secretion sites. The authors could use their CD63-pHluroin construct to count exocytic events in the filopodia vs the whole cell. Given the movie, I have a suspicion this would not be significant. The authors could also perform staining CD63 in non-permeabilized cells to capture and count exocytic events at the plasma membrane as well as their location between groups.

We thank the reviewer for these constructive suggestions and their critical assessment of our current data regarding the sites of sEV secretion. We agree that our CD63-pHluorin approach clearly indicates sEV secretion events in the soma at the cell–ECM interface, while we did not observe comparable events in filopodia. Accordingly, we have clarified these points in the revised manuscript.

(1) Results, page 6, Ln37 “We also attempted to visualise sEV release in filopodia using CD63-pHluorin where fluorescence is only observed upon the fusion of MVBs with the plasma membrane39. Using total internal reflection fluorescence microscopy (TIRF) we observed the typical “burst”-like appearance of sEV secretion at the cell-ECM interface in full agreement with an earlier report showing MVB recruitment to invadopodia-like structures in tumor cells18 (Fig S2B and Supplementary Video S1). Although we also observed an intense CD63-pHluorin staining along filopodia-like structures we were not able to detect typical “burst”-like events to confirm sEV secretion in filopodia. (Fig S2C and Supplemental Video S1)..”

(2) Discussion, page 12, Ln19 “Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells.”. 

• Fig. 2E and video 4. Again, the conclusions drawn from this data are very strained. First, no co-localization quantification is presented on the proportion of CD63 vesicles with actin. Once again, the movie, if anything convinces the reader that 95-99% of all CD63 vesicles are not associated with actin; therefore, this is an unlikely mechanism of transport.

We thank the reviewer for this valuable comment and for highlighting the need for quantitative co-localization analysis. In response, we developed a method to systematically quantify F-actin and CD63 co-localization in fixed VSMCs, as now presented in new Figures 2D and S2B. We acknowledge that the majority of CD63+ endosomes are not associated with F-actin, consistent with the reviewer’s interpretation. However, our quantitative data now show that a specific subpopulation of MVBs appears to utilize this actin-based mechanism for transport. We believe this addresses the concern and more accurately reflects the prevalence and significance of the mechanism described.

(1) Results, page 6 , Ln 19. “To quantify CD63 overlap with the actin tail-like structures, we extracted round-shaped actin structures and calculated the thresholded Manders colocalization coefficient (Fig S2B).  We observed overlap between F-actin tails and CD63 as well as close proximity of these markers in fixed VSMCs (Fig S2B). Approximately 50% of the F-actin tails were associated with 13% of all endosomes (tM1=0.44±0.23 and tM2= 0.13±0.06, respectively, N=3). Addition of 3-OMS enhanced this overlap further (tM1=0.75+/-0.18 and tM2=0.25+/-0.09) suggesting that Arp2/3-driven branched F-actin tails are involved in CD63+ MVB intracellular transport in VSMCs.”

• Are there perturbations that increase filopodia numbers? A gain of function experiment would be valuable here.

We thank the reviewer for this important suggestion regarding the potential value of gain-of-function experiments to clarify filopodia’s contribution to sEV release. In agreement with the reviewer’s scepticism, we have removed statements linking filopodia to sEV release from both the title and abstract to avoid overinterpretation. At present, our understanding of filopodia biology and the lack of robust tools to selectively and substantially increase filopodia numbers in VSMCs prevent us from directly addressing this question through gain-of-function assays. We acknowledge that future studies using established methods—such as overexpression of filopodia-inducing proteins (e.g., mDia2 or fascin)—could provide insight into whether an increased number of filopodia affects sEV release. However, such experiments are beyond the scope of the current manuscript. We have made the following changes to clarify these points:

(1) Results, page 6, Ln37 “We also attempted to visualise sEV release in filopodia using CD63-pHluorin where fluorescence is only observed upon the fusion of MVBs with the plasma membrane39. Using total internal reflection fluorescence microscopy (TIRF) we observed the typical “burst”-like appearance of sEV secretion at the cell-ECM interface in full agreement with an earlier report showing MVB recruitment to invadopodia-like structures in tumor cells18 (Fig S2B and Supplementary Video S1). Although we also observed an intense CD63-pHluorin staining along filopodia-like structures we were not able to detect typical “burst”-like events to confirm sEV secretion in filopodia. (Fig S2C and Supplemental Video S1)..”

(2) Discussion, page 12, Ln19 “Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells.”. 

Figure 3<br /> • Fig 3A. The CD63 staining is strongly associated with the entire plasma membrane. How are the authors distinguishing between normal membrane shedding and bona fida sEVs based on this staining alone (?)- this is insufficient as all membrane structures are seemingly positive. Additionally, there are very few sEVs in scrutinizing the provided images. For the "sEV secretion, fold change" graphs in previous figures, could the authors provide absolute values, or an indication of what these values are in absolute terms?

We thank the reviewer for raising this important point regarding the specificity of CD63 staining and the need to distinguish bona fide sEVs from membrane fragments or general membrane shedding. We agree that CD63 staining alone at the plasma membrane or in the extracellular matrix is not sufficient to unequivocally identify sEVs. To address this, we employed several complementary approaches to rigorously characterize ECM-associated sEVs:

First, using high-resolution iSIM imaging, we confirmed the association of CD63-positive particles specifically with the FN-rich matrix, and demonstrated that SMPD3 knockdown significantly reduced the number of CD63+ particles in the matrix (Fig. 3B; revised from Fig. S3A).

Second, by incubating FN matrices with purified and fluorescently labeled sEVs, we directly observed efficient entrapment of these labeled sEVs within the matrices (Fig. 3E), confirming that sEVs can interact with and be retained by the ECM.

Third, we developed and applied a sequential extraction protocol using mild salt buffer (0.5M NaCl) and strong denaturant (4M guanidine HCl) to selectively extract ECM-associated sEVs based on the strength of their association (see new Figs. S3A and S3B). Extracted vesicles were then characterized by ExoView analysis, which demonstrated a tetraspanin profile (CD63+/CD81+/CD9+) closely matching that of sEVs from conditioned media, providing evidence that these particles are true sEVs and not merely membrane debris. We also found that the more weakly bound (NaCl-extracted) fraction closely resembles media-derived sEVs, while the strongly bound (GuHCl-extracted) fraction is more enriched in CD63+ and CD63+/CD81+ sEVs but contains very few CD9+ vesicles, further supporting distinct extracellular vesicle subpopulations within the ECM.

In addition, the abundance of CD63+/CD81+ sEVs in both media and ECM-derived fractions was independently validated by CD63 bead-capture assay (Fig. S3B).

We hope these clarifications and the expanded data set address the reviewer’s concerns about sEV identification and quantification in the extracellular matrix:

(1) Results, page 7, Ln 16. To quantify ECM-trapped sEVs we applied a modified protocol for the sequential extraction of extracellular proteins using salt buffer (0.5M NaCl) to release sEVs which are loosely-attached to ECM via ionic interactions, followed by 4M guanidine HCl buffer (GuHCl) treatment to solubilize strongly-bound sEVs (Fig S3A) 42. We quantified total sEV and characterised the sEV tetraspanin profile in conditioned media, and the 0.5M NaCl and GuHCl fractions using ExoView. The total particle count showed that EVs are both loosely bound and strongly trapped within the ECM. sEV tetraspanin profiling showed differences between these 3 EV populations.  While there was close similarity between the conditioned media and the 0.5M NaCl fraction with high abundance of CD63+/CD81+ sEVs as well as CD63+/CD81+/CD9+ in both fractions (Fig S3A). In contrast, the GuHCl fraction was particularly enriched with CD63+ and CD63+/CD81+ sEVs with very low abundance of CD9+ EVs (Fig S3A). The abundance of CD63+/CD81+ sEVs was confirmed independently by a CD63+ bead capture assay in the media and loosely bound fractions (Fig S3B).

• A control of fig 3b would be helpful to parse out random uptake of extracellular debris verses targeted sEV internalization. It would be helpful if the authors added particles of similar size to that of the sEVs to test whether these structures are endocytosed/micropinocytosed at similar levels.

We thank the reviewer for this useful suggestion regarding the need for better controls to distinguish specific sEV uptake from nonspecific internalization of extracellular debris or similarly sized particles. As a comparison, in our study we analyzed the uptake of both sEVs and serum proteins such as fibronectin and fetuin-A (Figs S3C and S3D), and observed similar patterns of intracellular trafficking. However, we acknowledge that inert nanoparticles or beads of a similar size to sEVs could serve as potential controls to assess nonspecific micropinocytosis or endocytosis.

It is important to note, however, that the uptake of sEVs is strongly influenced by their surface protein composition and the so-called “protein corona.” Recent work from Prof. Khuloud T. Al-Jamal’s group underscores that exosome uptake mechanisms may be highly specific (Liam-Or et al., 2024), and studies from Mattias Belting’s lab have also shown the importance of heparan sulfate proteoglycans in exosome endocytosis (Cerezo-Magana et al., 2021). As a result, uptake comparisons with inert particles or beads may not fully recapitulate the specificity of sEV internalization, and distinct nanoparticle classes may rely on different uptake pathways.

Figure 4<br /> • Fig. 4E,F,G. How are the authors determining the neointima and media compartments without ancillary staining for basement membrane or endothelial markers? Anatomic specific markers need to be incorporated here for the reader to evaluate the specificity of the FN and CD81 staining. It is also hard to understand the severity of the atherosclerotic lesion without a companion H&E cross section.

We thank the reviewer for highlighting the need for more rigorous characterization of atherosclerotic lesion architecture and anatomical compartments in our study. In response, we have incorporated additional histological analyses and now provide ancillary staining and companion images to enable clear identification of the neointima and medial compartments, as well as to assess lesion severity (see new Figs S4A–S4D):

(1)Results, page  8, Ln 28. . “To test if FN associates with sEV markers in atherosclerosis, we investigated the spatial association of FN with sEV markers using the sEV-specific marker CD81. Staining of atherosclerotic plaques with haematoxylin and eosin revealed well-defined regions with the neointima as well as tunica media layers formed by phenotypically transitioned or contractile VSMCs, respectively (Fig S4A). Masson's trichrome staining of atherosclerotic plaques showed abundant haemorrhages in the neointima, and sporadic haemorrhages in the tunica media (Fig S4B). Staining of atherosclerotic plaques with orcein indicated weak connective tissue staining in the atheroma with a confluent extracellular lipid core, and strong specific staining at the tunica media containing elastic fibres which correlated well with the intact elastin fibrils in the tunica media (Figs S4C and S4D). Using this clear morphological demarcation, we found that FN accumulated both in the neointima and the tunica media where it was significantly colocalised with the sEV marker, CD81 (Fig. 4D, 4E and 4F). Notably CD81 and FN colocalization was particularly prominent in cell-free, matrix-rich plaque regions (Figs. 4E and 4F).”

• Figs s4c, S4d- proper controls are not provided. Again, a non-FN internalization control as well as a 4oC cold block negative control is required to interpret this data.

We thank the reviewer for this valuable suggestion. To enhance the rigor of our internalization assays, we have now included several additional controls using alternative treatments, fluorophore combinations, and internalization conditions:

a) We performed FN-Alexa568 uptake assays, followed by immunostaining for CD63 with a distinct fluorophore (Alexa488), to confirm the colocalization of internalized FN with CD63+ endosomal compartments in VSMCs (new Fig. S3E).

b) We also stained VSMCs, cultured under normal growth conditions, with an anti-FN antibody to visualize intracellular serum-derived FN and again observed colocalization with CD63 (new Figs. S3F and S3G). Notably, in cells grown to confluence, we observed a complete loss of intracellular FN staining and FN/CD63 colocalization, suggesting that FN recycling is prominent in sparse, motile cells, but not in confluent populations.

These additional controls strengthen our conclusions regarding FN internalization pathways and the conditions under which FN trafficking to the endosomal system occurs:

(1) Results, page 7, Ln 31  We treated serum-deprived primary human aortic VSMCs with FN-Alexa568 and found that it was endocytosed and subsequently delivered to early and late endosomes together with fetuin A, another abundant serum protein that is a recycled sEV cargo and elevated in plaques (Figs S3C and S3D). CD63 visualisation with a different fluorophore (Alexa488) confirmed FN colocalization with CD63+ MVBs (Fig S3E). Next, we stained non-serum deprived VSMC cultured in normal growth media (RPMI supplemented with 20% FBS) with an anti-FN antibody and observed colocalization of CD63 and serum-derived FN.  Co-localisation was reduced likely due to competitive bulk protein uptake by non-deprived cells (Fig S3F). Notably, when we compared FN distribution in sparsely growing VSMCs versus confluent cells we found that FN intracellular spots, as well as colocalization with CD63, completely disappeared in the confluent state (Fig S3F and S3G)..

• Can the authors please provide live and fixed imaging of FN and CD63-mediate filopodial secretion to amply support their conclusions.

We have observed CD63 MVBs in both fixed (Fig 2E) and live VSMCs (Fig 2F) yet we agree that further studies are required to establish the contribution of filopodia to sEV secretion. Therefore, we have added the following changes:

(1) Results, page 6, Ln37 “We also attempted to visualise sEV release in filopodia using CD63-pHluorin where fluorescence is only observed upon the fusion of MVBs with the plasma membrane39. Using total internal reflection fluorescence microscopy (TIRF) we observed the typical “burst”-like appearance of sEV secretion at the cell-ECM interface in full agreement with an earlier report showing MVB recruitment to invadopodia-like structures in tumor cells18 (Fig S2B and Supplementary Video S1). Although we also observed an intense CD63-pHluorin staining along filopodia-like structures we were not able to detect typical “burst”-like events to confirm sEV secretion in filopodia. (Fig S2C and Supplemental Video S1)..”

(2) Discussion, page 12, Ln19 “Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells.”. 

Figure 5

• Fig. 5A,B. The authors claim that sEV supplementation enhances VSMC migration speed and distance. The provided graphs show only a marginal increase in speed with sEV addition (A) but, concerningly, there is a four-star significant difference between the FN condition compared with FN+sEV (B) while the means appear the same. How are these conditions statistically different? The statistics seem off for these comparisons.

We thank the reviewer for highlighting concerns regarding the statistical analysis in Figures 5A and 5B. In response, we have carefully re-examined our data and statistical approach to ensure accuracy and transparency.

First, we have now included all individual cell migration tracks in the data representation for these figures. The statistical tests were repeated using the Kruskal–Wallis test with Dunn’s multiple comparison correction across all groups. This more stringent analysis confirmed our key findings: fibronectin (FN) stimulates VSMC migration speed, while inhibition of sEV secretion (with 3-OMS) reduces cellular speed (Fig. 5A). Addition of exogenous ECM-associated sEVs modestly restored cell speed in the presence of 3-OMS, but had no effect on baseline migration speed in 2D or 3D models (Figs. 5A, 5D).

Regarding the four-star significance observed in the original Fig. 5B, the previous result reflected an analysis based on pooled group averages, which may have overstated marginal differences. The revised analysis, based on individual cell tracks, does not support a substantial difference between FN and FN+sEV groups. The revised p-values and comparisons are now provided directly on the figures and described in the figure legends. We also clearly report the numbers of biological replicates, technical replicates, and individual data points for every condition.

Further, the modest effect of ECM-associated sEVs on speed is consistent with our observation that sEVs influence invasion directionality rather than baseline migration velocity, in agreement with previous findings in tumor models (Sung et al., 2015).

The manuscript has been revised accordingly, with updates in:

(1) Figures 5A and 5B: Individual cell track data are now shown, and statistical analyses have been repeated using the Kruskal–Wallis test with Dunn’s multiple comparisons.

(2) Figure legends and results sections: Numbers of biological and technical replicates, as well as individual data points, are now clearly stated.

(3) Results, page 9, line 14:  “FN as a cargo in sEVs promotes FA formation in tumour cells and increases cell speed14, 15. As we found that FN is loaded into VSMC-derived sEVs we hypothesized that ECM-entrapped sEVs can enhance cell migration by increasing cell adhesion and FA formation in the context of a FN-rich ECM. Therefore, we tested the effect of sEV deposition onto the FN matrix on VSMC migration in 2D and 3D models. We found that FN coating promoted VSMC velocity and inhibition of bulk sEV secretion with 3-OMS reduced VSMC speed in a 2D single-cell migration model (Figs. 5A, 5B) in agreement with previous studies using tumour cells14, 15. However, addition of sEVs to the ECM had no effect on VSMC speed at baseline but rescued cell speed and distance in the presence of the sEV secretion inhibitor, 3-OMS suggesting the EVs are not primarily regulating cell speed (Figs 5A and 5B).”

(4) Results, page 9, Ln 29 “Hence, ECM-associated sEVs have modest influence on VSMC speed but influence VSMC invasion directionality.”.

We hope that these changes address the reviewer’s concerns and improve the transparency and reproducibility of our data presentation

• Fig d-h. Generally, the magnitude of the difference between the presented conditions are biologically insignificant. Several of the graphs show a four-star difference with means that appear equivalent with overlapping error bars. Do the authors conclude that a 0.1%, or less, effect between groups is biologically meaningful?

We thank the reviewer for drawing attention to the apparent mismatch between statistical significance and biological relevance in Figures 5d–h. In response, we have reanalyzed the data using individual cell tracks and more stringent non-parametric statistical tests, as described above. This reanalysis confirmed that the magnitude of differences in migration speed and related parameters between the groups is minimal and not biologically meaningful. Thus, we no longer claim that sEVs significantly affect VSMC migration speed under these conditions in either 2D or 3D assays. Our revised manuscript now accurately reflects this finding in both the Results and Discussion sections, and the updated figures and legends clarify the true extent of any differences observed.

Figure 6

• Generally, the author's logic for looking into adhesion, focal adhesion and traction forces is hard to follow. If there are sEV-mediated migration differences, then there would inexorably be focal adhesion alterations. However, the data indicates few differences brought on by sEVs, which speaks to the lack of migration differences presented in Fig. 5. Overall, the sEV migration phenotype has so little of an effect, to then search for a mechanism seems destine to not turn up anything significant.

We thank the reviewer for highlighting the importance of connecting the observed phenotypic effects of sEVs to the investigation of adhesion and focal adhesion mechanisms. While our revised analysis confirms that sEVs have little to no effect on VSMC migration speed or distance in 2D and 3D models, we did observe a robust effect of sEVs on the directionality of cell invasion (Figs. 5G and 5H). This prompted us to look more closely at pathways involved in cell guidance rather than bulk cell motility.

Our proteomic comparison between larger EVs (10K fraction) and sEVs (100K fraction) revealed a unique adhesion complex present specifically on the sEVs—comprising collagen VI, TGFBI, LGALS3BP, and EDIL3 (Figs. 7A–C)—each of which has previously been implicated in integrin signaling, cell adhesion, or invasion. Functional blocking and knockdown studies further identified collagen VI as a key mediator in the regulation of cell adhesion and invasion directionality influenced by sEVs (Figs. 7F and 7I).

In response to this mechanistic insight, we have modified the graphical abstract and discussion to clarify our approach:

We now explicitly state that our focus has shifted from analyzing baseline migration speed to mechanisms guiding invasion directionality, in line with our key phenotypic findings.We highlight that the unique adhesion cluster identified on sEVs—including collagen VI and its cooperative partners—provides a strong mechanistic rationale for examining focal adhesion dynamics and ECM interactions, even in the absence of changes in migration velocity.Discussion excerpts (pages 13–14) have been updated to reflect this rationale and to summarize the potential significance of these findings for vascular biology and disease.

We hope this clarifies the logic underlying our approach and justifies the mechanistic studies performed in this context:

(1) Discussion, page 13, Ln 2  “Hence, it will be interesting in future studies to investigate whether sEVs can stimulate Rho activity by presenting adhesion modulators—particularly collagen VI—on their surface, thereby guiding cell directionality during invasion.”

(2) Discussion, page 13, Ln 30  “In addition to collagen VI the unique adhesion cluster in VSMC-derived sEVS also includes EGF-like repeat and discoidin I-like domain-containing protein (EDIL3), transforming growth factor-beta-induced protein ig-h3 (TGFBI) and the lectin galactoside-binding soluble 3 binding protein (LGALS3BP) and these proteins are also directly implicated in activation of integrin signalling and cellular invasiveness85-87. Although we found that collagen VI plays the key role in sEV-induced early formation of FAs in VSMCs, it is tempting to speculate that the high sEV efficacy in stimulating FA formation is driven by cooperative action of this unique adhesion complex on the sEVs surface and targeting this novel sEV-dependent mechanism of VSMC invasion may open-up new therapeutic opportunities to modulate atherosclerotic plaque development or even to prevent undesired VSMC motility in restenosis”.    . 

(3) Discussion, page 14, Ln 14 “In summary, cooperative activation of integrin signalling and F-actin cytoskeleton pathways results in the secretion of sEVs which associate with the ECM and play a signalling role by controlling FA formation and cell-ECM crosstalk. Further studies are needed to test these mechanisms across various cell types and ECM matrices.     ”.    

Figure 7<br /> • The authors need to provide additional evidence Col IV is harbored in sEVs and not a contaminant of sEV isolation as VSMCs secrete a copious amount of this in culture. For instance, IHC of isolated sEVs stained for CD63 and Col IV as well as single cell staining of the same sort.

We thank the reviewer for this important comment regarding the specificity of collagen VI detection in sEVs. To ensure that collagen VI is associated with bona fide sEVs—rather than being a contaminant resulting from high extracellular abundance—we performed a comparative analysis of vesicles isolated from the same conditioned media. Both proteomic mass spectrometry and western blotting revealed that collagen VI was exclusively present in the small EV (100K pellet) fraction and not in the larger EVs (10K pellet), as shown in Figs. 7B and 7C. Collagen VI was further identified in sEVs extracted from the ECM using our salt/guanidine protocol (new Fig. 7D).

Reviewer #2 (Recommendations For The Authors):

The authors have presented a nice collection of data with strong approaches to address their hypotheses. Nevertheless, an additional section within the Discussion would be welcome in addressing the potential limitations and important caveats to be considered alongside their study. These caveats and limitations could be reshaped by additional data supporting the ideas that: (1) small extracellular vesicles can be directly observed during their secretion from filopodia, (2) CD81 labeling in tissue can be interpreted clearly as extracellular vesicles and not the cell surface of other cell types (co-staining with an endothelial cell marker such as PECAM-1 perhaps), and (3) collagen VI within the vesicles is somehow accessed by adhesion molecules on the cell surface of migrating cells.

We thank the reviewer for these important suggestions and we have now added further studies and modified our conclusions to reflect the data more accurately:

(1) Results. Page 6, Ln37  “We also attempted to visualise sEV release in filopodia using CD63-pHluorin where fluorescence is only observed upon the fusion of MVBs with the plasma membrane39. Using total internal reflection fluorescence microscopy (TIRF) we observed the typical “burst”-like appearance of sEV secretion at the cell-ECM interface in full agreement with an earlier report showing MVB recruitment to invadopodia-like structures in tumor cells18 (Fig S2B and Supplementary Video S1). Although we also observed an intense CD63-pHluorin staining along filopodia-like structures we were not able to detect typical “burst”-like events to confirm sEV secretion in filopodia. (Fig S2C and Supplemental Video S1)”..  

(2) Discussion, page 12, Ln18: “Here we report that β1 integrin activation triggers sEV release followed by sEV entrapment by the ECM. Curiously we observed CD63+ MVB transport toward the filopodia tips as well as inhibition of sEV-secretion with filopodia formation inhibitors suggesting that sEV secretion can be directly linked to filopodia but further studies are needed to define the contribution of this pathway to the overall sEV secretion by cells”..

We quantified the colocalization of CD81 and CD31 to exclude the endothelial cell origin of sEVs and extended the characterisation of the atherosclerotic matrix as well as highlighting any limitations to interpretation ie re  CD81 ECM localisation: 

(1) Results, page 8, Ln 43 “An enhanced expression of CD81 by endothelial cells in early atheroma has been previously reported so to study the contribution of CD81+ sEVs derived from endothelial cells  we investigated the localisation of CD31 and CD8145. In agreement with a previous study, we found that the majority of CD31 colocalises with CD81 (Thresholded Mander's split colocalization coefficient 0.54±0.11, N=6) indicating that endothelial cells express CD81 (Fig 4G)45. However, only a minor fraction of total CD81 colocalised with CD31 (Thresholded Mander's split colocalization coefficient 0.24±0.06, N=6) confirming that the majority of CD81 in the neointima is originating from the most abundant VSMCs.. 

(2) Results, page 8, Ln 28: “To test if FN associates with sEV markers in atherosclerosis, we investigated the spatial association of FN with sEV markers using the sEV-specific marker CD81. Staining of atherosclerotic plaques with haematoxylin and eosin revealed well-defined regions with the neointima as well as tunica media layers formed by phenotypically transitioned or contractile VSMCs, respectively (Fig S4A). Masson's trichrome staining of atherosclerotic plaques showed abundant haemorrhages in the neointima, and sporadic haemorrhages in the tunica media (Fig S4B). Staining of atherosclerotic plaques with orcein indicated weak connective tissue staining in the atheroma with a confluent extracellular lipid core, and strong specific staining at the tunica media containing elastic fibres which correlated well with the intact elastin fibrils in the tunica media (Figs S4C and S4D). Using this clear morphological demarcation, we found that FN accumulated both in the neointima and the tunica media where it was significantly colocalised with the sEV marker, CD81 (Fig. 4D, 4E and 4F). Notably CD81 and FN colocalization was particularly prominent in cell-free, matrix-rich plaque regions (Figs. 4E and 4F). .”

We showed that collagen VI is presented on the surface of sEVs:

(1) Results, page 10, Ln43: “Collagen VI was the most abundant protein in VSMC-derived sEVs (Fig 7B, Table S7) and  was previously implicated in the interaction with the proteoglycan NG253 and suppression of cell spreading on FN54. To confirm the presence of collagen VI in ECM-associated sEVs we analysed sEVs extracted from the 3D matrix using 0.5M NaCl treatment and showed that both collagen VI and FN are present (Fig 7D). Next, we analysed the distribution of collagen VI using dot-blot. Alix staining was bright only upon permeabilization of sEV indicating that it is preferentially a luminal protein (Fig 7E). On the contrary, CD63 staining was similar in both conditions showing that it is surface protein (Fig 7E). Interestingly, collagen VI staining revealed that 40% of the protein is located on the outside surface with 60% in the sEV lumen (Fig 7E)

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

(1) Given the mechanical nature of the device and the propensity for mice to urinate on things, I also wonder how frequently the device breaks/needs to be repaired. Perhaps some details regarding the cost and reliability of the device would be helpful to include, as these are the two things that could make researchers hesitant to adopt immediately.

We thank the reviewer for their astute observations. We also noted the problem of mouse waste and incorporated this concern into the redesign we mention in the text.

“Mouse waste getting on mechanical parts was found to be a major concern for the initial version of the device. As part of the redesign, the linear stages were moved out from under the mice to avoid this problem. Despite this problem, the original version of the device has not had any of its stages break down yet. A common problem though was that stimulus tips would blunt or break if they hit the mesh of the mesh table, requiring replacement. This has been solved in the latest version through a new feature where the mesh is detected via the force sensor, prompting immediate stimulus withdrawal, avoiding damage.”

In regards to cost and adoption, we have added this sentence to the final line of the discussion:

“To promote wide adaptation of this device across as many labs as possible, a company, Tactorum Inc., has been formed.”

(2) The only major technical concern, which is easy to address, is whether the device generates ultrasonic sounds that rodents can hear when idle or operational, across the ultrasonic frequencies that are of biological relevance (20-110 kHz). These sounds are generally alarm vocalizations and can create stress in animals, and/or serve as cues of an impending stimulus (if indeed they are produced by the device).

The reviewer brings up an interesting question. The ARM does not make a lot of noise, but some of the noise it emits does range into the 20-110 kHz range, though besides this does not qualitatively have other similarities to a mouse vocalization. Based on this we tested whether the noise produced by the ARM causes stress in naïve mice.

“A concern was raised that the noise of the ARM may cause stress in the mice tested. To test this, the open field test was performed with naïve mice (n=10) 2 feet from the ARM while the ARM either sat silent or ran through its habituation program, producing noise. The mouse's center point movement was then tracked in relation to the chamber, its edges, and center. No significant differences were found in distance traveled, center entrances, center, time in center, and latency to center entrance based on a student’s two-tailed t-test (Figure S1D-G). Based on this, neither stress nor locomotion differences were detected by this test, indicating the ARM does not induce an increased stress state due to its noise, even in non-habituated mice.”

(3) This sentence in the intro may be inaccurate: "or the recent emergence of a therapeutic targeting voltage-gated sodium channels, that block pain in both rodents and humans such as VX-548 for NaV1.8 (Jones 2023)" Despite extensive searching, I have been unable to find a reference showing that VX-548 is antinociceptive in rodents (rats or mice). As for why this is the case, I do not know. One speculation: this drug may be selective for the human Nav1.8 channel (but again, I have found no references comparing specificity on human vs rodent Nav1.8 channels). To not mislead the field into thinking VX-548 works for rodents and humans, please remove "both rodents and" from the sentence above (unless you find a reference supporting VX-548 as being effective in pain assays with rodents. There is a PK/PD paper with rodents, but that only looks at drug metabolism, not efficacy with pain assays).

We agree with the reviewer and have removed mention of the new Nav1.8 therapeutic also working in rodents.

(4) In the intro paragraph where variability in measuring mechanical stimuli is described, there is a new reference from the Stucky lab that further supports the need for an automated way to measure allodynia, as they also found variability between experimenters. This would be a relevant reference to include: Rodriguez Garcia (2024) PMID: 38314814.

Thanks to the reviewer for this relevant citation and we have updated the text to incorporate this:

“Recent studies utilizing the manual highspeed analysis of withdrawal behavior analysis developed by Abdus-Saboor et al. 2019 has reproduced this sizable experimenter effect using the new technique. (Rodríguez García 2024)”

(5) "a simple sin wave motion": should be "sine", correct throughout (multiple instances of "sin")

Corrections made where relevant.

Reviewer #2 (Public review):

(1) ARM seems like a fantastic system that could be widely adopted, but no details are given on how a lab could build ARM, thus its usefulness is limited.

The reviewer raises a good point, unfortunately the authors are constrained by university policies around patent law. That said efforts are being made to make the ARM widely available to interested researchers. As mentioned above to Reviewer 1’s comments, we end the discussion section with this sentence:

“To promote wide adaptation of this device across as many labs as possible, a company, Tactorum Inc., has been formed.”

(2) The ARM system appears to stop short of hitting the desired forces that von Frey filaments are calibrated toward (Figure 2). This may affect the interpretation of results.

The reviewer gives an important observation. We amended the text to include more clarity on the max forces induced, and comments on causes beyond the delivery mechanism. It should be noted that a newly bought fresh set of von Frey’s was used.

“With the same 1.4 and 2 g von Frey filaments Researcher 1 delivered max average forces of 1.5 g and 2.7 g, and Researcher 2 1.35 g and 2.4 g. The ARM delivered average max forces closest to the targeted forces, with 1.36 g and 1.9 g. (Figure 2C) Some of the error observed could be due to the error rate (+/- 0.05 g) in the force gauge and the von Frey set used.”

(3) The authors mention that ARM generates minimal noise; however, if those sounds are paired with stimulus presentation, they could still prompt a withdrawal response. Including some 'catch' trials in an experiment could test for this.

The reviewer makes a very useful suggestion that we incorporated into our carrageenan experiments. This new data can be found in Supplemental Figure 3F.

“For the carrageenan model, three replicates of the force ramp stimulus were delivered to each paw, and catch trials were performed every 3^rd trial to test whether the mice would respond to the noise of the ARM alone. During catch trials, the stimulus was delivered to the open air behind the mouse, and any movement within 5 seconds of stimulus delivery was counted as a response. These trials found a 96% response rate in true trials, with only a 7% rate in catch trials, indicating responses were not being driven by device noise.”

(4) The experimental design in Figure 2 is unclear- did each experimenter have their own cohort of 10 mice, or was a single cohort of mice shared? If shared, there's some concern about repeat testing.

Further clarification was added to avoid confusion on the methods used here.

“Separate cohorts of 10 mice were used for ARM and manual delivery, with a week given between each researcher to avoid sensitization.”

(5) In Figure 5 and S4, the order of the legends does not match the order of the graphs. This can be particularly confusing as the color scheme is not colorblind-friendly. Please consider revising the presentation of these figures.

Corrections made where relevant.

Reviewer #3 (Public review):

(1) Limited details are provided for statistical tests and inappropriate claims are cited for individual tests. For example, in Figure 2, differences between researchers at specific forces are reported to be supported by a 2-way ANOVA; these differences should be derived from a post-hoc test that was completed only if the independent variable effects (or interaction effect) were found to be significant in the 2-way ANOVA. In other instances, statistical test details are not provided at all (e.g., Figures 3B, 3C, Figure 4, Figure 6G).

We would like to thank the reviewer for pointing out the lack of clarity in the text on these statistical methods. We have added further details across the manuscript and shown below here in order to address this concern.

“Both manual delivery and the ARM produced significant paw withdrawal percentage curves, a standard traditional measurement of mechanical sensitivity in the field (von Frey 1896, Dixon 1980, Chaplan 1994)(Figure 2E), with a 2-way ANOVA and a posthoc Tukey test detecting significant increases in comparing the 3 lower force VFH’s (0.02g, 0.07g, 0.16g) to the 2 highest force VFH’s (1g, 1.4g). This demonstrates that the ARM delivers results comparable to highly experienced researchers. However, a 2-way ANOVA and a posthoc Tukey test found that Researcher 2 elicited a significantly higher (p=0.0008) paw withdrawal frequency than Researcher 1 (Figure S2A) which corresponded with Researcher 2’s higher VFH application time as measured by the force sensor (Figure 2B).”

“Adjustments were then made to the PAWS software to automate the measurement of withdrawal latency based on pose tracking data of the withdrawal response and the trajectory of the stimulus delivery encoded into the ARM. Testing of C57/BL6J (n=15) at baseline found significant decreases in withdrawal latency for pinprick compared to cotton swab stimuli delivered in identical ways by the ARM (Figure 3B) based on a 2-tailed student t-test.”

“Mice injected with carrageenan (n=15) showed elevated shaking behavior (p=0.0385, 2-way ANOVA and a posthoc Tukey test) in response to pinprick stimuli in comparison to measurements at baseline (Figure 3C).”

“Remote habituated mice showed a significant decrease (p=0.0217, 2-way ANOVA) in time to rest over the 3 days (Figure 4B), but no significant differences for any single day. The number of turns was measured for each group during the first 10 minutes of day 1 to act as a baseline, and then from 20 to 30 minutes for each day. Turn counts were then compared as a percentage of the baseline count for each group. This period was chosen as it the period when experiments start after the day of habituation on experimental days. It was found that remote-habituated mice showed significantly less turning on day 2 compared to mice habituated with a researcher present (p=0.024, 2-way ANOVA posthoc Tukey test), and that only the remote-habituated mice showed significantly decreased turning behavior on day 3 compared to day 1 (p=0.0234, 2-way ANOVA posthoc Tukey test) (Figure 4C).”

“Sex-dependent differences were found in reflexive and affective behavioral components of the mouse withdrawal response when a researcher was present versus not for both reactions to innocuous and noxious stimuli. A 2-way ANOVA and a posthoc Tukey test found that cotton swab stimuli elicited increased male mouse reflexive paw withdrawal features, including max paw height (p=0.0413) and max paw velocity (Y-axis) (p=0.0424) when Researcher 1 was present compared to when no researcher was present (Figure 4E-F). Pinprick stimuli (Figure 4H-I) on the other hand led to increased max paw height (p=0.0436) and max paw velocity (Y-axis) (p=0.0406) in male mice compared to female mice when Researcher 1 was present.

Analysis of the shaking behavior elicited by cotton swab and pinprick stimuli found no significant differences in shaking behavior duration (Figure 4SA-B) but found sex-dependent differences in paw distance traveled after the initial withdrawal, including during shaking and guarding behaviors. For cotton swab (Figure 4G) male mice showed significantly increased paw distance traveled compared to female mice when Researcher 2 was present (p=0.0468, 2-way ANOVA posthoc Tukey test) but not when Researcher 2 was present or no researcher was present. Pinprick stimuli also elicited sex-based increases in paw distance traveled (Figure 4J) in male mice when Researcher 2 was present compared to both male mice when no researcher was present (p=0.0149, 2-way ANOVA posthoc Tukey test) and female mice when Researcher 1 was present (p=0.0038, 2-way ANOVA posthoc Tukey test).”

(2) In the current manuscript, the effects of the experimenter's presence on both habituation time and aspects of the withdrawal reflex are minimal for Researcher 2 and non-existent for Research 1. This is surprising given that Researcher 2 is female; the effect of experimenter presence was previously documented for male experiments as the authors appropriately point out (Sorge et al. PMID: 24776635). In general, this argument could be strengthened (or perhaps negated) if more than N=2 experiments were included in this assessment.

The reviewer makes an important point regarding this data and the need for further experiments. We designed a new set of experiments to examine the effect of male and female researchers overall. It should be noted that this is rather noisy data given it was collected by three sets of male and female researchers over 3 weeks. That said a significant difference was found between mouse sexes when a male researcher was present. This is consistent with previous data, but as we discuss this does not invalidate previous data as researcher gender appears to be only one of the factors at work in researcher presence effects on mouse behavior, leading to individuals having the potential for greater or lesser effects than their overall gender. Our new results can be found in Figure 4K.

“These results indicate that researcher presence at baseline can lead to significant differences in reflexive and affective pain behavior. In this case, male mice showed increased behavioral responses to both touch and pain behavior depending on whether the researcher was present. This led to sex differences in the affective and reflexive component of the withdrawal response when a researcher is present, which disappears when no researcher is present, or a different researcher is present. For this set of researchers, the female researcher elicited the greater behavioral effect. This appeared at first to contradict previous findings (Sorge 2024, Sorge 2014), but it was hypothesized that the effect of an individual researcher could easily vary compared to their larger gender group. To test this, 6 new researchers, half male and half female, were recruited and a new cohort of mice (n=15 male, n=15 female) was tested in each of their presence over the course of 3 weeks, controlling for circadian rhythms (Figure 4K). The newly added force ramp stimulus type was used for these experiments, with three replicates per trial, to efficiently measure mechanical threshold in a manner comparable to previous work. It was found that female mice showed significantly decreased mechanical threshold compared to male mice (p=0.034, Šídák's multiple comparisons test and student’s t-test) when a male researcher was present. This did not occur when a female researcher or no researcher was present. In the latter case of slight trend towards this effect was observed, but it was not significant (p=0.21), and may be the result of a single male researcher being responsible for handling and setting up the mice for all experiments.”

“These findings indicate that sex-dependent differences in evoked pain behavior can appear and disappear based on which researcher/s are in the room. There is a trend towards male researchers overall having a greater effect, but individuals may have a greater or lesser effect on mouse behavior, independent of the gender or sex. This presents a confound that must be considered in the analysis of sex differences in pain and touch behavior which may explain some of the variation in findings from different researchers. Together, these results suggest that remote stimulus delivery may be the best way to eliminate variation caused by experimenter presence while making it easier to compare with data from researchers in your lab and others.”

(3) The in vivo BLA calcium imaging data feel out of place in this manuscript. Is the point of Figure 6 to illustrate how the ARM can be coupled to Inscopix (or other external inputs) software? If yes, the following should be addressed: why do the up-regulated and down-regulated cell activities start increasing/decreasing before the "event" (i.e., stimulus application) in Figure 6F? Why are the paw withdrawal latencies and paw distanced travelled values in Figures 6I and 6J respectively so much faster/shorter than those illustrated in Figure 5 where the same approach was used?

Thanks to the reviewer for bringing up this concern. We have included further text discussing this behavioral data and how it compares to previous work in this study.

“Paw height and paw velocity were found to be consistent with data from figures 4E-I (male researcher and male mice) and 5C (stimulus intensity 2.5 and 4.5) for similar data, with slightly elevated measures of paw distance traveled and decreased paw withdrawal latency for the pinprick stimulus. This was likely caused by sensitization due to multiple stimulus deliveries over the course of the experiment, as due to logistics, 30 stimulus trials were delivered per session due to logistical constraints vs the max of 3 that were performed during previous experiments.”

“This data indicates that the ARM is an effective tool for efficiently correlating in vivo imaging data with evoked behavioral data, including sub-second behavior. One limitation is that the neural response appears to begin slightly before stimulus impact (Figure 6F, 6SB). This was likely caused by a combination of the imprecise nature of ARM v1 paw contact detection and slight delays in the paw contact signal reaching the Inscopix device due to flaws in the software and hardware used, slowing down the signal. Improvements have been made to eliminate this delay as part of the ARM v2, which have been shown to eliminate this delay in vivo fiber photometry data recorded as part of new projects using the device.”

(4) Another advance of this manuscript is the integration of a 500 fps camera (as opposed to a 2000 fps camera) in the PAWS platform. To convince readers that the use of this more accessible camera yields similar data, a comparison of the results for cotton swabs and pinprick should be completed between the 500 fps and 2000 fps cameras. In other words, repeat Supplementary Figure 3 with the 2000 fps camera and compare those results to the data currently illustrated in this figure.

The reviewer makes a good point about the need for direct comparison between 500 fps and 2000 fps data. To address this we added data from same mice, from 2 weeks prior with a comparable set up. These new results can be found in Supplemental Figure 3.

“Changes were made to PAWS to make it compatible with framerates lower than 2000 fps. This was tested using a 0.4 MP, 522 FPS, Sony IMX287 camera recording at 500 fps, and data recorded at 2000 fps by the previously used photron fastcam (Figure 3SC-F). The camera paired with PAWS was found to be sufficient to separate between cotton swab and pinprick withdrawal responses, suggesting it may be a useful tool for labs that cannot invest in a more expensive device. PAWS features measured from 500 fps video data were not significantly different from the 2000 fps data based on a 2 way ANOVA.”

(5) In Figure 2F, the authors demonstrate that a von Frey experiment can be completed much faster with the ARM vs. manually. I don't disagree with that fact - the data clearly show this. I do, however, wonder if the framing of this feature is perhaps too positive; many labs wait > 30 s between von Frey filament applications to prevent receptive field sensitization. The fact that an entire set of ten filaments can be applied in < 50 s (< 3 s between filaments given that each filament is applied for 2 s), while impressive, may never be a feature that is used in a real experiment.

The reviewer makes an important point about how different researchers perform these tests and the relevant timings. We have moderated the framing of these results to address this concern.

“Further, we found that the ARM decreased the time needed to apply a stimulus 10 times to a mouse paw by 50.9% compared to manual delivery (Figure 2F). This effect size may decrease for researchers who leave longer delays between stimulus delivery, but the device should still speed up experiments by reducing aiming time and allowing researchers to quickly switch to a new mouse while waiting for the first.”

(6) Why are different affective aspects of the hindpaw withdrawal shown in different figures? For example, the number of paw shakes is shown in Figure 3C, whereas paw shaking duration is shown in Figure 5D. It would be helpful - and strengthen the argument for either of these measures as being a reproducible, reliable measure of pain - if the same measure was used throughout.

Thanks to the reviewer for pointing out this discrepancy. We have adjusted the figures and text to only use the Number of Paw Shakes for better consistency (Figure 5D and Figure 5-figure supplement 1C).

(7) Is the distance the paw traveled an effective feature of the paw withdrawal (Figure 5E)? Please provide a reference that supports this statement.

A relevant citation and discussion of this metric based on previous studies has been added.

“Mice injected with carrageenan (n=15) showed elevated shaking behavior (p=0.0385) in response to pinprick stimuli in comparison to measurements at baseline (Figure 3C). This aligned with previous findings where PAWS has detected elevations in shaking and/or guarding behavior, examples of affective pain behavior, and post-peak paw distance traveled, which correlates with these behaviors in carrageenan pain models and has been to found to be a good measure of them in past studies (Bohic et al. 2023).”

(8) Dedek et al. (PMID: 37992707) recently developed a similar robot that can also be used to deliver mechanical stimuli. The authors acknowledge this device's ability to deliver optogenetic and thermal stimuli but fail to mention that this device can deliver mechanical stimuli in a similar manner to the device described in this paper, even without experimenter targeting. Additional discussion of the Dedek et al. device is warranted.

We would like to thank the reviewer for identifying  this omission. Discussion of this as well as further discussion of Dedek et al.’s automation prototyping work has been added.

“Previous attempts at automating mechanical stimulus delivery, including the electronic von Frey (Martinov 2013) and dynamic plantar asthesiometer (Nirogi 2012), have focused on eliminating variability in stimulus delivery. In contrast to the ARM, both of these devices rely upon a researcher being present to aim or deliver the stimulus, can only deliver vFH-like touch stimuli, and only measure withdrawal latency/force threshold. Additionally, progress has been made in automating stimulus assays by creating devices with the goal of delivering precise optogenetic and thermal stimuli to the mouse’s hind paw (Dedek 2023, Schorscher-Petchu 2021). The Prescott team went farther and incorporated a component into their design to allow for mechanical stimulation but this piece appears to be limited to a single filament type that can only deliver a force ramp. As a result these devices and those previously discussed lack of customization for delivering distinct modalities of mechanosensation that the ARM allows for. Moreover, in its current form the automated aiming of some of these devices may not provide the same resolution or reliability of the ARM in targeting defined targets (Figure 1C), such as regions of the mouse paw that might be sensitized during chronic pain states. Due to the nature of machine learning pose estimation, substantial work, beyond the capacity of a single academic lab, in standardizing the mouse environment and building a robust model based on an extensive and diverse training data set will be necessary for automated aiming to match the reliability or flexibility of manual aiming. That said, we believe this work along with that of that of the other groups mentioned has set the groundwork from which a new standard for evoked somatosensory behavior experiments in rodents will be built.”

(9) Page 2: von Frey's reference year should be 1896, not 1986.

This typo has been fixed, thanks to the reviewer for noting it.

“For more than 50 years, these stimuli have primarily been the von Frey hair (vFH) filaments that are delivered to the mouse paw from an experimenter below the rodent aiming, poking, and subsequently recording a paw lift (von Frey 1896, Dixon 1980, Chaplan 1994).”

(10) Page 2: Zumbusch et al. 2024 also demonstrated that experimenter identification can impact mechanical thresholds, not just thermal thresholds.

Text has been updated in order to note this important point.

“A meta-analysis of thermal and mechanical sensitivity testing (Chesler 2002, Zumbusch 2024) found that the experimenter has a greater effect on results than the mouse genotype, making data from different individual experimenters difficult to merge.”

(11) Page 2: One does not "deliver pain in the periphery". Noxious stimuli or injury can be delivered to the periphery, but by definition, pain is a sensation that requires a central nervous system.

Text has been updated for improved accuracy.

“Combining approaches to deliver painful stimuli with techniques mapping behavior and brain activity could provide important insights into brain-body connectivity that drives the sensory encoding of pain.”

Author response:

The following is the authors’ response to the current reviews.

Reviewer #1 (Public review): 

Summary: 

Ferreiro et al. present a method to simulate protein sequence evolution under a birth-death model where sequence evolution is guided by structural constraints on protein stability. The authors then use this model to explore the predictability of sequence evolution in several viral proteins. In principle, this work is of great interest to molecular evolution and phylodynamics, which has struggled to couple non-neutral models of sequence evolution to phylodynamic models like birth-death processes. Unfortunately, though, the model shows little improvement over neutral models in predicting protein sequence evolution, although it can predict protein stability better than models assuming neutral evolution. It appears that more work is needed to determine exactly what aspects of protein sequence evolution are predictable under such non-neutral phylogenetic models. 

We thank the reviewer for the positive comments about our work. We agree that further work is needed in the field of substitution models of molecular evolution to enable more accurate predictions of specific amino acid sequences in evolutionary processes.

Major concerns: 

(1) The authors have clarified the mapping between birth-death model parameters and fitness, but how fitness is modeled still appears somewhat problematic. The authors assume the death rate = 1 - birth rate. So a variant with a birth rate b = 1 would have a death rate d = 0 and so would be immortal and never die, which does not seem plausible. Also I'm not sure that this would "allow a constant global (birth-death) rate" as stated in line 172, as selection would still act to increase the population mean growth rate r = b - d. It seems more reasonable to assume that protein stability affects only either the birth or death rate and assume the other rate is constant, as in the Neher 2014 model. 

The model proposed by Neher, et al. (2014), which incorporates a death rate (d) higher than 0 for any variant, was implemented and applied in the present method. In general, this model did not yield results different from those obtained using the model that assumes d = 1 – b, suggesting that this aspect may not be crucial for the study system. Next, the imposition of arbitrary death events based on an arbitrary death rate could be a point of concern. Regarding the original model, a variant with d = 0 can experience a decrease in fitness through the mutation process. In an evolutionary process, each variant is subject to mutation, and Markov models allow for the incorporation of mutations that decrease fitness (albeit with lower probability than beneficial ones, but they can still occur). All this information is included in the manuscript.

(2) It is difficult to evaluate the predictive performance of protein sequence evolution. This is in part due to the fact that performance is compared in terms of percent divergence, which is difficult to compare across viral proteins and datasets. Some protein sequences would be expected to diverge more because they are evolving over longer time scales, under higher substitution rates or under weaker purifying selection. It might therefore help to normalize the divergence between predicted and observed sequences by the expected or empirically observed amount of divergence seen over the timescale of prediction. 

AU: The study protein datasets showed different levels of sequence divergence over their evolutionary times, as indicated for each dataset in the manuscript. For some metrics, we evaluated the accuracy (or error) of the predictions through direct comparisons between real and predicted protein variants using percentages to facilitate interpretation: 0% indicates a perfect prediction (no error), while 100% indicates a completely incorrect prediction (total error). Regarding normalization of these evaluations, we respectfully disagree with the suggestion because diverse factors can affect (not only the substitution rate, but also the sample size, structural features of the protein that may affect stability when accommodating different sequences, among others) and this complicates defining a consistent and meaningful normalization criterion. Given that the manuscript provides detailed information for each dataset, we believe that the presentation of the prediction accuracy through direct comparisons between real and predicted protein variants, expressed as percentages of similarity, is the clearest way.

(3) Predictability may also vary significantly across different sites in a protein. For example, mutations at many sites may have little impact on structural stability (in which case we would expect poor predictive performance) while even conservative changes at other sites may disrupt folding. I therefore feel that there remains much work to be done here in terms of figuring out where and when sequence evolution might be predictable under these types of models, and when sequence evolution might just be fundamentally unpredictable due to the high entropy of sequence space. 

We agree with this reflection. Mutations can have different effects on folding stability, which are accounted for by the model presented in this study. However, accurately predicting the exact sequences of protein variants with similar stability remains difficult with current structurally constrained substitution models, and therefore, further work is needed in this regard. This aspect is indicated in the manuscript.

We want to thank the reviewer again for taking the time to revise our work and for the insightful and helpful comments.

Reviewer #2 (Public review): 

In this study, the authors aim to forecast the evolution of viral proteins by simulating sequence changes under a constraint of folding stability. The central idea is that proteins must retain a certain level of structural stability (quantified by folding free energy, ΔG) to remain functional, and that this constraint can shape and restrict the space of viable evolutionary trajectories. The authors integrate a birth-death population model with a structurally constrained substitution (SCS) model and apply this simulation framework to several viral proteins from HIV-1, SARS-CoV-2, and influenza.

The motivation to incorporate biophysical constraints into evolutionary models is scientifically sound, and the general approach aligns with a growing interest in bridging molecular evolution and structural biology. The authors focus on proteins where immune pressure is limited and stability is likely to be a dominant constraint, which is conceptually appropriate. The method generates sequence variants that preserve folding stability, suggesting that stability-based filtering may capture certain evolutionary patterns. 

Correct. We thank the reviewer for the positive comments about our study.

However, the study does not substantiate its central claim of forecasting. The model does not predict future sequences with measurable accuracy, nor does it reproduce observed evolutionary paths. Validation is limited to endpoint comparisons in a few datasets. While KL divergence is used to compare amino acid distributions, this analysis is only applied to a single protein (HIV-1 MA), and there is no assessment of mutation-level predictive accuracy or quantification of how well simulated sequences recapitulate real evolutionary paths. No comparison is made to real intermediate variants available from extensive viral sequencing datasets which gather thousands of sequences with detailed collection date annotation (SARS-CoV-2, Influenza, RSV). 

There are several points in this comment.

The presented method accurately predicts folding stability of forecasted variants, as shown through comparisons between real and predicted protein variants. However, as the reviewer correctly indicates, predicting the exact amino acid sequences remains challenging. This limitation is discussed in detail in the manuscript, where we also suggest that further improvements in substitution models of protein evolution are needed to better capture the evolutionary signatures of amino acid change at the sequence level, even between amino acids with similar physicochemical properties. Regarding the time points used for validation, the studied influenza NS1 dataset included two validation points. A key limitation in increasing the number of time points is the scarcity of datasets derived from monitoring protein evolution with sufficient molecular diversity between samples collected at consecutive time points (i.e., at least more than five polymorphic amino acid sites). 

As described in the manuscript, calculating Kullback-Leibler (KL) divergence requires more than one sequence per studied time point. However, most datasets in the literature include only a single sequence per time point, typically a consensus sequence derived from bulk population sequencing. Generating multiple sequences per time point is experimentally more demanding, often requiring advanced methods such as single-virus sequencing or amplification of sublineages in viral subpopulations, as was done for the first dataset used in the study (Arenas, et al. 2016), which enabled the calculation of KL divergence. The extent to which the simulated sequences resemble real evolution is evaluated in the method validation. As noted, intermediate time point validation was performed using the influenza NS1 protein dataset. Although, as the reviewer indicates, thousands of viral sequences are available, these are usually consensus sequences from bulk sequencing. Indeed, many viral variants mainly differ through synonymous mutations, where the number of accumulated nonsynonymous mutations is small. For example, from the original Wuhan strain to the Omicron variant, the SARS-CoV-2 proteins Mpro and PLpro accumulated only 10 and 22 amino acid changes, respectively.

Analyzing intermediate variants of concern (i.e., Gamma or Delta) would reduce this number affecting statistics. In addition, many available viral sequences are not consecutive in evolutionary terms (one dataset does not represent the direct origin of another dataset at a subsequent time point), which further limits their applicability in this study. There is little data from monitored protein evolution with consecutive samples. The most suitable studies usually involve in vitro virus evolution, but the data from these studies often show low genetic variability between samples collected at different time points. Finally, it is important to note that the presented method can only be applied to proteins with known 3D structures, as it relies on selection based on folding stability. Non-structural proteins cannot be analyzed using this approach. Future work could incorporate additional selection constraints, which may improve the accuracy of predictions. These considerations and limitations are indicated in the manuscript.

The selection of proteins is narrow and the rationale for including or excluding specific proteins is not clearly justified. 

The viral proteins included in the study were selected based on two main criteria, general interest and data availability. In particular, we included proteins from viruses that affect humans and for which data from monitored protein evolution, with sufficient molecular diversity between consecutive time points, is available. These aspects are indicated in the manuscript.

The analyzed datasets are also under-characterized: we are not given insight into how variable the sequences are or how surprising the simulated sequences might be relative to natural diversity. Furthermore, the use of consensus sequences to represent timepoints is problematic, particularly in the context of viral evolution, where divergent subclades often coexist - a consensus sequence may not accurately reflect the underlying population structure. 

The manuscript indicates the sequence identity among protein datasets of different time points, along with other technical details. Next, the evaluation based on comparisons between simulated and real sequences reflects how surprising the simulated sequences might be relative to natural diversity, considering that the real dataset is representative. We believe that the diverse study real datasets are useful to evaluate the accuracy of the method in predicting different molecular patterns. Regarding the use of consensus sequences, we agree that they provide an approximation. However, as previously indicated, most of the available data from monitored protein evolution consist of consensus sequences obtained through bulk sequencing. Additionally, analyzing every individual viral sequence within a viral population, which is typically large, would be ideal but computationally intractable.

The fitness function used in the main simulations is based on absolute ΔG and rewards increased stability without testing whether real evolutionary trajectories tend to maintain, increase, or reduce folding stability over time for the particular systems (proteins) that are studied. While a variant of the model does attempt to center selection around empirical ΔG values, this more biologically plausible version is underutilized and not well validated.

The applied fitness function, based on absolute ΔG, is well stablished in the field (Sella and Hirsh 2005; Goldstein 2013). The present study independently predicts ΔG for the real and simulated protein variants at each sampling point. This ΔG prediction accounts not only for negative design, informed by empirical data, but also for positive design based on the study data (Arenas, et al. 2013; Minning, et al. 2013), thereby enabling the detection of variation in folding stability among protein variants. These aspects are indicated in the manuscript. Therefore, in our view, the study provides a proper comparison of real and predicted evolutionary trajectories in terms of folding stability.

Ultimately, the model constrains sequence evolution to stability-compatible trajectories but does not forecast which of these trajectories are likely to occur. It is better understood as a filter of biophysically plausible outcomes than as a predictive tool. The distinction between constraint-based plausibility and sequence-level forecasting should be made clearer. Despite these limitations, the work may be of interest to researchers developing simulation frameworks or exploring the role of protein stability in viral evolution, and it raises interesting questions about how biophysical constraints shape sequence space over time. 

The presented method estimates the fitness of each protein variant, which can reflect the relative survival capacity of the variant. Therefore, despite the error due to evolutionary constraints not considered by the method, it indicates which variants are more likely to become fixed over time. In our view, the method does not merely filter plausible variants, rather, it generates predictions of variant survival through predicted fitness based on folding stability and simulations of protein evolution under structurally constrained substitution models integrated with birth-death population genetics approaches. The use of simulation-based approaches for prediction is well established in population genetics. For example, approaches such as approximate Bayesian computation (Beaumont, et al. 2002) rely on this strategy, and it has also been applied in other studies of forecasting evolution (e.g., Neher, et al. 2014). We believe that the distinction between forecasting folding stability and amino acid sequence is clearly shown in the manuscript, including the main text and the figures.

Reviewer #2 (Recommendations for the authors): 

I thank the authors for addressing the question about template switching, their clarification was helpful. However, the core concerns I raised remain unresolved: the claim that the method is useful for forecasting is not substantiated.  In order to support the paper's central claims or to prove its usefulness, several key improvements could be incorporated: 

(1) Systematic analysis of more proteins: 

The manuscript would be significantly strengthened by a systematic evaluation of model performance across a broader set of viral proteins, beyond the examples currently shown. Many human influenza and SARS-CoV-2 proteins have wellcharacterized structures or high-quality homology templates, making them suitable candidates. In the light of limited success of the method, presenting the model's behavior across a more comprehensive protein set, including those with varying structural constraints and immune pressures, would help assess generalizability and clarify the specific conditions under which the model is applicable. 

Following a comment from the reviewer in a previous revision of the study, we included the analysis of an influenza NS1 protein dataset that contains two evaluation time points. Next, to validate the prediction method, it is necessary to have monitored protein sequences collected at least at two consecutive time points, with sufficient divergence between them to capture evolutionary signatures that allow for proper evaluation. Additionally, many data involve sequences that are not consecutive in evolutionary terms (one dataset is not a direct ancestor of another dataset existing at a posterior time point), which disallows their applicability in this study. Little data from monitored protein evolution with trustable consecutive (ancestor-descendant) samples exist. The most suitable studies often involve in vitro virus evolution, but they usually show low genetic variability between samples collected at different time points. Although thousands of sequences are available for some viruses, they are usually consensus sequences from bulk sequencing and often show a low number of nonsynonymous mutations at the study protein-coding gene between time points. For example, from the original Wuhan strain and the Omicron variant, the SARS-CoV-2 proteins Mpro and PLpro accumulated only 10 and 22 amino acid changes, respectively. Analyzing intermediate variants of concern (i.e., Gamma or Delta) would reduce this number affecting statistics. Thus, in practice, we found scarcity of data derived from monitoring protein evolution, with trustable ancestor and corresponding descendant data at consecutive time points and with sufficient molecular diversity between them (i.e., at least more than five polymorphic amino acid sites). In all, we believe that the diverse viral protein datasets used in the present study, along with the multiple analyzed datasets collected from monitored HIV-1 populations present in different patients, provide a representative application of the method, since notice that similar patterns were generally generated from the analysis of the different datasets.

(2) Present clear data statistics: For each analyzed dataset, the authors should provide basic information about the number of unique sequences, levels of variability, and evolutionary divergence between start and end sequences. This would contextualize the forecasting task and clarify whether the simulations are non-trivial. In particular, it should be shown that the consensus sequence is indeed representative of the viral population at a given time point. In viral evolution we frequently observe co-circulation of subclades and the consensus sequence is then not representative. 

For each dataset analyzed, the manuscript provides the sequence identity between samples at the study time points (which also informs about sequence variability), sample sizes, representative protein structure, and other technical details. The study assumes that consensus sequences, typically generated by bulk sequencing, are representative of the viral population. Next, samples at different time points should involve ancestor-descendant relationships, which is a requirement and one of the limitations to find appropriate data for this study, as noted in our previous response.

(3) Explore other metrics for population level sequence comparison: 

In the light of possible existence of subclades, mentioned above, the currently used metrics for sequence comparison may underestimate performance of the simulations. It would be sufficient to see some overlap of simulated clades and and the observed clades. 

We found this to be a good idea. However, in practice, we believe that the criteria used to define subclades could introduce biases into the results. For some metrics, we evaluated the accuracy of the predictions through direct comparisons between all real and predicted protein variants, using percentages to facilitate interpretation. We believe that using subclades could potentially reduce the current prediction errors, but this would complicate the interpretation of the results, as they would be influenced by the subjective criteria used to define the subclades.

Currently, the manuscript presents a plausible filtering framework rather than a predictive model. Without these additional analyses, the main claims remain only partially supported. 

Please see our reply to the comment of the reviewer just before the section titled “Recommendations for the authors”.

Response to some rebuttal statements: 

(1) "Sequence comparisons based on the KL divergence require, at the studied time point, an observed distribution of amino acid frequencies among sites and an estimated distribution of amino acid frequencies among sites. In the study datasets, this is only the case for the HIV-1 MA dataset, which belongs to a previous study from one of us and collaborators where we obtained at least 20 independent sequences at each sampling point (Arenas, et al. 2016)" 

The available Influenza and SARS-CoV-2 data gathers isolates annotated with exact collection dates, providing reach datasets for such analysis. 

The available influenza and SARS-CoV-2 sequences are typically derived from bulk sequencing and, therefore, they are consensus sequences. As a result, they cannot be used to calculate KL divergence. Additionally, many of the indicated sequences from databases are not demonstrated to be consecutive in evolutionary terms (one dataset is not a direct ancestor of another dataset existing at a posterior time point), which disallows their applicability in this study. The most suitable studies often involve in vitro virus evolution, but they usually show low genetic variability between samples collected at different time points.

(2) "Regarding extending the analysis to other time points (other variants of concern), we kindly disagree because Omicron is the variant of concern with the highest genetic distance to the Wuhan variant, and a high genetic distance is  required to properly evaluate the prediction method." 

There have been many more variants of concern subsequent to Omicron which circulated in 2021. 

A key aspect is the accumulation of diversity in the study proteins across different time points. The SARS-CoV-2 proteins Mpro and PLpro accumulated only 10 and 22 amino acid changes from the original Wuhan variant to Omicron, respectively.

Analyzing intermediate variants of concern (e.g., Gamma or Delta) or those closely related to Omicron would reduce the number of accumulated mutations even further.   

We want to thank the reviewer again for taking the time to revise our work and for the insightful and helpful comments.

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review): 

Summary: 

Ferreiro et al. present a method to simulate protein sequence evolution under a birth-death model where sequence evolution is constrained by structural constraints on protein stability. The authors then use this model to explore the predictability of sequence evolution in several viral structural proteins. In principle, this work is of great interest to molecular evolution and phylodynamics, which have struggled to couple non-neutral models of sequence evolution to phylodynamic models like birth-death. Unfortunately, though, the model shows little improvement over neutral models in predicting protein evolution, and this ultimately appears to be due to fundamental conceptual problems with how fitness is modeled and linked to the phylodynamic birth-death model. 

AU: We thank the reviewer for the positive comments about our work.

Regarding predictive power, the study showed a good accuracy in predicting the real folding stability of forecasted protein variants under a selection model, but not under a neutral model. Next, predicting the exact sequences was more challenging. In this revised version, where we added additional real data, we found that the accuracy of this prediction can vary among proteins (i.e., the SCS model was more accurate than the neutral model in predicting sequences of the influenza NS1 protein at different time points). Still, we consider that efforts are required in the field of substitution models of molecular evolution. For example, amino acids with similar physicochemical properties can result in predictions with appropriate folding stability while different specific sequence. The development of accurate substitution models of molecular evolution is an active area of research with ongoing progress, but further efforts are still needed. Next, forecasting the folding stability of future real proteins is fundamental for proper forecasting protein evolution, given the essential role of folding stability in protein function and its variety of applications. Regarding the conceptual concerns related to fitness modeling, we clarify them in detail in our responses to the specific comments below.

Major concerns:

(1) Fitness model: All lineages have the same growth rate r = b-d because the authors assume b+d=1. But under a birth-death model, the growth r is equivalent to fitness, so this is essentially assuming all lineages have the same absolute fitness since increases in reproductive fitness (b) will simply trade off with decreases in survival (d). Thus, even if the SCS model constrains sequence evolution, the birthdeath model does not really allow for non-neutral evolution such that mutations can feed back and alter the structure of the phylogeny. 

We thank the reviewer for this comment that aims to improve the realism of our model. In the model presented (but see later another model, derived from the proposal of the reviewer, that we have now implemented into the framework and applied it to the study data), the fitness predicted from a protein variant is used to obtain the corresponding birth rate of that variant. In this way, protein variants with high fitness have high birth rates leading to overall more birth events, while protein variants with low fitness have low birth rates resulting in overall more extinction events, which has biological meaning for the study system. The statement “All lineages have the same growth rate r = b-d” in our model is incorrect because, in our model, b and d can vary among lineages according to the fitness. For example, a lineage might have b=0.9, d=0.1, r=0.8, while another lineage could have b=0.6, d=0.4, r=0.2. Indeed, the statement “this is essentially assuming all lineages have the same absolute fitness” is incorrect. Clearly, assuming that all lineages have the same fitness would not make sense, in that situation the folding stability of the forecasted protein variants would be similar under any model, which is not the case as shown in the results. In our model, the fitness affects the reproductive success, where protein variants with a high fitness have higher birth rates leading to more birth events, while those with lower fitness have higher death rates leading to more extinction events. This parameterization is meaningful for protein evolution because the fitness of a protein variant can affect its survival (birth or extinction) without necessarily affecting its rate of evolution. While faster growth rate can sometimes be associated with higher fitness, a variant with high fitness does not necessarily accumulate substitutions at a faster rate. Regarding the phylogenetic structure, the model presented considers variable birth and death events across different lineages according to the fitness of the corresponding protein variants, and this affects the derived phylogeny (i.e., protein variants selected against can go extinct while others with high fitness can produce descendants). We are not sure about the meaning of the term “mutations can feed back” in the context of our system. Note that we use Markov models of evolution, which are well-stablished in the field (despite their limitations), and substitutions are fixed mutations, which still could be reverted later if selected by the substitution model (Yang 2006). Altogether, we find that the presented birth-death model is technically correct and appropriate for modeling our biological system. Its integration with structurally constrained substitution (SCS) models of protein evolution as Markov models follows general approaches of molecular evolution in population genetics (Yang 2006; Carvajal-Rodriguez 2010; Arenas 2012; Hoban, et al. 2012). We have now provided a more detailed description of the models in the manuscript.

Apart from these clarifications about the birth-death model used, we could understand the point of the reviewer and following the suggestion we have now incorporated an additional birth-death model that accounts for variable global birth-death rate among lineages. Specifically, we followed the model proposed by Neher et al (2014), where the death rate is considered as 1 and the birth rate is modeled as 1 + fitness. In this model, the global birth-death rate can vary among lineages. We implemented this model into the computer framework and applied it to the data used for the evaluation of the models. The results indicated that, in general, this model yields similar predictive accuracy compared to the previous birth-death model. Thus, accounting for variability in the global birth-death rate does not appear to play a major role in the studied systems of protein evolution. We have now presented this additional birth-death model and its results in the manuscript.

(2) Predictive performance: Similar performance in predicting amino acid frequencies is observed under both the SCS model and the neutral model. I suspect that this rather disappointing result owes to the fact that the absolute fitness of different viral variants could not actually change during the simulations (see comment #1). 

As indicated in our previous answer, our study shows a good accuracy in predicting the real folding stability of forecasted protein variants under a selection model, but not under a neutral model. Next, predicting the exact sequences was more challenging, which was not surprising considering previous studies. In particular, inferring specific sequences is considerably challenging even for ancestral molecular reconstruction (Arenas, et al. 2017; Arenas and Bastolla 2020). Indeed, observed sequence diversity is much greater than observed structural diversity (Illergard, et al. 2009; Pascual-Garcia, et al. 2010), and substitutions between amino acids with similar physicochemical properties can yield modeled protein variants with more accurate folding stability, even when the exact amino acid sequences differ. As indicated, further work is demanded in the field of substitution models of molecular evolution. Next, in this revised version, where we included analyses of additional real datasets, we found that the accuracy of sequence prediction can vary among datasets. Notably, the analysis of an influenza NS1 protein dataset, with higher diversity than the other datasets studied, showed that the SCS model was more accurate than the neutral model in predicting sequences across different time points. Datasets with relatively high sequence diversity can contain more evolutionary information, which can improve prediction quality. In any case, as previously indicated, we believe that efforts are required in the field of substitution models of molecular evolution. Apart from that, forecasting the folding stability of future real proteins is an important advance in forecasting protein evolution, given the essential role of folding stability in protein function (Scheiblhofer, et al. 2017; Bloom and Neher 2023) and its variety of applications.

Next, also as indicated in our previous response, the birth-death model used in this study accounts for variation in fitness among lineages producing variable reproductive success. The additional birth-death model that we have now incorporated, which considers variation of the global birth-death rate among lineages, produced similar prediction accuracy, suggesting a limited role in protein evolution modeling. Molecular evolution parameters, particularly the substitution model, appear to be more critical in this regard. We have now included these aspects in the manuscript.

(3) Model assessment: It would be interesting to know how much the predictions were informed by the structurally constrained sequence evolution model versus the birth-death model. To explore this, the authors could consider three different models: 1) neutral, 2) SCS, and 3) SCS + BD. Simulations under the SCS model could be performed by simulating molecular evolution along just one hypothetical lineage. Seeing if the SCS + BD model improves over the SCS model alone would be another way of testing whether mutations could actually impact the evolutionary dynamics of lineages in the phylogeny. 

In the present study, we compared the neutral model + birth-death (BD) with the SCS model + BD. Markov substitution models Q are applied upon an evolutionary time (i.e., branch length, t) and this allows to determine the probability of substitution events during that time period [P(t) = exp (Qt)]. This approach is traditionally used in phylogenetics to model the incorporation of substitution events over time. Therefore, to compare the neutral and SCS models in terms of evolutionary inference, an evolutionary time is required, in this case it is provided by the birth-death process. Thus, the cases 1) and 2) cannot be compared without an underlined evolutionary history. Next, comparisons in terms of likelihood, and other aspects, between models that ignore the protein structure and the implemented SCS models are already available in previous studies based on coalescent simulations or given phylogenetic trees (Arenas, et al. 2013; Arenas, et al. 2015). There, SCS models outperformed models that ignore evolutionary constraints from the protein structure, and those findings are consistent with the results obtained in the present study where we explored the application of these models to forecasting protein evolution. We would like to emphasize that forecasting the folding stability of future real proteins is a significant finding, folding stability is fundamental to protein function and has a variety of applications. We have now indicated these aspects in the manuscript.

(4) Background fitness effects: The model ignores background genetic variation in fitness. I think this is particularly important as the fitness effects of mutations in any one protein may be overshadowed by the fitness effects of mutations elsewhere in the genome. The model also ignores background changes in fitness due to the environment, but I acknowledge that might be beyond the scope of the current work. 

AU: This comment made us realize that more information about the features of the implemented SCS models should be included in the manuscript. In particular, the implemented SCS models consider a negative design based on the observed residue contacts in nearly all proteins available in the Protein Data Bank (Arenas, et al. 2013; Arenas, et al. 2015). This data is distributed with the framework, and it can be updated to incorporate new structures (further details are provided in the distributed framework documentation and practical examples). Therefore, the prediction of folding stability is a combination of positive design (direct analysis of the target protein) and negative design (consideration of background proteins from a database to improve the predictions), thus incorporating background molecular diversity. We have now indicated this important aspect in the manuscript. Regarding the fitness caused by the environment, we agree with the reviewer. This is a challenge for any method aiming to forecast evolution, as future environmental shifts are inherently unpredictable and may affect the accuracy of the predictions. Although one might attempt to incorporate such effects into the model, doing so risks overparameterization, especially when the additional factors are uncertain or speculative. We have now mentioned this aspect in the manuscript.

(5) In contrast to the model explored here, recent work on multi-type birth-death processes has considered models where lineages have type-specific birth and/or death rates and therefore also type-specific growth rates and fitness (Stadler and Bonhoeffer, 2013; Kunhert et al., 2017; Barido-Sottani, 2023). Rasmussen & Stadler (eLife, 2019) even consider a multi-type birth-death model where the fitness effects of multiple mutations in a protein or viral genome collectively determine the overall fitness of a lineage. The key difference with this work presented here is that these models allow lineages to have different growth rates and fitness, so these models truly allow for non-neutral evolutionary dynamics. It would appear the authors might need to adopt a similar approach to successfully predict protein evolution. 

We agree with the reviewer that robust birth-death models have been developed applying statistics and, in many cases, the primary aim of those studies is the development and refinement of the model itself. Regarding the study by Rasmussen and Stadler 2019, it incorporates an external evaluation of mutation events where the used fitness is specific for the proteins investigated in that study, which may pose challenges for users interested in analyzing other proteins. In contrast, our study takes a different approach. We implement a fitness function that can be predicted and evaluated for any type of structural protein (Goldstein 2013), making it broadly applicable. Actually, in this revised version we added the analysis of additional data of another protein (influenza NS1 protein) with predictions at different time points. In addition, we provide a freely available and well-documented computational framework to facilitate its use. The primary aim of our study is not the development of novel or complex birthdeath models. Rather, we aim to explore the integration of a standard birth-death model with SCS models for the purpose of predicting protein evolution. In the context of protein evolution, substitution models are a critical factor (Liberles, et al. 2012; Wilke 2012; Bordner and Mittelmann 2013; Echave, et al. 2016; Arenas, et al. 2017; Echave and Wilke 2017), and the presented combination with a birth-death model constitutes a first approximation upon which next studies can build to better understand this evolutionary system. We have now indicated these considerations in the manuscript.

Reviewer #2 (Public review): 

Summary: 

In this study, "Forecasting protein evolution by integrating birth-death population models with structurally constrained substitution models", David Ferreiro and coauthors present a forward-in-time evolutionary simulation framework that integrates a birth-death population model with a fitness function based on protein folding stability. By incorporating structurally constrained substitution models and estimating fitness from ΔG values using homology-modeled structures, the authors aim to capture biophysically realistic evolutionary dynamics. The approach is implemented in a new version of their open-source software, ProteinEvolver2, and is applied to four viral proteins from HIV-1 and SARS-CoV-2. 

Overall, the study presents a compelling rationale for using folding stability as a constraint in evolutionary simulations and offers a novel framework and software to explore such dynamics. While the results are promising, particularly for predicting biophysical properties, the current analysis provides only partial evidence for true evolutionary forecasting, especially at the sequence level. The work offers a meaningful conceptual advance and a useful simulation tool, and sets the stage for more extensive validation in future studies.

We thank the reviewer for the positive comments on our study. Regarding the predictive power, the results showed good accuracy in predicting the folding stability of the forecasted protein variants. In this revised version, where we included analyses of additional real datasets, we found that the accuracy of sequence prediction can vary among datasets. Notably, the analysis of an influenza NS1 protein dataset, with higher diversity than the other datasets studied, showed that the SCS model was more accurate than the neutral model in predicting sequences across different time points. Datasets with relatively high sequence diversity can contain more evolutionary information, which can improve prediction quality. Still, we believe that further efforts are required in the field in improving the accuracy of substitution models of molecular evolution. Altogether, accurately forecasting the folding stability of future real proteins is fundamental for predicting their protein function and enabling a variety of applications. Also, we implemented the models into a freely available computer framework, with detailed documentation and a variety of practical examples.

Strengths: 

The results demonstrate that fitness constraints based on protein stability can prevent the emergence of unrealistic, destabilized variants - a limitation of traditional, neutral substitution models. In particular, the predicted folding stabilities of simulated protein variants closely match those observed in real variants, suggesting that the model captures relevant biophysical constraints. 

We agree with the reviewer and appreciate the consideration that forecasting the folding stability of future real proteins is a relevant finding. For instance, folding stability is fundamental for protein function and affects several other molecular properties.

Weaknesses: 

The predictive scope of the method remains limited. While the model effectively preserves folding stability, its ability to forecast specific sequence content is not well supported. 

Our study showed a good accuracy in predicting the real folding stability of forecasted protein variants under a selection model, but not under a neutral model. Predicting the exact sequences was more challenging, which was not surprising considering previous studies. In particular, inferring specific sequences is considerably challenging even for ancestral molecular reconstruction (Arenas, et al. 2017; Arenas and Bastolla 2020). Indeed, observed sequence diversity is much greater than observed structural diversity (Illergard, et al. 2009; Pascual-Garcia, et al. 2010), and substitutions between amino acids with similar physicochemical properties can yield modeled protein variants with more accurate folding stability, even when the exact amino acid sequences differ. As indicated, further work is demanded in the field of substitution models of molecular evolution. Next, in this revised version, where we included analyses of additional real datasets, we found that the accuracy of sequence prediction can vary among datasets. Notably, the analysis of an influenza NS1 protein dataset, with higher diversity than the other datasets studied, showed that the SCS model was more accurate than the neutral model in predicting sequences across different time points. Datasets with relatively high sequence diversity can contain more evolutionary information, which can improve prediction quality. In any case, as previously indicated, we believe that efforts are required in the field of substitution models of molecular evolution. Apart from that, forecasting the folding stability of future real proteins is an important advance in forecasting protein evolution, given the essential role of folding stability in protein function (Scheiblhofer, et al. 2017; Bloom and Neher 2023) and its variety of applications. We have now expanded these aspects in the manuscript.

Only one dataset (HIV-1 MA) is evaluated for sequence-level divergence using KL divergence; this analysis is absent for the other proteins. The authors use a consensus Omicron sequence as a representative endpoint for SARS-CoV-2, which overlooks the rich longitudinal sequence data available from GISAID. The use of just one consensus from a single time point is not fully justified, given the extensive temporal and geographical sampling available. Extending the analysis to include multiple timepoints, particularly for SARS-CoV-2, would strengthen the predictive claims. Similarly, applying the model to other well-sampled viral proteins, such as those from influenza or RSV, would broaden its relevance and test its generalizability. 

The evaluation of forecasting evolution using real datasets is complex due to several conceptual and practical aspects. In contrast to traditional phylogenetic reconstruction of past evolutionary events and ancestral sequences, forecasting evolution often begins with a variant that is evolved forward in time and requires a rough fitness landscape to select among possible future variants (Lässig, et al. 2017). Another concern for validating the method is the need to know the initial variant that gives rise to the corresponding future (forecasted) variants, and it is not always known. Thus, we investigated systems where the initial variant, or a close approximation, is known, such as scenarios of in vitro monitored evolution. In the case of SARS-CoV-2, the Wuhan variant is commonly used as the starting variant of the pandemic. Next, since forecasting evolution is highly dependent on the used model of evolution, unexpected external factors can be dramatic for the predictions. For this reason, systems with minimal external influences provide a more controlled context for evaluating forecasting evolution. For instance, scenarios of in vitro monitored virus evolution avoid some external factors such as host immune responses. Another important aspect is the availability of data at two (i.e., present and future) or more time points along the evolutionary trajectory, with sufficient genetic diversity between them to identify clear evolutionary signatures. Additionally, using consensus sequences can help mitigate effects from unfixed mutations, which should not be modeled by a substitution model of evolution. Altogether, not all datasets are appropriate to properly evaluate or apply forecasting evolution. These aspects are indicated in the manuscript. Sequence comparisons based on the KL divergence require, at the studied time point, an observed distribution of amino acid frequencies among sites and an estimated distribution of amino acid frequencies among sites. In the study datasets, this is only the case for the HIV-1 MA dataset, which belongs to a previous study from one of us and collaborators where we obtained at least 20 independent sequences at each sampling point (Arenas, et al. 2016). This aspect is now more clearly indicated in the manuscript. Regarding the Omicron datasets, we used 384 curated sequences of the Omicron variant of concern to construct the study data and we believe that it is a representative sample. The sequence used for the initial time point was the Wuhan variant (Wu, et al. 2020), which is commonly assumed to be the origin of the pandemic in SARS-CoV-2 studies. As previously indicated, the use of consensus sequences is convenient to avoid variants with unfixed mutations. Regarding extending the analysis to other time points (other variants of concern), we kindly disagree because Omicron is the variant of concern with the highest genetic distance to the Wuhan variant, and a high genetic distance is required to properly evaluate the prediction method. Actually, we noted that earlier variants of concern show a small number of fixed mutations in the study proteins, despite the availability of large numbers of sequences in databases such as GISAID. Additionally, we investigated the evolutionary trajectories of HIV-1 protease (PR) in 12 intra-host viral populations with predictions for up to four different time points. Apart from those aspects, following the proposal of the reviewer, we have now incorporated the analysis of an additional dataset of influenza NS1 protein (Bao, et al. 2008), with predictions for two different time points, to further assess the generalizability of the method. We have now included details of this influenza NS1 protein dataset and the predictions derived from it in the manuscript.

It would also be informative to include a retrospective analysis of the evolution of protein stability along known historical trajectories. This would allow the authors to assess whether folding stability is indeed preserved in real-world evolution, as assumed in their model.

Our present study does not aim to investigate the evolution of the folding stability over time, although it provides this information indirectly at the studied time points. Instead, the present study shows that the folding stability of the forecasted protein variants is similar to the folding stability of the corresponding real protein variants for diverse viral proteins, which provides an important evaluation of the prediction method. Next, the folding stability can indeed vary over time in both real and modeled evolutionary scenarios, and our present study is not in conflict with this. In that regard, which is not the aim of our present study, some previous phylogenetic-based studies have reported temporal fluctuations in folding stability for diverse protein data (Arenas, et al. 2017; Olabode, et al. 2017; Arenas and Bastolla 2020; Ferreiro, et al. 2022).

Finally, a discussion on the impact of structural templates - and whether the fixed template remains valid across divergent sequences - would be valuable. Addressing the possibility of structural remodeling or template switching during evolution would improve confidence in the model's applicability to more divergent evolutionary scenarios.

This is an important point. For the datasets that required homology modeling (in several cases it was not necessary because the sequence was present in a protein structure of the PDB), the structural templates were selected using SWISS-MODEL, and we applied the best-fitting template. We have now included in a supplementary table details about the fitting of the structural templates. Indeed, our proposal assumes that the protein structure is maintained over the studied evolutionary time, which can be generally reasonable for short timescales where the structure is conserved (Illergard, et al. 2009; Pascual-Garcia, et al. 2010). Over longer evolutionary timescales, structural changes may occur and, in such cases, modeling the evolution of the protein structure would be necessary. To our knowledge, modeling the evolution of the protein structure remains a challenging task that requires substantial methodological developments. Recent advances in artificial intelligence, particularly in protein structure prediction from sequence, may offer promising tools for addressing this challenge. However, we believe that evaluating such approaches in the context of structural evolution would be difficult, especially given the limited availability of real data with known evolutionary trajectories involving structural change. In any case, this is probably an important direction for future research. We have now included this discussion in the manuscript.

Reviewer #1 (Recommendations for the authors): 

(1) Abstract: "expectedly, the errors grew up in the prediction of the corresponding sequences" <- Not entirely clear what is meant by "errors grew up" or what the errors grew with.

This sentence refers to the accuracy of sequence prediction in comparison to that of folding stability prediction. We have now clarified this aspect in the manuscript.

(2) Lines 162-165: "Alternatively, if the fitness is determined based on the similarity in folding stability between the modeled variant and a real variant, the birth rate is assumed to be 1 minus the root mean square deviation (RMSD) in folding stability." <- What is the biological motivation for using the RMSD? It seems like a more stable variant would always have higher fitness, at least according to Equation 1.

RMSD is commonly used in molecular biology to compare proteins in terms of structural distance, folding stability, kinetics, and other properties. It offers advantages such as minimizing the influence of small deviations while amplifying larger differences, thereby enhancing the detection of remarkable molecular changes. Additionally, RMSD would facilitate the incorporation of other biophysical parameters, such as structural divergences from a wild-type variant or entropy, which could be informative for fitness in future versions of the method. We have now included this consideration in the manuscript.

(3) Lines 165-166: "In both cases, the death rate (d) is considered as 1-b to allow a constant global (birth-death) rate" <- This would give a constant R = b / (1-b) over the entire phylogenetic tree. For applications to pathogens like viruses with epidemic dynamics, this is extremely implausible. Is there any need to make such a restrictive assumption? 

Regarding technical considerations of the model, we refer to our answer to the first public review comment. Next, a constant global rate of evolution was observed in numerous genes and proteins of diverse organisms, including viruses (Gojobori, et al.1990; Leitner and Albert 1999; Shankarappa, et al. 1999; Liu, et al. 2004; Lu, et al. 2018; Zhou, et al. 2019). However, following the comment of the reviewer, and as we indicated in our answer to the first public review comment, we have now implemented and evaluated an additional birth-death model that allows for variation in the global birth-death rate among lineages. We have implemented this additional model in the framework and described it along with its results in the manuscript.

(4) Lines 187-188: "As a consequence, since b+d=1 at each node, tn is consistent across all nodes, according to (Harmon, 2019)." <- This would also imply that all lineages have a growth rate r = b - d, which under a birth-death model is equivalent to saying all lineages have the same fitness! 

We clarified this aspect in our answer to the first public review comment. In particular, in the model presented, protein variants with higher fitness have higher birth rates, leading to more birth events, while protein variants with lower fitness have lower birth rates leading to more extinction events, which presents biological meaning for the study system. In our model b and d can vary among lineages according to the corresponding fitness (i.e., a lineage may have b=0.9, d=0.1, r=0.8; while another one may have b=0.6, d=0.4, r=0.2). Since the reproductive success varies among lineages in our model, the statement “this is essentially assuming all lineages have the same absolute fitness” is incorrect, although it could be interpreted like that in certain models. Fitness affects reproductive success, but fitness and growth rate of evolution are different biological processes (despite a faster growth rate can sometimes be associated with higher fitness, a variant with a high fitness not necessarily has to accumulate substitutions at a higher rate). An example in molecular adaptation studies is the traditional nonsynonymous to synonymous substitution rates ratio (dN/dS), where dN/dS (that informs about selection derived from fitness) can be constant at different rates of evolution (dN and dS). In any case, we thank the reviewer for raising this point, which led us to incorporate an additional birth-death model and inspired some ideas.  Thus, following the comment of the reviewer and as indicated in the answer to the first public review comment, we have now implemented and evaluated an additional birthdeath model that allows for variation in the global birth-death rate among lineages. The results indicated that this model yields similar predictive accuracy compared to the previous birth-death model. We have now included these aspects, along with the results from the additional model, in the manuscript.

(5) Line 321-322: "For the case of neutral evolution, all protein variants equally fit and are allowed, leading to only birth events," <- Why would there only be birth events? Lineages can die regardless of their fitness. 

AU: In the neutral evolution model, all protein variants have the same fitness, resulting in a flat fitness landscape. Since variants are observed, we allowed birth events. However, it assumed the absence of death events as no information independent of fitness is available to support their inclusion and quantification, thereby avoiding the imposition of arbitrary death events based on an arbitrary death rate. We have now provided a justification of this assumption in the manuscript.

Reviewer #2 (Recommendations for the authors): 

(1) Clarify the purpose of the alternative fitness mode ("ΔG similarity to a target variant"): 

The manuscript briefly introduces an alternative fitness function based on the similarity of a simulated protein's folding stability to that of a real protein variant, but does not provide a clear motivation, usage scenario, or results derived from it. 

The presented model provides two approaches for deriving fitness from predicted folding stability. The simpler approach assumes that a more stable protein variant has higher fitness than a less stable one. The alternative approach assigns high fitness to protein variants whose stability closely matches observed stability, acknowledging that the real observed stability is derived from the real selection process, and this approach considers negative design by contrasting the prediction with real information. For the analyses of real data in this study, we used the second approach, guided by these considerations. We have now clarified this aspect in the manuscript.

(2) Report structural template quality and modeling confidence: 

Since folding stability (ΔG) estimates rely on structural models derived from homology templates, the accuracy of these predictions will be sensitive to the choice and quality of the template structure. I recommend that the authors report, for each protein modeled, the template's sequence identity, coverage, and modeling quality scores. This will help readers assess the confidence in the ΔG estimates and interpret how template quality might impact simulation outcomes. 

We agree with the reviewer and we have now included additional information in a supplementary table regarding sequence identity, modeling quality and coverage of the structural templates for the proteins that required homology modeling. The selection of templates was performed using the well-established framework SWISS-MODEL and the best-fitting template was chosen. Next, a large number of protein structures are available in the PDB for the study proteins, which favors the accuracy of the homology modeling. For some datasets, homology modeling was not required, as the modeled sequence was already present in an available protein structure. We have now included this information in the manuscript and in a supplementary table.

(3) Clarify whether structural remodeling occurs during simulation: 

It appears that folding stability (ΔG) for all simulated protein variants is computed by mapping them onto a single initial homology model, without remodeling the structure as sequences evolve. If correct, this should be clearly stated, as it assumes that the structural fold remains valid across all simulated variants. A discussion on the potential impact of structural drift would be welcome.

We agree with the reviewer. As indicated in our answer to a previous comment, our method assumes that the protein structure is maintained over the studied evolutionary time, which is generally acceptable for short timescales where the structure is conserved (Illergard, et al. 2009; Pascual-Garcia, et al. 2010). At longer timescales the protein structure could change, requiring the modeling of structural evolution over the evolutionary time. To our knowledge, modeling the evolution of the protein structure remains a challenging task that requires substantial methodological developments. Recent advances in artificial intelligence, particularly in protein structure prediction from sequence, can be promising tools for addressing this challenge. However, we believe that evaluating such approaches in the context of structural evolution would be difficult, especially given the limited availability of real datasets with known evolutionary trajectories involving structural change. In any case, this is probably an important direction for future research. We have now included this discussion in the manuscript.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

Summary:

Lysosomal damage is commonly found in many diseases including normal aging and age-related disease. However, the transcriptional programs activated by lysosomal damage have not been thoroughly characterized. This study aimed to investigate lysosome damage-induced major transcriptional responses and the underlying signaling basis. The authors have convincingly shown that lysosomal damage activates a ubiquitination-dependent signaling axis involving TAB, TAK1, and IKK, which culminates in the activation of NF-kB and subsequent transcriptional upregulation of pro-inflammatory genes and pro-survival genes. Overall, the major aims of this study were successfully achieved.

Strengths:

This study is well-conceived and strictly executed, leading to clear and well-supported conclusions. Through unbiased transcriptomics and proteomics screens, the authors identified NF-kB as a major transcriptional program activated upon lysosome damage. TAK1 activation by lysosome damage-induced ubiquitination was found to be essential for NF-kB activation and MAP kinase signaling. The transcriptional and proteomic changes were shown to be largely driven by TAK1 signaling. Finally, the TAK1-IKK signaling was shown to provide resistance to apoptosis during lysosomal damage response. The main signaling axis of this pathway was convincingly demonstrated.

Weaknesses:

One weakness was the claim of K63-linked ubiquitination in lysosomal damage-induced NF-kB activation. While it was clear that K63 ubiquitin chains were present on damaged lysosomes, no evidence was shown in the current study to demonstrate the specific requirement of K63 ubiquitin chains in the signaling axis being studied. Clarifying the roles of K63-linked versus other types of ubiquitin chains in lysosomal damage-induced NF-kB activation may improve the mechanistic insights and overall impact of this study.

Another weakness was that the main conclusions of this study were all dependent on an artificial lysosomal damage agent. It will be beneficial to confirm key findings in other contexts involving lysosomal damage.

We would like to thank Reviewer #1 for the positive and constructive comments on our study. For a main concern regarding the molecular mechanism by which TAB proteins are activated in response to lysosomal damage, we have added the experimental results to support that the lysosomal accumulation of K63 ubiquitin chains serves as a trigger to activate the TAB-TAK1 pathway. We also investigated and discussed the role of LUBAC-mediated M1 ubiquitin chains in NF-kB activation and the effects of other lysosomal-damaging compounds. Please see the response to “Reviewer #3 (Public review): Suggestions:”.

Reviewer #2 (Public review):

Summary:

Endo et al. investigate the novel role of ubiquitin response upon lysosomal damage in activating cellular signaling for cell survival. The authors provide a comprehensive transcriptome and proteome analysis of aging-related cells experiencing lysosomal damage, identifying transcription factors involved in transcriptome and proteome remodeling with a focus on the NF-κB signaling pathway. They further characterized the K63-ubiquitin-TAB-TAK1-NF-κB signaling axis in controlling gene expression, inflammatory responses, and apoptotic processes.

Strengths:

In the aging-related model, the authors provide a comprehensive transcriptome and characterize the K63-ubiquitin-TAB-TAK1-NF-κB signaling axis. Through compelling experiments and advanced tools, they elucidate its critical role in controlling gene expression, inflammatory responses, and apoptotic processes.

Weaknesses:

The study lacks deeper connections with previous research, particularly:

• The established role of TAB-TAK1 in AMPK activation during lysosomal damage

• The potential significance of TBK1 in NF-κB signaling pathways

We would like to thank Reviewer #2 for the helpful comments on our study. To achieve a more comprehensive understanding of the signaling pathways involved in the lysosomal damage response, we investigated additional related signal mediators, such as TBK1 and LUBAC. The citations related to AMPK have been incorporated.

Reviewer #3 (Public review):

Summary:

The response to lysosomal damage is a fast-moving and timely field. Besides repair and degradation pathways, increasing interest has been focusing on damaged-induced signaling. The authors conducted both transcriptomics and proteomics to characterize the cellular response to lysosomal damage. They identify a signaling pathway leading to activation of NFkappaB. Based on this and supported by Western blot and microscopy data, the authors nicely show that TAB2/3 and TAK1 are activated at damaged lysosomes and kick off the pathway to alter gene expression, which induces cytokines and protect from cell death. TAB2/3 activation is proposed to occur through K63 ubiquitin chain formation. Generally, this is a careful and well conducted study that nicely delineates the pathway under lysosomal stress. The "omics" data serves as a valuable resource for the field. More work should be invested into how TAB2/3 are activated at the damaged lysosomes, also to increase novelty in light of previous reports.

Strengths:

Generally, this is a careful and well-conducted study that nicely delineates the pathway under lysosomal stress. The "omics" data serves as a valuable resource for the field.

Weaknesses:

More work should be invested into how TAB2/3 are activated at the damaged lysosomes, also to increase novelty in light of previous reports. Moreover, different damage types should be tested to probe relevance for different pathophysiological conditions.

We would like to thank Reviewer #3 for the valuable comments on our study. We have added the experimental results to address two concerns raised by Reviewer #3. Please see the response to “Reviewer #3 (Public review): Suggestions:”.

Suggestions:

(1) A recent paper claims that NFkappaB is activated by Otulin/M1 chains upon lysosome damage through TBK1 (PMID: 39744815). In contrast, Endo et al. nicely show that ubiquitylation is needed (shown by TAK-243) for NFkB activation but only have correlative data to link it specifically to K63 chains. On page 15, line 11, the authors even argue a "potential" involvement of K63. This point should be better dealt with. Can the authors specifically block K63 formation? K63R overexpression or swapping would be one way. Is the K63 ligase ITCH involved (PMID: 38503285) or any other NEDD4-like ligase? This could be compared to LUBAC inhibition. Also, the point needs to be dealt with more controversially in the discussion as these are alternative claims (M1 vs K63, TAB vs TBK1).

It is well-characterized that the NZF domain of TAB proteins preferentially associates with K63-linked ubiquitin chains. Therefore, we performed the add-back experiment using siRNA-resistant TAB2 WT and mutants incapable of binding to K63-linked ubiquitin chains, dNZF and E685A, to elucidate the requirement of K63 ubiquitin chains for TAK1 activation. We investigated whether the add-back of TAB2 mutants rescues the activation of TAK1 in TAB2-depleted cells (Fig. 2E). TAB2 WT, but not dNZF and E685A, rescued TAK1 activation in response to LLOMe, suggesting that the specific interaction of TAB proteins and K63 ubiquitin chains is a key mechanism to activate TAK1. We also found that the treatment of an E1 inhibitor TAK-243 effectively prevented the lysosomal accumulation of K63 ubiquitin chains, but TAB2 was recruited to damaged lysosomes (Fig. S2B). This suggests that the recruitment of TAB proteins to damaged lysosomes is independent of the association with K63 ubiquitin chains. Collectively, it is postulated that TAB proteins require interaction with K63 ubiquitin chains for TAK1 activation, but not for recruitment to damaged lysosomes. We have added the sentences (p9, lines 7-20, and p10, lines 8-10).

Next, we confirmed that LUBAC functions are essential for NF-kB activation in the lysosomal damage response. RNF31/HOIP is a component of LUBAC that catalyzes M1 ubiquitination. The depletion of RNF31 showed no significant effects on TAK1 activation, but abolished IKK activation (Fig. S4G). It is well-characterized that LUBAC-mediated M1 ubiquitin chains recruit IKK subunits and transduce the signaling to downstream in the canonical pathway. We assume that K63 ubiquitin chains in damaged lysosomes initially activate TAB-TAK1 and trigger LUBAC-mediated M1 ubiquitination, and subsequently, M1 ubiquitination functions to recruit the IKK complex. Consequently, activated TAK1 phosphorylates IKK subunits in damaged lysosomes, leading to NF-kB activation. We also examined whether TBK1 is involved in the activation of NF-kB. TBK1 was phosphorylated upon LLOMe, and depletion of TAB and TAK1 resulted in a slight reduction of TBK1 phosphorylation (Fig. S4D, E). The treatment of a TBK1 inhibitor BX-795 exhibited no or little effects on TAK1 activation, but abolished phosphorylation of IKK and IkBa (Fig. S4F). These suggest that TBK1 is required for the activation of NF-kB. We have added the sentences (p13, line 13-p14, line 10).

As mentioned by Reviewer #3, it is important to identify the E3 ligase responsible for K63 ubiquitination in the lysosomal damage response. We have been aiming to identify such E3 ligase(s). However, depletions of ITCH and other E3 ligases that have been tested exhibited no or little effects on K63 ubiquitination and TAK1 activation.  We would like to explore E3 ligase(s) in future study.

(2) It would be interesting to know what the trigger is that induces the pathway. Lipid perturbation by LLOMe is a good model, but does activation also occur with GPN (osmotic swelling) or lipid peroxidation (oxidative stress) that may be more broadly relevant in a pathophysiological way? Moreover, what damage threshold is needed? Does loss of protons suffice? Can activation be induced with a Ca2+ agonist in the absence of damage?

To further clarify the initial trigger that induces TAB-TAK1 activation coupled with lysosomal damage, we examined other damage sources, GPN and DC661, which induce hyperosmotic stress and lipid peroxidation in lysosomes, respectively, thereby resulting in lysosomal membrane damage. Under our experimental conditions, the treatment of these compounds did not result in significant accumulation of Gal-3, indicating a reduced level of lysosomal membrane permeabilization compared with LLOMe (Fig. S2C, D), and no or little TAK1 activation was observed (Fig. S2E). TAB proteins require their association with K63 ubiquitin chains for TAK1 activation. It is therefore postulated that the severe lysosomal membrane permeabilization that triggers the formation and cytosolic exposure of K63 ubiquitin chains may be a determinant of TAB-TAK1 activation. In our future work, we would like to examine broad stimulation of lysosomal damage and further elucidate the initial mechanism of TAB-TAK1 activation. We have added the sentences (p9, line 21-p10, line 7).

(3) The authors nicely define JNK and p38 activation. This should be emphasized more, possibly also in the abstract, as it may contribute to the claim of increased survival fitness.

We further tested whether the inhibition of JNK affects the anti-apoptotic effect (Fig. S5B). The inhibition of JNK resulted in an increase in the cleaved caspase-3. This suggests that the anti-apoptotic action in the lysosomal damage response requires JNK as well as IKK. We have added the sentences in results to emphasize the pivotal role of stress-induced MAPKs (p15, lines 7-11).

Reviewer #1 (Recommendations for the authors):

(1) Although the ubiquitination-TAB-TAK1-IKK axis was previously characterized in other contexts, specific evidence supporting lysosomal recruitment of these components by ubiquitination during lysosome damage would be beneficial.

We found that the treatment of an E1 inhibitor TAK-243 abolished the lysosomal accumulation of K63 ubiquitin chains, but TAB2 and TAK1 were recruited to damaged lysosomes (Fig. S2B). This suggests that the recruitment of TAB proteins to damaged lysosomes is independent of the association with K63-linked ubiquitin chains. Next, we investigated whether the add-back of TAB2 mutants incapable of binding K63 ubiquitin chains rescues the activation of TAK1 in TAB2-depleted cells (Fig. 2E). K63 ubiquitin binding of TAB2 was essential for TAK1 activation in response to LLOMe. Taken together, it is suggested that TAB proteins require their interaction with K63 ubiquitin chains for TAK1 activation, but not for recruitment to damaged lysosomes. We have added the sentences (p9, lines 7-20, and p10, lines 8-10). Please also see the response to “Reviewer #3 (Public review): Suggestions:”.

(2) The activation of p38 and JNK by lysosomal damage does not fit well into the main conclusions of the paper, since IKK knockdown was sufficient to block cellular resistance to apoptosis (caspase cleavage in Fig. 5f). Are p38 and JNK also important for cell survival during lysosomal damage?

We found that the inhibition of JNK resulted in an increase in the cleaved caspase-3, suggesting that the anti-apoptotic action in the lysosomal damage response requires both IKK and JNK (Fig. S5B). We have added the sentences (p15, lines 7-11).

(3) Cell death tests are recommended to support the conclusions related to apoptosis.

As suggested by Reviewer #1, we performed the cell death assay using propidium iodide (PI) and confirmed that HeLa cells co-treated with LLOMe and TAK-243 or HS-276 exhibited increased cell death (Fig. 5E). This indicates a direct correlation between the degree of caspase-3 cleavage and cell death, possibly apoptosis.

(4) Page 8, line 19-21, gal3 is not exposed upon lysosomal damage. It is recruited from the cytosol by the exposed beta-galactoside-containing glycans on lysosomal membrane proteins.

We have corrected the corresponding sentence (p7, lines 17-20).

(5) Carefully checking grammar throughout the text is recommended. Below are a few examples:

a) Page 4, line 10, remove "that".

b) "K63 ubiquitin" shall be replaced with "K63 ubiquitination" or "K63 ubiquitin chains".

c) Page 8, line 9, "remain" should be "remains".

We have carefully checked the revised manuscript.

Reviewer #2 (Recommendations for the authors):

Despite the novelty and significance of these findings in advancing the field, several technical and experimental limitations require further clarification:

We have responded to each comment. Please see below.

The manuscript should introduce or discuss previous research showing that TAB-TAK1 facilitates AMPK activation during lysosomal damage and TAK1's increased association with damaged lysosomes (PMID: 31995728).

We have added the reference (PMID: 31995728) and the sentences (p17, lines 15-20).

Figure 2A: The differential LAMP1 staining intensity between control and LLOMe-treated cells needs explanation. The weaker LAMP1 signal in control and puncta changes, especially during 5-minute LLOMe treatment, require detailed clarification

We have added the explanation (p8, lines 17-21).

Recent literature (PMID: 34585663) reports TBK1 activation during lysosomal damage. The authors should investigate or discuss whether TBK1 potentially contributes to NF-κB signaling in this context.

We experimentally investigated whether TBK1 is involved in the TAB-TAK1 pathway. We confirmed that TBK1 was activated upon LLOMe (Fig. S4D). Depletions of TAB and TAK1 exhibited a modest decrease in TBK1 phosphorylation (Fig. S4E). The inhibition of TBK1 by BX-795 did not affect TAK1 activation, but abolished phosphorylation of IKK and IkBa (Fig. S4F). This suggests that TBK1 is required for NF-kB activation. We have added the reference (PMID: 34585663) and the sentences (p13, lines 13-21, p14, lines 8-10, and p18, lines 15-20).

The introduction of lysosomal damage response lacks comprehensive mechanistic information. For example, while ESCRT is discussed, other critical mechanisms such as lipid transfer and stress granule formation in lysosomal repair should be incorporated. Moreover, mTOR and AMPK signaling pathways undergo significant changes upon lysosomal damage.

We have added the sentences (p3, lines 16-18, and p3, line 21-p4, line 1).

The statement "lysosomal permeabilization causes the dissociation of mTORC1 from lysosomes" should explicitly reference PMID: 29625033.

We have added the suggested reference (PMID: 29625033, p4, line 19).

The claim that "The elimination of damaged lysosomes through lysophagy requires a period of more than half a day" needs a specific publication citation.

We have added the reference (PMID: 23921551) to claim the time-scale of lysosomal clearance (p4, line 21).

Figure 1G: The label "WO after 2h" lacks explanation in the figure legend and requires detailed interpretation.

To simplify the figures, we have deleted the label “WO after 2 h” (Fig. 1G, 3F, 5D, F-J, S4G, S5A). Instead, we have added the explanation in the figure legends (Fig. 1G).

Reviewer #3 (Recommendations for the authors):

(1) page 8, line 13: it is recommended to phrase colocalisation "at" damaged lysosomes rather than "in" damaged lysosomes as the resolution does not allow the claim of influx into lysosomes.

We have corrected the word (p8, line 17).

(2) page 11, line 22: why is "whereas" used to link two events driven by the same mechanism.

We have corrected the word (p13, line 8).

Author response:

We thank the reviewers for their thoughtful and thorough consideration of the work. We appreciate the positive reception they give the work, and plan to address several of the comments with further experiments. To outline that work (and ensure that we are on the right track to addressing those concerns), we summarize the core concerns that prompt new experiments:

(1) Does the YFP tag on the ACRs interfere with simultaneous GCaMP imaging of RubyACR-expressing cells and could bleaching of the YFP complicate interpretation of the experiments here?

We will test whether 920 nm (2p) and 650 nm (1p) excitation cause YFP bleaching that interferes with interpretation of inhibitory calcium (i.e. GCaMP) signals. Because the YFP tag enhances opsin sensitivity, we prioritized these tagged RubyACRs for initial characterization. FLAG-tagged ACRs are in progress, but will take time to fully characterize. Considering that the RubyACR-EYFP versions work very well, and in many cases people will want the YFP tag, either for visualizing expression or to maximize sensitivity, we feel the current work is a valuable contribution on its own. Indeed several labs have already requested these lines.

(2) Are the ACRs activated by two-photon illumination?

We will examine GCaMP signals at increasing 2p intensities to determine whether imaging unintentionally activates RubyACRs, as well as whether 2p illumination could be used for intentional opsin activation.

(3) How toxic is the expression of these opsins?

We will update the quantification of toxicity in Table 1 to include all the drivers we used in this study. In fact the toxicity we observed was primarily with the vGlut driver, which was why that was the only information in the table. The other drivers we used did not appreciably reduce survival rate, but showing the one case where it did have a big effect left a strong and understandably inaccurate impression that toxicity was a big pitfall. We note that the widely used CSChrimson has similar % survival to the RubyACRs when expressed with these vGlut drivers.

We also plan to examine whether ACR expression leads to cell-autonomous perturbations. We will determine whether expression leads to some frequency of neuronal cell death, and we will evaluate whether any morphological effects occur.

We will also clarify in the Discussion that potential toxicity may be driver-specific (as it is here) and should be evaluated case-by-case by investigators using the tool.

(4) Use functional imaging to confirm inhibition of the neurons used only for behavioral experiments (pIP10 & PPL1-γ1pedc)

We will perform these imaging experiments. One caveat is that inhibition may not be readily detectable with GCaMP, as the resting calcium levels in pIP10 and PPL1-γ1pedc neurons may already be quite low. This differs from the non-spiking Mi1 neurons, where inhibition was clearly observed with GCaMP. For this reason, we consider the behavioral results stronger evidence of efficacy, but we agree that imaging could provide useful supporting evidence, recognizing that a negative result would be difficult to interpret.

(5) Confirm that the GtACR1 will inhibit locomotion in the flybowl when activated with green light, its spectral peak.

We will perform this benchmark experiment. Please note that our intention with this study was to find an effective red-light activated opto-inhibitor because these wavelengths are much less perturbing to behavior. In that respect, regardless of GtACR1’s performance with green light, the RubyACRs clearly provide important new tools for Drosophila behavioral neuroscience.

Author response:

The following is the authors’ response to the current reviews.

Public Reviews:

Reviewer #1 (Public review):

I applaud the authors' for providing a thorough response to my comments from the first round of review. The authors' have addressed the points I raised on the interpretation of the behavioral results as well as the validation of the model (fit to the data) by conducting new analyses, acknowledging the limitations where required and providing important counterpoints. As a result of this process, the manuscript has considerably improved. I have no further comments and recommend this manuscript for publication.

We are pleased that our revisions have addressed all the concerns raised by Reviewer #1.

Reviewer #2 (Public review):

Summary:

This manuscript proposes that the use of a latent cause model for assessment of memory-based tasks may provide improved early detection in Alzheimer's Disease as well as more differentiated mapping of behavior to underlying causes. To test the validity of this model, the authors use a previously described knock-in mouse model of AD and subject the mice to several behaviors to determine whether the latent cause model may provide informative predictions regarding changes in the observed behaviors. They include a well-established fear learning paradigm in which distinct memories are believed to compete for control of behavior. More specifically, it's been observed that animals undergoing fear learning and subsequent fear extinction develop two separate memories for the acquisition phase and the extinction phase, such that the extinction does not simply 'erase' the previously acquired memory. Many models of learning require the addition of a separate context or state to be added during the extinction phase and are typically modeled by assuming the existence of a new state at the time of extinction. The Niv research group, Gershman et al. 2017, have shown that the use of a latent cause model applied to this behavior can elegantly predict the formation of latent states based on a Bayesian approach, and that these latent states can facilitate the persistence of the acquisition and extinction memory independently. The authors of this manuscript leverage this approach to test whether deficits in production of the internal states, or the inference and learning of those states, may be disrupted in knock-in mice that show both a build-up of amyloid-beta plaques and a deterioration in memory as the mice age.

Strengths:

I think the authors' proposal to leverage the latent cause model and test whether it can lead to improved assessments in an animal model of AD is a promising approach for bridging the gap between clinical and basic research. The authors use a promising mouse model and apply this to a paradigm in which the behavior and neurobiology are relatively well understood - an ideal situation for assessing how a disease state may impact both the neurobiology and behavior. The latent cause model has the potential to better connect observed behavior to underlying causes and may pave a road for improved mapping of changes in behavior to neurobiological mechanisms in diseases such as AD.

The authors also compare the latent cause model to the Rescorla-Wagner model and a latent state model allowing for better assessment of the latent cause model as a strong model for assessing reinstatement.

Weaknesses:

I have several substantial concerns which I've detailed below. These include important details on how the behavior was analyzed, how the model was used to assess the behavior, and the interpretations that have been made based on the model.

(1) There is substantial data to suggest that during fear learning in mice separate memories develop for the acquisition and extinction phases, with the acquisition memory becoming more strongly retrieved during spontaneous recovery and reinstatement. The Gershman paper, cited by the authors, shows how the latent causal model can predict this shift in latent causes by allowing for the priors to decay over time, thereby increasing the posterior of the acquisition memory at the time of spontaneous recovery. In this manuscript, the authors suggest a similar mechanism of action for reinstatement, yet the model does not appear to return to the acquisition memory after reinstatement, at least based on the simulation and examples shown in figures 1 and 3. More specifically, in figure 1, the authors indicate that the posterior probability of the latent cause, zA (the putative acquisition memory), increases, partially leading to reinstatement. This does not appear to be the case as test 3 (day 36) appears to have similar posterior probabilities for zA as well as similar weights for the CS as compared to the last days of extinction. Rather, the model appears to mainly modify the weights in the most recent latent cause, zB - the putative the 'extinction state', during reinstatement. The authors suggest that previous experimental data have indicated that spontaneous recovery or reinstatement effects are due to an interaction of the acquisition and extinction memory. These studies have shown that conditioned responding at a later time point after extinction is likely due to a balance between the acquisition memory and the extinction memory, and that this balance can shift towards the acquisition memory naturally during spontaneous recovery, or through artificial activation of the acquisition memory or inhibition of the extinction memory (see Lacagnina et al. for example). Here the authors show that the same latent cause learned during extinction, zB, appears to dominate during the learning phase of reinstatement, with rapid learning to the context - the weight for the context goes up substantially on day 35 - in zB. This latent cause, zB, dominates at the reinstatement test, and due to the increased associative strength between the context and shock, there is a strong CR. For the simulation shown in figure 1, it's not clear why a latent cause model is necessary for this behavior. This leads to the next point.

We would like to first clarify that our behavioral paradigm did not last for 36 days, as noted by the reviewer. Our reinstatement paradigm contained 7 phases and 36 trials in total: acquisition (3 trials), test 1 (1 trial), extinction 1 (19 trials), extinction 2 (10 trials), test 2 (1 trial), unsignaled shock (1 trial), test 3 (1 trial). The day is labeled under each phase in Figure 2A. 

We have provided explanations on how the reinstatement is explained by the latent cause model in the first round of the review. Briefly, both acquisition and extinction latent causes contribute to the reinstatement (test 3). The former retains the acquisition fear memory, and the latter has the updated w_context from unsignaled shock. Although the reviewer is correct that the zB in Figure 1D makes a great contribution during the reinstatement, we would like to argue that the elevated CR from test 2 (trial 34) to test 3 (trial 36) is the result of the interaction between zA and zB.

We provided Author response image 1 using the same data in Figure 1D and 1E to further clarify this point. The posterior probability of zA increased after an unsignaled shock (trial 35), which may be attributed to the return of acquisition fear memory. The posterior probability of zA then decreased again after test 3 (trial 36) because there was no shock in this trial. Along with the weight change, the expected shock change substantially in these three trials, resulting in reinstatement. Note that the mapping of expected shock to CR in the latent cause model is controlled by parameter θ and λ. Once the expected shock exceeds the threshold θ, the CR will increase rapidly if λ is smaller.

Lastly, accepting the idea that separate memories are responsible for acquisition and extinction in the memory modification paradigm, the latent cause model (LCM) is a rational candidate modeling this idea. Please see the following reply on why a simple model like the Rescorla-Wagner (RW) model is not sufficient to fully explain the behaviors observed in this study.

Author response image 1.

The sum posterior probability (A), the sum of associative weight of CS (B), and the sum of associative weight of context (C) of acquisition and extinction latent causes in Figure 1D and 1E.

(2) The authors compared the latent cause model to the Rescorla-Wagner model. This is very commendable, particularly since the latent cause model builds upon the RW model, so it can serve as an ideal test for whether a more simplified model can adequately predict the behavior. The authors show that the RW model cannot successfully predict the increased CR during reinstatement (Appendix figure 1). Yet there are some issues with the way the authors have implemented this comparison:

(2A) The RW model is a simplified version of the latent cause model and so should be treated as a nested model when testing, or at a minimum, the number of parameters should be taken into account when comparing the models using a method such as the Bayesian Information Criterion, BIC.

We acknowledge that the number of parameters was not taken into consideration when we compared the models. We thank the reviewer for the suggestion to use the Bayesian Information Criterion (BIC). However, we did not use BIC in this study for the following reasons. We wanted a model that can explain fear conditioning, extinction and reinstatement, so our first priority is to fit the test phases. Models that simulate CRs well in non-test phases can yield lower BIC values even if they fail to capture reinstatement. When we calculate the BIC by using the half normal distribution (μ = 0, σ \= 0.3) as the likelihood for prediction error in each trial, the BIC of the 12-month-old control is -37.21 for the RW model (Appendix 1–figure 1C) and -11.60 for the LCM (Figure 3C). Based on this result, the RW model would be preferred, yet the LCM was penalized by the number of parameters, even though it fit better in trial 36. Because we did not think this aligned with our purpose to model reinstatement, we chose to rely on the practical criteria to determine whether the estimated parameter set is accepted or not for our purpose (see Materials and Methods). The number of accepted samples can thus roughly be seen as the model's ability to explain the data in this study. These exclusion criteria then created imbalances in accepted samples across models (Appendix 1–figure 2). In the RW model, only one or two samples met the criteria, preventing meaningful statistical comparisons of BIC within each group. Overall, though we agreed that BIC is one of the reasonable metrics in model comparison, we did not think it aligns with our purpose in this study.

(2B) The RW model provides the associative strength between stimuli and does not necessarily require a linear relationship between V and the CR. This is the case in the original RW model as well as in the LCM. To allow for better comparison between the models, the authors should be modeling the CR in the same manner (using the same probit function) in both models. In fact, there are many instances in which a sigmoid has been applied to RW associative strengths to predict CRs. I would recommend modeling CRs in the RW as if there is just one latent cause. Or perhaps run the analysis for the LCM with just one latent cause - this would effectively reduce the LCM to RW and keep any other assumptions identical across the models.

Regarding the suggestion to run the analysis using the LCM with one latent cause, we agree that this method is almost identical to the RW model, which is also mentioned in the original paper (Gershman et al., 2017). Importantly, it would also eliminate the RW model’s advantage of assigning distinct learning rates to different stimuli, highlighted in the next comment (2C).

We thank the reviewer for suggesting applying the transformation of associative strength (V) to CR as in the LCM. We examined this possibility by heuristically selecting parameter values to test how such a transformation would influence the RW model (Author response image 2A). Specifically, we set α_CS = 0.5, α_context \= 1, β = 1, and introduced the additional parameters θ and λ, as in the LCM. This parameter set is determined heuristically to address the reviewer’s concern about a higher learning rate of context. The dark blue line is the plain associative strength. The remaining lines are CR curves under different combinations of θ and λ.

Consistent with the reviewer’s comment, under certain parameter settings (θ \= 0.01, λ = 0.01), the extended RW model can reproduce higher CRs at test 3, thereby approximating the discrimination index observed in the 12-month-old control group. However, this modification changes the characteristics of CRs in other phases from those in the plain RW model. In the acquisition phase, the CRs rise more sharply. In the extinction phase, the CRs remain high when θ is small. Though changing λ can modulate the steepness, the CR curve is flat on the second day of the extinction phase, which does not reproduce the pattern in observed data (Figure 2B). These trade-offs suggest that the RW model with the sigmoid transformation does not improve fit quality and, in fact, sacrifices features that were well captured by simpler RW simulations (Appendix 1–figure 1A to 1D). To further evaluate this extended RW model (RW*), we applied the same parameter estimation method used in the LCM for individual data (see Materials and Methods). For each animal, α_CS, α_context, β, θ, and λ were estimated with their lower and upper bounds set as previously described (see Appendix 1, Materials and Methods). The results showed that the number of accepted samples slightly increased compared to the RW model without sigmoidal transformation of CR (RW* vs. RW in Author response image 2B, 2C). However, this improvement did not surpass the LCM (RW* vs. LCM in Author response image 2B, Author response image 1C). Overall, these results suggest that while using the same method to map the expected shock to CR, the RW model does not outperform the LCM. Practically, further extension, such as adding novel terms, might improve the fitting level. We would like to note that such extensions should be carefully validated if they are reasonable and necessary for an internal model, which is beyond the scope of this study. We hope this addresses the reviewer's concerns about the implementation of the RW model. 

Author response image 2.

Simulation (A) and parameter estimation (B and C) in the extended Rescorla-Wagner model.

(2C) In the paper, the model fits for the alphas in the RW model are the same across the groups. Were the alphas for the two models kept as free variables? This is an important question as it gets back to the first point raised. Because the modeling of the reinstatement behavior with the LCM appears to be mainly driven by latent cause zB, the extinction memory, it may be possible to replicate the pattern of results without requiring a latent cause model. For example, the 12-month-old App NL-G-F mice behavior may have a deficit in learning about the context. Within the RW model, if the alpha for context is set to zero for those mice, but kept higher for the other groups, say alpha_context = 0.8, the authors could potentially observe the same pattern of discrimination indices in figure 2G and 2H at test. Because the authors don't explicitly state which parameters might be driving the change in the DI, the authors should show in some way that their results cannot simply be due to poor contextual learning in the 12 month old App NL-G-F mice, as this can presumably be predicted by the RW model. The authors' model fits using RW don't show this, but this is because they don't consider this possibility that the alpha for context might be disrupted in the 12-month-old App NL-G-F mice. Of course, using the RW model with these alphas won't lead to as nice of fits of the behavior across acquisition, extinction, and reinstatement as the authors' LCM, the number of parameters are substantially reduced in the RW model. Yet the important pattern of the DI would be replicated with the RW model (if I'm not mistaken), which is the important test for assessment of reinstatement.

We would like to clarify that we estimated three parameters in the RW model for individuals:  α_CS,  α_context, and β. Even if we did so, many samples did not satisfy our criteria (Appendix 1–figure 2). Please refer to the “Evaluation of model fit” in Appendix 1 and the legend of Appendix 1–figure 1A to 1D, where we have written the estimated parameter values.

We did not agree that paralyzing the contextual learning by setting  α_context  as 0 in the RW model can explain the CR curve of 12-month-old AD mice well. Specifically, the RW model cannot capture the between-day extinction dynamics (i.e., the increase in CR at the beginning of day 2 extinction)  and the higher CR at test 3 relative to test 2 (i.e., DI between test 3 and test 2 is greater than 0.5). In addition, because the context input (= 0.2) was relatively lower than the CS input (= 1), and there is only a single unsignaled shock trial, even setting  α_context = 1 results in only a limited increase in CR (Appendix 1–figure 1A to 1D; see also Author response image 2 9). Thus, the RW model cannot replicate the reinstatement effect or the critical pattern of discrimination index, even under conditions of stronger contextual learning.  

(3) As stated by the authors in the introduction, the advantage of the fear learning approach is that the memory is modified across the acquisition-extinction-reinstatement phases. Although perhaps not explicitly stated by the authors, the post-reinstatement test (test 3) is the crucial test for whether there is reactivation of a previously stored memory, with the general argument being that the reinvigorated response to the CS can't simply be explained by relearning the CS-US pairing, because re-exposure the US alone leads to increase response to the CS at test. Of course there are several explanations for why this may occur, particularly when also considering the context as a stimulus. This is what I understood to be the justification for the use of a model, such as the latent cause model, that may better capture and compare these possibilities within a single framework. As such, it is critical to look at the level of responding to both the context alone and to the CS. It appears that the authors only look at the percent freezing during the CS, and it is not clear whether this is due to the contextual-US learning during the US re-exposure or to increased responding to the CS - presumably caused by reactivation of the acquisition memory. The authors do perform a comparison between the preCS and CS period, but it is not clear whether this is taken into account in the LCM. For example, the instance of the model shown in figure 1 indicates that the 'extinction cause', or cause z6, develops a strong weight for the context during the reinstatement phase of presenting the shock alone. This state then leads to increased freezing during the final CS probe test as shown in the figure. If they haven't already, I think the authors must somehow incorporate these different phases (CS vs ITI) into their model, particularly since this type of memory retrieval that depends on assessing latent states is specifically why the authors justified using the latent causal model. In more precise terms, it's not clear whether the authors incorporate a preCS/ITI period each day the cue is presented as a vector of just the context in addition to the CS period in which the vector contains both the context and the CS. Based on the description, it seemed to me that they only model the CRs during the CS period on days when the CS is presented, and thereby the context is only ever modeled on its own (as just the context by itself in the vector) on extinction days when the CS is not presented. If they are modeling both timepoints each day that the CS I presented, then I would recommend explicitly stating this in the methods section.

In this study, we did not model the preCS freezing rate, and we thank the reviewer for the suggestion to model preCS periods as separate context-only trials. In our view, however, this approach is not consistent with the assumptions of the LCM. Our rationale is that the available periods of context and the CS are different. We assume that observation of the context lasts from preCS to CS. If we simulate both preCS (context) and CS (context and tone), the weight of context would be updated twice. Instead, we follow the same method as described in the original code from Gershman et al. (2017) to consider the context effect. We agree that explicitly modeling preCS could provide additional insights, but we believe it would require modifying or extending the LCM. We consider this an important direction for future research, but it is outside the scope of this study.

(4) The authors fit the model using all data points across acquisition and learning. As one of the other reviewers has highlighted, it appears that there is a high chance for overfitting the data with the LCM. Of course, this would result in much better fits than models with substantially fewer free parameters, such as the RW model. As mentioned above, the authors should use a method that takes into account the number of parameters, such as the BIC.

Please refer to the reply to public review (2A) for the reason we did not take the suggestion to use BIC. In addition, we feel that we have adequately addressed the concern of overfitting in the first round of the review. 

(5) The authors have stated that they do not think the Barnes maze task can be modeled with the LCM. Whether or not this is the case, if the authors do not model this data with the LCM, the Barnes maze data doesn't appear valuable to the main hypothesis. The authors suggest that more sophisticated models such as the LCM may be beneficial for early detection of diseases such as Alzheimer's, so the Barnes maze data is not valuable for providing evidence of this hypothesis. Rather, the authors make an argument that the memory deficits in the Barnes maze mimic the reinstatement effects providing support that memory is disrupted similarly in these mice. Although, the authors state that the deficits in memory retrieval are similar across the two tasks, the authors are not explicit as to the precise deficits in memory retrieval in the reinstatement task - it's a combination of overgeneralizing latent causes during acquisition, poor learning rate, over differentiation of the stimuli.

We would like to clarify that we valued the latent cause model not solely because it is more sophisticated and fits more data points, but it is an internal model that implicates the cognitive process. Please also see the reply to the recommendations to authors (3) about the reason why we did not take the suggestion to remove this data.

Reviewer #3 (Public review):

Summary:

This paper seeks to identify underlying mechanisms contributing to memory deficits observed in Alzheimer's disease (AD) mouse models. By understanding these mechanisms, they hope to uncover insights into subtle cognitive changes early in AD to inform interventions for early-stage decline.

Strengths:

The paper provides a comprehensive exploration of memory deficits in an AD mouse model, covering early and late stages of the disease. The experimental design was robust, confirming age-dependent increases in Aβ plaque accumulation in the AD model mice and using multiple behavior tasks that collectively highlighted difficulties in maintaining multiple competing memory cues, with deficits most pronounced in older mice.

In the fear acquisition, extinction, and reinstatement task, AD model mice exhibited a significantly higher fear response after acquisition compared to controls, as well as a greater drop in fear response during reinstatement. These findings suggest that AD mice struggle to retain the fear memory associated with the conditioned stimulus, with the group differences being more pronounced in the older mice.

In the reversal Barnes maze task, the AD model mice displayed a tendency to explore the maze perimeter rather than the two potential target holes, indicating a failure to integrate multiple memory cues into their strategy. This contrasted with the control mice, which used the more confirmatory strategy of focusing on the two target holes. Despite this, the AD mice were quicker to reach the target hole, suggesting that their impairments were specific to memory retrieval rather than basic task performance.

The authors strengthened their findings by analyzing their data with a leading computational model, which describes how animals balance competing memories. They found that AD mice showed somewhat of a contradiction: a tendency to both treat trials as more alike than they are (lower α) and similar stimuli as more distinct than they are (lower σx) compared to controls.

Weaknesses:

While conceptually solid, the model struggles to fit the data and to support the key hypothesis about AD mice's inability to retain competing memories. These issues are evident in Figure 3:

(1) The model misses trends in the data, including the gradual learning of fear in all groups during acquisition, the absence of a fear response at the start of the experiment, and the faster return of fear during reinstatement compared to the gradual learning of fear during acquisition. It also underestimates the increase in fear at the start of day 2 of extinction, particularly in controls.

(2) The model explains the higher fear response in controls during reinstatement largely through a stronger association to the context formed during the unsignaled shock phase, rather than to any memory of the conditioned stimulus from acquisition (as seen in Figure 3C). In the experiment, however, this memory does seem to be important for explaining the higher fear response in controls during reinstatement (as seen in Author Response Figure 3). The model does show a necessary condition for memory retrieval, which is that controls rely more on the latent causes from acquisition. But this alone is not sufficient, since the associations within that cause may have been overwritten during extinction. The Rescorla-Wagner model illustrates this point: it too uses the latent cause from acquisition (as it only ever uses a single cause across phases) but does not retain the original stimulus-shock memory, updating and overwriting it continuously. Similarly, the latent cause model may reuse a cause from acquisition without preserving its original stimulus-shock association.

These issues lead to potential overinterpretation of the model parameters. The differences in α and σx are being used to make claims about cognitive processes (e.g., overgeneralization vs. over differentiation), but the model itself does not appear to capture these processes accurately.

The authors could benefit from a model that better matches the data and captures the retention and retrieval of fear memories across phases. While they explored alternatives, including the Rescorla-Wagner model and a latent state model, these showed no meaningful improvement in fit. This highlights a broader issue: these models are well-motivated but may not fully capture observed behavior.

Conclusion:

Overall, the data support the authors' hypothesis that AD model mice struggle to retain competing memories, with the effect becoming more pronounced with age. While I believe the right computational model could highlight these differences, the current models fall short in doing so.

We thank the reviewer for the insightful comments. For the comments (1) and (2), please refer to our previous author response to comments #26 and #27. We recognize that the models tested in this study have limitations and, as noted, do not fully capture all aspects of the observed behavioral data. We see this as an important direction for future research and value the reviewer’s suggestions.

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

I have maintained some of the main concerns included in the first round of reviews as I think they remain concerns with the new draft, even though the authors have included substantially more analysis of their data, which is appreciated. I particularly found the inclusion of the comparative modeling valuable, although I think the analysis comparing the models should be improved.

(1) This relates to point 1 in the public assessment or #16 in the response to reviewers from the authors. The authors raise the point that even a low posterior can drive behavioral expression (lines 361-365 in the response to authors), and so the acquisition latent cause may partially drive reinstatement. Yet in the stimulation shown in figure 1D, this does not seem to be the case. As I mentioned in the public response, in figure 1, the posteriors for zA are similar on day 34 and day 36, yet only on day 36 is there a strong CR. At least in this example, it does not appear that zA contributes to the increased responding from day 34 (test 2) to day 36 (test 3). There may be a slight increase in z1 in figure 3C, but the dominant change from day 34 to day 36 appears to be the increase in the posterior of z3 and the substantial increase in w3. The authors then cite several papers which have shown the shift in balance between what it is the putative acquisition memory and extinction memory (i.e. Lacagnina et al.). Yet I do not see how this modeling fits with most of the previous findings. For example, in the Lacagnina et al. paper, activation of the acquisition ensemble or inhibition of the extinction ensemble drives freezing, whereas the opposite pattern reduces freezing. What appears to be the pattern in the modeling in this paper is primarily learning of context in the extinction latent cause to predict the shock. As I mention in point 2C of the public review, it's not clear why this pattern of results would require a latent cause model. Would a high alpha for context and not the CS not give a similar pattern of results in the RW model? At least for giving similar results of the DIs in figure 2?

First, we would like to clarify that the x-axis in Figure 1D is labeled “Trial,” not “Day.” Please refer to the reply to public review (1), where we clarified the posterior probability of the latent cause from trials 34 to 36. Second, although we did not have direct neural circuit evidence in this study, we discussed the similarities between previous findings and the modeling in the first review. Briefly, our main point focuses on the interaction between acquisition and extinction memory. In other words, responses at different times arise from distinct internal states made up of competing memories. We assume that the reviewer expects a modeling result showing nearly full recovery of acquisition memory, which aligns with previous findings where optogenetic activation of the acquisition engram can partially mimic reinstatement (Zaki et al., 2022; see also the response to comment #12 in the first round of review). We acknowledge that such a modeling result cannot be achieved with the latent cause model and see it as a potential future direction for model improvement.

Please also refer to the reply to public review (2) about how a high alpha for context in the RW model cannot explain the pattern we observed in the reinstatement paradigm.

(2) This is related to point 3 in the public comments and #13 in the response to reviewers. I raised the question of comparing the preCS/ITI period with the CS period, but my main point was why not include these periods in the LCM itself as mentioned in more detail in point 3 in the current public review. The inclusion of the comparisons the authors performed helped, but my main point was that the authors could have a better measure of wcontext if they included the preCS period as a stimulus each day (when only the context is included in the stimulus). This would provide better estimates of wcontext. As stated in the public review, perhaps the authors did this, but my understanding of the methods this was not the case, rather, it seems the authors only included the CS period for CRs within the model (at least on days when the CS was present).

Please refer to the reply to public review (3) about the reason why we did not model the preCS freezing rate.

(3) This relates to point 4 in the public review and #15 and #24 in the response to authors. The authors have several points for why the two experiments are similar and how results may be extrapolated - lines 725-733. The first point is that associative learning is fundamental in spatial learning. I'm not sure that this broad connection between the two studies is particularly insightful for why one supports the other as associative learning is putatively involved in most behavioral tasks. In the second point about reversals, why not then use a reversal paradigm that would be easier to model with LCM? This data is certainly valuable and interesting, yet I don't think it's helpful for this paper to state qualitatively the similarities in the potential ways a latent cause framework might predict behavior on the Barnes maze. I would recommend that the authors either model the behavior with LCM, remove the experiment from the paper, or change the framing of the paper that LCM might be an ideal approach for early detection of dementia or Alzheimer's disease.

We would like to clarify that our aim was not to present the LCM as an ideal tool for early detection of AD symptoms. Rather, our focus is on the broader idea of utilizing internal models and estimating individual internal states in early-stage AD. Regarding using a reversal paradigm that would be easier to model with LCM, the most straightforward approach is to use another type of paradigm for fear conditioning, then to examine the extent to which similar behavioral characteristics are observed between paradigms within subjects. However, re-exposing the same mice to such paradigms is constrained by strong carry-over effects, limiting the feasibility of this experiment. Other behavioral tasks relevant to AD that avoid shock generally involve action selection for subsequent observation (Webster et al., 2014), which falls outside the structure of LCM. Our rationale for including the Barnes maze task is that spatial memory deficit is implicated in the early stage of AD, making it relevant for translational research. While we acknowledge that exact modeling of Barnes maze behavior would require a more sophisticated model (as discussed in the first round of review), our intention to use the reversal Barnes maze paradigm is to suggest a presumable memory modification learning in a non-fear conditioning paradigm. We also discussed whether similar deficits in memory modification could be observed across two behavioral tasks.

(4) Reviewer # mentioned that the change in pattern of behavior only shows up in the older mice questioning the clinical relevance of early detection. I do think this is a valid point and maybe should be addressed. There does seem to be a bit of a bump in the controls on day 23 that doesn't appear in the 6-month group. Perhaps this was initially a spontaneous recovery test indicated by the dotted vertical line? This vertical line does not appear to be defined in the figure 1 legend, nor in figures 2 and 3.

We would like to emphasize that the App^NL-G-F knock-in mouse is widely considered a model of early-stage AD, characterized by Aβ accumulation with little to no neurofibrillary tangle pathology or neuronal loss (see Introduction). By examining different ages, we can assess the contribution of both the amount and duration of Aβ accumulation as well as age-related factors. Modeling the deficit in the memory modification process in the older App^NL-G-F knock-in mice, we suggested a diverged internal state in early-stage AD in older age, and this does not diminish the relevance of the model for studying early cognitive changes in AD.

We would also like to clarify again that the x-axis in the figure is “Trial,” not “Day.” The vertical dashed lines in these figures indicate phase boundaries, and they were defined in the figure legend: in Figure 1C, “The vertical dashed lines separate the phases.”; in Figure 2B, “The dashed vertical line separates the extinction 1 and extinction 2 phases.”; in Figure 3, “The vertical dashed lines indicate the boundaries of phases.”

(5) Are the examples in figure 3 good examples? The example for the 12-month-old control shows a substantial increase in weights for the context during test 3, but not for the CS. Yet in the bar plots in Figure 4 G and H, this pattern seems to be different. The weights for the context appear to substantially drop in the "after extinction" period as compared to the "extinction" period. It's hard to tell the change from "extinction" to "after extinction" for the CS weights (the authors change the y-axis for the CS weights but not for the context weights from panels G to H).

We would like to clarify that in Figure 3C, the increase in weights for context is not presented during test 3 (trial 36), noted by the reviewer; rather, it is the unsignaled shock phase (trial 35).

We assumed that the reviewer might misunderstand that the labels on the left in Figure 4, “Acquisition”, “Extinction”, and “After extinction”, indicate the time point. However, the data shown in Figure 4C to 4H are all from the same time point: test 3 (trial 36). The grouping reflects the classification of latent causes based on the trial in which they were inferred. In addition, for Figures 4G and 4H, the y‐axis limits were not set identically because the data range for “Sum of w_CS” varied. This was done to ensure the visibility of all data points. In Figure 4, each dot represents one animal. Take Figure 3D as an example. The point in Figure 4G is the sum of w3 and w4 in trial 36, and the point in Figure 4H is w5 in trial 36, note that the subscript numerals indicate latent cause index. We hope this addresses the reviewer’s question about the difference between the two figures.

The following is the authors’ response to the original reviews

Reviewer #1 (Public review):

Summary:

The authors show certain memory deficits in a mouse knock-in model of Alzheimer's Disease (AD). They show that the observed memory deficits can be explained by a computational model, the latent cause model of associative memory. The memory tasks used include the fear memory task (CFC) and the 'reverse' Barnes maze. Research on AD is important given its known huge societal burden. Likewise, better characterization of the behavioral phenotypes of genetic mouse models of AD is also imperative to advance our understanding of the disease using these models. In this light, I applaud the authors' efforts.

Strengths:

(1) Combining computational modelling with animal behavior in genetic knock-in mouse lines is a promising approach, which will be beneficial to the field and potentially explain any discrepancies in results across studies as well as provide new predictions for future work.

(2) The authors' usage of multiple tasks and multiple ages is also important to ensure generalization across memory tasks and 'modelling' of the progression of the disease.

Weaknesses:

[#1] (1) I have some concerns regarding the interpretation of the behavioral results. Since the computational model then rests on the authors' interpretation of the behavioral results, it, in turn, makes judging the model's explanatory power difficult as well. For the CFC data, why do knock-in mice have stronger memory in test 1 (Figure 2C)? Does this mean the knock-in mice have better memory at this time point? Is this explained by the latent cause model? Are there some compensatory changes in these mice leading to better memory? The authors use a discrimination index across tests to infer a deficit in re-instatement, but this indicates a relative deficit in re-instatement from memory strength in test 1. The interpretation of these differential DIs is not straightforward. This is evident when test 1 is compared with test 2, i.e., the time point after extinction, which also shows a significant difference across groups, Figure 2F, in the same direction as the re-instatement. A clarification of all these points will help strengthen the authors' case.

We appreciate the reviewer for the critical comments. According to the latent cause framework, the strength of the memory is influenced by at least 2 parameters: associative weight between CS and US given a latent cause and posterior probability of the latent cause. The modeling results showed that a higher posterior probability of acquisition latent cause, but not higher associative weight, drove the higher test 1 CR in App^NL-G-F mice (Results and Discussion; Figure 4 – figure supplement 3B, 3C). In terms of posterior, we agree that App^NL-G-F mice have strong fear memory. On the other hand, this suggests that App^NL-G-F mice exhibited a tendency toward overgeneralization, favoring modification of old memories, which adversely affected the ability to retain competing memories. The strong memory in test 1 would be a compensatory effect of overgeneralization.    

To estimate the magnitude of reinstatement, at least, one would have to compare CRs between test 2 (extinction) and test 3 (reinstatement), as well as those between test 1 (acquisition) and test 3. These comparisons represent the extent to which the memory at the reinstatement is far from that in the extinction, and close to that in the acquisition. Since discrimination index (DI) has been widely used as a normalized measure to evaluate the extent to which the system can distinguish between two conditions, we applied DI consistently to behavioral and simulated data in the reinstatement experiment, and the behavioral data in the reversal Barnes maze experiment, allowing us to evaluate the discriminability of an agent in these experiments. In addition, we used DI to examine its correlation with estimated parameters, enabling us to explore how individual discriminability may relate to the internal state. We have already discussed the differences in DI between test 3 and test 1, as well as CR in test 1 between control and App^NL-G-F in the manuscript and further elaborated on this point in Line 232, 745-748.   

[#2] (2) I have some concerns regarding the interpretation of the Barnes maze data as well, where there already seems to be a deficit in the memory at probe test 1 (Figure 6C). Given that there is already a deficit in memory, would not a more parsimonious explanation of the data be that general memory function in this task is impacted in these mice, rather than the authors' preferred interpretation? How does this memory weakening fit with the CFC data showing stronger memories at test 1? While I applaud the authors for using multiple memory tasks, I am left wondering if the authors tried fitting the latent cause model to the Barnes maze data as well.

While we agree that the deficits shown in probe test 1 may imply impaired memory function in App^NL-G-F mice in this task, it would be difficult to explain this solely in terms of impairments in general memory function. The learning curve and the daily strategy changes suggested that App^NL-G-F mice would have virtually intact learning ability in the initial training phase (Figure 6B, 6F, Figure 6 – figure supplement 1 and 3). For the correspondence relationship between the reinstatement and the reversal Barnes maze learning from the aspect of memory modification process, please also see our reply to comment #24. We have explained why we did not fit the latent cause model to the Barnes maze data in the provisional response.

[#3] (3) Since the authors use the behavioral data for each animal to fit the model, it is important to validate that the fits for the control vs. experimental groups are similar to the model (i.e., no significant differences in residuals). If that is the case, one can compare the differences in model results across groups (Figures 4 and 5). Some further estimates of the performance of the model across groups would help.

We have added the residual (i.e., observed CR minus simulated CR) in Figure 3 – figure supplement 1D and 1E. The fit was similar between control and App^NL-G-F mice groups in the test trials, except test 3 in the 12-month-old group. The residual was significantly higher in the 12-month-old control mice than App^NL-G-F mice, suggesting the model underestimated the reinstatement in the control, yet the DI calculated from the simulated CR replicates the behavioral data (Figure 3 – figure supplement 1A to 1C). These results suggest that the latent cause model fits our data with little systematic bias such as an overestimation of CR for the control group in the reinstatement, supporting the validity of the comparisons in estimated parameters between groups. These results and discussion have been added in the manuscript Line 269-276.

One may notice that the latent cause model overestimated the CR in acquisition trials in all groups in Figure 3 – figure supplement 1D and 1E. We have discussed this point in the reply to comment #26, 34 questioned by reviewer 3.

[#4] (4) Is there an alternative model the authors considered, which was outweighed in terms of prediction by this model? 

Yes, we have further evaluated two alternative models: the Rescorla-Wagner (RW; Rescorla & Wagner, 1972) model and the latent state model (LSM; Cochran & Cisler, 2019). The RW model serves as a baseline, given its known limitations in explaining fear return after extinction. The LSM is another contemporary model that shares several concepts with the latent cause model (LCM) such as building upon the RW model, assuming a latent variable inferred by Bayes’ rule, and involving a ruminative update for memory modification. We evaluated the three models in terms of the prediction accuracy and reproducibility of key behavioral features. Please refer to the Appendix 1 for detailed methods and results for these two models.

As expected, the RW model fit well to the data till the end of extinction but failed to reproduce reinstatement (Appendix 1 – figure 1A to 1D). Due to a large prediction error in test 3, few samples met the acceptance criteria we set (Appendix 1 – figure 2 and 3A). Conversely, the LSM reproduced reinstatement, as well as gradual learning in acquisition and extinction phases, particularly in the 12month-old control (Appendix 1 – figure 1G). The number of accepted samples in the LSM was higher than in the RW model but generally lower than in the LCM (Appendix 1 – figure 2). The sum of prediction errors over all trials in the LSM was comparable to that in the LCM in the 6-month-old group (Appendix 1 – figure 4A), it was significantly lower in the 12-month-old group (Appendix 1 – figure 4B). Especially the LSM generated smaller prediction errors during the acquisition trials than in the LCM, suggesting that the LSM might be better at explaining the behaviors of acquisition (Appendix 1 – figure 4A and 4B; but see the reply for comment #34). While the LSM generated smaller prediction errors than the LCM in test 2 of the control group, it failed to replicate the observed DIs, a critical behavioral phenotype difference between control and App^NL-G-F mice (Appendix 1 – figure 6A to 6C; cf. Figure 2F to 2H, Figure 3 – figure supplement 1A to 1C).

Thus, although each model could capture different aspects of reinstatement, standing on the LCM to explain the reinstatement better aligns with our purpose. It should also be noted that we did not explore all parameter spaces of the LSM, hence we cannot rule out the possibility that alternative parameter sets could provide a better fit and explain the memory modification process well. A more comprehensive parameter search in the LSM may be a valuable direction for future research. 

[#5] One concern here is also parameter overfitting. Did the authors try leaving out some data (trials/mice) and predicting their responses based on the fit derived from the training data?

Following the reviewer’s suggestion, we confirmed if overfitting occurred using all trials to estimate parameters. Estimating parameters while actually leaving out trials would disorder the time lapse across trials, and thereby the prior of latent causes in each trial. Instead, we removed the constraint of prediction error by setting the error threshold to 1 for certain trials to virtually leave these trials out. We treated these trials as a virtual “training” dataset, while the rest of the trials were a “test” dataset. For the median CR data of each group (Figure 3), we estimated parameters under 6 conditions with unique training and test trials, then evaluated the prediction error for the training and test trials. Note that training and test trials were arbitrarily decided. Also, the error threshold for the acquisition trial was set to 1 as described in Materials and Methods, which we have further discussed the reason in the reply to comment #34 and treated acquisition trials separately from the test trials. We expect that the contribution of the data from the acquisition and test trials for parameter estimation could be discounted compared to those from the training trials with the constraint, and if overfitting occurred, the prediction error in the test data would be worse than that in the training trials.

Author response image 1A to 1F showed the simulated and observed CR under each condition, where acquisition trials were in light-shaded areas, test trials were in dark-shaded areas, and the rest of the trials were training trials. Author response image 1G showed mean squared prediction error across the acquisition, training and test trials under each condition. The dashed gray line showed the mean squared prediction error of training trials in Figure 3 as a baseline.

In conditions i and ii, where two or four trials in the extinction were used for training (Author response image 1A and 1B), the prediction error was generally higher in test trials than in training trials. In conditions iii and iv where ten trials in the extinction were used for training (Author response image 1C and 1D), the difference in prediction error between testing and training trials became smaller. These results suggest that providing more extinction trial data would reduce overfitting. In condition v (Author response image 1E), the results showed that using trials until extinction can predict reinstatement in control mice but not App^NL-G-F mice. Similarly, in condition vi (Author response image 1F), where test phase trials were left out, the prediction error differences were greater in App^NL-G-F mice. These results suggest that the test trials should be used for the parameter estimation to minimize prediction error for all groups. Overall, this analysis suggests that using all trials would reduce prediction error with few overfitting. 

Author response image 1.

Leaving trials out in parameter estimation in the latent cause model. (A – F) The observed CR (colored line) is the median freezing rate during the CS presentation over the mice within each group, which is the same as that in Figure 3. The colors indicate different groups: orange represents 6-month-old control, light blue represents 6-month-old App^NL-G-F mice, pink represents 12-month-old control, and dark blue represents 12-month-old App^NL-G-F mice. Under six different leave-out conditions (i – vi), parameters were estimated and used for generating simulated CR (gray line). In each condition, trials were categorized as acquisition (light-shaded area), training data (white area), and test data (dark-shaded area) based on the error threshold during parameter estimation. Only the error threshold of the test data trial was different from the original method (see Material and Method) and set to 1. In conditions i to vi, the number of test data trials is 27, 25, 19, and 19 in extinction phases. In condition v, the number of test data trials is 2 (trials 35 and 36). In condition vi, test data trials were the 3 test phases (trials 4, 34, and 36). (G) Each subplot shows the mean squared prediction error for the test data trial (gray circles), training data trial (white squares), and acquisition trial (gray triangles) in each group. The left y-axis corresponds to data from test and training trials, and the right y-axis corresponds to data from acquisition trials. The dashed line indicates the results calculated from Figure 3 as a baseline.  

Reviewer #1 (Recommendations for the authors):

Minor:

[#6] (1) I would like the authors to further clarify why 'explaining' the reinstatement deficit in the AD mouse model is important in working towards the understanding of AD i.e., which aspect of AD this could explain etc.

In this study, we utilized the reinstatement paradigm with the latent cause model as an internal model to illustrate how estimating internal states can improve understanding of cognitive alteration associated with extensive Aβ accumulation in the brain. Our findings suggest that misclassification in the memory modification process, manifesting as overgeneralization and overdifferentiation, underlies the memory deficit in the App^NL-G-F knock-in model mice. 

The parameters in the internal model associated with AD pathology (e.g., α and σ_x² in this study) can be viewed as computational phenotypes, filling the explanatory gap between neurobiological abnormalities and cognitive dysfunction in AD. This would advance the understanding of cognitive symptoms in the early stages of AD beyond conventional behavioral endpoints alone.

We further propose that altered internal states in App^NL-G-F knock-in mice may underlie a wide range of memory-related symptoms in AD as we observed that App^NL-G-F knock-in mice failed to retain competing memories in the reversal Barnes maze task. We speculate on how overgeneralization and overdifferentiation may explain some AD symptoms in the manuscript:

- Line 565-569: overgeneralization may explain deficits in discriminating highly similar visual stimuli reported in early-stage AD patients as they misclassify the lure as previously learned object

- Line 576-579: overdifferentiation may explain impaired ability to transfer previously learned association rules in early-stage AD patients as they misclassify them as separated knowledge. 

- Line 579-582: overdifferentiation may explain delusions in AD patients as an extended latent cause model could simulate the emergence of delusional thinking

We provide one more example here that overgeneralization may explain that early-stage AD patients are more susceptible to proactive interference than cognitively normal elders in semantic memory tests (Curiel Cid et al., 2024; Loewenstein et al., 2015, 2016; Valles-Salgado et al., 2024), as they are more likely to infer previously learned material. Lastly, we expect that explaining memory-related symptoms within a unified framework may facilitate future hypothesis generation and contribute to the development of strategies for detecting the earliest cognitive alteration in AD.  

[#7] (2) The authors state in the abstract/introduction that such computational modelling could be most beneficial for the early detection of memory disorders. The deficits observed here are pronounced in the older animals. It will help to further clarify if these older animals model the early stages of the disease. Do the authors expect severe deficits in this mouse model at even later time points?

The early stage of the disease is marked by abnormal biomarkers associated with Aβ accumulation and neuroinflammation, while cognitive symptoms are mild or absent. This stage can persist for several years during which the level of Aβ may reach a plateau. As the disease progresses, tau pathology and neurodegeneration emerge and drive the transition into the late stage and the onset of dementia. The App^NL-G-F knock-in mice recapitulate the features present in the early stage (Saito et al., 2014), where extensive Aꞵ accumulation and neuroinflammation worsen along with ages (Figure 2 – figure supplement 1). Since App^NL-G-F knock-in mice are central to Aβ pathology without tauopathy and neurodegeneration, it should be noted that it does not represent the full spectrum of the disease even at advanced ages. Therefore, older animals still model the early stages of the diseases and are suitable to study the long-term effect of Aβ accumulation and neuroinflammation. 

The age tested in previous reports using App^NL-G-F mice spanned a wide range from 2 months old to 24 months old. Different behavioral tasks have varied sensitivity but overall suggest the dysfunction worsens with aging (Bellio et al., 2024; Mehla et al., 2019; Sakakibara et al., 2018). We have tested the reinstatement experiment with 17-month-old App^NL-G-F mice before (Author response image 2). They showed more advanced deficits with the same trends observed in 12-month-old App^NL-G-F mice, but their freezing rates were overall at a lower level. There is a concern that possible hearing loss may affect the results and interpretation, therefore we decided to focus on 12-month-old data.

Author response image 2.

Freezing rate across reinstatement paradigm in the 17-month-old App^NL-G-F mice. Dashed and solid lines indicate the median freezing rate over 34 mice before (preCS) and during (CS) tone presentation, respectively. Red, blue, and yellow backgrounds represent acquisition, extinction, and unsignaled shock in Figure 2A. The dashed vertical line separates the extinction 1 and extinction 2 phases.

[#8] (3) There are quite a few 'marginal' p-values in the paper at p>0.05 but near it. Should we accept them all as statistically significant? The authors need to clarify if all the experimental groups are sufficiently powered.

For our study, we decided a priori that p < 0.05 would be considered statistically significant, as described in the Materials and Methods. Therefore, in our Results, we did not consider these marginal values as statistically significant but reported the trend, as they may indicate substantive significance.

We described our power analysis method in the manuscript Line 897-898 and have provided the results in Tables S21 and S22.

[#9] (4) The authors emphasize here that such computational modelling enables us to study the underlying 'reasoning' of the patient (in the abstract and introduction), I do not see how this is the case. The model states that there is a latent i.e. another underlying variable that was not previously considered.

Our use of the term “reasoning” was to distinguish the internal model, which describes how an agent makes sense of the world, from other generative models implemented for biomarker and disease progression prediction. However, we agree that using “reasoning” may be misleading and imprecise, so to reduce ambiguity we have removed this word in our manuscript Line 27: Nonetheless, internal models of the patient remain underexplored in AD; Line 85: However, previous approaches did not suppose an internal model of the world to predict future from current observation given prior knowledge.   

[#10] (5) The authors combine knock-in mice with controls to compute correlations of parameters of the model with behavior of animals (e.g. Figure 4B and Figure 5B). They run the risk of spurious correlations due to differences across groups, which they have indeed shown to exist (Figure 4A and 5A). It would help to show within-group correlations between DI and parameter fit, at least for the control group (which has a large spread of data).

We agree that genotype (control, App^NL-G-F) could be a confounder between the estimated parameters and DI, thereby generating spurious correlations. To address this concern, we have provided withingroup correlation in Figure 4 – figure supplement 2 for the 12-month-old group and Figure 5 – figure supplement 2 for the 6-month-old group.

In the 12-month-old group, the significant positive correlation between σx2 and DI remained in both control and App^NL-G-F mice even if we adjusted the genotype effect, suggesting that it is very unlikely that the correlations in Figure 4B are due to the genotype-related confounding. On the other hand, the positive correlation between α and DI was found to be significant in the control mice but not in the App^NL-G-F mice. Most of α were distributed around the lower bound in App^NL-G-F mice, which possibly reduced the variance and correlation coefficient. These results support our original conclusion that α and σ_x² are parameters associated with a lower magnitude of reinstatement in aged App^NL-G-F mice.

In the 6-month-old group, the correlations shown in Figure 5B were not preserved within subgroups, suggesting genotype would be a confounder for α, σ_x², and DI. We recognized that significant correlations in Figure 5B may arise from group differences, increased sample size, or greater variance after combining control and App^NL-G-F mice. 

Therefore, we concluded that α and σ_x² are associated with the magnitude of reinstatement but modulated by the genotype effect depending on the age. 

We have added interpretations of within-group correlation in the manuscript Line 307-308, 375-378.

[#11] (6) It is unclear to me why overgeneralization of internal states will lead to the animals having trouble recalling a memory. Would this not lead to overgeneralization of memory recall instead?

We assume that the reviewer is referring to “overgeneralization of internal states,” a case in which the animal’s internal state remained the same regardless of the observation, thereby leading to “overgeneralization of memory recall.” We agree that this could be one possible situation and appears less problematic than the case in which this memory is no longer retrievable. 

However, in our manuscript, we did not deal with the case of “overgeneralization of internal states”. Rather, our findings illustrated how the memory modification process falls into overgeneralization or overdifferentiation and how it adversely affects the retention of competing memories, thereby causing App^NL-G-F mice to have trouble recalling the same memory as the control mice. 

According to the latent cause model, retrieval failure is explained by a mismatch of internal states, namely when an agent perceives that the current cue does not match a previously experienced one, the old latent cause is less likely to be inferred due to its low likelihood (Gershman et al., 2017). For example, if a mouse exhibited higher CR in test 2, it would be interpreted as a successful fear memory retrieval due to overgeneralization of the fear memory. However, it reflects a failure of extinction memory retrieval due to the mismatch between the internal states at extinction and test 2. This is an example that overgeneralization of memory induces the failure of memory retrieval. 

On the other hand, App^NL-G-F mice exhibited higher CR in test 1, which is conventionally interpreted as a successful fear memory retrieval. When estimating their internal states, they would infer that their observation in test 1 well matches those under the acquisition latent causes, that is the overgeneralization of fear memory as shown by a higher posterior probability in acquisition latent causes in test 1 (Figure 4 – figure supplement 3). This is an example that over-generalization of memory does not always induce retrieval failure as we explained in the reply to comment #1. 

Reviewer #2 (Public review):

Summary:

This manuscript proposes that the use of a latent cause model for the assessment of memory-based tasks may provide improved early detection of Alzheimer's Disease as well as more differentiated mapping of behavior to underlying causes. To test the validity of this model, the authors use a previously described knock-in mouse model of AD and subject the mice to several behaviors to determine whether the latent cause model may provide informative predictions regarding changes in the observed behaviors. They include a well-established fear learning paradigm in which distinct memories are believed to compete for control of behavior. More specifically, it's been observed that animals undergoing fear learning and subsequent fear extinction develop two separate memories for the acquisition phase and the extinction phase, such that the extinction does not simply 'erase' the previously acquired memory. Many models of learning require the addition of a separate context or state to be added during the extinction phase and are typically modeled by assuming the existence of a new state at the time of extinction. The Niv research group, Gershman et al. 2017, have shown that the use of a latent cause model applied to this behavior can elegantly predict the formation of latent states based on a Bayesian approach, and that these latent states can facilitate the persistence of the acquisition and extinction memory independently. The authors of this manuscript leverage this approach to test whether deficits in the production of the internal states, or the inference and learning of those states, may be disrupted in knock-in mice that show both a build-up of amyloid-beta plaques and a deterioration in memory as the mice age.

Strengths:

I think the authors' proposal to leverage the latent cause model and test whether it can lead to improved assessments in an animal model of AD is a promising approach for bridging the gap between clinical and basic research. The authors use a promising mouse model and apply this to a paradigm in which the behavior and neurobiology are relatively well understood - an ideal situation for assessing how a disease state may impact both the neurobiology and behavior. The latent cause model has the potential to better connect observed behavior to underlying causes and may pave a road for improved mapping of changes in behavior to neurobiological mechanisms in diseases such as AD.

Weaknesses:

I have several substantial concerns which I've detailed below. These include important details on how the behavior was analyzed, how the model was used to assess the behavior, and the interpretations that have been made based on the model.

[#12] (1) There is substantial data to suggest that during fear learning in mice separate memories develop for the acquisition and extinction phases, with the acquisition memory becoming more strongly retrieved during spontaneous recovery and reinstatement. The Gershman paper, cited by the authors, shows how the latent causal model can predict this shift in latent states by allowing for the priors to decay over time, thereby increasing the posterior of the acquisition memory at the time of spontaneous recovery. In this manuscript, the authors suggest a similar mechanism of action for reinstatement, yet the model does not appear to return to the acquisition memory state after reinstatement, at least based on the examples shown in Figures 1 and 3. Rather, the model appears to mainly modify the weights in the most recent state, putatively the 'extinction state', during reinstatement. Of course, the authors must rely on how the model fits the data, but this seems problematic based on prior research indicating that reinstatement is most likely due to the reactivation of the acquisition memory. This may call into question whether the model is successfully modeling the underlying processes or states that lead to behavior and whether this is a valid approach for AD.

We thank the reviewer for insightful comments. 

We agree that, as demonstrated in Gershman et al. (2017), the latent cause model accounts for spontaneous recovery via the inference of new latent causes during extinction and the temporal compression property provided by the prior. Moreover, it was also demonstrated that even a relatively low posterior can drive behavioral expression if the weight in the acquisition latent cause is preserved. For example, when the interval between retrieval and extinction was long enough that acquisition latent cause was not dominant during extinction, spontaneous recovery was observed despite the posterior probability of acquisition latent cause (C1) remaining below 0.1 in Figure 11D of Gershman et al. (2017). 

In our study, a high response in test 3 (reinstatement) is explained by both acquisition and extinction latent cause. The former preserves the associative weight of the initial fear memory, while the latter has w_context learned in the unsignaled shock phase. These positive w were weighted by their posterior probability and together contributed to increased expected shock in test 3. Though the posterior probability of acquisition latent cause was lower than extinction latent cause in test 3 due to time passage, this would be a parallel instance mentioned above. To clarify their contributions to reinstatement, we have conducted additional simulations and the discussion in reply to the reviewer’s next comment (see the reply to comment #13).

We recognize that our results might appear to deviate from the notion that reinstatement results from the strong reactivation of acquisition memory, where one would expect a high posterior probability of the acquisition latent cause. However, we would like to emphasize that the return of fear emerges from the interplay of competing memories. Previous studies have shown that contextual or cued fear reinstatement involves a neural activity switch back to fear state in the medial prefrontal cortex (mPFC), including the prelimbic cortex and infralimbic cortex, and the amygdala, including ventral intercalated amygdala neurons (ITCv), medial subdivision of central nucleus of the amygdala (CeM), and the basolateral amygdala (BLA) (Giustino et al., 2019; Hitora-Imamura et al., 2015; Zaki et al., 2022). We speculate that such transition is parallel to the internal states change in the latent cause model in terms of posterior probability and associative weight change.

Optogenetic manipulation experiments have further revealed how fear and extinction engrams contribute to extinction retrieval and reinstatement. For instance, Gu et al. (2022) used a cued fear conditioning paradigm and found that inhibition of extinction engrams in the BLA, ventral hippocampus (vHPC), and mPFC after extinction learning artificially increased freezing to the tone cue. Similar results were observed in contextual fear conditioning, where silencing extinction engrams in the hippocampus dentate gyrus (DG) impaired extinction retrieval (Lacagnina et al., 2019). These results suggest that the weakening extinction memory can induce a return of fear response even without a reminder shock. On the other hand, Zaki et al. (2022) showed that inhibition of fear engrams in the BLA, DG, or hippocampus CA1 attenuated contextual fear reinstatement. However, they also reported that stimulation of these fear engrams was not sufficient to induce reinstatement, suggesting these fear engram only partially account for reinstatement. 

In summary, reinstatement likely results from bidirectional changes in the fear and extinction circuits, supporting our interpretation that both acquisition and extinction latent causes contribute to the reinstatement. Although it remains unclear whether these memory engrams represent latent causes, one possible interpretation is that w_context update in extinction latent causes during unsignaled shock indicates weakening of the extinction memory, while preservation of w in acquisition latent causes and their posterior probability suggests reactivation of previous fear memory. 

[#13] (2) As stated by the authors in the introduction, the advantage of the fear learning approach is that the memory is modified across the acquisition-extinction-reinstatement phases. Although perhaps not explicitly stated by the authors, the post-reinstatement test (test 3) is the crucial test for whether there is reactivation of a previously stored memory, with the general argument being that the reinvigorated response to the CS can't simply be explained by relearning the CS-US pairing, because re-exposure the US alone leads to increase response to the CS at test. Of course there are several explanations for why this may occur, particularly when also considering the context as a stimulus. This is what I understood to be the justification for the use of a model, such as the latent cause model, that may better capture and compare these possibilities within a single framework. As such, it is critical to look at the level of responding to both the context alone and to the CS. It appears that the authors only look at the percent freezing during the CS, and it is not clear whether this is due to the contextual US learning during the US re-exposure or to increased response to the CS - presumably caused by reactivation of the acquisition memory. For example, the instance of the model shown in Figure 1 indicates that the 'extinction state', or state z6, develops a strong weight for the context during the reinstatement phase of presenting the shock alone. This state then leads to increased freezing during the final CS probe test as shown in the figure. By not comparing the difference in the evoked freezing CR at the test (ITI vs CS period), the purpose of the reinstatement test is lost in the sense of whether a previous memory was reactivated - was the response to the CS restored above and beyond the freezing to the context? I think the authors must somehow incorporate these different phases (CS vs ITI) into their model, particularly since this type of memory retrieval that depends on assessing latent states is specifically why the authors justified using the latent causal model.

To clarify the contribution of context, we have provided preCS freezing rate across trials in Figure 2 – figure supplement 2. As the reviewer pointed out, the preCS freezing rate did not remain at the same level across trials, especially within the 12-month-old control and App^NL-G-F group (Figure 2 – figure supplement 2A and 2B), suggesting the effect context. A paired samples t-test comparing preCS freezing (Figure 2 – figure supplement 2E) and CS freezing (Figure 2E) in test 3 revealed significant differences in all groups: 6-month-old control, t(23) = -6.344, p < 0.001, d = -1.295; 6-month-old App^NL-G-F, t(24) = -4.679, p < 0.001, d = -0.936; 12-month-old control, t(23) = -4.512, p < 0.001, d = 0.921; 12-month-old App^NL-G-F, t(24) = -2.408, p = 0.024, d = -0.482. These results indicate that the response to CS was above and beyond the response to context only. We also compared the change in freezing rate (CS freezing rate minus preCS freezing rate) in test 2 and test 3 to examine the net response to the tone. The significant difference was found in the control group, but not in the App^NL-GF group (Author response image 3). The increased net response to the tone in the control group suggested that the reinstatement was partially driven by reactivation of acquisition memory, not solely by the contextual US learning during the unsignaled shock phase. We have added these results and discussion in the manuscript Line 220-231.

Author response image 3.

Net freezing rate in test 2 and test 3. Net freezing rate is defined as the CS freezing rate (i.e., freezing rate during 1 min CS presentation) minus the preCS freezing rate (i.e., 1 min before CS presentation). The dashed horizontal line indicates no freezing rate change from the preCS period to the CS presentation. *p < 0.05 by paired-sample Student’s t-test, and the alternative hypothesis specifies that test 2 freezing rate change is less than test 3. Colors indicate different groups: orange represents 6-month-old control (n = 24), light blue represents 6-month-old App^NL-G-F mice (n = 25), pink represents 12-month-old control (n = 24), and dark blue represents 12-month-old App^NL-G-F mice (n = 25). Each black dot represents one animal. Statistical results were as follows: t(23) = -1.927, p = 0.033, Cohen’s d = -0.393 in 6-month-old control; t(24) = -1.534, p = 0.069, Cohen’s d = -0.307 in 6-month-old App^NL-G-F; t(23) = -1.775, p = 0.045, Cohen’s d = -0.362 in 12-month-old control; t(24) = 0.86, p = 0.801, Cohen’s d = 0.172 in 12-monthold App^NL-G-F. 

According to the latent cause model, if the reinstatement is merely induced by an association between the context and the US in the unsignaled shock phase, the CR given context only and that given context and CS in test 3 should be equal. However, the simulation conducted for each mouse using their estimated parameters confirmed that this was not the case in this study. The results showed that simulated CR was significantly higher in the context+CS condition than in the context only condition (Author response image 4). This trend is consistent with the behavioral results we mentioned above.

Author response image 4.

Simulation of context effect in test 3. Estimated parameter sets of each sample were used to run the simulation that only context or context with CS was present in test 3 (trial 36). The data are shown as median with interquartile range, where white bars with colored lines represent CR for context only and colored bars represent CR for context with CS. Colors indicate different groups: orange represents 6-month-old control (n = 15), light blue represents 6-month-old App^NL-G-F mice (n = 12), pink represents 12-month-old control (n = 20), and dark blue represents 12-month-old App^NL-G-F mice (n = 18). Each black dot represents one animal. **p < 0.01, and ***p < 0.001 by Wilcoxon signed-rank test comparing context only and context + CS in each group, and the alternative hypothesis specifies that CR in context is not equal to CR in context with CS. Statistical results were as follows: W = 15, p = 0.008, effect size r = -0.66 in 6-month-old control; W = 0, p < 0.001, effect size r = -0.88 in 6-month-old App^NL-G-F; W = 25, p = 0.002, effect size r = -0.67 in 12-month-old control; W = 9, p = 0.002 , effect size r = -0.75 in 12-month-old App^NL-G-F. 

[#14] (3) This is related to the second point above. If the question is about the memory processes underlying memory retrieval at the test following reinstatement, then I would argue that the model parameters that are not involved in testing this hypothesis be fixed prior to the test. Unlike the Gershman paper that the authors cited, the authors fit all parameters for each animal. Perhaps the authors should fit certain parameters on the acquisition and extinction phase, and then leave those parameters fixed for the reinstatement phase. To give a more concrete example, if the hypothesis is that AD mice have deficits in differentiating or retrieving latent states during reinstatement which results in the low response to the CS following reinstatement, then perhaps parameters such as the learning rate should be fixed at this point. The authors state that the 12-month-old AD mice have substantially lower learning rate measures (almost a 20-fold reduction!), which can be clearly seen in the very low weights attributed to the AD mouse in Figure 3D. Based on the example in Figure 3D, it seems that the reduced learning rate in these mice is most likely caused by the failure to respond at test. This is based on comparing the behavior in Figures 3C to 3D. The acquisition and extinction curves appear extremely similar across the two groups. It seems that this lower learning rate may indirectly be causing most of the other effects that the authors highlight, such as the low σx, and the changes to the parameters for the CR. It may even explain the extremely high K. Because the weights are so low, this would presumably lead to extremely low likelihoods in the posterior estimation, which I guess would lead to more latent states being considered as the posterior would be more influenced by the prior.

We thank the reviewer for the suggestion about fitting and fixing certain parameters in different phases.

However, this strategy may not be optimal for our study for the following scientific reasons.

Our primary purpose is to explore internal states in the memory modification process that are associated with the deficit found in App^NL-G-F mice in the reinstatement paradigm. We did not restrict the question to memory retrieval, nor did we have a particular hypothesis such that only a few parameters of interest account for the impaired associative learning or structure learning in App^NL-G-F mice while all other parameters are comparable between groups. We are concerned that restricting questions to memory retrieval at the test is too parsimonious and might lead to misinterpretation of the results. As we explain in reply to comment #5, removing trials in extinction during parameter estimation reduces the model fit performance and runs the risk of overfitting within the individual. Therefore, we estimated all parameters for each animal, with the assumption that the estimated parameter set represents individual internal state (i.e., learning and memory characteristics) and should be fixed within the animal across all trials.  

Figure 3 is the parameter estimation and simulation results using the median data of each group as an individual. The estimated parameter value is one of the possible cases in that group to demonstrate how a typical learning curve fits the latent cause model. The reviewer mentioned “20-fold reduction in learning rate” is the comparison of two data points, not the actual comparison between groups. The comparison between control and App^NL-G-F mice in the 12-month-old group for all parameters was provided in Table S7. The Mann-Whitney U test did not reveal a significant difference in learning rate (η): 12-month-old control (Mdn = 0.09, IQR=0.23) vs. 12-month-old App^NL-G-F (Mdn = 0.12, IQR=0.23), U = 199, p = 0.587.  

We agree that lower learning rate could bias the learning toward inferring a new latent cause. However, this tendency may depend on the value of other parameters and varied in different phases in the reinstatement paradigm. Here, we used ⍺ as an example and demonstrate their interaction in Appendix 2 – table 2 with relatively extreme values: ⍺ \= {1, 3} and η \= {0.01, 0.5} while the rest of the parameters fixed at the initial guess value. 

When ⍺ = 1, the number of latent causes across phases (K_acq, K_ext, K_rem) remain unchanged and their posterior probability in test 3 were comparable even if η increased from 0.01 to 0.5. This is an example that lower η does not lead to inferring new latent causes because of low ⍺. The effect of low learning rate manifests in test 3 CR due to low w_{context, acq} and w_{context, ext}. 

When ⍺ = 3, the number of acquisition latent causes (K_acq) was higher in the case of η = 0.01 than that of η = 0.5, showing the effect mentioned by the reviewer. However, test 1 CR is much lower when η = 0.01, indicating unsuccessful learning even after inferring a new latent cause. This is none of the cases observed in this study. During extinction phases, the effect of η is surpassed by the effect of high ⍺, where the number of extinction latent causes (K_ext) is high and not affected by η. After the extinction phases, the effect of K kicks in as the total number of latent causes reaches its value (K = 33 in this example), especially in the case of η = 0.01. A new latent cause is inferred after extinction in the condition of η = 0.5, but the CR 3 is still high as the w_{context, acq} and w_{context_, ext_} are high. This is an example that a new latent cause is inferred in spite of higher η. 

Overall, the learning rate would not have a prominent effect alone throughout the reinstatement paradigm, and it has a joint effect with other parameters. Note that the example here did not cover our estimated results, as the estimated learning rate was not significantly different between control and App^NL-G-F mice (see above). Please refer to the reply to comment #31 for more discussion about the interaction among parameters when the learning rate is fixed. We hope this clarifies the reviewer’s concern.

[#15] (4) Why didn't the authors use the latent causal model on the Barnes maze task? The authors mention in the discussion that different cognitive processes may be at play across the two tasks, yet reversal tasks have been suggested to be solved using latent states to be able to flip between the two different task states. In this way, it seems very fitting to use the latent cause model. Indeed, it may even be a better way to assess changes in σx as there are presumably 12 observable stimuli/locations.

Please refer to our provisional response about the application of the latent cause model to the reversal Barnes maze task. Briefly, it would be difficult to directly apply the latent cause model to the Barnes maze data because this task involves operant learning, and thereby almost all conditions in the latent cause model are not satisfied. Please also see our reply to comment #24 for the discussion of the link between the latent cause model and Barnes maze task. 

Reviewer #2 (Recommendations for the authors):

[#16] (1) I had a bit of difficulty finding all the details of the model. First, I had to mainly rely on the Gershman 2017 paper to understand the model. Even then, there were certain aspects of the model that were not clear. For instance, it's not quite clear to me when the new internal states are created and how the maximum number of states is determined. After reading the authors' methods and the Gershman paper, it seems that a new internal state is generated at each time point, aka zt, and that the prior for that state decays onwards from alpha. Yet because most 'new' internal states don't ever take on much of a portion of the posterior, most of these states can be ignored. Is that a correct understanding? To state this another way, I interpret the equation on line 129 to indicate that the prior is determined by the power law for all existing internal states and that each new state starts with a value of alpha, yet I don't see the rule for creating a new state, or for iterating k other than that k iterates at each timestep. Yet this seems to not be consistent with the fact that the max number of states K is also a parameter fit. Please clarify this, or point me to where this is better defined.

I find this to be an important question for the current paper as it is unclear to me when the states were created. Most notably, in Figure 3, it's important to understand why there's an increase in the posterior of z⁵ in the AD 12-month mice at test. Is state z⁵ generated at trial 5? If so, the prior would be extremely small by trial 36, making it even more perplexing why z⁵ has such a high posterior. If its weights are similar to z³ and z⁴, and they have been much more active recently, why would z⁵ come into play?

We assume that the “new internal state" the reviewer is referring to is the “new latent cause." We would like to clarify that “internal state" in our study refers to all the latent causes at a given time point and observation. As this manuscript is submitted as a Research Advance article in eLife, we did not rephrase all the model details. Here, we explain when a new latent cause is created (i.e., the prior probability of a new latent cause is greater than 0) with the example of the 12-month-old group (Figure 3C and 3D). 

Suppose that before the start of each trial, an agent inferred the most likely latent cause with maximum posterior, and it inferred k latent causes so far. A new latent cause can be inferred at the computation of the prior of latent causes at the beginning of each trial.  

In the latent cause model, it follows a distance-dependent Chinese Restaurant Process (CRP; Blei and Frazier, 2011). The prior of each old latent cause is its posterior probability, which is the final count of the EM update before the current. In addition, the prior of old latent causes is sensitive to the time passage so that it exponentially decreases as a forgetting function modulated by g (see Figure 2 in Gershman et al., 2017). Simultaneously, the prior of a new cause is assigned ⍺. The new latent cause is inferred at this moment. Hence, the prior of latent causes is jointly determined by ⍺, g and its posterior probability. The maximum number of latent causes K is set a priori and does not affect the prior while k < K (see also reply to comment #30 for the discussion of boundary set for K and comment #31 for the discussion of the interaction between ⍺ and K). Note that only one new latent cause can be inferred in each trial, and (k+1)^th latent cause, which has never been inferred so far, is chosen as the new latent cause.

In our manuscript, the subscript number in zₖ denotes the order in which they were inferred, not the trial number. In Figures 3C and 3D, z₃ and z₄ were inferred in trials 5 and 6 during extinction; z₅ is a new latent cause inferred in trial 36. Therefore, the prior of z₅ is not extremely small compared to z₄ and z₃.

In both control and App^NL-G-F mice in the 12-month-old (Figures 3C and 3D), z³ is dominant until trial 35. The unsignaled shock at trial 35 generates a large prediction error as only context is presented and followed by the US. This prediction error reduces posterior of z³, while increasing the posterior of z⁴ and w_context in z³ and z⁴. This decrease of posterior of z³ is more obvious in the App^NL-G-F than in the control group, prompting them to infer a new latent cause z⁵ (Figure 3C and 3D). Although Figure 3C and 3D are illustrative examples as we explained in the reply to comment #14, this interpretation would be plausible as the App^NL-G-F group inferred a significantly larger number of latent causes after the extinction with slightly higher posteriors of them than those in the control group (Figure 4E).

[#17] (2) Related to the above, Are the states zA and zB defined by the authors to help the reader group the states into acquisition and extinction states, or are they somehow grouped by the model? If the latter is true, I don't understand how this would occur based on the model. If the former, could the authors state that these states were grouped together by the author?

We used zA and zB annotations to assist with the explanation, so this is not grouped by the model. We have stated this in the manuscript Line 181-182.

[#18] (3) This expands on the third point above. In Figure 3D, internal states z³, z⁴, and z⁵ appear to be pretty much identical in weights in the App group. It's not clear to me why then the posterior of z⁵ would all of a sudden jump up. If I understand correctly, the posterior is the likelihood of the observations given the internal state (presumably this should be similar across z³,z⁴, and z⁵), multiplied by the prior of the state. Z3 and Z4 are the dominant inferred states up to state 36. Why would z⁵ become more likely if there doesn't appear to be any error? I'm inferring no error because there are little or no changes in weights on trial 36, most prominently no changes in z³ which is the dominant internal state in step 36. If there's little change in weights, or no errors, shouldn't the prior dominate the calculation of the posterior which would lead to z³ and z⁴ being most prominent at trial 36?

We have explained how z⁵ of the 12-month-old App^NL-G-F was inferred in the reply to comment #16. Here, we explain the process underlying the rapid changes of the posterior of z³, z⁴, and z⁵ from trial 35 to 36.

During the extinction, the mice inferred z³ given the CS and the context in the absence of US. In trial 35, they observed the context and the unsignaled shock in the absence of the CS. This reduced the likelihood for the CS under z³ and thereby the posterior of z³, while relatively increasing the posterior of z⁴. The associative weight between the context and the US , w_context, indeed increased in both z³ and z⁴, but w_context of z⁴ was updated more than that of z³ due to its higher posterior probability. At the beginning of trial 36, a new latent cause z⁵ was inferred with a certain prior (see also the reply for comment #16), and w₅ = w₀, where w₀ is the initial value of weight. After normalizing the prior over latent causes, the emergence of z⁵ reduced the prior probability of other latent causes compared to the case where the prior of z⁵ is 0. Since the CS was presented while the US was absent in trial 36, the likelihood of the CS and that of the US under z³, and especially z⁴, given the cues and w became lower than the case in which z⁵ has not been inferred yet. Consequently, the posterior of z⁵ became salient (Figure 3D).

To maintain consistency across panels, we used a uniform y-axis range. However, we acknowledge that this may make it harder to notice the changes of associative weights in Figure 3D. We have provided the subpanel in Figure 3D with a smaller y-axis limit to reveal the weight changes at trial 35 in Author response image 5.

Author response image 5.

Magnified view of w_context and wCS in the last 3 trials in Figure 3D. The graph format is the same as in Figure 3D. The weight for CS (_w_CS) and that for context (w_context) in each latent cause across trial 34 (test 2), 35 (unsignaled shock), and 36 (test 3) in 12-month-old App^NL-G-F in Figure 3D was magnified in the upper and lower magenta box, respectively.

[#19] (8) In Figure 4B - The figure legend didn't appear to indicate at which time points the DIs are plotted.

We have amended the figure legend to indicate that DI between test 3 and test 1 is plotted.

[#20] (9) Lines 301-303 state that the posterior probabilities of the acquisition internal states in the 12month AD mice were much higher at test 1 and that this resulted in different levels of CR across the control and 12-month App group. This is shown in the Figure 4A supplement, but this is not apparent in Figure 3 panels C and D. Is the example shown in panel D not representative of the group? The CRs across the two examples in Figure 3 C and D look extremely similar at test 1. Furthermore, the posteriors of the internal states look pretty similar across the two groups for the first 4 trials. Both the App and control have substantial posterior probabilities for the acquisition period, I don't see any additional states at test 1. The pattern of states during acquisition looks strikingly similar across the two groups, whereas the weights of the stimuli are considerably different. I think it would help the authors to use an example that better represents what the authors are referring to, or provide data to illustrate the difference. Figure 4C partly shows this, but it's not very clear how strong the posteriors are for the 3rd state in the controls.

Figure 3 serves as an example to explain the internal states in each group (see also the third paragraph in the reply to comment #14). Figure 4C to H showed the results from each sample for between-group comparison in selected features. Therefore, the results of direct comparisons of the parameter values and internal states between genotypes in Figure 3 are not necessarily the same as those in Figure 4. Both examples in Figure 3C and 3D inferred 2 latent causes during the acquisition. In terms of posterior till test 1 (trial 4), the two could be the same. However, such examples were not rare, as the proportion of the mice that inferred 2 latent causes during the acquisition was slightly lower than 50% in the control, and around 90% in the App^NL-G-F mice (Figure 4C). The posterior probability of acquisition latent cause in test 1 showed a similar pattern (Figure 4 – figure supplement 3), with values near 1 in around 50% of the control mice and around 90% of the App^NL-G-F mice.  

[#21] (10) Line 320: This is a confusing sentence. I think the authors are saying that because the App group inferred a new state during test 3, this would protect the weights of the 'extinction' state as compared to the controls since the strength of the weight updates depends on the probability of the posterior.

In order to address this, we have revised this sentence to “Such internal states in App^NL-G-F mice would diverge the associative weight update from those in the control mice after extinction.” in the manuscript Line 349-351.

[#22] (11) In lines 517-519 the authors address the difference in generalizing the occurrence of stimuli across the App and control groups. It states that App mice with lower alpha generalized observations to an old cause rather than attributing it as a new state. Going back to statement 3 above, I think it's important to show that the model fit of a reduction in alpha does not go hand-in-hand with a reduction in the learning rates and hence the weights. Again, if the likelihoods are diminished due to the low weights, then the fit of alpha might be reduced as well. To reiterate my point above, if the observations in changes in generalization and differentiation occur because of a reduction in the learning rate, the modeling may not be providing a particularly insightful understanding of AD, other than that poor learning leads to ineffectual generalization and differentiation. Do these findings hold up if the learning rates are more comparable across the control and App group?

These findings were explained on the basis of comparable learning rates between control and App^NL-GF mice in the 12-month-old group (see the reply to comment #14). In addition, we have conducted simulation for different ⍺ and σ_x² values under the condition of the fixed learning rate, where overgeneralization and overdifferentaiton still occurred (see the reply to comment #26).  

[#23] (12) Lines 391 - 393. This is a confusing sentence. "These results suggest that App NL-G-F mice could successfully form a spatial memory of the target hole, while the memory was less likely to be retrieved by a novel observation such as the absence of the escape box under the target hole at the probe test 1." The App mice show improved behavior across days of approaching the correct hole. Is this statement suggesting that once they've approached the target hole, the lack of the escape box leads to a reduction in the retention of that memory?

We speculated that when the mice observed the absence of the escape box, a certain prediction error would be generated, which may have driven the memory modification. In App^NL-G-F mice, such modification, either overgeneralization or overdifferentiation, could render the memory of the target hole vulnerable; if overgeneralization occurred, the memory would be quickly overwritten as the goal no longer exists in this position in this maze, while if overdifferentiation occurred, a novel memory such that the goal does not exist in the maze different from previous one would be formed. In either case of misclassification, the probability of retrieving the goal position would be reduced. To reduce ambiguity in this sentence, we have revised the description in the manuscript Line 432-434 as follows: “These results suggest that App^NL-G-F mice could successfully form a spatial memory of the target hole, while they did not retrieve the spatial memory of the target hole as strongly as control mice when they observed the absence of the escape box during the probe test.”

[#24] (13) The connection between the results of Barnes maze and the fear learning paradigm is weak. How can changes in overgeneralization due to a reduction in the creation of inferred states and differentiation due to a reduced σx lead to the observations in the Barnes maze experiment?

We extrapolated our interpretation in the reinstatement modeling to behaviors in a different behavioral task, to explore the explanatory power of the latent cause framework formalizing mechanisms of associative learning and memory modification. Here, we explain the results of the reversal Barnes maze paradigm in terms of the latent cause model, while conferring the reinstatement paradigm.

Whilst we acknowledge that fear conditioning and spatial learning are not fully comparable, the reversal Barnes maze paradigm used in our study shares several key learning components with the reinstatement paradigm. 

First, associative learning is fundamental in spatial learning (Leising & Blaisdell, 2009; Pearce, 2009). Although we did not make any specific assumptions of what kind of associations were learned in the Barnes maze, performance improvements in learning phases likely reflect trial-and-error updates of these associations involving sensory preconditioning or secondary conditioning. Second, the reversal training phases could resemble the extinction phase in the reinstatement paradigm, challenge previously established memory. In terms of the latent cause model, both the reversal learning phase in the reversal Barnes maze paradigm and the extinction phase in the reinstatement paradigm induce a mismatch of the internal state. This process likely introduces large prediction errors, triggering memory modification to reconcile competing memories.  

Under the latent cause framework, we posit that the mice would either infer new memories or modify existing memories for the unexpected observations in the Barnes maze (e.g., changed location or absence of escape box) as in the reinstatement paradigm, but learn a larger number of association rules between stimuli in the maze compared to those in the reinstatement. In the reversal Barnes maze paradigm, the animals would infer that a latent cause generates the stimuli in the maze at certain associative weights in each trial, and would adjust behavior by retaining competing memories.

Both overgeneralization and overdifferentiation could explain the lower exploration time of the target hole in the App^NL-G-F mice in probe test 1. In the case of overgeneralization, the mice would overwrite the existing spatial memory of the target hole with a memory that the escape box is absent. In the case of overdifferentiation, the mice would infer a new memory such that the goal does not exist in the novel field, in addition to the old memory where the goal exists in the previous field. In both cases, the App^NL-G-F mice would not infer that the location of the goal is fixed at a particular point and failed to retain competing spatial memories of the goal, leading to relying on a less precise, non-spatial strategy to solve the task.  

Since there is no established way to formalize the Barnes maze learning in the latent cause model, we did not directly apply the latent cause model to the Barnes maze data. Instead, we used the view above to explore common processes in memory modification between the reinstatement and the Barnes maze paradigm. 

The above description was added to the manuscript on page 13 (Line 410-414) and page 19-20 (Line 600-602, 626-639).

[#25] (14) In the fear conditioning task, it may be valuable to separate responding to the context and the cue at the time of the final test. The mice can learn about the context during the reinstatement, but there must be an inference to the cue as it's not present during the reinstatement phase. This would provide an opportunity for the model to perhaps access a prior state that was formed during acquisition. This would be more in line with the original proposal by Gershman et al. 2017 with spontaneous recovery.

Please refer to the reply to comment #13 regarding separating the response to context in test 3.  

Reviewer #3 (Public review):

Summary:

This paper seeks to identify underlying mechanisms contributing to memory deficits observed in Alzheimer's disease (AD) mouse models. By understanding these mechanisms, they hope to uncover insights into subtle cognitive changes early in AD to inform interventions for early-stage decline.

Strengths:

The paper provides a comprehensive exploration of memory deficits in an AD mouse model, covering the early and late stages of the disease. The experimental design was robust, confirming age-dependent increases in Aβ plaque accumulation in the AD model mice and using multiple behavior tasks that collectively highlighted difficulties in maintaining multiple competing memory cues, with deficits most pronounced in older mice.

In the fear acquisition, extinction, and reinstatement task, AD model mice exhibited a significantly higher fear response after acquisition compared to controls, as well as a greater drop in fear response during reinstatement. These findings suggest that AD mice struggle to retain the fear memory associated with the conditioned stimulus, with the group differences being more pronounced in the older mice.

In the reversal Barnes maze task, the AD model mice displayed a tendency to explore the maze perimeter rather than the two potential target holes, indicating a failure to integrate multiple memory cues into their strategy. This contrasted with the control mice, which used the more confirmatory strategy of focusing on the two target holes. Despite this, the AD mice were quicker to reach the target hole, suggesting that their impairments were specific to memory retrieval rather than basic task performance.

The authors strengthened their findings by analyzing their data with a leading computational model, which describes how animals balance competing memories. They found that AD mice showed somewhat of a contradiction: a tendency to both treat trials as more alike than they are (lower α) and similar stimuli as more distinct than they are (lower σx) compared to controls.

Weaknesses:

While conceptually solid, the model struggles to fit the data and to support the key hypothesis about AD mice's ability to retain competing memories. These issues are evident in Figure 3:

[#26] (1) The model misses key trends in the data, including the gradual learning of fear in all groups during acquisition, the absence of a fear response at the start of the experiment, the increase in fear at the start of day 2 of extinction (especially in controls), and the more rapid reinstatement of fear observed in older controls compared to acquisition.

We acknowledge these limitations and explained why they arise in the latent cause model as follows.

a. Absence of a fear response at the start of the experiment and the gradual learning of fear during acquisition 

In the latent cause model, the CR is derived from a sigmoidal transformation from the predicted outcome with the assumption that its mapping to behavioral response may be nonlinear (see Equation 10 and section “Conditioned responding” in Gershman et al., 2017). 

The magnitude of the unconditioned response (trial 1) is determined by w₀, θ, and λ. An example was given in Appendix 2 – table 3. In general, a higher w₀ and a lower θ produce a higher trial 1 CR when other parameters are fixed. During the acquisition phase, once the expected shock exceeds θ, CR rapidly approaches 1, and further increases in expected shock produce few changes in CR. This rapid increase was also evident in the spontaneous recovery simulation (Figure 11) in Gershman et al. (2017). The steepness of this rapid increase is modulated by λ such that a higher value produces a shallower slope. This is a characteristic of the latent cause model, assuming CR follows a sigmoid function of expected shock, while the ordinal relationship over CRs is maintained with or without the sigmoid function, as Gershman et al. (2017) mentioned. If one assumes that the CR should be proportional to the expected shock, the model can reproduce the gradual response as a linear combination of w and posteriors of latent causes while omitting the sigmoid transformation (Figure 3). 

b. Increase in fear at the start of day 2 extinction

This point is partially reproduced by the latent cause model. As shown in Figure 3, trial 24 (the first trial of day 2 extinction) showed an increase in both posterior probability of latent cause retaining fear memory and the simulated CRs in all groups except the 6-month-old control group, though the increase in CR was small due to the sigmoid transformation (see above). This can be explained by the latent cause model as 24 h time lapse between extinction 1 and 2 decreases the prior of the previously inferred latent cause, leading to an increase of those of other latent causes. 

Unlike other groups, the 6-month-old control did not exhibit increased observed CR at trial 24

but at trial 25 (Figure 3A). The latent cause model failed to reproduce it, as there was no increase in posterior probability in trial 24 (Figure 3A). This could be partially explained by the low value of g, which counteracts the effect of the time interval between days: lower g keeps prior of the latent causes at the same level as those in the previous trial. Despite some failures in capturing this effect, our fitting policy was set to optimize prediction among the test trials given our primary purpose of explaining reinstatement.

c. more rapid reinstatement of fear observed in older controls compared to acquisition

We would like to point out that this was replicated by the latent cause model as shown in Figure 3 – figure supplement 1C. The DI between test 3 and test 1 calculated from the simulated CR was significantly higher in 12-month-old control than in App^NL-G-F mice (cf. Figure 2C to E).  

[#27] (2) The model attributes the higher fear response in controls during reinstatement to a stronger association with the context from the unsignaled shock phase, rather than to any memory of the conditioned stimulus from acquisition. These issues lead to potential overinterpretation of the model parameters. The differences in α and σx are being used to make claims about cognitive processes (e.g., overgeneralization vs. overdifferentiation), but the model itself does not appear to capture these processes accurately. The authors could benefit from a model that better matches the data and that can capture the retention and recollection of a fear memory across phases.

First, we would like to clarify that the latent cause model explains the reinstatement not only by the extinction latent cause with increased w_context but also the acquisition latent cause with preserved wCS and w_context (see also reply to comment #13). Second, the latent cause model primarily attributes the higher fear reinstatement in control to a lower number of latent causes inferred after extinction (Figure 4E) and higher w_context in extinction latent cause (Figure 4G). We noted that there was a trend toward significance in the posterior probability of latent causes inferred after extinction (Figure 4E), which in turn influences those of acquisition latent causes. Although the posterior probability of acquisition latent cause appeared trivial and no significance was detected between control and App^NL-G-F mice (Figure 4C), it was suppressed by new latent causes in App^NL-G-F mice (Author response image 6).

This indicates that App^NL-G-F mice retrieved acquisition memory less strongly than control mice. Therefore, we argue that the latent cause model attributed a higher fear response in control during reinstatement not solely to the stronger association with the context but also to CS fear memory from acquisition. Although we tested whether additional models fit the reinstatement data in individual mice, these models did not satisfy our fitting criteria for many mice compared to the latent cause model (see also reply to comment #4 and #28).

Author response image 6.

Posterior probability of acquisition, extinction, and after extinction latent causes in test 3. The values within each bar indicate the mean posterior probability of acquisition latent cause (darkest shade), extinction latent cause (medium shade), and latent causes inferred after extinction (lightest shade) in test 3 over mice within genotype. Source data are the same as those used in Figure 4C–E (posterior of z).

Conclusion:

Overall, the data support the authors' hypothesis that AD model mice struggle to retain competing memories, with the effect becoming more pronounced with age. While I believe the right computational model could highlight these differences, the current model falls short in doing so.

Reviewer #3 (Recommendations for the authors):

[#28] Other computational models may better capture the data. Ideally, I'd look for a model that can capture the gradual learning during acquisition, and, in some mice, the inferring of a new latent cause during extinction, allowing the fear memory to be retained and referenced at the start of day 2 extinction and during later tests.

We have further evaluated another computational model, the latent state model, and compared it with the latent cause model. The simulation of reinstatement and parameter estimation method of the latent state model were described in the Appendix.

The latent state model proposed by Cochran and Cisler (2019) shares several concepts with the latent cause model, and well replicates empirical data under certain conditions. We expect that it can also explain the reinstatement. 

Following the same analysis flow for the latent cause model, we estimated the parameters and simulated reinstatement in the latent state model from individual CRs and median of them. In the median freezing rate data of the 12-month-old control mice, the simulated CR replicated the observed CR well and exhibited the ideal features that the reviewer looked for: gradual learning during acquisition and an increased fear at the start of the second-day extinction (Appendix 1 – figure 1G). However, a lot of samples did not fit well to the latent state model. The number of anomalies was generally higher than that in the latent cause model (Appendix 1 – figure 2). Within the accepted samples, the sum of squared prediction error in all trials was significantly lower in the latent state model, which resulted from lower prediction error in the acquisition trials (Appendix 1 – figure 4A and 4B). In the three test trials, the squared prediction error was comparable between the latent state model and the latent cause model except for the test 2 trials in the control group (Appendix 1 – figure 4A and 4B, rightmost panel). On the other hand, almost all accepted samples continued to infer the acquisition latent states during extinction without inferring new states (Appendix 1 – figure 5B and 5E, left panel), which differed from the ideal internal states the reviewer expected. While the latent state model fit performance seems to be better than the latent cause model, the accepted samples cannot reproduce the lower DI between test 3 and test 1 in aged App^NL-G-F mice (Appendix 1 – figure 6C). These results make the latent state model less suitable for our purpose and therefore we decided to stay with the latent cause model. It should also be noted that we did not explore all parameter spaces of the latent state model hence we cannot rule out the possibility that alternative parameter sets could provide a better fit and explain the memory modification process well. A more comprehensive parameter search in the LSM may be a valuable direction for future research.

If you decide not to go with a new model, my preference would be to drop the current modeling. However, if you wish to stay with the current model, I'd like to see justification or acknowledgment of the following:

[#29] (1) Lower bound on alpha of 1: This forces the model to infer new latent causes, but it seems that some mice, especially younger AD mice, might rely more on classical associative learning (e.g., Rescorla-Wagner) rather than inferring new causes.

We acknowledge that the default value set in Gershman et al. (2017) is 0.1, and the constraint we set is a much higher value. However, ⍺ = 1 does not always force the model to infer new latent causes.

In the standard form Chinese restaurant process (CRP), the prior that n^th observation is assigned to a new cluster is given by ⍺ / (n - 1 + ⍺) (Blei & Gershman, 2012). When ⍺ = 1, the prior of the new cluster for the 2nd observation will be 0.5; when ⍺ = 3, this prior increases to 0.75. Thus, when ⍺ > 1, the prior of the new cluster is above chance early in the sequence, which may relate to the reviewer’s concern. However, this effect diminishes as the number of observations increases. For instance, the prior of the new cluster drops to 0.1 and 0.25 for the 10th observation when ⍺ = 1 and 3, respectively. Furthermore, the prior in the latent cause model is governed by not only α but also g, a scaling parameter for the temporal difference between successive observations (see Results in the manuscript) following “distance-dependent” CRP, then normalized over all latent causes including a new latent cause. Thus, it does not necessarily imply that ⍺ greater than 1 forces agents to infer a new latent cause_. As shown in Appendix 2 – table 4, the number of latent causes does not inflate in each trial when _α = 1. On the other hand, the high number of latent causes due to α = 2 can be suppressed when g = 0.01. More importantly, the driving force is the prediction error generated in each trial (see also comment #31 about the interaction between ⍺ and σ_x²). Raising the value of ⍺ per se can be viewed as increasing the probability to infer a new latent cause, not forcing the model to do so by higher α alone. 

During parameter exploration using the median behavioral data under a wider range of ⍺ with a lower boundary at 0.1, the estimated value eventually exceeded 1. Therefore, we set the lower bound of ⍺ to be 1 is to reduce inefficient sampling. 

[#30] (2) Number of latent causes: Some mice infer nearly as many latent causes as trials, which seems unrealistic.

We set the upper boundary for the maximum number of latent causes (K) to be 36 to align with the infinite features of CRP. This allowed some mice to infer more than 20 latent causes in total. When we checked the learning curves in these mice, we found that they largely fluctuated or did not show clear decreases during the extinction (Author response image 7, colored lines). The simulated learning curves were almost flat in these trials (Author response image 7, gray lines). It might be difficult to estimate the internal states of such atypical mice if the sampling process tried to fit them by increasing the number of latent causes. Nevertheless, most of the samples have a reasonable total number of latent causes: 12-month-old control mice, Mdn = 5, IQR = 4; 12-month-old App^NL-G-F mice, Mdn = 5, IQR = 1.75; 6-month-old control mice, Mdn = 7, IQR = 12.5; 6-month-old App^NL-G-F mice, Mdn = 5, IQR = 5.25. These data were provided in Tables S9 and S12.  

Author response image 7.

Samples with a high number of latent causes. Observed CR (colored line) and simulated CR (gray line) for individual samples with a total number of inferred latent causes exceeding 20. 

[#31] (3) Parameter estimation: With 10 parameters fitting one-dimensional curves, many parameters (e.g., α and σx) are likely highly correlated and poorly identified. Consider presenting scatter plots of the parameters (e.g., α vs σx) in the Supplement.

We have provided the scatter plots with a correlation matrix in Figure 4 – figure supplement 1 for the 12-month-old group and Figure 5 – figure supplement 1 for the 6-month-old group. As pointed out by the reviewer, there are significant rank correlations between parameters including ⍺ and σ_x² in both the 6 and 12-month-old groups. However, we also noted that there are no obvious linear relationships between the parameters.

The correlation above raises a potential problem of non-identifiability among parameters. First, we computed the variance inflation index (VIF) for all parameters to examine the risk of multicollinearity, though we did not consider a linear regression between parameters and DI in this study. All VIF values were below the conventional threshold 10 (Appendix 2 – table 5), suggesting that severe multicollinearity is unlikely to bias our conclusions. Second, we have conducted the simulation with different combinations of ⍺, σ_x², and K to clarify their contribution to overgeneralization and overdifferentiation observed in the 12-month-old group. 

In Appendix 2 – table 6, the values of ⍺ and σ_x² were either their upper or lower boundary set in parameter estimation, while the value K was selected heuristically to demonstrate its effect. Given the observed positive correlation between alpha and σ_x², and their negative correlation with K (Figure 4 - figure supplement 1), we consider the product of K \= {4, 35}, ⍺ \= {1, 3} and σ_x² \= {0.01, 3}. Among these combinations, the representative condition for the control group is α = 3, σ_x² = 3, and that for the App^NL-G-F group is α = 1, σ_x² = 0.01. In the latter condition, overgeneralization and overdifferentiation, which showed higher test 1 CR, lower number of acquisition latent causes (K_acq), lower test 3 CR, lower DI between test 3 and test 1, and higher number of latent causes after extinction (K_rem), was extremely induced. 

We found conditions that fall outside of empirical correlation, such as ⍺ = 3, σ_x² = 0.01, also reproduced overgeneralization and overdifferentiation. Similarly, the combination, ⍺ = 1, σ_x² = 3, exhibited control-like behavior when K = 4 but shifted toward App^NL-G-F-like behavior when K = 36. The effect of K was also evident when ⍺ = 3 and σ_x² = 3, where K = 36 led to over-differentiation. We note that these conditions were artificially set and likely not representative of biologically plausible. These results underscore the non-identifiability concern raised by the reviewer. Therefore, we acknowledge that merely attributing overgeneralization to lower ⍺ or overdifferentiation to lower σ_x² may be overly reductive. Instead, these patterns likely arise from the joint effect of ⍺, σ_x², and K. We have revised the manuscript accordingly in Results and Discussion (page 11-13, 18-19).

[#32] (4) Data normalization: Normalizing the data between 0 and 1 removes the interpretability of % freezing, making mice with large changes in freezing indistinguishable seem similar to mice with small changes.

As we describe in our reply to comment #26, the conditioned response in the latent cause model was scaled between 0 and 1, and we assume 0 and 1 mean the minimal and maximal CR within each mouse, respectively. Furthermore, although we initially tried to fit simulated CRs to raw CRs, we found that the fitting level was low due to the individual difference in the degree of behavioral expression: some mice exhibited a larger range of CR, while others showed a narrower one. Thus, we decided to normalize the data. We agree that this processing will make the mice with high changes in freezing% indistinguishable from those with low changes. However, the freezing% changes within the mouse were preserved and did not affect the discrimination index.

[#33] (5) Overlooking parameter differences: Differences in parameters, like w₀, that didn't fit the hypothesis may have been ignored.

Our initial hypothesis is that internal states were altered in App^NL-G-F mice, and we did not have a specific hypothesis on which parameter would contribute to such a state. We mainly focus on the parameters (1) that are significantly different between control and App</sup>NL-G</sup>- mice and (2) that are significantly correlated to the empirical behavioral data, DI between test 3 and test 1. 

In the 12-month-old group, besides ⍺ and σ_x², w₀ and K showed marginal p-value in Mann-Whitney U test (Table S7) and moderate correlation with the DI (Table S8). While differences in K were already discussed in the manuscript, we did miss the point that w₀ could contribute to the differences in w between control and App^NL-G-F (Figure 4G) in the previous manuscript. We explain the contribution of w₀ on the reinstatement results here. When other parameters are fixed, higher w₀ would lead to higher CR in test 3, because higher w₀ would allow increasing w_context by the unsignaled shock, leading to reinstatement (Appendix 2 – table 7). It is likely that higher w₀ would be sampled through the parameter estimation in the 12-month-old control but not App^NL-G-F. On the other hand, the number of latent causes is not sensitive to w₀ when other parameters were fixed at the initial guess value (Appendix 2 – table 1), suggesting w₀ has a small contribution to memory modification process. 

Thus, we speculate that although the difference in w₀ between control and App^NL-G-F mice may arise from the sampling process, resulting in a positive correlation with DI between test 3 and test 1, its contribution to diverged internal states would be smaller relative to α or σ_x² as a wide range of w₀ has no effect on the number of latent causes (Appendix 2 – table 7). We have added the discussion of differences in w₀ in the 12-month-old group in manuscript Line 357-359.

In the 6-month-old group, besides ⍺ and σ_x², 𝜃 is significantly higher in the AD mice group (Table S10) but not correlated with the DI (Table S11). We have already discussed this point in the manuscript.  

[#34] (6) Initial response: Higher initial responses in the model at the start of the experiment may reflect poor model fit.

Please refer to our reply to comment #26 for our explanation of what contributes to high initial responses in the latent cause model.

In addition, achieving a good fit for the acquisition CRs was not our primary purpose, as the response measured in the acquisition phase includes not only a conditioned response to the CS and context but also an unconditioned response to the novel stimuli (CS and US). This mixed response presumably increased the variance of the measured freezing rate over individuals, therefore we did not cover the results in the discussion.

Rather, we favor models at least replicating the establishment of conditioning, extinction and reinstatement of fear memory in order to explain the memory modification process. As we mentioned in the reply for comment #4, alternative models, the latent state model and the Rescorla-Wagner model, failed to replicate the observation (cf. Figure 3 – figure supplement 1A-1C). Thus, we chose to stand on the latent cause model as it aligns better with the purpose of this study. 

[#35] In addition, please be transparent if data is excluded, either during the fitting procedure or when performing one-way ANCOVA. Avoid discarding data when possible, but if necessary, provide clarity on the nature of excluded data (e.g., how many, why were they excluded, which group, etc?).

We clarify the information of excluded data as follows. We had 25 mice for the 6-month-old control group, 26 mice for the 6-month-old App^NL-G-F group, 29 mice for the 12-month-old control group, and 26 mice for the 12-month-old App^NL-G-F group (Table S1). 

Our first exclusion procedure was applied to the freezing rate data in the test phase. If the mouse had a freezing rate outside of the 1.5 IQR in any of the test phases, it is regarded as an outlier and removed from the analysis (see Statistical analysis in Materials and Methods). One mouse in the 6-month-old control group, one mouse in the 6-month-old App^NL-G-F group, five mice in the 12-month-old control group, and two mice in the 12-month-old App^NL-G-F group were excluded.

Our second exclusion procedure was applied during the fitting and parameter estimation (see parameter estimation in Materials and Methods). We have provided the number of anomaly samples during parameter estimation in Appendix 1 – figure 2.   

Lastly, we would like to state that all the sample sizes written in the figure legends do not include outliers detected through the exclusion procedure mentioned above.

[#36] Finally, since several statistical tests were used and the differences are small, I suggest noting that multiple comparisons were not controlled for, so p-values should be interpreted cautiously.

We have provided power analyses in Tables S21 and S22 with methods described in the manuscript (Line 897-898) and added a note that not all of the multiple comparisons were corrected for in the manuscript (Line 898-899).

References cited in the response letter only 

Bellio, T. A., Laguna-Torres, J. Y., Campion, M. S., Chou, J., Yee, S., Blusztajn, J. K., & Mellott, T. J. (2024). Perinatal choline supplementation prevents learning and memory deficits and reduces brain amyloid Aβ42 deposition in App^NL-G-F Alzheimer’s disease model mice. PLOS ONE, 19(2), e0297289. https://doi.org/10.1371/journal.pone.0297289

Blei, D. M., & Frazier, P. I. (2011). Distance Dependent Chinese Restaurant Processes. Journal of Machine Learning Research, 12(74), 2461–2488.

Cochran, A. L., & Cisler, J. M. (2019). A flexible and generalizable model of online latent-state learning. PLOS Computational Biology, 15(9), e1007331. https://doi.org/10.1371/journal.pcbi.1007331

Curiel Cid, R. E., Crocco, E. A., Duara, R., Vaillancourt, D., Asken, B., Armstrong, M. J., Adjouadi, M., Georgiou, M., Marsiske, M., Wang, W., Rosselli, M., Barker, W. W., Ortega, A., Hincapie, D., Gallardo, L., Alkharboush, F., DeKosky, S., Smith, G., & Loewenstein, D. A. (2024). Different aspects of failing to recover from proactive semantic interference predicts rate of progression from amnestic mild cognitive impairment to dementia. Frontiers in Aging Neuroscience, 16. https://doi.org/10.3389/fnagi.2024.1336008

Giustino, T. F., Fitzgerald, P. J., Ressler, R. L., & Maren, S. (2019). Locus coeruleus toggles reciprocal prefrontal firing to reinstate fear. Proceedings of the National Academy of Sciences, 116(17), 8570–8575. https://doi.org/10.1073/pnas.1814278116

Gu, X., Wu, Y.-J., Zhang, Z., Zhu, J.-J., Wu, X.-R., Wang, Q., Yi, X., Lin, Z.-J., Jiao, Z.-H., Xu, M., Jiang, Q., Li, Y., Xu, N.-J., Zhu, M. X., Wang, L.-Y., Jiang, F., Xu, T.-L., & Li, W.-G. (2022). Dynamic tripartite construct of interregional engram circuits underlies forgetting of extinction memory. Molecular Psychiatry, 27(10), 4077–4091. https://doi.org/10.1038/s41380-022-01684-7

Lacagnina, A. F., Brockway, E. T., Crovetti, C. R., Shue, F., McCarty, M. J., Sattler, K. P., Lim, S. C., Santos, S. L., Denny, C. A., & Drew, M. R. (2019). Distinct hippocampal engrams control extinction and relapse of fear memory. Nature Neuroscience, 22(5), 753–761. https://doi.org/10.1038/s41593-019-0361-z

Loewenstein, D. A., Curiel, R. E., Greig, M. T., Bauer, R. M., Rosado, M., Bowers, D., Wicklund, M., Crocco, E., Pontecorvo, M., Joshi, A. D., Rodriguez, R., Barker, W. W., Hidalgo, J., & Duara, R. (2016). A Novel Cognitive Stress Test for the Detection of Preclinical Alzheimer’s Disease: Discriminative Properties and Relation to Amyloid Load. The American Journal of Geriatric Psychiatry : Official Journal of the American Association for Geriatric Psychiatry, 24(10), 804–813. https://doi.org/10.1016/j.jagp.2016.02.056

Loewenstein, D. A., Greig, M. T., Curiel, R., Rodriguez, R., Wicklund, M., Barker, W. W., Hidalgo, J., Rosado, M., & Duara, R. (2015). Proactive Semantic Interference Is Associated With Total and Regional Abnormal Amyloid Load in Non-Demented Community-Dwelling Elders: A Preliminary Study. The American Journal of Geriatric Psychiatry : Official Journal of the American Association for Geriatric Psychiatry, 23(12), 1276–1279. https://doi.org/10.1016/j.jagp.2015.07.009

Valles-Salgado, M., Gil-Moreno, M. J., Curiel Cid, R. E., Delgado-Á lvarez, A., Ortega-Madueño, I., Delgado-Alonso, C., Palacios-Sarmiento, M., López-Carbonero, J. I., Cárdenas, M. C., MatíasGuiu, J., Díez-Cirarda, M., Loewenstein, D. A., & Matias-Guiu, J. A. (2024). Detection of cerebrospinal fluid biomarkers changes of Alzheimer’s disease using a cognitive stress test in persons with subjective cognitive decline and mild cognitive impairment. Frontiers in Psychology, 15. https://doi.org/10.3389/fpsyg.2024.1373541

Zaki, Y., Mau, W., Cincotta, C., Monasterio, A., Odom, E., Doucette, E., Grella, S. L., Merfeld, E., Shpokayte, M., & Ramirez, S. (2022). Hippocampus and amygdala fear memory engrams reemerge after contextual fear relapse. Neuropsychopharmacology, 47(11), 1992–2001. https://doi.org/10.1038/s41386-022-01407-0

Author response:

Reviewer #1 (Public review):

Summary:

Review of the manuscript titled " Mycobacterial Metallophosphatase MmpE acts as a nucleomodulin to regulate host gene expression and promotes intracellular survival".

The study provides an insightful characterization of the mycobacterial secreted effector protein MmpE, which translocates to the host nucleus and exhibits phosphatase activity. The study characterizes the nuclear localization signal sequences and residues critical for the phosphatase activity, both of which are required for intracellular survival.

Strengths:

(1) The study addresses the role of nucleomodulins, an understudied aspect in mycobacterial infections.

(2) The authors employ a combination of biochemical and computational analyses along with in vitro and in vivo validations to characterize the role of MmpE.

Weaknesses:

(1) While the study establishes that the phosphatase activity of MmpE operates independently of its NLS, there is a clear gap in understanding how this phosphatase activity supports mycobacterial infection. The investigation lacks experimental data on specific substrates of MmpE or pathways influenced by this virulence factor.

We thank the reviewer for this insightful comment and agree that identification of the substrate of MmpE is important to fully understand its role in mycobacterial infection.

MmpE is a putative purple acid phosphatase (PAP) and a member of the metallophosphoesterase (MPE) superfamily. Enzymes in this family are known for their catalytic promiscuity and broad substrate specificity, acting on phosphomonoesters, phosphodiesters, and phosphotriesters (Matange et al., Biochem J., 2015). In bacteria, several characterized MPEs have been shown to hydrolyze substrates such as cyclic nucleotides (e.g., cAMP) (Keppetipola et al., J Biol Chem, 2008; Shenoy et al., J Mol Biol, 2007), nucleotide derivatives (e.g., AMP, UDP-glucose) (Innokentev et al., mBio, 2025), and pyrophosphate-containing compounds (e.g., Ap4A, UDP-DAGn) (Matange et al., Biochem J., 2015). Although the binding motif of MmpE has been identified, determining its physiological substrates remains challenging due to the low abundance and instability of potential metabolites, as well as the limited sensitivity and coverage of current metabolomic technologies in mycobacteria.

(2) The study does not explore whether the phosphatase activity of MmpE is dependent on the NLS within macrophages, which would provide critical insights into its biological relevance in host cells. Conducting experiments with double knockout/mutant strains and comparing their intracellular survival with single mutants could elucidate these dependencies and further validate the significance of MmpE's dual functions.

We thank the reviewer for the comment. In our study, we demonstrate that both the nuclear localization and phosphatase activity of MmpE are required for full virulence (Figure 3D–E). Importantly, deletion of the NLS motifs did not impair MmpE’s phosphatase activity in vitro (Figure 2F), indicating that its enzymatic function is structurally independent of its nuclear localization. These findings suggest that MmpE functions as a bifunctional protein, with distinct and non-overlapping roles for its nuclear trafficking and phosphatase activity. We have expanded on this point in the Discussion section “MmpE Functions as a Bifunctional Protein with Nuclear Localization and Phosphatase Activity”.

(3) The study does not provide direct experimental validation of the MmpE deletion on lysosomal trafficking of the bacteria.

We thank the reviewer for the comment. The role of Rv2577/MmpE in phagosome maturation has been demonstrated in M. tuberculosis, where its deletion increases colocalization with lysosomal markers such as LAMP-2 and LAMP-3 (Forrellad et al., Front Microbiol, 2020). In our study, we found that mmpE deletion in M. bovis BCG led to upregulation of lysosomal genes, including TFEB, LAMP1, LAMP2, and v-ATPase subunits, compared to the wild-type strain. These results suggest that MmpE may regulate lysosomal trafficking by interfering with phagosome–lysosome fusion.

To further validate MmpE’s role in phagosome maturation, we will perform fluorescence colocalization assays in THP-1 macrophages infected with BCG/wt, ∆mmpE, complemented, and NLS-mutant strains. Co-staining with LAMP1 and LysoTracker will allow us to assess whether the ∆mmpE mutant is more efficiently trafficked to lysosomes.

(4) The role of MmpE as a mycobacterial effector would be more relevant using virulent mycobacterial strains such as H37Rv.

We thank the reviewer for the comment. Previously, the role of Rv2577/MmpE as a virulence factor has been demonstrated in M. tuberculosis CDC 1551, where its deletion significantly reduced bacterial replication in mouse lungs at 30 days post-infection (Forrellad et al., Front Microbiol, 2020). However, that study did not explore the underlying mechanism of MmpE function. In our work, we found that MmpE enhances M. bovis BCG survival in both macrophages (THP-1 and RAW264.7) and mice (Figure 2A-B, Figure 6A), consistent with its proposed role in virulence. To investigate the molecular mechanism by which MmpE promotes intracellular survival, we used M. bovis BCG as a biosafe surrogate and this model is widely accepted for studying mycobacterial pathogenesis (Wang et al., Nat Immunol, 2025; Wang et al., Nat Commun, 2017; Péan et al., Nat Commun, 2017).

Reviewer #2 (Public review):

Summary:

In this paper, the authors have characterized Rv2577 as a Fe3+/Zn2+ -dependent metallophosphatase and a nucleomodulin protein. The authors have also identified His348 and Asn359 as critical residues for Fe3+ coordination. The authors show that the proteins encode for two nuclease localization signals. Using C-terminal Flag expression constructs, the authors have shown that the MmpE protein is secretory. The authors have prepared genetic deletion strains and show that MmpE is essential for intracellular survival of M. bovis BCG in THP-1 macrophages, RAW264.7 macrophages, and a mouse model of infection. The authors have also performed RNA-seq analysis to compare the transcriptional profiles of macrophages infected with wild-type and MmpE mutant strains. The relative levels of ~ 175 transcripts were altered in MmpE mutant-infected macrophages and the majority of these were associated with various immune and inflammatory signalling pathways. Using these deletion strains, the authors proposed that MmpE inhibits inflammatory gene expression by binding to the promoter region of a vitamin D receptor. The authors also showed that MmpE arrests phagosome maturation by regulating the expression of several lysosome-associated genes such as TFEB, LAMP1, LAMP2, etc. These findings reveal a sophisticated mechanism by which a bacterial effector protein manipulates gene transcription and promotes intracellular survival.

Strength:

The authors have used a combination of cell biology, microbiology, and transcriptomics to elucidate the mechanisms by which Rv2577 contributes to intracellular survival.

Weakness:

The authors should thoroughly check the mice data and show individual replicate values in bar graphs.

We kindly appreciate the reviewer for the advice. We will update the relevant mice data in the revised manuscript.

Reviewer #3 (Public review):

Summary:

In this manuscript titled "Mycobacterial Metallophosphatase MmpE Acts as a Nucleomodulin to Regulate Host Gene Expression and Promote Intracellular Survival", Chen et al describe biochemical characterisation, localisation and potential functions of the gene using a genetic approach in M. bovis BCG and perform macrophage and mice infections to understand the roles of this potentially secreted protein in the host cell nucleus. The findings demonstrate the role of a secreted phosphatase of M. bovis BCG in shaping the transcriptional profile of infected macrophages, potentially through nuclear localisation and direct binding to transcriptional start sites, thereby regulating the inflammatory response to infection.

Strengths:

The authors demonstrate using a transient transfection method that MmpE when expressed as a GFP-tagged protein in HEK293T cells, exhibits nuclear localisation. The authors identify two NLS motifs that together are required for nuclear localisation of the protein. A deletion of the gene in M. bovis BCG results in poorer survival compared to the wild-type parent strain, which is also killed by macrophages. Relative to the WT strain-infected macrophages, macrophages infected with the ∆mmpE strain exhibited differential gene expression. Overexpression of the gene in HEK293T led to occupancy of the transcription start site of several genes, including the Vitamin D Receptor. Expression of VDR in THP1 macrophages was lower in the case of ∆mmpE infection compared to WT infection. This data supports the utility of the overexpression system in identifying potential target loci of MmpE using the HEK293T transfection model. The authors also demonstrate that the protein is a phosphatase, and the phosphatase activity of the protein is partially required for bacterial survival but not for the regulation of the VDR gene expression.

Weaknesses:

(1)   While the motifs can most certainly behave as NLSs, the overexpression of a mycobacterial protein in HEK293T cells can also result in artefacts of nuclear localisation. This is not unprecedented. Therefore, to prove that the protein is indeed secreted from BCG, and is able to elicit transcriptional changes during infection, I recommend that the authors (i) establish that the protein is indeed secreted into the host cell nucleus, and (ii) the NLS mutation prevents its localisation to the nucleus without disrupting its secretion.

We kindly appreciate the reviewer for the advice and will include the relevant experiments in the revised manuscript. The localization of WT MmpE and the NLS mutated MmpE will be tested in the BCG infected macrophages.

Demonstration that the protein is secreted: Supplementary Figure 3 - Immunoblotting should be performed for a cytosolic protein, also to rule out detection of proteins from lysis of dead cells. Also, for detecting proteins in the secreted fraction, it would be better to use Sauton's media without detergent, and grow the cultures without agitation or with gentle agitation. The method used by the authors is not a recommended protocol for obtaining the secreted fraction of mycobacteria.

We agree with the reviewer and we will further validate the secretion of MmpE using the tested protocol.

Demonstration that the protein localises to the host cell nucleus upon infection: Perform an infection followed by immunofluorescence to demonstrate that the endogenous protein of BCG can translocate to the host cell nucleus. This should be done for an NLS1-2 mutant expressing cell also.

We will add this experiment in the revised manuscript.

(2) In the RNA-seq analysis, the directionality of change of each of the reported pathways is not apparent in the way the data have been presented. For example, are genes in the cytokine-cytokine receptor interaction or TNF signalling pathway expressed more, or less in the ∆mmpE strain?

We thank the reviewer for pointing this out and fully agree that conventional KEGG pathway enrichment diagrams do not convey the directionality of individual gene expression changes within each pathway. While KEGG enrichment analysis identifies pathways that are statistically overrepresented among differentially expressed genes, it does not indicate whether individual genes within those pathways are upregulated or downregulated.

To address this, we re-analyzed the expression trends of DEGs within each significantly enriched KEGG pathway. The results show that key immune-related pathways, including cytokine–cytokine receptor interaction, TNF signaling, NF-κB signaling, and chemokine signaling, are collectively upregulated in THP-1 macrophages infected with ∆mmpE strain compared to those infected with the wild-type BCG strain. The full list of DEGs will be provided in the supplementary materials. The complete RNA-seq dataset has been deposited in the GEO database, and the accession number will be included in the revised manuscript.

(3) Several of these pathways are affected as a result of infection, while others are not induced by BCG infection. For example, BCG infection does not, on its own, produce changes in IL1β levels. As the author s did not compare the uninfected macrophages as a control, it is difficult to interpret whether ∆mmpE induced higher expression than the WT strain, or simply did not induce a gene while the WT strain suppressed expression of a gene. This is particularly important because the strain is attenuated. Does the attenuation have anything to do with the ability of the protein to induce lysosomal pathway genes? Does induction of this pathway lead to attenuation of the strain? Similarly, for pathways that seem to be downregulated in the ∆mmpE strain compared to the WT strain, these might have been induced upon infection with the WT strain but not sufficiently by the ∆mmpE strain due to its attenuation/ lower bacterial burden.

We thank the reviewer for the comment. We will update qRT-PCR data with the uninfected macrophages as a control in the revised manuscript.

Wild-type Mycobacterium bovis BCG strain still has the function of inhibiting phagosome maturation (Branzk et al., Nat Immunol, 2014; Weng et al., Nat Commun, 2022). Forrellad et al. previously identified Rv2577/MmpE as a virulence factor in M. tuberculosis and disruption of the MmpE gene impairs the ability of M. tuberculosis to arrest phagosome maturation (Forrellad et al., Front Microbiol, 2020). In our study, transcriptomic and qRTPCR data (Figures 4C and G, S4C) show that deletion of mmpE in M. bovis BCG leads to upregulation of lysosomal biogenesis and acidification genes, including TFEB, LAMP1, and vATPase. To further validate MmpE’s role in phagosome maturation, we will perform fluorescence colocalization assays in THP-1 macrophages infected with BCG/wt, ∆mmpE, complemented, and NLS-mutant strains. Co-staining with LAMP1 and LysoTracker will assess whether the ∆mmpE mutant is more efficiently trafficked to lysosomes.

Furthermore, CFU assays demonstrated that the ∆mmpE strain exhibits markedly reduced bacterial survival in both human THP-1 and murine RAW264.7 macrophages, as well as in mice, compared to the wild-type strain (Figures 4A and C, 6A). These findings suggest that the loss of MmpE compromises bacterial survival, likely due to enhanced lysosomal trafficking and acidification. This supports previous studies showing that increased lysosomal activity promotes mycobacterial clearance (Gutierrez et al., Cell, 2004; Pilli et al., Immunity, 2012).

(4) CHIP-seq should be performed in THP1 macrophages, and not in HEK293T. Overexpression of a nuclear-localised protein in a non-relevant line is likely to lead to several transcriptional changes that do not inform us of the role of the gene as a transcriptional regulator during infection.

We thank the reviewer for the comment. We performed ChIP-seq in HEK293T cells is based on the fact that this cell line is widely used in ChIP-based assays due to its high transfection efficiency, robust nuclear protein expression, and well-annotated genome (Lampe et al., Nat Biotechnol, 2024; Marasco et al., Cell, 2022). These features make HEK293T an ideal system for the initial identification of genome wide chromatin binding profiles of novel nuclear effectors such as MmpE.

Furthermore, we validated the major observations in THP-1 macrophages, including (i) RNAseq of THP-1 cells infected with either WT BCG or ∆mmpE strains revealed significant transcriptional changes in immune and lysosomal pathways (Figure 4A); (ii) Integrated analysis of CUT&Tag and RNA-seq data identified 298 genes in infected THP-1 cells that exhibited both MmpE binding and corresponding expression changes. Among these, VDR was validated as a direct transcriptional target of MmpE using EMSA and ChIP-PCR (Figures 5E-J, S5D-F). Notably, the signaling pathways associated with MmpE-bound genes, including PI3K-Akt-mTOR signaling and lysosomal function, substantially overlap with those transcriptionally modulated in infected THP-1 macrophages (Figures 4B-G, S4B-C, S5C-D), further supporting the biological relevance of the ChIP-seq data obtained from HEK293T cells.

(5) I would not expect to see such large inflammatory reactions persisting 56 days postinfection with M. bovis BCG. Is this something peculiar for an intratracheal infection with 1x107 bacilli? For images of animal tissue, the authors should provide images of the entire lung lobe with the zoomed-in image indicated as an inset.

We thank the reviewer for the comment. The lung inflammation peaked at days 21–28 and had clearly subsided by day 56 across all groups (Figure 6B), consistent with the expected resolution of immune responses to an attenuated strain like M. bovis BCG. This temporal pattern is in line with previous studies using intravenous or intratracheal BCG vaccination in mice and macaques, which also demonstrated robust early immune activation followed by resolution over time (Smith et al., Nat Microbiol, 2025; Darrah et al., Nature, 2020).

In this study, the infectious dose (1×10⁷ CFU intratracheally) was selected based on previous studies in which intratracheal delivery of 1×10⁷CFU produced consistent and measurable lung immune responses and pathology without causing overt illness or mortality (Xu et al., Sci Rep, 2017; Niroula et al., Sci Rep, 2025). We will provide whole-lung lobe images with zoomed-in insets in the revised manuscript.

(6) For the qRT-PCR based validation, infections should be performed with the MmpEcomplemented strain in the same experiments as those for the WT and ∆mmpE strain so that they can be on the same graph, in the main manuscript file. Supplementary Figure 4 has three complementary strains. Again, the absence of the uninfected, WT, and∆mmpE infected condition makes interpretation of these data very difficult.

We thank the reviewer for the comment. As suggested, we will conduct the qRT-PCR experiment including the uninfected, WT, ∆mmpE, Comp-MmpE, and the three complementary strains infecting THP-1 cells. The updated data will be provided in the revised manuscript.

(7) The abstract mentions that MmpE represses the PI3K-Akt-mTOR pathway, which arrests phagosome maturation. There is not enough data in this manuscript in support of this claim. Supplementary Figure 5 does provide qRT-PCR validation of genes of this pathway, but the data do not indicate that higher expression of these pathways, whether by VDR repression or otherwise, is driving the growth restriction of the ∆mmpE strain.

We thank the reviewer for the comment. The role of MmpE in phagosome maturation was previously characterized. Disruption of mmpE impairs the ability of M. tuberculosis to arrest lysosomal trafficking (Forrellad et al., Front Microbiol, 2020). In this study, we further found that MmpE suppresses the expression of key lysosomal genes, including TFEB, LAMP1, LAMP2, and ATPase subunits (Figure 4G), suggesting MmpE is involved in arresting phagosome maturation. As noted, the genes in the PI3K–Akt–mTOR pathway are upregulated in ∆mmpE-infected macrophages (Figure S5C).

To functionally validate this, we will conduct two complementary experimental approaches:

(i) Immunofluorescence assays: We will assess phagosome maturation and lysosomal fusion in THP-1 cells infected with BCG/wt, ∆mmpE, Comp-MmpE, and NLS mutant strains. Colocalization of intracellular bacteria with LAMP1 and LysoTracker will be quantified to determine whether the ∆mmpE strain is more efficiently trafficked to lysosomes.

(ii) CFU assays: We will perform CFU assays in THP-1 cells infected with BCG/wt or ∆mmpE in the presence or absence of PI3K-Akt-mTOR pathway inhibitors (e.g., Dactolisib), to assess whether activation of this pathway contributes to the intracellular growth restriction observed in the ∆mmpE strain.

(8) The relevance of the NLS and the phosphatase activity is not completely clear in the CFU assays and in the gene expression data. Firstly, there needs to be immunoblot data provided for the expression and secretion of the NLS-deficient and phosphatase mutants. Secondly, CFU data in Figure 3A, C, and E must consistently include both the WT and ∆mmpE strain.

We thank the reviewer for the comment. We will provide immunoblot data for the expression and secretion of the NLS-deficient and phosphatase mutants. Additionally, we will revise Figure 3A, 3C, and 3E to consistently include both the WT and ΔmmpE strains in the CFU assays.

Reference

Branzk N, Lubojemska A, Hardison SE, Wang Q, Gutierrez MG, Brown GD, Papayannopoulos V (2014) Neutrophils sense microbe size and selectively release neutrophil extracellular traps in response to large pathogens Nat Immunol 15:1017-25.

Darrah PA, Zeppa JJ, Maiello P, Hackney JA, Wadsworth MH 2nd, Hughes TK, Pokkali S, Swanson PA 2nd, Grant NL, Rodgers MA, Kamath M, Causgrove CM, Laddy DJ, Bonavia A, Casimiro D, Lin PL, Klein E, White AG, Scanga CA, Shalek AK, Roederer M, Flynn JL, Seder RA (2020) Prevention of tuberculosis in macaques after intravenous BCG immunization Nature 577:95-102.

Forrellad MA, Blanco FC, Marrero Diaz de Villegas R, Vázquez CL, Yaneff A, García EA, Gutierrez MG, Durán R, Villarino A, Bigi F (2020) Rv2577 of Mycobacterium tuberculosis Is a virulence factor with dual phosphatase and phosphodiesterase functions Front Microbiol 11:570794.

Gutierrez MG, Master SS, Singh SB, Taylor GA, Colombo MI, Deretic V (2004) Autophagy is a defense mechanism inhibiting BCG and Mycobacterium tuberculosis survival in infected macrophages Cell 119:753-66.

Innokentev A, Sanchez AM, Monetti M, Schwer B, Shuman S (2025) Efn1 and Efn2 are extracellular 5'-nucleotidases induced during the fission yeast response to phosphate starvation mBio 16: e0299224.

Keppetipola N, Shuman S (2008) A phosphate-binding histidine of binuclear metallophosphodiesterase enzymes is a determinant of 2',3'-cyclic nucleotide phosphodiesterase activity J Biol Chem 283:30942-9.

Lampe GD, King RT, Halpin-Healy TS, Klompe SE, Hogan MI, Vo PLH, Tang S, Chavez A, Sternberg SH (2024) Targeted DNA integration in human cells without double-strand breaks using CRISPR-associated transposases Nat Biotechnol 42:87-98.

Marasco LE, Dujardin G, Sousa-Luís R, Liu YH, Stigliano JN, Nomakuchi T, Proudfoot NJ, Krainer AR, Kornblihtt AR (2022) Counteracting chromatin effects of a splicing-correcting antisense oligonucleotide improves its therapeutic efficacy in spinal muscular atrophy Cell 185:2057-2070.e15.

Matange N, Podobnik M, Visweswariah SS (2015) Metallophosphoesterases: structural fidelity with functional promiscuity Biochem J 467:201-16.

Niroula N, Ghodasara P, Marreros N, Fuller B, Sanderson H, Zriba S, Walker S, Shury TK, Chen JM (2025) Orally administered live BCG and heat-inactivated Mycobacterium bovis protect bison against experimental bovine tuberculosis Sci Rep 15:3764.

Péan CB, Schiebler M, Tan SW, Sharrock JA, Kierdorf K, Brown KP, Maserumule MC,

Menezes S, Pilátová M, Bronda K, Guermonprez P, Stramer BM, Andres Floto R, Dionne MS (2017) Regulation of phagocyte triglyceride by a STAT-ATG2 pathway controls mycobacterial infection Nat Commun 8:14642.

Pilli M, Arko-Mensah J, Ponpuak M, Roberts E, Master S, Mandell MA, Dupont N, Ornatowski W, Jiang S, Bradfute SB, Bruun JA, Hansen TE, Johansen T, Deretic V (2012) TBK-1 promotes autophagy-mediated antimicrobial defense by controlling autophagosome maturation Immunity 37:223-34.

Shenoy AR, Capuder M, Draskovic P, Lamba D, Visweswariah SS, Podobnik M (2007) Structural and biochemical analysis of the Rv0805 cyclic nucleotide phosphodiesterase from Mycobacterium tuberculosis J Mol Biol 365:211-25.

Smith AA, Su H, Wallach J, Liu Y, Maiello P, Borish HJ, Winchell C, Simonson AW, Lin PL, Rodgers M, Fillmore D, Sakal J, Lin K, Vinette V, Schnappinger D, Ehrt S, Flynn JL (2025) A BCG kill switch strain protects against Mycobacterium tuberculosis in mice and non-human primates with improved safety and immunogenicity Nat Microbiol 10:468-481.

Wang J, Ge P, Qiang L, Tian F, Zhao D, Chai Q, Zhu M, Zhou R, Meng G, Iwakura Y, Gao GF, Liu CH (2017) The mycobacterial phosphatase PtpA regulates the expression of host genes and promotes cell proliferation Nat Commun 8:244.

Wang J, Li BX, Ge PP, Li J, Wang Q, Gao GF, Qiu XB, Liu CH (2015) Mycobacterium tuberculosis suppresses innate immunity by coopting the host ubiquitin system Nat Immunol 16:237–245

Weng Y, Shepherd D, Liu Y, Krishnan N, Robertson BD, Platt N, Larrouy-Maumus G, Platt FM (2022) Inhibition of the Niemann-Pick C1 protein is a conserved feature of multiple strains of pathogenic mycobacteria Nat Commun 13:5320.

Xu X, Lu X, Dong X, Luo Y, Wang Q, Liu X, Fu J, Zhang Y, Zhu B, Ma X (2017) Effects of hMASP2 on the formation of BCG infection-induced granuloma in the lungs of BALB/c mice Sci Rep 7:2300.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review): 

Summary: 

McDougal et al. aimed to characterize the antiviral activity of mammalian IFIT1 orthologs. They first performed three different evolutionary selection analyses within each major mammalian clade and identified some overlapping positive selection sites in IFIT1. They found that one site that is positively selected in primates is in the RNA-binding exit tunnel of IFIT1 and is tolerant of mutations to amino acids with similar biochemical properties. They then tested 9 diverse mammalian IFIT1 proteins against VEEV, VSV, PIV3, and SINV and found that each ortholog has distinct antiviral activities. Lastly, they compared human and chimpanzee IFIT1 and found that the determinant of their differential anti-VEEV activity may be partly attributed to their ability to bind Cap0 RNA. 

Strengths: 

The study is one of the first to test the antiviral activity of IFIT1 from diverse mammalian clades against VEEV, VSV, PIV3, and SINV. Cloning and expressing these 39 IFIT1 orthologs in addition to single and combinatorial mutants is not a trivial task. The positive connection between anti-VEEV activity and Cap0 RNA binding is interesting, suggesting that differences in RNA binding may explain differences in antiviral activity. 

Weaknesses: 

The evolutionary selection analyses yielded interesting results, but were not used to inform follow-up studies except for a positively selected site identified in primates. Since positive selection is one of the two major angles the authors proposed to investigate mammalian IFIT1 orthologs with, they should integrate the positive selection results with the rest of the paper more seamlessly, such as discussing the positive selection results and their implications, rather than just pointing out that positively selected sites were identified. The paper should elaborate on how the positive selection analyses PAML, FUBAR, and MEME complement one another to explain why the tests gave them different results. Interestingly, MEME which usually provides more sites did not identify site 193 in primates that was identified by both PAML and FUBAR. The authors should also provide the rationale for choosing to focus on the 3 sites identified in primates only. One of those sites, 193, was also found to be positively selected in bats, although the authors did not discuss or integrate that finding into the study. In Figure 1A, they also showed a dN/dS < 1 from PAML, which is confusing and would suggest negative selection instead of positive selection. Importantly, since the authors focused on the rapidly evolving site 193 in primates, they should test the IFIT1 orthologs against viruses that are known to infect primates to directly investigate the impact of the evolutionary arms race at this site on IFIT1 function. 

We thank the reviewer for their assessment and for acknowledging the breadth of our dataset regarding diverse IFIT1s, number of viruses tested, and the functional data that may correlate biochemical properties of IFIT1 orthologous proteins with antiviral function. We have expanded the introduction and results sections to better explain and distinguish between PAML, FUBAR, and MEME analyses. Furthermore, we have expanded the discussion to incorporate the observation that site 193 is rapidly evolving in bats, as well as the observation that nearby sites to the TPR4 loop were identified as rapidly evolving in all clades of mammals tested. We also do observe an overall gene dN/dS of <1, however this is simply the average across all codons of the entire gene and does not rule out positive selection at specific sites. This is observed for other restriction factors, as many domains are undergoing purifying selection to retain core functions (e.g enzymatic function, structural integrity) while other domains (e.g. interfaces with viral antagonists or viral proteins) show strong positive selection. Specific examples include the restriction factors BST-2/Tetherin (PMID: 19461879) and MxA (PMID: 23084925). Furthermore, we agree that testing more IFIT1-sensitive viruses that naturally infect primates with our IFIT1 193 mutagenesis library would shed light on the influence of host-virus arms races at this site. However, VEEV naturally does also infect humans as well as at least one other species of primate (PMID: 39983680).

Below we individually address the reviewers' claims of inaccurate data interpretation.

Some of the data interpretation is not accurate. For example: 

(1) Lines 232-234: "...western blot analysis revealed that the expression of IFIT1 orthologs was relatively uniform, except for the higher expression of orca IFIT1 and notably lower expression of pangolin IFIT1 (Figure 4B)." In fact, most of the orthologs are not expressed in a "relatively uniform" manner e.g. big brown bat vs. shrew are quite different. 

We have now included quantification of the western blots to allow the reader to compare infection results with the infection data (Updated Figure 4B and 4G). We have also removed the phrase “relatively uniform” from the text and have instead included text describing the quantified expression differences.

(2) Line 245: "...mammalian IFIT1 species-specific differences in viral suppression are largely independent of expression differences." While it is true that there is no correlation between protein expression and antiviral activity in each species, the authors cannot definitively conclude that the species-specific differences are independent of expression differences. Since the orthologs are clearly not expressed in the same amounts, it is impossible to fully assess their true antiviral activity. At the very least, the authors should acknowledge that the protein expression can affect antiviral activity. They should also consider quantifying the IFIT1 protein bands and normalizing each to GAPDH for readers to better compare protein expression and antiviral activity. The same issue is in Line 267. 

We have now included quantification and normalization of the western blots to allow the reader to compare infection results with the infection data (Updated Figure 4B and 4G). Furthermore, we acknowledge in the text that expression differences may affect antiviral potency in infection experiments.

(3) Line 263: "SINV... was modestly suppressed by pangolin, sheep, and chinchilla IFIT1 (Figure 4E)..." The term "modestly suppressed" does not seem fitting if there is 60-70% infection in cells expressing pangolin and chinchilla IFIT1. 

We have modified the text to say “significantly suppressed” rather than “modestly suppressed.”

(4) The study can be significantly improved if the authors can find a thread to connect each piece of data together, so the readers can form a cohesive story about mammalian IFIT1. 

We appreciate the reviewer’s suggestion and have tried to make the story including more cohesive through commentary on positive selection and by using the computational analysis to first inform potential evolutionary consequences of IFIT1 functionality first by an intraspecies (human) approach, and then later an interspecies approach with diverse mammals that have great sequence diversity. Furthermore, we point out that almost all IFIT1s tested in the ortholog screen were also included in our computational analysis allowing for the potential to connect functional observations with those seen in the evolutionary analyses.

Reviewer #2 (Public review): 

McDougal et al. describe the surprising finding that IFIT1 proteins from different mammalian species inhibit the replication of different viruses, indicating that the evolution of IFIT1 across mammals has resulted in host speciesspecific antiviral specificity. Before this work, research into the antiviral activity and specificity of IFIT1 had mostly focused on the human ortholog, which was described to inhibit viruses including vesicular stomatitis virus (VSV) and Venezuelan equine encephalitis virus (VEEV) but not other viruses including Sindbis virus (SINV) and parainfluenza virus type 3 (PIV3). In the current work, the authors first perform evolutionary analyses on IFIT1 genes across a wide range of mammalian species and reveal that IFIT1 genes have evolved under positive selection in primates, bats, carnivores, and ungulates. Based on these data, they hypothesize that IFIT1 proteins from these diverse mammalian groups may show distinct antiviral specificities against a panel of viruses. By generating human cells that express IFIT1 proteins from different mammalian species, the authors show a wide range of antiviral activities of mammalian IFIT1s. Most strikingly, they find several IFIT1 proteins that have completely different antiviral specificities relative to human IFIT1, including IFIT1s that fail to inhibit VSV or VEEV, but strongly inhibit PIV3 or SINV. These results indicate that there is potential for IFIT1 to inhibit a much wider range of viruses than human IFIT1 inhibits. Electrophoretic mobility shift assays (EMSAs) suggest that some of these changes in antiviral specificity can be ascribed to changes in the direct binding of viral RNAs. Interestingly, they also find that chimpanzee IFIT1, which is >98% identical to human IFIT1, fails to inhibit any tested virus. Replacing three residues from chimpanzee IFIT1 with those from human IFIT1, one of which has evolved under positive selection in primates, restores activity to chimpanzee IFIT1. Together, these data reveal a vast diversity of IFIT1 antiviral specificity encoded by mammals, consistent with an IFIT1-virus evolutionary "arms race". 

Overall, this is a very interesting and well-written manuscript that combines evolutionary and functional approaches to provide new insight into IFIT1 antiviral activity and species-specific antiviral immunity. The conclusion that IFIT1 genes in several mammalian lineages are evolving under positive selection is supported by the data, although there are some important analyses that need to be done to remove any confounding effects from gene recombination that has previously been described between IFIT1 and its paralog IFIT1B. The virology results, which convincingly show that IFIT1s from different species have distinct antiviral specificity, are the most surprising and exciting part of the paper. As such, this paper will be interesting for researchers studying mechanisms of innate antiviral immunity, as well as those interested in species-specific antiviral immunity. Moreover, it may prompt others to test a wide range of orthologs of antiviral factors beyond those from humans or mice, which could further the concept of host-specific innate antiviral specificity. Additional areas for improvement, which are mostly to clarify the presentation of data and conclusions, are described below. 

Strengths: 

(1) This paper is a very strong demonstration of the concept that orthologous innate immune proteins can evolve distinct antiviral specificities. Specifically, the authors show that IFIT1 proteins from different mammalian species are able to inhibit the replication of distinct groups of viruses, which is most clearly illustrated in Figure 4G. This is an unexpected finding, as the mechanism by which IFIT1 inhibits viral replication was assumed to be similar across orthologs. While the molecular basis for these differences remains unresolved, this is a clear indication that IFIT1 evolution functionally impacts host-specific antiviral immunity and that IFIT1 has the potential to inhibit a much wider range of viruses than previously described. 

(2) By revealing these differences in antiviral specificity across IFIT1 orthologs, the authors highlight the importance of sampling antiviral proteins from different mammalian species to understand what functions are conserved and what functions are lineage- or species-specific. These results might therefore prompt similar investigations with other antiviral proteins, which could reveal a previously undiscovered diversity of specificities for other antiviral immunity proteins. 

(3) The authors also surprisingly reveal that chimpanzee IFIT1 shows no antiviral activity against any tested virus despite only differing from human IFIT1 by eight amino acids. By mapping this loss of function to three residues on one helix of the protein, the authors shed new light on a region of the protein with no previously known function. 

(4) Combined with evolutionary analyses that indicate that IFIT1 genes are evolving under positive selection in several mammalian groups, these functional data indicate that IFIT1 is engaged in an evolutionary "arms race" with viruses, which results in distinct antiviral specificities of IFIT1 proteins from different species. 

Weaknesses: 

(1) The evolutionary analyses the authors perform appear to indicate that IFIT1 genes in several mammalian groups have evolved under positive selection. However, IFIT1 has previously been shown to have undergone recurrent instances of recombination with the paralogous IFIT1B, which can confound positive selection analyses such as the ones the authors perform. The authors should analyze their alignments for evidence of recombination using a tool such as GARD (in the same HyPhy package along with MEME and FUBAR). Detection of recombination in these alignments would invalidate their positive selection inferences, in which case the authors need to either analyze individual non-recombining domains or limit the number of species to those that are not undergoing recombination. While it is likely that these analyses will still reveal a signature of positive selection, this step is necessary to ensure that the signatures of selection and sites of positive selection are accurate. 

(2) The choice of IFIT1 homologs chosen for study needs to be described in more detail. Many mammalian species encode IFIT1 and IFIT1B proteins, which have been shown to have different antiviral specificity, and the evolutionary relationship between IFIT1 and IFIT1B paralogs is complicated by recombination. As such, the assertion that the proteins studied in this manuscript are IFIT1 orthologs requires additional support than the percent identity plot shown in Figure 3B. 

(3) Some of the results and discussion text could be more focused on the model of evolution-driven changes in IFIT1 specificity. In particular, the chimpanzee data are interesting, but it would appear that this protein has lost all antiviral function, rather than changing its antiviral specificity like some other examples in this paper. As such, the connection between the functional mapping of individual residues with the positive selection analysis is somewhat confusing. It would be more clear to discuss this as a natural loss of function of this IFIT1, which has occurred elsewhere repeatedly across the mammalian tree. 

(4) In other places in the manuscript, the strength of the differences in antiviral specificity could be highlighted to a greater degree. Specifically, the text describes a number of interesting examples of differences in inhibition of VSV versus VEEV from Figure 3C and 3D, but it is difficult for a reader to assess this as most of the dots are unlabeled and the primary data are not uploaded. A few potential suggestions would be to have a table of each ortholog with % infection by VSV and % infection by VEEV. Another possibility would be to plot these data as an XY scatter plot. This would highlight any species that deviate from the expected linear relationship between the inhibition of these two viruses, which would provide a larger panel of interesting IFIT1 antiviral specificities than the smaller number of species shown in Figure 4. 

We thank the reviewer for their fair assessment of our manuscript. As the reviewer requested, we performed GARD analysis on our alignments used for PAML, FUBAR, and MEME (New Supp Fig 1). By GARD, we found 1 or 2 predicted breakpoints in each clade. However, much of the sequence was after or between the predicted breakpoints. Therefore, we were able to reanalyze for sites undergoing positive selection in the large region of the sequence that do not span the breakpoints. We were able to validate almost all sites originally identified as undergoing positive selection still exhibit signatures of positive selection taking these breakpoints into account: primates (11/12), bats (14/16), ungulates (30/37), and carnivores (2/4). To further validate our positive selection analysis, we used Recombination Detection Program 4 (RDP4) to remove inferred recombinant sequences from the primate IFIT1 alignment and performed PAML, FUBAR, and MEME. Once again, the sites in our original anlaysis were largely validated by this method. Importantly, sites 170, 193, and 366 in primates, which are discussed in our manuscript, were found to be undergoing positive selection in 2 of the 3 analyses using alignments after the indicated breakpoint in GARD and after removal of recombinant sequences by RDP4. We have updated the text to acknowledge IFIT1/IFIT1B recombination more clearly and include the GARD analysis as well as PAML, FUBAR, and MEME reanalysis taking into account predicted breakpoints by GARD and RDP4. Furthermore, to increase evidence that the sequences used in this study for both computational and functional analysis are IFIT1 orthologs rather than IFIT1B, we have included a maximum likelihood tree after aligning coding sequences on the C-terminal end (corresponding to bases 907-1437 of IFIT1). In Daughtery et al. 2016 (PMID: 27240734) this strategy was used to distinguish between IFIT1 and IFITB. All sequences used in our study grouped with IFIT1 sequences (including many confirmed IFIT1 sequences used in Daughterty et al.) rather than IFIT1B sequences or IFIT3. This new data, including the GARD, RDP4, and maximum likelihood tree is included as a new Supplementary Figure 1.

We also agree with the reviewer that it is possible that chimpanzee IFIT1 has lost antiviral function due to the residues 364 and 366 that differ from human IFIT1. We have updated the discussion sections to include the possibility that chimpanzee IFIT1 is an example of a natural loss of function that has occurred in other species over evolution as well as the potential consequences of this occurrence. Regarding highlighting the strength of differences in antiviral activity between IFIT1 orthologs, we have included several updates to strengthen the ability of the reader to assess these differences. First, we have included a supplementary table that includes the infection data for each ortholog from the VEEV and VSV screen to allow for readers to evaluate ranked antiviral activity of the species that suppress these viruses. In addition, the silhouettes next to the dot plots indicate the top ranked hits in order of viral inhibition (with the top being the most inhibitory) giving the reader a visual representation in the figure of top antiviral orthologs during our screen. We have also updated the figure legend to inform the reader of this information.

Reviewer #3 (Public Review):  

Summary: 

This manuscript by McDougal et al, demonstrates species-specific activities of diverse IFIT1 orthologs and seeks to utilize evolutionary analysis to identify key amino acids under positive selection that contribute to the antiviral activity of this host factor. While the authors identify amino acid residues as important for the antiviral activity of some orthologs and propose a possible mechanism by which these residues may function, the significance or applicability of these findings to other orthologs is unclear. However, the subject matter is of interest to the field, and these findings could be significantly strengthened with additional data.

Strengths:

Assessment of multiple IFIT1 orthologs shows the wide variety of antiviral activity of IFIT1, and identification of residues outside of the known RNA binding pocket in the protein suggests additional novel mechanisms that may regulate IFIT1 activity.

Weaknesses:

Consideration of alternative hypotheses that might explain the variable and seemingly inconsistent antiviral activity of IFIT1 orthologs was not really considered. For example, studies show that IFIT1 activity may be regulated by interaction with other IFIT proteins but was not assessed in this study.

Given that there appears to be very little overlap observed in orthologs that inhibited the viruses tested, it's possible that other amino acids may be key drivers of antiviral activity in these other orthologs. Thus, it's difficult to conclude whether the findings that residues 362/4/6 are important for IFIT1 activity can be broadly applied to other orthologs, or whether these are unique to human and chimpanzee IFIT1. Similarly, while the hypothesis that these residues impact IFIT1 activity in an allosteric manner is an attractive one, there is no data to support this.  

We thank the reviewer for their fair assessment of our manuscript. To address the weaknesses that the reviewer has pointed out we have expanded the discussion to more directly address alternate hypotheses, such as the possibility of IFIT1 activity being regulated by interaction with other IFIT proteins. Furthermore, we expanded the discussion to include an alternate hypothesis for the role of residues 364 and 366 in primate IFIT1 besides allosteric regulation. In addition, we did not intend to claim or imply that residues 364/6 are the key drivers of antiviral activity for all IFITs tested. However, we speculate that within primates these residues may play a key role as these residues differ between chimpanzee IFIT1 (which lacks significant antiviral activity towards the viruses tested in this study) and human IFIT1 (which possesses significant antiviral activity). In addition, these residues seem to be generally conserved in primate species, apart from chimpanzee IFIT1. We have included changes to the text to more clearly indicate that we highlight the importance of these residues specifically for primate IFIT1, but not necessarily for all IFIT1 proteins in all clades.

Reviewer #1 (Recommendations for the authors): 

(1) The readers would benefit from a more detailed background on the concept and estimation of positive selection for the readers, including the M7/8 models in PAML. 

We have included more information in the text to provide a better background for the concepts of positive selection and how PAML tests for this using M7 and M8 models.

(2) Presentation of data 

a) Figure 3C and 3D: is there a better way to present the infection data so the readers can tell the ranked antiviral activity of the species that suppress VEEV? 

We have included a supplementary table that includes the infection data for each ortholog from the VEEV and VSV screen to allow for readers to evaluate ranked antiviral activity of the species that suppress these viruses. In addition, the silhouettes next to the dot plots indicate the top ranked hits in order of viral inhibition (with the top being the most inhibitory). We have updated the figure legend to inform the reader of this information as well.

b) Figure 4C and 4D: consider putting the western blot in Supplementary Figure 1 underneath the infection data or with the heatmap so readers can compare it with the antiviral activity. 

We have also included quantification of the western blots performed to evaluate IFIT1 expression during the experiments shown in Figure 4C and 4D in an updated Figure 4B. We have also included normalized expression values with the heatmap shown in an updated Figure 4G so the reader can evaluate potential impact of protein expression on antiviral activity for all infection experiments shown in figure 4.

(3) Line 269-270: as a rationale for narrowing the species to human, black flying fox, and chimp IFIT1, human and black flying fox were chosen because they strongly inhibit VEEV, but pangolin wasn't included even though it had the strongest anti-VEEV activity? 

The rationale for narrowing the species to human, black flying fox, and chimpanzee IFIT1 was related to the availability of biological tools, high quality genome/transcriptome sequencing databases, and other factors. Specifically human and chimp IFIT1 are closely related but have variable antiviral activities, making their comparison highly relevant. Bats are well established as reservoirs for diverse viruses, whereas the reservoir status of many other mammals is less well defined. Furthermore, purifying large amounts of high quality IFIT1 protein after bacterial expression was another limitation to functional studies. We have added this information into the manuscript text.

(4) Figure 5A: to strengthen the claim that "species-specific antiviral activities of IFIT1s can be partly explained by RNA binding potential", it would be good to include one more positive and one more negative control. In other words, test the cap0 RNA binding activity of an IFIT1 ortholog that strongly inhibits VEEV and an ortholog that does not. It would also be good to discuss why chimp IFIT1 still shows dose-dependent RNA binding yet it is one of the weakest at inhibiting VEEV. 

We appreciate the reviewer's suggestion to include more controls and expand the dataset. While we understand the potential value of expanding the dataset, we believe that human IFIT1 serves as a robust positive control and human IFIT1 R187 (RNA-binding deficient) serves as an established negative control. Future experiments with other purified IFITs from other species will indeed strengthen evidence linking IFIT1 species-specific activity and RNA-binding.

Regarding chimpanzee IFIT1, we acknowledge there appears to be some dose-dependent Cap0 RNA-binding. However, the binding affinity is much weaker than that of human or black flying fox IFIT1. We speculate that during viral infection reduced binding affinity could impair the ability of chimpanzee IFIT1 to efficiently sequester viral RNA and inhibit viral translation. This reduction in binding affinity may, therefore, allow the cell to be overwhelmed by the exponential increase in viral RNA during replication resulting in an ineffective antiviral IFIT1. In the literature, a similar phenomenon is observed by Hyde et. al (PMID: 24482115). In this study, the authors test mouse Ifit1 Cap0 RNA binding by EMSA of the 5’ UTR sequence of VEEV RNA containing an A or G at nucleotide position 3. EMSA shows binding of both the A3 and G3 Cap0 VEEV RNA sequences, however stronger Ifit1 binding is observed for A3 Cap0 RNA sequence. The consequences of the reduced Ifit1 binding of the G3 Cap0 VEEV RNA are observed in vitro by a substantial increase in viral titers produced from cells as well as an increase in protein produced in a luciferase-based translation assay. The authors also show in vivo relevance of this reduction of Ifit1 binding as WT B6 mice infected with VEEV containing the A3 UTR exhibited 100% survival, while WT B6 mice infected with VEEV containing the G3 UTR survived at a rate of only ~25%. Therefore, the literature supports that a decrease in Cap0 RNA binding by an IFIT protein (while still exhibiting Cap0 RNA binding) observed by EMSA can result in considerable alterations of viral infection both in vitro and in vivo.

Minor: 

(1) Line 82: "including 5' triphosphate (5'-ppp-RNA), or viral RNAs..." having a comma here will make the sentence clearer. 

We have improved the clarity of this sentence. It now reads, “IFIT1 binds uncapped 5′triphosphate RNA (5′-ppp-RNA) and capped but unmethylated RNA (Cap0, an m⁷G cap lacking 2′-O methylation).”

(2) Line 100: "...similar mechanisms have been at least partially evolutionarily conserved in IFIT proteins to restrict viral infection by IFIT proteins". 

We have updated the text to improve clarity by revising the sentence to “VEEV TC-83 is sensitive to human IFIT1 and mouse Ifit1B, indicating at least partial conservation of antiviral function by IFIT proteins."

(3) Line 109: "signatures of rapid evolution or positive selection" would put positive selection second because that is the more technical term that can benefit from the more layperson term (rapid evolution). 

We have updated this sentence incorporating this suggestion. “Positive selection, or rapid evolution, is denoted by a high ratio of nonsynonymous to synonymous substitutions (dN/dS >1).”

(4) Lines 116-117: "However, this was only assessed in a few species" would benefit from a citation. 

We have inserted the citation.

(5) Line 127 heading: "IFIT1 is rapidly evolving in mammals" would be more accurate to say "in major clades of mammals". 

We have updated the text to include this suggestion.

(6) Line 165: "IFIT1 L193 mutants". 

We have updated the text to rephrase this for clarity.

(7) Line 170: two strains of VEEV were mentioned in the Intro, so it would be good to specify which strain of VEEV was used?

We have updated the text to clarify the VEEV strain. In this study, all experiments were performed using the VEEV TC-83 strain.

(8) Line 174: "Indeed, all mutants at position 193, whether hydrophobic or positively charged, inhibited VEEV similarly to the WT..." It should read "all hydrophobic and positively charged mutants inhibited VEEV similarly to the WT...". 

We corrected as suggested. 

(9) Line 204: what are "control cells"? Cells that are mock-infected, or cells without IFIT1? 

We have updated the text to improve clarity. What we refer to as control cells, were cells expressing an empty vector control rather than an IFIT1.

(10) Need to clarify n=2 and n=3 replicates throughout the manuscript. Does that refer to three independent experiments? Or an experiment with triplicate wells/samples? 

We have updated the text to say “independent experiments” instead of “biological replicates” to prevent any confusion.  All n=2 or n=3 replicates denote independent experiments.

(11) Line 254: "dominant antiviral effector against the related human parainfluenza virus type 5..." 

We have updated the text to improve clarity.

(12) Line 271: "The black flying fox (Pteropus alecto), is a model megabat species..." scientific name was italicized here but not elsewhere. Remove comma.

We have updated the text accordingly.

(13) Line 293: "...chimpanzee IFIT1 lacked these properties" but chimp IFIT1 can bind cap0 RNA, just at a lower level. 

We have updated the text to acknowledge that chimpanzee IFIT1 can bind cap0 RNA, albeit at a lower level than human IFIT1.

(14) Figure 6B: please fix the x-axis labels. They're very cramped. 

We have updated the x-axis labels for figure 6B and figure 6D to improve clarity.

(15) Line 609: "...trimmed and aligned"? 

Our phrasing is to indicate that coding sequences were aligned, and gaps were removed to reduce the chance of false positive signal by underrepresented codons such as gaps or short insertions. We have removed “trimmed” from the text and changed the text to say “aligned sequences” to increase clarity.

Reviewer #2 (Recommendations for the authors): 

(1) Numbers less than 10 should be spelled out throughout the manuscript (e.g. line 138). 

We have updated the text to reflect the request.

(2) Line 165: "expression of IFIT1 193 mutants" should be rephrased. 

We have updated the text to rephrase this sentence for clarity.

(3) A supplemental table or file should be included that contains the accession number and species names of sequences used for evolutionary analyses and for functional testing. In addition, the alignments that were used for positive selection can be included.  

We have included a supplemental file containing accession numbers, species names for evolutionary analysis and functional studies. In addition, this table includes the infection data for each IFIT1 homolog for the screen performed in figure 3.

(4) The discussion of potential functions of the C-terminus of IFIT1 should include possible interactions with other proteins. In particular, the C-terminus of IFIT1 has been shown to interact with IFIT3 in a way that modulates its activity (PMID: 29525521). Although residues 362-366 were not shown in that paper to interact with a fragment of IFIT3, it is possible that these residues may be important for interaction with full-length IFIT3 or some other IFIT1 binding partner. 

We thank the reviewer for their suggestion. We have expanded the discussion to explore the possibility that residues 364 and 366 of IFIT1 may be involved in IFIT1-IFIT3 interactions and consequently Cap0 RNA-binding and antiviral activity.

(5) The quantification of the EMSAs should be described in more detail. In particular, from looking at the images shown in Figure 5A, it would appear that human and chimpanzee IFIT1 show similar degrees of probe shift, while the human R187H panel shows no shifting at all. However, the quantification shows chimpanzee IFIT1 as being statistically indistinguishable from human R187H. Additional information on how bands were quantified and whether they were normalized to unshifted RNA would be helpful in attempting to resolve this visual discordance. 

EMSAs were quantified by determining Adj. Vol. Intensity in ImageLab (BioRad), which subtracts background signal, after imaging at the same exposure and SYBR Gold staining time. To determine Adj. Vol. Intensity, we drew a box (same size for each gel and lane for each replicate) for each lane above the free probe. These values were not normalized to unshifted RNA, however equal RNA was loaded. While the ANOVA shows no significant difference, between human R187H and chimpanzee IFIT1 band shift intensity, this is potentially due to the between group variance in the ANOVA. The increase in the AUC value for chimpanzee IFIT1 is 36.4% higher than R187H.

The AUC of Adj. Vol. Intensity of human IFIT1 band shift is roughly 2-fold more than that of chimpanzee IFIT1. We believe this matches with the visual representation as well, as human IFIT1 has a darker “upper” band in the shift, as well as a clear dark “lower” band that is not well defined in the chimpanzee shift. Furthermore, the upper band of the chimpanzee IFIT1 shift appears to be as intense in the 400nM as the upper band in the 240nM human IFIT1 lane, without taking into account the lower band seen for human IFIT1 as well. We included this quantification as kD was unable to be calculated due to no clear probe disappearance and we do not intend for this quantification to act as a substitute for binding affinity calculations, rather to aid the reader in data interpretation.

Reviewer #3 (Recommendations for the authors): 

(1) IFIT1 has been demonstrated to function in conjunction with other IFIT proteins, do you think the absence of antiviral activity is due to isolated expression of IFIT1 without these cofactors, and therefore might explain why there was little overlap observed in orthologs that inhibited the viruses tested (Figure 3, lines 209-210). 

We do not believe that isolated expression of IFIT1 without cofactors (such as orthologous IFIT proteins) would fully explain the disparities in antiviral activity as many IFIT1s that expressed inhibited either VSV or VEEV in our screen. However, we acknowledge that the expression of IFIT1 alone does create a limitation in our study as IFIT1 antiviral activity and RNA-binding can be modulated by interactions with other IFIT proteins. Therefore, we do believe that it is possible that co-expression of IFIT1 with other IFITs from a given species might potentially enhance antiviral activity. Future studies may shed light on this.

(2) Figure 5 - Calculating the Kd for each protein would be more informative. How does the binding affinity of these IFIT1 proteins compare to that which has previously been reported? 

We are unable to accurately determine kD as there is not substantial diminished signal of the free probe. Therefore, we are only able to compare IFIT1 protein binding between species without accurate mathematical calculation of binding affinity. Our result does appear similar to that of mouse Ifit1 binding to VEEV RNA (PMID: 24482115), in which the authors also do not calculate a kD for their RNA EMSA.

(3) Mutants 364 and 366 may not have direct contact with RNA, but RNA EMSA data presented suggest that the binding affinity may be different (though this is hard to conclude without Kd data). Additional biochemical data with these mutants might provide more insight here. 

We agree that further studies using 364 and 366 double mutant human and chimpanzee protein in EMSAs would provide additional biochemical data and provide insight into the role of these residues in direct RNA binding. We acknowledge this is a limitation of our study as we provide only genetic data demonstrating the importance of these residues.

(4) Given that there appears to be very little overlap observed in orthologs that inhibited the viruses tested, it's possible that other amino acids may be key drivers of antiviral activity in these other orthologs. Thus, it's difficult to conclude whether the findings that residues 362/4/6 are important for IFIT1 activity can be broadly applied to other orthologs. A more systematic assessment of the role of these mutations across multiple diverse orthologs would provide more insight here. Do other antiviral proteins show this trend (ie exhibit little overlap in orthologs that inhibit these viruses). What do you think might be driving this? 

We agree that other residues outside of 364 and 366 may be key drivers of antiviral activity across the IFTI1 orthologs tested. We do not hypothesize that this will broadly apply across IFIT1 from diverse clades of mammals as overall amino acid identity can differ by over 30%. However, based on the chimpanzee and human IFIT1 data, as well as sequence alignment within primates specifically, we believe these residues may be key for primate (but not necessarily other clades of mammals) IFIT1 antiviral activity.

Regarding if other antiviral proteins show little overlap in orthologs that inhibit a given virus, to our knowledge such a functional study with this large and divergent dataset of orthologs has not been performed. However, there are many examples of restriction factors exhibiting speciesspecific antiviral activity when ortholog screens have been performed. For example, HIV was reported to be suppressed by MX2 orthologs from human, rhesus macaque, and African green monkey, but not sheep or dog MX2 (PMID: 24760893). In addition, foamy virus was inhibited by the human and rhesus macaque orthologs of PHF11, but not the mouse and feline orthologs (PMID: 32678836). Furthermore, studies from our lab have shown variability in RTP4 ortholog antiviral activity inhibition towards viruses much as hepatitis C virus (HCV), West Nile virus (WNV), and Zika virus (ZIKV) (PMID: 33113352).

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public Review): 

Summary: 

Weiss and co-authors presented a versatile probabilistic tool. aTrack helps in classifying tracking behaviors and understanding important parameters for different types of single particle motion types: Brownian, Confined, or Directed motion. The tool can be used further to analyze populations of tracks and the number of motion states. This is a stand-alone software package, making it user-friendly for a broad group of researchers. 

Strengths: 

This manuscript presents a novel method for trajectory analysis. 

Weaknesses: 

(1) In the results section, is there any reason to choose the specific range of track length for determining the type of motion? The starting value is fine, and would be short enough, but do the authors have anything to report about how much is too long for the model? 

We chose to test the range of track lengths (five-to-hundreds of steps) to cover the broad range of scenarios arising from single proteins or fluorophores to brighter objects with more labels.  While there is no upper-limit per se, the computation time of our method scales linearly with track length, 100 time-points takes ~2 minutes to run on a standard consumer-level desktop CPU. We have added the following sentence to note the time-cost with trajectory length:  

“The recurrent formula enables our model computation time to scale linearly with the number of time points.”

(2) Robustness to model mismatches is a very important section that the authors have uplifted diligently. Understanding where and how the model is limited is important. For example, the authors mentioned the limitation of trajectory length, do the authors have any information on the trajectory length range at which this method works accurately? This would be of interest to readers who would like to apply this method to their own data. 

We agree that limitations are important to estimate, and trajectory length is an important consideration when choosing how to analyze a dataset. We report the categorization certainty, i.e. the likelihood differences, for a range of track lengths (Fig. 2 a,c, Fig. 3c-d, and Fig. 4 c,g.).

For example, here are the key plots from Fig. 2 quantifying the relative likelihoods, where being within the light region is necessary. The light areas represent a useful likelihood ratio.

We only performed analysis up to track lengths of 600 time steps but parameter estimations and significance can only improve when increasing the track length as long as the model assumptions are verified. The broader limitations and future opportunities for new methods are now expanded upon in the discussion, for example switching between states and model and state and model ambiguities (bound vs very slow diffusion vs very slow motion).

(3) aTrack extracts certain parameters from the trajectories to determine the motion types. However, it is not very clear how certain parameters are calculated. For example, is the diffusion coefficient D calculated from fitting, and how is the confinement factor defined and estimated, with equations? This information will help the readers to understand the principles of this algorithm.

We apologize for the confusion. All the model parameters are fit using the maximum likelihood approach. To make this point clearer in the manuscript, we have made three changes:

(1) We modified the following sentence to replace “determined” with "fit”:

“Finally, Maximum Likelihood Estimation (MLE) is used to fit the underlying parameter value”

(2) We added the following sentence in the main text :

“In our model, the velocity is the characteristic parameter of directed motion and the confinement factor represents the force within a potential well. More precisely, the confinement factor $l$ is defined such that at each time step the particle position is updated by $l$ times the distance particle/potential well center (see the Methods section for more details).”.

(3) We have added a new section in the methods, called Fitting Method, where we have added the explanation below:

“For the pure Brownian model, the parameters are the diffusion coefficient and the localization error. For the confinement model, the parameters are the diffusion coefficient, the localization error, confinement factor, and the diffusion coefficientof the potential well. For the directed model, the parameters are the diffusion coefficient, the localization error, the initial velocity and the acceleration variance.

These parameters are estimated using the maximum likelihood approach which consists in finding the parameters that maximize the likelihood. We realize this fitting step using gradient descent via a TensorFlow model. All the estimates presented in this article are obtained from a single set of initial parameters to demonstrate that the convergence capacity of aTrack is robust to the initial parameter values.”

(4) The authors mentioned the scenario where a particle may experience several types of motion simultaneously. How do these motions simulated and what do they mean in terms of motion types? Are they mixed motion (a particle switches motion types in the same trajectory) or do they simply present features of several motion types? It is not intuitive to the readers that a particle can be diffusive (Brownian) and direct at the same time. 

In the text, we present an example where one can observe this type of motion to help the reader understand when this type of motion can be met: “Sometimes, particles undergo diffusion and directed motion simultaneously, for example, particles diffusing in a flowing medium (Qian 1991).”

This is simulated by the addition of two terms affecting the hidden position variable before adding a localization term to create the observed variable. In the analysis, this manifests as non-zero values for the diffusion coefficient and the linear velocity. For example, Figure 4g and the associated text, where a single particle moves with a directed component and a Brownian diffusion component at each step.

We did not simulate transitions between types of motion. Switching is not treated by this current model; however, this limitation is described in the discussion and our team and others are currently working on addressing this challenge.

Reviewer #2 (Public Review): 

Summary: 

The authors present a software package "aTrack" for identification of motion types and parameter estimation in single-particle tracking data. The software is based on maximum likelihood estimation of the time-series data given an assumed motion model and likelihood ratio tests for model selection. They characterized the performance of the software mostly on simulated data and showed that it is applicable to experimental data. 

Strengths: 

A potential advantage of the presented method is its wide applicability to different motion types. 

Weaknesses: 

(1) There has been a lot of similar work in this field. Even though the authors included many relevant citations in the introduction, it is still not clear what this work uniquely offers. Is it the first time that direct MLE of the time-series data was developed? Suggestions to improve would include (a) better wording in the introduction section, (b) comparing to other popular methods (based on MSD, step-size statistics (Spot-On, eLife 2018;7:e33125), for example) using the simulated dataset generated by the authors, (c) comparing to other methods using data set in challenges/competitions (Nat. Comm (2021) 12:6253).  

We thank the reviewer for this suggestion and agree that the explanation of the innovative aspects of our method in the introduction was not clear enough. We have now modified the introduction to better explain what is improved here compared to previous approaches.

“The main innovations of this model are: 1) it uses analytical recurrence formulas to perform the integration step for complex motion, improving speed and accuracy; 2) it handles both confined and directed motion; 3) anomalous parameters, such as the center of the potential well and the velocity vector are allowed to change through time to better represent tracks with changing directed motion or confinement area; and lastly 4) for a given track or set of tracks, aTrack can determine whether tracks can be statistically categorized as confined or directed, and the parameters that best describe their behavior, for example, diffusion coefficient, radius of confinement, and speed of directed motion.”

Regarding alternatives, we compare our method in the text to the best-performing algorithm of the

2021 Anomalous Diffusion (AnDi) Challenge challenge mentioned by the reviewer in Figure 6 (RANDI, Argun et al, arXiv, 2021, Muñoz-Gil et al, Nat Com. 2021). Notably, both methods performed similarly on fBm, but ours was more robust in cases where there were small differences between the process underlying the data and the model assumptions, a likely scenario in real datasets. Regarding Spot-On, this was not mentioned as it only deals with multiple populations of Brownian diffusers, preventing a quantitative comparison.

(2) The Hypothesis testing method presented here has a number of issues: first, there is no definition of testing statistics. Usually, the testing statistics are defined given a specific (Type I and/or Type II) error rate. There is also no discussion of the specificity and sensitivity of the testing results (i.e. what's the probability of misidentification of a Brownian trajectory as directed? etc).

We now explain our statistical approach and how to perform hypothesis testing with our metric in a new supplementary section, Statistical test. 

We use the likelihood ratio as a more conservative alternative to the p-value. In Fig S2, we show that our metric is an upper bound of the p-value and can be used to perform hypothesis testing with a chosen type I error rate. 

Related, it is not clear what Figure 2e (and other similar plots) means, as the likelihood ratio is small throughout the parameter space. Also, for likelihood ratio tests, the authors need to discuss how model complexity affects the testing outcome (as more complex models tend to be more "likely" for the data) and also how the likelihood function is normalized (normalization is not an issue for MLE but critical for ratio tests). 

We present the likelihood ratio as an upper bound of the p-value. Therefore, we can reject the null hypothesis if it is smaller than a given threshold, e.g. 0.05, but this number should be decreased if multiple tests are performed. The colorscale we show in the figure is meant to highlight the working range (light), and ambiguous range (dark) of the method.

As the reviewer mentions, we expect the alternative hypothesis to result in higher likelihoods than the simpler null hypothesis for null hypothesis tracks, but, as seen in the Fig S2, the likelihood ratio of a dataset corresponding to the null hypothesis is strongly skewed toward its upper limit 1. This means that for most of the tracks, the likelihood is not (or little) affected by the model complexity. The likelihoods of all the models are normalized so their integrals over the data equals 1/A with A the area of the field of view which is independent of the model complexity.

(3) Relating to the mathematical foundation (Figure 1b). The measured positions are drawn as direct arrows from the real position states: this infers instantaneous localization. In reality, there is motion blur which introduces a correlation of the measured locations. Motion blur is known to introduce bias in SPT analysis, how does it affect the method here? 

The reviewer raises an important point as our model does not explicitly consider motion blur. We have now added a paragraph that presents how our model performs in case of motion blur in the section called Robustness to model mismatches. This section and the corresponding new Supplemental Fig. S7 demonstrate that the estimated diffusion length is accurate so long as the static localization error is higher than the dynamic localization error. If the dynamic localization error is higher, our model systematically underestimates the diffusion length by a factor 0.81 = (2/3)^0.5 which can be corrected for with an added post-processing step.  

(4) The authors did not go through the interpretation of the figure. This may be a matter of style, but I find the figures ambiguous to interpret at times.  

We thank the reviewer for their feedback on improving the readability. To avoid overly repetitive and lengthy sections of text, we have opted for a concise approach. This allows us to present closely related panels at the same point in the text, while not ignoring important variations and tests. Considering this feedback and the reviewers, we have added more information and interpretation throughout our manuscript to improve interpretability.

(5) It is not clear to me how the classification of the 5 motion types was accomplished. 

We have modified the specific text related to this figure to describe an illustrative example to show how one could use aTrack on a dataset where not that much is known: First, we present the method to determine the number of states; second, we verify the parameter estimates correspond to the different states.  

Classifying individual tracks is possible. While not done in the section corresponding to Fig. 5, this is done in Fig. 7 and a new supplementary plot, Fig. S9b (shown below). In brief, this is accomplished with our method by computing the likelihood of each track given each state. The probability that a given track is in state k equals the likelihood of the track given the state divided by the sum of the likelihoods given the different states. 

(6) Figure 3. Caption: what is ((d_{est}-0.1)/0.1)? Also panel labeled as "d" should be "e". 

Thank you for bringing these errors to our attention, the panel and caption have been corrected.

Reviewer #3 (Public Review): 

Summary: 

In this work, Simon et al present a new computational tool to assess non-Brownian single-particle dynamics (aTrack). The authors provide a solid groundwork to determine the motion type of single trajectories via an analytical integration of multiple hidden variables, specifically accounting for localization uncertainty, directed/confined motion parameters, and, very novel, allowing for the evolution of the directed/confined motion parameters over time. This last step is, to the best of my knowledge, conceptually new and could prove very useful for the field in the future. The authors then use this groundwork to determine the motion type and its corresponding parameter values via a series of likelihood tests. This accounts for obtaining the motion type which is statistically most likely to be occurring (with Brownian motion as null hypothesis). Throughout the manuscript, aTrack is rigorously tested, and the limits of the methods are fully explored and clearly visualised. The authors conclude with allowing the characterization of multiple states in a single experiment with good accuracy and explore this in various experimental settings. Overall, the method is fundamentally strong, wellcharacterised, and tested, and will be of general interest to the single-particle-tracking field. 

Strengths: 

(1) The use of likelihood ratios gives a strong statistical relevance to the methodology. There is a sharp decrease in likelihood ratio between e.g. confinement of 0.00 and 0.05 and velocity of 0.0 and 0.002 (figure 2c), which clearly shows the strength of the method - being able to determine 2nm/timepoint directed movement with 20 nm loc. error and 100 nm/timepoint diffusion is very impressive. 

We apologize for the confusion, the directed tracks in Fig 2 have no Brownian-motion component, i.e. D=0. We have made this clearer in the main text. Specifically, this section of the text refers to a track in linear motion with 2 nm displacements per step. With 70 time points (69 steps), a single particle which moved from 138 nm with a localization error of 20 nm (95% uncertainty range of 80 nm) can be statistically distinguished from slow diffusive motion.

In Fig. 4g, we explore the capabilities of our method to detect if a diffusive particle also has a directed motion component. 

(2) Allowing the hidden variables of confinement and directed motion to change during a trajectory (i.e. the q factor) is very interesting and allows for new interpretations of data. The quantifications of these variables are, to me, surprisingly accurate, but well-determined. 

(3) The software is well-documented, easy to install, and easy to use. 

Weaknesses: 

(1) The aTrack principle is limited to the motions incorporated by the authors, with, as far as I can see, no way to add new analytical non-Brownian motion. For instance, being able to add a dynamical stateswitching model (i.e. quick on/off switching between mobile and non-mobile, for instance, repeatable DNA binding of a protein), could be of interest. I don't believe this necessarily has to be incorporated by the authors, but it might be of interest to provide instructions on how to expand aTrack.  

We agree that handling dynamic state switching is very useful and highlight this potential future direction in the discussion. The revised text reads:

“An important limitation of our approach is that it presumes that a given track follows a unique underlying model with fixed parameters. In biological systems, particles often transition from one motion type to another; for example, a diffusive particle can bind to a static substrate or molecular motor (46). In such cases, or in cases of significant mislinkings, our model is not suitable. However, this limitation can be alleviated by implicitly allowing state transitions with a hidden Markov Model (15) or alternatives such as change-point approaches (30, 47, 48), and spatial approaches (49).”

(2) The experimental data does not very convincingly show the usefulness of aTrack. The authors mention that SPBs are directed in mitosis and not in interphase. This can be quantified and studied by microscopy analysis of individual cells and confirming the aTrack direction model based on this, but this is not performed. Similarly, the size of a confinement spot in optical tweezers can be changed by changing the power of the optical tweezer, and this would far more strongly show the quantitative power of aTrack. 

We agree with the reviewer and have revised the biological experiment section significantly to better illustrate the potential of aTrack in various use cases.

Now, we show an experiment to quantify the effect of LatA, an actin inhibitor, on the fraction of directed tracks obtained with aTrack. We find that LatA significantly decreases directed motion while a LatA-resistant mutant is not affected (Fig7a-c).

As suggested by the reviewer, we have expanded the optical tweezer experiment by varying the laser power. As expected, increasing the laser power decreases the confinement radius.

(3) The software has a very strict limit on the number of data points per trajectory, which is a user input. Shorter trajectories are discarded, while longer trajectories are cut off to the set length. It is not explained why this is necessary, and I feel it deletes a lot of useful data without clear benefit (in experimental conditions).

We thank the reviewer for this recommendation; we have now modified the architecture of our model to enable users to consider tracks of multiple lengths. Note that the computation time is proportional to the longest track length times the number of tracks.  

Reviewer #2 (Recommendations For The Authors): 

Develop a better mathematical foundation for the likelihood ratio tests. 

We added more explanation of the likelihood ratio tests and their interpretation a new section entitled Statistical test in the supplementary information to address this recommendation.

Place this work in clearer contexts. 

We have now revised the introduction to better contextualize this work.

Improve manuscript clarity. 

Based on reviewer feedback and input from others, we have addressed this point throughout the article to improve readability.

Make the code available. 

The code is available on https://github.com/FrancoisSimon/aTrack, now including code for track generation.

Reviewer #3 (Recommendations For The Authors): 

(1) I believe the underlying model presented in Figure 1 is of substantial impact, especially when considering it as a simulation tool. I would suggest the authors make their method also available as a simulator (as far as I can tell, this is not explicitly done in their code repository, although logically the code required for the simulator should already be in the codebase somewhere). 

Thank you for this suggestion, the simulation scripts are now on the Github repository together with the rest of the analysis method. https://github.com/FrancoisSimon/aTrack

(2) The authors should explore and/or discuss the effects of wrong trajectory linking to their method. Throughout the text, fully correct trajectory linking is assumed and assessed, while in real experiments, it is often the case that trajectory linking is wrong, e.g. due to blinking emitters, imaging artefacts, high-density localizations, etc etc. This would have a major impact on the accuracy of trajectories, and it is extremely relevant to explore how this is translated to the output of aTrack. 

As the reviewer notes, our current model does not account for track mislinking. This limits the method to data with lower fluorophore-densities, which is the typical use-case for SPT. We have added a brief description of the issue into the discussion of limitations.  

(3) aTrack only supports 2D-tracking, but I don't believe there is a conceptual reason not to have this expanded to three dimensions. 

The stand-alone software is currently limited to 2D tracks, however, the aTrack Python package works for any number of dimensions (i.e. 1-3). Note that since the current implementation assumes a single localization error for all axes, more modifications may be required for some types of 3D tracking. See https://github.com/FrancoisSimon/aTrack for more details about aTrack implementations.

(4) Crucial information is missing in the experimental demonstrations. Especially in the NP-bacteria dataset, I miss scalebars, and information on the number of tracks. It is not explained why 5 different states are obtained - especially because I would naively expect three states: immobile NPs (e.g. stuck to glass), diffusing NPs, and NPs attached to bacteria, and thus directed. Figure 7e shows three diffusive states (why more than one?), no immobile states (why?), and two directed states (why?). 

We thank the reviewer for pointing out these issues. We have now added scalebars and more experimental details to the figure and text as well as modifying the plot to more clearly emphasize the directed nanoparticles that are attached to cells from the diffusive nanoparticles.  

Likely, our focal plane was too high to see the particles stuck on glass. The multiple diffusive states may be caused by different sizes of nanoparticle complexes, the multiple directed states can be caused by the fact that directed motion of the cell-attached-nanoparticles occasionally shows drastic changes of orientations. We have also clarified in the text how multiple states can help handle a heterogeneous population as was shown by Prindle et al. 2022, Microbiol Spectr. The characterization and phenotyping of microbial populations by nanoparticle tracking was published in Zapata et al. 2022, Nanoscale. 

(5) I don't think I agree that 'robustness to model mismatches' is a good thing. Very crudely, the fact that aTrack finds fractional Brownian motion to be normal Brownian motion is technically a downside - and this should be especially carefully positioned if (in the future) a fractional Brownian motion model would be added to aTrack. I think that the author's point can be better tested by e.g. widely varying simulated vs fitted loc precision/diffusion coefficient (which are somewhat interchangeable).

In this context, our intention in describing the robustness to “model mismatches” refers to classifying subdiffusion as subdiffusive irrespective of the exact subdiffusion motion physics (as well as superdiffusion), that is, to use aTrack how MSD analysis is often deployed. This is important in the context of real-world applications where simple mathematical models cannot perfectly represent real tracks with greater complexity. 

Inevitably, some fraction of tracks with a pure Brownian motion may appear to match with a fractional Brownian motion, and thus statistical tests are needed to determine if this is significant. In general, aTrack finds fBm to be normal Brownian motion only when the anomalous coefficient is near 1, i.e. when the two models are indeed the same. When analysing fBm tracks with anomalous coefficients of 0.5 or 1.5, aTrack find that these tracks are better explained by our confined diffusion model or directed motion model, respectively (Please see Fig. 6a, copied below). 

To better clarify our objective, the section now has a brief introduction that reads:

“One of the most important features of a method is its robustness to deviations from its assumptions. Indeed, experimental tracking data will inevitably not match the model assumptions to some degree, and models need to be resilient to these small deviations.”  

Smaller points: 

(1) It is not clear what a biological example is of rotational diffusion. 

We modified the text to better explain the use of rotational diffusion.

(2) The text in the section on experimental data should be expanded and clarified, there currently are multiple 'floating sentences' that stop halfway, and it does not clearly describe the biological relevance and observed findings.  

We thank the reviewer for pointing out this issue. We have reworked the experimental section to better and more clearly explain the biological relevance of the findings.

(3) Caption of figure 3: 'd' should be 'e'. 

(4) Caption of Figure 7: log-likelihood should be Lconfined - Lbrownian, I believe. 

(5) Equation number missing in SI first sentence. 

(6) Supplementary Figure 1 top part access should be Lc-Lb instead of Ld-Lb. 

We have made these corrections, thank you for bringing them to our attention.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

In the present work, Chen et al. investigate the role of short heat shock factors (S-HSF), generated through alternative splicing, in the regulation of the heat shock response (HSR). The authors focus on S-HsfA2, an HSFA2 splice variant containing a truncated DNA-binding domain (tDBD) and a known transcriptional-repressor leucin-rich domain (LRD). The authors found a two-fold effect of S-HsfA2 on gene expression. On the one hand, the specific binding of S-HsfA2 to the heat-regulated element (HRE), a novel type of heat shock element (HSE), represses gene expression. This mechanism was also shown for other S-HSFs, including HsfA4c and HsfB1. On the other hand, S-HsfA2 is shown to interact with the canonical HsfA2, as well as with a handful of other HSFs, and this interaction prevents HsfA2 from activating gene expression. The authors also identified potential S-HsfA2 targets and selected one, HSP17.6B, to investigate the role of the truncated HSF in the HSR. They conclude that S-HsfA2-mediated transcriptional repression of HSP17.6B helps avoid hyperactivation of the HSR by counteracting the action of the canonical HsfA2.

The manuscript is well written and the reported findings are, overall, solid. The described results are likely to open new avenues in the plant stress research field, as several new molecular players are identified. Chen et al. use a combination of appropriate approaches to address the scientific questions posed. However, in some cases, the data are inadequately presented or insufficient to fully support the claims made. As such, the manuscript would highly benefit from tackling the following issues:

(1) While the authors report the survival phenotypes of several independent lines, thereby strengthening the conclusions drawn, they do not specify whether the presented percentages are averages of multiple replicates or if they correspond to a single repetition. The number of times the experiment was repeated should be reported. In addition, Figure 7c lacks the quantification of the hsp17.6b-1 mutant phenotype, which is the background of the knock-in lines. This is an essential control for this experiment

For the seedling survival rates and gene expression levels, we added statistical analysis based on at least two independent experiments. Figure 6E of the revised manuscript shows the phenotypes of the WT, hsp17.6b-1, HSP17.6B-KI, and HSP17.6B-OE plants and the statistical analysis of their seedling survival rates after heat exposure.

(2) In Figure 1c, the transcript levels of HsfA2 splice variants are not evident, as the authors only show the quantification of the truncated variant. Moreover, similar to the phenotypes discussed above, it is unclear whether the reported values are averages and, if so, what is the error associated with the measurements. This information could explain the differences observed in the rosette phenotypes of the S-HsfA2-KD lines. Similarly, the gene expression quantification presented in Figures 4 and 5, as well as the GUS protein quantification of Figure 3F, also lacks this crucial information.

RT‒qPCR analysis of the expression of these genes from at least two independent experiments was performed. We also added these missing information to the figure legends.

(3) The quality of the main figures is low, which in some cases prevents proper visualization of the data presented. This is particularly critical for the quantification of the phenotypes shown in Figure 1b and for the fluorescence images in Figures 4f and 5b. Also, Figure 9b lacks essential information describing the components of the performed experiments.

We apologize; owing to the limitations of equipment and technology, we will attempt to obtain high-quality images in the future. A detailed description of Figure 9b is provided in the methods section.

(4) Mutants with low levels of S-HsfA2 yield smaller plants than the corresponding wild type. This appears contradictory, given that the proposed role of this truncated HSF is to counteract the growth repression induced by the canonical HSF. What would be a plausible explanation for this observation? Was this phenomenon observed with any of the other tested S-HSFs?

We found that the constitutive expression of S-HsfA2 inhibits Arabidopsis growth. Considering this, Arabidopsis plants do not produce S-HsfA2 under normal conditions to avoid growth inhibition. However, under heat stress, Arabidopsis plants generate S-HsfA2, which contributes to heat tolerance and growth balance. In the revised manuscript, we provided supporting data indicating that S-HsfA4c-GFP or S-HsfB1-RFP constitutive expression confers Arabidopsis extreme heat stress sensitivity but inhibits root growth (Supplemental Figure S8). Therefore, this phenomenon is also observed in S-HsfA4c-GFP or S-HsfB1-RFP.

(5) In some cases, the authors make statements that are not supported by the results:<br /> (i) the claim that only the truncated variant expression is changed in the knock-down lines is not supported by Figure 1c;

In three S-HsfA2-KD lines, RT‒PCR splicing analysis revealed that HsfA2-II but not HsfA2-III is easily detected. In the revised manuscript, we added RT‒qPCR analysis, and the results revealed that the abundance of HsfA2-III and HsfA2-II but not that of the full-length HsfA2 mRNA significantly decreased under extreme heat (Figure 1C). Considering that HsfA2-III but not HsfA2-II is a predominant splice variant under extreme heat (Liu et al., 2013), S-HsfA2-KD may lead to the knockdown of alternative HsfA2 splicing transcripts, especially HsfA2-III.

(ii) the increase in GUS signal in Figure 3a could also result from local protein production;

We included this possibility in the results analysis.

(iii) in Figure 6b, the deletion of the HRE abolishes heat responsiveness, rather than merely altering the level of response; and

In the revised manuscript, we added new data concerning the roles of HREs and HSEs in the response of the HSP17.6B promoter to heat stress (Figure 6A). These results suggest that the HRE and HSE elements are responsible for the response of the HSP17.6B promoter to heat stress and that the HRE negatively regulates the HSP17.6B promoter at 37°C, whereas the HSE is positive at 42°C.

(iv) the phenotypes in Figure 8b are not clear enough to conclude that HSP17.6B overexpressors exhibit a dwarf but heat-tolerant phenotype.

When grown in soil, the HSP17.6B-OE seedlings presented a dwarf phenotype compared with the WT control. Heat stress resulted in browning of the WT leaves, but the leaves of the HSP17.6B-OE plants remained green, suggesting that the HSP17.6B-OE seedlings also presented a heat-tolerant phenotype in the soil. These results are qualitative but not quantitative experimental data; therefore, the conclusions are adjusted in the abstract and results sections.

Reviewer #2 (Public review):

Summary:

The authors report that Arabidopsis short HSFs S-HsfA2, S-HsfA4c, and S-HsfB1 confer extreme heat. They have truncated DNA binding domains that bind to a new heat-regulated element. Considering Short HSFA2, the authors have highlighted the molecular mechanism by which S-HSFs prevent HSR hyperactivation via negative regulation of HSP17.6B. The S-HsfA2 protein binds to the DNA binding domain of HsfA2, thus preventing its binding to HSEs, eventually attenuating HsfA2-activated HSP17.6B promoter activity. This report adds insights to our understanding of heat tolerance and plant growth.

Strengths:

(1) The manuscript represents ample experiments to support the claim.

(2) The manuscript covers a robust number of experiments and provides specific figures and graphs in support of their claim.

(3) The authors have chosen a topic to focus on stress tolerance in a changing environment.

Weaknesses:

(1) One s-HsfA2 represents all the other s-Hsfs; S-HsfA4c, and S-HsfB1. s-Hsfs can be functionally different. Regulation may be positive or negative. Maybe the other s-hsfs may positively regulate for height and be suppressed by the activity of other s-hsfs.

In this study, we used S-HsfA2, S-HsfA4c, and S-HsfB1 data to support the view that “splice variants of HSFs generate new plant HSFs”. We also noted that S-HsfA2 cannot represent a traditional S-HSF. S-HsfA4c and S-HsfB1 may have functions other than S-HsfA2 because of their different C-terminal motifs or domains. Different S-HSFs might participate in the same biological process, such as heat tolerance, through the coregulation of downstream genes. We added this information to the discussion section.

(2) Previous reports on gene regulations by hsfs can highlight the mechanism.

In the introduction section, we included these references concerning HSFs and S-HSFs.

(3) The Materials and Methods section could be rearranged so that it is based on the correct flow of the procedure performed by the authors.

The materials and methods and results sections are arranged in the logical order.

(4) Graphical representation could explain the days after sowing data, to provide information regarding plant growth.

The days after sowing (DAS) for the age of the Arabidopsis seedlings are stated in the Materials and Methods section and figure legends.

(5) Clear images concerning GFP and RFP data could be used.

We provided high-quality images of S-HsfA2-GFP and the GFP control (Figure 3 in the revised manuscript).

Reviewing Editor comments:

The EMSA shown in Figures 2, 3, 4, and 5, which are critical to support the manuscript's claims, are of poor quality, without any repeats to support. In addition, there is not much information about how these EMSA were done. I suggest including better EMSA in a new version of this manuscript.

Thank you for your suggestion. We added the missing information, including the detailed EMSA method and experiment repeat times in the methods section and figure legends. We provide high-quality images of HRE probes binding to nuclear proteins (Figure 4E).

Reviewer  #1 (Recommendations for the authors):

(1) The paper is overall well-written, but it could greatly benefit from reorganizing the results subsections. Currently, there are entire subsections dedicated to supplementary figures (e.g., lines 177-191) and main figures split into different subsections (e.g., lines 237-246). It is recommended to organize all the information related to a main figure into a single subsection and to incorporate the description of the corresponding supplementary figures. This would imply a general reorganization of the figures, moving some information to the supplementary data (for instance, the data in Figure 4 could be supplementary to Figure 5) and vice versa (Supplementary Figure 4 should be incorporated into main Figure 2, as it presents very important results). Also, Figures 7 and 8 would be better presented if merged into a single figure/subsection.

Thank you for your suggestion. We have merged some figures into a single figure according to the main information. In the current version, there are 8 main figures, which includes a new figure.

(2) Survival phenotypes vary widely, making reliable statistical analysis challenging. The chlorophyll and fresh weight quantifications presented in figures such as Figure 5 appear to effectively describe the phenomenon and allow for statistical comparisons. Figures 1 and 7 would benefit from including these measurements if the variability in survival percentages is too high to calculate statistical differences reliably. Also, in Figure 8, all chlorophyll measurements should be normalized to fresh weight rather than seedling number due to the dwarfism observed in the overexpressor lines.

Thank you for pointing out your concerns. We added statistical analysis based on at least two independent experiments, including Figures 1 and 7, to the original manuscript. In Figure 8 in the original manuscript, chlorophyll measurements were normalized to fresh weight.

(3) Typos: in Figure 3a it should be "min" not "mim"; in Supplementary Figure 3, the GFP and merge images are swapped.

We apologize for these errors, and we have corrected them. Supplementary Figure 3 was replaced with new images and was included in Figure 3 in the revised manuscript.

Reviewer  #2 (Recommendations for the authors):

(1) The abstract states "How this process is prevented to ensure proper plant growth has not been determined." The authors can be the first to do this, by adding graphical data on the height difference in hSfA2-arabidopis and wild-type Arabidopsis.

Thank you and agree with you. We have added this information to the new working model (Figure 8)

(2) The authors claim that Arabidopsis S-HsfA2, S-HsfA4c, and S-HsfB1; but have used S-HsfA2 to understand the action. The mechanisms being unknown for S-HsfA4c, and S-HsfB1 cannot be represented by S-HsfA2 to represent the mechanism.

Thank you for your valuable comments. In this study, we used S-HsfA2, S-HsfA4c, and S-HsfB1 data to support the view that “splice variants of HSFs generate new plant HSFs”. We also noted that S-HsfA2 cannot represent a traditional S-HSF because S-HsfA4c and S-HsfB1 may have functions other than S-HsfA2. Therefore, we deleted “representative S-HSF” from the revised manuscript. In the future, we will conduct in-depth research on the relevant mechanisms of S-HsfA4c and S-HsfB1 under your guidance.

(3) The authors can include which of the HSFs interacted with other genes of Arabidopsis reported by other researchers are positively or negatively regulated in heat response/ growth or the balance.

In the introduction section, we included these genes. AtHsfA2, AtHsfA3, and BhHsf1 confer heat tolerance in Arabidopsis but also result in a dwarf phenotype in plants (Ogawa et al., 2007; Yoshida et al., 2008; Zhu et al., 2009).

(4) The authors have started from the subsection plant materials and growth conditions. It is unclear from where the authors have found these HSF mutant Arabidopsis? Is it a continuation of some other work? As a reader, I am utterly confused because of the arrangement of the materials and methods section.

We apologize for the lack of detailed information in the Materials and methods section. These mutants were purchased from AraShare (Fuzhou, China) and verified via PCR and RT‒qPCR. We added the missing information.

(5) Is the DAS - Days After Sowing - represented as a graph or table? This will add data to the plant growth section to clearly state the difference between the mutants and the wild-type.

In this study, the age of the Arabidopsis seedlings was calculated as days after sowing (DAS), as stated in the Materials and Methods section and figure legends.

(6) Heat stress treatment after gus staining looks absurd. Should it not follow after plant materials and growth conditions, which should ideally be after the plant transformation and cloning section? The initial step is definitely about plasmid construction. Kindly rearrange.

Thank you for your valuable suggestions. We have rearranged the logical order of the materials and methods.

(7) The expression of GFP and RFP was not clearly seen in the images. This could be because of the poor resolution of the images added.

We obtained high-quality images of S-HsfA2-GFP (Figure 3 in the revised manuscript).

(8) We live in an age where it is widely known that genes are not functioning independently but are coregulated and coregulate other proteins. The authors can address the role of these spliced variants on gene regulation and compare them with the HSFs.

We agree with your suggestion. In this study, HSP17.6B was identified as a direct gene of S-HsfA2 and HsfA2, which can partly explain the role of S-HsfA2 in heat resistance and growth balance. However, the mechanical mechanism by which S-HsfA2 regulates heat tolerance and growth balance may not be limited to HSP17.6B. On the basis of the current data, we propose that the putative S-HsfA2-DERB2A-HsfA3 module might be associated with the roles of S-HsfA2 in heat tolerance and growth balance. Please refer to the discussion section for a detailed explanation.

(9) Regulatory elements can be validated in relation to their interaction with proven HSFs.

Supplemental Figure S3 shows that His6-HsfA2 failed to bind to the HRE in vitro.

(10) The authors seem to be biased toward heat stress and have not worked enough on plant growth. Biochemical data and images on plant growth could be added to bring out the novelty of this manuscript.

Thank you for your suggestion. We added new data indicating that, compared with the wild-type control, S-HsfA2-GFP, S-HsfA4c-GFP, or S-HsfB1-GFP overexpression inhibited root length (Supplemental Figure 8).

(11) Line 251 on page 11 of the submitted manuscript says that the s-Hsfs were previously identified by Liu et al. (2013) yet in the abstract the authors claim that these s-HsFs are NEW kinds of HSF with a unique truncated DNA-binding domain (tDBD) that binds a NEW heat-regulated element (HRE).

In our previous report, several S-HSFs, including S-HsfA2, S-HsfA4, S-HsfB1, and S-HsfB2a, were identified primarily in Arabidopsis (Liu et al., 2013). In this study, we further characterized S-HsfA2, S-HsfA4, and S-HsfB1 and revealed several features of S-HSFs. Therefore, we claim that these S-HSFs are new kinds of HSFs.

(12) What are these NEW kinds of HRE? Which genes have these HRE? Was an in silico study conducted to study it or can any reports can be cited?

HREs, i.e., heat-regulated elements, are newly identified heat-responsive elements in this study. The sequences of HREs are partially related to traditional heat shock elements (HSEs). Because we did not identify the essential nucleic acids required for t-DBD binding to the HRE, we did not perform an in silico study.

(13) S-HSFs may interact with existing HSFs. Have the authors thought in this direction? It can have a role in positively regulating other sHSFs or regulating multiple expressing genes related to plant growth and other functions. This needs to be explored.

Thank you for this point. Given that the overexpression of Arabidopsis HsfA2 or HsfA3 inhibits growth under nonstress conditions, we discussed this direction from the perspective of the interaction of S-HsfA2 with HsfA2 or HsfA3 in the revised manuscript.

(14) The authors need to concentrate on the presentation and arrangement of both their materials and methods and result section and write them in a systematic manner (or following a workflow).

The materials, methods and results sections are arranged in logical order.

(15) The authors have used references in the results section which can be added to the discussion section to make it more accurate.

Thank you for your suggestions. We have moved some references to the discussion section, but the necessary references remain in the results section.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Plasmodium vivax can persist in the liver of infected individuals in the form of dormant hypnozoites, which cause malaria relapses and are resistant to most current antimalarial drugs. This highlights the need to develop new drugs active against hypnozoites that could be used for radical cure. Here, the authors capitalize on an in vitro culture system based on primary human hepatocytes infected with P. vivax sporozoites to screen libraries of repurposed molecules and compounds acting on epigenetic pathways. They identified a number of hits, including hydrazinophthalazine analogs. They propose that some of these compounds may act on epigenetic pathways potentially involved in parasite quiescence. To provide some support to this hypothesis, they document DNA methylation of parasite DNA based on 5-methylcytosine immunostaining, mass spectrometry, and bisulfite sequencing.

Strengths:

-The drug screen itself represents a huge amount of work and, given the complexity of the experimental model, is a tour de force.

-The screening was performed in two different laboratories, with a third laboratory being involved in the confirmation of some of the hits, providing strong support that the results were reproducible.

-The screening of repurposing libraries is highly relevant to accelerate the development of new radical cure strategies.

We thank the reviewer for pointing out the strengths of our report.

Weaknesses:

The manuscript is composed of two main parts, the drug screening itself and the description of DNA methylation in Plasmodium pre-erythrocytic stages. Unfortunately, these two parts are loosely connected. First, there is no evidence that the identified hits kill hypnozoites via epigenetic mechanisms. The hit compounds almost all act on schizonts in addition to hypnozoites, therefore it is unlikely that they target quiescence-specific pathways. At least one compound, colforsin, seems to selectively act on hypnozoites, but this observation still requires confirmation. Second, while the description of DNA methylation is per se interesting, its role in quiescence is not directly addressed here. Again, this is clearly not a specific feature of hypnozoites as it is also observed in P. vivax and P. cynomolgi hepatic schizonts and in P. falciparum blood stages. Therefore, the link between DNA methylation and hypnozoite formation is unclear. In addition, DNA methylation in sporozoites may not reflect epigenetic regulation occurring in the subsequent liver stages.

We agree our report lacks direct evidence that hydrazinophthalazines are interacting with parasite epigenetic mechanisms. We spent significant resources attempting several novel approaches to establish a direct connection, but technological advances are needed to enable such studies, which we mention in the introduction and discussion. We disagree that schizonticidal activity automatically excludes the possibility a hypnozonticidal hit is acting on quiescence-specific pathways because both hypnozoites and schizonts are under epigenetic control and these pathways are likely performing different functions in different stages. Also important is the use of the word ‘specific’ as this term could be used to indicate parasite versus host (a drug that clears a parasite infection with a safety margin), parasite-directed effect versus host-directed effect (a drug acting via an agonistic or antagonistic effect on parasite or host pathway(s), but leading to parasite death in either case), hypnozoite versus schizont, or P. vivax versus other Plasmodium species. We were careful to indicate the usage of ‘specific’ throughout the text. Given the almost-nonexistent hit rate when screening diverse small molecule libraries screening against P. vivax hypnozoites, and remarkable increase in hits when screening epigenetic inhibitors as described in this report, our data suggests epigenetic pathways are important to the regulation of hypnozoite dormancy in addition to regulation of other parasite stages, but those effects are outside the scope of this report.

-The mode of action of the hit compounds remains unknown. In particular, it is not clear whether the drugs act on the parasite or on the host cell. Merely counting host cell nuclei to evaluate the toxicity of the compounds is probably acceptable for the screen but may not be sufficient to rule out an effect on the host cell. A more thorough characterization of the toxicity of the selected hit compounds is required.

We agree, and mention in the results and discussion, that the effect could be mediated through host pathways. This is not unlike the 8-aminoquinolones, which are activated by host cytochromes and kill via ROS, which is a nonspecific mechanism (that is, the compound is not directly interacting with a parasite target) leading to a parasite-specific effect (the parasite cannot tolerate the ROS produced, but the host can). During screening, it is generally the case that detecting hits with direct effects on the target organism are more desirable, so hits are counterscreened for general cytotoxicity. In this report, we show an effect on the parasite in direct comparison to the effect on host primary hepatocytes in the P. vivax assay itself, and follow up on hits with general counterscreens using two mammalian cell lines using CellTiter Glo, which does not rely on nuclei counts. Some compounds did show general cytotoxic effects, but with selectivity (more potency) against P. vivax liver stages, while other hits like the hydrazinophthalazines did not show an effect against primary hepatocytes and show only weak toxicity against mammalian cells at the highest dose tested. Further studies are needed to determine if the effect is indeed host- or parasite-directed and, if hydrazinophthalazines are to be developed into marketed antimalarials, extensive safety testing would be part of the development process.

-There is no convincing explanation for the differences observed between P. vivax and P. cynomolgi. The authors question the relevance of the simian model but the discrepancy could also be due to the P. vivax in vitro platform they used.

Fully characterizing the chemo-sensitivity of P. vivax and P. cynomolgi liver stages is outside the scope of this report. Rather, we report tool compounds which could be used in future studies to further characterize these sister species. We also make the point that P. cynomolgi is the gold standard for in vivo antirelapse activity, but it is still a model species, not a target species, and so few experimental hypnozonticidal compounds have been reported that the predictive value of P. cynomolgi is not fully understood. We found that several of our hits were species-specific using our in vitro platforms, thus future studies are needed to ensure this predictive value.

-Many experiments were performed only once, not only during the screen (where most compounds were apparently tested in a single well) but also in other experiments. The quality of the data would be increased with more replication.

Due to their size, compound library screens are typically performed once, with confirmation in dose-response assays, which were repeated several times. Rhesus PK studies was performed once on three animals, which is typical. All other studies were performed at least twice and most were performed three times or more. We provide a data table showing readers the source material for all replication as well as other source data tables showing the raw data for dose-response and other assays.

-While the extended assay (12 days versus 8 days) represents an improvement of the screen, the relevance of adding inhibitors of core cytochrome activity is less clear, as under these conditions the culture system deviates from physiological conditions.

We agree that cytochrome inhibitors render the platform less physiologically relevant, but the goal of screening is to detect hits which could be improved upon using medicinal chemistry, including metabolic stability. Metabolic stability is better assessed using standard assays such as liver microsomes, thus our goal was to characterize the effects of test compounds on the parasite without the confounding effect of hepatic metabolism.

Reviewer #2 (Public Review):

Summary:

In this manuscript, inhibitors of the P. vivax liver stages are identified from the Repurposing, Focused Rescue, and Accelerated Medchem (ReFRAME) library as well as a 773-member collection of epigenetic inhibitors. This study led to the discovery that epigenetics pathway inhibitors are selectively active against P. vivax and P. cynomolgi hypnozoites. Several inhibitors of histone post-translational modifications were found among the hits and genomic DNA methylation mapping revealed the modification on most genes. Experiments were completed to show that the level of methylation upstream of the gene (promoter or first exon) may impact gene expression. With the limited number of small molecules that act against hypnozoites, this work is critically important for future drug leads. Additionally, the authors gleaned biological insights from their molecules to advance the current understanding of essential molecular processes during this elusive parasite stage.

Strengths:

-This is a tremendously impactful study that assesses molecules for the ability to inhibit Plasmodium hypnozoites. The comparison of various species is especially relevant for probing biological processes and advancing drug leads.

-The SI is wonderfully organized and includes relevant data/details. These results will inspire numerous studies beyond the current work.

We thank the reviewer for pointing out the strengths of our report.

Reviewer #3 (Public Review):

Although this work represents a massive screening effort to find new drugs targeting P. vivax hypnozoites, the authors should balance their statement that they identified targetable epigenetic pathways in hypnozoites.

-They should emphasize the potential role of the host cell in the presentation of the results and the discussion, as it is known that other pathogens modify the epigenome of the host cell (i.e. toxoplasma, HIV) to prevent cell division. Also, hydrazinophtalazines target multiple pathways (notably modulation of calcium flux) and have been shown to inhibit DNA-methyl transferase 1 which is lacking in Plasmodium.

-In a drug repurposing approach, the parasite target might also be different than the human target.

-The authors state that host-cell apoptotic pathways are downregulated in P. vivax infected cells (p. 5 line 162). Maybe the HDAC inhibitors and DNA-methyltransferase inhibitors are reactivating these pathways, leading to parasite death, rather than targeting parasites directly.

We agree caution must be taken as we did not directly confirm the mechanism of our hits. Many follow up studies will be needed to do so. We do point out in the discussion that the mechanism of hits could be host-directed. We agree with the notion that some of these hits could be affecting parasitized host cell pathways, which lead to death of the parasitized cell, with the parasite being collateral damage, yet such a mechanism could lead to a safe and effective novel antimalarial.

It would make the interpretation of the results easier if the authors used EC50 in µM rather than pEC50 in tables and main text. It is easy to calculate when it is a single-digit number but more complicated with multiple digits.

We apologize for the atypical presentation of potency data. However, there is growing concern in drug discovery when Standard Deviation is applied to Potency data because Standard Deviation is a linear calculation and Potency is a log effect, making the math incompatible. We understand thousands of papers are reported every year using this mathematically incorrect method, making our presentation of these data less familiar. However, we define pEC50 in its use in the text and table legends and hope to increase its use in the broader scientific community.

Authors mention hypnozoite-specific effects but in most cases, compounds are as potent on hypnozoite and schizonts. They should rather use "liver stage specific" to refer to increased activity against hypnozoites and schizonts compared to the host cell. The same comment applies to line 351 when referring to MMV019721. Following the same idea, it is a bit far-fetched to call MMV019721 "specific" when the highest concentration tested for cytotoxicity is less than twice the EC50 obtained against hypnozoites and schizonts.

We have reviewed and revised statements in the manuscript to ensure the effect we are describing is accurate in terms of parasite versus parasite form.

Page 5 lines 187-189, the authors state "...hydrazinophtalazines were inactive when tested against P. berghei liver schizonts and P. falciparum asexual blood stages, suggesting that hypnozoite quiescence may be biologically distinct from developing schizonts". The data provided in Figure 1B show that these hydrazinophtalazines are as potent in P. vivax schizonts than in P. vivax hypnozoites, so the distinct activity seems to be Plasmodium species specific and/or host-cell specific (primary human hepatocytes rather than cell lines for P. berghei) rather than hypnozoite vs schizont specific.

We agree the effect of hydrazinophtalazine could be more species specific than stage specific, but the context of our comment has to do with current methods in antimalarial discovery and development. Given the biological uniqueness of the various Plasmodium species and stages, any hypnozonticidal hit may or may not have pan-species or pan-stage activity; our goal was to characterize this. Regardless of the mechanism, we found it interesting that the hydrazinophtalazines kill P. vivax hypnozoites, but not P. cynomolgi hypnozoites nor other species and stages used in antimalarial drug development. This result makes the point that hypnozoite-focused assays may be required to detect and develop hypnozonticidal hits, regardless of what other species or stages they may or may not act on.

Why choose to focus on cadralazine if abandoned due to side effects? Also, why test the pharmacokinetics in monkeys? As it was a marketed drug, were no data available in humans?

Cadralazine was found more potent than hydralazine and PK data was available from humans, thus dose prediction calculations showed an efficacious dose was more achievable with cadralazine than hydralazine. Side effects are often dependent on dose and regimen, which are very likely to be much different for treating malaria versus hypertension. Thus, the potential side effects of cadralazine if it was to be used as an antimalarial are simply unknown and are not disqualifying at this step. The PK study was done in Rhesus macaques so we could calculate the dose needed to achieve coverage of EC90 during a planned follow up in a Rhesus-P. cynomolgi relapse model. However, this planned in vivo efficacy study was not justified once we concurrently discovered cadralazine was inactive on P. cynomolgi in vitro.

In the counterscreen mentioned on page 6, the authors should mention that the activity of poziotinib in P. berghei and P. cynomolgi is equivalent to cell toxicity, so likely not due to parasite specificity.

Poziotinib shows activity against mammalian cell lines but not against the primary hepatocyte cultures supporting dose-response assays against P. vivax liver forms, which do not replicate. Thus, poziotinib appears selective in the liver stage assay but also may have a much more potent effect in continuously replicating cell lines.

To improve the clarity and flow of the manuscript, could the authors make a recapitulative table/figure for all the data obtained for poziotinib and hydrazinophtalazines in the different assays (8-days vs 12-days) and laboratory settings rather than separate tables in main and supplementary figures. Maybe also reorder the results section notably moving the 12-day assay before the DNA methylation part.

We apologize for the large amount of data presented but believe we are presenting it in the clearest way possible. All raw data is available if readers wish to re-analyze or re-organize our findings.

The isobologram plot shows an additive effect rather than a synergistic effect between cadralazine and 5-azacytidine, please modify the paragraph title accordingly. Please put the same axis scale for both fractional EC50 in the isobologram graph (Figure 2A).

The isobologram shows the effect approaching synergy at some combinations. The isobologram was rendered using standard methods. The raw data is available if readers wish to re-analyze it.

Concerning the immunofluorescence detection of 5mC and 5hmC, the authors should be careful with their conclusions. The Hoechst signal of the parasites is indistinguishable because of the high signal given by the hepatocyte nuclei. The signal obtained with the anti-5hmC in hepatocyte nuclei is higher than with the anti-5mC, thus if a low signal is obtained in hypnozoites and schizonts, it might be difficult to dissociate from the background. In blood stages (Figure S18), the best to obtain a good signal is to lyse the red blood cell using saponin, before fixation and HCl treatment.

We spent many hours using high resolution imaging of hundreds of parasites trying to detect clear 5hmC signal in both hypnozoites and schizonts but never saw a clearly positive signal. Indeed, the host signal can be confounding, thus we felt the most clear and unbiased way to quantify and present these data was using HCI. We appreciate the suggestion to lyse cells first for detecting in the blood stage.

To conclude that 5mC marks are the predominate DNA methylation mark in both P. falciparum and P. vivax, authors should also mention that they compare different stages of the life cycle, that might have different methylation levels.

We do mention at the start of this section our reasoning that quantifying marks in sporozoites was technically achievable, but not in a mixed culture of parasites and hepatocytes. We agree they could have different marks at these different stages.

Also, the authors conclude that "[...] 5mC is present at low level in P. vivax and P. cynomolgi sporozoites and could control liver stage development and hypnozoite quiescence". Based on the data shown here, nothing, except presence the of 5mC marks, supports that DNA methylation could be implicated in liver stage development or hypnozoite quiescence.

We clearly show sporozoite and liver stage DNA is methylated, which implicates this fundamental cell function exists in P. vivax liver stages, and that compounds with characterized activity against DNMT are active on liver stages. We acknowledge we were unable to show a direct effect and use the qualifier ‘could’ for this very reason.

How many DNA-methyltransferase inhibitors were present in the epigenetic library? Out of those, none were identified as hits, maybe the hydrazinophtalazines effect is not linked to DNMT inhibition but another target pathway of these molecules like calcium transport?

We supply the complete list of inhibitors in the epigenetic library as a supplemental file, the library contained 773 compounds. Hydrazinophtalazines were not included in the library, but several other DNA methyltransferase inhibitors were inactive. It is possible that hydrazinophtalazine activity is linked to other mechanisms but the inactivity of other DNMT inhibitors does not preclude the possibility hydrazinophtalazines are acting through DNMT.

The authors state (line 344): "These results corroborate our hypothesis that epigenetic pathways regulate hypnozoites". This conclusion should be changed to "[...] that epigenetic pathways are involved in P. vivax liver stage survival" because:

-The epigenetic inhibitors described here are as active on hypnozoite than liver schizonts.

-Again, we cannot rule out that the host cell plays a role in this effect and that the compound may not act directly on the parasite.

The same comment applies to the quote in lines 394 to 396. There is no proof in the results presented here that DNA methylation plays any role in the effect of hydrazinophtalazines in the anti-plasmodial activity obtained in the assay.

We maintain that we use words throughout the text that express uncertainty about the mechanisms involved. It is important to point out that, prior to this paper, the number of hypnozonticidal hits was incredibly low and this field is just emerging. The fundamental role of epigenetic mechanisms is regulation of gene expression. Finding several hypnozonticial hits when screening epigenetic libraries implies epigenetic pathways are important for hypnozoite survival. We intentionally do not specify exact mechanisms or if they are host or parasite pathways. Host-parasite interactions in the liver stage are incredibly difficult to resolve and are outside the scope of this report. Furthermore, this statement is not exclusive to schizonts, but since screens of diversity sets against schizonts result in a much higher hit rate, the focus of this comment is unearthing rare hypnozonticidal hits.

Author response:

The following is the authors’ response to the current reviews.

We thank the editors and the reviewers for their helpful comments. We have provided a response to reviewers’ recommendations and made some revisions on the manuscript. 

Reviewer #1 (Recommendations for the authors): 

In the newly added population factor analysis, several methodological decisions remain unclear to me:

In Figure 7, why do the authors compare the mean distance between conditions in the latent spaces of MIo and SIo? Since these latent spaces are derived separately, they exist on di@erent scales (with MIo appearing roughly four times larger than SIo), and this discrepancy is reflected in the reported mean distances (Figure 7, inset plots). Wouldn't this undermine a direct comparison?

Thank you for this helpful feedback. The reviewer is correct that the latent spaces are derived separately for MIo and SIo, thus they exist on diGerent scales as we have noted in the caption of Figure 7: “Axes for SIo are 1/4 scale of MIo.” 

To allow for a direct comparison between MIo and SIo, we corrected the analysis by comparing their normalized mean inter-trajectory distances obtained by first calculating the geometric index (GI) of the inter-trajectory distances, d, between each pair of population trajectories per region as: GI= (d₁-d₂)/ (d₁+d₂). We then performed the statistics on the GIs and found a significant diGerence between mean inter-trajectory distances in MIo vs. SIo. We performed the same analysis comparing the distance travelled between MIo and SIo trajectories by getting the normalized diGerence in distances travelled and still found a significant diGerence in both tasks. We have updated the results and figure inset to reflect these changes.

In Figure 12, unlike Figure 7 which shows three latent dimensions, only two factors are plotted. While the methods section describes a procedure for selecting the optimal number of latent factors, Figure 7 - figure supplement 3 shows that variance explained continues to increase up to about five latent dimensions across all areas. Why, then, are fewer dimensions shown?

Thank you for the opportunity to clarify the figure. The m obtained from the 3-fold crossvalidation varied for the full sample and was 20 factors for the subsample. We clarify that all statistical analyses were done using 20 latent factors. Using the full sample of neurons, the first 3 factors explained 81% of variance in feeding data compared to 71% in drinking data. When extended to 5 factors, feeding maintained its advantage with 91% variance explained versus 82% for drinking. Because feeding showed higher variance explained than drinking across 3 or 5 factors, only three factors were shown in Figure 7 for better visualization. We added this clarification to the Methods and Results.

Figure 12 shows the diGerences in the neural trajectories between the control and nerve block conditions. The control vs. nerve block comparison complicated the visualization of the results. Thus, we plotted only the two latent factors with the highest separation between population trajectories. This was clarified in the Methods and caption of Figure 12.

In Figure 12, factor 2 and 3 are plotted against each other? and factor 1 is left out?

This observation is incorrect; Factor 1 was included: Top subplots (feeding) show Factor 1 vs 3 (MIo) and Factor 1 vs 2 (SIo) while the bottom subplots (drinking) show Factor 2 vs 3 (MIo) and Factor 1 vs 2 (SIo).  We have clarified this in the Methods and caption of Figure 12.

Finally, why are factor analysis results shown only for monkey R? 

Factor analysis results were performed on both animals, but the results were shown only for monkey R to decrease the number of figures in the manuscript. Figure 7- figure supplement 1 shows the data for both monkeys. Here are the equivalent Figure 7 plots for monkey Y. 

Author response image 1.

Reviewer #2 (Recommendations for the authors): 

Overall, the manuscript has been improved. 

New analyses provide improved rigor (as just one example, organizing the feeding data into three-category split to better match the three-direction drinking data decoding analysis and also matching the neuron counts).

The updated nerve block change method (using an equal number of trials with a similar leftright angle of movement in the last 100 ms of the tongue trajectory) somewhat reduces my concern that kinematic diGerences could account for the neural changes, but on the other hand the neural analyses use 250 ms (meaning that the neural diGerences could be related to behavioral diGerences earlier in the trial). Why not subselect to trials with similar trajectories throughout the whole movement(or at least show that as an additional analysis, albeit one with lower trial counts). 

As the reviewer pointed out, selecting similar trajectories throughout the whole movement would result in lower trial counts that lead to poor statistical power. We think that the 100 ms prior to maximum tongue protrusion is a more important movement segment to control for similar kinematics between the control and nerve block conditions since this represents the subject’s intended movement endpoint. 

A lot of the Results seemed like a list of measurements without suGicient hand-holding or guide-posting to explain what the take-away for the reader should be. Just one example to make concrete this broadly-applicable feedback: "Cumulative explained variance for the first three factors was higher in feeding (MIo: 82%, SIo: 81%) than in drinking (MIo: 74%, SIo: 63%) when all neurons were used for the factor analysis (Fig. 7)": why should we care about 3 factors specifically? Does this mean that in feeding, the neural dimensionality is lower (since 3 factors explain more of it)? Does that mean feeding is a "simpler" behavior (which is counter-intuitive and does not conform to the authors' comments about the higher complexity of feeding). And from later in that paragraph: what are we do make of the diGerences in neural trajectory distances (aside from quantifying using a diGerent metric the same larger changes in firing rates that could just as well be quantified as statistics across single-neuron PETHs)?

Thank you for the feedback on the writing style. We have made some revisions to describe the takeaway for the reader. That fewer latent factors explain 80% of the variance in the feeding data means that the underlying network activity is relatively simple despite apparent complexity. When neural population trajectories are farther away from each other in state space, it means that the patterns of activity across tongue directions are more distinct and separable, thus, less likely to be confused with each other. This signifies that neural representations of 3D tongue directions are more robust. When there is better neural discrimination and more reliable information processing, it is easier for downstream brain regions to distinguish between diGerent tongue directions.  

The addition of more population-level analyses is nice as it provides a more eGicient summary of the neural measurements. However, it's a surface-level dive into these methods; ultimately the goal of ensemble "computation through dynamics" analyses is to discover simpler structure / organizational principles at the ensemble level (i.e., show things not evidence from single neurons), rather than just using them as a way to summarize data. For instance, here neural rotations are remarked upon in the Results, without referencing influential prior work describing such rotations and why neural circuits may use this computational motif to separate out conditions and shape muscle activity-generating readouts (Churchland et al. Nature 2012 and subsequent theoretical iterations including the Russo et al.). That said, the Russo et al tangling study was well-referenced and the present tangling results were eGectively contextualized with respect to that paper in terms of the interpretation. I wish more of the results were interpreted with comparable depth. 

Speaking of Russo et al: the authors note qualitative diGerences in tangling between brain areas, but do not actually quantify tangling in either. These observations would be stronger if quantified and accompanied with statistics.

Contrary to the reviewer’s critique, we did frame these results in the context of structure/organizational principles at the ensemble level. We had already cited prior work of Churchland et al., 2012; Michaels et al., 2016and Russo et al., 2018. In the Discussion, DiGerences across behaviors, we wrote: “In contrast, MIo trajectories in drinking exhibited a consistent rotational direction regardless of spout location (Fig. 7). This may reflect a predominant non-directional information such as condition-independent time-varying spiking activity during drinking (Kaufman et al., 2016; Kobak et al., 2016; Arce-McShane et al., 2023).” 

Minor suggestions: 

Some typos, e.g. 

• no opening parenthesis in "We quantified directional diGerences in population activity by calculating the Euclidean distance over m latent factors)"

• missing space in "independent neurons(Santhanam et al., 2009;..."); 

• missing closing parentheses in "followed by the Posterior Inferior (Figure 3 - figure supplement 1."

There is a one-page long paragraph in the Discussion. Please consider breaking up the text into more paragraphs each organized around one key idea to aid readability.

Thank you, we have corrected these typos.

Could it be that the Kaufman et al 2013 reference was intended to be Kaufman et al 2015 eNeuro (the condition-invariant signal paper)?

Thank you, we have corrected this reference.

At the end of the Clinical Implications subsection of the Discussion, the authors note the growing field of brain-computer interfaces with references for motor read-out or sensory write-in of hand motor/sensory cortices, respectively. Given that this study looks at orofacial cortices, an even more clinically relevant development is the more recent progress in speech BCIs (two     recent reviews: https://www.nature.com/articles/s41583-024-00819-9, https://www.annualreviews.org/content/journals/10.1146/annurev-bioeng-110122012818) many of which record from human ventral motor cortex and aspirations towards FES-like approaches for orofacial movements (e.g., https://link.springer.com/article/10.1186/s12984-023-01272-y).  

Thank you, we have included these references.

Reviewer #3 (Recommendations for the authors): 

Major Suggestions 

(1) For the factor analysis of feeding vs licking, it appears that the factors were calculated separately for the two behaviors. It could be informative to calculate the factors under both conditions and project the neural data for the two behaviors into that space. The overlap/separations of the subspace could be informative. 

We clarify that we performed a factor analysis that included both feeding and licking for MIo, as stated in the Results: “To control for factors such as diGerent neurons and kinematics that might influence the results, we performed factor analysis on stable neurons across both tasks using all trials (Fig. 7- figure supplement 2A) and using trials with similar kinematics (Fig. 7- figure supplement 2B).” We have revised the manuscript to reflect this more clearly.

(2) For the LSTM, the Factor analyses and the decoding it is unclear if the firing rates are mean subtracted and being normalized (the methods section was a little unclear). Typically, papers in the field either z-score the data or do a softmax.

The firing rates were z-scored for the LSTM and KNN. For the factor analysis, the spike counts were not z-scored, but the results were normalized. We clarified this in the Methods section.

Minor: 

Page 1: Abstract- '... how OSMCx contributes to...' 

Since there are no direct causal manipulations of OSMCx in this manuscript, this study doesn't directly study the OSMCx's contribution to movement - I would recommend rewording this sentence.

Similarly, Page 2: 'OSMCx plays an important role in coordination...' the citations in this paragraph are correlative, and do not demonstrate a causal role.

There are similar usages of 'OSMCx coordinates...' in other places e.g. Page 8. 

Thank you, we revised these sentences.

Page 7: the LSTM here has 400 units, which is a very large network and contains >12000 parameters. Networks of this size are prone to memorization, it would be wise to test the rsquare of the validation set against a shuGled dataset to see if the network is actually working as intended. 

Thank you for bringing up this important point of verifying that the network is learning meaningful patterns versus memorizing. Considering the size of our training samples, the ratio of samples to parameters is appropriate and thus the risk of memorization is low. Indeed, validation tests and cross-validation performed indicated expected network behavior and the R squared values obtained here were similar to those reported in our previous paper (Laurence-Chasen et al., 2023).

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

In their paper, Hosack and Arce-McShane investigate how the 3D movement direction of the tongue is represented in the orofacial part of the sensory-motor cortex and how this representation changes with the loss of oral sensation. They examine the firing patterns of neurons in the orofacial parts of the primary motor cortex (MIo) and somatosensory cortex (SIo) in non-human primates (NHPs) during drinking and feeding tasks. While recording neural activity, they also tracked the kinematics of tongue movement using biplanar videoradiography of markers implanted in the tongue. Their findings indicate that most units in both MIo and SIo are directionally tuned during the drinking task. However, during the feeding task, directional turning was more frequent in MIo units and less prominent in SIo units. Additionally, in some recording sessions, they blocked sensory feedback using bilateral nerve block injections, which resulted in fewer directionally tuned units and changes in the overall distribution of the preferred direction of the units.

Strengths:

The most significant strength of this paper lies in its unique combination of experimental tools. The author utilized a video-radiography method to capture 3D kinematics of the tongue movement during two behavioral tasks while simultaneously recording activity from two brain areas. Moreover, they employed a nerve-blocking procedure to halt sensory feedback. This specific dataset and experimental setup hold great potential for future research on the understudied orofacial segment of the sensory-motor area.

Weaknesses:

Aside from the last part of the result section, the majority of the analyses in this paper are focused on single units. I understand the need to characterize the number of single units that directly code for external variables like movement direction, especially for less-studied areas like the orofacial part of the sensory-motor cortex. However, as a field, our decadelong experience in the arm region of sensory-motor cortices suggests that many of the idiosyncratic behaviors of single units can be better understood when the neural activity is studied at the level of the state space of the population. By doing so, for the arm region, we were able to explain why units have "mixed selectivity" for external variables, why the tuning of units changes in the planning and execution phase of the movement, why activity in the planning phase does not lead to undesired muscle activity, etc. See (Gallego et al. 2017; Vyas et al. 2020; Churchland and Shenoy 2024) for a review. Therefore, I believe investigating the dynamics of the population activity in orofacial regions can similarly help the reader go beyond the peculiarities of single units and in a broader view, inform us if the same principles found in the arm region can be generalized to other segments of sensorymotor cortex.

We thank and agree with the reviewer on the value of information gained from studying population activity. We also appreciate that population analyses have led to the understanding that individual neurons have “mixed selectivity”. We have shown previously that OSMCx neurons exhibit mixed selectivity in their population activity and clear separation between latent factors associated with gape and bite force levels (Arce-McShane FI, Sessle BJ, Ram Y, Ross CF, Hatsopoulos NG (2023) Multiple regions of primate orofacial sensorimotor cortex encode bite force and gape. Front Systems Neurosci. doi: 10.3389/fnsys.2023.1213279. PMID: 37808467 PMCID: 10556252), and chew-side and food types (Li Z & Arce-McShane FI (2023). Cortical representation of mastication in the primate orofacial sensorimotor cortex. Program No. NANO06.05. 2023 Neuroscience Meeting Planner. Washington, D.C.: Society for Neuroscience, 2023. Online.). 

The primary goal of this paper was to characterize single units in the orofacial region and to do a follow-up paper on population activity. In the revised manuscript, we have now incorporated the results of population-level analyses. The combined results of the single unit and population analyses provide a deeper understanding of the cortical representation of 3D direction of tongue movements during natural feeding and drinking behaviors. 

Further, for the nerve-blocking experiments, the authors demonstrate that the lack of sensory feedback severely alters how the movement is executed at the level of behavior and neural activity. However, I had a hard time interpreting these results since any change in neural activity after blocking the orofacial nerves could be due to either the lack of the sensory signal or, as the authors suggest, due to the NHPs executing a different movement to compensate for the lack of sensory information or the combination of both of these factors. Hence, it would be helpful to know if the authors have any hint in the data that can tease apart these factors. For example, analyzing a subset of nerve-blocked trials that have similar kinematics to the control.

Thank you for bringing this important point. We agree with the reviewer that any change in the neural activity may be attributed to lack of sensory signal or to compensatory changes or a combination of these factors. To tease apart these factors, we sampled an equal number of trials with similar kinematics for both control and nerve block feeding sessions. We added clarifying description of this approach in the Results section of the revised manuscript: “To confirm this e ect was not merely due to altered kinematics, we conducted parallel analyses using carefully subsampled trials with matched kinematic profiles from both control and nerve-blocked conditions.”

Furthermore, we ran additional analysis for the drinking datasets by subsampling a similar distribution of drinking movements from each condition. We compared the neural data from an equal number of trials with a similar left-right angle of movement in the last 100 ms of the tongue trajectory, nearest the spout. We compared the directional tuning across an equal number of trials with a similar left-right angle of movement in the last 100 ms of the tongue trajectory, nearest the spout. These analyses that control for similar kinematics showed that there was still a decrease in the proportion of directionally modulated neurons with nerve block compared to the control. This confirms that the results may be attributed to the lack of tactile information. These are now integrated in the revised paper under Methods section: Directional tuning of single neurons, as well as Results section: E ects of nerve block: Decreased directional tuning of MIo and SIo neurons and Figure 10 – figure supplement 1.

Reviewer #2 (Public review):

Summary:

This manuscript by Hosack and Arce-McShane examines the directional tuning of neurons in macaque primary motor (MIo) and somatosensory (SIo) cortex. The neural basis of tongue control is far less studied than, for example, forelimb movements, partly because the tongue's kinematics and kinetics are difficult to measure. A major technical advantage of this study is using biplanar video-radiography, processed with modern motion tracking analysis software, to track the movement of the tongue inside the oral cavity. Compared to prior work, the behaviors are more naturalistic behaviors (feeding and licking water from one of three spouts), although the animals were still head-fixed.

The study's main findings are that:

• A majority of neurons in MIo and a (somewhat smaller) percentage of SIo modulated their firing rates during tongue movements, with different modulations depending on the direction of movement (i.e., exhibited directional tuning). Examining the statistics of tuning across neurons, there was anisotropy (e.g., more neurons preferring anterior movement) and a lateral bias in which tongue direction neurons preferred that was consistent with the innervation patterns of tongue control muscles (although with some inconsistency between monkeys).

• Consistent with this encoding, tongue position could be decoded with moderate accuracy even from small ensembles of ~28 neurons.

• There were di erences observed in the proportion and extent of directional tuning between the feeding and licking behaviors, with stronger tuning overall during licking. This potentially suggests behavioral context-dependent encoding.

• The authors then went one step further and used a bilateral nerve block to the sensory inputs (trigeminal nerve) from the tongue. This impaired the precision of tongue movements and resulted in an apparent reduction and change in neural tuning in Mio and SIo.

Strengths:

The data are difficult to obtain and appear to have been rigorously measured, and provide a valuable contribution to this under-explored subfield of sensorimotor neuroscience. The analyses adopt well-established methods, especially from the arm motor control literature, and represent a natural starting point for characterizing tongue 3D direction tuning.

Weaknesses:

There are alternative explanations for some of the interpretations, but those interpretations are described in a way that clearly distinguishes results from interpretations, and readers can make their own assessments. Some of these limitations are described in more detail below.

One weakness of the current study is that there is substantial variability in results between monkeys, and that only one session of data per monkey/condition is analyzed (8 sessions total). This raises the concern that the results could be idiosyncratic. The Methods mention that other datasets were collected, but not analyzed because the imaging pre-processing is very labor-intensive. While I recognize that time is precious, I do think in this case the manuscript would be substantially strengthened by showing that the results are similar on other sessions.

We acknowledge the reviewer’s concern about inter-subject variability. Animal feeding and drinking behaviors are quite stable across sessions, thus, we do not think that additional sessions will address the concern that the results could be idiosyncratic. Each of the eight datasets analyzed here have su icient neural and kinematic data to capture neural and behavioral patterns.  Nevertheless, we performed some of the analyses on a second feeding dataset from Monkey R. The results from analyses on a subset of this data were consistent across datasets; for example, (1) similar proportions of directionally tuned neurons, (2) similar distances between population trajectories (t-test p > 0.9), and (3) a consistently smaller distance between Anterior-Posterior pairs than others in MIo (t-test p < 0.05) but not SIo (p > 0.1). 

This study focuses on describing directional tuning using the preferred direction (PD) / cosine tuning model popularized by Georgopoulous and colleagues for understanding neural control of arm reaching in the 1980s. This is a reasonable starting point and a decent first-order description of neural tuning. However, the arm motor control field has moved far past that viewpoint, and in some ways, an over-fixation on static representational encoding models and PDs held that field back for many years. The manuscript benefits from drawing the readers' attention (perhaps in their Discussion) that PDs are a very simple starting point for characterizing how cortical activity relates to kinematics, but that there is likely much richer population-level dynamical structure and that a more mechanistic, control-focused analytical framework may be fruitful. A good review of this evolution in the arm field can be found in Vyas S, Golub MD, Sussillo D, Shenoy K. 2020. Computation Through Neural Population Dynamics. Annual Review of Neuroscience. 43(1):249-75

Thank you for highlighting this important point. Research on orofacial movements hasn't progressed at the same pace as limb movement studies. Our manuscript focused specifically on characterizing the 3D directional tuning properties of individual neurons in the orofacial area—an analysis that has not been conducted previously for orofacial sensorimotor control. While we initially prioritized this individual neuron analysis, we recognize the value of broader population-level insights.

Based on your helpful feedback, we have incorporated additional population analyses to provide a more comprehensive picture of orofacial sensorimotor control and expanded our discussion section. We appreciate your expertise in pushing our work to be more thorough and aligned with current neuroscience approaches.

Can the authors explain (or at least speculate) why there was such a large difference in behavioral e ect due to nerve block between the two monkeys (Figure 7)?

We acknowledge this as a variable inherent to this type of experimentation. Previous studies have found large kinematic variation in the effect of oral nerve block as well as in the following compensatory strategies between subjects. Each animal’s biology and response to perturbation vary naturally. Indeed, our subjects exhibited different feeding behavior even in the absence of nerve block perturbation (see Figure 2 in Laurence-Chasen et al., 2022). This is why each individual serves as its own control.

Do the analyses showing a decrease in tuning after nerve block take into account the changes (and sometimes reduction in variability) of the kinematics between these conditions? In other words, if you subsampled trials to have similar distributions of kinematics between Control and Block conditions, does the effect hold true? The extreme scenario to illustrate my concern is that if Block conditions resulted in all identical movements (which of course they don't), the tuning analysis would find no tuned neurons. The lack of change in decoding accuracy is another yellow flag that there may be a methodological explanation for the decreased tuning result.

Thank you for bringing up this point. We accounted for the changes in the variability of the kinematics between the control and nerve block conditions in the feeding dataset where we sampled an equal number of trials with similar kinematics for both control and nerve block. However, we did not control for similar kinematics in the drinking task. In the revised manuscript, we have clarified this and performed similar analysis for the drinking task. We sampled a similar distribution of drinking movements from each condition. We compared the neural data from an equal number of trials with a similar left-right angle of movement in the last 100 ms of the tongue trajectory, nearest the spout. There was a decrease in the percentage of neurons that were directionally modulated (between 30 and 80%) with nerve block compared to the control. These results have been included in the revised paper under Methods section: Directional tuning of single neurons, as well as Results section: E ects of nerve block: Decreased directionality of MIo and SIo neurons.

While the results from decoding using KNN did not show significant differences between decoding accuracies in control vs. nerve block conditions, the results from the additional factor analysis and decoding using LSTM were consistent with the decrease in directional tuning at the level of individual neurons.  

The manuscript states that "Our results suggest that the somatosensory cortex may be less involved than the motor areas during feeding, possibly because it is a more ingrained and stereotyped behavior as opposed to tongue protrusion or drinking tasks". Could an alternative explanation be more statistical/technical in nature: that during feeding, there will be more variability in exactly what somato sensation afferent signals are being received from trial to trial (because slight differences in kinematics can have large differences in exactly where the tongue is and the where/when/how of what parts of it are touching other parts of the oral cavity)? This variability could "smear out" the apparent tuning using these types of trial-averaged analyses. Given how important proprioception and somatosensation are for not biting the tongue or choking, the speculation that somatosensory cortical activity is suppressed during feedback is very counter-intuitive to this reviewer.

Thank you for bringing up this point. We have now incorporated this in our revised Discussion (see Comparison between MIo and SIo). We agree with the reviewer that trialby-trial variability in the a erent signals may account for the lower directional signal in SIo during feeding than in drinking. Indeed, SIo’s mean-matched Fano factor in feeding was significantly higher than those in drinking (Author response image 1). Moreover, the results of the additional population and decoding analyses also support this.  

Author response image 1.

Comparison of mean-matched Fano Factor between Sio neurons during feeding and drinking control tasks across both subjects (Wilcoxon rank sum test, p < 0.001).

Reviewer #3 (Public review):

Summary:

In this study, the authors aim to uncover how 3D tongue direction is represented in the Motor (M1o) and Somatosensory (S1o) cortex. In non-human primates implanted with chronic electrode arrays, they use X-ray-based imaging to track the kinematics of the tongue and jaw as the animal is either chewing food or licking from a spout. They then correlate the tongue kinematics with the recorded neural activity. Using linear regressions, they characterize the tuning properties and distributions of the recorded population during feeding and licking. Then, they recharacterize the tuning properties after bilateral lidocaine injections in the two sensory branches of the trigeminal nerve. They report that their nerve block causes a reorganization of the tuning properties. Overall, this paper concludes that M1o and S1o both contain representations of the tongue direction, but their numbers, their tuning properties, and susceptibility to perturbed sensory input are different.

Strengths:

The major strengths of this paper are in the state-of-the-art experimental methods employed to collect the electrophysiological and kinematic data.

Weaknesses:

However, this paper has a number of weaknesses in the analysis of this data.

It is unclear how reliable the neural responses are to the stimuli. The trial-by-trial variability of the neural firing rates is not reported. Thus, it is unclear if the methods used for establishing that a neuron is modulated and tuned to a direction are susceptible to spurious correlations. The authors do not use shuffling or bootstrapping tests to determine the robustness of their fits or determining the 'preferred direction' of the neurons. This weakness colors the rest of the paper.

Thank you for raising these points. We have performed the following additional analyses: (1) We have added analyses to ensure that the results could not be explained by neural variability. To show the trial-by-trial variability of the neural firing rates, we have calculated the Fano factor (mean overall = 1.34747; control = 1.46471; nerve block = 1.23023). The distribution was similar across directions, suggesting that responses of MIo and SIo neurons to varying 3D directions were reliable. (2) We have used a bootstrap procedure to ensure that directional tuning cannot be explained by mere chance. (3) To test the robustness of our PDs we also performed a bootstrap test, which yielded the same results for >90% of neurons, and a multiple linear regression test for fit to a cosine-tuning function. In the revised manuscript, the Methods and Results sections have been updated to include these analyses.  

Author response image 2.

Comparison of Fano Factor across directions for MIo and SIo Feeding Control (Kruskal-Wallis, p > 0.7).

The authors compare the tuning properties during feeding to those during licking but only focus on the tongue-tip. However, the two behaviors are different also in their engagement of the jaw muscles. Thus many of the differences observed between the two 'tasks' might have very little to do with an alternation in the properties of the neural code - and more to do with the differences in the movements involved. 

Using the tongue tip for the kinematic analysis of tongue directional movements was a deliberate choice as the anterior region of the tongue is highly mobile and sensitive due to a higher density of mechanoreceptors. The tongue tip is the first region that touches the spout in the drinking task and moves the food into the oral cavity for chewing and subsequent swallowing. 

We agree with the reviewer that the jaw muscles are engaged differently in feeding vs. drinking (see Fig. 2). For example, a wider variety of jaw movements along the three axes are observed in feeding compared to the smaller amplitude and mostly vertical jaw movements in drinking. Also, the tongue movements are very different between the two behaviors. In feeding, the tongue moves in varied directions to position the food between left-right tooth rows during chewing, whereas in the drinking task, the tongue moves to discrete locations to receive the juice reward. Moreover, the tongue-jaw coordination differs between tasks; maximum tongue protrusion coincides with maximum gape in drinking but with minimum gape in the feeding behavior. Thus, the different tongue and jaw movements required in each behavior may account for some of the differences observed in the directional tuning properties of individual neurons and population activity. These points have been included in the revised Discussion.

Author response image 3.

Tongue tip position (mm) and jaw pitch(degree) during feeding (left) and drinking (right) behaviors. Most protruded tongue position coincides with minimum gape (jaw pitch at 0°) during  feeding but with maximum gape during drinking.

Many of the neurons are likely correlated with both Jaw movements and tongue movements - this complicates the interpretations and raises the possibility that the differences in tuning properties across tasks are trivial.

We thank the reviewer for raising this important point. In fact, we verified in a previous study whether the correlation between the tongue and jaw kinematics might explain di erences in the encoding of tongue kinematics and shape in MIo (see Supplementary Fig. 4 in Laurence-Chasen et al., 2023): “Through iterative sampling of sub-regions of the test trials, we found that correlation of tongue kinematic variables with mandibular motion does not account for decoding accuracy. Even at times where tongue motion was completely un-correlated with the jaw, decoding accuracy could be quite high.” 

The results obtained from population analyses showing distinct properties of population trajectories in feeding vs. drinking behaviors provide strong support to the interpretation that directional information varies between these behaviors.

The population analyses for decoding are rudimentary and provide very coarse estimates (left, center, or right), it is also unclear what the major takeaways from the population decoding analyses are. The reduced classification accuracy could very well be a consequence of linear models being unable to account for the complexity of feeding movements, while the licking movements are 'simpler' and thus are better accounted for.

We thank the reviewer for raising this point. The population decoding analyses provide additional insight on the directional information in population activity,  as well as a point of comparison with the results of numerous decoding studies on the arm region of the sensorimotor cortex. In the revised version, we have included the results from decoding tongue direction using a long short-term memory (LSTM) network for sequence-tosequence decoding. These results di ered from the KNN results, indicating that a linear model such as KNN was better for drinking and that a non-linear and continuous decoder was better suited for feeding.  These results have been included in the revised manuscript.

The nature of the nerve block and what sensory pathways are being affected is unclear - the trigeminal nerve contains many different sensory afferents - is there a characterization of how e ectively the nerve impulses are being blocked? Have the authors confirmed or characterized the strength of their inactivation or block, I was unable to find any electrophysiological evidence characterizing the perturbation.

The strength of the nerve block is characterized by a decrease in the baseline firing rate of SIo neurons, as shown in Supplementary Figure 6 of “Loss of oral sensation impairs feeding performance and consistency of tongue–jaw coordination” (Laurence-Chasen et al., 2022)..

Overall, while this paper provides a descriptive account of the observed neural correlations and their alteration by perturbation, a synthesis of the observed changes and some insight into neural processing of tongue kinematics would strengthen this paper.

We thank the reviewer for this suggestion. We have revised the Discussion to provide a synthesis of the results and insights into the neural processing of tongue kinematics.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

(1) The procedure for anesthesia explained in the method section was not clear to me. The following information was missing: what drug/dose was used? How long the animal was under anesthesia? How long after the recovery the experiments were done?

The animals were fully sedated with ketamine (100 mg/ml, 10 mg/kg) for less than 30 minutes, and all of the data was collected within 90 minutes after the nerve block was administered.

(2) In Figure 10, panels A and B are very close together, it was not at first clear whether the text "Monkey R, Monkey Y" belongs to panel A or B.

We have separated the two panels further in the revised figure.

(3) I found Figure 11 very busy and hard to interpret. Separating monkeys, fitting the line for each condition, or using a bar plot can help with the readability of the figure.

Thank you for the suggestion. We agree with you and have reworked this figure. To simplify it we have shown the mean accuracy across iterations.

(4) I found the laterality discussions like "This signifies that there are more neurons in the left hemisphere contributes toward one direction of tongue movement, suggesting that there is some laterality in the PDs of OSMCx neurons that varies between individuals" bit of an over-interpretation of data, given the low n value and the dissimilarity in how strongly the nerve blocking altered monkies behavior.

Thank you for sharing this viewpoint. We do think that laterality is a good point of comparison with studies on M1 neurons in the arm/hand region. In our study, we found that the peak of the PD distribution coincides with leftward tongue movements in feeding. The distribution of PDs provides insight into how tongue muscles are coordinated during movement. Intrinsic and extrinsic tongue muscles are involved in shaping the tongue (e.g., elongation, broadening) and positioning the tongue (e.g., protrusion/retraction, elevation/depression), respectively. These muscles receive bilateral motor innervation except for genioglossus. Straight tongue protrusion requires the balanced action of the right and left genioglossi while the lateral protrusion involves primarily the contralateral genioglossus. Given this unilateral innervation pattern, we hypothesized that left MIo/SIo neurons would preferentially respond to leftward tongue movements, corresponding to right genioglossus activation. 

Reviewer #2 (Recommendations for the authors):

Are the observation of tuning peaks being most frequently observed toward the anterior and superior directions consistent with the statistics of the movements the tongue typically makes? This could be analogous to anisotropies previously reported in the arm literature, e.g., Lillicrap TP, Scott SH. 2013. Preference Distributions of Primary Motor Cortex Neurons Reflect Control Solutions Optimized for Limb Biomechanics. Neuron. 77(1):168-79

Thank you for bringing our attention to analogous findings by Lillicrap & Scott, 2013. Indeed, we do observe the highest number of movements in the Anterior Superior directions, followed by the Posterior Inferior. This does align with the distribution of tuning peaks that we observed. Author response image 4 shows the proportions of observed movements in each group of directions across all feeding datasets. We have incorporated this data in the Results section: Neuronal modulation patterns di er between MIo and SIo, as well as added this point in the Discussion.

Author response image 4.

Proportion of feeding trials in each group of directions. Error bars represent ±1 standard deviation across datasets (n = 4).

"The Euclidean distance was used to identify nearest neighbors, and the number of nearest neighbors used was K = 7. This K value was determined after testing different Ks which yielded comparable results." In general, it's a decoding best practice to tune hyperparameters (like K) on fully held-out data from the data used for evaluation. Otherwise, this tends to slightly inflate performance because one picks the hyperparameter that happened to give the best result. It sounds like that held-out validation set wasn't used here. I don't think that's going to change the results much at all (especially given the "comparable results" comment), but providing this suggestion for the future. If the authors replicate results on other datasets, I suggest they keep K = 7 to lock in the method.

K = 7 was chosen based on the size of our smallest training dataset (n = 55). The purpose of testing different K values was not to select which value gave the best result, but to demonstrate that similar K values did not affect the results significantly. We tested the di erent K values on a subset of the feeding data, but that data was not fully held-out from the training set. We will keep your suggestion in mind for future analysis.

The smoothing applied to Figure 2 PSTHs appears perhaps excessive (i.e., it may be obscuring interesting finer-grained details of these fast movements). Can the authors reduce the 50 ms Gaussian smoothing (I assume this is the s.d.?) ~25 ms is often used in studying arm kinematics. It also looks like the movement-related modulation may not be finished in these 200 ms / 500 ms windows. I suggest extending the shown time window. It would also be helpful to show some trial-averaged behavior (e.g. speed or % displacement from start) under or behind the PSTHs, to give a sense of what phase of the movement the neural activity corresponds to.

Thank you for the suggestion. We have taken your suggestions into consideration and modified Figure 2 accordingly. We decreased the Gaussian kernel to 25 ms and extended the time window shown. The trial-averaged anterior/posterior displacement was also added to the drinking PSTHs.

Reviewer #3 (Recommendations for the authors):

The major consideration here is that the data reported for feeding appears to be very similar to that reported in a previous study:

"Robust cortical encoding of 3D tongue shape during feeding in macaques"

Are the neurons reported here the same as the ones used in this previous paper? It is deeply concerning that this is not reported anywhere in the methods section.

These are the same neurons as in our previous paper, though here we include several additional datasets of the nerve block and drinking sessions. We have now included this in the methods section.

Second, I strongly recommend that the authors consider a thorough rewrite of this manuscript and improve the presentation of the figures. As written, it was not easy to follow the paper, the logic of the experiments, or the specific data being presented in the figures.

Thank you for this suggestion. We have done an extensive rewrite of the manuscript and revision of the figures.

A few recommendations:

(1) Please structure your results sections and use descriptive topic sentences to focus the reader. In the current version, it is unclear what the major point being conveyed for each analysis is.

Thank you for this suggestion. We have added topic sentences to the begin each section of the results.

(2) Please show raster plots for at least a few example neurons so that the readers have a sense of what the neural responses look like across trials. Is all of Figure 2 one example neuron or are they different neurons? Error bars for PETH would be useful to show the reliability and robustness of the tuning.

Figure 2 shows different neurons, one from MIo and one from SIo for each task. There is shading showing ±1 standard error around the line for each direction, however this was a bit difficult to see. In addition to the other changes we have made to these figures, we made the lines smaller and darkened the error bar shading to accentuate this. We also added raster plots corresponding to the same neurons represented in Figure 2 as a supplement.

(3) Since there are only two data points, I am not sure I understand why the authors have bar graphs and error bars for graphs such as Figure 3B, Figure 5B, etc. How can one have an error bar and means with just 2 data points?

Those bars represent the standard error of the proportion. We have changed the y-axis label on these figures to make this clearer.

(4) Results in Figure 6 could be due to differential placement of the electrodes across the animals. How is this being accounted for?

Yes, this is a possibility which we have mentioned in the discussion. Even with careful placement there is no guarantee to capture a set of neurons with the exact same function in two subjects, as every individual is different. Rather we focus on analyses of data within the same animal. The purpose of Figure 6 is to show the di erence between MIo and SIo, and between the two tasks, within the same subject. The more salient result from calculating the preferred direction is that there is a change in the distribution between control and nerve block within the same exact population. Discussions relating to the comparison between individuals are speculative and cannot be confirmed without the inclusion of many more subjects.

(5) For Figure 7, I would recommend showing the results of the Sham injection in the same figure instead of a supplement.

Thank you for the suggestion, we have added these results to the figure.

(6) I think the e ects of the sensory block on the tongue kinematics are underexplored in Figure 7 and Figure 8. The authors could explore the deficits in tongue shape, and the temporal components of the trajectory.

Some of these effects on feeding have been explored in a previous paper, LaurenceChasen et al., 2022. We performed some additional analyses on changes to kinematics during drinking, including the number of licks per 10 second trial and the length of individual licks. The results of these are included below. We also calculated the difference in the speed of tongue movement during drinking, which generally decreased and exhibited an increase in variance with nerve block (f-test, p < 0.001). However, we have not included these figures in the main paper as they do not inform us about directionality.

Author response image 5.

Left halves of hemi-violins (black) are control and right halves (red) are nerve block for an individual. Horizontal black lines represent the mean and horizontal red lines the median. Results of two-tailed t-test and f-test are indicated by asterisks and crosses, respectively: *,† p < 0.05; **,†† p < 0.01; ***,††† p < 0.001.

(9) In Figures 9 and 10. Are the same neurons being recorded before and after the nerve block? It is unclear if the overall "population" properties are different, or if the properties of individual neurons are changing due to the nerve block.

Yes, the same neurons are being recorded before and after nerve block. Specifically, Figure 9B shows that the properties of many individual neurons do change due to the nerve block. Di erences in the overall population response may be attributed to some of the units having reduced/no activity during the nerve block session.

Additionally, I recommend that the authors improve their introduction and provide more context to their discussion. Please elaborate on what you think are the main conceptual advances in your study, and place them in the context of the existing literature. By my count, there are 26 citations in this paper, 4 of which are self-citations - clearly, this can be improved upon.

Thank you for this suggestion. We have done an extensive rewrite of the Introduction and Discussion. We discussed the main conceptual advances in our study and place them in the context of the existing literature.

Author response:

We sincerely thank the reviewers for the time and care they have invested in evaluating our manuscript. We greatly appreciate their thoughtful feedback, which highlights both the strengths and the areas where the work can be improved. We recognize the importance of the concerns raised, particularly regarding the TMS analyses and interpretation, as well as aspects of the manuscript structure and clarity. The authors are committed to transparency and a rigorous scientific process, and we will therefore carefully consider all reviewer comments. In the coming months, we will revise the manuscript to incorporate additional analyses, provide clearer methodological detail, and refine the interpretation of the stimulation results.

Author response:

eLife Assessment:

This paper performs a valuable critical reassessment of anatomical and functional data, proposing a reclassification of the mouse visual cortex in which almost all the higher visual areas are consolidated into a single area V2. However, the evidence supporting this unification is incomplete, as the key experimental observations that the model attempts to reproduce do not accurately reflect the literature . This study will likely be of interest to neuroscientists focused on the mouse visual cortex and the evolution of cortical organization.

We do not agree or understand which 'key experimental observations' that the model attempts to reproduce do not accurately reflect the literature. The model reproduces a complete map of the visual field, with overlap in certain regions. When reversals are used to delineate areas, as is the current custom, multiple higher order areas are generated, and each area has a biased and overlapping visual field coverage. These are the simple outputs of the model, and they are consistent with the published literature, including recent publications such as Garrett et al. 2014 and Zhuang et al. 2017, a paper published in this journal. The area boundaries produced by the model are not identical to area boundaries in the literature, because the model is a simplification.

Reviewer #1 (Public review):

Summary:

In this manuscript, the authors argue that defining higher visual areas (HVAs) based on reversals of retinotopic tuning has led to an over-parcellation of secondary visual cortices. Using retinotopic models, they propose that the HVAs are more parsimoniously mapped as a single area V2, which encircles V1 and exhibits complex retinotopy. They reanalyze functional data to argue that functional differences between HVAs can be explained by retinotopic coverage. Finally, they compare the classification of mouse visual cortex to that of other species to argue that our current classification is inconsistent with those used in other model species.

Strengths:

This manuscript is bold and thought-provoking, and is a must-read for mouse visual neuroscientists. The authors take a strong stance on combining all HVAs, with the possible exception of area POR, into a single V2 region. Although I suspect many in the field will find that their proposal goes too far, many will agree that we need to closely examine the assumptions of previous classifications to derive a more accurate areal map. The authors' supporting analyses are clear and bolster their argument. Finally, they make a compelling argument for why the classification is not just semantic, but has ramifications for the design of experiments and analysis of data.

Weaknesses:

Although I enjoyed the polemic nature of the manuscript, there are a few issues that weaken their argument.

(1) Although the authors make a compelling argument that retinotopic reversals are insufficient to define distinct regions, they are less clear about what would constitute convincing evidence for distinct visual regions. They mention that a distinct area V3 has been (correctly) defined in ferrets based on "cytoarchitecture, anatomy, and functional properties", but elsewhere argue that none of these factors are sufficient to parcellate any of the HVAs in mouse cortex, despite some striking differences between HVAs in each of these factors. It would be helpful to clearly define a set of criteria that could be used for classifying distinct regions.

We agree the revised manuscript would benefit from a clear discussion of updated rules of area delineation in the mouse. In brief, we argue that retinotopy alone should not be used to delineate area boundaries in mice, or any other species. Although there is some evidence for functional property, architecture, and connectivity changes across mouse HVAs, area boundaries continue to be defined primarily, and sometimes solely (Garrett et al., 2014; Juavinett et al., 2018; Zhuang et al., 2017), based on retinotopy. We acknowledge that earlier work (Wang and Burkhalter, 2007; Wang et al., 2011) did consider cytoarchitecture and connectivity alongside retinotopy, but more recent work has shifted to a focus on retinotopy as indicated by the currently accepted criterion for area delineation.  

As reviewer #2 points out, the present criteria for mouse visual area delineation can be found in the Methods section of: [Garrett, M.E., Nauhaus, I., Marshel, J.H., and Callaway, E.M. (2014)].

Criterion 1: Each area must contain the same visual field sign at all locations within the area.

Criterion 2: Each visual area cannot have a redundant representation of visual space.

Criterion 3: Adjacent areas of the same visual field sign must have a redundant representation.

Criterion 4: An area's location must be consistently identifiable across experiments.

As discussed in the manuscript, recent evidence in higher order visual cortex of tree shrews and rats led us to question the universality of these criteria across species. Specifically, tree shrew V2, macaque V2, and marmoset DM, exhibit reversals in visual field-sign in what are defined as single visual areas. This suggests that criterion 1 should be updated. It also suggests that Criterion 2 and 3 should be updated since visual field sign reversals often co-occur with retinotopic redundancies, since reversing course in the direction of progression along the visual field can easily lead to coverage of visual field regions already traveled.  

More broadly, we argue that topography is just one of several criteria that should be considered in area delineation. We understand that few visual areas in any species meet all criteria, but we emphasize that topography cannot consistently be the sole satisfied criterion – as it currently appears to be for many mouse HVAs. Inspired by a recent perspective on cortical area delineation (Petersen et al., 2024), we suggest the following rules, that will be worked into the revised version of the manuscript. Topography is a criterion, but it comes after considerations of function, architectonics and connectivity.

(1) Function—Cortical areas differ from neighboring areas in their functional properties  

(2) Architectonics—Cortical areas often exhibit distinctions from neighboring areas in multiple cyto- and myeloarchitectonic markers

(3) Connectivity—Cortical areas are characterized by a specific set of connectional inputs and outputs from and to other areas

(4) Topography—Cortical areas often exhibit a distinct topography that balances maximal coverage of the sensory field with minimal redundancy of coverage within an area.

As we discuss in the manuscript, although there are functional, architectonic, and connectivity differences across mouse HVAs, they typically vary smoothly across multiple areas – such that neighboring areas share the same properties and there are no sharp borders. For instance, sharp borders in cytoarchitecture are generally lacking in the mouse HVAs. A notable exceptions to this is the clear and sharp change in m2AChR expression that occurs between LM and AL (Wang et al., 2011). 

(2) On a related note, although the authors carry out impressive analyses to show that differences in functional properties between HVAs could be explained by retinotopy, they glossed over some contrary evidence that there are functional differences independent of retinotopy. For example, axon projections to different HVAs originating from a single V1 injection - presumably including neurons with similar retinotopy - exhibit distinct functional properties (Glickfeld LL et al, Nat Neuro, 2013). As another example, interdigitated M2+/M2- patches in V1 show very different HVA connectivity and response properties, again independent of V1 location/retinotopy (Meier AM et al., bioRxiv). One consideration is that the secondary regions might be considered a single V2 with distinct functional modules based on retinotopy and connectivity (e.g., V2LM, V2PM, etc).

Thank you for the correction. We will revise the text to discuss (Glickfeld et al., 2013), as it remains some of the strongest evidence in favor of retinotopy-independent functional specialization of mouse HVAs. However, one caveat of this study is the size of the V1 injection that is the source of axons studied in the HVAs. As apparent in Figure 1B, the large injection covers nearly a quarter of V1. It is worth nothing that (Han et al., 2018) found, using single-cell reconstructions and MAPseq, that the majority of V1 neurons project to multiple nearby HVA targets. In this experiment the tracing does not suffer from the problem of spreading over V1’s retinotopic map, and suggests that, presumably retinotopically matched, locations in each area receive shared inputs from the V1 population rather than a distinct but spatially interspersed subset. In fact, the authors conclude “Interestingly, the location of the cell body within V1 was predictive of projection target for some recipient areas (Extended Data Fig. 8). Given the retinotopic organization of V1, this suggests that visual information from different parts of visual field may be preferentially distributed to  specific target areas, which is consistent with recent findings (Zhuang et al., 2017)”. Given an injection covering a large portion of the retinotopic map, and the fact that feed-forward projections from V1 to HVAs carry coarse retinotopy - it is difficult to prove that functional specializations noted in the HVA axons are retinotopyindependent. This would require measurement of receptive field location in the axonal boutons, which the authors did not perform (possibly because the SNR of calcium indicators prevented such measurements at the time).  

Another option would be to show that adjacent neurons in V1, that project to far-apart HVAs, exhibit distinct functional properties on par with differences exhibited by neurons in very different parts of V1 due to retinotopy. In other words, the functional specificity of V1 inputs to HVAs at retinotopically identical locations is of the same order as those that might be gained by retinotopic biases. To our knowledge, such a study has not been conducted, so we have decided to measure the data in collaboration with the Allen Institute. As part of the Allen Institute’s pioneering OpenScope project, we will make careful two-photon and electrophysiology measurements of functional properties, including receptive field location, SF, and TF in different parts of the V1 retinotopic map. Pairing this data with existing Allen Institute datasets on functional properties of neurons in the HVAs will allow us to rule in, or rule-out, our hypotheses regarding retinotopy as the source of functional specialization in mouse HVAs. We will update the discussion in the revised manuscript to better reflect the need for additional evidence to support or refute our proposal.

Meier AM et al., bioRxiv 2025 (Meier et al., 2025) was published after our submission, but we are thankful to the reviewers for guiding our attention to this timely paper. Given the recent findings on the influence of locomotion on rodent and primate visual cortex, it is very exciting to see clearly specialized circuits for processing self-generated visual motion in V1. However, it is difficult to rule out the role of retinotopy as the HVA areas (LM, AL, RL) participating in the M2+ network less responsive to self-generated visual motion exhibit a bias for the medial portion of the visual field and the HVA area (PM) involved in the M2- network responsive to self-generated visual motion exhibit a bias for the lateral (or peripheral) parts of the visual field. For instance, a peripheral bias in area PM has been shown using retrograde tracing as in Figure 6 of (Morimoto et al., 2021), single-cell anterograde tracing  as in Extended Data Figure 8 of (Han et al., 2018), and functional imaging studies (Zhuang et al., 2017). Recent findings in the marmoset also point to visual circuits in the peripheral, but not central, visual field being significantly modulated by selfgenerated movements (Rowley et al., 2024). 

However, a visual field bias in area PM that selectively receive M2- inputs is at odds with the clear presence of modular M2+/M2- patches across the entire map of V1 (Ji et al., 2015).  One possibility supported by existing data is that neurons in M2- patches, as well as those in M2+ patches, in the central representation of V1 make fewer or significantly weaker connections with area PM compared to areas LM, AL and RL. Evidence to the contrary would support retinotopy-independent and functionally specialized inputs from V1 to HVAs.

(3) Some of the HVAs-such as AL, AM, and LI-appear to have redundant retinotopic coverage with other HVAS, such as LM and PM. Moreover, these regions have typically been found to have higher "hierarchy scores" based on connectivity (Harris JA et al., Nature, 2019; D'Souza RD et al., Nat Comm, 2022), though unfortunately, the hierarchy levels are not completely consistent between studies. Based on existing evidence, there is a reasonable argument to be made for a hybrid classification, in which some regions (e.g., LM, P, PM, and RL) are combined into a single V2 (though see point #2 above) while other HVAs are maintained as independent visual regions, distinct from V2. I don't expect the authors to revise their viewpoint in any way, but a more nuanced discussion of alternative classifications is warranted.

We understand that such a proposal would combine a subset of areas with matched field sign (LM, P, PM, and RL) would be less extreme and received better by the community. This would create a V2 with a smooth map without reversals or significant redundant retinotopic coverage. However, the intuition we have built from our modeling studies suggest that both these areas, and the other smaller areas with negative field sign (AL, AM, LI), are a byproduct of a complex single map of the visual field that exhibits reversals as it contorts around the triangular and tear-shaped boundaries of V1. In other words, we believe the redundant coverage and field-sign changes/reversals are a byproduct of a single secondary visual field in V2 constrained by the cortical dimensions of V1. That being said, we understand that area delineations are in part based on a consensus by the community. Therefore we will continue to discuss our proposal with community members, and we will incorporate new evidence supporting or refuting our hypothesis, before we submit our revised manuscript.

Reviewer #2 (Public review):

Summary:

The study by Rowley and Sedigh-Sarvestani presents modeling data suggesting that map reversals in mouse lateral extrastriate visual cortex do not coincide with areal borders, but instead represent borders between subregions within a single area V2. The authors propose that such an organization explains the partial coverage in higher-order areas reported by Zhuang et al., (2017). The scheme revisits an organization proposed by Kaas et al., (1989), who interpreted the multiple projection patches traced from V1 in the squirrel lateral extrastriate cortex as subregions within a single area V2. Kaas et al's interpretation was challenged by Wang and Burkhalter (2007), who used a combination of topographic mapping of V1 connections and receptive field recordings in mice. Their findings supported a different partitioning scheme in which each projection patch mapped a specific topographic location within single areas, each containing a complete representation of the visual field. The area map of mouse visual cortex by Wang and Burkhalter (2007) has been reproduced by hundreds of studies and has been widely accepted as ground truth (CCF) (Wang et al., 2020) of the layout of rodent cortex. In the meantime, topographic mappings in marmoset and tree shew visual cortex made a strong case for map reversals in lateral extrastriate cortex, which represent borders between functionally diverse subregions within a single area V2. These findings from non-rodent species raised doubts about whether during evolution, different mammalian branches have developed diverse partitioning schemes of the cerebral cortex. Rowley and Sedigh-Sarvestani favor a single master plan in which, across evolution, all mammalian species have used a similar blueprint for subdividing the cortex.

Strengths:

The story illustrates the enduring strength of science in search of definitive answers.

Weaknesses:

To me, it remains an open question whether Rowley and Sedigh-Sarvestani have written the final chapter of the saga. A key reason for my reservation is that the areas the maps used in their model are cherry-picked. The article disregards published complementary maps, which show that the entire visual field is represented in multiple areas (i.e. LM, AL) of lateral extrastriate cortex and that the map reversal between LM and AL coincides precisely with the transition in m2AChR expression and cytoarchitecture (Wang and Burkhalter, 2007; Wang et al., 2011). Evidence from experiments in rats supports the gist of the findings in the mouse visual cortex (Coogan and Burkhalter, 1993).

We would not claim to have written the final chapter of the saga. Our goal was to add an important piece of new evidence to the discussion of area delineations across species. We believe this new evidence supports our unification hypothesis.  We also believe that there are several missing pieces of data that could support or refute our hypothesis. We have begun a collaboration to collect some of this data.  

(1) The selective use of published evidence, such as the complete visual field representation in higher visual areas of lateral extrastriate cortex (Wang and Burkhalter, 2007; Wang et al., 2011) makes the report more of an opinion piece than an original research article that systematically analyzes the area map of mouse visual cortex we have proposed. No direct evidence is presented for a single area V2 with functionally distinct subregions.

This brings up a nuanced issue regarding visual field coverage. Wang & Burkhalter, 2007 Figure 6 shows the receptive field of sample neurons in area LM that cover the full range between 0 and 90 degrees of azimuth, and -40 to 80 degree of elevation – which essentially matches the visual field coverage in V1. However, we do not know whether these neurons are representative of most neurons in area LM. In other words, while these single-cell recordings along selected contours in cortex show the span of the visual field coverage, they may not be able to capture crucial information about its shape, missing regions of the visual field or potential bias. To mitigate this, visual field maps measured with electrophysiology are commonly produced by even sampling across the two dimensions of the visual area, either by moving a single electrode along a grid-pattern (e.g. (Manger et al., 2002)), or using a grid-liked multi-electrode probe (e.g. (Yu et al., 2020)). This was not carried out either in Wang & Burkhalter 2007 or Wang et al. 2011.  Even sampling of cortical space is time consuming and difficult with electrophysiology, but efficient with functional imaging. Therefore, despite the likely under-estimation of visual field coverage, imaging techniques are valuable in that they can efficiently exhibit not only the span of the visual field of a cortical region, but also its shape and bias.  

Multiple functional imaging studies that simultaneously measure visual field coverage in V1 and HVAs report a bias in the coverage of HVAs, relative to that in V1 (Garrett et al., 2014; Juavinett et al., 2018; Zhuang et al., 2017). While functional imaging will likely underestimate receptive fields compared to electrophysiology, the consistent observation of an orderly bias for distinct parts of the visual field across the HVAs suggests that at least some of the HVAs do not have full and uniform coverage of the visual field comparable to that in V1. For instance, (Garrett et al., 2014) show that the total coverage in HVAs, when compared to V1, is typically less than half (Figure 6D) and often irregularly shaped.

Careful measurements of single-cell receptive fields, using mesoscopic two-photon imaging across the HVAs would settle this question. As reviewer #1 points out, this is technically feasible, though no dataset of this kind exists to our knowledge.

(2) The article misrepresents evidence by commenting that m2AChR expression is mainly associated with the lower field. This is counter to published findings showing that m2AChR spans across the entire visual field (Gamanut et al., 2018; Meier et al., 2021). The utility of markers for delineating areal boundaries is discounted, without any evidence, in disregard of evidence for distinct areal patterns in early development (Wang et al., 2011). Pointing out that markers can be distributed non-uniformly within an area is well-familiar. m2AChR is non-uniformly expressed in mouse V1, LM and LI (Ji et al., 2015; D'Souza et al., 2019; Meier et al., 2021). Recently, it has been found that the patchy organization within V1 plays a role in the organization of thalamocortical and intracortical networks (Meier et al., 2025). m2AChR-positive patches and m2AChR-negative interpatches organize the functionally distinct ventral and dorsal networks, notably without obvious bias for upper and lower parts of the visual field.

We wrote that “Future work showed boundaries in labeling of histological markers such as SMI-32 and m2ChR labeling, but such changes mostly delineated area LM/AL (Wang et al., 2011) and seemed to be correlated with the representation of the lower visual field.” The latter statement regarding the representation of the lower visual field is directly referencing the data in Figure 1 of (Wang et al., 2011), which is titled “Figure 1: LM/AL border identified by the transition of m2AChR expression coincides with receptive field recordings from lower visual field.” Similar to the Wang et al., we were simply referring to the fact that the border of area LM/AL co-exhibits a change in m2AChR expression as well as lower-visual field representation.  

(3) The study has adopted an area partitioning scheme, which is said to be based on anatomically defined boundaries of V2 (Zhuang et al., 2017). The only anatomical borders used by Zhuang et al. (2017) are those of V1 and barrel cortex, identified by cytochrome oxidase staining. In reality, the partitioning of the visual cortex was based on field sign maps, which are reproduced from Zhuang et al., (2017) in Figure 1A. It is unclear why the maps shown in Figures 2E and 2F differ from those in Figure 1A. It is possible that this is an oversight. But maintaining consistent areal boundaries across experimental conditions that are referenced to the underlying brain structure is critical for assigning modeled projections to areas or sub-regions. This problem is evident in Figure 2F, which is presented as evidence that the modeling approach recapitulates the tracings shown in Figure 3 of Wang and Burkhalter (2007). The dissimilarities between the modeling and tracing results are striking, unlike what is stated in the legend of Figure 2F.

Thanks for this correction. By “anatomical boundaries of higher visual cortex”, we meant the cortical boundary between V1 and higher order visual areas on one end, and the outer edge of the envelope that defines the functional boundaries of the HVAs in cortical space (Zhuang et al., 2017). The reviewer is correct that we should have referred to these as functional boundaries. The word ‘anatomical’ was meant to refer to cortical space, rather than visual field space.

More generally though, there is no disagreement between the partitioning of visual cortex in Figure 1 and 2. Rather, the portioning in Figure 1 is directly taken from Zhuang et al., (2017) whereas those in Figure 2 are produced by mathematical model simulation. As such, one would not expect identical areal boundaries between Figure 2 and Figure 1. What we aimed to communicate with our modeling results, is that a single area can exhibit multiple visual field reversals and retinotopic redundancies if it is constrained to fit around V1 and cover a visual field approximately matched to the visual field coverage in V1. We defined this area explicitly as a single area with a single visual field (boundaries shown in Figure 2A). So  the point of our simulation is to show that even an explicitly defined single area can appear as multiple areas if it is constrained by the shape of mouse V1, and if visual field reversals are used to indicate areal boundaries. As in most models, different initial conditions and parameters produce a complex visual field which will appear as multiple HVAs when delineated by areal boundaries. What is consistent however, is the existence of complex single visual field that appears as multiple HVAs with partially overlapping coverage.

Similarly, we would not expect a simple model to exactly reproduce the multi-color tracer injections in Wang and Burkhalter (2007). However, we find it quite compelling that the model can produce multiple groups of multi-colored axonal projections beyond V1 that can appear as multiple areas each with their own map of the visual field using current criteria, when the model is explicitly designed to map a single visual field. We will explain the results of the model, and their implications, better in the revised manuscript.

(4) The Rowley and Sedigh-Sarvestani find that the partial coverage of the visual field in higher order areas shown by Zhuang et al (2017) is recreated by the model. It is important to caution that Zhuang et al's (2017) maps were derived from incomplete mappings of the visual field, which was confined to -25-35 deg of elevation. This underestimates the coverage we have found in LM and AL. Receptive field mappings show that LM covers 0-90 deg of azimuth and -30-80 elevation (Wang and Burkhalter, 2007). AL covers at least 0-90 deg of azimuth and -30-50 deg of elevation (Wang and Burkhalter, 2007; Wang et al., 2011). These are important differences. Partial coverage in LM and AL underestimates the size of these areas and may map two projection patches as inputs to subregions of a single area rather than inputs to two separate areas. Complete, or nearly complete, visual representations in LM and AL support that each is a single area. Importantly, both areas are included in a callosal-free zone (Wang and Burkhalter, 2007). The surrounding callosal connections align with the vertical meridian representation. The single map reversal is marked by a transition in m2AChR expression and cytoarchitecture (Wang et al., 2011).

This is a good point. We do not expect that expanding the coverage of V1 will change the results of the model significantly. However, for the revised manuscript, we will update V1 coverage to be accurate, repeat our simulations, and report the results.  

(5) The statement that the "lack of visual field overlap across areas is suggestive of a lack of hierarchical processing" is predicated on the full acceptance of the mappings by Zhuang et al (2017). Based on the evidence reviewed above, the reclassification of visual areas proposed in Figure 1C seems premature.

The reviewer is correct. In the revised manuscript, we will be careful to distinguish bias in visual field coverage across areas from presence or lack of visual field overlap.  

(6) The existence of lateral connections is not unique to rodent cortex and has been described in primates (Felleman and Van Essen, 1991).

(7) Why the mouse and rat extrastriate visual cortex differ from those of many other mammals is unclear. One reason may be that mammals with V2 subregions are strongly binocular.

This is an interesting suggestion, and careful visual topography data from rabbits and other lateral eyed animals would help to evaluate it. For what it’s worth, tree shrews are lateral eyed animals with only 50 degrees of binocular visual field and also show V2 subregions.

Reviewer #3 (Public review):

Summary:

The authors review published literature and propose that a visual cortical region in the mouse that is widely considered to contain multiple visual areas should be considered a single visual area.

Strengths:

The authors point out that relatively new data showing reversals of visual-field sign within known, single visual areas of some species require that a visual field sign change by itself should not be considered evidence for a border between visual areas.

Weaknesses:

The existing data are not consistent with the authors' proposal to consolidate multiple mouse areas into a single "V2". This is because the existing definition of a single area is that it cannot have redundant representations of the visual field. The authors ignore this requirement, as well as the data and definitions found in published manuscripts, and make an inaccurate claim that "higher order visual areas in the mouse do not have overlapping representations of the visual field". For quantification of the extent of overlap of representations between 11 mouse visual areas, see Figure 6G of Garrett et al. 2014. [Garrett, M.E., Nauhaus, I., Marshel, J.H., and Callaway, E.M. (2014). Topography and areal organization of mouse visual cortex. The Journal of neuroscience 34, 12587-12600. 10.1523/JNEUROSCI.1124-14.2014.

Thank you for this correction, we admit we should have chosen our words more carefully. In the revised manuscript, we will emphasize that higher order visual areas in the mouse do have some overlap in their representations but also exhibit bias in their coverage. This is consistent with our proposal and in fact our model simulations in Figure 2E also show overlapping representations along with differential bias in coverage. However, we also note Figure 6 of Garret et al. 2014 provides several pieces of evidence in support of our proposal that higher order areas are sub-regions of a single area V2. Specifically, the visual field coverage of each area is significantly less than that in V1 (Garret et al. 2014, Figure 6D). While the imaging methods used in Garret et al. likely under-estimate receptive fields, one would assume they would similarly impact measurements of coverage in V1 and HVAs. Secondly, each area exhibits a bias towards a different part of the visual field (Figure 6C and E), that this bias is distinct for different areas but proceeds in a retinotopic manner around V1 - with adjacent areas exhibiting biases for nearby regions of the visual field (Figure 6E). Thus, the biases in the visual field coverage across HVAs appear to be related and not independent of each other. As we show in our modeling and in Figure 2, such orderly and inter-related biases can be created from a single visual field constrained to share a border with mouse V1.   

With regards to the existing definition of a single area: we did not ignore the requirement that single areas cannot have redundant representations of the visual field. Rather, we believe that this requirement should be relaxed considering new evidence collected from other species, where multiple visual field reversals exist within the same visual area. We understand this issue is nuanced and was not made clear in the original submission.  

In the revised manuscript, we will clarify that visual field reversals often exhibit redundant retinotopic representation on either side of the reversal. In the revised manuscript we will clarify that our argument that multiple reversals can exist within a single visual area in the mouse, is an argument that some retinotopic redundancy can exist with single visual areas. Such a re-classification would align how we define visual areas in mice with existing classification in tree shrews, ferrets, cats, and primates – all of whom have secondary visual areas with complex retinotopic maps exhibiting multiple reversals and redundant retinotopic coverage.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

Summary:

Parise presents another instantiation of the Multisensory Correlation Detector model that can now accept stimulus-level inputs. This is a valuable development as it removes researcher involvement in the characterization/labeling of features and allows analysis of complex stimuli with a high degree of nuance that was previously unconsidered (i.e., spatial/spectral distributions across time). The author demonstrates the power of the model by fitting data from dozens of previous experiments, including multiple species, tasks, behavioral modalities, and pharmacological interventions.

Thanks for the kind words!

Strengths:

One of the model's biggest strengths, in my opinion, is its ability to extract complex spatiotemporal co-relationships from multisensory stimuli. These relationships have typically been manually computed or assigned based on stimulus condition and often distilled to a single dimension or even a single number (e.g., "-50 ms asynchrony"). Thus, many models of multisensory integration depend heavily on human preprocessing of stimuli, and these models miss out on complex dynamics of stimuli; the lead modality distribution apparent in Figures 3b and c is provocative. I can imagine the model revealing interesting characteristics of the facial distribution of correlation during continuous audiovisual speech that have up to this point been largely described as "present" and almost solely focused on the lip area.

Another aspect that makes the MCD stand out among other models is the biological inspiration and generalizability across domains. The model was developed to describe a separate process - motion perception - and in a much simpler organism - Drosophila. It could then describe a very basic neural computation that has been conserved across phylogeny (which is further demonstrated in the ability to predict rat, primate, and human data) and brain area. This aspect makes the model likely able to account for much more than what has already been demonstrated with only a few tweaks akin to the modifications described in this and previous articles from Parise.

What allows this potential is that, as Parise and colleagues have demonstrated in those papers since our (re)introduction of the model in 2016, the MCD model is modular - both in its ability to interface with different inputs/outputs and its ability to chain MCD units in a way that can analyze spatial, spectral, or any other arbitrary dimension of a stimulus. This fact leaves wide open the possibilities for types of data, stimuli, and tasks a simplistic, neutrally inspired model can account for.

And so it's unsurprising (but impressive!) that Parise has demonstrated the model's ability here to account for such a wide range of empirical data from numerous tasks (synchrony/temporal order judgement, localization, detection, etc.) and behavior types (manual/saccade responses, gaze, etc.) using only the stimulus and a few free parameters. This ability is another of the model's main strengths that I think deserves some emphasis: it represents a kind of validation of those experiments, especially in the context of cross-experiment predictions (but see some criticism of that below).

Finally, what is perhaps most impressive to me is that the MCD (and the accompanying decision model) does all this with very few (sometimes zero) free parameters. This highlights the utility of the model and the plausibility of its underlying architecture, but also helps to prevent extreme overfitting if fit correctly (but see a related concern below).

We sincerely thank the reviewer for their thoughtful and generous comments. We are especially pleased that the core strengths of the model—its stimulus-computable architecture, biological grounding, modularity, and cross-domain applicability—were clearly recognized. As the reviewer rightly notes, removing researcher-defined abstractions and working directly from naturalistic stimuli opens the door to uncovering previously overlooked dynamics in complex multisensory signals, such as the spatial and temporal richness of audiovisual speech.

We also appreciate the recognition of the model’s origins in a simple organism and its generalization across species and behaviors. This phylogenetic continuity reinforces our view that the MCD captures a fundamental computation with wide-ranging implications. Finally, we are grateful for the reviewer’s emphasis on the model’s predictive power across tasks and datasets with few or no free parameters—a property we see as key to both its parsimony and explanatory utility.

We have highlighted these points more explicitly in the revised manuscript, and we thank the reviewer for their generous and insightful endorsement of the work.

Weaknesses:

There is an insufficient level of detail in the methods about model fitting. As a result, it's unclear what data the models were fitted and validated on. Were models fit individually or on average group data? Each condition separately? Is the model predictive of unseen data? Was the model cross-validated? Relatedly, the manuscript mentions a randomization test, but the shuffled data produces model responses that are still highly correlated to behavior despite shuffling. Could it be that any stimulus that varies in AV onset asynchrony can produce a psychometric curve that matches any other task with asynchrony judgements baked into the task? Does this mean all SJ or TOJ tasks produce correlated psychometric curves? Or more generally, is Pearson's correlation insensitive to subtle changes here, considering psychometric curves are typically sigmoidal? Curves can be non-overlapping and still highly correlated if one is, for example, scaled differently. Would an error term such as mean-squared or root mean-squared error be more sensitive to subtle changes in psychometric curves? Alternatively, perhaps if the models aren't cross-validated, the high correlation values are due to overfitting?

The reviewer is right: the current version of the manuscript only provides limited information about parameter fitting. In the revised version of the manuscript, we included a parameter estimation and generalizability section that includes all information requested by the reviewer.

To test whether using the MSE instead of Pearson correlation led to a similar estimated set of parameter values, we repeated the fitting using the MSE. The parameter estimated with this method (TauV, TauA, TauBim) closely followed those estimated using Pearson correlation (TauV, TauA, TauBim). Given the similarity of these results, we have chosen not to include further figures, however this analysis is now included in the new section (pages 23-24).

Regarding the permutation test, it is expected that different stimuli produce analogous psychometric functions: after all, all studies relied on stimuli containing identical manipulation of lags. As a result, MCD population responses tend to be similar across experiments. Therefore, it is not a surprise that the permuted distribution of MCD-data correlation in Supplementary Figure 1K has a mean as high as 0.97. However, what is important is to demonstrate that the non-permuted dataset has an even higher goodness of fit. Supplementary Figure 1K demonstrates that none of the permuted stimuli could outperform the non-permuted dataset; the mean of the non-permuted distribution is 4.7 (standard deviations) above the mean of the already high  permuted distribution.

We believe the new section, along with the present response, fully addresses the legitimate concerns of the reviewer.

While the model boasts incredible versatility across tasks and stimulus configurations, fitting behavioral data well doesn't mean we've captured the underlying neural processes, and thus, we need to be careful when interpreting results. For example, the model produces temporal parameters fitting rat behavior that are 4x faster than when fitting human data. This difference in slope and a difference at the tails were interpreted as differences in perceptual sensitivity related to general processing speeds of the rat, presumably related to brain/body size differences. While rats no doubt have these differences in neural processing speed/integration windows, it seems reasonable that a lot of the differences in human and rat psychometric functions could be explained by the (over)training and motivation of rats to perform on every trial for a reward - increasing attention/sensitivity (slope) - and a tendency to make mistakes (compression evident at the tails). Was there an attempt to fit these data with a lapse parameter built into the decisional model as was done in Equation 21? Likewise, the fitted parameters for the pharmacological manipulations during the SJ task indicated differences in the decisional (but not the perceptual) process and the article makes the claim that "all pharmacologically-induced changes in audiovisual time perception" can be attributed to decisional processes "with no need to postulate changes in low-level temporal processing." However, those papers discuss actual sensory effects of pharmacological manipulation, with one specifically reporting changes to response timing. Moreover, and again contrary to the conclusions drawn from model fits to those data, both papers also report a change in psychometric slope/JND in the TOJ task after pharmacological manipulation, which would presumably be reflected in changes to the perceptual (but not the decisional) parameters.

Fitting or predicting behaviour does not in itself demonstrate that a model captures the underlying neural computations—though it may offer valuable constraints and insights. In line with this, we were careful not to extrapolate the implications of our simulations to specific neural mechanisms.

Temporal sensitivity is, by definition, a behavioural metric, and—as the reviewer correctly notes—its estimation may reflect a range of contributing factors beyond low-level sensory processing, including attention, motivation, and lapse rates (i.e., stimulus-independent errors). In Equation 21, we introduced a lapse parameter specifically to account for such effects in the context of monkey eye-tracking data. For the rat datasets, however, the inclusion of a lapse term was not required to achieve a close fit to the psychometric data (ρ = 0.981). While it is likely that adding a lapse component would yield a marginally better fit, the absence of single-trial data prevents us from applying model comparison criteria such as AIC or BIC to justify the additional parameter. In light of this, and to avoid unnecessary model complexity, we opted not to include a lapse term in the rat simulations.

With respect to the pharmacological manipulation data, we acknowledge the reviewer’s point that observed changes in slope and bias could plausibly arise from alterations at either the sensory or decisional level—or both. In our model, low-level sensory processing is instantiated by the MCD architecture, which outputs the MCDcorr and MCDlag signals that are then scaled and integrated during decision-making. Importantly, this scaling operation influences the slope of the resulting psychometric functions, such that changes in slope can arise even in the absence of any change to the MCD’s temporal filters. In our simulations, the temporal constants of the MCD units were fixed to the values estimated from the non-pharmacological condition (see parameter estimation section above), and only the decision-related parameters were allowed to vary. From this modelling perspective, the behavioural effects observed in the pharmacological datasets can be explained entirely by changes at the decisional level. However, we do not claim that such an explanation excludes the possibility of genuine sensory-level changes. Rather, we assert that our model can account for the observed data without requiring modifications to early temporal tuning.

To rigorously distinguish sensory from decisional effects, future experiments will need to employ stimuli with richer temporal structure—e.g., temporally modulated sequences of clicks and flashes that vary in frequency, phase, rhythm, or regularity (see Fujisaki & Nishida, 2007; Denison et al., 2012; Parise & Ernst, 2016, 2025; Locke & Landy, 2017; Nidiffer et al., 2018). Such stimuli engage the MCD in a more stimulus-dependent manner, enabling a clearer separation between early sensory encoding and later decision-making processes. Unfortunately, the current rat datasets—based exclusively on single click-flash pairings—lack the complexity needed for such disambiguation. As a result, while our simulations suggest that the observed pharmacologically induced effects can be attributed to changes in decision-level parameters, they do not rule out concurrent sensory-level changes.

In summary, our results indicate that changes in the temporal tuning of MCD units are not necessary to reproduce the observed pharmacological effects on audiovisual timing behaviour. However, we do not assert that such changes are absent or unnecessary in principle. Disentangling sensory and decisional contributions will ultimately require richer datasets and experimental paradigms designed specifically for this purpose. We have now modified the results section (page 6) and the discussion (page 11) to clarify these points.

The case for the utility of a stimulus-computable model is convincing (as I mentioned above), but its framing as mission-critical for understanding multisensory perception is overstated, I think. The line for what is "stimulus computable" is arbitrary and doesn't seem to be followed in the paper. A strict definition might realistically require inputs to be, e.g., the patterns of light and sound waves available to our eyes and ears, while an even more strict definition might (unrealistically) require those stimuli to be physically present and transduced by the model. A reasonable looser definition might allow an "abstract and low-dimensional representation of the stimulus, such as the stimulus envelope (which was used in the paper), to be an input. Ultimately, some preprocessing of a stimulus does not necessarily confound interpretations about (multi)sensory perception. And on the flip side, the stimulus-computable aspect doesn't necessarily give the model supreme insight into perception. For example, the MCD model was "confused" by the stimuli used in our 2018 paper (Nidiffer et al., 2018; Parise & Ernst, 2025). In each of our stimuli (including catch trials), the onset and offset drove strong AV temporal correlations across all stimulus conditions (including catch trials), but were irrelevant to participants performing an amplitude modulation detection task. The to-be-detected amplitude modulations, set at individual thresholds, were not a salient aspect of the physical stimulus, and thus only marginally affected stimulus correlations. The model was of course, able to fit our data by "ignoring" the on/offsets (i.e., requiring human intervention), again highlighting that the model is tapping into a very basic and ubiquitous computational principle of (multi)sensory perception. But it does reveal a limitation of such a stimulus-computable model: that it is (so far) strictly bottom-up.

We appreciate the reviewer’s thoughtful engagement with the concept of stimulus computability. We agree that the term requires careful definition and should not be taken as a guarantee of perceptual insight or neural plausibility. In our work, we define a model as “stimulus-computable” if all its inputs are derived directly from the stimulus, rather than from experimenter-defined summary descriptors such as temporal lag, spatial disparity, or cue reliability. In the context of multisensory integration, this implies that a model must account not only for how cues are combined, but also for how those cues are extracted from raw inputs—such as audio waveforms and visual contrast sequences.

This distinction is central to our modelling philosophy. While ideal observer models often specify how information should be combined once identified, they typically do not address the upstream question of how this information is extracted from sensory input. In that sense, models that are not stimulus-computable leave out a key part of the perceptual pipeline. We do not present stimulus computability as a marker of theoretical superiority, but rather as a modelling constraint that is necessary if one’s aim is to explain how structured sensory input gives rise to perception. This is a view that is also explicitly acknowledged and supported by Reviewer 2.

Framed in Marr’s (1982) terms, non–stimulus-computable models tend to operate at the computational level, defining what the system is doing (e.g., computing a maximum likelihood estimate), whereas stimulus-computable models aim to function at the algorithmic level, specifying how the relevant representations and operations might be implemented. When appropriately constrained by biological plausibility, such models may also inform hypotheses at the implementational level, pointing to potential neural substrates that could instantiate the computation.

Regarding the reviewer’s example illustrating a limitation of the MCD model, we respectfully note that the account appears to be based on a misreading of our prior work. In Parise & Ernst (2025), where we simulated the stimuli from Nidiffer et al. (2018), the MCD model reproduced participants’ behavioural data without any human intervention or adjustment. The model was applied in a fully bottom-up, stimulus-driven manner, and its output aligned with observer responses as-is. We suspect the confusion may stem from analyses shown in Figure 6 - Supplement Figure 5 of Parise & Ernst (2025), where we investigated the lack of a frequency-doubling effect in the Nidiffer et al. data. However, those analyses were based solely on the Pearson correlation between auditory and visual stimulus envelopes and did not involve the MCD model. No manual exclusion of onset/offset events was applied, nor was the MCD used in those particular figures. We also note that Parise & Ernst (2025) is a separate, already published study and is not the manuscript currently under review. 

In summary, while we fully agree that stimulus computability does not resolve all the complexities of multisensory perception (see comments below about speech), we maintain that it provides a valuable modelling constraint—one that enables robust, generalisable predictions when appropriately scoped. 

The manuscript rightly chooses to focus a lot of the work on speech, fitting the MCD model to predict behavioral responses to speech. The range of findings from AV speech experiments that the MCD can account for is very convincing. Given the provided context that speech is "often claimed to be processed via dedicated mechanisms in the brain," a statement claiming a "first end-to-end account of multisensory perception," and findings that the MCD model can account for speech behaviors, it seems the reader is meant to infer that energetic correlation detection is a complete account of speech perception. I think this conclusion misses some facets of AV speech perception, such as integration of higher-order, non-redundant/correlated speech features (Campbell, 2008) and also the existence of top-down and predictive processing that aren't (yet!) explained by MCD. For example, one important benefit of AV speech is interactions on linguistic processes - how complementary sensitivity to articulatory features in the auditory and visual systems (Summerfield, 1987) allow constraint of linguistic processes (Peelle & Sommers, 2015; Tye-Murray et al., 2007).

We thank the reviewer for their thoughtful comments, and especially for the kind words describing the range of findings from our AV speech simulations as “very convincing.”

We would like to clarify that it is not our view that speech perception can be reduced to energetic correlation detection. While the MCD model captures low- to mid-level temporal dependencies between auditory and visual signals, we fully agree that a complete account of audiovisual speech perception must also include higher-order processes—including linguistic mechanisms and top-down predictions. These are critical components of AV speech comprehension, and lie beyond the scope of the current model.

Our use of the term “end-to-end” is intended in a narrow operational sense: the model transforms raw audiovisual input (i.e., audio waveforms and video frames) directly into behavioural output (i.e., button press responses), without reliance on abstracted stimulus parameters such as lag, disparity or reliability. It is in this specific technical sense that the MCD offers an end-to-end model. We have revised the manuscript to clarify this usage to avoid any misunderstanding.

In light of the reviewer’s valuable point, we have now edited the Discussion to acknowledge the importance of linguistic processes (page 13) and to clarify what we mean by end-to-end account (page 11). We agree that future work will need to explore how stimulus-computable models such as the MCD can be integrated with broader frameworks of linguistic and predictive processing (e.g., Summerfield, 1987; Campbell, 2008; Peelle & Sommers, 2015; Tye-Murray et al., 2007).

References

Campbell, R. (2008). The processing of audio-visual speech: empirical and neural bases. Philosophical Transactions of the Royal Society B: Biological Sciences, 363(1493), 1001-1010. https://doi.org/10.1098/rstb.2007.2155

Nidiffer, A. R., Diederich, A., Ramachandran, R., & Wallace, M. T. (2018). Multisensory perception reflects individual differences in processing temporal correlations. Scientific Reports 2018 8:1, 8(1), 1-15. https://doi.org/10.1038/s41598-018-32673-y

Parise, C. V, & Ernst, M. O. (2025). Multisensory integration operates on correlated input from unimodal transient channels. ELife, 12. https://doi.org/10.7554/ELIFE.90841

Peelle, J. E., & Sommers, M. S. (2015). Prediction and constraint in audiovisual speech perception. Cortex, 68, 169-181. https://doi.org/10.1016/j.cortex.2015.03.006

Summerfield, Q. (1987). Some preliminaries to a comprehensive account of audio-visual speech perception. In B. Dodd & R. Campbell (Eds.), Hearing by Eye: The Psychology of Lip-Reading (pp. 3-51). Lawrence Erlbaum Associates.

Tye-Murray, N., Sommers, M., & Spehar, B. (2007). Auditory and Visual Lexical Neighborhoods in Audiovisual Speech Perception: Trends in Amplification, 11(4), 233-241. https://doi.org/10.1177/1084713807307409

Reviewer #2 (Public review):

Summary:

Building on previous models of multisensory integration (including their earlier correlation-detection framework used for non-spatial signals), the author introduces a population-level Multisensory Correlation Detector (MCD) that processes raw auditory and visual data. Crucially, it does not rely on abstracted parameters, as is common in normative Bayesian models," but rather works directly on the stimulus itself (i.e., individual pixels and audio samples). By systematically testing the model against a range of experiments spanning human, monkey, and rat data, the authors show that their MCD population approach robustly predicts perception and behavior across species with a relatively small (0-4) number of free parameters.

Strengths:

(1) Unlike prior Bayesian models that used simplified or parameterized inputs, the model here is explicitly computable from full natural stimuli. This resolves a key gap in understanding how the brain might extract "time offsets" or "disparities" from continuously changing audio-visual streams.

(2) The same population MCD architecture captures a remarkable range of multisensory phenomena, from classical illusions (McGurk, ventriloquism) and synchrony judgments, to attentional/gaze behavior driven by audio-visual salience. This generality strongly supports the idea that a single low-level computation (correlation detection) can underlie many distinct multisensory effects.

(3) By tuning model parameters to different temporal rhythms (e.g., faster in rodents, slower in humans), the MCD explains cross-species perceptual data without reconfiguring the underlying architecture.

We thank the reviewer for their positive evaluation of the manuscript, and particularly for highlighting the significance of the model's stimulus-computable architecture and its broad applicability across species and paradigms. Please find our responses to the individual points below.

Weaknesses:

(1) The authors show how a correlation-based model can account for the various multisensory integration effects observed in previous studies. However, a comparison of how the two accounts differ would shed light on the correlation model being an implementation of the Bayesian computations (different levels in Marr's hierarchy) or making testable predictions that can distinguish between the two frameworks. For example, how uncertainty in the cue combined estimate is also the harmonic mean of the unimodal uncertainties is a prediction from the Bayesian model. So, how the MCD framework predicts this reduced uncertainty could be one potential difference (or similarity) to the Bayesian model.

We fully agree with the reviewer that a comparison between the correlation-based MCD model and Bayesian accounts is valuable—particularly for clarifying how the two frameworks differ conceptually and where they may converge.

As noted in the revised manuscript, the key distinction lies in the level of analysis described by Marr (1982). Bayesian models operate at the computational level, describing what the system is aiming to compute (e.g., optimal cue integration). In contrast, the MCD functions at the algorithmic level, offering a biologically plausible mechanism for how such integration might emerge from stimulus-driven representations.

In this context, the MCD provides a concrete, stimulus-grounded account of how perceptual estimates might be constructed—potentially implementing computations with Bayesian-like characteristics (e.g., reduced uncertainty, cue weighting). Thus, the two models are not mutually exclusive but can be seen as complementary: the MCD may offer an algorithmic instantiation of computations that, at the abstract level, resemble Bayesian inference.

We have now updated the manuscript to explicitly highlight this relationship (pages 2 and 11). In the revised manuscript, we also included a new figure (Figure 5) and movie (Supplementary Movie 3), to show how the present approach extends previous Bayesian models for the case of cue integration (i.e., the ventriloquist effect).

(2) The authors show a good match for cue combination involving 2 cues. While Bayesian accounts provide a direction for extension to more cues (also seen empirically, for eg, in Hecht et al. 2008), discussion on how the MCD model extends to more cues would benefit the readers.

We thank the reviewer for this insightful comment: extending the MCD model to include more than two sensory modalities is a natural and valuable next step. Indeed, one of the strengths of the MCD framework lies in its modularity. Let us consider the MCDcorr​ output (Equation 6), which is computed as the pointwise product of transient inputs across modalities. Extending this to include a third modality, such as touch, is straightforward: MCD units would simply multiply the transient channels from all three modalities, effectively acting as trimodal coincidence detectors that respond when all inputs are aligned in time and space.

By contrast, extending MCDlag is less intuitive, due to its reliance on opponency between two subunits (via subtraction). A plausible solution is to compute MCDlag in a pairwise fashion (e.g., AV, VT, AT), capturing relative timing across modality pairs.

Importantly, the bulk of the spatial integration in our framework is carried by MCDcorr, which generalises naturally to more than two modalities. We have now formalised this extension and included a graphical representation in a supplementary section of the revised manuscript.

Likely Impact and Usefulness:

The work offers a compelling unification of multiple multisensory tasks- temporal order judgments, illusions, Bayesian causal inference, and overt visual attention - under a single, fully stimulus-driven framework. Its success with natural stimuli should interest computational neuroscientists, systems neuroscientists, and machine learning scientists. This paper thus makes an important contribution to the field by moving beyond minimalistic lab stimuli, illustrating how raw audio and video can be integrated using elementary correlation analyses.

Reviewer #1 (Recommendations for the authors):

Recommendations:

My biggest concern is a lack of specificity about model fitting, which is assuaged by the inclusion of sufficient detail to replicate the analysis completely or the inclusion of the analysis code. The code availability indicates a script for the population model will be included, but it is unclear if this code will provide the fitting details for the whole of the analysis.

We thank the reviewer for raising this important point. A new methodological section has been added to the manuscript, detailing the model fitting procedures used throughout the study. In addition, the accompanying code repository now includes MATLAB scripts that allow full replication of the spatiotemporal MCD simulations.

Perhaps it could be enlightening to re-evaluate the model with a measure of error rather than correlation? And I think many researchers would be interested in the model's performance on unseen data.

The model has now been re-evaluated using mean squared error (MSE), and the results remain consistent with those obtained using Pearson correlation. Additionally, we have clarified which parts of the study involve testing the model on unseen data (i.e., data not used to fit the temporal constants of the units). These analyses are now included and discussed in the revised fitting section of the manuscript (pages 23-24).

Otherwise, my concerns involve the interpretation of findings, and thus could be satisfied with minor rewording or tempering conclusions.

The manuscript has been revised to address these interpretative concerns, with several conclusions reworded or tempered accordingly. All changes are marked in blue in the revised version.

Miscellanea:

Should b0 in equation 10 be bcrit to match the below text?

Thank you for catching this inconsistency. We have corrected Equation 10 (and also Equation 21) to use the more transparent notation bcrit instead of b0, in line with the accompanying text.

Equation 23, should time be averaged separately? For example, if multiple people are speaking, the average correlation for those frames will be higher than the average correlation across all times.

We thank the reviewer for raising this thoughtful and important point. In response, we have clarified the notation of Equation 23 in the revised manuscript (page 20). Specifically, we now denote the averaging operations explicitly as spatial means and standard deviations across all pixel locations within each frame.

This equation computes the z-score of the MCD correlation value at the current gaze location, normalized relative to the spatial distribution of correlation values in the same frame. That is, all operations are performed at the frame level, not across time. This ensures that temporally distinct events are treated independently and that the final measure reflects relative salience within each moment, not a global average over the stimulus. In other words, the spatial distribution of MCD activity is re-centered and rescaled at each frame, exactly to avoid the type of inflation or confounding the reviewer rightly cautioned against.

Reviewer #2 (Recommendations for the authors):

The authors have done a great job of providing a stimulus computable model of cue combination. I had just a few suggestions to strengthen the theoretical part of the paper:

(1) While the authors have shown a good match between MCD and cue combination, some theoretical justification or equivalence analysis would benefit readers on how the two relate to each other. Something like Zhang et al. 2019 (which is for motion cue combination) would add to the paper.

We agree that it is important to clarify the theoretical relationship between the Multisensory Correlation Detector (MCD) and normative models of cue integration, such as Bayesian combination. In the revised manuscript, we have now modified the introduction and added a paragraph in the Discussion addressing this link more explicitly. In brief, we see the MCD as an algorithmic-level implementation (in Marr’s terms) that may approximate or instantiate aspects of Bayesian inference.

(2) Simulating cue combination for tasks that require integration of more than two cues (visual, auditory, haptic cues) would more strongly relate the correlation model to Bayesian cue combination. If that is a lot of work, at least discussing this would benefit the paper

This point has now been addressed, and a new paragraph discussing the extension of the MCD model to tasks involving more than two sensory modalities has been added to the Discussion section.

Author response:

The following is the authors’ response to the previous reviews.

Reviewer #1 (Recommendations for the authors):

(1) The onus of making the revisions understandable to the reviewers lies with the authors. In its current form, how the authors have approached the review is hard to follow, in my opinion. Although the authors have taken a lot of effort in answering the questions posed by reviewers, parallel changes in the manuscript are not clearly mentioned. In many cases, the authors have acknowledged the criticism in response to the reviewer, but have not changed their narrative, particularly in the results section.

We fully acknowledge your concern regarding the narrative linking EB-induced GluCl expression to JH biosynthesis and fecundity enhancement, particularly the need to address alternative interpretations of the data. Below, we outline the specific revisions made to address your feedback and ensure the manuscript’s narrative aligns more precisely with the experimental evidence:

(1) Revised Wording in the Results Section

To avoid overinterpretation of causality, we have modified the language in key sections of the Results (e.g., Figure 5 and related text):

Original phrasing:

“These results suggest that EB activates GluCl which induces JH biosynthesis and release, which in turn stimulates reproduction in BPH (Figure 5J).”

Revised phrasing:

“We also examined whether silencing Gluclα impacts the AstA/AstAR signaling pathway in female adults. Knock-down of Gluclα in female adults was found to have no impact on the expression of AT, AstA, AstB, AstCC, AstAR, and AstBR. However, the expression of AstCCC and AstCR was significantly upregulated in dsGluclα-injected insects (Figure 5-figure supplement 2A-H). Further studies are required to delineate the direct or indirect mechanisms underlying this effect of Gluclα-knockdown.” (line 643-649). And we have removed Figure 5J in the revised manuscript.

(2) Expanded Discussion of Alternative Mechanisms

In the Discussion section, we have incorporated a dedicated paragraph to explore alternative pathways and compensatory mechanisms:

Key additions:

“This EB action on GluClα expression is likely indirect, and we do not consider EB as transcriptional regulator of GluClα. Thus, the mechanism behind EB-mediated induction of GluClα remains to be determined. It is possible that prolonged EB exposure triggers feedback mechanisms (e.g. cellular stress responses) to counteract EB-induced GluClα dysfunction, leading to transcriptional upregulation of the channel. Hence, considering that EB exposure in our experiments lasts several days, these findings might represent indirect (or secondary) effects caused by other factors downstream of GluCl signaling that affect channel expression.” (line 837-845).

(2) In the response to reviewers, the authors have mentioned line numbers in the main text where changes were made. But very frequently, those lines do not refer to the changes or mention just a subsection of changes done. As an example please see point 1 of Specific Points below. The problem is throughout the document making it very difficult to follow the revision and contributing to the point mentioned above.

Thank you for highlighting this critical oversight. We sincerely apologize for the inconsistency in referencing line numbers and incomplete descriptions of revisions, which undoubtedly hindered your ability to track changes effectively. We have eliminated all vague or incomplete line number references from the response letter. Instead, revisions are now explicitly tied to specific sections, figures, or paragraphs.

(3) The authors need to infer the performed experiments rationally without over interpretation. Currently, many of the claims that the authors are making are unsubstantiated. As a result of the first review process, the authors have acknowledged the discrepancies, but they have failed to alter their interpretations accordingly.

We fully agree that overinterpretation of data undermines scientific rigor. In response to your feedback, we have systematically revised the manuscript to align claims strictly with experimental evidence and to eliminate unsubstantiated assertions. We sincerely apologize for the earlier overinterpretations and appreciate your insistence on precision. The revised manuscript now rigorously distinguishes between observations (e.g., EB-GluCl-JH correlations) and hypotheses (e.g., GluCl’s mechanistic role). By tempering causal language and integrating competing explanations, we aimed to present a more accurate and defensible narrative.

SPECIFIC POINTS (to each question initially raised and their rebuttals)

(1a) "Actually, there are many studies showing that insects treated with insecticides can increase the expression of target genes". Please note what is asked for is that the ligand itself induces the expression of its receptor. Of course, insecticide treatment will result in the changes expression of targets. Of all the evidences furnished in rebuttal, only Peng et al. 2017 fits the above definition. Even in this case, the accepted mode of action of chlorantraniliprole is by inducing structural change in ryanodine receptor. The observed induction of ryanodine receptor chlorantraniliprole can best be described as secondary effect. All others references do not really suffice the point asked for.

We appreciate the reviewers’ suggestions for improving the manuscript. First, we have supplemented additional studies supporting the notion that " There are several studies showing that insects treated with insecticides display increases in the expression of target genes. For example, the relative expression level of the ryanodine receptor gene of the rice stem borer, Chilo suppressalis was increased 10-fold after treatment with chlorantraniliprole, an insecticide which targets the ryanodine receptor (Peng et al., 2017). In Drosophila, starvation (and low insulin) elevates the transcription level of the receptors of the neuropeptides short neuropeptide F and tachykinin (Ko et al., 2015; Root et al., 2011). In BPH, reduction in mRNA and protein expression of a nicotinic acetylcholine receptor α8 subunit is associated with resistance to imidacloprid (Zhang et al., 2015). Knockdown of the α8 gene by RNA interference decreased the sensitivity of N. lugens to imidacloprid (Zhang et al., 2015). Hence, the expression of receptor genes may be regulated by diverse factors, including insecticide exposure.” We have inserted text in lines 846-857 to elaborate on these possibilities.

Second, we would like to reiterate our position: we have merely described this phenomenon, specifically that EB treatment increases GluClα expression. “This EB action on GluClα expression is likely indirect, and we do not consider EB as transcriptional regulator of GluClα. Thus, the mechanism behind EB-mediated induction of GluClα remains to be determined. It is possible that prolonged EB exposure triggers feedback mechanisms (e.g. cellular stress responses) to counteract EB-induced GluClα dysfunction, leading to transcriptional upregulation of the channel. Hence, considering that EB exposure in our experiments lasts several days, these findings might represent indirect (or secondary) effects caused by other factors downstream of GluCl signaling that affect channel expression.” We have inserted text in lines 837-845 to elaborate on these possibilities.

Once again, we sincerely appreciate this discussion, which has provided us with a deeper understanding of this phenomenon.

b. The authors in their rebuttal accepts that they do not consider EB to a transcriptional regulator of Gluclα and the induction of Gluclα as a result of EB can best be considered as a secondary effect. But that is not reflected in the manuscript, particularly in the result section. Current state of writing implies EB up regulation of Gluclα to an important event that contributes majorly to the hypothesis. So much so that they have retained the schematic diagram (Fig. 5J) where EB -> Gluclα is drawn. Even the heading of the subsection says "EB-enhanced fecundity in BPHs is dependent on its molecular target protein, the Gluclα channel". As mentioned in the general points, it is not enough to have a good rebuttal written to the reviewer, the parent manuscript needs to reflect on the changes asked for.

Thank you for your comments. We have carefully addressed your suggestions and made corresponding revisions to the manuscript.

We fully acknowledge the reviewer's valid concern. In this revised manuscript, “However, we do not propose that EB is a direct transcriptional regulator of Gluclα, since EB and other avermectins are known to alter the channel conformation and thus their function (Wolstenholme, 2012; Wu et al., 2017). Thus, it is likely that the observed increase in Gluclα transcipt is a secondary effect downstream of EB signaling.” (Line 625-629). We agree that the original presentation in the manuscript, particularly within the Results section, did not adequately reflect this nuance and could be misinterpreted as suggesting a direct regulatory role for EB on Gluclα transcription.

Regarding Fig. 5J, we have removed the figure and all mentions of Fig. 5J and its legend in the revised manuscript.

c. "We have inserted text on lines 738 - 757 to explain these possibilities." Not a single line in the section mentioned above discussed the topic in hand. This is serious undermining of the review process or carelessness to the extreme level.

In the Results section, we have now added descriptions “Taken together, these results reveal that EB exposure is associated with an increase in JH titer and that this elevated JH signaling contributes to enhanced fecundity in BPH.” (line 375-377).

For the figures, we have removed Fig. 4N and all mentions of Fig. 4N and its legend in the revised manuscript.

Lastly, regarding the issue of locating specific lines, we deeply regret any inconvenience caused. Due to the track changes mode used during revisions, line numbers may have shifted, resulting in incorrect references. We sincerely apologize for this and have now corrected the line numbers.

(2) The section written in rebuttal should be included in the discussion as well, explaining why authors think a nymphal treatment with JH may work in increasing fecundity of the adults. Also, the authors accept that EBs effect on JH titer in Indirect. The text of the manuscript, results section and figures should be reflective of that. It is NOT ok to accept that EB impacts JH titer indirectly in a rebuttal letter while still continuing to portray EB direct effect on JH titer. In terms of diagrams, authors cannot put a -> sign until and unless the effect is direct. This is an accepted norm in biological publications.

We appreciate the reviewer’s valuable suggestions here. We have now carefully revised the manuscript to address all concerns, particularly regarding the mechanism linking nymphal EB exposure to adult fecundity and the indirect nature of EB’s effect on JH titers. Below are our point-by-point responses and corresponding manuscript changes. Revised text is clearly marked in the resubmitted manuscript.

(1) Clarifying the mechanism linking nymphal EB treatment to adult fecundity:

Reviewer concern: Explain why nymphal EB treatment increases adult fecundity despite undetectable EB residues in adults.

Response & Actions Taken:

We agree this requires explicit discussion. We now propose that nymphal EB exposure triggers developmental reprogramming (e.g., metabolic/epigenetic changes) that persist into adulthood, indirectly enhancing JH synthesis and fecundity. This is supported by two key findings:

(1) No detectable EB residues in adults after nymphal treatment (new Figure 1–figure supplement 1C).

(2) Increased adult weight and nutrient reserves (Figure 1–figure supplement 3E,F), suggesting altered resource allocation.

Added to Discussion (Lines 793–803): Notably, after exposing fourth-instar BPH nymphs to EB, no EB residues were detected in the subsequent adult stage. This finding indicates that the EB-induced increase in adult fecundity is initiated during the nymphal stage and s manifests in adulthood - a mechanism distinct from the direct fecundity enhancement of fecundity observed when EB is applied to adults. We propose that sublethal EB exposure during critical nymphal stages may reprogram metabolic or endocrine pathways, potentially via insulin/JH crosstalk. For instance, increased nutrient storage (e.g., proteins, sugars; Figure 2–figure supplement 2) could enhance insulin signaling, which in turn promotes JH biosynthesis in adults (Ling and Raikhel, 2021; Mirth et al., 2014; Sheng et al., 2011). Future studies should test whether EB alters insulin-like peptide expression or signaling during development.

(3) Emphasizing EB’s indirect effect on JH titers:Reviewer concern: The manuscript overstated EB’s direct effect on JH. Arrows in figures implied causality where only correlation exists.

Response & Actions

Taken:We fully agree. EB’s effect on JH is indirect and multifactorial (via AstA/AstAR suppression, GluCl modulation, and metabolic changes). We have:

Removed oversimplified schematics (original Figures 3N, 4N, 5J).

Revised all causal language (e.g., "EB increases JH" → "EB exposure is associated with increased circulating JH III "). (Line 739)

Clarified in Results/Discussion that EB-induced JH changes are likely secondary to neuroendocrine disruption.

Key revisions:

Results (Lines 375–377):

"Taken together, these results reveal that EB exposure is associated with an increase in JH titer and that JH signaling contributes to enhanced fecundity in BPH."

Discussion (Lines 837–845):

This EB action on GluClα expression is likely indirect, and we do not consider EB as transcriptional regulator of GluClα. Thus, the mechanism behind EB-mediated induction of GluClα remains to be determined. It is possible that prolonged EB exposure triggers feedback mechanisms (e.g. cellular stress responses) to counteract EB-induced GluClα dysfunction, leading to transcriptional upregulation of the channel. Hence, considering that EB exposure in our experiments lasts several days, these findings might represent indirect (or secondary) effects caused by other factors downstream of GluCl signaling that affect channel expression.

a. Lines 281-285 as mentioned, does not carry the relevant information.

Thank you for your careful review of our manuscript. We sincerely apologize for the confusion regarding line references in our previous response. Due to extensive revisions and tracked changes during the revision process, the line numbers shifted, resulting in incorrect citations for Lines 281–285. The correct location for the added results (EB-induced increase in mature eggs in adult ovaries) is now in lines 253-258: “We furthermore observed that EB treatment of female adults also increases the number of mature eggs in the ovary (Figure 2-figure supplement 1).”

b. Lines 351-356 as mentioned, does not carry the relevant information. Lines 281-285 as mentioned, does not carry the relevant information.

Thank you for your careful review of our manuscript. We sincerely apologize for the confusion regarding line references in our previous response. The correct location for the added results is now in lines 366-371: “We also investigated the effects of EB treatment on the JH titer of female adults. The data indicate that the JH titer was also significantly increased in the EB-treated female adults compared with controls (Figure 3-figure supplement 3A). However, again the steroid 20-hydroxyecdysone, was not significantly different between EB-treated BPH and controls (Figure 3-figure supplement 3B).”

c. Lines 378-379 as mentioned, does not carry the relevant information. Lines 387-390 as mentioned, does not carry the relevant information.

We sincerely apologize for the confusion regarding line references in our previous response.

The correct location for the added results is now in lines 393-394: We furthermore found that EB treatment in female adults increases JHAMT expression (Figure 3-figure supplement 3C).

The other correct location for the added results is now in lines 405-408: We found that Kr-h1 was significantly upregulated in the adults of EB-treated BPH at the 5M, 5L nymph and 4 to 5 DAE stages (4.7-fold to 27.2-fold) when 4th instar nymph or female adults were treated with EB (Figure 3H and Figure 3-figure supplement 3D)..

(3) The writing quality is still extremely poor. It does not meet any publication standard, let alone elife.

We fully understand your concerns and frustrations, and we sincerely apologize for the deficiencies in our writing quality, which did not meet the high standards expected by you and the journal. We fully accept your criticism regarding the writing quality and have rigorously revised the manuscript according to your suggestions.

(4) I am confused whether Figure 2B was redone or just edited. Otherwise this seems acceptable to me.

Regarding Fig. 2B, we have edited the text on the y-axis. The previous wording included the term “retention,” which may have caused misunderstanding for both the readers and yourself, leading to the perception of contradiction. We have now revised this wording to ensure accurate comprehension.

(5) The rebuttal is accepted. However, still some of the lines mentioned does not hold relevant information.

This error has been corrected.

The correct location for the added results is now in lines 255-258 and lines 279-282: “Hence, although EB does not affect the normal egg developmental stages (see description in next section), our results suggest that EB treatment promotes oogenesis and, as a result the insects both produce more eggs in the ovary and a larger number of eggs are laid.” and “However, considering that the number of eggs laid by EB treated females was larger than in control females (Figure 1 and Figure 1-figure supplement 1), our data indicates that EB treatment of BPH can both promote both oogenesis and oviposition.”

(6) Thank you for the clarification. Although now discussed extensively in discussion section, the nuances of indirect effect and minimal change in expression should also be reflected in the result section text. This is to ensure that readers have clear idea about content of the paper.

Corrected. To ensure readers gain a clear understanding of our data, we have briefly presented these discussions in the Results section. Please see line 397-402: The levels of met mRNA slightly increased in EB-treated BPH at the 5M and 5L instar nymph and 1 to 5 DAE adult stages compared to controls (1.7-fold to 2.9-fold) (Figure 3G). However, it should be mentioned that JH action does not result in an increase of Met. Thus, it is possible that other factors (indirect effects), induced by EB treatment cause the increase in the mRNA expression level of Met.

(7) As per the author's interpretation, it becomes critical to quantitate the amount of EB present at the adult stages after a 4th instar exposure to it. Only this experiment will unambiguously proof the authors claim. Also, since they have done adult insect exposure to EB, such experiments should be systematically performed for as many sections as possible. Don't just focus on few instances where reviewers have pointed out the issue.

Thank you for raising this critical point. To address this concern, we have conducted new supplementary experiments. The new experimental results demonstrate that residual levels of emamectin benzoate (EB) in adult-stage brown planthoppers (BPH) were below the instrument detection limit following treatment of 4th instar nymphs with EB. Line 172-184: “To determine whether EB administered during the fourth-instar larval stage persists as residues in the adult stage, we used HPLC-MS/MS to quantify the amount of EB present at the adult stage after exposing 4th-instar nymphs to this compound. However, we found no detectable EB residues in the adult stage following fourth-instar nymphal treatment (Figure 1-figure supplement 1C). This suggests that the mechanism underlying the increased fecundity of female adults induced by EB treatment of nymphs may differ from that caused by direct EB treatment of female adults. Combined with our previous observation that EB treatment significantly increased the body weight of adult females (Figure 1—figure supplement 3E and F), a possible explanation for this phenomenon is that EB may enhance food intake in BPH, potentially leading to elevated production of insulin-like peptides and thus increased growth. Increased insulin signaling could potentially also stimulate juvenile hormone (JH) biosynthesis during the adult stage (Badisco et al., 2013).”

(8) Thank you for the revision. Lines 725-735 as mentioned, does not carry the relevant information. However, since the authors have decided to remove this systematically from the manuscript, discussion on this may not be required.

Thank you for identifying the limited relevance of the content in Lines 725–735 of the original manuscript. As recommended, we have removed this section in the revised version to improve logical coherence and maintain focus on the core findings.

(9) Normally, dsRNA would last for some time in the insect system and would down-regulate any further induction of target genes by EB. I suggest the authors to measure the level of the target genes by qPCR in KD insects before and after EB treatment to clear the confusion and unambiguously demonstrate the results. Please Note- such quantifications should be done for all the KD+EB experiments. Additionally, citing few papers where such a rescue effect has been demonstrated in closely related insect will help in building confidence.

We appreciate the reviewer’s suggestion to clarify the interaction between RNAi-mediated gene knockdown (KD) and EB treatment. To address this, we performed additional experiments measuring Kr-h1 expression via qPCR in dsKr-h1-injected insects before and after EB exposure.

The results (now Figure 3–figure supplement 4) show that:

(1) EB did not rescue *Kr-h1* suppression at 24h post-treatment (*p* > 0.05).

(2) Partial recovery of fecundity occurred later (Figure 3M), likely due to:

a) Degradation of dsRNA over time, reducing KD efficacy (Liu et al., 2010).

b) Indirect effects of EB (e.g., hormonal/metabolic reprogramming) compensating for residual Kr-h1 suppression.

Please see line 441-453: “Next, we investigated whether EB treatment could rescue the dsRNA-mediated gene silencing effect. To address this, we selected the Kr-h1 gene and analyzed its expression levels after EB treatment. Our results showed that Kr-h1 expression was suppressed by ~70% at 72 h post-dsRNA injection. However, EB treatment did not significantly rescue Kr-h1 expression in gene knock down insects (*p* > 0.05) at 24h post-EB treatment (Figure 3-figure supplement 4). While dsRNA-mediated Kr-h1 suppression was robust initially, its efficacy may decline during prolonged experiments. This aligns with reports in BPH, where effects of RNAi gradually diminish beyond 7 days post-injection (Liu et al., 2010a). The late-phase fecundity increase might reflect partial Kr-h1 recovery due to RNAi degradation, allowing residual EB to weakly stimulate reproduction. In addition, the physiological impact of EB (e.g., neurotoxicity, hormonal modulation) could manifest via compensatory feedback loops or metabolic remodeling.”

(10) Not a very convincing argument. Besides without a scale bar, it is hard for the reviewers to judge the size of the organism. Whole body measurements of JH synthesis enzymes will remain as a quite a drawback for the paper.

In response to your suggestion, we have also included images with scale bars (see next Figure 1). The images show that the head region is difficult to separate from the brown thoracic sclerite region. Furthermore, the anatomical position of the Corpora Allata in brown planthoppers has never been reported, making dissection uncertain and highly challenging. To address this, we are now attempting to use Drosophila as a model to investigate how EB regulates JH synthesis and reproduction.

Author response image 1.<br /> This illustration provides a visual representation of the brown planthopper (BPH), a major rice pest.<br />

Figure 1. This illustration provides a visual representation of the brown planthopper (BPH), a major rice pest.).

(11) "The phenomenon reported was specific to BPH and not found in other insects. This limits the implications of the study". This argument still holds. Combined with extreme species specificity, the general effect that EB causes brings into question the molecular specificity that the authors claim about the mode of action.

We acknowledge that the specificity of the phenomenon to BPH may limit its broader implications, but we would like to emphasize that this study provides important insights into the unique biological mechanisms in BPH, a pest of significant agricultural importance. The molecular specificity we described in the manuscript is based on rigorous experimental evidence. We believe that it contributes to valuable knowledge to understand the interaction of external factors such as EB and BPH and resurgence of pests. We hope that this study will inspire further research into the mechanisms underlying similar phenomena in other insects, thereby broadening our understanding of insect biology. Since EB also has an effect on fecundity in Drosophila, albeit opposite to that in BPHs (Fig. 1 suppl. 2), it seems likely that EB actions may be of more general interest in insect reproduction.

(12) The authors have added a few lines in the discussion but it does not change the overall design of the experiments. In this scenario, they should infer the performed experiments rationally without over interpretation. Currently, many of the claims that the authors are making are unsubstantiated. As a result of the first review process, the authors have acknowledged the discrepancies, but they have failed to alter their interpretations accordingly.

We appreciate your concern regarding the experimental design and the need for rational inference without overinterpretation. In response, we would like to clarify that our discussion is based on the experimental data we have collected. We acknowledge that our study focuses on BPH and the specific effects of EB, and while we agree that broader generalizations require further research, we believe the new findings we present are valid and contribute to the understanding of this specific system.

We also acknowledge the discrepancies you mentioned and have carefully considered your suggestions. In this revised version, we believe our interpretations are reasonable and consistent with the data, and we have adjusted our discussion to better reflect the scope of our findings. We hope that these revisions address your concerns. Thank you again for your constructive feedback.

ADDITIONAL POINTS

(1) Only one experiment was performed with Abamectin. No titration for the dosage were done for this compound, or at least not provided in the manuscript. Inclusion of this result will confuse readers. While removing this result does not impact the manuscript at all. My suggestion would be to remove this result.

We acknowledge that the abamectin experiment lacks dose-titration details and that its standalone presentation could lead to confusion. However, we respectfully request to retain these results for the following reasons:

Class-Specific Mechanism Validation:

Abamectin and emamectin benzoate (EB) are both macrocyclic lactones targeting glutamate-gated chloride channels (GluCls). The observed similarity in their effects on BPH fecundity (e.g., Figure 1—figure supplement 1B) supports the hypothesis that GluCl modulation, rather than compound-specific off-target effects, drives the reproductive enhancement. This consistency strengthens the mechanistic argument central to our study.

(2) The section "The impact of EB treatment on BPH reproductive fitness" is poorly described. This needs elaboration. A line or two should be included to describe why the parameters chosen to decide reproductive fitness were selected in the first place. I see that the definition of brachypterism has undergone a change from the first version of the manuscript. Can you provide an explanation for that? Also, there is no rationale behind inclusion of statements on insulin at this stage. The authors have not investigated insulin. Including that here will confuse readers. This can be added in the discussion though.

Thank you for your suggestion. We have added an explanation regarding the primary consideration of evaluating reproductive fitness. In the interaction between sublethal doses of insecticides and pests, reproductive fitness is a key factor, as it accurately reflects the potential impact of insecticides on pest control in the field. Among the reproductive fitness parameters, factors such as female Nilaparvata lugens body weight, lifespan, and brachypterous ratio (as short-winged N. lugens exhibit higher oviposition rates than long-winged individuals) are critical determinants of reproductive success. Therefore, we comprehensively assessed the effects of EB on these parameters to elucidate the primary mechanism by which EB influences reproduction. We sincerely appreciate your constructive feedback.

(3) "EB promotes ovarian maturation in BPH" this entire section needs to be rewritten and attention should be paid to the sequence of experiments described.

Thank you for your suggestion. Based on your recommendation, we have rewritten this section (lines 267–275) and adjusted the sequence of experimental descriptions to improve the structural clarity of this part.

(4) Figure 3N is outright wrong and should be removed or revised.

In accordance with your recommendation, we have removed the figure.

(5) When you are measuring hormonal titers, it is important to mention explicitly whether you are measuring hemolymph titer or whole body.

We believe we have explicitly stated in the Methods section (line 1013) that we measured whole-body hormone titers. However, we now added this information to figure legends.

(6)  EB induces JH biosynthesis through the peptidergic AstA/AstAR signaling pathway- this section needs attention at multiple points. Please check.

We acknowledge that direct evidence for EB-AstA/AstAR interaction is limited and have framed these findings as a hypothesis for future validation.

References

Liu, S., Ding, Z., Zhang, C., Yang, B., Liu, Z., 2010. Gene knockdown by intro-thoracic injection of double-stranded RNA in the brown planthopper, Nilaparvata lugens. Insect Biochem. Mol. Biol. 40, 666-671

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public Review):

In this manuscript, Roy et al. used the previously published deep transfer learning tool, DEGAS, to map disease associations onto single-cell RNA-seq data from bulk expression data. The authors performed independent runs of DEGAS using T2D or obesity status and identified distinct β-cell subpopulations. β-cells with high obese-DEGAS scores contained two subpopulations derived largely from either non-diabetic or T2D donors. Finally, immunostaining using human pancreas sections from healthy and T2D donors validated the heterogeneous expression and depletion of DLK1 in T2D islets.

Strengths:

(1) This meta-analysis of previously published scRNA-seq data using a deep transfer learning tool.

(2) Identification of novel beta cell subclusters.

(3) Identified a relatively innovative role of DLK1 in T2D disease progression.

Thank you for your comments on the strengths of our work.

Weaknesses :

“There is little overlap of the DE list of bulk RNA-seq analysis in Figure 1D and 1E overlap with the DE list of pseudo-bulk RNA-seq analysis of all cells in Figure S2C. “

Thank you for pointing this out. To clarify, we did not perform pseudo-bulk analysis on the scRNAseq data. Instead, we used the Seurat FindClusterMarkers function to identify differentially enriched genes between T2D and ND single cells. Indeed, there are many significant genes in new Fig S2D (original S2C). There is some overlap between those data and the DEGS from bulk RNAseq data in Fig 1D, including IAPP, ENTPD3, and FFAR4. However, the limited overlap supports the notion that improved approaches are necessary to identify candidate DEGs from single cell data, as simply performing a comparison of T2D to ND of all β-cells may miss important genes or include many false positives. We have now added clarification to the text to highlight this point.

The biological meaning of "beta cells had the lowest scores compared to other cell types" is not clear.

The relatively lower T2D-DEGAS scores for beta cells overall compared to all other cell types (alpha cells, acinar cells, etc) likely reflects the fact that in T2D, beta cell-specific genes can be downregulated. This affects the DEGAS model which is reflected in the scores of all cells in the scRNAseq data. By subsetting the beta cells and replotting them on their own, we can analyze the relative differences in DEGAS scores between different subsets of beta cells. We have now amended the text to clarify, as follows:

“We next mapped the T2D-association scores onto the single cells (Fig 3A). β-cells had a wide distribution of scores, possibly reflecting β-cell heterogeneity or altered β-cell gene expression after onset of T2D (Fig 3B).”

The figures and supplemental figures were not cited following the sequence, which makes the manuscript very difficult to read. Some supplemental figures, such as Figures S1C-S1D, S2B-S2E, S3A-S3B, were not cited or mentioned in the text.

We apologize for this oversight and have now amended the text to call out all figures/panels in order of first introduction.

In Figure 7, the current resolution is too low to determine the localization of DLK1.

We have confirmed that in our Adobe Illustrator file, each microscopy panel has a DPI of >600. We have also provided the highest quality TIFF file versions of our figure set. We hope the reviewer will have access to download the high-quality TIFF file for Fig 7 if possible, or the editorial staff can provide it.

As a result of addressing the critiques, we identified CDKN1C as another promising candidate enriched in the β^T2D-DEGAS and β^obese-DEGAS subpopulations of β-cells. We found that CDKN1C is heterogeneously expressed at the protein level in β-cells and that it is increased in T2D in agreement with the DEGAS predictions. We have amended the manuscript to highlight CDKN1C more prominently while still discussing DLK1. DLK1 is very interesting, but exhibits greater donor to donor variability in its alterations in T2D.

Reviewer #2 (Public Review):

Summary:

The manuscript by Gitanjali Roy et al. applies deep transfer learning (DEGAS) to assign patient-level disease attributes (metadata) to single cells of T2D and non-diabetic patients, including obese patients. This led to the identification of a singular cluster of T2D-associated β-cells; and two subpopulations of obese- β-cells derived from either non-diabetic or T2D donors. The objective was to identify novel and established genes implicated in T2D and obesity. Their final goal is to validate their findings at the protein level using immunohistochemistry of pancreas tissue from non-diabetic and T2D organ donors.

Strengths:

This paper is well-written, and the findings are relevant for β-cell heterogeneity in T2D and obesity.

Thank you for your comments on the positive aspects of our work.

Weaknesses:

The validation they provide is not sufficiently strong: no DLK1 immunohistochemistry is shown of obese patient-derived sections.

We have acquired additional FFPE pancreas samples from the Integrated Islet Distribution Program (IIDP) from lean, overweight, and obese humans with and without T2D. We have now stained for CDKN1C and DLK1 in these samples and have integrated the data into Fig 7 and Fig S5.

Because the data with CDKN1C was more striking and consistent with the DEGAS predictions, we have chosen to highlight CDKN1C in the main figure and text. The DLK1 data is still quite interesting, although there is substantial variability between T2D donors when it comes to altered staining intensity. DLK1 presents an interesting challenge, given multiple isoforms and cleavage products, and will require further investigation as the focus of a different manuscript.

Additional presumptive relevant candidates from this transcriptomic analysis should be screened for, at the protein level.

Thank you for this suggestion. We also identified CDKN1C as promising candidate enriched in the β^T2D-DEGAS and β^obese-DEGAS subpopulations of β-cells. We found that CDKN1C is heterogeneously expressed at the protein level in β-cells and that it is increased in T2D in agreement with the DEGAS predictions. We have amended the manuscript to highlight CDKN1C more prominently while still discussing DLK1. DLK1 is very interesting but exhibits greater donor to donor variability in its alterations in T2D.

Reviewer #1 (Recommendations For The Authors):

Please explain and provide the detailed information on what percentage of the DE list of bulk RNA-seq analysis in Figures 1D and 1E overlap with the DE list of pseudo-bulk RNA-seq analysis of all cells in Figure S2C.

Addressed in response to R1 Comment 1.

Please provide the definition of each cluster of UMAP of the merged human islet scRNA-seq data.

In figure panels 2A-B,D-G and 3A, the clusters are now labeled according to the marker genes described in Fig 2C.

The integrative UMAP needs to be included in the main figure.

We have now moved previous Fig S2A and S2B into the main figures as new Fig 2A-B.

All figures and supplemental figures need to be cited following sequence.

Addressed in response to R1 Comment 3.

In Figure 7, high-resolution images are needed to determine the colocalization of INS and DLK1.

Addressed in response to R1 Comment 4.

Reviewer #2 (Recommendations For The Authors):

Results: 124-128: Fig 1H_The error bars seem high, please include whether the boxplots are SEM or SD. Also, more detail on statistics is missing.

Thank you for pointing out the need for clarification here. The whiskers on the box and whiskers plots are not error bars. By default, in geom_boxplot() and stat_boxplot(), the whiskers extend to 1.5 times the interquartile range. The box itself represents 50% of the data, the bottom of the box is the first quartile, the middle horizontal line is the median, and the top line of the box is the third quartile. We have now added a clearer description of this to the figure legend and in the methods section.

The genes shown in Fig 1H were selected because they are found in the T2D Knowledge Portal, illustrating a clear link to T2D. At the T2DKP (https://t2d.hugeamp.org/research.html?pageid=mccarthy_t2d_247), PAX4 and APOE are listed as causal, SLC2A2 has strong evidence, and CYTIP has a linked SNP. This is now discussed in the results section before the Fig 1H callout. These genes are significantly differentially expressed using edgeR in panel 1D with FDR<0.05. The individual data points for each human are shown.

Figure 6: In general, the representation of the data is quite misleading. It would be nice to have an alternative way of presenting the data, especially when comparing beta-obese differentially expressed genes and pathways and T2D beta obese. Maybe an additional Venn diagram can help. Also, it would be nice to compare data from T2D beta nonobese to ND beta obese, especially given how the story is presented in the paper.

Thank you for pointing out this clarity issue. We agree that additional alternate ways to present the data would be helpful. When we performed DEGAS using BMI as the disease feature we noted two major and one minor clusters of high-scoring cells in Fig 6A .

Author response image 1.

Author response image 2.<br />

This contrasted with the score map when we ran DEGAS with T2D as the disease feature

The main difference seems to be the low scoring β^T2D-DEGAS cluster is different from the low β^obese-DEGAS cluster.

Therefore, we could not easily apply thresholding to the β^obese-DEGAS scores, so instead we subsetted them for comparison. It was also apparent from the metadata that single cells from the left-hand side of the β-cell cluster came from donors that had T2D.

To clarify these points and address the reviewer’s concerns, we have added a comparison of the DEGs identified for β^T2D-DEGAS high vs. low and T2D-β^obese-DEGAS vs ND-β^obese-DEGAS in Fig S4J, also shown below. DLK1 and CDKNC1C fall within the intersection, in addition to being two of the most enriched candidates in each DEGAS run (Fig 4C and Fig 6D).

220-222: Figure 7C_ Is one of the nondiabetic beta samples obese? If so, please clearly label it; if not, that info is missing. One would expect that the DLK1 expression in ND obese beta cells resembles the T2D beta cell and not ND non-obese beta cells. That's a big point of this entire work, and experimentally missing. Additional candidate proteins should be checked.

We have amended the entire Fig 7 to include more data for DLK1 staining as well as adding staining for CDKN1C. We also used CellProfiler to quantify the intensity distribution of DLK1 staining in β-cells and overall found that our initial conclusions were not supported when considering an increased sample size. DLK1 expression is heterogeneous both within and between donors. While we have data from T2D donors that shows DLK1 is lost, other T2D samples indicate that DLK1 is not always lost. At least in the current sample set we have analyzed, we cannot conclude that there is a clear correlation between diabetes or BMI for DLK1. Why DLK1 labels some β-cells and not others and what the role of this subpopulation is an open question.

Alternatively, we greatly appreciate the reviewer’s suggestion to validate additional candidates, as this led us to CDKN1C. In new Fig 7E-H we now show that CDKN1C is increased in T2D β-cells, in agreement with the DEGAS predictions.

This work shows that machine learning approaches are powerful for identifying potential candidates, but it also highlights the need for these predictions to be validated at the protein level in human samples.

Discussion: Based on lack of supporting IHC data, this is an overstatement:

“DLK1 expression highly overlapped with high scoring βT2D DEGAS cells (Figure 7A) and with T2D βobese-DEGAS cells (Figure 7B). DLK1 immunostaining primarily colocalized with β-cells in non-diabetic human pancreas (Figure 7C). DLK1 showed heterogeneous expression within islets and between islets within the same pancreas section, wherein some islets had DLK1/INS co-staining in most β-cells and other islets had only a few DLK1+ β-cells. In the T2D pancreas, DLK1 staining was much less intense and in fewer β-cells, yet DLK1+/INS+ cells were observed (Figure 7C). This contrasts with the relatively higher DLK1 gene expression seen in the β-cells from the βT2D-DEGAS and T2D-βobese-DEGAS subpopulations (Figure 4D & 6C) as highlighted in Figure 7A,B. which were up- or down-regulated in subpopulations of β-cells identified by DEGAS, and to validate our findings at the protein level using immunohistochemistry of pancreas tissue from non-diabetic and T2D organ donors.”

This part was at the very end of the last results subsection. This section has been largely rewritten to better describe the new figure and the language has been tempered to not overinterpret the data shown.

“Our current findings applying DEGAS to islet data have implications for β-cell heterogeneity in T2D and obesity. The abundance of T2D-related factors and functional β-cell genes in our analysis validates applying DEGAS to islet data to identify disease-associated phenotypes and increase confidence in the novel candidate.”

This part was found at the end of the Background section. We have removed the second sentence to temper the language.

Author response:

Reviewer #1 (Public review):

This study established a C921Y OGT-ID mouse model, systematically demonstrating in mammals the pathological link between O-GlcNAc metabolic imbalance and neurodevelopmental disorders (cortical malformation, microcephaly) as well as behavioral abnormalities (hyperactivity, impulsivity, learning/memory deficits). However, critical flaws in the current findings require resolution to ensure scientific rigor.

The most concerning finding appears in Figure S12. While Supplementary Figure S12 demonstrates decreased OGA expression without significant OGT level changes in C921Y mutants via Western blot/qPCR, previous reports (Florence Authier, et al., Dis Model Mech. 2023) described OGT downregulation in Western blot and an increase in qPCR in the same models. The opposite OGT expression outcomes in supposedly identical mouse models directly challenge the model's reliability. This discrepancy raises serious concerns about either the experimental execution or the interpretation of results. The authors must revalidate the data with rigorous controls or provide a molecular biology-based explanation.

The referee’s assessment is based on a misunderstanding – these are certainly not the same experiment repeated twice with different answers. In the previous report of the OGT-C921Y mutant mice (Florence Authier, et al., Dis Model Mech. 2023), OGT and OGA mRNA/protein expression have been assessed in total brain protein extract from 3 months old male mice. In that study we observed a significant reduction in OGT protein levels while OGT mRNA levels were significantly increased in the mutant compared to WT controls. However, in our the current study (Figure S12), OGA and OGT mRNA/protein expression have been a) restricted to the pre-frontal cortex and b) are from 4 months old male mice, which does not allow a direct comparison of the two studies. In the pre-frontal cortex, OGT protein levels are not changed while OGT mRNA levels are increased (similarly to the total brain data), albeit not significantly. The different outcomes of OGT protein levels in both total brain and prefrontal cortex could suggest regional differences in OGT protein levels/stability as OGT mRNA levels are increased in both cases. Three other brain regions (hippocampus, striatum and cerebellum) have now also been assessed for OGT mRNA/protein expression, supporting such regional differences in OGT protein levels and these data will be included in the new version of the manuscript.

A few additional comments to the author may be helpful to improve the study.

Major

(1) While this study systematically validated multi-dimensional phenotypes (including neuroanatomical abnormalities and behavioral deficits) in OGT C921Y mutant mice, there is a lack of relevant mechanisms and intervention experiments. For example, the absence of targeted intervention studies on key signaling pathways prevents verification of whether proteomics-identified molecular changes directly drive phenotypic manifestations.

We agree with the referee that these experiments would further strenghten the work. They would, however, result in a 1-5 year delay in sharing this work with the scientific and patient communities. We will continue to work along these lines and report separately in the future.

(2) Although MRI detected nodular dysplasia and heterotopia in the cingulate cortex, the cellular basis remains undefined. Spatiotemporal immunofluorescence analysis using neuronal (NeuN), astrocytic (GFAP), and synaptic (Synaptophysin) markers is recommended to identify affected cell populations (e.g., radial glial migration defects or intermediate progenitor differentiation abnormalities).

We are currently performing these experiments so that they can be included in the version of record of this manuscript.

(3) While proteomics revealed dysregulation in pathways including Wnt/β-catenin and mTOR signaling, two critical issues remain unresolved: a) O-GlcNAc glycoproteomic alterations remain unexamined; b) The causal relationship between pathway changes and O-GlcNAc imbalance lacks validation. It is recommended to use co-immunoprecipitation or glycosylation sequencing to confirm whether the relevant proteins undergo O-GlcNAc modification changes, identify specific modification sites, and verify their interactions with OGT.

We agree with the referee that these experiments would further strenghten the work and will perform further experiments to explore whether these pathways are functionally affected. However, it is important to note that the inference that these proteins must themselves be O-GlcNAc modified is incorrect – indeed, O-GlcNAcylation of unknown protein kinase X, E3 ligase/DUB, Y or transcription factor Z could indirectly affect these pathways/proteins.

(4) Given that OGT-ID neuropathology likely originates embryonically, we recommend serial analyses from E14.5 to P7 to examine cellular dynamics during critical corticogenesis phases.

We agree with the referee that these experiments would further strenghten the work. They would, however, result in a significant delay in sharing this work with the scientific and patient communities. We will continue to work along these lines and report separately in the future.

(5) The interpretation of Figure 8A constitutes overinterpretation. Current data fail to conclusively demonstrate impairment of OGT's protein interaction network and lack direct evidence supporting the proposed mechanisms of HCF1 misprocessing or OGA loss.

For clarity, we will remove panel A from Figure 8 in the version of record – this panel was only ever meant to represent a priori hypotheses for OGT-CDG mechanisms, none of which have been either excluded or confirmed.

Reviewer #2 (Public review):

Summary:

The authors are trying to understand why certain mutants of O-GlcNAc transferase (OGT) appear to cause developmental disorders in humans. As an important step towards that goal, the authors generated a mouse model with one of these mutations that disrupts OGT activity. They then go on to test these mice for behavioral differences, finding that the mutant mice exhibit some signs of hyperactivity and differences in learning and memory. They then examine alterations to the structure of the brain and skull, and again find changes in the mutant mice that have been associated with developmental disorders. Finally, they identify proteins that are up- or down-regulated between the two mice as potential mechanisms to explain the observations.

Strengths:

The major strength of this manuscript is the creation of this mouse model, as a key step in beginning to understand how OGT mutants cause developmental disorders. This line will prove important for not only the authors but other investigators as well, enabling the testing of various hypotheses and potentially treatments. The experiments are also rigorously performed, and the conclusions are well supported by the data.

Weaknesses:

The only weakness identified is a lack of mechanistic insight. However, this certainly may come in the future through more targeted experimentation using this mouse model.

We agree with the referee that these experiments would further strenghten the work. They would, however, result in a 1-5 year delay in sharing this work with the scientific and patient communities. We will continue to work along these lines and report separately in the future.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public Review):

Summary:

Previous studies have shown that treatment with 17α-estradiol (a stereoisomer of the 17β-estradiol) extends lifespan in male mice but not in females. The current study by Li et al, aimed to identify cell-specific clusters and populations in the hypothalamus of aged male rats treated with 17α-estradiol (treated for 6 months). This study identifies genes and pathways affected by 17α-estradiol in the aged hypothalamus.

Strengths:

Using single-nucleus transcriptomic sequencing (snRNA-seq) on the hypothalamus from aged male rats treated with 17α-estradiol they show that 17α-estradiol significantly attenuated age-related increases in cellular metabolism, stress, and decreased synaptic activity in neurons.

Thanks.

Moreover, sc-analysis identified GnRH as one of the key mediators of 17α-estradiol's effects on energy homeostasis. Furthermore, they show that CRH neurons exhibited a senescent phenotype, suggesting a potential side effect of the 17α-estradiol. These conclusions are supported by supervised clustering by neuropeptides, hormones, and their receptors.

Thanks.

Weaknesses:

However, the study has several limitations that reduce the strength of the key claims in the manuscript. In particular:

(1) The study focused only on males and did not include comparisons with females. However, previous studies have shown that 17α-estradiol extends lifespan in a sex-specific manner in mice, affecting males but not females. Without the comparison with the female data, it's difficult to assess its relevance to the lifespan.

This study was originally designed based on previous findings indicating that lifespan extension is only effective in males, leading to the exclusion of females from the analysis. The primary focus of our research was on the transcriptional changes and serum endocrine alterations induced by 17α-estradiol in aged males compared to untreated aged males. We believe that even in the absence of female subjects, the significant effects of 17α-estradiol on metabolism in the hypothalamus, synapses, and endocrine system remain evident, particularly regarding the expression levels of GnRH and testosterone. Notably, lower overall metabolism, increased synaptic activity, and elevated levels of GnRH and testosterone are strong indicators of health and well-being in males, supporting the validity of our primary conclusions. However, including female controls would enhance the depth of our findings. If female controls were incorporated, we propose redesigning the sample groups to include aged male control, aged female control, aged female treated, aged male treated, as well as young male control, young male treated, young female control, and young female treated. We regret that we cannot provide this data in the short term. Nevertheless, we believe this reviewer’s creative idea presents a valuable avenue for future research on this topic. In this study, we emphasize the role of 17α-estradiol in overall metabolism, synaptic function, GnRH, and testosterone in aged males and underscore the importance of supervised clustering of neuropeptide-secreting neurons in the hypothalamus.

(2) It is not known whether 17α-estradiol leads to lifespan extension in male rats similar to male mice. Therefore, it is not possible to conclude that the observed effects in the hypothalamus, are linked to the lifespan extension.

Thanks for the reminding. 17α-estradiol was reported to extend lifespan in male rats similar to male mice (PMID: 33289482). We have added the valuable reference to introduction in the new version.  

(3) The effect of 17α-estradiol on non-neuronal cells such as microglia and astrocytes is not well-described (Figure 1). Previous studies demonstrated that 17α-estradiol reduces microgliosis and astrogliosis in the hypothalamus of aged male mice. Current data suggest that the proportion of oligo, and microglia were increased by the drug treatment, while the proportions of astrocytes were decreased. These data might suggest possible species differences, differences in the treatment regimen, or differences in drug efficiency. This has to be discussed.

We have reviewed reports describing changes in cell numbers following 17α-estradiol treatment in the brain, using the keywords "17α-estradiol," "17alpha-estradiol," and "microglia" or "astrocyte." Only a limited amount of data was obtained. We found one article indicating that 17α-estradiol treatment in Tg (AβPP(swe)/PS1(ΔE9)) model mice resulted in a decreased microglial cell number compared to the placebo (AβPP(swe)/PS1(ΔE9) mice), but this change was not significant when compared to the non-transgenic control (PMID: 21157032). The transgenic AβPP(swe)/PS1(ΔE9) mouse model may differ from our wild-type aging rat model in this context.

Moreover, the calculation of cell numbers was based on visual observation under a microscope across several brain tissue slices. This traditional method often yields controversial results. For example, oligodendrocytes in the corpus callosum, fornix, and spinal cord have been reported to be 20-40% more numerous in males than in females based on microscopic observations (PMID: 16452667). In contrast, another study found no significant difference in the number of oligodendrocytes between sexes when using immunohistochemistry staining (PMID: 18709647). Such discrepancies arising from traditional observational methods are inevitable.

We believe the data presented in this article are reliable because the cell number and cell ratio data were derived from high-throughput cell counting of the entire hypothalamus using single-cell suspension and droplet wrapping (10x Genomics).

(4) A more detailed analysis of glial cell types within the hypothalamus in response to drugs should be provided.

We provided more enrichment analysis data of differentially expressed genes between Y, O, and O.T in microglia and astrocytes in Figure 2—figure supplement 3. In this supplemental data, we found unlike that in neurons, Micro displayed lower levels of synapse-related cellular processes in O.T. compared to O.

(5) The conclusion that CRH neurons are going into senescence is not clearly supported by the data. A more detailed analysis of the hypothalamus such as histological examination to assess cellular senescence markers in CRH neurons, is needed to support this claim.

We also noted the inappropriate claim and have changed "senescent phenotype" to "stressed phenotype" and "abnormal phenotype" in both the abstract and results sections. The stressed phenotype could be induced by heightened functional activity in the cells, potentially indicating higher cellular activity. The GnRH and CRH neurons discussed in this paper may represent such a case, as illustrated by the observed high serum GnRH, testosterone, and cortisol levels. This revision suggestion is highly valuable and constructive for our understanding of the unique physiological characteristics revealed by these data.

Reviewer #2 (Public Review):

Summary:

Li et al. investigated the potential anti-ageing role of 17α-Estradiol on the hypothalamus of aged rats. To achieve this, they employed a very sophisticated method for single-cell genomic analysis that allowed them to analyze effects on various groups of neurons and non-neuronal cells. They were able to sub-categorize neurons according to their capacity to produce specific neurotransmitters, receptors, or hormones. They found that 17α-Estradiol treatment led to an improvement in several factors related to metabolism and synaptic transmission by bringing the expression levels of many of the genes of these pathways closer or to the same levels as those of young rats, reversing the ageing effect. Interestingly, among all neuronal groups, the proportion of Oxytocin-expressing neurons seems to be the one most significantly changing after treatment with 17α-Estradiol, suggesting an important role of these neurons in mediating its anti-ageing effects. This was also supported by an increase in circulating levels of oxytocin. It was also found that gene expression of corticotropin-releasing hormone neurons was significantly impacted by 17α-Estradiol even though it was not different between aged and young rats, suggesting that these neurons could be responsible for side effects related to this treatment. This article revealed some potential targets that should be further investigated in future studies regarding the role of 17α-Estradiol treatment in aged males.

Strengths:

(1) Single-nucleus mRNA sequencing is a very powerful method for gene expression analysis and clustering. The supervised clustering of neurons was very helpful in revealing otherwise invisible differences between neuronal groups and helped identify specific neuronal populations as targets.

Thanks.

(2) There is a variety of functions used that allow the differential analysis of a very complex type of data. This led to a better comparison between the different groups on many levels.

Thanks.

(3) There were some physiological parameters measured such as circulating hormone levels that helped the interpretation of the effects of the changes in hypothalamic gene expression

Thanks.

Weaknesses

(1) One main control group is missing from the study, the young males treated with 17α-Estradiol.

Given that the treatment period lasts six months, which extends beyond the young male rats' age range, we aimed to investigate the perturbation of 17α-Estradiol on the normal aging process. Including data from young males could potentially obscure the treatment's effects in aged males due to age effects, though similar effects between young and aged animals may exist. Long-term treatment of hormone may exert more developmental effects on the young than the old. Consequently, we decided to exclude this group from our initial sample design. We apologize for this omission.

(2) Even though the technical approach is a sophisticated one, analyzing the whole rat hypothalamus instead of specific nuclei or subregions makes the study weaker.

The precise targets of 17α-Estradiol within the hypothalamus remain unresolved. Selecting a specific nucleus for study is challenging. The supervised clustering method described in this manuscript allows us to identify the more sensitive neuron subtypes influenced by 17α-Estradiol and aging across the entire hypothalamus, without the need to isolate specific nuclei in a disturbed hypothalamic environment.

(3) Although the authors claim to have several findings, the data fail to support these claims. You may mean the claim as the senescent phenotype in Crh neuron induced by 17a-estradiol.

Thanks. We have changed the "senescent phenotype" to "stressed phenotype" in the abstract and results to avoid such claim. The stressed phenotype may be induced by heightened functional activity in the cells, potentially indicating higher cellular activity.

(4) The study is about improving ageing but no physiological data from the study demonstrated such a claim with the exception of the testes histology which was not properly analyzed and was not even significantly different between the groups.

The primary objective of this study is to elucidate the effects of 17α-Estradiol on the endocrine system in the aging hypothalamus; exploring anti-aging effects is not the main focus. From the characteristics of the aging hypothalamus, we know that down-regulated GnRH and testosterone levels, along with elevated mTOR signaling, are indicators of aging in these organs from previous publications (PMID: 37886966, PMID: 37048056, PMID: 22884327). The contrasting signaling networks related to metabolism and synaptic processes significantly differentiate young and aging hypothalami, and 17α-Estradiol helps rebalance these networks, suggesting its potential anti-aging effects.

(5) Overall, the study remains descriptive with no physiological data to demonstrate that any of the effects on hypothalamic gene expression are related to metabolic, synaptic, or other functions.

The study focuses on investigating cellular responses and endocrine changes in the aging hypothalamus induced by 17α-estradiol, utilizing single-nucleus RNA sequencing (snRNA-seq) and a novel data mining methodology to analyze various neuron subtypes. It is important to note that this study does not mainly aim to explore the anti-aging effects. Consequently, we have revised the claim in the abstract from “the effects of 17α-estradiol in anti-aging in neurons” to “the effects of 17α-estradiol on aging neurons.” We observed that the lower overall metabolism and increased expression levels of cellular processes in the synapses align with findings previously reported regarding 17α-estradiol. To address the lack of physiological data and the challenges in measuring multiple endocrine factors due to their volatile nature, we employed several bidirectional Mendelian analyses of various genome-wide association study (GWAS) data related to these serum endocrine factors to identify their mutual causal effects.

Reviewing Editor Comment:

Based on the Public Reviews and Recommendations for Authors, the Reviewers strongly recommend that revisions include an experimental demonstration of the physiological effects of the treatment on ageing in rats as well as the CRH-senescence link. Additional analysis of the glia would greatly strengthen the study, as would inclusion of females and young male controls. The important point was also raised that the work linking 17a-estradiol was performed in mice, and the link with lifespan in rats is not known. Discussion of this point is recommended.

We thank the reviewers for their constructive feedback. Regarding the recommendations in the Public Reviews and Recommendations for Authors:

a)  Physiological effects & CRH-senescence link:

We acknowledge that 17α-estradiol has been reported to extend lifespan in male rats, consistent with findings in male mice (PMID: 33289482). This point has now been noted in the Introduction. We regret that further experimental validation of the treatment's physiological effects on aging in rats was beyond the scope of this study.

b) Phenotype terminology:

In response to concerns about the "senescent" characterization of CRH neurons, we have revised this terminology to "stressed phenotype" throughout the abstract and results. While we were unable to conduct additional experiments to confirm senescence markers, this revised description better reflects the heightened cellular activity observed (as evidenced by elevated serum GnRH and testosterone levels), without implying confirmed senescence.

c) Glial cell analysis:

To address questions about glial cell function during treatment, we have added new enrichment analysis data of differentially expressed genes in microglia and astrocytes from young (Y), old (O), and old treated (O.T) groups in Figure 2—figure supplement 3. This analysis reveals that microglia exhibit contrasting synaptic-related cellular processes compared to total neurons.

d) Female and young controls:

We sincerely apologize for the absence of female subjects and young male controls in the current study. The reviewers' suggestion to examine the male-specific effects of 17α-estradiol using female controls represents an excellent direction for future research, which we plan to pursue in upcoming studies.

Reviewer #2 (Recommendations For The Authors):

General comments:

(1) The manuscript is very hard to read. Proofreading and editing by software or a professional seems necessary. The words "enhanced", "extensive" etc. are not always used in the right way.

Thanks for the suggestion. We have revised the proofreading and editing. The words "enhanced" and "extensive" were also revised in most sentences.

(2) The numbers of animals and samples are not well explained. Is it 9 rats overall or per group? If there are 8 testes samples per group, should we assume that there were 4 rats per group? The pooling of the hypothalamic how was it done? Were all the hypothalamic from each group pooled together? A small table with the animals per group and the samples would help.

We appreciate your reminder regarding the initial mistake in our manuscript preparation. In the preliminary submission, we reported 9 rats based solely on sequencing data and data mining. The revised version (v1) now includes additional experimental data, with an effective total of 12 animals (4 per group). Unfortunately, we overlooked updating this information in the v1 submission. We have since added detailed information in the Materials and Methods sections: Animals, Treatment and Tissues, and snRNA-seq Data Processing, Batch Effect Correction, and Cell Subset Annotation.

(3) The Clustering is wrong. There are genes in there that do not fall into any of the 3 categories: Neurotransmitters, Receptors, Hormones.

We acknowledge the error in gene clustering and have implemented the following corrections:

(a) The description has been updated to state: 'Vast majority of these subtypes were clustered by neuropeptides, hormones, and their receptors among all neurons.'

(b) Genes not belonging to these three categories have been substantially removed.

(c) The neuropeptide category (now including several growth hormones) has been expanded to 104 genes, while their corresponding receptors (including several sex hormone receptors) now comprise 105 genes.

(4) The coloring of groups in the graphs is inconsistent. It must be more homogeneous to make it easier to identify.

We have changed the colors of groups in Fig. 1D to make the color of cell clusters consistent in Fig. 1A-D.

(5) The groups c1-c4 are not well explained. How did the authors come up with these?

We have added more descriptions of c1-c4 in materials and methods in the new version.

(6) In most cases it's not clear if the authors are talking about cell numbers that express a certain mRNA, the level of expression of a certain mRNA, or both. They need to do a better job using more precise descriptions instead of using general terms such as "signatures", "expression profiles", "affected neurons" etc. It is very hard to understand if the number of neurons is compared between the groups or the gene expression.

We have changed the "signatures" to "gene signatures" to make it more accurate in meaning. The "affected neurons" were also changed to "sensitive neurons". But sorry that we were not able to find better alternatives to the "expression profiles".

(7) Sometimes there are claims made without justification or a reference. For example, the claim about the senescence of CRH neurons due to the upregulation of mitochondrial genes and downregulation of adherence junction genes (lines 326-328) should be supported by a reference or own findings.

The "senescence" here is not appropriate. We have changed it to "stressed phenotype" or "aberrant changes" in abstract and results.

(8) Young males treated with Estradiol as a control group is necessary and it is missing.

Your suggestion is appreciated; however, the treatment duration for aged mice (O.T) was set at 6 months, while the young mice were only 4 months old. This disparity makes it challenging to align treatment timelines for the young animals. The primary aim of this study is to investigate the perturbation of 17α-estradiol on the aging process, and any distinct effects due to age effect observed in young males might complicate our understanding of its role in aged males, though similar endocrine effects may exist in the young animals. Long-term treatment of hormone may exert more developmental effects on the young than the old. Therefore, we made the decision to exclude the young samples in our initial study design. We apologize for any confusion this may have caused.

Specific Comments:

Line 28: "elevated stresses and decreased synaptic activity": Please make this clearer. Can't claim changes in synaptic activity by gene expression.

We have changed it to "the expression level of pathways involved in synapse"

Line 32: "increased Oxytocin": serum Oxytocin.

We have added the “serum”.

Line 52 - 54: Any studies from rats?

Thanks. In rats there is also reported that 17α-estradiol has similar metabolic roles as that in mice (PMID: 33289482) and we have added it to the refences. It’s very useful for this manuscript.

Line 62 - 65: It wasn't investigated thoroughly in this paper so why was it suggested in the introduction?

We have deleted this sentence as being suggested.

Line 70: "synaptic activity" Same as line 28.

We have changed it to "pathways involved in synaptic activity".

Line 79: Why were aged rats caged alone and young by two? Could that introduce hypothalamic gene expression effects?

The young males were bred together in peace. But the aged males will fight and should be kept alone.

Lines 78, 99, 109-110: It is not clear how many animals per group were used and how many samples per group were used separately and/or grouped. Please be more specific.

We have added these information to Materials and methods/Animals, treatment and tissues and Materials and methods/snRNA-seq data processing, batch effect correction, and cell subset annotation.

Line 205: "in O" please add "versus young.".

We have changed accordingly.

Line 207: replace "were" with "was"

We have alternatively changed the "proportion" to "proportions".

Line 208: replace "that" with "compared to" and after "in O.T." add "compared to?"

We have changed accordingly.

Line 223: "O.T." compared to what? Figure?

We have changed it accordingly.

Line 227: Figure?

We have added (Figure 1E) accordingly.

Line 229: "synaptic activity" Same as line 28.

We have revised it.

Line 235: "synaptic activity" and "neuropeptide secretion" Same as line 28.

We have revised it.

Line 256:" interfered" please revise.

We changed to "exerted".

Line 263: "on the contrary" please revise.

We have changed "on the contrary" to "opposite".

Line 270: "conversed" did you mean "conserved"?

We have changed "conversed" to "inversed".

Line 296-298: Please explain. Why would these be side effects?

It’s hard to explain, therefore, we deleted the words "side effects".

Line 308: "synaptic activity" Same as line 28.

We have changed it to "expression levels of synapse-related cellular processes".

Line 314: "and sex hormone secretion and signaling"Isn't this expected?

Yes, it is expected. We have added it to the sentence "and, as expected, sex hormone secretion and signaling".

Line 325-328: Why is this senescence? Reference?

We have added “potent” to it.

Line 360-361: This doesn't show elevated synaptic activity.

"elevated synaptic activity" was changed to "The elevated expression of synapse-related pathways"

Line 363-364: "Unfortunately" is not a scientific expression and show bias.

We have changed it to "Notably".

Line 376: Similar as above.

Yes, we have change it to "in contrast".

Lines 382-385: This is speculation. Please move to discussion.

Sorry for that. We think the causal effects derived from MR result is evidence. As such, we have not changed it.

Line 389: Please revise "hormone expressing".

We have changed it accordingly.

Line 401: Isn't this effect expected due to feedback inhibition of the biochemical pathway? Please comment.

The binding capability of 17alpha-estradiol to estrogen receptors and its role in transcriptional activation remain core questions surrounded by controversy. Earlier studies suggest that 17alpha-estradiol exhibits at least 200 times less activity than 17beta-estradiol (PMID: 2249627, PMID: 16024755). However, recent data indicate that 17alpha-estradiol shows comparable genomic binding and transcriptional activation through estrogen receptor α (Esr1) to that of 17beta-estradiol (PMID: 33289482). Additionally, there is evidence that 17alpha-estradiol has anti-estrogenic effects in rats (PMID: 16042770). These findings imply possible feedback inhibition via estrogen receptors. Furthermore, 17alpha-estradiol likely differs from 17beta-estradiol due to its unique metabolic consequences and its potential to slow aging in males, an effect not attributed to 17beta-estradiol. For instance, neurons are also targets of 17alpha-estradiol, with Esr1 not being the sole target (PMID: 38776045). Intriguingly, neurons expressing Ar and Esr1 ranked among the top 20 most perturbed receptor subtypes during aging (O vs Y), but were no longer ranked in this group following treatment (O.T vs Y and O.T vs O comparisons). This indicates that 17α-estradiol administration attenuated age-associated perturbation in these neuronal subtypes, which may be a consequence of potential feedback (Figure 3D). Nevertheless, the precise effective targets of 17alpha-estradiol are still unresolved.

Line 409: This conclusion cannot be made because the effect is not statistically significant. Can say "trend" etc.

Thanks for the recommendation. We have added "potential" in front of the conclusion.

Line 426: "suggesting" please revise.

sorry, it’s a verb.

Lines 426-428: This is speculation. Please move to discussion.

The elevated GnRH levels in O.T., observed through EIA analysis, suggest a deduction regarding the direct causal effects of 17alpha-estradiol on various endocrine factors related to feeding, energy homeostasis, reproduction, osmotic regulation, stress response, and neuronal plasticity through MR analysis. Thus, we have not amended our position. We apologize for any confusion.

Lines 431-432: improved compared to what?

The statement have been revised as " The most striking role of 17α-estradiol treatment revealed in this study showed that HPG axis was substantially improved in the levels of serum Gnrh and testosterone".

Line 435: " Estrogen Receptor Antagonists". Please revise.

Thanks for the recommendation. We have changed it to "estrogen receptor antagonists".

Line 438" "Secrete". Please revise

Sorry, it is "secret".

Lines 439-449: None of this has been demonstrated. Please remove these conclusions.

We appreciate the reviewer's scrutiny regarding lines 439-449. While these statements should not be interpreted as definitive conclusions from our current data, we propose they serve as clinically relevant discussion points worthy of exploration. Our findings demonstrate 17α-estradiol's role in modulating testosterone levels in aged males. This mechanistic insight warrants consideration of its therapeutic potential for age-related hypogonadism - a hypothesis we believe merits discussion given the compound's specific endocrine effects.

Lines 450-457: No females were included in this study. Why? Also, why is this discussed? It is relevant but doesn't belong in this manuscript since it was not studied here.

Testosterone levels are crucial for male health, while estradiol levels are essential for the health and fertility of females. Previous studies have demonstrated that 17α-estradiol does not contribute to lifespan extension in females. Given the effects of 17α-estradiol on males—specifically, its role in promoting testosterone and reducing estradiol levels—we believe it is important to discuss the potential sex-biased effects of 17α-estradiol, as this could inform future investigations. We have refined this section to clarify that these points represent mechanistic hypotheses derived from our male data and existing literature, not conclusions about unstudied female physiology. This framing maintains the discussion's scientific value while respecting the study's scope.

Lines 458-459: This was not demonstrated in this article. Please remove.

We have restricted the claim to "expression level of energy metabolism in hypothalamic neurons".

Line 464: "Promoted lifespan extension" Not demonstrated. Please remove.

At the end of the sentence it was revised as "which may be a contributing factor in promoting lifespan extension".

Line 466: "Showed" No.

The whole sentence was deleted in the new version.

Line 483: "the sex-based effects". Not studied here.

Since the changes in testosterone levels are significant in this dataset and this hormone has a sex-biased nature, we find it worthwhile to suggest this as a topic for future investigation. We have added "which needs further verification in the future" at the end of this sentence.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review): 

In this manuscript, Dillard and colleagues integrate cross-species genomic data with a systems approach to identify potential driver genes underlying human GWAS loci and establish the cell type(s) within which these genes act and potentially drive disease. Specifically, they utilize a large single-cell RNA-seq (scRNA-seq) dataset from an osteogenic cell culture model - bone marrow-derived stromal cells cultured under osteogenic conditions (BMSC-OBs) - from a genetically diverse outbred mouse population called the Diversity Outbred (DO) stock to discover network driver genes that likely underlie human bone mineral density (BMD) GWAS loci. The DO mice segregate over 40M single nucleotide variants, many of which affect gene expression levels, therefore making this an ideal population for systems genetic and co-expression analyses. The current study builds on previously published work from the same group that used co-expression analysis to identify co-expressed "modules" of genes that were enriched for BMD GWAS associations. In this study, the authors utilize a much larger scRNA-seq dataset from 80 DO BMSC-OBs, infer co-expression-based and Bayesian networks for each identified mesenchymal cell type, focused on networks with dynamic expression trajectories that are most likely driving differentiation of BMSC-OBs, and then prioritized genes ("differentiation driver genes" or DDGs) in these osteogenic differentiation networks that had known expression or splicing QTLs (eQTL/sQTLs) in any GTEx tissue that colocalized with human BMD GWAS loci. The systems analysis is impressive, the experimental methods are described in detail, and the experiments appear to be carefully done. The computational analysis of the single-cell data is comprehensive and thorough, and the evidence presented in support of the identified DDGs, including Tpx2 and Fgfrl1, is for the most part convincing. Some limitations in the data resources and methods hamper enthusiasm somewhat and are discussed below. Overall, while this study will no doubt be valuable to the BMD community, the cross-species data integration and analytical framework may be more valuable and generally applicable to the study of other diseases, especially for diseases with robust human GWAS data but for which robust human genomic data in relevant cell types is lacking. 

Specific strengths of the study include the large scRNA-seq dataset on BMSC-OBs from 80 DO mice, the clustering analysis to identify specific cell types and sub-types, the comparison of cell type frequencies across the DO mice, and the CELLECT analysis to prioritize cell clusters that are enriched for BMD heritability (Figure 1). The network analysis pipeline outlined in Figure 2 is also a strength, as is the pseudotime trajectory analysis (results in Figure 3). One weakness involves the focus on genes that were previously identified as having an eQTL or sQTL in any GTEx tissue. The authors rightly point out that the GTEx database does not contain data for bone tissue, but the reason that eQTLs can be shared across many tissues - this assumption is valid for many cis-eQTLs, but it could also exclude many genes as potential DDGs with effects that are specific to bone/osteoblasts. Indeed, the authors show that important BMD driver genes have cell-type-specific eQTLs. Furthermore, the mesenchymal cell type-specific co-expression analysis by iterative WGCNA identified an average of 76 co-expression modules per cell cluster (range 26-153). Based on the limited number of genes that are detected as expressed in a given cell due to sparse per-cell read depth (400-6200 reads/cell) and dropouts, it's hard to believe that as many as 153 co-expression modules could be distinguished within any cell cluster. I would suspect some degree of model overfitting here and would expect that many/most of these identified modules have very few gene members, but the methods list a minimum module size of 20 genes. How do the numbers of modules identified in this study compare to other published scRNA-seq studies that use iterative WGCNA? 

In the section "Identification of differentiation driver genes (DDGs)", the authors identified 408 significant DDGs and found that 49 (12%) were reported by the International Mouse Knockout [sic] Consortium (IMPC) as having a significant effect on whole-body BMD when knocked out in mice. Is this enrichment significant? E.g., what is the background percentage of IMPC gene knockouts that show an effect on whole-body BMD? Similarly, they found that 21 of the 408 DDGs were genes that have BMD GWAS associations that colocalize with GTEx eQTLs/sQTLs. Given that there are > 1,000 BMD GWAS associations, is this enrichment (21/408) significant? Recommend performing a hypergeometric test to provide statistical context to the reported overlaps here. 

We thank the reviewer for their constructive feedback and thoughtful questions. In regards to the iterativeWGCNA, a larger number of modules is sometimes an outcome of the analysis, as reported in the iterativeWGCNA preprint (Greenfest-Allen et al., 2017). While we did not make a comparison to other works leveraging this tool for scRNA-seq, it has been used broadly across other published studies, such as PMID: 39640571, 40075303, 33677398, 33653874. While model overfitting, as you mention, may be a cause for more modules, our Bayesian network analysis we perform after iterativeWGCNA highlights smaller aspects of coexpression modules, as opposed to focusing on the entirety of any given module.

We did not perform enrichment or statistical tests as our goal was to simply highlight attributes or unique features of these genes for additional context.

Reviewer #2 (Public review): 

Summary: 

In this manuscript, Farber and colleagues have performed single-cell RNAseq analysis on bone marrow-derived stem cells from DO Mice. By performing network analysis, they look for driver genes that are associated with bone mineral density GWAS associations. They identify two genes as potential candidates to showcase the utility of this approach. 

Strengths: 

The study is very thorough and the approach is innovative and exciting. The manuscript contains some interesting data relating to how cell differentiation is occurring and the effects of genetics on this process. The section looking for genes with eQTLs that differ across the differentiation trajectory (Figure 4) was particularly exciting. 

Weaknesses: 

The manuscript is in parts hard to read due to the use of acronyms and there are some questions about data analysis that need to be addressed. 

We thank the reviewer for their feedback and shared enthusiasm for our work. We tried to minimize the use of technical acronyms as much as we could without compromising readability. Additionally, we addressed questions regarding aspects of data analysis. 

Reviewer #1 (Recommendations for the authors):

(1) For increased transparency and to allow reproducibility, it would be necessary for the scripts used in the analysis to be shared along with the publication of the preprint. Also, where feasible, sharing the processed data in addition to the raw data would allow the community greater access to the results and be highly beneficial. 

Thank you for this suggestion. The raw data will be available via GEO accession codes listed in the data availability statement. We will make available scripts for some analyses on our Github (https://github.com/Farber-Lab/DO80_project) and processed scRNA-seq data in a Seurat object (.rds) on Zenodo (https://zenodo.org/records/15299631)

(2) Lines 55-76: I think the summary of previous work here is too long. I understand that they would like to cover what has been done previously, but this seems like overkill. 

Good suggestion. We have streamlined some of the summary of our previous work.

(3) Did the authors try to map QTL for cell-type proportion differences in their BMSC-OBs? While 80 samples certainly limit mapping power, the data shown in Figs 4C/D suggest that you might identify a large-effect modifier of LMP/OB1 proportions. 

We did try to map QTL for cell type proportion differences, but no significant associations were identified. 

(4) Methods question: Does the read alignment method used in your analysis account for SNPs/indels that segregate among the DO/CC founder strains? If not, the authors may wish to include this in their discussion of study limitations and speculate on how unmapped reads could affect expression results. 

The read alignment method we used does not account for SNPs/indels from the DO founder strains that fall in RNA transcripts captured in the scRNA-seq data. We have included this as a limitation in our discussion (line 422-424). 

(5) Much of the discussion reads as an overview of the methods, while a discussion of the results and their context to the existing BMD literature is relatively lacking in comparison.

We have added additional explanation of the results and context to the discussion (line 381-382, 396-407). 

(6) Figure 1E and lines 146-149: Adjusted p values should be reported in the figure and accompanying text instead of switching between unadjusted and adjusted p values. 

We updated Figure 1e to portray adjusted p-values, listed the adjusted p-values in legend of Figure 1e, and listed them in the main text (line 153-154).

(7) Why do the authors bring the IMPC KO gene list into the analysis so late? This seems like a highly relevant data resource (moreso than the GTEx eQTLs/sQTLs) that could have been used much earlier to help identify DDGs. 

Given that our scRNA-seq data is also from mice, we did choose to integrate information from the IMPC to highlight supplemental features of genes in networks (i.e., genes that have an experimentally-tested and significant effect on BMD in mice). However, our primary goal was to inform human GWAS and leverage our previous work in which we identified colocalizations between human BMD GWAS and eQTL/sQTL in a human GTEx tissue, which is why this information was used to guide our network analysis.

(8) Does Fgfrl1 and/or Tpx2 have a cis-eQTL in your BMSC-OB scRNA-seq dataset? 

We did not identify cis-eQTL effects for Fgfrl1 and Tpx2.

(9) Figure 4B-C: These eQTLs may be real, but based on the diplotype patterns in Figure 4C, I suspect they are artifacts of low mapping power that are driven by rare genotype classes with one or two samples having outlier expression results. For example, if you look at the results in Fig 4C for S100a1 expression, the genotype classes with the highest/lowest expression have lower sample numbers. In the case of Pkm eQTL showing a PWK-low effect, the PWK genome has many SNPs that differ from the reference genome in the 3' UTR of this gene, and I wonder if reads overlapping these SNPs are not aligning correctly (see point 4 above) and resulting (falsely) in lower expression values for samples with a PWK haplotype. 

As mentioned above, our alignment method did not consider DO founder genetic variation that is specifically located in the 3’ end of RNA transcripts in the scRNA-seq data. We have included this as a limitation in our discussion (line 422-424).

In future studies, we intend to include larger populations of mice to potentially overcome, as you mention, any artifacts that may be attributable to low statistical power, rare genotype classes, or outlier expression.

Reviewer #2 (Recommendations for the authors):

Major Points 

(1) The authors hypothesize "that many genes impacting BMD do so by influencing osteogenic differentiation or possibly bone marrow adipogenic differentiation". However, cell type itself does not correlate with any bone trait. Does this indicate that the hypothesis is not entirely correct, as genes that drive these phenotypes would not be enriched in one particular cell type? The authors have previously identified "high-priority target genes". So, are there any cell types that are enriched for these target genes? If not, this would indicate that all these genes are more ubiquitously expressed and this is probably why they would have a greater effect on the overall bone traits. Furthermore, are the 73 eGenes (so genes with eQTLs in a particular cell type that change around cell type boundaries) or the DDGs (Table 1) enriched for these high-priority target genes? 

The bone traits measured in the DO mice are complex and impacted by many factors, including the differentiation propensity and abundance of certain cell types, both within and outside of bone. Though we did not identify correlations between cell type abundance and the bone traits we measured, we tailored our investigations to focus on cellular differentiation using the scRNA-seq data. However, future studies would need to be performed to investigate any connections between cellular differentiation, cell type abundance, and bone traits.

We did not perform enrichment analyses of either the target genes identified from our other work or eGenes identified here, but instead used the target gene list to center our network analysis and the eGenes to showcase the utility of the DO mouse population.

(2) The readability of the paper could be improved by minimising the use of acronyms and there are several instances of confusing wording throughout the paper. In many cases, this can be solved by re-organising sentences and adding a bit more detail. For example, it was unclear how you arrived at Fgfrl1 or Tpx2.

One of the goals of our study was to identify genes that have (to our knowledge) little to no known connection to BMD. We chose to highlight Fgfrl1 and Tpx2 because there is minimal literature characterizing these genes in the context of bone, which we speak to in the results (line 296-297). Additionally, we prioritized these genes in our previous work and they were identified in this study by using our network analyses using the scRNA-seq data, which we mention in the results (line 276-279).

(3) Technical aspects of the assay. In Figure 1d you show that the cell populations vary considerably between different DO mice. It would be useful to give some sense of the technical variance of this assay given that the assay involves culturing the cells in an exogenous environment. This could take the form of tests between mice within the same inbred strain, or even between different legs of the same DO mice to show that results are technically very consistent. It might also be prudent to identify that this is a potential limitation of the approach as in vitro culturing has the potential to substantially change the cell populations that are present. 

We agree that in vitro culturing, in addition to the preparation of single cells for scRNA-seq, are unavoidable sources of technical variation in this study. However, the total number of cells contributed by each of the 80 DO mice after data processing does not appear to be skewed and the distribution appears normal (see added figures, now included as Supplemental Figure 3). Therefore, technical variation is at least consistent across all samples. Nevertheless, we have mentioned the potential for technical variation artifacts in our study in the discussion (line 414-416).

(4) Need for permutation testing. "We identified 563 genes regulated by a significant eQTL in specific cell types. In total, 73 genes with eQTLs were also tradeSeq-identified genes in one or more cell type boundaries". These types of statements are fine but they need to be backed up with permutation testing to show that this level of enrichment is greater than one would expect by chance. 

We did not perform enrichment tests as our only goal was to 1. determine if eQTL could be resolved in the DO mouse population using our scRNA-seq data and 2. predict in what cell type the associated eQTL and associated eGene may have an effect.

(5) The main novelty of the paper seems to be that you have used single-cell RNA seq (given that you appear to have already detailed the candidates at the end). I don't think this makes the paper less interesting, but I think you need to reframe the paper more about the approach, and not the specific results. How you landed on these candidates is also not clear. So the paper might be improved by more robustly establishing the workflow and providing guidelines for how studies like this should be conducted in the future. 

We sought to not only devise a rigorous approach to analyze our single cell data, but also showcase the utility of the approach in practice by highlighting targets for future research (i.e., Fgfrl1 and Tpx2).

Our goal was to identify novel genes and we landed on these candidate genes (Fgfrl1 and Tpx2) because they had substantial data supporting their causality and they have yet to be fully characterized in the context of bone and BMD (line 295-297).

In regards to establishing the workflow, we have included rationale for specific aspects of our approach throughout the paper. For example, Figure 2 itemizes each step of our network analysis and we explain why each step is utilized throughout various parts results (e.g., lines 168-170, 179-181, 191-193, 202-203, 257-260, 276-277).

We have added a statement advocating for large-scale scRNA-seq from genetically diverse samples and network analyses for future studies (line 436-438).

Minor Points 

(1) In the summary you use the word "trajectory". Trajectories for what? I assume the transition between cell types, but this is not clear. 

We added text to clarify the use of trajectory in the summary (line 34).

(2) This sentence: "By 60 identifying networks enriched for genes implicated in GWAS we predicted putatively causal genes 61 for hundreds of BMD associations based on their membership in enriched modules." is also not clear. Do you mean: we predicted putatively causal genes by identifying clusters of co-expressed genes that were enriched for GWAS genes?" It is not clear how you identify the causal gene in the network. Is this just based on the hub gene? 

The aforementioned sentence has since been removed to streamline the introduction, as suggested by Reviewer 1.

In regards to causal gene identification, it is not based on whether it is hub gene. We prioritized a DDG (and their associated networks) if it was a causal gene that we identified in our previous work as having eQTL/sQTL in a GTEx tissue that colocalizes with human BMD GWAS.

(3) Figure 3C. This is good but the labels are quite small. Would be good to make all the font sizes larger. 

We have enlarged Figure 3C.

(4) Line 341 in the Discussion should be "pseudotemporal". 

We have edited “temporal” to “pseduotemporal”.

Author response:

The following is the authors’ response to the original reviews

Reviewer #1 (Public review):

This is a well-designed and very interesting study examining the impact of imprecise feedback on outcomes in decision-making. I think this is an important addition to the literature, and the results here, which provide a computational account of several decision-making biases, are insightful and interesting.

We thank the reviewer for highlighting the strengths of this work.

I do not believe I have substantive concerns related to the actual results presented; my concerns are more related to the framing of some of the work. My main concern is regarding the assertion that the results prove that non-normative and non-Bayesian learning is taking place. I agree with the authors that their results demonstrate that people will make decisions in ways that demonstrate deviations from what would be optimal for maximizing reward in their task under a strict application of Bayes' rule. I also agree that they have built reinforcement learning models that do a good job of accounting for the observed behavior. However, the Bayesian models included are rather simple, per the author's descriptions, applications of Bayes' rule with either fixed or learned credibility for the feedback agents. In contrast, several versions of the RL models are used, each modified to account for different possible biases. However, more complex Bayes-based models exist, notably active inference, but even the hierarchical Gaussian filter. These formalisms are able to accommodate more complex behavior, such as affect and habits, which might make them more competitive with RL models. I think it is entirely fair to say that these results demonstrate deviations from an idealized and strict Bayesian context; however, the equivalence here of Bayesian and normative is, I think, misleading or at least requires better justification/explanation. This is because a great deal of work has been done to show that Bayes optimal models can generate behavior or other outcomes that are clearly not optimal to an observer within a given context (consider hallucinations for example), but which make sense in the context of how the model is constructed as well as the priors and desired states the model is given.

As such, I would recommend that the language be adjusted to carefully define what is meant by normative and Bayesian and to recognize that work that is clearly Bayesian could potentially still be competitive with RL models if implemented to model this task. An even better approach would be to directly use one of these more complex modelling approaches, such as active inference, as the comparator to the RL models, though I would understand if the authors would want this to be a subject for future work.

We thank the reviewer for raising this crucial and insightful point regarding the framing of our results and the definitions of 'normative' and 'Bayesian' learning. Our primary aim in this work was to characterize specific behavioral signatures that demonstrate deviations from predictions generated by a strict, idealized Bayesian framework when learning from disinformation (which we term “biases”). We deliberately employed relatively simple Bayesian models as benchmarks to highlight these specific biases. We fully agree that more sophisticated Bayes-based models (as mentioned by the reviewer, or others) could potentially offer alternative mechanistic explanations for participant behavior. However, we currently do not have a strong notion about which Bayesian models can encompass our findings, and hence, we leave this important question for future work.

To enhance clarity within the current manuscript we now avoided the use of the term “normative” to refer to our Bayesian models, using the term “ideal” instead. We also define more clearly what exactly we mean by that notion when the idea model is described:

“This model is based on an idealized assumptions that during the feedback stage of each trial, the value of the chosen bandit is updated (based on feedback valence and credibility) according to Bayes rule reflecting perfect adherence to the instructed task structure (i.e., how true outcomes and feedback are generated).”

Moreover, we have added a few sentences in the discussion commenting on how more complex Bayesian models might account for our empirical findings:

“However, as hypothesized, when facing potential disinformation, we also find that individuals exhibit several important biases i.e., deviations from strictly idealized Bayesian strategies. Future studies should explore if and under what assumptions, about the task’s generative structure and/or learner’s priors and objectives, more complex Bayesian models (e.g., active inference (58)) might account for our empirical findings.”

Abstract:

The abstract is lacking in some detail about the experiments done, but this may be a limitation of the required word count. If word count is not an issue, I would recommend adding details of the experiments done and the results.

We thank the reviewer for their valuable suggestion. We have now included more details about the experiment in the abstract:

“In two experiments, participants completed a two-armed bandit task, where they repeatedly chose between two lotteries and received outcome-feedback from sources of varying credibility, who occasionally disseminated disinformation by lying about true choice outcome (e.g., reporting non reward when a reward was truly earned or vice versa).”

One comment is that there is an appeal to normative learning patterns, but this suggests that learning patterns have a fixed optimal nature, which may not be true in cases where the purpose of the learning (e.g. to confirm the feeling of safety of being in an in-group) may not be about learning accurately to maximize reward. This can be accommodated in a Bayesian framework by modelling priors and desired outcomes. As such, the central premise that biased learning is inherently non-normative or non-Bayesian, I think, would require more justification. This is true in the introduction as well.

Introduction:

As noted above, the conceptualization of Bayesian learning being equivalent to normative learning, I think requires further justification. Bayesian belief updating can be biased and non-optimal from an observer perspective, while being optimal within the agent doing the updating if the priors/desired outcomes are set up to advantage these "non-optimal" modes of decision making.

We appreciate the reviewer's thoughtful comment regarding the conceptualization of "normative" and "Bayesian" learning. We fully agree that the definition of "normative" is nuanced and can indeed depend on whether one considers reward-maximization or the underlying principles of belief updating. As explained above we now restrict our presentation to deviations from “ideal Bayes” learning patterns and we acknowledge the reviewer’s concern in a caveat in our discussion.

Results:

I wonder why the agent was presented before the choice, since the agent is only relevant to the feedback after the choice is made. I wonder if that might have induced any false association between the agent identity and the choice itself. This is by no means a critical point, but it would be interesting to get the authors' thoughts.

We thank the reviewer for raising this interesting point regarding the presentation of the agent before the choice. Our decision to present the agent at this stage was intentional, as our original experimental design aimed to explore the possible effects of "expected source credibility" on participants' choices (e.g., whether knowledge of feedback credibility will affect choice speed and accuracy). However, we found nothing that would be interesting to report.

The finding that positive feedback increases learning is one that has been shown before and depends on valence, as the authors note. They expanded their reinforcement learning model to include valence, but they did not modify the Bayesian model in a similar manner. This lack of a valence or recency effect might also explain the failure of the Bayesian models in the preceding section, where the contrast effect is discussed. It is not unreasonable to imagine that if humans do employ Bayesian reasoning that this reasoning system has had parameters tuned based on the real world, where recency of information does matter; affect has also been shown to be incorporable into Bayesian information processing (see the work by Hesp on affective charge and the large body of work by Ryan Smith). It may be that the Bayesian models chosen here require further complexity to capture the situation, just like some of the biases required updates to the RL models. This complexity, rather than being arbitrary, may be well justified by decision-making in the real world.

Thanks for these additional important ideas which speak more to the notion that more complex Bayesian frameworks may account for biases we report.

The methods mention several symptom scales- it would be interesting to have the results of these and any interesting correlations noted. It is possible that some of the individual variability here could be related to these symptoms, which could introduce precision parameter changes in a Bayesian context and things like reward sensitivity changes in an RL context.

We included these questionnaires for exploratory purposes, with the aim of generating informed hypotheses for future research into individual differences in learning. Given the preliminary nature of these analyses, we believe further research is required about this important topic.

Discussion:

(For discussion, not a specific comment on this paper): One wonders also about participants' beliefs about the experiment or the intent of the experimenters. I have often had participants tell me they were trying to "figure out" a task or find patterns even when this was not part of the experiment. This is not specific to this paper, but it may be relevant in the future to try and model participant beliefs about the experiment especially in the context of disinformation, when they might be primed to try and "figure things out".

We thank the reviewer for this important recommendation. We agree and this point is included in our caveat (cited above) that future research should address what assumptions about the generative task structure can allow Bayesian models to account for our empirical patterns.

As a general comment, in the active inference literature, there has been discussion of state-dependent actions, or "habits", which are learned in order to help agents more rapidly make decisions, based on previous learning. It is also possible that what is being observed is that these habits are at play, and that they represent the cognitive biases. This is likely especially true given, as the authors note, the high cognitive load of the task. It is true that this would mean that full-force Bayesian inference is not being used in each trial, or in each experience an agent might have in the world, but this is likely adaptive on the longer timescale of things, considering resource requirements. I think in this case you could argue that we have a departure from "normative" learning, but that is not necessarily a departure from any possible Bayesian framework, since these biases could potentially be modified by the agent or eschewed in favor of more expensive full-on Bayesian learning when warranted.<br /> Indeed, in their discussion on the strategy of amplifying credible news sources to drown out low-credibility sources, the authors hint at the possibility of longer-term strategies that may produce optimal outcomes in some contexts, but which were not necessarily appropriate to this task. As such, the performance on this task- and the consideration of true departure from Bayesian processing- should be considered in this wider context.

Another thing to consider is that Bayesian inference is occurring, but that priors present going in produce the biases, or these biases arise from another source, for example, factoring in epistemic value over rewards when the actual reward is not large. This again would be covered under an active inference approach, depending on how the priors are tuned. Indeed, given the benefit of social cohesion in an evolutionary perspective, some of these "biases" may be the result of adaptation. For example, it might be better to amplify people's good qualities and minimize their bad qualities in order to make it easier to interact with them; this entails a cost (in this case, not adequately learning from feedback and potentially losing out sometimes), but may fulfill a greater imperative (improved cooperation on things that matter). Given the right priors/desired states, this could still be a Bayes-optimal inference at a social level and, as such, may be ingrained as a habit that requires effort to break at the individual level during a task such as this.

We thank the reviewer for these insightful suggestions speaking further to the point about more complex Bayesian models.

The authors note that this task does not relate to "emotional engagement" or "deep, identity-related issues". While I agree that this is likely mostly true, it is also possible that just being told one is being lied to might elicit an emotional response that could bias responses, even if this is a weak response.

We agree with the reviewer that a task involving performance-based bonuses, and particularly one where participants are explicitly told they are being lied to, might elicit weak emotional response. However, our primary point is that the degree of these responses is expected to be substantially weaker than those typically observed in the broader disinformation literature, which frequently deals with highly salient political, social, or identity-related topics that inherently carry strong emotional and personal ties for participants, leading to much more pronounced affective engagement and potential biases. Our task deliberately avoids such issues thus minimizing the potential for significant emotion-driven biases. We have toned down the discussion accordingly:

“This occurs even when the decision at hand entails minimal emotional engagement or pertinence to deep, identity-related, issues.”

Reviewer #2 (Public review):

This valuable paper studies the problem of learning from feedback given by sources of varying credibility. The solid combination of experiment and computational modeling helps to pin down properties of learning, although some ambiguity remains in the interpretation of results.

Summary:

This paper studies the problem of learning from feedback given by sources of varying credibility. Two banditstyle experiments are conducted in which feedback is provided with uncertainty, but from known sources. Bayesian benchmarks are provided to assess normative facets of learning, and alternative credit assignment models are fit for comparison. Some aspects of normativity appear, in addition to deviations such as asymmetric updating from positive and negative outcomes.

Strengths:

The paper tackles an important topic, with a relatively clean cognitive perspective. The construction of the experiment enables the use of computational modeling. This helps to pinpoint quantitatively the properties of learning and formally evaluate their impact and importance. The analyses are generally sensible, and parameter recovery analyses help to provide some confidence in the model estimation and comparison.

We thank the reviewer for highlighting the strengths of this work.

Weaknesses:

(1) The approach in the paper overlaps somewhat with various papers, such as Diaconescu et al. (2014) and Schulz et al. (forthcoming), which also consider the Bayesian problem of learning and applying source credibility, in terms of theory and experiment. The authors should discuss how these papers are complementary, to better provide an integrative picture for readers.

Diaconescu, A. O., Mathys, C., Weber, L. A., Daunizeau, J., Kasper, L., Lomakina, E. I., ... & Stephan, K. E. (2014). Inferring the intentions of others by hierarchical Bayesian learning. PLoS computational biology, 10(9), e1003810.

Schulz, L., Schulz, E., Bhui, R., & Dayan, P. Mechanisms of Mistrust: A Bayesian Account of Misinformation Learning. https://doi.org/10.31234/osf.io/8egxh

We thank the reviewers for pointing us to this relevant work. We have updated the introduction, mentioning these precedents in the literature and highlighting our specific contributions:

“To address these questions, we adopt a novel approach within the disinformation literature by exploiting a Reinforcement Learning (RL) experimental framework (36). While RL has guided disinformation research in recent years (37–41), our approach is novel in using one of its most popular tasks: the “bandit task”.”

We also explain in the discussion how these papers relate to the current study:

“Unlike previous studies wherein participants had to infer source credibility from experience (30,37,72), we took an explicit-instruction approach, allowing us to precisely assess source-credibility impact on learning, without confounding it with errors in learning about the sources themselves. More broadly, our work connects with prior research on observational learning, which examined how individuals learn from the actions or advice of social partners (72–75). This body of work has demonstrated that individuals integrate learning from their private experiences with learning based on others’ actions or advice—whether by inferring the value others attribute to different options or by mimicking their behavior (57,76). However, our task differs significantly from traditional observational learning. Firstly, our feedback agents interpret outcomes rather than demonstrating or recommending actions (30,37,72).”

(2) It isn't completely clear what the "cross-fitting" procedure accomplishes. Can this be discussed further?

We thank the reviewer for requesting further clarification on the cross-fitting procedure. Our study utilizes two distinct model families: Bayesian models and CA models. The credit assignment parameters from the CA models can be treated as “data/behavioural features” corresponding to how choice feedback affects choice-propensities. The cross fitting-approach allows us in effect to examine whether these propensity features are predicted from our Bayesian models. To the extent they are not, we can conclude empirical behavior is “biased”.

Thus, in our cross-fitting procedure we compare the CA model parameters extracted from participant data (empirical features) with those that would be expected if our Bayesian agents performed the task. Specifically, we first fit participant behavior with our Bayesian models, then simulate this model using the best-fitted parameters and fit those simulations with our CA models. This generates a set of CA parameters that would be predicted if participants behavior is reduced to a Bayesian account. By comparing these predicted Bayesian CA parameters with the actual CA parameters obtained from human participants, the cross-fitting procedure allows us to quantitatively demonstrate that the observed participant parameters are indeed statistically significant deviations from normative Bayesian processing. This provides a robust validation that the biases we identify are not artifacts of the CA model's structure but true departures from normative learning.

We also note that Reviewer 3 suggested an intuitive way to think about the CA parameters—as analogous to logistic regression coefficients in a “sophisticated regression” of choice on (recencyweighted) choice-feedback. We find this suggestion potentially helpful for readers. Under this interpretation, the purpose of the cross-fitting method can be seen simply as estimating the regression coefficients that would be predicted by our Bayesian agents, and comparing those to the empirical coefficients.

In our manuscript we now explain this issues more clearly by explaining how our model is analogous to a logistic regression:

“The probability to choose a bandit (say A over B) in this family of models is a logistic function of the contrast choice-propensities between these two bandits. One interpretation of this model is as a “sophisticated” logistic regression, where the CA parameters take the role of “regression coefficients” corresponding to the change in log odds of repeating the just-taken action in future trials based on the feedback (+/- CA for positive or negative feedback, respectively; the model also includes gradual perseveration which allows for constant log-odd changes that are not affected by choice feedback) . The forgetting rate captures the extent to which the effect of each trial on future choices diminishes with time. The Q-values are thus exponentially decaying sums of logistic choice propensities based on the types of feedback a bandit received.”

We also explain our cross-fitting procedure in more detail:

“To further characterise deviations between behaviour and our Bayesian learning models, we used a “crossfitting” method. Treating CA parameters as data-features of interest (i.e., feedback dependent changes in choice propensity), our goal was to examine if and how empirical features differ from features extracted from simulations of our Bayesian learning models. Towards that goal, we simulated synthetic data based on Bayesian agents (using participants’ best fitting parameters), but fitted these data using the CA-models, obtaining what we term “Bayesian-CA parameters” (Fig. 2d; Methods). A comparison of these BayesianCA parameters, with empirical-CA parameters obtained by fitting CA models to empirical data, allowed us to uncover patterns consistent with, or deviating from, ideal-Bayesian value-based inference. Under the sophisticated logistic-regression interpretation of the CA-model family the cross-fitting method comprises a comparison between empirical regression coefficients (i.e., empirical CA parameters) and regression coefficient based on simulations of Bayesian models (Bayesian CA parameters).”

(3) The Credibility-CA model seems to fit the same as the free-credibility Bayesian model in the first experiment and barely better in the second experiment. Why not use a more standard model comparison metric like the Bayesian Information Criterion (BIC)? Even if there are advantages to the bootstrap method (which should be described if so), the BIC would help for comparability between papers.

We thank the reviewer for this important comment regarding our model comparison approach. We acknowledge that classical information criteria like AIC and BIC are widely used in RL studies. However, we argue our method for model-comparison is superior.

We conducted a model recovery analysis demonstrating a significant limitation of using AIC or BIC for model-comparison in our data. Both these methods are strongly biased in favor of the Bayesian models. Our PBCM method, on the other hand, is both unbiased and more accurate. We believe this is because “off the shelf” methods like AIC and BIC rely on strong assumptions (such as asymptotic sample size and trial-independence) that are not necessarily met in our tasks (Data is finite; Trials in RL tasks depend on previous trials). PBCM avoids such assumptions to obtain comparison criteria specifically tailored to the structure and size of our empirical data. We have now mentioned this fact in the results section of the main text:

“We considered using AIC and BIC, which apply “off-the shelf” penalties for model-complexity. However, these methods do not adapt to features like finite sample size (relying instead on asymptotic assumption) or temporal dependence (as is common in reinforcement learning experiments). In contrast, the parametric bootstrap cross-fitting method replaces these fixed penalties with empirical, data-driven criteria for modelselection. Indeed, model-recovery simulations confirmed that whereas AIC and BIC were heavily biased in favour of the Bayesian models, the bootstrap method provided excellent model-recovery (See Fig. S20).”

We have also included such model recovery in the SI document:

(4) As suggested in the discussion, the updating based on random feedback could be due to the interleaving of trials. If one is used to learning from the source on most trials, the occasional random trial may be hard to resist updating from. The exact interleaving structure should also be clarified (I assume different sources were shown for each bandit pair). This would also relate to work on RL and working memory: Collins, A. G., & Frank, M. J. (2012). How much of reinforcement learning is working memory, not reinforcement learning? A behavioral, computational, and neurogenetic analysis. European Journal of Neuroscience, 35(7), 10241035.

We thank the reviewer for this point. The specific interleaved structure of the agents is described in the main text:

“Each agent provided feedback for 5 trials for each bandit pair (with the agent order interleaved within the bandit pair).”

As well as in the methods section:

“Feedback agents were randomly interleaved across trials subject to the constraint that each agent appeared on 5-trials for each bandit pair.”

We also thank the reviewer for mentioning the relevant work on working memory. We have now added it to our discussion point:

“In our main study, we show that participants revised their beliefs based on entirely non-credible feedback, whereas an ideal Bayesian strategy dictates such feedback should be ignored. This finding resonates with the “continued-influence effect” whereby misleading information continues to influence an individual's beliefs even after it has been retracted (59,60). One possible explanation is that some participants failed to infer that feedback from the 1-star agent was statistically void of information content, essentially random (e.g., the group-level credibility of this agent was estimated by our free-credibility Bayesian model as higher than 50%). Participants were instructed that this feedback would be “a lie” 50% of the time but were not explicitly told that this meant it was random and should therefore be disregarded. Notably, however, there was no corresponding evidence random feedback affected behaviour in our discovery study. It is possible that an individual’s ability to filter out random information might have been limited due to a high cognitive load induced by our main study task, which required participants to track the values of three bandit pairs and juggle between three interleaved feedback agents (whereas in our discovery study each experimental block featured a single bandit pair). Future studies should explore more systematically how the ability to filter random feedback depends on cognitive load (61).”

(5) Why does the choice-repetition regression include "only trials for which the last same-pair trial featured the 3-star agent and in which the context trial featured a different bandit pair"? This could be stated more plainly.

We thank the reviewer for this question. When we previously submitted our manuscript, we thought that finding enhanced credit-assignment for fully credible feedback following potential disinformation from a different context would constitute a striking demonstration of our “contrast effect”. However, upon reexamining this finding we found out we had a coding error (affecting how trials were filtered). We have now rerun and corrected this analysis. We have assessed the contrast effect for both "same-context" trials (where the contextual trial featured the same bandit pair as the learning trial) and "different-context" trials (where the contextual trial featured a different bandit pair). Our re-analysis reveals a selective significant contrast effect in the samecontext condition, but no significant effect in the different-context condition. We have updated the main text to reflect these corrected findings and provide a clearer explanation of the analysis:

“A comparison of empirical and Bayesian credit-assignment parameters revealed a further deviation from ideal Bayesian learning: participants showed an exaggerated credit-assignment for the 3-star agent compared with Bayesian models [Wilcoxon signed-rank test, instructed-credibility Bayesian model (median difference=0.74, z=11.14); free-credibility Bayesian model (median difference=0.62, z=10.71), all p’s<0.001] (Fig. 3a). One explanation for enhanced learning for the 3-star agents is a contrast effect, whereby credible information looms larger against a backdrop of non-credible information. To test this hypothesis, we examined whether the impact of feedback from the 3-star agent is modulated by the credibility of the agent in the trial immediately preceding it. More specifically, we reasoned that the impact of a 3-star agent would be amplified by a “low credibility context” (i.e., when it is preceded by a low credibility trial). In a binomial mixed effects model, we regressed choice-repetition on feedback valence from the last trial featuring the same bandit pair (i.e., the learning trial) and the feedback agent on the trial immediately preceding that last trial (i.e., the contextual credibility; see Methods for model-specification). This analysis included only learning trials featuring the 3-star agent, and context trials featuring the same bandit pair as the learning trial (Fig. 4a). We found that feedback valence interacted with contextual credibility (F(2,2086)=11.47, p<0.001) such that the feedback-effect (from the 3-star agent) decreased as a function of the preceding context-credibility (3-star context vs. 2-star context: b= -0.29, F(1,2086)=4.06, p=0.044; 2star context vs. 1-star context: b=-0.41, t(2086)=-2.94, p=0.003; and 3-star context vs. 1-star context: b=0.69, t(2086)=-4.74, p<0.001) (Fig. 4b). This contrast effect was not predicted by simulations of our main models of interest (Fig. 4c). No effect was found when focussing on contextual trials featuring a bandit pair different than the one in the learning trial (see SI 3.5). Thus, these results support an interpretation that credible feedback exerts a greater impact on participants’ learning when it follows non-credible feedback, in the same learning context.”

We have modified the discussion accordingly as well:

“A striking finding in our study was that for a fully credible feedback agent, credit assignment was exaggerated (i.e., higher than predicted by our Bayesian models). Furthermore, the effect of fully credible feedback on choice was further boosted when it was preceded by a low-credibility context related to current learning. We interpret this in terms of a “contrast effect”, whereby veridical information looms larger against a backdrop of disinformation (21). One upshot is that exaggerated learning might entail a risk of jumping to premature conclusions based on limited credible evidence (e.g., a strong conclusion that a vaccine is produces significant side-effect risks based on weak credible information, following non-credible information about the same vaccine). An intriguing possibility, that could be tested in future studies, is that participants strategically amplify the extent of learning from credible feedback to dilute the impact of learning from noncredible feedback. For example, a person scrolling through a social media feed, encountering copious amounts of disinformation, might amplify the weight they assign to credible feedback in order to dilute effects of ‘fake news’. Ironically, these results also suggest that public campaigns might be more effective when embedding their messages in low-credibility contexts , which may boost their impact.”

And we have included some additional analyses in the SI document:

“3.5 Contrast effects for contexts featuring a different bandit

Given that we observed a contrast effect when both the learning and the immediately preceding "context trial” involved the same pair of bandits, we next investigated whether this effect persisted when the context trial featured a different bandit pair – a situation where the context would be irrelevant to the current learning. Again, we used in a binomial mixed effects model, regressing choice-repetition on feedback valence in the learning trial and the feedback agent in the context trial. This analysis included only learning trials featuring the 3-star agent, and context trials featuring a different bandit pair than the learning trial (Fig. S22a). We found no significant evidence of an interaction between feedback valence and contextual credibility (F(2,2364)=0.21, p=0.81) (Fig. S22b). This null result was consistent with the range of outcomes predicted by our main computational models (Fig. S22c).

We aimed to formally compare the influence of two types of contextual trials: those featuring the same bandit pair as the learning trial versus those featuring a different pair. To achieve this, we extended our mixedeffects model by incorporating a new predictor variable, "CONTEXT_TYPE" which coded whether the contextual trial involved the same bandit pair (coded as -0.5) or a different bandit pair (+0.5) compared to the learning trial. The Wilkinson notation for this expanded mixed-effects model is:

𝑅𝐸𝑃𝐸𝐴𝑇 ~ 𝐶𝑂𝑁𝑇𝐸𝑋𝑇_𝑇𝑌𝑃𝐸 ∗ 𝐹𝐸𝐸𝐷𝐵𝐴𝐶𝐾 ∗ (𝐶𝑂𝑁𝑇𝐸𝑋𝑇_2-star + 𝐶𝑂𝑁𝑇𝐸𝑋𝑇_3-star) + 𝐵𝐸𝑇𝑇𝐸𝑅 + (1|𝑝𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑛𝑡)

This expanded model revealed a significant three-way interaction between feedback valence, contextual credibility, and context type (F(2,4451) = 7.71, p<0.001). Interpreting this interaction, we found a 2-way interaction between context-source and feedback valence when the context was the same (F(2,4451) = 12.03, p<0.001), but not when context was different (F(2,4451) = 0.23, p = 0.79). Further interpreting the double feedback-valence * context-source interaction (for the same context) we obtained the same conclusions as reported in the main text.”

(6) Why apply the "Truth-CA" model and not the Bayesian variant that it was motivated by?

Thanks for this very useful suggestion. We are unsure if we fully understand the question. The Truth-CA model was not motivated by a new Bayesian model. Our Bayesian models were simply used to make the point that participants may partially discriminate between truthful and untruthful feedback (for a given source). This led to the idea that perhaps more credit is assigned for truth (than lie) trials, which is what we found using our Truth-CA model. Note we show that our Bayesian models cannot account for this modulation.

We have now improved our "Truth-CA" model. Previously, our Truth-CA model considered whether feedback on each trial was true or not based on realized latent true outcomes. However, it is possible that the very same feedback would have had an opposite truth-status if the latent true outcome was different (recall true outcomes are stochastic). This injects noise into the trial classification in our previous model. To avoid this, in our new model feedback is modulated by the probability the reported feedback is true (marginalized over stochasticity of true outcome).

We have described this new model in the methods section:

“Additionally, we formulated a “Truth-CA” model, which worked as our Credibility-CA model, but incorporated a free truth-bonus parameter (TB). This parameter modulates the extent of credit assignment for each agent based on the posterior probability of feedback being true (given the credibility of the feedback agent, and the true reward probability of the chosen bandit). The chosen bandit was updated as follows:

𝑄 ← (1 – 𝑓_Q) ∗ 𝑄 + [𝐶𝐴(𝑎𝑔𝑒𝑛𝑡) + 𝑇𝐵 ∗ (𝑃(𝑡𝑟𝑢𝑡ℎ) − 0.5)] ∗ 𝐹

where P(truth) is the posterior probability of the feedback being true in the current trial (for exact calculation of P(truth) see “Methods: Bayesian estimation of posterior belief that feedback is true”).”

All relevant results have been updated accordingly in the main text:

“To formally address whether feedback truthfulness modulates credit assignment, we fitted a new variant of the CA model (the “Truth-CA” model) to the data. This variant works as our Credibility-CA model but incorporated a truth-bonus parameter (TB) which increases the degree of credit assignment for feedback as a function of the experimenter-determined likelihood the feedback is true (which is read from the curves in Fig 6a when x is taken to be the true probability the bandit is rewarding). Specifically, after receiving feedback, the Q-value of the chosen option is updated according to the following rule: 𝑄 ← (1 – 𝑓_Q) ∗ 𝑄 + [𝐶𝐴(𝑎𝑔𝑒𝑛𝑡) + 𝑇𝐵 ∗ (𝑃(𝑡𝑟𝑢𝑡ℎ) − 0.5)] ∗ 𝐹 where 𝑇𝐵 is the free parameter representing the truth bonus, and 𝑃(𝑡𝑟𝑢𝑡ℎ) is the probability the received feedback being true (from the experimenter’s perspective). We acknowledge that this model falls short of providing a mechanistically plausible description of the credit assignment process, because participants have no access to the experimenter’s truthfulness likelihoods (as the true bandit reward probabilities are unknown to them). Nonetheless, we use this ‘oracle model’ as a measurement tool to glean rough estimates for the extent to which credit assignment Is boosted as a function of its truthfulness likelihood. Fitting this Truth-CA model to participants' behaviour revealed a significant positive truth-bonus (mean=0.21, t(203)=3.12, p=0.002), suggesting that participants indeed assign greater weight to feedback that is likely to be true (Fig. 6c; see SI 3.3.1 for detailed ML parameter results). Notably, simulations using our other models (Methods) consistently predicted smaller truth biases (compared to the empirical bias) (Fig. 6d). Moreover, truth bias was still detected even in a more flexible model that allowed for both a positivity bias and truth-bias (see SI 3.7). The upshot is that participants are biased to assign higher credit based on feedback that is more likely to be true in a manner that is inconsistent with out Bayesian models and above and beyond the previously identified positivity biases.“

Finally, the Supplementary Information for the discovery study has also been revised to feature this analysis:

“We next assessed whether participants infer whether the feedback they received on each trial was true or false and adjust their credit assignment based on this inference. We again used the “Truth-CA” model to obtain estimates for the truth bonus (TB), the increase in credit assignment as a function of the posterior probability of feedback being true. As in our main study, the fitted truth bias parameter was significantly positive, indicating that participants assign greater weight to feedback they believe is likely to be true (Fig, S4a; see SI 3.3.1 for detailed ML parameter results). Strikingly, model-simulations (Methods) predicted a lower truth bonus than the one observed in participants (Fig. S4b).”

(7) "Overall, the results from this study support the exact same conclusions (See SI section 1.2) but with one difference. In the discovery study, we found no evidence for learning based on 50%-credibility feedback when examining either the feedback effect on choice repetition or CA in the credibility-CA model (SI 1.2.3)" - this seems like a very salient difference, when the paper reports the feedback effect as a primary finding of interest, though I understand there remains a valence-based difference.

We agree with the reviewer and thank them for this suggestion. We now state explicitly throughout the manuscript that this finding was obtained only in one of our two studies. In the section “Discovery study” of the results we state explicitly this finding was not found in the discovery study:

“However, we found no evidence for learning based on 50%-credibility feedback when examining either the feedback effect on choice repetition or CA in the credibility-CA model (SI 1.2.3).”

We also note that related to another concern from R3 (that perseveration may masquerade as positivity bias) we conducted additional analyses (detailed in SI 3.6.2). These analyses revealed that the observed positivity bias for the 1-star agent in the discovery study falls within the range predicted by simple choice-perseveration. Consequently, we have removed the suggestion that participants still learn from the random agent in the discovery study. Furthermore, we have modified the discussion section to include a possible explanation for this discrepancy between the two studies:

“Notably, however, there was no corresponding evidence random feedback affected behaviour in our discovery study. It is possible that an individual’s ability to filter out random information might have been limited due to a high cognitive load induced by our main study task, which required participants to track the values of three bandit pairs and juggle between three interleaved feedback agents (whereas in our discovery study each experimental block featured a single bandit pair). Future studies should explore more systematically how the ability to filter random feedback depends on cognitive load (61).”

(8) "Participants were instructed that this feedback would be "a lie 50% of the time but were not explicitly told that this meant it was random and should therefore be disregarded." - I agree that this is a possible explanation for updating from the random source. It is a meaningful caveat.

Thank you for this thought. While this can be seen as a caveat—since we don’t know what would have happened with explicit instructions—we also believe it is interesting from another perspective. In many real-life situations, individuals may have all the necessary information to infer that the feedback they receive is uninformative, yet still fail to do so, especially when they are not explicitly told to ignore it.

In future work, we plan to examine how behaviour changes when participants are given more explicit instructions—for example, that the 50%-credibility agent provides purely random feedback.

(9) "Future studies should investigate conditions that enhance an ability to discard disinformation, such as providing explicit instructions to ignore misleading feedback, manipulations that increase the time available for evaluating information, or interventions that strengthen source memory." - there is work on some of this in the misinformation literature that should be cited, such as the "continued influence effect". For example: Johnson, H. M., & Seifert, C. M. (1994). Sources of the continued influence effect: When misinformation in memory affects later inferences. Journal of experimental psychology: Learning, memory, and cognition, 20(6), 1420.

We thank the reviewer for pointing us towards the relevant literature. We have now included citations about the “continued influence effect” of misinformation in the discussion:

“In our main study, we show that participants revised their beliefs based on entirely non-credible feedback, whereas an ideal Bayesian strategy dictates such feedback should be ignored. This finding resonates with the “continued-influence effect” whereby misleading information continues to influence an individual's beliefs even after it has been retracted (59,60).”

(10) Are the authors arguing that choice-confirmation bias may be at play? Work on choice-confirmation bias generally includes counterfactual feedback, which is not present here.

We agree with the reviewer that a definitive test for choice-confirmation bias typically requires counterfactual feedback, which is not present in our current task. In our discussion, we indeed suggest that the positivity bias we observe may stem from a form of choice-confirmation, drawing on the extensive literature on this bias in reinforcement learning (Lefebvre et al., 2017; Palminteri et al., 2017; Palminteri & Lebreton, 2022). However, we fully acknowledge that this link is a hypothesis and that explicitly testing for choice-confirmation bias would necessitate a future study specifically incorporating counterfactual feedback. We have included a clarification of this point in the discussion:

“Previous reinforcement learning studies, report greater credit-assignment based on positive compared to negative feedback, albeit only in the context of veridical feedback (43,44,62). Here, supporting our a-priori hypothesis we show that this positivity bias is amplified for information of low and intermediate credibility (in absolute terms in the discovery study, and relative to the overall extent of CA in both studies) . Of note, previous literature has interpreted enhanced learning for positive outcomes in reinforcement learning as indicative of a confirmation bias (42,44). For example, positive feedback may confirm, to a greater extent than negative feedback one’s choice as superior (e.g., “I chose the better of the two options”). Leveraging the framework of motivated cognition (35), we posited that feedback of uncertain veracity (e.g., low credibility) amplifies this bias by incentivising individuals to self-servingly accept positive feedback as true (because it confers positive, desirable outcomes), and explain away undesirable, choice-disconfirming, negative feedback as false. This could imply an amplified confirmation bias on social media, where content from sources of uncertain credibility, such as unknown or unverified users, is more easily interpreted in a self-serving manner, disproportionately reinforcing existing beliefs (63). In turn, this could contribute to an exacerbation of the negative social outcomes previously linked to confirmation bias such as polarization (64,65), the formation of ‘echo chambers’ (19), and the persistence of misbelief regarding contemporary issues of importance such as vaccination (66,67) and climate change (68–71). We note however, that further studies are required to determine whether positivity bias in our task is indeed a form of confirmation bias.”

Reviewer #3 (Public review):

Summary

This paper investigates how disinformation affects reward learning processes in the context of a two-armed bandit task, where feedback is provided by agents with varying reliability (with lying probability explicitly instructed). They find that people learn more from credible sources, but also deviate systematically from optimal Bayesian learning: They learned from uninformative random feedback, learned more from positive feedback, and updated too quickly from fully credible feedback (especially following low-credibility feedback). Overall, this study highlights how misinformation could distort basic reward learning processes, without appeal to higher-order social constructs like identity.

Strengths

(1) The experimental design is simple and well-controlled; in particular, it isolates basic learning processes by abstracting away from social context.

(2) Modeling and statistics meet or exceed the standards of rigor.

(3) Limitations are acknowledged where appropriate, especially those regarding external validity.

(4) The comparison model, Bayes with biased credibility estimates, is strong; deviations are much more compelling than e.g., a purely optimal model.

(5) The conclusions are interesting, in particular the finding that positivity bias is stronger when learning from less reliable feedback (although I am somewhat uncertain about the validity of this conclusion)

We deeply thank the reviewer for highlighting the strengths of this work.

Weaknesses

(1) Absolute or relative positivity bias?

In my view, the biggest weakness in the paper is that the conclusion of greater positivity bias for lower credible feedback (Figure 5) hinges on the specific way in which positivity bias is defined. Specifically, we only see the effect when normalizing the difference in sensitivity to positive vs. negative feedback by the sum. I appreciate that the authors present both and add the caveat whenever they mention the conclusion (with the crucial exception of the abstract). However, what we really need here is an argument that the relative definition is the right way to define asymmetry....

Unfortunately, my intuition is that the absolute difference is a better measure. I understand that the relative version is common in the RL literature; however previous studies have used standard TD models, whereas the current model updates based on the raw reward. The role of the CA parameter is thus importantly different from a traditional learning rate - in particular, it's more like a logistic regression coefficient (as described below) because it scales the feedback but not the decay. Under this interpretation, a difference in positivity bias across credibility conditions corresponds to a three-way interaction between the exponentially weighted sum of previous feedback of a given type (e.g., positive from the 75% credible agent), feedback positivity, and condition (dummy coded). This interaction corresponds to the nonnormalized, absolute difference.

Importantly, I'm not terribly confident in this argument, but it does suggest that we need a compelling argument for the relative definition.

We thank the reviewer for raising this important point about the definition of positivity bias, and for their thoughtful discussion on the absolute versus relative measures. We believe that the relative valence bias offers a distinct and valuable perspective on positivity bias. Conceptually, this measure describes positivity bias in a manner akin to a “percentage difference” relative to the overall level of learning which allows us to control for the overall decreases in the overall amount of credit assignment as feedback becomes less credible. We are unsure if one measure is better or more correct than the other and we believe that reporting both measures enriches the understanding of positivity bias and allows for a more comprehensive characterization of this phenomenon (as long as these measures are interpreted carefully). We have stated the significance of the relative measure in the results section:

“Following previous research, we quantified positivity bias in 2 ways: 1) as the absolute difference between credit-assignment based on positive or negative feedback, and 2) as the same difference but relative to the overall extent of learning. We note that the second, relative, definition, is more akin to “percentage change” measurements providing a control for the overall lower levels of credit-assignment for less credible agent.”

We also wish to point out that in our discovery study we had some evidence for amplification of positivity bias in absolute sense.

(2) Positivity bias or perseveration?

A key challenge in interpreting many of the results is dissociating perseveration from other learning biases. In particular, a positivity bias (Figure 5) and perseveration will both predict a stronger correlation between positive feedback and future choice. Crucially, the authors do include a perseveration term, so one would hope that perseveration effects have been controlled for and that the CA parameters reflect true positivity biases. However, with finite data, we cannot be sure that the variance will be correctly allocated to each parameter (c.f. collinearity in regressions). The fact that CA- is fit to be negative for many participants (a pattern shown more strongly in the discovery study) is suggestive that this might be happening. A priori, the idea that you would ever increase your value estimate after negative feedback is highly implausible, which suggests that the parameter might be capturing variance besides that it is intended to capture.

The best way to resolve this uncertainty would involve running a new study in which feedback was sometimes provided in the absence of a choice - this would isolate positivity bias. Short of that, perhaps one could fit a version of the Bayesian model that also includes perseveration. If the authors can show that this model cannot capture the pattern in Figure 5, that would be fairly convincing.

We thank the reviewer for this very insightful and crucial point regarding the potential confound between positivity bias and perseveration. We entirely agree that distinguishing these effects can be challenging. To rigorously address this concern and ascertain that our observed positivity bias, particularly its inflation for low-credibility feedback, is not merely an artifact of perseveration, we conducted additional analyses as suggested.

First, following the reviewer’s suggestion we simulated our Bayesian models, including a perseveration term, for both our main and discovery studies. Crucially, none of these simulations predicted the specific pattern of inflated positivity bias for low-credibility feedback that we identified in participants.

Additionally, taking a “devil’s advocate” approach, we tested whether our credibility-CA model (which includes perseveration but not a feedback valence bias) can predict our positivity bias findings. Thus, we simulated 100 datasets using our Credibility-CA model (based on empirical best-fitting parameters). We then fitted each of these simulated datasets using our CredibilityValence CA model. By examining the distribution of results across these synthetic datasets fits and comparing them to the actual results from participants, we found that while perseveration could indeed lead (as the reviewer suspected) to an artifactual positivity bias, it could not predict the magnitude of the observed inflation of positivity bias for low-credibility feedback (whether measured in absolute or relative terms).

Based on these comprehensive analyses, we are confident that our main results concerning the modulation of a valence bias as a function of source-credibility cannot be accounted by simple choice-perseveration. We have briefly explained these analyses in the main results section:

“Previous research has suggested that positivity bias may spuriously arise from pure choice-perseveration (i.e., a tendency to repeat previous choices regardless of outcome) (49,50). While our models included a perseveration-component, this control may not be preferent. Therefore, in additional control analyses, we generated synthetic datasets using models including choice-perseveration but devoid of feedback-valence bias, and fitted them with our credibility-valence model (see SI 3.6.1). These analyses confirmed that perseveration can masquerade as an apparent positivity bias. Critically, however, these analyses also confirmed that perseveration cannot account for our main finding of increased positivity bias, relative to the overall extent of CA, for low-credibility feedback.”

Additionally, we have added a detailed description of these additional analyses and their findings to the Supplementary Information document:

“3.6 Positivity bias results cannot be explained by a pure perseveration

3.6.1 Main study

Previous research has suggested it may be challenging to dissociate between a feedback-valence positivity bias and perseveration (i.e., a tendency to repeat previous choices regardless of outcome). While our Credit Assignment (CA) models already include a perseveration mechanism to account for this, this control may not be perfect. We thus conducted several tests to examine if our positivity-bias related results could be accounted for by perseveration.

First we examined whether our Bayesian-models, augmented by a perseveration mechanism (as in our CA model) can generate predictions similar to our empirical results. We repeated our cross-fitting procedure to these extended Bayesian models. To briefly recap, this involved fitting participant behavior with them, generating synthetic datasets based on the resulting maximum likelihood (ML) parameters, and then fitting these simulated datasets with our Credibility-Valence CA model (which is designed to detect positivity bias). This test revealed that adding perseveration to our Bayesian models did not predict a positivity bias in learning. In absolute terms there was a small negativity bias (instructed-credibility Bayesian: b=−0.19, F(1,1218)=17.78, p<0.001, Fig. S23a-b; free-credibility Bayesian: b=−0.17, F(1,1218)=13.74, p<0.001, Fig. S23d-e). In relative terms we detected no valence related bias (instructed-credibility Bayesian: b=−0.034, F(1,609)=0.45, p=0.50, Fig. S22c; free-credibility Bayesian: b=−0.04, F(1,609)=0.51, p=0.47, Fig. S23f). More critically, these simulations also did not predict a change in the level of positivity bias as a function of feedback credibility, neither at an absolute level (instructed-credibility Bayesian: F(2,1218)=0.024, p=0.98, Fig. S23b; free-credibility Bayesian: F(2,1218)=0.008, p=0.99, Fig. S23e), nor at a relative level (instructedcredibility Bayesian: F(2,609)=1.57, p=0.21, Fig. S23c; free-credibility Bayesian: F(2,609)=0.13, p=0.88, Fig. S23f). The upshot is that our positivity-bias findings cannot be accounted for by our Bayesian models even when these are augmented with perseveration.

However, it is still possible that empirical CA parameters from our credibility-valence model (reported in main text Fig. 5) were distorted, absorbing variance from a perseveration. To address this, we took a “devil's advocate” approach testing the assumption that CA parameters are not truly affected by feedback valance and that there is only perseveration in our data. Towards that goal, we simulated data using our CredibilityCA model (which includes perseveration but does not contain a valence bias in its learning mechanism) and then fitted these synthetic datasets using our Credibility-Valence CA model to see if the observed positivity bias could be explained by perseveration alone. Specifically, we generated 101 “group-level” synthetic datasets (each including one simulation for each participant, based on their empirical ML parameters), and fitted each dataset with our Credibility-Valence CA model. We then analysed the resulting ML parameters in each dataset using the same mixed-effects models as described in the main text, examining the distribution of effects of interest across these simulated datasets. Comparing these simulation results to the data from participants revealed a nuanced picture. While the positivity bias observed in participants is within the range predicted by a pure perseveration account when measured in absolute terms (Fig. S24a), it is much higher than predicted by pure perseveration when measured relative to the overall level of learning (Fig. S24c). More importantly, the inflation in positivity bias for lower credibility feedback is substantially higher in participants than what would be predicted by a pure perseveration account, a finding that holds true for both absolute (Fig. S24b) and relative (Fig. S24d) measures.”

“3.6.2 Discovery study

We then replicated these analyses in our discovery study to confirm our findings. We again checked whether extended versions of the Bayesian models (including perseveration) predicted the positivity bias results observed. Our cross-fitting procedure showed that the instructed-credibility Bayesian model with perseveration did predict a positivity bias for all credibility levels in this discovery study, both when measured in absolute terms [50% credibility (b=1.74,t(824)=6.15), 70% credibility (b=2.00,F(1,824)=49.98), 85% credibility (b=1.81,F(1,824)=40.78), 100% credibility (b=2.42,F(1,824)=72.50), all p's<0.001], and in relative terms [50% credibility (b=0.25,t(412)=3.44), 70% credibility (b=0.31,F(1,412)=17.72), 85% credibility (b=0.34,F(1,412)=21.06), 100% credibility (b=0.42,F(1,412)=31.24), all p's<0.001]. However, importantly, these simulations did not predict a change in the level of positivity bias as a function of feedback credibility, neither at an absolute level (F(3,412)=1.43,p=0.24), nor at a relative level (F(3,412)=2.06,p=0.13) (Fig. S25a-c). In contrast, simulations of the free-credibility Bayesian model (with perseveration) predicted a slight negativity bias when measured in absolute terms (b=−0.35,F(1,824)=5.14,p=0.024), and no valence bias when measured relative to the overall degree of learning (b=0.05,F(1,412)=0.55,p=0.46). Crucially, this model also did not predict a change in the level of positivity bias as a function of feedback credibility, neither at an absolute level (F(3,824)=0.27,p=0.77), nor at a relative level (F(3,412)=0.76,p=0.47) (Fig. S25d-f).

As in our main study, we next assessed whether our Credibility-CA model (which includes perseveration but no valence bias) predicted the positivity bias results observed in participants in the discovery study. This analysis revealed that the average positivity bias in participants is higher than predicted by a pure perseveration account, both when measured in absolute terms (Fig. S26a) and in relative terms (Fig. S26c). Specifically, only the aVBI for the 70% credibility agent was above what a perseveration account would predict, while the rVBI for all agents except the completely credible one exceeded that threshold. Furthermore, the inflation in positivity bias for lower credibility feedback (compared to the 100% credibility agent) is significantly higher in participants than would be predicted by a pure perseveration account, in both absolute (Fig. S26b) and relative (Fig. S26d) terms.

Together, these results show that the general positivity bias observed in participants could be predicted by an instructed-credibility Bayesian model with perseveration, or by a CA model with perseveration. Moreover, we find that these two models can predict a positivity bias for the 50% credibility agent, raising a concern that our positivity bias findings for this source may be an artefact of not-fully controlled for perseveration. However, the credibility modulation of this positivity bias, where the bias is amplified for lower credibility feedback, is consistently not predicted by perseveration alone, regardless of whether perseveration is incorporated into a Bayesian or a CA model. This finding suggests that participants are genuinely modulating their learning based on feedback credibility, and that this modulation is not merely an artifact of choice perseveration.”

(3) Veracity detection or positivity bias?

The "True feedback elicits greater learning" effect (Figure 6) may be simply a re-description of the positivity bias shown in Figure 5. This figure shows that people have higher CA for trials where the feedback was in fact accurate. But assuming that people tend to choose more rewarding options, true-feedback cases will tend to also be positive-feedback cases. Accordingly, a positivity bias would yield this effect, even if people are not at all sensitive to trial-level feedback veracity. Of course, the reverse logic also applies, such that the "positivity bias" could actually reflect discounting of feedback that is less likely to be true. This idea has been proposed before as an explanation for confirmation bias (see Pilgrim et al, 2024 https://doi.org/10.1016/j.cognition.2023.105693and much previous work cited therein). The authors should discuss the ambiguity between the "positivity bias" and "true feedback" effects within the context of this literature....

Before addressing these excellent comments, we first note that we have now improved our "TruthCA" model. Previously, our Truth-CA model considered whether feedback on each trial was true or not based on realized latent true outcomes. However, it is possible that the very same feedback would have had an opposite truth-status if the latent true outcome was different (recall true outcomes are stochastic). This injects noise into the trial classification in our former model. To avoid this, in our new model feedback is modulated by the probability the reported feedback is true (marginalized over stochasticity of true outcome). Please note in our responses below that we conducted extensive analysis to confirm that positivity bias doesn’t in fact predict the truthbias we detect using our truth biased model

We have described this new model in the methods section:

“Additionally, we formulated a “Truth-CA” model, which worked as our Credibility-CA model, but incorporated a free truth-bonus parameter (TB). This parameter modulates the extent of credit assignment for each agent based on the posterior probability of feedback being true (given the credibility of the feedback agent, and the true reward probability of the chosen bandit). The chosen bandit was updated as follows:

𝑄 ← (1 – 𝑓_Q) ∗ 𝑄 + [𝐶𝐴(𝑎𝑔𝑒𝑛𝑡) + 𝑇𝐵 ∗ (𝑃(𝑡𝑟𝑢𝑡ℎ) − 0.5)] ∗ 𝐹

where P(truth) is the posterior probability of the feedback being true in the current trial (for exact calculation of P(truth) see “Methods: Bayesian estimation of posterior belief that feedback is true”).”

All relevant results have been updated accordingly in the main text:

To formally address whether feedback truthfulness modulates credit assignment, we fitted a new variant of the CA model (the “Truth-CA” model) to the data. This variant works as our Credibility-CA model, but incorporated a truth-bonus parameter (TB) which increases the degree of credit assignment for feedback as a function of the experimenter-determined likelihood the feedback is true (which is read from the curves in Fig 6a when x is taken to be the true probability the bandit is rewarding). Specifically, after receiving feedback, the Q-value of the chosen option is updated according to the following rule:

𝑄 ← (1 – 𝑓_Q) ∗ 𝑄 + [𝐶𝐴(𝑎𝑔𝑒𝑛𝑡) + 𝑇𝐵 ∗ (𝑃(𝑡𝑟𝑢𝑡ℎ) − 0.5)] ∗ 𝐹

where 𝑇𝐵 is the free parameter representing the truth bonus, and 𝑃(𝑡𝑟𝑢𝑡ℎ) is the probability the received feedback being true (from the experimenter’s perspective). We acknowledge that this model falls short of providing a mechanistically plausible description of the credit assignment process, because participants have no access to the experimenter’s truthfulness likelihoods (as the true bandit reward probabilities are unknown to them). Nonetheless, we use this ‘oracle model’ as a measurement tool to glean rough estimates for the extent to which credit assignment Is boosted as a function of its truthfulness likelihood.

Fitting this Truth-CA model to participants' behaviour revealed a significant positive truth-bonus (mean=0.21, t(203)=3.12, p=0.002), suggesting that participants indeed assign greater weight to feedback that is likely to be true (Fig. 6c; see SI 3.3.1 for detailed ML parameter results). Notably, simulations using our other models (Methods) consistently predicted smaller truth biases (compared to the empirical bias) (Fig. 6d). Moreover, truth bias was still detected even in a more flexible model that allowed for both a positivity bias and truth-bias (see SI 3.7). The upshot is that participants are biased to assign higher credit based on feedback that is more likely to be true in a manner that is inconsistent with out Bayesian models and above and beyond the previously identified positivity biases.”

Finally, the Supplementary Information for the discovery study has also been revised to feature this analysis:

“We next assessed whether participants infer whether the feedback they received on each trial was true or false and adjust their credit assignment based on this inference. We again used the “Truth-CA” model to obtain estimates for the truth bonus (TB), the increase in credit assignment as a function of the posterior probability of feedback being true. As in our main study, the fitted truth bias parameter was significantly positive, indicating that participants assign greater weight to feedback they believe is likely to be true (Fig, S4a; see SI 3.3.1 for detailed ML parameter results). Strikingly, model-simulations (Methods) predicted a lower truth bonus than the one observed in participants (Fig. S4b).”

Additionally, we thank the reviewer for pointing us to the relevant work by Pilgrim et al. (2024). We agree that the relationship between "true feedback" and "positivity bias" effects is nuanced, and their potential overlap warrants careful consideration. Note our analyses suggest that this is not solely the case. Firstly, simulations of our Credibility-Valence CA model predict only a small "truth bonus" effect, which is notably smaller than what we observed in participants. Secondly, we formulated an extension of our "Truth-CA" model that includes a valence bias in credit assignment. If our truth bonus results were merely an artifact of positivity bias, this extended model should absorb that variance, producing a null truth bonus parameter. However, fitting this model to participant data still revealed a significant positive truth bonus, which again exceeds the range predicted by simulations of our Credibility CA model:

“3.7 Truth inference is still detected when controlling for valence bias

Given that participants frequently select bandits that are, on average, mostly rewarding, it is reasonable to assume that positive feedback is more likely to be objectively true than negative feedback. This raises a question if the "truth inference" effect we observed in participants might simply be an alternative description of a positivity bias in learning. To directly test this idea, we extended our Truth-CA model to explicitly account for a valence bias in credit assignment. This extended model features separate CA parameters for positive and negative feedback for each agent. When we fitted this new model to participant behavior, it still revealed a significant truth bonus in both the main study (Wilkoxon’s signrank test: median = 0.09, z(202)=2.12, p=0.034; Fig. S27a) and the discovery study (median = 3.52, z(102)=7.86, p<0.001; Fig. S27c). Moreover, in the main study, this truth bonus remained significantly higher than what was predicted by all the alternative models, with the exception of the instructed-credibility bayesian model (Fig. S27b). In the discovery study, the truth bonus was significantly higher than what was predicted by all the alternative models (Fig. S27d).”

Together, these findings suggest that our truth inference results are not simply a re-description of a positivity bias.

Conversely, we acknowledge the reviewer's point that our positivity bias results could potentially stem from a more general truth inference mechanism. We believe that this possibility should be addressed in a future study where participants rate their belief that received feedback is true (rather than a lie).We have extended our discussion to clarify this possibility and to include the suggested citation:

“Our findings show that individuals increase their credit assignment for feedback in proportion to the perceived probability that the feedback is true, even after controlling for source credibility and feedback valence. Strikingly, this learning bias was not predicted by any of our Bayesian or credit-assignment (CA) models. Notably, our evidence for this bias is based on a “oracle model” that incorporates the probability of feedback truthfulness from the experimenter's perspective, rather than the participant’s. This raises an important open question: how do individuals form beliefs about feedback truthfulness, and how do these beliefs influence credit assignment? Future research should address this by eliciting trial-by-trial beliefs about feedback truthfulness. Doing so would also allow for testing the intriguing possibility that an exaggerated positivity bias for non-credible sources reflects, to some extent, a truth-based discounting of negative feedback—i.e., participants may judge such feedback as less likely to be true. However, it is important to note that the positivity bias observed for fully credible sources (here and in other literature) cannot be attributed to a truth bias—unless participants were, against instructions, distrustful of that source.”

The authors get close to this in the discussion, but they characterize their results as differing from the predictions of rational models, the opposite of my intuition. They write:

“Alternative "informational" (motivation-independent) accounts of positivity and confirmation bias predict a contrasting trend (i.e., reduced bias in low- and medium credibility conditions) because in these contexts it is more ambiguous whether feedback confirms one's choice or outcome expectations, as compared to a full-credibility condition.”

I don't follow the reasoning here at all. It seems to me that the possibility for bias will increase with ambiguity (or perhaps will be maximal at intermediate levels). In the extreme case, when feedback is fully reliable, it is impossible to rationally discount it (illustrated in Figure 6A). The authors should clarify their argument or revise their conclusion here.

We apologize for the lack of clarity in our previous explanation. We removed the sentence you cited (it was intended to make a different point which we now consider non-essential). Our current narration is consistent with the point you are making.

(4) Disinformation or less information?

Zooming out, from a computational/functional perspective, the reliability of feedback is very similar to reward stochasticity (the difference is that reward stochasticity decreases the importance/value of learning in addition to its difficulty). I imagine that many of the effects reported here would be reproduced in that setting. To my surprise, I couldn't quickly find a study asking that precise question, but if the authors know of such work, it would be very useful to draw comparisons. To put a finer point on it, this study does not isolate which (if any) of these effects are specific to disinformation, rather than simply less information. I don't think the authors need to rigorously address this in the current study, but it would be a helpful discussion point.

We thank the reviewer for highlighting the parallel (and difference) between feedback reliability and reward stochasticity. However, we have not found any comparable results in the literature. We also note that our discussion includes a paragraph addressing the locus of our effects making the point that more studies are necessary to determine whether our findings are due to disinformation per se or sources being less informative. While this paragraph was included in the previous version it led us to infer our Discussion was too long and we therefore shortened it considerably:

“An important question arises as to the psychological locus of the biases we uncovered. Because we were interested in how individuals process disinformation—deliberately false or misleading information intended to deceive or manipulate—we framed the feedback agents in our study as deceptive, who would occasionally “lie” about the true choice outcome. However, statistically (though not necessarily psychologically), these agents are equivalent to agents who mix truth-telling with random “guessing” or “noise” where inaccuracies may arise from factors such as occasionally lacking access to true outcomes, simple laziness, or mistakes, rather than an intent to deceive. This raises the question of whether the biases we observed are driven by the perception of potential disinformation as deceitful per se or simply as deviating from the truth. Future studies could address this question by directly comparing learning from statistically equivalent sources framed as either lying or noisy. Unlike previous studies wherein participants had to infer source credibility from experience (30,37,72), we took an explicit-instruction approach, allowing us to precisely assess source-credibility impact on learning, without confounding it with errors in learning about the sources themselves. More broadly, our work connects with prior research on observational learning, which examined how individuals learn from the actions or advice of social partners (72–75). This body of work has demonstrated that individuals integrate learning from their private experiences with learning based on others’ actions or advice—whether by inferring the value others attribute to different options or by mimicking their behavior (57,76). However, our task differs significantly from traditional observational learning. Firstly, our feedback agents interpret outcomes rather than demonstrating or recommending actions (30,37,72). Secondly, participants in our study lack private experiences unmediated by feedback sources. Finally, unlike most observational learning paradigms, we systematically address scenarios with deliberately misleading social partners. Future studies could bridge this by incorporating deceptive social partners into observational learning, offering a chance to develop unified models of how individuals integrate social information when credibility is paramount for decision-making.”

(5) Over-reliance on analyzing model parameters

Most of the results rely on interpreting model parameters, specifically, the "credit assignment" (CA) parameter. Exacerbating this, many key conclusions rest on a comparison of the CA parameters fit to human data vs. those fit to simulations from a Bayesian model. I've never seen anything like this, and the authors don't justify or even motivate this analysis choice. As a general rule, analyses of model parameters are less convincing than behavioral results because they inevitably depend on arbitrary modeling assumptions that cannot be fully supported. I imagine that most or even all of the results presented here would have behavioral analogues. The paper would benefit greatly from the inclusion of such results. It would also be helpful to provide a description of the model in the main text that makes it very clear what exactly the CA parameter is capturing (see next point).

We thank the reviewer for this important suggestion which we address together with the following point.

(6) RL or regression?

I was initially very confused by the "RL" model because it doesn't update based on the TD error. Consequently, the "Q values" can go beyond the range of possible reward (SI Figure 5). These values are therefore not Q values, which are defined as expectations of future reward ("action values"). Instead, they reflect choice propensities, which are sometimes notated $h$ in the RL literature. This misuse of notation is unfortunately quite common in psychology, so I won't ask the authors to change the variable. However, they should clarify when introducing the model that the Q values are not action values in the technical sense. If there is precedent for this update rule, it should be cited.

Although the change is subtle, it suggests a very different interpretation of the model.

Specifically, I think the "RL model" is better understood as a sophisticated logistic regression, rather than a model of value learning. Ignoring the decay term, the CA term is simply the change in log odds of repeating the just-taken action in future trials (the change is negated for negative feedback). The PERS term is the same, but ignoring feedback. The decay captures that the effect of each trial on future choices diminishes with time. Importantly, however, we can re-parameterize the model such that the choice at each trial is a logistic regression where the independent variables are an exponentially decaying sum of feedback of each type (e.g., positive-cred50, positive-cred75, ... negative-cred100). The CA parameters are simply coefficients in this logistic regression.

Critically, this is not meant to "deflate" the model. Instead, it clarifies that the CA parameter is actually not such an assumption-laden model estimate. It is really quite similar to a regression coefficient, something that is usually considered "model agnostic". It also recasts the non-standard "cross-fitting" approach as a very standard comparison of regression coefficients for model simulations vs. human data. Finally, using different CA parameters for true vs false feedback is no longer a strange and implausible model assumption; it's just another (perfectly valid) regression. This may be a personal thing, but after adopting this view, I found all the results much easier to understand.

We thank the reviewer for their insightful and illuminating comments, particularly concerning the interpretation of our model parameters and the nature of our Credit assignment model. We believe your interpretation of the model is accurate and we now narrate it to readers in the hope that our modelling will become clearer and more intuitively. We also present to readers how these recasts our “cross-fitting” approach in the way you suggested (we return to this point below).

Broadly, while we agree that modelling results depend on underlying assumptions, we believe that “model-agnostic” approaches also have important limitations—especially in reinforcement learning (RL), where choices are shaped by histories of past events, which such approaches often fail to fully account for. As students of RL, we are frequently struck by how careful modelling demonstrates that seemingly meaningful “model-agnostic” patterns can emerge as artefacts of unaccounted-for variables. We also note that the term “model-agnostic” is difficult to define—after all, even regression models rely on assumptions, and some computational models make richer or more transparent assumptions than others. Ideally, we aim to support our findings using converging methods wherever possible.

We want to clarify that many of our reported findings indeed stem from straightforward behavioral analyses (e.g., simple regressions of choice-repetition), which do not rely on complex modeling assumptions. The two key results that primarily depend on the analysis of model parameters are our findings related to positivity bias and truth inference.

Regarding the positivity bias, identifying truly model-agnostic behavioral signatures, distinct from effects like choice-perseveration, has historically been a significant challenge in the literature. Classical research on this bias rests on the interpretation of model parameters (Lefebvre et al., 2017; Palminteri et al., 2017), or at least on the use of models to assess what an “unbiased learner” baseline should look like (Palminteri & Lebreton, 2022). Some researchers have suggested possible regressions incorporating history effects to detect positivity bias from choicerepetition behavior, but these regressions (as our model) rely on subtle assumptions about forgetting and history effects (Toyama et al., 2019). Specifically, in our case, this issue is also demonstrated by analysis we conducted related to the previous point the reviewer made (about perseveration masquerading as positivity bias). We believe that dissociating clearly positivity bias from perseveration is an important challenge for the field going forward.

For our truth inference results, obtaining purely behavioral signatures is similarly challenging due to the intricate interdependencies (the reviewer has identified in previous points) between agent credibility, feedback valence, feedback truthfulness, and choice accuracy within our task design.

Finally, we agree with the reviewer that regression coefficients are often interpreted as a “modelagnostic” pattern. From this perspective even our findings regarding positivity and truth bias are not a case of over-reliance on complex model assumptions but are rather a way to expose deviations between empirical “sophisticated” regression coefficients and coefficients predicted from Bayesian models.

We have now described the main learning rule of our model in the main text to ensure that the meaning of the CA parameters is clearer for readers:

“Next, we formulated a family of non-Bayesian computational RL models. Importantly, these models can flexibly express non-Bayesian learning patterns and, as we show in following sections, can serve to identify learning biases deviating from an idealized Bayesian strategy. Here, an assumption is that during feedback, the choice propensity for the chosen bandit (which here is represented by a point estimate, “Q value“, rather than a distribution) either increases or decreases (for positive or negative feedback, respectively) according to a magnitude quantified by the free “Credit-Assignment (CA)” model parameters (47):

𝑄(𝑐ℎ𝑜𝑠𝑒𝑛) ← (1 – 𝑓_Q) ∗ 𝑄(𝑐ℎ𝑜𝑠𝑒𝑛) + 𝐶𝐴(𝑎𝑔𝑒𝑛𝑡, 𝑣𝑎𝑙𝑒𝑛𝑐𝑒) ∗ 𝐹

where F is the feedback received from the agents (coded as 1 for reward feedback and -1 for non-reward feedback), while fQ (∈[0,1]) is the free parameter representing the forgetting rate of the Q-value (Fig. 2a, bottom panel; Fig. S5b; Methods). The probability to choose a bandit (say A over B) in this family of models is a logistic function of the contrast choice-propensities between these two bandits. One interpretation of this model is as a “sophisticated” logistic regression, where the CA parameters take the role of “regression coefficients” corresponding to the change in log odds of repeating the just-taken action in future trials based on the feedback (+/- CA for positive or negative feedback, respectively; the model also includes gradual perseveration which allows for constant log-odd changes that are not affected by choice feedback; see “Methods: RL models”) . The forgetting rate captures the extent to which the effect of each trial on future choices diminishes with time. The Q-values are thus exponentially decaying sums of logistic choice propensities based on the types of feedback a bandit received.”

We also explain the implications of this perspective for our cross-fitting procedure:

“To further characterise deviations between behaviour and our Bayesian learning models, we used a “crossfitting” method. Treating CA parameters as data-features of interest (i.e., feedback dependent changes in choice propensity), our goal was to examine if and how empirical features differ from features extracted from simulations of our Bayesian learning models. Towards that goal, we simulated synthetic data based on Bayesian agents (using participants’ best fitting parameters), but fitted these data using the CA-models, obtaining what we term “Bayesian-CA parameters” (Fig. 2d; Methods). A comparison of these BayesianCA parameters, with empirical-CA parameters obtained by fitting CA models to empirical data, allowed us to uncover patterns consistent with, or deviating from, ideal-Bayesian value-based inference. Under the sophisticated logistic-regression interpretation of the CA-model family the cross-fitting method comprises a comparison between empirical regression coefficients (i.e., empirical CA parameters) and regression coefficient based on simulations of Bayesian models (Bayesian CA parameters). Using this approach, we found that both the instructed-credibility and free-credibility Bayesian models predicted increased BayesianCA parameters as a function of agent credibility (Fig. 3c; see SI 3.1.1.2 Tables S8 and S9). However, an in-depth comparison between Bayesian and empirical CA parameters revealed discrepancies from ideal Bayesian learning, which we describe in the following sections.”

Recommendations for the authors:

Reviewer #3 (Recommendations for the authors):

(1) Keep terms consistent, e.g., follow-up vs. main; hallmark vs. traditional.

We have now changed the text to keep terms consistent.

(2) CA model is like a learning rate; but it's based on the raw reward, not the TD error - this seems strange.

We thank the reviewer for this comment. We understand that the use of a CA model instead of a TD error model may seem unusual at first glance. However, the CA model offers an important advantage: it more easily accommodates what we term "negative learning rates". This means that some participants may treat certain agents (especially the random one) as consistently deceitful, leading them to effectively increase/reduce choice tendencies following negative/positive feedback. A CA model handles this naturally by allowing negative CA parameters as a simple extension of positive ones. In contrast, adapting a TD error model to account for this is more complex. For instance, attempting to introduce a "negative learning rate" makes the RW model behave in a non-stable manner (e.g., Q values become <0 or >1). At the initial stages of our project, we explored different approaches to dealing with this issue and we found the CA model provides the best approach. For these reasons, we decided to proceed with our CA model.

Additionally, we used the CA model in previous studies (e.g., Moran, Dayan & Dolan (2021)) where we included (in SI) a detailed discussion of the similarities and difference between creditassignment and Rescorla-Wagner models

(3) Why was the follow-up study not pre-registered?

We appreciate the reviewer's comment regarding preregistration, which we should have done. Unfortunately, this is now “water under the bridge” but going forward we hope to pre-register increasing parts of our work.

(4) Other work looking at reward stochasticity?

As noted in point 4 of the main weaknesses, previous work on reward stochasticity primarily focused on explaining the increase/decrease in learning and its mechanistic bases under varying stochasticity levels. In our study, we uniquely characterize several specific learning biases that are modulated by source credibility, a topic not extensively explored within the existing reward stochasticity framework, as far as we know.

(5) Equation 1 is different from the one in the figure?

The reviewer is completely correct. The figure provides a simplified visual representation, primarily focusing on the feedback-based update of the Q-value, and for simplicity, it omits the forgetting term present in the full Equation 1. To ensure complete clarity and prevent any misunderstanding, we have now incorporated a more detailed explanation of the model, including the complete Equation 1 and its components, directly within the main text. This comprehensive description will ensure that readers are fully aware of how the model operates.

“Next, we formulated a family of non-Bayesian computational RL models. Importantly, these models can flexibly express non-Bayesian learning patterns and, as we show in following sections, can serve to identify learning biases deviating from an idealized Bayesian strategy. Here, an assumption is that during feedback, the choice propensity for the chosen bandit (which here is represented by a point estimate, “Q value“, rather than a distribution) either increases or decreases (for positive or negative feedback, respectively) according to a magnitude quantified by the free “Credit-Assignment (CA)” model parameters (47):

𝑄(𝑐ℎ𝑜𝑠𝑒𝑛) ← (1 – 𝑓_Q) ∗ 𝑄(𝑐ℎ𝑜𝑠𝑒𝑛) + 𝐶𝐴(𝑎𝑔𝑒𝑛𝑡, 𝑣𝑎𝑙𝑒𝑛𝑐𝑒) ∗ 𝐹

where F is the feedback received from the agents (coded as 1 for reward feedback and -1 for non-reward feedback), while fQ (∈[0,1]) is the free parameter representing the forgetting rate of the Q-value (Fig. 2a, bottom panel; Fig. S5b; Methods).”

(6) Please describe/plot the distribution of all fitted parameters in the supplement. I would include the mean and SD in the main text (methods) as well.

Following the reviewer’s suggestions, we have included in the Supplementary Document tables displaying the mean and SD of fitted parameters from participants for our main models of interest. We have also plotted the distributions of such parameters. Both for the main study:

(7) "A novel approach within the disinformation literature by exploiting a Reinforcement Learning (RL) experimental framework".

The idea of applying RL to disinformation is not new. Please tone down novelty claims. It would be nice to cite/discuss some of this work as well.

https://arxiv.org/abs/2106.05402?utm_source=chatgpt.com https://www.scirp.org/pdf/jbbs_2022110415273931.pdf https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4173312

We thank the reviewer for pointing us towards relevant literature. We have now toned down the sentence in the introduction and cited the references provided:

“To address these questions, we adopt a novel approach within the disinformation literature by exploiting a Reinforcement Learning (RL) experimental framework (36). While RL has guided disinformation research in recent years (37–40), our approach is novel in using one of its most popular tasks: the “bandit task”.”

(8) Figure 3a - The figures should be in the order that they're referenced (3 is referenced before 2).

We generally try to stick to this important rule but, in this case, we believe that our ordering serves better the narrative and hope the reviewer will excuse this small violation.

(9) "Additionally, we found a positive feedback-effect for the 3-star agent"

What is the analysis here? To avoid confusion with the "positive feedback" effect, consider using "positive effect of feedback". The dash wasn't sufficient to avoid confusion in my case.

We have now updated the terms in the text to avoid confusion.

(10) The discovery study revealed even stronger results supporting a conclusion that the credibility-CA model was superior to both Bayesian models for most subjects

This is very subjective, but I'll just mention that my "cherry-picking" flag was raised by this sentence. Are you only mentioning cases where the discovery study was consistent with the main study? Upon a closer read, I think the answer is most likely "no", but you might consider adopting a more systematic (perhaps even explicit) policy on when and how you reference the discovery study to avoid creating this impression in a more casual reader.

We thank the reviewer for this valuable suggestion. To prevent any impression of "cherry-picking", we have removed specific references to the discovery study from the main body of the text. Instead, all discussions regarding the convergence and divergence of results between the two studies are now in the dedicated section focusing on the discovery study:

“The discovery study (n=104) used a disinformation task structurally similar to that used in our main study, but with three notable differences: 1) it included 4 feedback agents, with credibilities of 50%, 70%, 85% and 100%, represented by 1, 2, 3, and 4 stars, respectively; 2) each experimental block consisted of a single bandit pair, presented over 16 trials (with 4 trials for each feedback agent); and 3) in certain blocks, unbeknownst to participants, the two bandits within a pair were equally rewarding (see SI section 1.1). Overall, this study's results supported similar conclusions as our main study (see SI section 1.2) with a few differences. We found convergent support for increased learning from more credible sources (SI 1.2.1), superior fit for the CA model over Bayesian models (SI 1.2.2) and increased learning from feedback inferred to be true (SI 1.2.6). Additionally, we found an inflation of positivity bias for low-credibility both when measured relative to the overall level of credit assignment (as in our main study), or in absolute terms (unlike in our main study) (Fig. S3; SI 1.2.5). Moreover, choice-perseveration could not predict an amplification of positivity bias for low-credibility sources (see SI 3.6.2). However, we found no evidence for learning based on 50%-credibility feedback when examining either the feedback effect on choice repetition or CA in the credibility-CA model (SI 1.2.3).”

(11) An in-depth comparison between Bayesian and empirical CA parameters revealed discrepancies from normative Bayesian learning.

Consider saying where this in-depth comparison can be found (based on my reading, I think you're referring to the next section?

We have now modified the sentence for better clarity:

“However, an in-depth comparison between Bayesian and empirical CA parameters revealed discrepancies from ideal Bayesian learning, which we describe in the following sections.”

(12) "which essentially provides feedback" Perhaps you meant "random feedback"?

We have modified the text as suggested by the reviewer.

<(13) Essentially random

Why "essentially"? Isn't it just literally random?

We have modified the text as suggested by the reviewer.

(14) Both Bayesian models predicted an attenuated credit-assignment for the 3-star agent

Attenuated relative to what? I wouldn't use this word if you mean weaker than what we see in the human data. Instead, I would say people show an exaggerated credit-assignment, since Bayes is the normative baseline.

We changed the text according to the reviewer’s suggestion:

“A comparison of empirical and Bayesian credit-assignment parameters revealed a further deviation from ideal Bayesian learning: participants showed an exaggerated credit-assignment for the 3-star agent compared with Bayesian models.”

(15) "there was no difference between 2-star and 3-star agent contexts (b=0.051, F(1,2419)=0.39, p=0.53)"

You cannot confirm the null hypothesis! Instead, you can write "The difference between 2-star and 3-star agent contexts was not significant". Although even with this language, you should be careful that your conclusions don't rest on the lack of a difference (the next sentence is somewhat ambiguous on this point).

Additionally, the reported b coefs do not match the figure, which if anything, suggests a larger drop from 0.75 (2-star) to 1 (3-star). Is this a mixed vs fixed effects thing? It would be helpful to provide an explanation here.

We thank the reviewer for this question. When we previously submitted our manuscript, we thought that finding enhanced credit-assignment for fully credible feedback following potential disinformation from a DIFFERENT context would constitute a striking demonstration of our “contrast effect”. However, upon reexamining this finding we found out we had a coding error (affecting how trials were filtered). We have now rerun and corrected this analysis. We have assessed the contrast effect for both "same-context" trials (where the contextual trial featured the same bandit pair as the learning trial) and "different-context" trials (where the contextual trial featured a different bandit pair). Our re-analysis reveals a selective significant contrast effect in the same-context condition, but no significant effect in the different-context condition. We have updated the main text to reflect these corrected findings and provide a clearer explanation of the analysis:

“A comparison of empirical and Bayesian credit-assignment parameters revealed a further deviation from ideal Bayesian learning: participants showed an exaggerated credit-assignment for the 3-star agent compared with Bayesian models [Wilcoxon signed-rank test, instructed-credibility Bayesian model (median difference=0.74, z=11.14); free-credibility Bayesian model (median difference=0.62, z=10.71), all p’s<0.001] (Fig. 3a). One explanation for enhanced learning for the 3-star agents is a contrast effect, whereby credible information looms larger against a backdrop of non-credible information. To test this hypothesis, we examined whether the impact of feedback from the 3-star agent is modulated by the credibility of the agent in the trial immediately preceding it. More specifically, we reasoned that the impact of a 3-star agent would be amplified by a “low credibility context” (i.e., when it is preceded by a low credibility trial). In a binomial mixed effects model, we regressed choice-repetition on feedback valence from the last trial featuring the same bandit pair (i.e., the learning trial) and the feedback agent on the trial immediately preceding that last trial (i.e., the contextual credibility; see Methods for model-specification). This analysis included only learning trials featuring the 3-star agent, and context trials featuring the same bandit pair as the learning trial (Fig. 4a). We found that feedback valence interacted with contextual credibility (F(2,2086)=11.47, p<0.001) such that the feedback-effect (from the 3-star agent) decreased as a function of the preceding context-credibility (3-star context vs. 2-star context: b= -0.29, F(1,2086)=4.06, p=0.044; 2star context vs. 1-star context: b=-0.41, t(2086)=-2.94, p=0.003; and 3-star context vs. 1-star context: b=0.69, t(2086)=-4.74, p<0.001) (Fig. 4b). This contrast effect was not predicted by simulations of our main models of interest (Fig. 4c). No effect was found when focussing on contextual trials featuring a bandit pair different than the one in the learning trial (see SI 3.5). Thus, these results support an interpretation that credible feedback exerts a greater impact on participants’ learning when it follows non-credible feedback, in the same learning context.”

We have modified the discussion accordingly as well:

“A striking finding in our study was that for a fully credible feedback agent, credit assignment was exaggerated (i.e., higher than predicted by our Bayesian models). Furthermore, the effect of fully credible feedback on choice was further boosted when it was preceded by a low-credibility context related to current learning. We interpret this in terms of a “contrast effect”, whereby veridical information looms larger against a backdrop of disinformation (21). One upshot is that exaggerated learning might entail a risk of jumping to premature conclusions based on limited credible evidence (e.g., a strong conclusion that a vaccine produces significant side-effect risks based on weak credible information, following non-credible information about the same vaccine). An intriguing possibility, that could be tested in future studies, is that participants strategically amplify the extent of learning from credible feedback to dilute the impact of learning from noncredible feedback. For example, a person scrolling through a social media feed, encountering copious amounts of disinformation, might amplify the weight they assign to credible feedback in order to dilute effects of ‘fake news’. Ironically, these results also suggest that public campaigns might be more effective when embedding their messages in low-credibility contexts, which may boost their impact.”

And we have included some additional analyses in the SI document:

“3.5 Contrast effects for contexts featuring a different bandit Given that we observed a contrast effect when both the learning and the immediately preceding "context trial” involved the same pair of bandits, we next investigated whether this effect persisted when the context trial featured a different bandit pair – a situation where the context would be irrelevant to the current learning. Again, we used in a binomial mixed effects model, regressing choice-repetition on feedback valence in the learning trial and the feedback agent in the context trial. This analysis included only learning trials featuring the 3-star agent, and context trials featuring a different bandit pair than the learning trial (Fig. S22a). We found no significant evidence of an interaction between feedback valence and contextual credibility (F(2,2364)=0.21, p=0.81) (Fig. S22b). This null result was consistent with the range of outcomes predicted by our main computational models (Fig. S22c).”

We aimed to formally compare the influence of two types of contextual trials: those featuring the same bandit pair as the learning trial versus those featuring a different pair. To achieve this, we extended our mixedeffects model by incorporating a new predictor variable, "CONTEXT_TYPE" which coded whether the contextual trial involved the same bandit pair (coded as -0.5) or a different bandit pair (+0.5) compared to the learning trial. The Wilkinson notation for this expanded mixed-effects model is:

𝑅𝐸𝑃𝐸𝐴𝑇 ~ 𝐶𝑂𝑁𝑇𝐸𝑋𝑇_𝑇𝑌𝑃𝐸 ∗ 𝐹𝐸𝐸𝐷𝐵𝐴𝐶𝐾 ∗ (𝐶 𝐶𝑂𝑁𝑇𝐸𝑋𝑇_2-star + 𝐶𝑂𝑁𝑇𝐸𝑋𝑇_3-star) + 𝐵𝐸𝑇𝑇𝐸𝑅 + (1|𝑝𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑛𝑡)

This expanded model revealed a significant three-way interaction between feedback valence, contextual credibility, and context type (F(2,4451) = 7.71, p<0.001). Interpreting this interaction, we found a 2-way interaction between context-source and feedback valence when the context was the same (F(2,4451) = 12.03, p<0.001), but not when context was different (F(2,4451) = 0.23, p = 0.79). Further interpreting the double feedback-valence * context-source interaction (for the same context) we obtained the same conclusions as reported in the main text.”

(16) "Strikingly, model-simulations (Methods) showed this pattern is not predicted by any of our other models"

Why doesn't the Bayesian model predict this?

Thanks for the comment. Overall, Bayesian models do predict a slight truth inference effect (see Figure 6d). However, these effects are not as strong as the ones observed in participants, suggesting that our results go beyond what would be predicted by a Bayesian model.

Conceptually, it's important to note that the Bayesian model can infer (after controlling for source credibility and feedback valence) whether feedback is truthful based solely on prior beliefs about the chosen bandit. Using this inferred truth to amplify the weight of truthful feedback would effectively amount to “bootstrapping on one’s own beliefs.” This is most clearly illustrated with the 50% agent: if one believes that a chosen bandit yields rewards 70% of the time, then positive feedback is more likely to be truthful than negative feedback. However, a Bayesian observer would also recognize that, given the agent’s overall unreliability, such feedback should be ignored regardless.

(17) "A striking finding in our study was that for a fully credible feedback agent, credit assignment was exaggerated (i.e., higher than predicted by a Bayesian strategy)".

"Since we did not find any significant interactions between BETTER and the other regressors, we decided to omit it from the model formulation".

Was this decision made after seeing the data? If so, please report the original analysis as well.

We have included the BETTER regressor again, and we have re-run the analyses. We now report the results of such regression. We have also changed the methods section accordingly:

“We used a different mixed-effects binomial regression model to test whether value learning from the 3-star agent was modulated by contextual credibility. We focused this analysis on instances where the previous trial with the same bandit pair featured the 3-star agent. We regressed the variable REPEAT, which indicated whether the current trial repeated the choice from the previous trial featuring the same bandit-pair (repeated choice=1, non-repeated choice=0). We included the following regressors: FEEDBACK coding the valence of feedback in the previous trial with the same bandit pair (positive=0.5, negative=-0.5), CONTEXT2-star indicating whether the trial immediately preceding the previous trial with the same bandit pair (context trial) featured the 2-star agent (feedback from 2-star agent=1, otherwise=0), and CONTEXT3star indicating whether the trial immediately preceding the previous trial with the same bandit pair featured the 3-star agent. We also included a regressor (BETTER) coding whether the bandit chosen in the learning trial was the better -mostly rewarding- or the worse -mostly unrewarding- bandit within the pair. We included in this analysis only current trials where the context trial featured a different bandit pair. The model in Wilkinson’s notation was:

𝑅𝐸𝑃𝐸𝐴𝑇~ 𝐹𝐸𝐸𝐷𝐵𝐴𝐶𝐾 ∗ (𝐶𝑂𝑁𝑇𝐸𝑋𝑇_2-star + 𝐶𝑂𝑁𝑇𝐸𝑋𝑇_3-star) + 𝐵𝐸𝑇𝑇𝐸𝑅 + (1|𝑝𝑎𝑟𝑡𝑖𝑐𝑖𝑝𝑎𝑛𝑡) ( 13 )

In figure 4c, we independently calculate the repeat probability difference for the better (mostly rewarding) and worse (mostly non-rewarding) bandits and averaged across them. This calculation was done at the participants level, and finally averaged across participants.”

Author response:

We thank the reviewers for recognizing the strengths of our work, as well as for their thoughtful and constructive feedback. In this provisional response, we focus on the main concern raised—namely, the need for stronger evidence that the effect is specific to suicide. A full revision of the manuscript will follow, in which we will address this point in greater depth and respond carefully to all additional comments in a point-by-point manner.

More specifically, reviewer 3 points out that “The main analyses control for illness duration and medication but not for symptom severity. The supplementary analysis in Figure S11 is insufficient as it mistakes the absence of evidence (i.e., p > 0.05) for evidence of absence.”. This is indeed an important point that we address below.

(1) Correction for symptom severity.

To address the request for evidence on specificity to suicidality beyond general symptom severity, we performed separate linear regressions to explain in gambling behaviour, value-insensitive approach parameter (β_gain), and mood sensitivity to certain rewards (β_CR) with group as a predictor (1 for S⁺ group and 0 for S^- group) and scores for anxiety and depression as covariates. Results remained significant after controlling anxiety and depression (ps < 0.027).

Author response table 1.

Given high correlations among anxiety and depression questionnaires (rs > 0.753, ps < 0.001), we performed Principal Components Analysis (PCA) on the clinical questionnaire to extract the orthogonal components, where each component explained 86.95%, 7.09%, 3.27%, and 2.68% variance, respectively. We then performed linear regressions using these components as covariates to control for anxiety and depression. Our main results remained significant (ps < 0.027).

Author response table 2.

We believe that these analyses provide evidence that the main effects on gambling and on mood were specific to suicide.

(2) Evidence of absence of effect of symptom severity

Based on clinical interviews, we included patients with and without suicidality (S⁺ and S^- groups). However, in line with suicidal-related literature (e.g., Tsypes et al., 2024), S⁺ and S^- differed substantially in the severity of symptoms (see Table 1). Although we median-split patients by the scores of general symptoms (e.g., depression and anxiety) and verified no significant differences in these severities (Figure S11), the “absence of evidence” cannot provide insights of “evidence of absence”. We, therefore, additionally conducted Bayesian statistics in gambling behavior, value-insensitive approach parameter, and mood sensitivity to certain rewards. BF₀₁ is a Bayes factor comparing the null model (M₀) to the alternative model (M₁), where M₀ assumes no group difference. BF₀₁ > 1 indicates that evidence favors M₀. As can be seen below, most results supported null hypothesis, suggesting that general symptoms of anxiety and depression overall did not influence our main results.

Author response table 3.

Overall, we believe that these analyses provide compelling evidence for the specificity of the effect to suicide, above and beyond depression and anxiety.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #2 (Recommendations for the authors):

Comments on revised version:

The authors addressed the concerns adequately. The three remaining concerns are:

(1) The use of one-way ANOVA is not well justified.

The statement about statistical test in “Statistical analysis” section is as follows in the revised manuscript, “Data sets were tested for normality and direct comparisons between two groups were made using two-tailed Student’s t test (t test, for normally distributed data) as indicated. To evaluate statistical significance of three or more groups of samples, one-way ANOVA analysis with a Tukey test was used or repeated measures ANOVA analysis with a Tukey test was used in behavior assays. Statistical parameters are reported in the figures and the corresponding legends”.

We used a one-way ANOVA for the data about one categorical independent variable and one quantitative dependent variable. The independent variable should have at least three different groups or categories. And we conducted repeated measures ANOVA analysis for the data about behavioral tests according to the suggestion by Reviewer #1 (Point 18) in revised manuscript.

(2) The use of superplots to show culture to culture variability would make it more transparent.

Thanks for the nice suggestion. While superplots could more transparently show culture to culture variability, it is difficult to add more colors or even shades to the scatterplots in the current form, which have already been color coded for multiple groups of samples. The scatterplots we used effectively illustrate the variability across all collected data and do not affect the conclusions of our study. Therefore, we prefer not to change the way of data presentation in the revised manuscript.

(3) Change EEN1 in Figure 8B to EndoA1.

Thanks a lot for the sharp eye. Corrected.

Reviewer #3 (Recommendations for the authors):

Specific comments:

The authors have made a substantial effort to improve their manuscript. A number of issues, related to numbers of observations mentioned by the reviewers, are clarified in the revised manuscript. The authors have also clarified some of the other questions from the reviewers. The long list of issues brought up by the reviewers and the many corrections needed still raise questions about data quality in this manuscript.

In response to my comments (Point 2), the added experiment with PSD95.FingR and GPN.FingR in cultured neurons (Fig. S5A-D) is a good addition; the in vivo data using FingRs in Figure S3 look less convincing however. In response to my Point 5, the authors have added a cell-free binding assay (Figure 5I). This is a useful addition, but to convincingly make the point of interaction between Gephyrin and EndoA1, more rigorous biophysical quantitation of binding is needed. The legend in Figure 5I states that 4 independent experiments were performed, but the graph only shows 3 dots. This needs to be corrected.

We sincerely appreciate your comments and apologize for any concerns raised. As suggested (Point 2), we made many efforts to visualize endogenous postsynaptic proteins using recombinant probes. However, due to much lower expression of GPN.FingR compared with PSD95.FingR in P21 brain slices following viral infection (Figure S3), we were unable to obtain better imaging results. To strengthen our data and conclusions, we additionally performed experiments with PSD95.FingR and GPN.FingR in cultured neurons (Fig. S5A-D) in the revised manuscript.

Regarding the biophysical quantification of gephyrin–endophilin A1 binding, we do not have the equipment for this type of experiment (surface plasmon resonance or isothermal titration calorimetry). Instead, we performed a pull-down assay as an alternative to confirm their interaction (Figure 5I). We also apologize for the error in the number of independent experiments stated in the figure legend and have corrected it in the revised manuscript.

Author response:

The following is the authors’ response to the original reviews

Reviewer #1 (Public review): 

Summary: 

This manuscript uses a diverse isolate collection of Streptococcus pneumoniae from hospital patients in the Netherlands to understand the population-level genetic basis of growth rate variation in this pathogen, which is a key determinant of S. pneumoniae within-host fitness. Previous efforts have studied this phenomenon in strain-specific comparisons, which can lack the statistical power and scope of population-level studies. The authors collected a rigorous set of in vitro growth data for each S. pneumoniae isolate and subsequently paired growth curve analysis with whole-genome analyses to identify how phylogenetics, serotype, and specific genetic loci influence in vitro growth. While there were noticeable correlations between capsular serotype and phylogeny with growth metrics, they did not identify specific loci associated with altered in vitro growth, suggesting that these phenotypes are controlled by the collective effect of the entire genetic background of a strain. This is an important finding that lays the foundation for additional, more highly-powered studies that capture more S. pneumoniae genetic diversity to identify these genetic contributions.

Thank you for an excellent summary of our manuscript.

Strengths: 

(1) The authors were able to completely control the experimental and genetic analyses to ensure all isolates underwent the same analysis pipeline to enhance the rigor of their findings.

(2) The isolate collection captures an appreciable amount of S. pneumoniae diversity and, importantly, enables disentangling the contributions of the capsule and phylogenetic background to growth rates.

(3) This study provides a population-level, rather than strain-specific, view of how genetic background influences the growth rate in S. pneumoniae. This is an advance over previous studies that have only looked at smaller sets of strains.

(4) The methods used are well-detailed and robust to allow replication and extension of these analyses. Moreover, the manuscript is very well written and includes a thoughtful and thorough discussion of the strengths and limitations of the current study.

Thank you for excellently summarising the strengths of our manuscript.

Weaknesses: 

(1) As acknowledged by the authors, the genetic diversity and sample size of this newly collected isolate set are still limited relative to the known global diversity of S. pneumoniae, which evidently limits the power to detect loci with smaller/combinatorial contributions to growth rate (and ultimately infection). 

Indeed, while larger pneumococcal datasets exist globally, most of these datasets do not have reliable metadata on in vitro growth rates and other phenotypes, as the intention, for the most part, is to conduct population-level surveillance to track the changes in the serotype distribution to assess the impact of introducing pneumococcal conjugate vaccines. In this study, we adopted a different approach to phenotypically characterising the samples collected from these surveillance studies to understand the genetic features that influence the intrinsic growth characteristics of the isolates. While our dataset size is modest, it exemplifies how we can combine whole-genome sequencing and phenotypic characterisation of bacterial isolates to understand the genetic determinants that may drive intrinsic phenotypic differences between strains.

(2) The in vitro growth data is limited to a single type of rich growth medium, which may not fully reflect the nutritional and/or selective pressures present in the host.

We agree that our study focused on a single type of rich growth medium, which may not fully reflect the nutritional or selective pressures present in the host. The rationale and the representativeness of the selected culture conditions were more extensively discussed in Arends et al. (10.1128/spectrum.00050-22). Considering that this was a proof-of-concept study to assess the feasibility of our approach, future studies by us and others will evaluate the impact of using different media. Besides the media, complementary techniques such as transcriptome sequencing will help uncover additional insights into potential factors that influence differences in pneumococcal growth kinetics. 

(3) The current study does not use genetic manipulation or in vitro/in vivo infection models to experimentally test whether alteration of growth rates as observed in this study is linked to virulence or successful infection. The availability of a naturally diverse collection with phylogenetic and serotype combinations already identified as interesting by the authors provides a strong rationale for wet-lab studies of these phenotypes.

We concur that additional genetic manipulation studies to assess the impact of altering growth rates on virulence and infection would have provided further insights. While this was beyond the scope of this study, we plan to conduct follow-up work to assess this using carefully selected strains from our pneumococcal collection. Because our current study demonstrates that genetic determinants of pneumococcal growth features are not simply confined to single loci, such experimental validation would require novel wet-lab approaches that consider epistatic interactions. In addition, in vivo infection models that allow the study of dissemination from the bloodstream are not yet well established.

Reviewer #2 (Public review): 

Summary: 

The study by Chaguza et al. presents a novel perspective on pneumococcal growth kinetics, suggesting that the overall genetic background of Streptococcus pneumoniae, rather than specific loci, plays a more dominant role in determining growth dynamics. Through a genome-wide association study (GWAS) approach, the authors propose a shift in how we understand growth regulation, differing from earlier findings that pinpointed individual genes, such as wchA or cpsE, as key regulators of growth kinetics. This study highlights the importance of considering the cumulative impact of the entire genetic background rather than focusing solely on individual genetic loci.

The study emphasizes the cumulative effects of genetic variants, each contributing small individual impacts, as the key drivers of pneumococcal growth. This polygenic model moves away from the traditional focus on single-gene influences. Through rigorous statistical analyses, the authors persuasively advocate for a more holistic approach to understanding bacterial growth regulation, highlighting the complex interplay of genetic factors across the entire genome. Their findings open new avenues for investigating the intricate mechanisms underlying bacterial growth and adaptation, providing fresh insights into bacterial pathogenesis.

Thank you for an excellent summary of our manuscript.

Strengths: 

This study exemplifies a holistic approach to unraveling key factors in bacterial pathogenesis. By analyzing a large dataset of whole-genome sequences and employing robust statistical methodologies, the authors provide strong evidence to support their main findings. Which is a leap forward from previous studies focused on a relatively smaller number of strains. Their integration of genome-wide association studies (GWAS) highlights the cumulative, polygenic influences on pneumococcal growth kinetics, challenging the traditional focus on individual loci. This comprehensive strategy not only advances our understanding of bacterial growth regulation but also establishes a foundation for future research into the genetic underpinnings of bacterial pathogenesis and adaptation. The amount of data generated and corresponding approaches to analyze the data are impressive as well as convincing. The figures are convincing and comprehensible too.

Thank you for pointing out the strengths of our manuscript excellently.

Weaknesses: 

Despite the strong outcomes of the GWAS approach, this study leaves room for differing interpretations. A key point of contention lies in the title, which initially gives the impression that the research addresses growth kinetics under both in vitro and in vivo conditions. However, the study is limited to in vitro growth kinetics, with the assumption that these findings are equally applicable to in vivo scenarios-a premise that is not universally valid. To more accurately reflect the study's scope and avoid potential misrepresentation, the title should explicitly specify "in vitro" growth kinetics. This clarification would better align the title with the study's actual focus and findings.

Thank you for these suggestions. We have updated the title to include "in vitro" to avoid confusion. The new title now reads, “The capsule and genetic background, rather than specific loci, strongly influence in vitro pneumococcal growth kinetics.” While our study used in vitro data, our goal is to highlight that such in vitro differences in pneumococcal growth may influence in vivo dynamics, as highlighted in several papers referenced in the introduction and discussion. 

This study suggests that the entire genetic background significantly influences bacterial growth kinetics. However, to transform these predictions into established facts, extensive experimental validation is necessary. This would involve "bench experiments" focusing on generating and studying mutant variants of serotypes or strains with diverse genomic variations, such as targeted deletions. The growth phenotypes of these mutants should be analyzed, complemented by complementation assays to confirm the specific roles of the deleted regions. These efforts would provide critical empirical evidence to support the findings from the GWAS approach and enhance understanding of the genetic basis of bacterial growth kinetics.

We fully agree with this assessment. As reviewer #1 similarly highlighted, additional genetic manipulation studies would provide further helpful information to assess the impact of altering growth rates on virulence and infection. However, the experimental studies were beyond the scope of this study due to several factors beyond our control. However, we intend to conduct follow-up experimental work to provide additional insights into how the combination of serotypes and genetic background influences pneumococcal growth in vitro and virulence in vivo. Because our current study demonstrates that genetic determinants of pneumococcal growth features are not simply confined to single loci, such experimental validation would require novel wet-lab approaches that consider epistatic interactions. 

In the discussion section, the authors state that "the influence of serotype appeared to be higher than the genetic background for the average growth rate" (lines 296-298). Alongside references 13-15, this emphasizes the important role of capsular variability, which is a key determinant of serotypes, in influencing growth kinetics. However, this raises the question: why isn't a specific locus like cps, which is central to capsule biogenesis, considered a strong influencer of growth kinetics in this study?

Thank you for highlighting the point above. Indeed, the capsule biosynthesis (cps) locus is associated with pneumococcal growth kinetics, as seen in the analysis of individual serotypes. However, the cps locus does not come up as a hit in the GWAS because we controlled for the population structure of the pneumococcal strains. The absence of the hits in the cps locus is because serotypes, hence cps loci, tend to be tightly associated with lineages despite occasional capsule switches, which introduce serotypes to different lineages. Therefore, controlling for population structure, which is critical for GWAS analyses, virtually eliminates the detection of potential hits within the cps locus. However, detecting such hits with larger datasets may still be possible. For this reason, we performed a separate analysis of the individual serotypes and lineages shown in Figure 3.

One plausible explanation could be the absence of "elevated signals" for cps in the GWAS analysis. GWAS relies on identifying loci with statistically significant associations to phenotypes. The lack of such signals for cps may indicate that its contribution, while biologically important, does not stand out genome-wide. This might be due to the polygenic nature of growth kinetics, where the overall genetic background exerts a cumulative effect, potentially diluting the apparent influence of individual loci like cps in statistical analyses. 

We fully agree with this point. We mentioned in the abstract and discussion that the absence of the signals for specific individual loci within the pneumococcal genome may imply that the growth kinetics are polygenic. We have edited the discussion to emphasise the suggested point.

Reviewer #3 (Public review): 

This study provides insights into the growth kinetics of a diverse collection of Streptococcus pneumoniae, identifying capsule and lineage differences. It was not able to identify any specific loci from the genome-wide association studies (GWAS) that were associated with the growth features. It does provide a useful study linking phenotypic data with large-scale genomic population data. The methods for the large part were appropriately written in sufficient detail, and data analysis was performed with rigour. The interpretation of the results was supported by the data, although some additional explanation of the significance of e.g. ancestral state reconstruction would be useful. Efforts were made to make the underlying data fully accessible to the readers although some of the supplementary material could be formatted and explained a bit better. 

Thank you for the excellent summary of the manuscript. We have added some text to clarify the significance of some approaches, including ancestral state reconstruction and supplementary material.

Reviewer #1 (Recommendations for the authors): 

(1) Since the PCBN was collected pre and post-vaccine introduction, did the authors stratify their analyses other than Figure 7 (disease correlations) to assess how vaccine status may influence growth rates? Is the assertion in Lines 238-239 supported by the in vitro data? 

We have done this analysis. Overall, there was no association between vaccine introduction and pneumococcal growth rates. In lines 238-239, we assumed that in vaccinated populations, the host may be more capable of suppressing bacterial replication due to vaccination. However, there was no in vitro data to back this statement. Therefore, we have edited the statement to remove the text regarding vaccination policy. 

We considered vaccination status when analysing the data presented in Figure 7. As mentioned in the legend, we only analysed the dataset collected before vaccine introduction to avoid confounding due to vaccination status. To fully assess the impact of vaccination, we would need additional information besides the date of isolation, including vaccine doses and time since vaccination, which was not available for our study.

(2) Similarly, do any of the growth rate metrics correlate with other aspects of the clinical dataset, like the year of isolation or the sex/age of the patient?

We did not include these assessments in the manuscript, as these aspects of the clinical dataset are mostly related to the patient and not necessarily the intrinsic characteristics of the pneumococcus. However, upon revising the manuscript, we compared the growth characteristics against the vaccination period, and we did not find any statistically significant association. The relationship between pneumococcal growth features of the isolates used in the current study and their corresponding clinical manifestations of invasive  disease was described in Arends  et al. (10.1128/spectrum.00050-22).

(3) When evaluating the impact of serotype on growth rates, did the directionality of some of the described impacts match with those previously reported in other studies?

We were unable to assess the directionality of the serotype’s impact on growth rates. In part, we did not conduct this analysis because our study used different strains from those used in other studies. Such differences in the genetic backgrounds, growth media, and analytical approaches made assessing the consistencies between the studies difficult.

(4) Did the authors expect that a specific growth metric would be more likely to correlate with specific genetic variants? The reader would benefit from a brief discussion of how the metrics (e.g., maximum growth or lag phase duration) are biologically meaningful beyond the overall growth rate. 

We indeed expected that specific growth metrics might correlate with certain genetic variants based on their distinct biological roles. The lag phase duration can potentially reflect the ability of the pneumococcus to adapt to environmental conditions, such as nutrient availability or stress, and may be more influenced by regulatory genes involved in sensing and responding to environmental cues (PMID: 30642990, PMID: 22139505). In contrast, maximum growth rate is more likely to be impacted by core metabolic or biosynthetic genes that control the rate of cell division under optimal conditions (PMID: 31053828). Maximum optical density, which reflects the final cell density, might be shaped by factors related to nutrient utilization efficiency, waste tolerance, or quorum sensing. The duration of the stationary phase is related to the switch from lipoteichoic acids to wall teichoic acids, permitting the initiation of the lytic growth phase (PMID: 239401). It is unclear whether this switch is mediated by external triggers or also by intrinsic features of the pneumococcus. Including this type of analysis allows for a more nuanced understanding of how genetic variants contribute to different physiological aspects of microbial growth. The relevance of the lag phase and the stationary phase in relation to the clinical phenotypes of invasive disease (such as pleural empyema and meningitis) of our pneumococcal isolates has been studied and discussed in Arends et al. (PMID: 35678554). The observed associations are summarized in Table 2 of that article. We have added some text in the discussion on the biological relevance of each bacterial growth metric.

(5) For the GWAS analyses, have similar analyses been performed for other S. pneumoniae collections? Are there known "control" loci that the authors could replicate in the current collection to verify the robustness of the approach?

Others have undertaken GWAS analyses of other S. pneumoniae collections elsewhere. Unlike our study, none of the GWAS analyses elsewhere focused on bacterial growth kinetics. Therefore, considering this is the first GWAS study in pneumococcus and bacteria, in general, to focus on growth kinetics, we do not have “control” loci that we could replicate to verify the robustness of the approach. However, we hope that future studies will be able to utilise our findings to compare their approach as more and more similar analyses of in vitro growth data become available.

(6) Is there a statistical method that could predict the sample size necessary to detect the proposed combinatorial or small contributions from various genetic loci to growth rate? This reviewer is not an expert in statistical genetics but would appreciate an indication of the scale required by future studies to identify these regions.

We are unaware of a statistical approach that could predict sample sizes to detect small or combinatorial effect sizes. However, we intend to conduct simulations in future studies to gain insights into the required sample sizes.

(7) WGS and genome assembly metrics should be provided for each sequenced genome especially since only short-read assemblies were performed. If not already deposited, the assemblies should be deposited for data sharing as well.

We have deposited the sequence reads to the European Nucleotide Archive (ENA) and provided the accession numbers, WGS, and assembly metrics in Supplementary Data 1. We have described the tools used to generate the assemblies from the reads.

(8) Please include the specific ethics approval numbers for the sample collection protocol.

Study procedures were approved by the Medical Ethical committees of the participating hospitals, including a waiver for individual informed consent (file number 2020–6644 Radboudumc).  

Reviewer #3 (Recommendations for the authors): 

Certain aspects of the manuscript could be clarified and extended to improve the manuscript.

(1) Introduction 

a) The authors assume knowledge by the reader on Streptococcus pneumoniae, specifically the genetic diversity of lineages and capsules. This diversity is highlighted in the discussion L368 that there are >100 serotypes. The authors should consider backgrounding the number of serotypes and the importance of serotype switching in these bacteria, as well as explaining the diversity of the lineages (GPSC) that are increasingly used as standard nomenclature for Streptococcus pneumonia.

Thank you for bringing this to our attention. We have included a brief description of the GPSC lineages and capsule switching in the introduction.

b) The last paragraph of the introduction is lengthy and gets into the methods and results of the manuscript. These could be edited down.

We have revised the paragraph to remove the methods and results.

(2) Methods 

a) The authors should provide details on the QC undertaken and any exclusion criteria of genomes based on the QC. The supplement material has tabs e.g. read and assembly metrics but unclear how determined and impacted the study.

We utilised all the genomes available for this study, which had in vitro phenotypic data available. We excluded no genomes due to poor sequence quality.

Additional information about the genomes is available from previous studies, which are referenced in the methods section.

b) Why did the authors map draft assemblies to the reference genome for the SNP alignment (from which the ML tree was inferred)? Draft genome assemblies usually contain errors so there is potential for false positive SNPs. Further, there is a lack of perbase quality information using the draft genome assemblies. Given the short read data are available - why were the reads not used as input for snippy (which is the standard input for snippy)? This may have impacted the results reliant on the SNP calls.

We mapped a combination of reads and draft assemblies to the reference genome to generate the SNP alignment using Snippy (https://github.com/tseemann/snippy). For the pneumococcal isolates, we mapped the reads, while for the included outgroup, we mapped the assembly as we did not have sequence reads available. We have edited the methods section to clarify this.

c) SNP alignment. the authors explain the decision to not undertake recombination detection later in the discussion. Did the authors mask any phage or repeat regions? And how was the outgroup S. oralis included in the analyses e.g what genome was used?

We included the outgroup genome in the alignment generated by SNIPPY, which involved generating aligned consensus sequences for each isolate after mapping the reads to the pneumococcal ATCC 700669 reference genome (GenBank accession: NC_011900), as described in the methods. We have now included the accession number for the S. oralis genome, which was used as an outgroup in our phylogenetic analysis. Phages are not typically common in pneumococcal genomes compared to other species. Similarly, although repeats are present in the pneumococcal genome, the consensus in the field is that these do not particularly bias the pneumococcal phylogeny. Therefore, the consensus in the field has been not to explicitly mask these regions as done for highly clonal bacterial pathogens, such as Mycobacterium tuberculosis. Overall, our approach to building the phylogenetic tree is robust compared to alternative methods (PMID:

29774245).

d) Should the presence/absence of unitigs that were used as the input for the GWAS be included as a supp dataset?

We have now provided the presence/absence matrix for the unitigs used in the  analysis as a supplementary dataset available at GitHub(https://github.com/ChrispinChaguza/SpnGrowthKinetics). We have revised the methods section to include a section on data availability.

e) For the annotation of unitigs, the authors used their bespoke script with features from complete public genomes. Please provide accession/ identifying information of the complete genomes (not only the ATCC 700669) reference in the methods. Also, why did the authors choose not to annotate with annotate_hits_pyseer from pyseer? 

We annotated the hits using our bespoke script because we understood our approach better and could control the information generated from the script. Annotating with “annotate_hits_pyseer” from pyseer would produce similar results to both approaches, as they compared the unitigs to annotated reference genomes.

(3) Results 

a) The authors could consider providing an overview of the diversity (e.g. lineages and capsules) in the study and contextualising it in the broader context of Streptococcus pneumoniae population genomics. This would help readers who are less familiar with this pathogen to understand the diversity included in this study. 

We included this information in the first paragraph of the results section. Considering that population-level analyses based on this dataset have already been published, we have referenced the corresponding papers to provide additional information to readers.

b) Did the timespan of the study pre and post-PCV7 introduction need to be briefly touched on in the results? For example, did the serotypes and lineages vary over the two collection periods and does this need to be considered in the interpretation of the results at all? 

The prevalence of serotypes and lineages varied over time, partly due to the introduction of vaccines and random temporal fluctuations in the distribution of strains. We did not explicitly adjust for time, as this is not likely to influence the intrinsic biology of the strains. However, we adjusted for the population structure of the strains, whose changes would most likely affect the distribution of strains in the population. For other analyses, including that in Figure 7, we considered the vaccination status by restricting the analysis to the isolates collected before vaccine introduction.

c) Figures. Some of the figures had very small text (especially Figure 1) that was difficult to read and Figure 2 and Figure 4 were mentioned once, while several paragraphs of results were used to discuss Figure 3. Is Figure 1 required as a main figure? Could Figure 3 be split? e.g. one with the chord diagram, one with panels b-e, and one with panels jq? Figure 4 - the ancestral state reconstruction analyses could be expanded upon in the results.

We have increased the text in some figures where possible. However, for figures that show more information, smaller text is more suitable. 

Figure 1 is essential to the manuscript as it provides a visual overview of the approach used in this study. Without this figure, it may be difficult for some readers, especially those unfamiliar with bacterial genomic analyses, to understand our study approach and how we estimated the pneumococcal growth parameters used for the GWAS. 

For Figure 4, we prefer to keep it as it is, to have the information in one place, as splitting it will mean including some of the panels in the supplementary material, considering that we already have seven figures in the manuscript. 

We have added additional text to the results regarding the ancestral reconstruction analyses. We included them mainly to demonstrate the correlation between the pneumococcal growth rates and the phylogeny.

(4) Discussion 

a) Why was 15 hours for culture undertaken and not 24? The authors discuss the impact that this may have had on their results.

The 15-hour incubation period was deliberately chosen, as the growth curves indicate that most isolates had reached the stationary phase by that time. Extending the culture duration would likely not have yielded additional meaningful data. As is well established, Streptococcus pneumoniae undergoes autolysis upon reaching a certain cell density, which could distort growth measurements and complicate interpretation if incubation were prolonged. For clarification, we have changed the sentences related to this topic in the Discussion.

b) Some paragraphs in the discussion were very long e.g. L347-381. The authors could consider breaking long paragraphs down into shorter ones to improve the readability of the manuscript.

We agree with this assessment. We initially wanted to include all the information on the study’s limitations in the same paragraph. However, as suggested, we have now split the highlighted paragraph into two shorter paragraphs. 

(5) Supplementary Data 

a) Providing information in each tab of each supp data file would be useful. For example - including a table header that explained what was in each sheet rather than relying on the tab names. Formatting for some of the underlying supplementary data could be improved e.g. in supplementary data 2 no explanation is given to interpret the data included in these files.

Thank you for the suggestions. For clarity, we have included a header in each tab of the spreadsheet that describes what is included in each dataset. We have also removed the previous Supplementary Data 2. We realised that the information presented in this spreadsheet was redundant, as it was already available in Supplementary Data 1.

Author response:

Reviewer #1 (Public review):

Summary:

This work by Govorunova et al. identified three naturally blue-shifted channelrhodopsins (ChRs) from ancyromonads, namely AnsACR, FtACR, and NlCCR. The phylogenetic analysis places the ancyromonad ChRs in a distinct branch, highlighting their unique evolutionary origin and potential for novel applications in optogenetics. Further characterization revealed the spectral sensitivity, ionic selectivity, and kinetics of the newly discovered AnsACR, FtACR, and NlCCR. This study also offers valuable insights into the molecular mechanism underlying the function of these ChRs, including the roles of specific residues in the retinal-binding pocket. Finally, this study validated the functionality of these ChRs in both mouse brain slices (for AnsACR and FtACR) and in vivo in Caenorhabditis elegans (for AnsACR), demonstrating the versatility of these tools across different experimental systems.

In summary, this work provides a potentially valuable addition to the optogenetic toolkit by identifying and characterizing novel blue-shifted ChRs with unique properties.

Strengths:

This study provides a thorough characterization of the biophysical properties of the ChRs and demonstrates the versatility of these tools in different ex vivo and in vivo experimental systems. The mutagenesis experiments also revealed the roles of key residues in the photoactive site that can affect the spectral and kinetic properties of the channel.

We thank the Reviewer for his/her positive evaluation of our work.

Weaknesses:

While the novel ChRs identified in this work are spectrally blue-shifted, there still seems to be some spectral overlap with other optogenetic tools. The authors should provide more evidence to support the claim that they can be used for multiplex optogenetics and help potential end-users assess if they can be used together with other commonly applied ChRs. Additionally, further engineering or combination with other tools may be required to achieve truly orthogonal control in multiplexed experiments.

To demonstrate the usefulness of ancyromonad ChRs for multiplex optogenetics as a proof of principle, we co-expressed AnsACR with the red-shifted cation-conducting ChR Chrimson and measured net photocurrent generated by this combination as a function of the wavelength. We found that it is hyperpolarizing in the blue region of the spectrum, and depolarizing at the red region. In the revision, we added a new panel (Figure 1D) showing these results and the following paragraph to the main text:

“To test the possibility of using AnsACR in multiplex optogenetics, we co-expressed it with the red-shifted CCR Chrimson (Klapoetke et al., 2014) fused to an EYFP tag in HEK293 cells. We measured the action spectrum of the net photocurrents with 4 mM Cl^- in the pipette, matching the conditions in the neuronal cytoplasm (Doyon, Vinay et al. 2016). Figure 1D, black shows that the direction of photocurrents was hyperpolarizing upon illumination with λ<500 nm and depolarizing at longer wavelengths. A shoulder near 520 nm revealed a FRET contribution from EYFP (Govorunova, Sineshchekov et al. 2020), which was also observed upon expression of the Chrimson construct alone (Figure 1D, red)”.

In the C. elegans experiments, partial recovery of pharyngeal pumping was observed after prolonged illumination, indicating potential adaptation. This suggests that the effectiveness of these ChRs may be limited by cellular adaptation mechanisms, which could be a drawback in long-term experiments. A thorough discussion of this challenge in the application of optogenetics tools would prove very valuable to the readership.

We added the following paragraph to the revised Discussion:

“One possible explanation of the partial recovery of pharyngeal pumping that we observed after 15-s illumination, even at the highest tested irradiance, is continued attenuation of photocurrent during prolonged illumination (desensitization). However, the rate of AnsACR desensitization (Figure 1 – figure supplement 4A and Figure 1 – figure supplement 5A) is much faster than the rate of the pumping recovery, reducing the likelihood that desensitization is driving this phenomenon. Another possible reason for the observed adaptation is an increase in the cytoplasmic Cl^- concentration owing to AnsACR activity and hence a breakdown of the Cl^- gradient on the neuronal membrane. The C. elegans pharynx is innervated by 20 neurons, 10 of which are cholinergic (Pereira, Kratsios et al. 2015). A pair of MC neurons is the most important for regulation of pharyngeal pumping, but other pharyngeal cholinergic neurons, including I1, M2, and M4, also play a role (Trojanowski, Padovan-Merhar et al. 2014). Moreover, the pharyngeal muscles generate autonomous contractions in the presence of acetylcholine tonically released from the pharyngeal neurons (Trojanowski, Raizen et al. 2016). Given this complexity, further elucidation of pharyngeal pumping adaptation mechanisms is beyond the scope of this study.”

Reviewer #2 (Public review):

Summary:

Govorunova et al present three new anion opsins that have potential applications in silencing neurons. They identify new opsins by scanning numerous databases for sequence homology to known opsins, focusing on anion opsins. The three opsins identified are uncommonly fast, potent, and are able to silence neuronal activity. The authors characterize numerous parameters of the opsins.

Strengths:

This paper follows the tradition of the Spudich lab, presenting and rigorously characterizing potentially valuable opsins. Furthermore, they explore several mutations of the identified opsin that may make these opsins even more useful for the broader community. The opsins AnsACR and FtACR are particularly notable, having extraordinarily fast onset kinetics that could have utility in many domains. Furthermore, the authors show that AnsACR is usable in multiphoton experiments having a peak photocurrent in a commonly used wavelength. Overall, the author's detailed measurements and characterization make for an important resource, both presenting new opsins that may be important for future experiments, and providing characterizations to expand our understanding of opsin biophysics in general.

We thank the Reviewer for his/her positive evaluation of our work.

Weaknesses:

First, while the authors frequently reference GtACR1, a well-used anion opsin, there is no side-by-side data comparing these new opsins to the existing state-of-the-art. Such comparisons are very useful to adopt new opsins.

GtACR1 exhibits the peak sensitivity at 515 nm and therefore is poorly suited for combination with red-shifted CCRs or fluorescent sensors, unlike blue-light-absorbing ancyromonad ACRs. Nevertheless, we conducted side-by-side comparison of ancyromonad ChRs, GtACR1 and GtACR2, the latter of which has the spectral maximum at 470 nm. The results are shown in the new Figures 1E and F, and the new multipanel Figure 1 – figure supplement 4 added in the revision. We also added the following text, describing these results, to the revised Results section:

“Figures 1E and F show the dependence of the peak photocurrent amplitude and reciprocal peak time, respectively, on the photon flux density for ancyromonad ChRs and GtACRs. The current amplitude saturated earlier than the time-to-peak for all tested ChRs. Figure 1 – figure supplement 4A-E shows normalized photocurrent traces recorded at different photon densities. Quantitation of desensitization at the end of 1-s illumination revealed a complex light dependence (Figure 1, Figure Supplement 4F). Figure 1 – figure supplement 5 shows normalized photocurrent traces recorded in response to a 5-s light pulse of the maximal available intensity and the magnitude of desensitization at its end.”

Next, multiphoton optogenetics is a promising emerging field in neuroscience, and I appreciate that the authors began to evaluate this approach with these opsins. However, a few additional comparisons are needed to establish the user viability of this approach, principally the photocurrent evoked using the 2p process, for given power densities. Comparison across the presented opsins and GtACR1 would allow readers to asses if these opsins are meaningfully activated by 2P.

We carried out additional 2P experiments in ancyromonad ChRs, GtACR1 and GtACR2 and added their results to a new main-text Figure 6 and Figure 6 – figure supplement 1. We added the new section describing these results, “Two-photon excitation”, to the main text in the revision:

“To determine the 2P activation range of AnsACR, FtACR, and NlCCR, we conducted raster scanning using a conventional 2P laser, varying the excitation wavelength between 800 and 1,080 nm (Figure 6 – figure supplement 1). All three ChRs generated detectable photocurrents with action spectra showing maximal responses at ~925 nm for AnsACR, 945 nm for FtACR, and 890 nm for NlCCR (Figure 6A). These wavelengths fall within the excitation range of common Ti:Sapphire lasers, which are widely used in neuroscience laboratories and can be tuned between ~700 nm and 1,020-1,300 nm. To assess desensitization, cells expressing AnsACR, FtACR, or NlCCR were illuminated at the respective peak wavelength of each ChR at 15 mW for 5 seconds. GtACR1 and GtACR2, previously used in 2P experiments (Forli, Vecchia et al. 2018, Mardinly, Oldenburg et al. 2018), were included for comparison. The normalized photocurrent traces recorded under these conditions are shown in Figure 6B-F. The absolute amplitudes of 2P photocurrents at the peak time and at the end of illumination are shown in Figure 6G and H, respectively. All five tested variants exhibited comparable levels of desensitization at the end of illumination (Figure 6I).”

Reviewer #3 (Public review):

Summary:

The authors aimed to develop Channelrhodopsins (ChRs), light-gated ion channels, with high potency and blue action spectra for use in multicolor (multiplex) optogenetics applications. To achieve this, they performed a bioinformatics analysis to identify ChR homologues in several protist species, focusing on ChRs from ancyromonads, which exhibited the highest photocurrents and the most blue-shifted action spectra among the tested candidates. Within the ancyromonad clade, the authors identified two new anion-conducting ChRs and one cation-conducting ChR. These were characterized in detail using a combination of manual and automated patch-clamp electrophysiology, absorption spectroscopy, and flash photolysis. The authors also explored sequence features that may explain the blue-shifted action spectra and differences in ion selectivity among closely related ChRs.

Strengths:

A key strength of this study is the high-quality experimental data, which were obtained using well-established techniques such as manual patch-clamp and absorption spectroscopy, complemented by modern automated patch-clamp approaches. These data convincingly support most of the claims. The newly characterized ChRs expand the optogenetics toolkit and will be of significant interest to researchers working with microbial rhodopsins, those developing new optogenetic tools, as well as neuro- and cardioscientists employing optogenetic methods.

We thank the Reviewer for his/her positive evaluation of our work.

Weaknesses:

This study does not exhibit major methodological weaknesses. The primary limitation of the study is that it includes only a limited number of comparisons to known ChRs, which makes it difficult to assess whether these newly discovered tools offer significant advantages over currently available options.

We conducted side-by-side comparison of ancyromonad ChRs and GtACRs, wildly used for optical inhibition of neuronal activity. The results are shown in the new Figures 1E and F, and the new multipanel Figure 1 – figure supplement 4 and Figure 1 – figure supplement 5 added in the revision. We also added the following text, describing these results, to the revised Results section:

“Figures 1E and F show the dependence of the peak photocurrent amplitude and reciprocal peak time, respectively, on the photon flux density for ancyromonad ChRs and GtACRs. The current amplitude saturated earlier than the time-to-peak for all tested ChRs. Figure 1 – figure supplement 4A-E shows normalized photocurrent traces recorded at different photon densities. Quantitation of desensitization at the end of 1-s illumination revealed a complex light dependence (Figure 1, Figure Supplement 4F). Figure 1 – figure supplement 5 shows normalized photocurrent traces recorded in response to a 5-s light pulse of the maximal available intensity and the magnitude of desensitization at its end.”

Additionally, although the study aims to present ChRs suitable for multiplex optogenetics, the new ChRs were not tested in combination with other tools. A key requirement for multiplexed applications is not just spectral separation of the blue-shifted ChR from the red-shifted tool of interest but also sufficient sensitivity and potency under low blue-light conditions to avoid cross-activation of the respective red-shifted tool. Future work directly comparing these new ChRs with existing tools in optogenetic applications and further evaluating their multiplexing potential would help clarify their impact.

As a proof of principle, we co-expressed AnsACR with the red-shifted cation-conducting CCR Chrimson and demonstrated that the net photocurrent generated by this combination is hyperpolarizing in the blue region of the spectrum, and depolarizing at the red region. In the revision, we added a new panel (Figure 1D) showing these results and the following paragraph to the main text:

“To test the possibility of using AnsACR in multiplex optogenetics, we co-expressed it with the red-shifted CCR Chrimson (Klapoetke et al., 2014) fused to an EYFP tag in HEK293 cells. We measured the action spectrum of the net photocurrents with 4 mM Cl^- in the pipette, matching the conditions in the neuronal cytoplasm (Doyon, Vinay et al. 2016). Figure 1D, black shows that the direction of photocurrents was hyperpolarizing upon illumination with λ<500 nm and depolarizing at longer wavelengths. A shoulder near 520 nm revealed a FRET contribution from EYFP (Govorunova, Sineshchekov et al. 2020), which was also observed upon expression of the Chrimson construct alone (Figure 1D, red)”.

Reviewing Editor Comments:

The reviewers suggest that direct comparison to GtACR1 is the most important step to make this work more useful to the community.

We followed the Reviewers’ recommendations and carried out side-by-side comparison of ancyromonad ChRs and GtACR1 as well as GtACR2 (Figure 1E and F, Figure 1 – figure supplement 4, Figure 1 – figure supplement 5, and Figure 6). Note, however, that GtACR1’s spectral maximum is at 515 nm, which makes it poorly suitable for blue light excitation. Also, ChRs are known to perform very differently in different cell types and upon expression of their genes in different vector backbones, so our results cannot be generalized for all experimental systems. Each ChR user needs to select the most appropriate tool for his/her purpose by testing several candidates in his/her own experimental setting.

Reviewer #1 (Recommendations for the authors):

(1) The figure legend for Figure 2D-I appears to be incomplete. Please provide a detailed explanation of the panels.

In the revision, we have expanded the legend of Figure 2 to explain all individual panels.

(2) The meaning of the Vr shift (Y-axis in Figure 2H-I) should be clarified in the main text to aid reader understanding.

In the revision, we added the phrase “which indicated higher relative permeability to NO₃ than to Cl^-“ to explain the meaning of the Vr shift upon replacement of Cl^- with NO₃-.

(3) Adding statistical analysis for the peak and end photocurrent values in Figure 2D-F would strengthen the claim that there is minimal change in relative permeability during illumination.

In the revision, we added the V_r values for the peak photocurrent to Figure 2H-I, which already contained the V_r values for the end photocurrent, and carried out a statistical analysis of their comparison. The following sentence was added to the text in the revision:

“The V_r values of the peak current and that at the end of illumination were not significantly different by the two-tailed Wilcoxon signed-rank test (Fig. 2G), indicating no change in the relative permeability during illumination.”

(4) Figure 4H and I seem out of place in Figure 4, as the title suggests a focus on wild-proteins and AnsACR mutants. The authors could consider moving these panels to Figure 3 for better alignment with the content.

As noted below, we changed the panel order in Figure 4 upon the Reviewer’s request. In particular, former Figure 4I is Figure 4C in the revision, and former Figure 4H is now panel C in Figure 3 – figure supplement 1 in the revision. We rearranged the corresponding section of the text (highlighted yellow in the manuscript).

(5) The characterization section could be strengthened by including data on the pH sensitivity of FtACR, which is currently missing from the main figures.

Upon the Reviewer’s request, we carried out pH titration of FtACR absorbance and added the results as Figure 4B in the revision.

(6) The logic in Figure 4A-G appears somewhat disjointed. For example, Figure 4A shows pH sensitivity for WT AnsACR and the G86E mutant, while Figure 4 B-D shifts to WT AnsACR and the D226N mutant, and Figure 4E returns to the G86E mutant. Reorganizing or clarifying the flow would improve readability.

We followed the Reviewer’s advice and changed the panel order in Figure 4. In the revised version, the upper row (panels A-C) shows the pH titration data of the three WTs, the middle row (panels D-F) shows analysis of the AnsACR_D226N mutant, and the lower row (panels G-I) shows analysis of the AnsACR_G88E mutant. We also rearranged accordingly the description of these panels in the text.

(7) In Figure 5A, "NIACR" should likely be corrected to "NlCCR".

We corrected the typo in the revision.

(8) The statistical significance in Figure 6C and D is somewhat confusing. Clarifying which groups are being compared and using consistent symbols would improve interoperability.

In the revision, we improved the figure panels and legend to clarify that the comparisons are between the dark and light stimulation groups within the same current injection.

(9) The authors pointed out that at rest or when a small negative current was injected, the neurons expressing Cl- permeable ChRs could generate a single action potential at the beginning of photostimulation, as has been reported before. The authors could help by further discussing if and how this phenomenon would affect the applicability of such tools.

We mentioned in the revised Discussion section that activation of ACRs in the axons could depolarize the axons and trigger synaptic transmission at the onset of light stimulation, and this undesired excitatory effect need to be taken into consideration when using ACRs.

Reviewer #2 (Recommendations for the authors):

Govorunova et al present three new anion opsins that have potential applications in silencing neurons. This paper follows the tradition of the Spudich lab, presenting and rigorously characterizing potentially valuable opsins. Furthermore, they explore several mutations of the identified opsin that may make these opsins even more useful for the broader community. In general, I feel positively about this manuscript. It presents new potentially useful opsins and provides characterization that would enable its use. I have a few recommendations below, mostly centered around side-by-side comparisons to existing opsins.

(1) My primary concern is that while there is a reference to GtACR1, a highly used opsin first described by this team, they do not present any of this data side by side.

When evaluating opsins to use, it is important to compare them to the existing state of the art. As a potential user, I need to know where these opsins differ. Citing other papers does not solve this as, even within the same lab, subtle methodological differences or data plotting decisions can obscure important differences.

As we explained in the response to the public comments, we carried out side-by-side comparison of ancyromonad ChRs and GtACRs as requested by the Reviewer. The results are shown in the new Figures 1E and F, and the new multipanel Figure 1 – figure supplement 4 and Figure 1 – figure supplement 5, added in the revision. However, we would like to emphasize a limited usefulness of such comparative analysis, as ChRs are known to perform very differently in different cell types and upon expression of their genes in different vector backbones, so our results cannot be generalized for all experimental systems. Each ChR user needs to select the most appropriate tool for his/her purpose by testing several candidates in his/her own experimental setting.

(2) Multiphoton optogenetics is an emerging field of optogenetics, and it is admirable that the authors address it here. The authors should present more 2p characterization, so that it can be established if these new opsins are viable for use with 2P methods, the way GtACR1 is. The following would be very useful for 2P characterization:

Photocurrents for a given power density, compared to GtACR1 and GtACR2.

The new Figure 6 (B-F) added in the revision shows photocurrent traces recorded from the three ancyromonad ChRs and  two GtACRs upon 2P excitation of a given power density.

Comparing NICCR and FtACR's wavelength specificity and photocurrent. If these opsins are too weak to create reasonable 2P spectra, this difference should be discussed.

The new Figure 6A shows the 2P action spectra of all three ancyromonad ChRs.

A Trace and calculated photocurrent kinetics to compare 1P and 2P. This need not be the flash-based absorption characterization of Figure 3, but a side-by-side photocurrent as in Figure 2.

As mentioned above, photocurrent traces recorded from ancyromonad ChRs and GtACRs upon 2P excitation are shown in the new Figure 6 (B-F). However, direct comparison of the 2P data with the 1P data is not possible, as we used laser scanning illumination for the former and wild-field illumination for the latter.

Characterization of desensitization. As the authors mention, many opsins undergo desensitization, presenting the ratio of peak photocurrent vs that at multiple time points (probably up to a few seconds) would provide evidence for how effectively these constructs could be used in different scenarios.

We conducted a detailed analysis of desensitization under both 1P and 2P excitation. The new Figure 1 – figure supplement 4 and Figure 1 – figure supplement 5 show the data obtained under 1P excitation, and the new Figure 6 shows the data for 2P conditions.

I have to admit, that by the end of the paper, I was getting confused as to which of the three original constructs had which property, and how that was changing with each mutation. I would suggest that a table summarizing each opsin and mutation with its onset and offset kinetics, peak wavelength, photocurrent, and ion selectivity would greatly increase the ability to select and use opsins in the future.

In the revision, we added a table of the spectroscopic properties of all tested mutants as Supplementary File 2. This study did not aim to analyze other parameters listed by the Reviewer. We added the following sentence referring to this table to the main text:

“Supplementary File 2 contains the λ values of the half-maximal amplitude of the long-wavelength slope of the spectrum, which can be estimated more accurately from the action spectra than the λ of the maximum.”

It may be out of the scope of this manuscript, but if a soma localization sequence can be shown to remove the 'axonal spiking' (as described in line 441), this would be a significant addition to the paper.

Our previous study (Messier et al., 2018, doi: 10.7554/eLife.38506) showed that a soma localization sequence can reduce, but not eliminate, the axonal spiking. We plan to test these new ACRs with the trafficking motifs in the future.

NICCR appears to have the best photocurrents of all tested opsins in this paper. It seems odd that it was omitted from the mouse cortical neurons experiments.

We have not included analysis of NlCCR behavior in neurons because we are preparing a separate manuscript on this ChR.

Figure 6 would benefit from more gradation in the light powers used to silence and would benefit from comparison to GtACR. I suggest using a fixed current with a series of illumination intensities to see which of the three opsins (or GtACR) is most effective at silencing. At present, it looks binary, and a user cannot evaluate if any of these opsins would be better than what is already available.

In the revision, we added the data comparing the light sensitivity of AnsACR and FtACR with previously identified GtACR1 and GtACR2 (new Figure 1E and F) to help users compare these ACRs. Although they are less sensitive to light comparing to GtACR1 and GtACR2, they could still be activated by commercially available light sources if the expression levels are similar. Less sensitive ACRs may have less unwanted activation when using with other optogenetic tools.

Reviewer #3 (Recommendations for the authors):

Suggested Improvements to Experiments, Data, or Analyses:

(1) Line 25: "significantly exceeding those by previously known tools" and Line 408: "NlCCR is the most blue-shifted among ancyromonad ChRs and generates larger photocurrents than the earlier known CCRs with a similar absorption maximum." As noted in the public review, this statement applies only to a very specific subgroup of ChRs with spectral maxima below 450 nm. If the goal was to claim that NlCCR is a superior tool among a broader range of blue-light-activated ChRs, direct comparisons with state-of-the-art ChRs such as ChR2 T159C (Berndt et al., 2011), CatCh (Kleinlogel et al., 2014), CoChR (Klapoetke et al., 2014), CoChR-3M (Ganjawala et al., 2019), or XXM 2.0 (Ding et al., 2022) would be beneficial. If the goal was to demonstrate superiority among tools with spectra below 450 nm, I suggest explicitly stating this in the paper.

The Reviewer correctly inferred that we emphasized the superiority of NlCCR among tools with similar spectral maxima, not all blue-light-activated ChRs available for neuronal photoexcitation, most of which exhibit absorption maxima at longer wavelengths. To clarify this, we added “with similar spectral maxima” to the sentence in the original Line 25. The sentence in Line 408 already contains this clarification: “with a similar absorption maximum”.

(2) Lines 111-113: "The absorption spectra of the purified proteins were slightly blue-shifted from the respective photocurrent action spectra (Figure 1D), likely due to the presence of non-electrogenic cis-retinal-bound forms." I would be skeptical of this statement. The spectral shifts in NlCCR and AnsACR are small and may fall within the range of experimental error. The shift in FtACR is more apparent; however, if two forms coexist in purified protein, this should be reflected as two Gaussian peaks in the absorption spectrum (or at least as a broader total peak reflecting two states with close maxima and similar populations). On the contrary, the action spectrum appears to have two peaks, one potentially below 465 nm. Generally, neither spectrum appears significantly broader than a typical microbial rhodopsin spectrum. This question could be clarified by quantifying the widths of the absorption and action spectra or by overlaying them on the same axis. In my opinion, the two spectra seem very similar, and just appearance of the "bump" in the action spectum shifts the apparent maximum of the action spectrum to the red. If there were two states, then they should both be electrogenic, and the slight difference in spectra might be explained by something else (e.g. by a slight difference in the quantum yields of the two states).

As the Reviewer suggested, in the revision we added a new figure (Figure 1 – figure supplement 2), showing the overlay of the absorption and action spectra of each ancyromonad ChR. This figure shows that the absorption spectra are wider than the action spectra (especially in AnsACR and FtACR), which confirms our interpretation (contribution of the non-electrogenic blue-shifted cis-retinal-bound forms to the absorption spectrum). Note that the presence of such forms explaining a blue shift of the absorption spectrum has been experimentally verified in HcKCR1 (doi: 10.1016/j.cell.2023.08.009; 10.1038/s41467-025-56491-9). Therefore, we revised the text as follows:

“The absorption spectra of the purified proteins (Figure 1C) were slightly blue-shifted from the respective photocurrent action spectra (Figure 1 – figure supplement 3), likely due to the presence of non-electrogenic cis-retinal-bound forms. The presence of such forms, explaining the discrepancy between the absorption and the action spectra, was verified by HPLC in KCRs (Tajima et al. 2023, Morizumi et al., 2025).”

(3) Lines 135-136: "The SyncroPatch enables unbiased estimation of the photocurrent amplitude because the cells are drawn into the wells without considering their tag fluorescence." While SyncroPatch does allow unbiased selection of patched cells, it does not account for the fraction of transfected cells. Without a method to exclude non-transfected cells, which are always present in transient transfections, the comparison of photocurrents may be affected by the proportion of untransfected cells, which could vary between constructs. To clarify whether the statistically significant difference in the Kolmogorov-Smirnov test could indicate that the fraction of transfected cells after 48-72h differs between constructs, I suggest analyzing only transfected cells or reporting fractions of transfected cells by each construct.

The Reviewer correctly states that non-transfected cells are always present in transiently transfected cell populations. However, his/her suggestion to “exclude non-transfected cells” is not feasible in the absence of a criterion for such exclusion. As it is evident from our data, transient transfection results in a continuum of the amplitude values, and it is not possible to distinguish a small photocurrent from no photocurrent, considering the noise level. We would like, however, to emphasize that not excluding any cells provides an estimate of the overall potency of each ChR variant, which depends on both the fraction of transfected cells and their photocurrents. This approach mimics the conditions of in vivo experiments, when non-expressing cells also cannot be excluded.

(4) Line 176: "AnsACR and FtACR photocurrents exhibited biphasic rise." The fastest characteristic time is very close to the typical resolution of a patch-clamp experiment (RC = 50 μs for a 10 pF cell with a 5 MΩ series resistance). Thus, I am skeptical that the faster time constant of the biphasic opening represents a protein-specific characteristic time. It may not be fully resolved by patch-clamp and could simply result from low-pass filtering of a specific cell. I suggest clarifying this for the reader.

The Reviewer is right that the patch clamp setup acts as a lowpass filter. Earlier, we directly measured its time resolution (~15 μs) by recording the ultrafast (occurring on the ps time scale) charge movements related to the trans-cis isomerization (doi: 10.1111/php.12558). However, the lowpass filter of the setup can only slow the entire signal, but cannot lead to the appearance of a separate kinetic component (i.e. a monophasic process cannot become biphasic). Therefore, we believe that the biphasic photocurrent rise reflects biphasic channel opening rather than a measurement artifact. Two phases in the channel opening have also been detected in GtACR1 (doi: 10.1073/pnas.1513602112) and CrChR2 (10.1073/pnas.1818707116).

(5) Line 516: "The forward LED current was 900 mA." It would be more informative to report the light intensity rather than the forward current, as many readers may not be familiar with the specific light output of the used LED modules at this forward current.

We have added the light intensity value in the revision:

“The forward LED current was 900 mA (which corresponded to the irradiance of ~2 mW mm^-2)…”

(6) Lines 402-403: "The NlCCR ... contains a neutral residue in the counterion position (Asp85 in BR), which is typical of all ACRs. Yet, NlCCR does not conduct anions, instead showing permeability to Na+." This is not atypical for CCRs and has been demonstrated in previous works of the authors (CtCCR in Govorunova et al. 2021, ChvCCR1 in Govorunova et al. 2022). What is unique is the absence of negatively charged residues in TM2, as noted later in the current study. However, the absence of negatively charged residues in TM2 appears to be rare for ACRs as well. Not as a strong point of criticism, but to enhance clarity, I suggest analyzing the frequency of carboxylate residues in TM2 of ACRs to determine whether the unique finding is relevant to ion selectivity or to another property.

The Reviewer is correct that some CCRs lack a carboxylate residue in the D85 position, so this feature alone cannot be considered as a differentiating criterion. However, the complete absence of glutamates in TM2 is not rare in ACRs and is found, for example, in HfACR1 and CarACR2. We have discussed this issue in our earlier review (doi: 10.3389/fncel.2021.800313) and do not think that repeating this discussion in this manuscript is appropriate.

Recommendations for Writing and Presentation:

(1) Some figures contain incomplete or missing labels:

Figure 2: Panels D to I lack labels.

In the revision, we have expanded the legend of Figure 2 to explain all individual panels.

Figure 3 - Figure Supplement 1: Missing explanations for each panel.

In the revision, we changed the order of panes and explained all individual panels in the legend.

Figure 5 - Figure Supplement 1: Missing explanations for each panel.

No further explanation for individual panels in this Figure is needed because all panels show the action spectra of various mutants, the names of which are provided in the panels themselves. Repeating this information in the figure legend would be redundant.

(2) In Figure 2, "sem" is written in lowercase, whereas "SEM" is capitalized in other figures. Standardizing the format would improve consistency.

In the revision, we changed the font of the SEM abbreviation to the uppercase in all instances.

(3) Line 20: "spectrally separated molecules must be found in nature." There is no proof that they cannot be developed synthetically; rather, it is just difficult. I suggest softening this statement, as the findings of this study, together with others, will probably allow designing molecules with specified spectral properties in the future.

In the revision, we changed the cited sentence to the following:

“Multiplex optogenetic applications require spectrally separated molecules, which are difficult to engineer without disrupting channel function”.

(4) Line 216-219: "Acidification increased the amplitude of the fast current ~10-fold (Figure 4F) and shifted its Vr ~100 mV (Figure 3 - figure supplement 1D), as expected of passive proton transport. The number of charges transferred during the fast peak current was >2,000 times smaller than during the channel opening, from which we concluded that the fast current reflects the movement of the RSB proton." The claim about passive transport of the RSB proton should be clarified, as typically, passive transport is not limited to exactly one proton per photocycle, and the authors observe the increase in the fast photocurrents upon acidification.

We thank the Reviewer for pointing out the confusing character of our description. To clarify the matter, we added a new photocurrent trace to Figure 4I in the revision recorded from AnsACR_G86E at 0 mV and pH 7.4. We have rewritten the corresponding section of Results as follows:

“Its rise and decay τ corresponded to the rise and decay τ of the fast positive current recorded from AnsACR_G86E at 0 mV and neutral pH, superimposed on the fast negative current reflecting the chromophore isomerization (Figure 4I, upper black trace). We interpret this positive current as an intramolecular proton transfer to the mutagenetically introduced primary acceptor (Glu86), which was suppressed by negative voltage (Figure 4I, lower black trace). Acidification increased the amplitude of the fast negative current ~10-fold (Figure 4I, black arrow) and shifted its V_r ~100 mV to more depolarized values (Figure 4 – figure supplement 2A). This can be explained by passive inward movement of the RSB proton along the large electrochemical gradient.”

Minor Corrections:

(1) Line 204: Missing bracket in "phases in the WT (Figure 4D."

The quoted sentence was deleted during the revision.

(2) Line 288: Typo-"This Ala is conserved" should probably be "This Met is conserved."

We mean here the Ala four residues downstream from the first Ala. To avoid confusion, we changed the cited sentence to the following:

“The Ala corresponding to BR’s Gly122 is also found in AnsACR and NlCCR (Figure 5A)…”

(3) Lines 702-704: Missing Addgene plasmid IDs in "(plasmids #XXX and #YYY, respectively)."

In the revision, we added the missing plasmid IDs.

Author response:

The following is the authors’ response to the previous reviews.

Minor Issues:

(1) As the authors mention, MKs have been suggested to mature rapidly at the sinusoids, and both integrin KO and laminin KO MKs appear mislocalized away from the sinusoids. Additionally, average MK distances from the sinusoid may also help separate whether the maturation defects could be in part due to impaired migration towards CXCL12 at the sinusoid. Presumably, MKs could appear mislocalized away from the sinusoid given the data presented suggesting they are leaving the BM and entering circulation. Additional commentary on intrinsic (ex-vivo) MK maturation phenotypes may help strengthen the author's conclusions

Thank you for your insightful suggestion regarding intrinsic MK maturation defects in integrin KO and laminin KO mice. This indeed could be the case. We have now addressed this possibility in the revised discussion section (page 14; lines 14-15), acknowledging intrinsic maturation defects as a potential contributor to observed maturation issues.

(2) It would be helpful if the authors could comment as to whether MKs are detectable in blood.

We appreciate the opportunity to clarify this point. Intact Itgb1^-/-/Itgb3^-/- MKs were not detected in the peripheral blood by either flow cytometry or blood smear analysis. This indicates that megakaryocytes do not normally circulate in the systemic bloodstream. Instead, we observed large MK nuclei trapped specifically within the lung capillaries, consistent with their known physical retention in the pulmonary circulation during platelet release. This explanation is now better explained on page 10, lines 14-19.

(3) Supplementary Figure 6 - shows no effect on in vitro MK maturation and proplt, or MK area - But Figures 6B/6C demonstrate an increase in total MK number in MMP-inhibitor treated mice compared to control. This discrepancy should be better discussed.

We have now expanded the discussion in the revised manuscript to address the different results obtained in vitro and in vivo, emphazing that the in vitro model may not fully recapitulate the complex and dynamic bone marrow ECM niche. Additionally, differences in the source and regulation of MMPs likely contribute to the differing outcomes, underlining the importance of studying these processes within their physiological context. For instance, non-megakaryocytic sources of MMPs and paracrine regulatory mechanisms may play a critical role within the physiological microenvironment, ultimately affecting MK proliferation and maturation in a manner not observed in simplified culture systems. This clarifications can be found on page 12, lines 6-17.

(4) A function of the ECM discussed relates to MK maturation but in the B1/3 integrin KO mice, the presence of the ECM cage is reduced but there appears to be no significant impact upon maturation (Supplementary Figure 4). By contrast, MMP inhibition in vivo (but not in vitro) reduces MK maturation. These data could be better clarified in the text.

Thank you for raising this important point. While Suppl. Figure 4 shows normal size and ploidy in DKO MK, a critical defect is revealed at the ultrastructural level. Mature DKO MKs exhibit severe dysplasia of the demarcation membrane system (DMS), characterized by extensive membrane accumulation and abnormal archirecture, with no typical platelet territories visible. This DMS defect directly impairs MK maturation and explains the thrombocytopenia observed in these mice. Increased emperipolesis further indicated disrupted maturation processes. These observations confirm the essential role of the ECM cage in supporting proper DMS organization and overall MK maturation in vivo, consistent with findings from MMP inhibition experiments. We have clarified and emphasized the significance of these DMS abnormalities in the revised manuscripts, including updated results (Page 9, lines 17-21) and a new EM image in Suppl. Figure 4.

Reviewer #1 (Public review):

The authors report on a thorough investigation of the interaction of megakaryocytes (MK) with their associated ECM during maturation. They report convincing evidence to support the existence of a dense cage-like pericellular structure containing laminin γ1 and α4 and collagen IV, which interacts with integrins β1 and β3 on MK and serve to fix the perisinusoidal localization of MK and prevent their premature intravasation. As with everything in nature, the authors support a Goldilocks range of MK-ECM interactions - inability to digest the ECM via inhibition of MMPs leads to insufficient MK maturation and development of smaller MK. This important work sheds light into the role of cell-matrix interactions in MK maturation, and suggests that higher-dimensional analyses are necessary to capture the full scope of cellular biology in the context of their microenvironment. The authors have responded appropriately to the majority of my previous comments.

We sincerely thank the reviewer for their insightful comments.

Some remaining points:

In a previous critique, I had suggested that "it is unclear how activation of integrins allows the MK to become "architects for their ECM microenvironment" as the authors posit. A transcriptomic analysis of control and DKO MKs may help elucidate these effects". The authors pointed out the technical difficulty of obtained sufficient numbers of MK for such analysis, which I accept, and instead analyzed mature platelets, finding no difference between control and DKO platelets. This is not necessarily surprising, since mature circulating platelets have no need to engage an ECM microenvironment, and for the same reason I would suggest that mature platelet analyses are not representative of MK behavior as regards ECM interactions.

We fully agree with the reviewer that platelet analyses do not accurately reflect the behavior of MKs in the context of interactions with the ECM. This understanding is also one of the reasons why we chose not to include RT-PCR data on platelets in our manuscript. Instead, we emphasize the role of integrins as essential regulators of ECM remodeling, as they transmit traction forces that can significantly influence this process. We also report reduced RhoA activation in DKO MK, which is likely to affect ECM organization. We believe that these explanations contribute to a clearer understanding of how integrin activation enables megakaryocytes to act as "architects" of their ECM microenvironment.

Reviewer #2 (Public review):

This study makes a significant contribution to understanding the microenvironment of megakaryocytes (MKs) in the bone marrow, identifying an extracellular matrix (ECM) cage structure that influences MK localization and maturation. The authors provide compelling evidence for the presence of this ECM cage and its role in MK homeostasis, employing an array of sophisticated imaging techniques and molecular analyses.The authors have addressed most of the concerns raised in the previous review, providing clarifications and additional data that strengthen their conclusion.

More broadly, this work adds to a growing recognition of the ECM as an active participant in haematopoietic cell regulation in the bone marrow microenvironment. This work could pave the way to future studies investigating how the megakaryocytes' ECM cage affects their function as part of the haematopoietic stem cell niche, and by extension, influences global haematopoiesis.

We thank this reviewer for providing such constructive feedback.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review): 

Summary: 

This study shows a novel role for SCoR2 in regulating metabolic pathways in the heart to prevent injury following ischemia/reperfusion. It combines a new multi-omics method to determine SCoR2 mediated metabolic pathways in the heart. This paper would be of interest to cardiovascular researchers working on cardioprotective strategies following ischemic injury in the heart. 

Strengths:

(1) Use of SCoR2KO mice subjected to I/R injury. 

(2) Identification of multiple metabolic pathways in the heart by a novel multi-omics approach.

We thank the Reviewer for the positive review of our manuscript.

Weaknesses:

(1) Use of a global SCoR2KO mice is a limitation since the effects in the heart can be a combination of global loss of SCoR2. 

(2) Lack of a cell type specific effect. 

We agree that global KOs limit the cell type-specific mechanistic conclusions that can be drawn. Global knockouts are nonetheless informative in their own right and serve to identify phenotypes worthy of further study.

Reviewer #2 (Public review):

Summary: 

This manuscript addresses the gap in knowledge related to the cardiac function of the S-denitrosylase SNOCoA Reductase 2 (SCoR2; product of the Akr1a1 gene). Genetic variants in SCoR2 have been linked to cardiovascular disease, yet their exact role in the heart remains unclear. This paper demonstrates that mice deficient in SCoR2 show significant protection in a myocardial infarction (MI) model. SCoR2 influenced ketolytic energy production, antioxidant levels, and polyol balance through the S-nitrosylation of crucial metabolic regulators. 

Strengths: 

(1) Addresses a well-defined gap in knowledge related to the cardiac function of SNO-CoA Reductase 2. Besides the in-depth case for this specific player, the manuscript sheds more light on the links between Snitrosylation and metabolic reprogramming in the heart.

(2) Rigorous proof of requirement through the combination of gene knockout and in vivo myocardial ischemia/reperfusion. 

(3) Identification of precise Cys residue for SNO-modification of BDH1 as SCoR2 target in cardiac ketolysis 

We thank the Reviewer for their kind words.

Weaknesses: 

(1) The experiments with BDH1 stability were performed in mutant 293 cells. Was there a difference in BDH1 stability in myocardial tissue or primary cardiomyocytes from SCoR2-null vs -WT mice? The same question extends to PKM2. 

We have not assessed BDH1 stability directly in cardiomyocytes. However, S-nitrosylation increased BDH1 stability in HEK293 cells, and BDH1 expression was increased in (injured) hearts of SCoR2KO mice, together with increased SNO-BDH1. 

For PKM2, there is a wealth of published evidence from us and others that S-nitrosylation does not regulate protein stability but rather inhibits tetramerization required for full activity.  

(2) In the absence of tracing experiments, the cross-sectional changes in ketolysis, glycolysis, or polyol intermediates presented in Figures 4 and 5 are suggestive at best. This needs to be stressed while describing and interpreting these results. 

We now acknowledge this limitation in the ‘Limitations’ section of the manuscript and in edits made to the text. 

(3) The findings from human samples with ischemic and non-ischemic cardiomyopathy do not seem immediately or linearly in line with each other and with the model proposed from the KO mice. While the correlation holds up in the non-ischemic cardiomyopathy (increased SNO-BDH1, SNO-PKM2 with decreased SCoR2 expression), how do the authors explain the decreased SNO-BDH1 with preserved SCoR2 expression in ischemic cardiomyopathy? This seems counterintuitive as activation of ketolysis is a quite established myocardial response to ischemic stress. It may help the overall message clarity to focus the human data part on only NICM patients. 

We find it interesting and important that SNO-BDH1 is readily detected in human heart tissue and its level is correlated to disease state. Our findings suggest conservation of this mechanism in human heart failure. However, we caution against drawing further conclusions related to NICM or ICM. Our animal model (based on a single time point) cannot faithfully recapitulate patients with chronic heart disease or differences between NICM and ICM. 

(4) This is partially linked to the point above. An important proof that is lacking at present is the proof of sufficiency for SCoR2 in S-nitrosylation of targets and cardiac remodeling. Does SCoR2 overexpression in the heart or isolated cardiomyocytes reduce S-nitrosylation of BDH1 and other targets, undermining heart function at baseline or under stress? 

The Reviewer proposes to test the effect of SCoR2 overexpression on cardioprotection. This is an interesting experiment for future study with the following caveats. First, it presupposes that native expression of SCoR2 is insufficient to control basal steady state S-nitrosylation of SNO-BDH1 and SNO-PKM2 (this does not seem to be the case). Second, overexpressed SCoR2 may be mislocalized within cells or associated with unnatural targets. Thank you.

Reviewer #3 (Public review): 

Summary: 

This manuscript demonstrates that mice lacking the denitrosylase enzyme SCoR2/AKR1A1 demonstrate a robust cardioprotection resulting from reprogramming of multiple metabolic pathways, revealing widespread, coordinated metabolic regulation by SCoR2. 

Strengths: 

(1) The extensive experimental evidence. 

(2) The use of the knockout model. 

We thank the Reviewer for identifying strengths in our work.

Weaknesses: 

(1) The connection of direct evidence for the mechanism. 

We believe we have identified a novel mechanism for cardioprotection entailing coordinate reprogramming of multiple metabolic pathways and suggesting a widescale role for SCoR2 in metabolic regulation. This is the key message we convey. While genetic dissection of individual pathways may be worthwhile, these investigations will have their own limitations. 

(2) The mouse model used is not tissue-specific. 

Please see our response to Reviewer 1, above. 

Reviewer #1 (Recommendations for the authors):

In the study, titled "The denitrosylase SCoR2 controls cardioprotective metabolic reprogramming", Grimmett ZW et al., describe a role for SNO-CoA Reductase 2 (SCoR2) in promoting cardioprotection via metabolic reprogramming in the heart after I/R injury. Authors show that loss SCoR2 coordinates multiple metabolic pathways to limit infarct size. Overall, the hypothesis is interesting, however there are some limitations as described below: 

(1) It is unclear whether SCoR2 mice are global or cardiomyocyte specific. 

We apologize for any confusion. These are global SCoR2^-/- mice. This is now stated in the Results when first identifying the strain, as well as in the Methods.  

(2) Can the authors clarify how divergent metabolic pathways such as Ketone oxidation, glycolysis, PPP and polyol metabolism work downstream of SCoR2 to impact cardioprotection in mice with I/R. 

The metabolic pathways of ketone oxidation, glycolysis, PPP and polyols appear to converge to support ischemic cardioprotection in SCoR2^-/- mice, as depicted in the model shown in Fig. 5L. Subsequent to SNO-PKM2 blockade of flux through glycolysis (detailed in this manuscript and in Zhou et al, 2019, PMID: 30487609, as well as by others), substrates of ketolysis and glycolysis are funneled into the PPP, producing the antioxidant NADPH and energy precursor phosphocreatine, which are well-known to be cardioprotective. This occurs more readily in SCoR2^-/- mice due to elevated SNO-BDH1 (detailed in this manuscript). 

Polyols, thought to be products of the PPP carbohydrate intermediates arabinose, ribulose, xylulose (among others), have recently been shown to be harmful to cardiovascular health in humans. These polyols are uniformly downregulated in SCoR2^-/- mice. We suggest this is likely the result of S-nitrosylation of SCoR2-substrate enzymes that form polyols (SCoR2/Akr1a1 is unable to directly reduce carbohydrates to their corresponding polyols). Regulation of endogenous polyol production in humans is a new concept and the mechanisms whereby these compounds increase risk of cardiac events are a subject of active investigation. This is detailed in the final paragraph of both the Results and Discussion sections, and in Fig. 5L. 

(3) The only functional outcome of SCoR2 loss in echocardiography and measurements for apoptosis. However, it would be important to determine whether the cardioprotective effect persists. It seems cardiac function was recorded 24hours post injury and whether the benefit remains till later time point such as 2 or 4 weeks is not shown. Without this time point, loss of SCoR2 only leads to an acute increment in function. 

Loss of SCoR2 reduced post-MI mortality at 4 hr; cardiac functional changes (plus troponin, LDH, and apoptosis) were studied in surviving animals at 24 hr post-MI. Cardiac response to acute injury and to chronic injury (weeks post-MI) are not the same metabolically. This is well elucidated in the literature and exemplified by the role of PKM2, which is protective in the chronic response to MI (28 days post-MI; PMID: 32078387), but implicated in injury at shorter timepoints post-MI (PMID: 33288902, 28964797). All that said, functional changes at 2-4 weeks will be important to determine in the future, as the Reviewer indicates. 

Reviewer #2 (Recommendations for the authors): 

(1) The last paragraph of the Results section should be divided into the statement related to Table S2 in the Results section, and the rest of the paragraph should be put somewhere in the Discussion. 

Thank you for this suggestion, which we have taken. 

(2) The number of mice alive/dead should be reported in the histogram in Figure 1G. 

Done.

(3) A concise Graphical Abstract will be useful to grasp the overall logic and message of the manuscript from the beginning. 

We thank you for this suggestion and have added a graphical abstract to the manuscript.

Reviewer #3 (Recommendations for the authors): 

I would suggest having more evidence on the effect of metabolic reprogramming on which cell type. The use of a global knockout is a major limitation, and probably some in vitro experiments with shRNA knockdown in endothelial cells and fibroblasts would provide more insights. 

The reviewer suggests one direction for future study. We identify a novel mechanism for cardioprotection entailing coordinate reprogramming of multiple metabolic pathways and suggesting a widescale role for SCoR2 in metabolic regulation. This is the message we wish to convey. The role of cardiomyocytes vs contributing cell types is a thoughtful direction for future study. Thank you. 

Editor's additional comment:

The editors wish to highlight a critical issue concerning the characterization of the SCoR2−/− mice employed in this study. 

In the Methods section (page 20), the manuscript states that "SCoR2+/− mice were made by Deltagen, Inc. as described previously (33)." However, reference 33 does not describe SCoR2−/− mice; instead, it refers to other genetically modified strains, including Akr1a1+/−, eNOS−/−, and PKM2−/− mice, with no mention of a SCoR2-targeted model. 

The editors fully acknowledge that the authors may be using the term "SCoR2" as a functional synonym for Akr1a1, based on its described role as a mammalian homologue of yeast SCoR. If this is the case, such equivalence should be explicitly stated in the manuscript to prevent potential confusion. Moreover, considering that the genetic deletion of Akr1a1 (i.e., SCoR2) underlies the key mechanistic findings presented, it is essential that the manuscript include a clear and comprehensive description of the generation and validation of the mouse model used. 

We therefore ask the authors to (1) clarify the nomenclature and relationship between "SCoR2" and Akr1a1, and (2) provide full details on the generation of the knockout mice, including the targeting strategy and the genotyping procedures. This information is necessary not only to ensure transparency and reproducibility but also to allow readers to fully appreciate the biological relevance of the findings.

Thank you for identifying this inconsistency. We have adjusted the manuscript text accordingly to clearly state that SCoR2 is a functional name for the product of the Akr1a1 gene and that these SCoR2^-/- mice are the same as Akr1a1^-/- mice described in Ref 33. We have augmented the Methods text to describe the generation and genotyping of these SCoR2/Akr1a1 knockout mice.

Author response:

The following is the authors’ response to the original reviews

Reviewer #1 (Public review):

Summary:

There is growing appreciation for the important of luminal (apical) ECM in tube development, but such matrices are much less well understood than basal ECMs. Here the authors provide insights into the aECM that shapes the Drosophila salivary gland (SG) tube and the importance of PAPSS-dependent sulfation in its organization and function.

The first part of the paper focuses on careful phenotypic characterization of papss mutants, using multiple markers and TEM. This revealed reduced markers of sulfation and defects in both apical and basal ECM organization, Golgi (but not ER) morphology, number and localization of other endosomal compartments, plus increased cell death. The authors focus on the fact that papss mutants have an irregular SG lumen diameter, with both narrowed regions and bulged regions. They address the pleiotropy, showing that preventing the cell death and resultant gaps in the tube did not rescue the SG luminal shape defects and discussing similarities and differences between the papss mutant phenotype and those caused by more general trafficking defects. The analysis uses a papss nonsense mutant from an EMS screen - I appreciate the rigorous approach the authors took to analyze transheterozygotes (as well as homozygotes) plus rescued animals in order to rule out effects of linked mutations. Importantly, the rescue experiments also demonstrated that sulfation enzymatic activity is important.

The 2nd part of the paper focuses on the SG aECM, showing that Dpy and Pio ZP protein fusions localize abnormally in papss mutants and that these ZP mutants (and Np protease mutants) have similar SG lumen shaping defects to the papss mutants. A key conclusion is that SG lumen defects correlate with loss of a Pio+Dpy-dependent filamentous structure in the lumen. These data suggest that ZP protein misregulation could explain this part of the papss phenotype.

Overall, the text is very well written and clear. Figures are clearly labeled. The methods involve rigorous genetic approaches, microscopy, and quantifications/statistics and are documented appropriately. The findings are convincing.

Significance:

This study will be of interest to researchers studying developmental morphogenesis in general and specifically tube biology or the aECM. It should be particularly of interest to those studying sulfation or ZP proteins (which are broadly present in aECMs across organisms, including humans).

This study adds to the literature demonstrating the importance of luminal matrix in shaping tubular organs and greatly advances understanding of the luminal matrix in the Drosophila salivary gland, an important model of tubular organ development and one that has key matrix differences (such as no chitin) compared to other highly studied Drosophila tubes like the trachea.

The detailed description of the defects resulting from papss loss suggests that there are multiple different sulfated targets, with a subset specifically relevant to aECM biology. A limitation is that specific sulfated substrates are not identified here (e.g. are these the ZP proteins themselves or other matrix glycoproteins or lipids?); therefore, it's not clear how direct or indirect the effects of papss are on ZP proteins. However, this is clearly a direction for future work and does not detract from the excellent beginning made here.

Comments on revised version:

Overall, I am pleased with the authors' revisions in response to my original comments and those of the other reviewers

Reviewer #2 (Public review):

Summary

This study provides new insights into organ morphogenesis using the Drosophila salivary gland (SG) as a model. The authors identify a requirement for sulfation in regulating lumen expansion, which correlates with several effects at the cellular level, including regulation of intracellular trafficking and the organization of Golgi, the aECM and the apical membrane. In addition, the authors show that the ZP proteins Dumpy (Dpy) and Pio form an aECM regulating lumen expansion. Previous reports already pointed to a role for Papss in sulfation in SG and the presence of Dpy and Pio in the SG. Now this work extends these previous analyses and provides more detailed descriptions that may be relevant to the fields of morphogenesis and cell biology (with particular focus on ECM research and tubulogenesis). This study nicely presents valuable information regarding the requirements of sulfation and the aECM in SG development.

Strengths

-The results supporting a role for sulfation in SG development are strong. In addition, the results supporting the involvement of Dpy and Pio in the aECM of the SG, their role in lumen expansion, and their interactions, are also strong.

-The authors have made an excellent job in revising and clarifying the many different issues raised by the reviewers, particularly with the addition of new experiments and quantifications. I consider that the manuscript has improved considerably.

-The authors generated a catalytically inactive Papss enzyme, which is not able to rescue the defects in Papss mutants, in contrast to wild type Papss. This result clearly indicates that the sulfation activity of Papss is required for SG development.

Weaknesses

-The main concern is the lack of clear connection between sulfation and the phenotypes observed at the cellular level, and, importantly, the lack of connection between sulfation and the Pio-Dpy matrix. Indeed, the mechanism/s by which sulfation affects lumen expansion are not elucidated and no targets of this modification are identified or investigated. A direct (or instructive) role for sulfation in aECM organization is not clearly supported by the results, and the connection between sulfation and Pio/Dpy roles seems correlative rather than causative. As it is presented, the mechanisms by which sulfation regulates SG lumen expansion remains elusive in this study.

-In my opinion the authors overestimate their findings with several conclusions, as exemplified in the abstract:

"In the absence of Papss, Pio is gradually lost in the aECM, while the Dpy-positive aECM structure is condensed and dissociates from the apical membrane, leading to a thin lumen. Mutations in dpy or pio, or in Notopleural, which encodes a matriptase that cleaves Pio to form the luminal Pio pool, result in a SG lumen with alternating bulges and constrictions, with the loss of pio leading to the loss of Dpy in the lumen. Our findings underscore the essential role of sulfation in organizing the aECM during tubular organ formation and highlight the mechanical support provided by ZP domain proteins in maintaining luminal diameter."

The findings leading to conclude that sulfation organizes the aECM and that the absence of Papss leads to a thin lumen due to defects in Dpy/Pio are not strong. The authors certainly show that Papss is required for proper Pio and Dpy accumulation. They also show that Pio is required for Dpy accumulation, and that Pio and Dpy form an aECM required for lumen expansion. However, the absence of Pio and Dpy do not fully recapitulate Papss mutant defects (thin lumen). I wonder whether other hypothesis and models could account for the observed results. For instance, a role for Papss affecting secretion, in which case sulfation would have an indirect role in aECM organization. This study does not address the mechanical properties of Dpy in normal and mutant salivary glands.

-Minor issues relate to the genotype/phenotype analysis. It is surprising that the authors detect only mild effects on sulfation in Papss mutants using an anti-sulfoTyr antibody, as Papss is the only Papss synthathase. Generating germ line clones (which is a feasible experiment) would have helped to prove that this minor effect is due to the contribution of maternal product. The loss of function allele used in this study seems problematic, as it produces effects in heterozygous conditions difficult to interpret. Cleaning the chromosome or using an alternative loss of function condition (another allele, RNAi, etc...) would have helped to present a more reliable explanation.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Overall, I am pleased with the authors' revisions in response to my original comments and those of the other reviewers. The addition of the sulfation(-) mutant to Fig. 1 is particularly nice. I have just a few additional suggestions for text changes to improve clarity/precision.

(1) The current title of this manuscript is quite broad, making it sound like a review article. I recommend adding sulfation and salivary gland to the title to convey the main points more clearly. e.g. Sulfation affects apical extracellular matrix organization during development of the Drosophila salivary gland tube.

Thank you for the suggestion. We agree and have changed the title of the paper as suggested.

(2) Figure 1B shows very striking enrichment of papss expression in the salivary gland compared to other tubes like the trachea that also contain Pio and Dpy. To me, this implies that the key substrate(s) of Papss are likely to be unique, or at least more highly enriched, in the salivary gland aECM compared to the tracheal aECM (e.g. probably not Pio or Dpy themselves). I suggest that the authors address the implications of this apparent SG specificity in the discussion (paragraph beginning on p. 21, line 559).

Yes, we agree that there may be other key substrates of Papss in the SG, such as mucins, which play an important role in organizing the aECM and expanding the lumen. We have included a discussion.

(3) p. 15, lines 374-376 "The Pio protein is known to be cleaved, at one cleavage site after the ZP domain by the furin protease and at another cleavage site within the ZP domain by the matriptase Notopleural (Np) (Drees et al., 2019; Drees et al., 2023; Figure 5B)." As far as I can see, the Drees papers show that Pio is cleaved somewhere in the vicinity of a consensus furin cleavage site, but do not actually establish that the cleavage happens at this exact site or is done by a furin protease (this is just an assumption). Please word more carefully, e.g. "at one cleavage site after the ZP domain, possibly by a furin protease".

Thank you for pointing this out. We have edited the text.

Reviewer #2 (Recommendations for the authors):

Throughout the paper, I find a bit confusing the description of the lumen phenotype and their interpretations.

Papss mutants produce SG that are either "thin" or show "irregular lumen with bulges". Do the authors think that these are two different manifestations of the same effect? or do they think that there are different causes behind?

The thin lumen phenotype appears to occur when the Pio-Dpy matrix is significantly condensed. When this matrix is less condensed in one region of the lumen than in other regions, the lumen appears irregular with bulges.

Are the defects in Grasp65 mutants categorized as "irregular lumen with bulges" similar to those in Papss mutants? Why do these mutants don't show a "thin lumen" defect?

Grasp65 mutant phenotypes are milder than those of Papss mutants. Multiple mutations in several Golgi components that more significantly disrupt Golgi structures and function may cause more severe defects in lumen expansion and shape.

How the defects described for Pio ("multiple constrictions with a slight expansion between constrictions") and Dpy mutants ("lumen with multiple bulges and constrictions") relate to the "irregular lumen with bulges" in Papss mutants?

pio and dpy mutants show more stereotypical phenotypes, while Papss mutants exhibit more irregular and random phenotypes. The irregular lumen phenotypes in Papss mutants are associated with a condensed Pio-Dpy matrix.

Author response:

Reviewer #1 (Public review):

Summary:

The authors of this study propose a model in which NPY family regulators antagonize the activity of the pid mutation in the context of floral development and other auxin-related phenotypes. This is hypothesized to occur through regulation of or by PID and its action on the PIN1 auxin transporter.

Strengths:

The findings are intriguing.

We are pleased that the reviewer found the work interesting!

Weaknesses and Major Comments:

(1) While the findings are indeed intriguing, the mechanism of action and interaction among these components remains poorly understood. The study would benefit from significantly more thorough and focused experimental analyses to truly advance our understanding of pid phenotypes and the interplay among PID, NPYs, and PIN1.

Elucidating the mechanism of action and interaction among these components will require years of additional research. As key steps toward these goals, our work clearly established that 1) NPY1 functions downstream of PID, as overexpression of NPY1 completely suppressed pid phenotypes. This is surprising because the predominant model is that PID functions by directly phosphorylating and activating PINs without the need of NPY1 involvement.  2) In the absence of PID, NPY1 protein accumulated less in the NPY1 OE lines, suggesting that PID plays a role in affecting NPY1 stability/degradation/accumulation. We are not sure what are the exact experiments this reviewer is proposing.

Regarding pid phenotypes, pid is completely sterile in our conditions, while the suppression by NPY1 OE is very clear and the lines are fertile.

(2) The manuscript appears hastily assembled, with key methodological and conceptual details either missing or inconsistent. Although issues with figure formatting and clarity (e.g., lack of scale bars and inconsistent panel layout) may alone warrant revision, the content remains the central concern and must take precedence over presentation.

We did not include scale bars in our figures because the phenotype of interest is presence/absence of flowers. Readers should compare the mutants with the rescued plants and the WT plants.

(3) Given that fertile progeny are obtained from pid-TD pin1/PIN1 and pid NPY OE lines, it would be important to analyze whether mutations and associated phenotypes are heritable. This is especially relevant since CRISPR lines can be mosaic. Comprehensive genotyping and inheritance studies are required.

We only use stable, heritable, Cas9-free mutants in our studies.  We genotype our mutants in every generation.  More details have been added to the Materials and Methods section. We provide the genetic materials we use to the scientific community when requested to enable verification and extension of our results. 

(4) The Materials and Methods section lacks essential information on how the lines were generated, genotyped, propagated, and scored. There is also generally no mention of how reproducible the observations were. These genetic experiments need to be described in detail, including the number of lines analyzed and consistency across replicates.

More details have been added to the Materials and Methods section

The criticism is not fully accurate. For example, we stated in the main text: “We genotyped T2 progenies from two pid-c1 heterozygous T1 plants (#68 and # 83) for the presence of pid-c1 and for pid-c1 zygosity. We used mCherry signal, which was included in the NPY1 OE construct, as a proxy to determine the presence and absence of the NPY1 transgene. For each line, we identified T2 plants without the NPY1 transgene and without the pid-c1 mutation (called WT-68 and WT-83, respectively). We also isolated T2 plants that contained the NPY1 overexpression construct, but did not have the pid-c1 mutation (called NPY1 OE #68 in WT, and NPY1 OE #83 in WT). Finally, we identified T2 plants that were pid-c1 homozygous and that had the NPY1 transgene (called NPY1 OE #68 in pid-c1 and NPY1 OE #83 in pid-c1). These genetic materials enabled us to compare the same NPY1 OE transgenic event in different genetic backgrounds.”

The genetic materials used are freely available to the scientific community.  We would like to point out that we used several pin1 and pid alleles to make sure that the phenotypes are caused by the genes of interest.

(5) The nature of the pid alleles used in the study is not described. This is essential for interpretation.

The mutants were described in a previous paper (M. Mudgett, Z. Shen, X. Dai, S.P. Briggs, & Y. Zhao, Suppression of pinoid mutant phenotypes by mutations in PIN-FORMED 1 and PIN1-GFP fusion, Proc. Natl. Acad. Sci. U.S.A. 120 (48) e2312918120, https://doi.org/10.1073/pnas.2312918120 (2023).  We have added the relevant information to Materials and Methods.

(6) The authors measure PIN1 phosphorylation in response to NPY overexpression and conclude that the newly identified phosphorylation sites are inhibitory because they do not overlap with known activating sites. This conclusion is speculative without functional validation. Functional assays are available and must be included to substantiate this claim.

We concluded that the phosphorylation of PINs in NPY1 OE is inhibitory on the basis of the following: 1) pid is suppressed in pin1 heterozygous backgrounds and by PIN1-GFP_HDR, demonstrating that partial loss of function of PIN1 or a decrease in PIN1 gene dosage, which decreases PIN1 protein expression, caused the suppression of pid. 2) pid is completely suppressed by NPY1 OE, which caused an increase of PIN phosphorylation, suggesting that phosphorylation of PINs in NPY1 OE lines is inhibitory.  It is true that we do not have biochemical data to support the conclusion. We would like to point out that the phosphorylation sites in PINs identified in this work do overlap with previously identified sites.

PIN activity assays are conducted in heterologous systems that do not include NPY proteins. Since NPY is important for PIN activities, we believe that these assays may provide misleading results. Moreover, PIN1 is likely part of a large protein complex.  Without knowing the composition of the complex, functional assays in heterologous systems will not be interpretable.

(7) Figure 5 implies that NPY1 acts downstream of PID, but there is no biochemical evidence supporting this hierarchy. Additional experiments are needed to demonstrate the epistatic or regulatory relationship.

We show that overexpression of NPY1 completely suppressed the pid phenotype, and this epistatic relationship indicates that NPY1 functions downstream of PID. Moreover, we report that PID is required for NPY1 accumulation, indicating that PID is upstream of NPY1.

(8) The authors should align their genetic observations with cell biological data on PIN1, PIN2, and PID localization and distribution.

We are hesitating in using traditional PIN1-GFP, PIN2-GFP lines, as they are not stable in our hands. Localization of PID is still not clear. We have generated PID-GFP_HDR lines, but we could not detect any fluorescent signals (unpublished results).  In addition, maize PINOID (BIF2) localizes to the nucleus, cytoplasm and cell periphery (Skirpan, A., Wu, X. and McSteen, P. (2008), Genetic and physical interaction suggest that BARREN STALK1 is a target of BARREN INFLORESCENCE2 in maize inflorescence development. The Plant Journal, 55: 787-797. https://doi.org/10.1111/j.1365-313X.2008.03546.x)

We would rather wait for the proper genetic materials before devoting our effort to this.

Reviewer #2 (Public review):

Summary:

The study is well-conducted, revealing that NPY1, with previously less-characterized molecular functions, can suppress pid mutant phenotypes with a phosphorylation-based mechanism. Overexpression of NPY1 (NPY1-OE) results in PIN phosphorylation at unique sites and bypasses the requirement for PID for this event. Conversely, a C-terminal deleted form of NPY1 (NPY1-dC) fails to rescue pid despite promoting a certain phospho-profile in PIN proteins.

Strengths:

(1) The careful genetic analyses of pid suppression by NPY1-OE and the inability of NPY1dC to do the same.

(2) Phospho-proteomics approaches reveal that NPY1-OE induces phosphorylation of PINs at non-canonical sites, independent of PID.

Thank you for having accurately summarized the main findings

Weaknesses:

(1) The native role of NPY1 is not tested by phospho-proteomics in loss-of-function npy1 mutants. Such analysis would be crucial to demonstrate that NPY1 is required for the observed phosphorylation events.

This is an excellent point and we agree with the reviewer that analyzing loss-of-function npy mutants is important. The challenge is that we need to knockout NPY1, NPY3, and NPY5 to phenocopy pid. We will also need to find a way to suppress the npy triple mutants, which are sterile, so that we can have meaningful comparisons.

(2) The functional consequences of the newly identified phosphorylation sites in PINs remain speculative. Site-directed mutagenesis (phospho-defective and phospho-mimetic) would help clarify their physiological roles.

We agree with the reviewer on this point as well. However, this is not trivial, as we have uncovered so many phosphorylation sites.

(3) The kinase responsible for NPY1-mediated phosphorylation remains unidentified. Since NPY1 is a non-kinase protein, a model involving recruitment of partner kinases (e.g., PIN-phosphorylating kinases other than PID) should be considered or discussed.

we will add a sentence to mention D6PK and other kinases in the Discussion in the revised version.  We are hoping that the kinases will come out of future forward genetic screens.

Reviewer #3 (Public review):

Summary:

This manuscript from Mudgett et al. explores the relative roles of PID and NPY1 in auxin-dependent floral initiation in Arabidopsis. Micro vectorial auxin flows directed by PIN1 are essential to flower initiation, and loss of PIN1 or two of its regulators, PID and NPY1 (in a yucca-deficient background) phenocopies the pinformed phenotype. This group has previously shown that PID-PIN1 interactions and function are dosage-dependent. The authors pick up this thread by demonstrating that a heterozygote containing a CRISPR deletion of one copy of PIN1 can restore quasi-wild type floral initiation to pid.

The authors then show that overexpression of NPY1 is sufficient to more or less restore wild-type floral initiation to the pid mutant. The authors claim that this result demonstrates that NPY1 functions downstream of PID, as this ectopic abundance of NPY1 resulted in phosphorylation of PIN1 at sites that differ from sites of action of PID. The authors pursue evidence that PID action via NPY1 is analogous to the mode of action by which phot1/2 act on NPH3 in seedling phototropism. Such a model is supported by the evidence presented herein that the C terminus of NPY1, which has abundant Ser/Thr content, is phosphorylated, and that the deletion of this domain prevents overexpression compensation of the pinformed phenotype.

While the results presented support evidence in the literature that PID acts on NPY1 to regulate PIN1 function, it is also possible that NPY1 overexpression results in limited expansion of phosphorylation targets observed with other AGC kinases. And if the phot model is any indication, there may be other PID targets that modulate PIN1-dependent floral initiation.

However, overexpression of the NPY1 C-terminal deletion construct resulted in phosphorylation of both PIN1 and PIN2 and agravitropic root growth similar to what is observed in pin2 mutants. This suggests that direct PID phosphorylation of PINs and action via NPY1 can be distinguished by phosphorylation sites and by growth phenotypes.

Strengths:

A very important effort that places NPY1 downstream of PID in floral initiation.

We thank the reviewer for the comments.

Weaknesses:

As PID has been shown to act on sites that regulate PIN protein polarity as well as PIN protein function, it would be useful if the authors consider how their results would fit/not fit with a model where combinatorial function of NPY1 and PID regulate PIN1 in a manner similar to the way that PID appears to function combinatorially with D6PK on PIN3

We agree with the reviewer that we do not have a complete picture of how NPY, PID, PIN work together to control flower initiation. Some aspects of our results are difficult to reconcile with the model of PIN1 and PID acting in tandem, i.e., by PID directly phosphorylating and activating PIN1. Indeed, our results suggest that PIN1 and PID have opposite effects on organogenesis. For example, heterozygous pin1 (or PIN1-GFP_HDR, which is presumably less active than wild type PIN1) suppresses the pid phenotype.  Moreover, pid and pin1 have opposite effects on cotyledon number and true leaf number. Mutations in PID lead to more cotyledons and more true leaves than WT whereas pin1 mutants make fewer cotyledons and fewer true leaves than WT (Bennett SRM, Alvarez J, Bossinger G, Smyth DR (1995) Morphogenesis in pinoid mutants of Arabidopsis thaliana. The Plant Journal 8: 505-520).  We have elaborated on this point in the last paragraph of the Discussion.

The genetic materials we have generated may allow us to uncover additional components in the pathway from forward genetic screens, which may eventually lead to a clear picture.

Author response:

(1) General Statements

Our manuscript studies mechanisms of planar polarity establishment in vivo in the Drosophila pupal wing. Specifically we seek to understand mechanisms of ‘cell-scale signalling’ that is responsible for segregating core pathway planar polarity proteins to opposite cell edges. This is an understudied question, in part because it is difficult to address experimentally.

We use conditional and restrictive expression tools to spatiotemporally manipulate core protein activity, combined with quantitative measurement of core protein distribution, polarity and stability. Our results provide evidence for a robust cell-scale signal, while arguing against mechanisms that depend on depletion of a limited pool of a core protein or polarised transport of core proteins on microtubules. Furthermore, we show that polarity propagation across a tissue is hard, highlighting the strong intrinsic capacity of individual cells to establish and maintain planar polarity.

The original manuscript received three fair and thorough peer-reviews, which raised many important points. In response, we decided to embark on a full revision that attempts to answer all of the points. We have included new data to support our conclusions in Supplemental Figures 1, 2 and 5.

Additionally in response to the reviewers we have revised the manuscript title, which is now ‘Characterisation of cell-scale signalling by the core planar polarity pathway during Drosophila wing development’.

(2) Point-by-point description of the revisions

We thank all of the reviewers for their thorough and thoughtful review of our manuscript. They raise many helpful points which have been extremely useful in assisting us to revise the manuscript.

In response we have carried out a major revision of the manuscript, making numerous changes and additions to the text and also adding new experimental data. Specific changes are listed after our detailed response to each comment.

Reviewer #1:

Summary

The authors use inducible Fz::mKate2-sfGFP to explore "cell-scale signaling" in PCP. They reach several conclusions. First, they conclude that cell-scale signaling does not depend on limiting pools of core components (other than Fz). Second, they conclude that cell-scale signaling does not depend on microtubule orientation, and third, they conclude that cell-scale signaling is strong relative to cell to cell coupling of polarity. 

There are some interesting inferences that can be drawn from the manuscript, but there are also some significant challenges in interpreting the results and conclusions from the work as presented. I suggest that the authors 1) define "cell-scale signaling," as the precise meaning must be inferred, 2) reconsider some premises upon which some conclusions depend, 3) perform an essential assay validation, and 4) explain some other puzzling inconsistencies.

Major points

The exact meaning of cell-scale signaling is not defined, but I infer that the authors use this term to describe how what happens on one side of a cell affects another side. The remainder of my critique depends on this understanding of the intended meaning.

As the reviewer points out, it is important that the meaning of the term ‘cell-scale signalling’ is clear to the reader and in response to their comment we have had another go at defining it explicitly in the Introduction to the manuscript.

Specifically, we use the term ‘cell-scale signalling’ to describe possible intracellular mechanisms acting on core protein segregation to opposite cell membranes during core pathway dependent planar polarisation. For example, this could be a signal from distal complexes at one side of the cell leading to segregation of proximal complexes to the opposite cell edge, or vice versa. See also our response to Reviewer #2 regarding the distinction between ‘molecular-scale’ and ‘cell-scale’ signalling. 

Changes to manuscript: Revised definition of ‘cell-scale signalling’ in Introduction.

The authors state that any tissue wide directional information comes from pre-existing polarity and its modification by cell flow, such that the de novo signaling paradigm "bypasses" these events and should therefore not be responsive to any further global cues. It is my understanding that this is not a universally accepted model, and indeed, the authors' data seem to suggest otherwise. For example, the image in Fig 5B shows that de novo induction restores polarity orientation to a predominantly proximal to distal orientation. If no global cue is active, how is this orientation explained?

We assume that the reviewer’s point is that it is not universally accepted that de novo induction after hinge contraction leads to uncoupling from global cues (rather than that it is not accepted that hinge contraction remodels radial polarity to a proximodistal pattern). We are (we believe) the only lab that has used de novo induction as a tool, and we’re not aware of any debate in the literature about whether this bypasses global cues. Nevertheless, we accept that it is hard to prove there is no influence of global cues, when the nature of those cues and the time at which they act remain unclear. Below we summarise the reasons why we believe there are not significance effects of global cues in our experiments that would influence the interpretation of our results.

First, our reading of the literature supports a broad consensus that an early radial core planar polarity pattern is realigned by cell flow produced by hinge contraction beginning at around 16h APF (e.g. Aigouy et al., 2010; Strutt and Strutt, 2015; Aw and Devenport, 2017; Butler and Wallingford, 2017; Tan and Strutt, 2025). Taken at face value, this suggests that there are ‘radial’ cues present prior to hinge contraction, maybe coming from the wing margin – arguably these radial cues could be Ft-Ds or Wnts or both, given they are expressed in patterns consistent with such a role (notwithstanding the published evidence arguing against roles for either of these cues). It then appears that hinge contraction supercedes these cues to convert a radial pattern to a proximodistal pattern – whether the radial cues that affect the core pathway earlier remain active after hinge contraction is unclear, although both Ft-Ds and Wnts appear to maintain their ‘radial’ patterns beyond the beginning of hinge contraction (e.g. Merkel et al., 2014; Ewen-Campen et al., 2020; Yu et al., 2020).

We think that the reviewer is proposing the presence of a proximodistal cue that is active in the proximal region of the wing that we use for our experiments shown e.g. in Fig.5, and that this cue orients core polarity here (but not elsewhere in the wing) in a time window after 18h APF. Ft-Ds and Wnts do not seem to be plausible candidates as they are still in ‘radial’ patterns. This leaves either an unknown proximodistal cue (a gradient of some unknown signalling molecule?), or possibly some ability of hinge contraction to align proximodistal polarity specifically in this wing region but not elsewhere. We cannot definitively rule out either of these possibilities, but neither do we think there is sufficient evidence to justify invoking their existence to explain our observations.

In particular, the reason that we don’t think there is a proximodistal cue in the proximal part of the wing after 18h APF, is that work from our lab shows that induction of Fz or Stbm expression at times around or after the start of hinge contraction (i.e. >16 h APF) results in increasing levels of trichome swirling with polarity not being coordinated with the tissue axis either proximally or distally (Strutt and Strutt, 2002; Strutt and Strutt 2007). Our simplest interpretation for this is that induction at these stages fails to establish the early radial pattern of core pathway polarity and hence hinge contraction cannot reorient radial to proximodistal. If hinge contraction alone could specify proximodistal polarity in the absence of the earlier radial polarity, then we would not expect to see swirling over much of the proximal wing (where the forces from hinge contraction are strongest (Etournay et al., 2015)).

In this manuscript, our earliest de novo experiments begin with Fz induction at 18h APF (de novo 10h), then at 20h APF (de novo 8h) and at 22h APF (de novo 6h). The image in Fig. 5B, referred to by the reviewer, is of a wing where Fz is induced de novo at 22 h APF. In these wings, as expected, the core proteins localise asymmetrically in stereotypical swirling patterns throughout the wing surface (see Fig. 2M and also Strutt and Strutt, 2002; Strutt and Strutt 2007), but – usefully for our experiments – they broadly localise along the proximal-distal axis in the region analysed in Fig. 5B. Given the strong swirling in surrounding regions when inducing at >20h APF, we feel reasonably confident in assuming that the pattern is not due to a proximodistal cue present in the proximal wing.

We appreciate that the original manuscript did not show images including the trichome pattern in adjacent regions, so this point would not have been clear, but we now include these in Supplementary Fig. 5. We have also added a note in the legend to Fig. 5B to clarify that the proximodistal pattern seen is local to this wing region. We apologise for this oversight and the confusion caused and appreciate the feedback.

The 6 hr condition, that has only partial polarity magnitude, is quite disordered. Do the patterns at 8 and 10 hrs become more proximally-distally oriented? It is stated that they all show swirls, but please provide adult wing images, and the corresponding orientation outputs from QuantifyPolarity to help validate the notion that the global cues are indeed bypassed by this paradigm.

In all three ‘normal’ de novo conditions (6h, 8h and 10h), regardless of the time of induction, the polarity orientation patterns of Fz-mKate2 in pupal and adult wings are very similar in the experimentally analysed region (Fig. S5B-E). The strong local hair swirling agrees with the previous published data (Strutt and Strutt, 2002; Strutt and Strutt 2007). Overall, we don’t see any evidence that the 10h de novo induction results in more proximodistally coordinated polarity than the 8h or 6h conditions. This is consistent with our contention that there is no global cue present at these stages, which presumably would have a stronger effect when core pathway activity was induced at earlier stages.

Changes to manuscript: Added additional explanation of the ‘de novo induction’ paradigm and why we believe the resulting polarity patterns are unlikely to be influenced by any global signals in Introduction and Results section ‘Induced core protein relocalisation…’. Added quantification of polarity in the experiment region proximal to the anterior cross-vein in pupal wings (Fig.S5E-E’’’) and zoomed-out images of the surrounding region in adult wings showing that the polarity pattern does not become more proximodistal when induction time is longer, and also that there is not overall proximodistal polarity in proximal regions of the wing (Fig.S5B-D), arguing against an unknown proximodistal polarity cue at these stages of development.

In the de novo paradigm, polarization is initiated immediately or shortly after heat shock induction. However, the results should be differently interpreted if the level of available Fz protein does not rise rapidly and then stabilize before the 6 hr time point, and instead continues to rise throughout the experiment. Western blots of the Fz::mKate2-sfGFP at time points after induction should be performed to demonstrate steady state prior to measurements. Otherwise, polarity magnitude could simply reflect the total available pool of Fz at different times after induction. Interpreting stability is complex, and could depend on the same issue, as well as the amount of recycling that may occur. Prior work from this lab using FRAP suggested that turnover occurs, and could result from recycling as well as replenishment from newly synthesized protein. 

The reviewer raises an important point, which we agree could confound our experimental interpretations. As suggested we have now carried out western blotting and quantitation for Fz::mKate2-sfGFP levels and added these data to Fig.S1 (Fig. S1C,D). Quantified Fz is not significantly different between the three de novo polarity induction timings and not significantly different compared to constitutive Fz::mKate2-sfGFP expression (although there is a trend towards increasing Fz::mKate2-sfGFP protein levels with increasing induction times). These data are consistent with Fz::mKate2-sfGFP being at steady state in our experiments and that levels are sufficient to achieve normal polarity (as constitutive Fz::mKate2-sfGFP does so). Therefore it is unlikely that differing protein levels explain the differing polarity magnitudes at the different induction times. Interestingly, Fz::mKate2-sfGFP levels are lower than endogenous Fz levels, possibly due to lower expression or increased turnover/reduced recycling.

Changes to manuscript: Added western blot analysis of Fz::mKate2-sfGFP expression under 10h, 8h and 6h induction conditions vs endogenous Fz expression and constitutive Fz::mKate2sfGFP expression (Fig.S1C-D) and discussed in Results section ‘Planar polarity establishment is…’.

From the Fig 3 results, the authors claim that limiting pools of core proteins do not explain cellscale signaling, a result expected based on the lack of phenotypes in heterozygotes, but of course they do not test the possibility that Fz is limiting. They do note that some other contributing protein could be. 

Previously published results from our lab (Strutt et al., 2016 Cell Reports; Supplemental Fig. S6E) show that in a heterozygous fz mutant background, Fz protein levels are not affected by halving the gene dosage when compared to wt, suggesting that Fz is most likely produced in excess and is not normally limiting, but that protein that cannot form complexes may be rapidly degraded. We have now added this information to the text.

Changes to manuscript: Added explanation in text that Fz levels had previously been shown to not be dosage sensitive in Results section ‘Planar polarity establishment is…’ and also added a caveat to the Discussion about not directly testing Fz.

In Fig 3, it is unclear why the authors chose to test dsh1/+ rather than dsh[null]/+. In any case, the statistically significant effect of Dsh dose reduction is puzzling, and might indicate that the other interpretation is correct. Ideally, a range including larger and smaller reductions would be tested. As is, I don't think limiting Dsh is ruled out. 

Concerning the choice of dsh allele, we appreciate the query of the reviewer regarding use of dsh[1] instead of a null, as there might be a concern that dsh[1] would give a less strong phenotype. The answer is that over more than two decades we and others have never found any evidence that dsh[1] does not act as a ‘null’ for planar polarity in the pupal wing, and furthermore use of dsh[1] preserves function in Wg signalling – and we would prefer to rule out any phenotypic effects due to any potential cross-talk between the two pathways that might be seen using a complete null. To expand on this point, dsh[1] mutant protein is never seen at cell junctions (Axelrod 2001; Shimada et al., 2001; our own work), and by every criteria we have used, planar polarity is completely disrupted in hemizygous or homozygous mutants e.g. see quantifications of polarity in (Warrington et al., 2017 Curr Biol).

In terms of the broader point, whether we can rule out Dsh being limiting, we were very careful to be clear that we did not see evidence for Dsh (or other core proteins) being limiting in terms of ‘rates of core pathway de novo polarisation’. When the reviewer says ‘the statistically significant effect of Dsh dose reduction is puzzling’ we believe they are referring to the data in Fig. 3J, showing a small but significantly different reduction in stable Fz in de novo 6h conditions (also seen in 8h de novo conditions, Fig. S3I). As Dsh is known to stabilise Fz in complexes (Strutt et al., 2011 Dev Cell; Warrington et al., 2017 Curr Biol), in itself this result is not wholly surprising. Nevertheless, while this shows that halving Dsh levels does modestly reduce Fz stability, it does not alter our conclusion that halving Dsh levels does not affect Fz polarisation rate under either 6h or 8h de novo conditions.

Unfortunately, we do not have available to us a practical way of achieving consistent intermediate reductions in Dsh levels (e.g. a series of verified transgenes expressing at different levels). Levels of all the core proteins could be dialled down using transgenes, to see when the system breaks, and indeed we have previously published that lower levels of polarity are seen if Fmi levels are <<50% or if animals are transheterozygous for pk, stbm, dgo or dsh, pk, stbm, dgo simultaneously (Strutt et al., 2016 Cell Reports). However, it seems to be a trivial result that eventually the ability to polarise is lost if insufficient core proteins are present at the junctions. For this reason we have focused on a simple set of experiments reducing gene dosage singly by 50% under two de novo induction conditions, and have been careful to state our results cautiously. The assays we carried out were a great deal of work even for just the 5 heterozygous conditions tested.

We believe that the experiments shown effectively make the point that there is no strong dosage sensitivity – and it remains our contention that if protein levels were the key to setting up cell-scale polarity, then a 50% reduction would be expected to show an effect on the rate of polarisation. We further note that as Fz::mKate2-sfGFP levels are lower than endogenous Fz levels (see above), the system might be expected to be sensitised to further dosage reductions, and despite this we failed to see an effect on rate of polarisation.

We note that Reviewer #3 made a similar point about whether we can rule out dosage sensitivity on the basis of 50% reductions in protein level. To address the comments of both reviewers we had now added some further narrative and caveats in the text.

In a similar vein, Reviewer #2 requested data on whether dosage reduction altered protein levels by the expected amount. We have now added further explanation/references and western blot data to address this.

Changes to manuscript: Added more explanation of our choice of dsh[1] as an appropriate mutant allele to use in Results section ‘Planar polarity establishment is…’. Added some narrative and caveats regarding whether lowering levels more than 50% would add to our findings in the Discussion. Revised conclusions to be more cautious including altering section title to read ‘Planar polarity establishment is not highly sensitive to variation in protein levels of core complex components’.

Also added westerns and text/references showing that for the tested proteins there is a reduction in protein levels upon removal of one gene dosage in Results section ‘Planar polarity establishment is…’ and Fig.S2.

The data in Fig 5 are somewhat internally inconsistent, and inconsistent with the authors' interpretation. In both repolarization conditions, the authors claim that repolarization extends only to row 1, and row 1 is statistically different from non-repolarized row 1, but so too is row 3. Row 2 is not. This makes no sense, and suggests either that the statistical tests are inappropriate and/or the data is too sparse to be meaningful. 

As we’re sure the reviewer appreciates, this was an extremely complex experiment to perform and analyse. We spent a lot of time trying to find the best way to illustrate the results (finally settling on a 2D vector representation of polarity) and how to show the paired statistical comparisons between different groups. Moreover, in the end we were only able to detect generally quite modest (statistically significant) changes in cell polarity under the experimental conditions.

However, we note that failure to see large and consistent changes in polarity is exactly the expected result if it is hard to repolarise from a boundary – and this is of course the conclusion that we draw. Conversely, if repolarisation were easy, which was our expectation at least under de novo conditions without existing polarity, then we would have expected large and highly statistically significant changes in polarity across multiple cell rows. Hence we stand by our conclusion that ‘it is hard to repolarise from a boundary of Fz overexpression in both control and de novo polarity conditions’.

Overall, we were trying to establish three points:

(1) to demonstrate that repolarisation occurs from a boundary of overexpression i.e. from boundary 0 to row 0

(2) to establish whether a wave of repolarisation occurs across rows 1, 2 and 3

(3) to determine if in repolarisation in de novo condition it is easier to repolarise than in repolarisation in the control (already polarised) condition Taking each in turn:

(1) To detect repolarisation from a boundary relative to the control condition, we have to compare row 0 in repolarisation condition (Fig.5G,K) vs control condition (Fig.5F,J). This comparison shows a significative repolarisation (p=0.0014). From now, row 0 in repolarisation condition is our reference for repolarisation occurring.

(2) To determine if there is a wave of repolarisation in the repolarisation condition we have to compare row 0 vs row 1 to 3 in the repolarisation condition (Fig.5K). Row 1 is not significantly different to row 0, but rows 2 and 3 are different and the vectors show obviously lower polarity than row 0. Hence no wave of repolarisation is detected over rows 1 to 3.

(3) To determine if it is easier to repolarise in the de novo condition, our reference for establishment of a repolarisation pattern is the polarisation condition in rows 0 to 3. So, we compare repolarisation condition vs repolarisation in de novo condition, row 0 vs row 0, row 1 vs row 1, row 2 vs row 2 and row 3 vs row 3 – in each case no significative difference in polarity is detected, supporting our conclusion that it is not easier to repolarise in the de novo condition.

We agree that the variations in row 3 are puzzling, but there is no evidence that this is due to propagation of polarity from row 0, and so in terms of our three questions, it does not alter our conclusions.

Changes to manuscript: We have extensively revised the text describing the results in Fig.5 to hopefully make the reasons for our conclusions clearer and also be more cautious in our conclusions in Results section ‘Induced core protein relocalisation…’. 

For the related boundary intensity data in Fig 6, the authors need to describe exactly how boundaries were chosen or excluded from the analysis. Ideally, all boundaries would be classified as either meido-lateral (meaning anterior-posterior) or proximal-distal depending on angle. 

We thank the reviewer for pointing out that this was not clear.

All boundaries were classified following their orientation compared to the Fz over-expression boundary using hh-GAL4 expressed in the wing posterior compartment. Horizontal junctions were defined as parallel to the Fz over-expression boundary (between 0 and 45 degrees) and mediolateral junctions as junctions linking two horizontal boundaries (between 45 and 90 degrees).

Changes to manuscript: The boundary classification detailed above has been added in the Materials and Methods.

If the authors believe their Fig 5 and 6 analyses, how do they explain that hairs are reoriented well beyond where the core proteins are not? This would be a dramatic finding, because as far as I know, when core proteins are polarized, prehair orientation always follows the core protein distribution. Surprisingly, the authors do not so much as comment about this. The authors should age their wings just a bit more to see whether the prehair pattern looks more like the adult hair pattern or like that predicted by their protein orientation results.

Again the reviewer makes an interesting point, and we agree that this is something that we should have more directly addressed in the manuscript.

There are three reasons why we might expect adult trichomes to show a different effect from the measured core protein polarity pattern seen in our experiments:

(i) we are assaying core protein polarity at 28h APF, but trichomes emerge at >32h APF, so there is still time for polarity to propagate a bit further from the boundary. We now have added data showing that by the point of trichome initiation, the wave of polarisation extends 3-4 cell rows (Fig.S5A).

(ii) it has long been known that a strong localisation of core proteins at a cell edge is not required for polarisation of trichome polarity from a boundary. For instance, in Strutt & Strutt 2007 we show clones of cells overexpressing Fz causing propagation through pk[pk-sple] mutant tissue where there is no detectable core protein polarity. We were following up prior observations of Adler et al., 2000 in the wing and Lawrence et al., 2004 in the abdomen.

(iii) there is evidence to suggest that the polarity of adult trichomes is locally coupled, possibly mechanically. This point is hard to prove without live imaging taking in both initial core protein localisation, the site of actin-rich trichome initiation and then the final orientation of the much larger microtubule filled trichome, and we’re not aware that such data exist. However, Wong & Adler 1993 (JCB) showed that over a number of hours trichomes become much larger and move towards the centre of the cell, presumably becoming decoupled from any core protein cue. The images in Guild … & Tilney, 2005 (MBoC)  are also interesting to look at in this regard. Finally, septate junction proteins have been implicated in local alignment of trichomes, independently of the core pathway (Venema … & Auld, 2004 Dev Biol).

Changes to manuscript: Added new data in Fig.S5A showing where trichomes initiate under 6h de novo induction conditions, for comparison to core protein localisation and adult trichome data in Fig.5. Added some text explaining why adult trichome repolarisation might be stronger than the observed effects on core protein localisation in Discussion. 

Minor points

As the authors know, there is a model in the literature that suggests microtubule trafficking provides a global cue to orient PCP. The authors' repolarization data in Fig 4 make a reasonably convincing case against a role for no role for microtubules in cell-scale signaling, but do not rule out a role as a global cue. The authors should be careful of language such as "...MTs and core proteins being oriented independently of each other" that would appear to possibly also refer to a role as a global cue. 

Thank you for pointing out that this was not clear. We have now modified the text to hopefully address this.

Changes to manuscript: Text updated in Results section ‘Microtubules do not provide…’.

Significance:

There are two negative conclusions and one positive conclusion made by the authors. Provided the above points are addressed, the negative conclusions, that core proteins are not limiting and that microtubules are not involved in cell-scale signaling are solid. The positive conclusion is more nebulous - the authors say that cell-scale signaling is strong relative to cell-cell signaling - but how strong is strong? Strong relative to their prior expectations? I'm not sure how to interpret such a conclusion. Overall, we learn something from these results, though it fails to reveal anything about mechanism. These results will be of some interest to those studying PCP.

The reviewer raises an interesting point, which is how do you compare the strength of two different processes, even if both processes affect the same outcome (in this case cell polarity). Repolarisation from a boundary has not been carefully studied at the level of core protein localisation in any previous study to our knowledge – this is one of the important novel aspects of this study. Hence there is not a baseline for defining strong repolarisation. Similarly, there has been no investigation of the nature of ‘cell-scale signalling’. This was a considerable challenge for us in writing the manuscript, and we have done our best to find appropriate language that hopefully conveys our message adequately. Minimally our work may provide a baseline for helping to define the ‘strengths’ of these processes in future studies.

One of our main points is that we can generate an artificial boundary of Fz expression, where Fz levels are at least several fold higher than in the neighbouring cell (e.g. compare Fig.4N’ and O’) and only two rows of cells show a significant change in polarity relative to controls. Even when the tissue next to the overexpression domain is still in the process of generating polarity (de novo condition) then the boundary has little effect on polarity in neighbouring cell rows. This was a result that surprised us, and we tried to convey that by using language to suggest cell-scale signalling was stronger than cell-cell signalling i.e. stronger in terms of the ability to define the final direction of polarity.

Changes to manuscript: In the revised manuscript we have reviewed our use of language and now avoid saying ‘strong’ but instead use terms such as ‘effective’ and ‘robust’ in e.g. Results section ‘Induced core protein relocalisation…’, the Discussion and we have also changed the title of the manuscript to avoid claiming a ‘strong’ signal.

Reviewer #2:

Overview

This paper aims to dissect the relative importance of the various cues that establish PCP in the wing disc of Drosophila, which remains a prominent and relevant model for PCP. The authors suggest that one must consider cues at three scales (molecular, cell and tissue) and specifically design tests for the importance of cell-level cues, which they call non-local cell scale signalling. They develop clever experimental approaches that allow them to track complex stability and also to induce polarity at experimentally defined times. In a first set of experiments, they restore PCP after the global cues have disappeared (de novo polarisation) and conclude from the results that another (cell scale) cue must exist. In another set of experiments, they show that de novo repolarization is robust to the dosage of various components of core PCP, leading them to conclude that there must be an underlying cell scale polarity, which, apparently, has nothing to do with microtubule or cell shape polarity. They then describe nice evidence that de novo polarisation is relatively short range both in a polarised and unpolarised field. They conclude by there is a strong cell-intrinsic polarity that remains to be characterised.

Critique

The experiments described in this paper are of high quality with a sophisticated level of design and analysis. However, there needs to be some recalibration of the extent of the conclusions that can be drawn (see below). Moreover, a limitation of this paper is that, despite the quality of their data, they cannot give a molecular hint about the nature of their proposed cell-scale signal. Below are a two key points that the authors may want to clarify.

(1) The first set of repolarisation experiment is performed after the global cell rearrangements that have been shown to act as global signal. However, this approach does not exclude the possible contribution of an unknown diffusible global signal.

A similar point was raised by Reviewer 1. For the convenience of this reviewer, we’ll summarise the arguments against such an unknown cue again below. More broadly, both reviewers asking a similar question indicates that we have failed to lay out the evidence in sufficient detail. In our defence, we have used the same ‘de novo’ paradigm in three previous publications (Strutt and Strutt 2002, 2007; Brittle et al 2022) without attracting (overt) controversy. We have now added text to the Introduction and Results that goes into more detail, as well as more experimental evidence (Fig.S5).

Firstly, it is worth noting that the global cues acting in the wing are poorly understood, with mostly negative evidence against particular cues accruing in recent years. This makes it a hard subject to succinctly discuss. Secondly, we accept that it is hard to prove there is no influence of global cues, when the nature of those cues and the time at which they act remain unclear. Below we summarise the reasons why we believe there are not significance effects of global cues in our experiments that would influence the interpretation of our results.

First, our reading of the literature supports a broad consensus that an early radial core planar polarity pattern is realigned by cell flow produced by hinge contraction beginning at around 16h APF (e.g. Aigouy et al., 2010; Strutt and Strutt, 2015; Aw and Devenport, 2017; Butler and Wallingford, 2017; Tan and Strutt, 2025). Taken at face value, this suggests that there are ‘radial’ cues present prior to hinge contraction, maybe coming from the wing margin – arguably these radial cues could be Ft-Ds or Wnts or both, given they are expressed in patterns consistent with such a role (notwithstanding the published evidence arguing against roles for either of these cues). It then appears that hinge contraction supercedes these cues to convert a radial pattern to a proximodistal pattern – whether the radial cues that affect the core pathway earlier remain active after hinge contraction is unclear, although both Ft-Ds and Wnts appear to maintain their ‘radial’ patterns beyond the beginning of hinge contraction (e.g. Merkel et al., 2014; Ewen-Campen et al.,2020; Yu et al., 2020).

We think that the reviewers are proposing the presence of a proximodistal cue that is active in the proximal region of the wing that we use for our experiments shown e.g. in Fig.5, and that this cue orients core polarity here (but not elsewhere in the wing) in a time window after 18h APF. Ft-Ds and Wnts do not seem to be plausible candidates as they are still in ‘radial’ patterns. This leaves either an unknown proximodistal cue (a gradient of some unknown signalling molecule?), or possibly some ability of hinge contraction to align proximodistal polarity specifically in this wing region but not elsewhere. We cannot definitively rule out either of these possibilities, but neither do we think there is sufficient evidence to justify invoking their existence to explain our observations.

In particular, the reason that we don’t think there is a proximodistal cue in the proximal part of the wing after 18h APF, is that work from our lab shows that induction of Fz or Stbm expression at times around or after the start of hinge contraction (i.e. >16 h APF) results in increasing levels of trichome swirling with polarity not being coordinated with the tissue axis either proximally or distally (Strutt and Strutt, 2002; Strutt and Strutt 2007). Our simplest interpretation of this is that induction at these stages fails to result in the early radial pattern of core pathway polarity being established and hence a failure of hinge contraction to reorient radial to proximodistal. If hinge contraction alone could specify proximodistal polarity in the absence of the earlier radial polarity, then we would not expect to see swirling over much of the proximal wing (where the forces from hinge contraction are strongest, Etournay et al., 2015).

In this manuscript, our earliest de novo experiments begin at 18h APF (de novo 10h), then at 20h APF (de novo 8h) and at 22h APF (de novo 6h). The image in Fig. 5B referred to by Reviewer 1, is of a wing where Fz is induced de novo at 22 h APF. In these wings, as expected, the core proteins localise asymmetrically in stereotypical swirling patterns throughout the wing surface (see Fig. 2M and also Strutt and Strutt, 2002; Strutt and Strutt 2007), but – usefully for our experiments – they broadly localise along the proximal-distal axis in the region analysed in Fig. 5B. Given the strong swirling in surrounding regions when inducing at >20h APF, we feel reasonably confident in assuming that the pattern is not due to a proximodistal cue present in the proximal wing. We appreciate that the original manuscript did not show images including the trichome pattern in adjacent regions, so this point would not have been clear, but we now include these in Supplementary Fig.S5. We have also added a note in the legend to Fig. 5B to clarify that the proximodistal pattern seen is local to this wing region.

Changes to manuscript: Text extended in Introduction and Results to better explain why we believe the de novo conditions that we use most likely result in a polarity pattern that is not significantly influenced by ‘global cues’. Now show zoomed-out images of the surrounding region around the experiment region proximal to the anterior cross-vein region in adult wings, showing that the polarity pattern does not become more proximodistal when induction time is longer, and also that there is not overall proximodistal polarity in proximal regions of the wing, arguing against an unknown proximodistal polarity cue at these stages of development (Fig.S5B-E’’’).

(2) The putative non-local cell scale signal must be more precisely defined (maybe also given a better name). It is not clear to me that one can separate cell-scale from molecular-scale signal.

Local signals can redistribute within a cell (or membrane) so local signals are also cell-scale. Without a clear definition, it is difficult to interpret the results of the gene dosage experiments. The link between gene dosage and cell-scale signal is not rigorously stated. Related to this, the concluding statement of the introduction is too cryptic.

We thank the reviewer for raising this, as again a similar comment was made by Reviewer 1, so we are clearly falling short in defining the term. We have now had another attempt in the Introduction.

To more specifically answer the point made by the reviewer regarding molecular vs cellular, we are essentially being guided here by the prior computational modelling work, as at the biological level the details are still being worked out. A specific class of previous models only allowed ‘signals’ between core proteins to act ‘locally’, meaning within a cell junction, and within the models there was no explicit mechanism by which proteins on other junctions could ‘detect’ the polarity of a neighbouring junction (e.g. Amonlirdviman et al., 2005; Le Garrec et al., 2006; Fischer et al., 2013). Other models implicitly or explicitly encode a mechanism by which cell junctions can be influenced by the polarity of other junctions (e.g. Meinhardt, 2007; Burak and Shraiman, 2009; Abley et al., 2013; Shadkhoo and Mani, 2019), for instance by diffusion of a factor produced by localisation of particular planar polarity proteins.

We agree with the reviewer that a cell-scale signal will depend on ‘molecules’ and thus could be called ‘molecular-scale’, but here by ‘molecular-scale’ we mean signals that at the range of the sizes of molecules i.e. nanometers, rather than cell-scale signals that act at the size of cells i.e. micrometers. A caveat to our definition is that we implicitly include interactions that occur locally on cell junctions (<1 µm range) within ‘molecular-scale’, but this is a shorter range than ‘cellular-scale’ which requires signals acting over the diameter of a cell (3-5 µm). Nevertheless, we think the concept of ‘molecular-scale’ vs ‘cell-scale’ is a helpful one in this context, and have attempted to address the issue through a more careful definition of the terms.

Changes to manuscript: Text revised in Introduction and legend to Fig.1 to more carefully define ‘cell-scale signalling’ and to distinguish it from ‘molecular-scale signalling’. Final sentence of Introduction also altered so we no longer cryptically speculate on the nature of the cell-scale signal but leave this to the Discussion.

Minor comments. 

Some of the (clever) genetic manipulation may need more details in the text. For example:

- Need to specify if the hs-flp approach induces expression throughout the tissue.

We apologise for the lack of clarity. In all the experiments, the hs-FLP transgene is present in all cells, and heat-shock results in ubiquitous expression. 

Changes to manuscript: We have clarified this in the Results and Materials and Methods.

- Need to specify in the text that in the unpolarised condition the tissue is both dsh and fz mutant.

The reviewer is of course correct and we have updated this point in the text. The full genotype for the unpolarised condition is: w dsh¹ hsFLP22/y;; Act>>fz-mKate2sfGFP, fz^P21/fz^P21 (see Table S1). So this line is mutant for dsh and fz with induced expression of Fz-mKate2sfGFP. 

Changes to manuscript: We have clarified this in the relevant part of the Results.

- Need to specify in the text that the experiment illustrated in Fig 5 is with hh-gal4. 

As noted by the reviewer, we continued to use the same hh-GAL4 repolarisation paradigm as in Fig.4 and this info was in the legend to Fig.5 legend. However, we agree it is helpful to be explicit about this in the main text.

Changes to manuscript: We have added this to this section of the Results.

- Need to address a possible shortcoming of the hh experiment, that the AP boundary is a region of high tension.

It is true that the AP boundary is under high tension in the wing disc (e.g. Landsberg et al., 2009). But we are not aware of any evidence that this higher tension persists into the pupal wing. In separate studies we have labelled for Myosin II in pupal wings (Trinidad et al 2025 Curr Biol; Tan & Strutt 2025 Nature Comms), and as far as we have noticed have not seen preferentially higher levels on the AP boundary. We think if tension were higher, the cell boundaries would appear straighter than in surrounding cells (as seen in the wing disc) and this is not evident in our images.

- Need to dispel the possibility that there is no residual polarisation (e.g. of other components) in fz1 mutant (I assume this is the case).

We use the null allele fz[P21] through this work, and we and others have consistently reported a complete loss of polarisation of other core proteins or downstream components in this background. The caveat to this is that core proteins that persist at cell junctions always appear at least slightly punctate in mutant backgrounds for other core proteins, and so any automated detection algorithm will always find evidence of individual cell polarity above a baseline level of uniform distribution. Hence we tend to use lack of local coordination of polarity (variance of cell polarity angle) as an additional measure of loss of polarisation, in addition to direct measures of average cell polarity. (We discuss this in the QuantifyPolarity manuscript Tan et al 2021 e.g. Fig.S6).

Changes to manuscript: We now include in the Materials and Methods section ‘Fly genetics…’ a much more extensive explanation of the evidence for specific mutant alleles being ‘null’ for planar polarity function (including dsh1 as raised by Reviewer 1), specifically that they result in no detectable planar polarisation of either other core proteins or downstream effectors, and added appropriate references.

- Need to provide evidence that 50% gene dosage commensurately affect protein level. 

This is a good suggestion. In the case of Stbm, we have already published a western blot showing that a reduction in gene dosage results in reduced protein levels (Strutt et al 2016, Fig.S6). We have now performed western blots to quantify protein levels upon reduction of fmi, pk and dgo levels (we actually used EGFP-dgo for the latter, as we don’t have antibodies that can detect endogenous Dgo on western blots).

Changes to manuscript: When presenting the dosage reduction experiments, we now refer back to Strutt et al., 2016 explicitly for Stbm, and have added western blot data for Fmi, Pk and EGFPDgo in new Fig.S2.

- I am surprised that the relationship with microtubule polarity was never investigated. Is this true? 

We agree this is a point that needed further clarification, as Reviewer 1 made a related point regarding the two possible roles for microtubules, one being as a mediator of a global cue upstream of the core pathway, and the second (which we investigate in this manuscript) as a mediator of a cell-scale signal downstream of the core pathway.

Both the Uemura and Axelrod groups have published on potential upstream function as a global cue mediator in the Drosophila wing (e.g. Shimada et al., 2006; Harumoto et al., 2010; Matis et al., 2014).

Both groups have also looked out whether core pathway components could affect orientation of microtubules (Harumoto et al., 2010; Olofsson at al., 2014; Sharp and Axelrod 2016). Notably Harumoto et al., 2010 observed that in 24h APF wings, loss of Fz or Stbm did not alter microtubule polarity from a proximodistal orientation consistent with the microtubules aligning along the long cell axis in the absence of other cues. However, this did not rule out an instructive effect of Fz or Stbm on microtubule polarity during core pathway cell-scale signalling. The Axelrod lab manuscripts saw interesting effects of Pk protein isoforms on microtubule polarity, albeit not throughout the entire wing, which hinted at a potential role in cell-scale signalling. Taken together this prior work was the motivation for our directed experiments to specifically test whether the core pathway might generate cell-scale polarity by instructing microtubule polarity.

Changes to manuscript: We have revised the Results section ‘Microtubules do not…’ to make a clearer distinction regarding possible ‘upstream’ and ‘downstream’ roles of microtubules in Drosophila core pathway planar polarity and the motivation for our experiments investigating the latter.

- The authors suggest that polarity does not propagate as a wave. And yet the range measured in adult is longer than in the pupal wing. Explain. 

Again an excellent point, also made by Reviewer 1, which we have now addressed explicitly in the manuscript. For the convenience of this reviewer, we lay out the reasons why we think the propagation of polarity seen in the adult is further than seen for core protein localisation.

There are three reasons why we might expect adult trichomes to show a different effect from the measured core protein polarity pattern seen in our experiments:

(i) we are assaying core protein polarity at 28h APF, but trichomes emerge at >32h APF, so there is still time for polarity to propagate a bit further from the boundary. We now have added data showing that by the point of trichome initiation, the wave of polarisation extends 3-4 cell rows (Fig.S5A).  

(ii) it has long been known that a strong localisation of core proteins at a cell edge is not required for polarisation of trichome polarity from a boundary. For instance, in Strutt & Strutt 2007 we show clones of cells overexpressing Fz causing propagation through pk[pk-sple] mutant tissue where there is no detectable core protein polarity. We were following up prior observations of Adler et al 2000 in the wing and Lawrence et al 2004 in the abdomen.

(iii) there is evidence to suggest that the polarity of adult trichomes is locally coupled, possibly mechanically. This point is hard to prove without live imaging taking in both initial core protein localisation, the site of actin-rich trichome initiation and then the final orientation of the much larger microtubule filled trichome, and we’re not aware that such data exist. However, Wong & Adler 1993 (JCB) showed that over a number of hours trichomes become much larger and move towards the centre of the cell, presumably becoming decoupled from any core protein cue. The images in Guild … & Tilney, 2005 (MBoC)  are also interesting to look at in this regard. Finally, septate junction proteins have been implicated in local alignment of trichomes, independently of the core pathway (Venema … & Auld, 2004 Dev Biol).

Changes to manuscript: Added new data in Fig.S5A showing where trichomes initiate under 6h de novo induction conditions, for comparison to core protein localisation and adult trichome data in Fig.5. Added some text explaining why adult trichome repolarisation might be stronger than the observed effects on core protein localisation in Discussion. 

- The discussion states that the cell-intrinsic system remains to be fully characterised, implying that it has been partially characterised. What do we know about it? 

As the reviewer probably realises, we were attempting to side-step a long speculative discussion about the various hints and ideas in the literature by grouping them under the umbrella of ‘remaining to be fully characterised’. We would argue that this current manuscript is the first to attempt to systematically investigate the nature of ‘cell-scale signalling’. The lack of prior work is probably due to two factors (i) pioneering theoretical work showed that a sufficiently strong global signal coupled with ‘local’ (i.e. confined to one cell junction) protein interactions was sufficient to polarise cells without the need to invoke the existence of a cell-scale signal; (ii) there is no easy way to identify cell-scale signals as their loss results in loss of polarity which will also occur if other (i.e. more locally acting) core pathway functions are compromised.

The main investigation of the potential for cell-scale signalling has been another set of theory studies (Burak and Shraiman 2009; Abley et al., 2013; Shadkhoo and Mani 2019) which have considered the possibility of diffusible signals. In our present work we have further considered the possibility of a ‘depletion’ model, based on the pioneering theory work of Hans Meinhardt, and as discussed above the possibility that microtubules could mediate a cell-scale signal.

Changes to manuscript: We have revised the Discussion to hopefully be clearer about the current state of knowledge.

Reviewer #3:

The manuscript by Carayon and Strutt addresses the role of cell-scale signaling during the establishment of planar cell polarity (PCP) in the Drosophila pupal wing. The authors induce locally the expression of a tagged core PCP protein, Frizzled, and observe and analyze the de novo establishment of planar cell polarity. Using this system, the authors show that PCP can be established within several hours, that PCP is robust towards variation in core PCP protein levels, that PCP proteins do not orient microtubules, and that PCP is robust towards 'extrinsic' repolarization. The authors conclude that the polarization at the cell-scale is strongly intrinsic and only weakly affected by the polarity of neighboring cells. 

Major comments

The data are clearly presented and the manuscript is well written. The conclusions are well supported by the data. 

(1) The authors use a system to de novo establish PCP, which has the advantage of excluding global cues orienting PCP and thus to focus on the cell-intrinsic mechanisms. At the same time, the system has the limitation that it is unclear to what extent de novo PCP establishment reflects 'normal' cell scale PCP establishment, in particular because the Gal4/UAS expression system that is used to induce Fz expression will likely result in much higher Fz levels compared with the endogenous levels. The authors should briefly discuss this limitation. 

We apologise if this wasn’t clear. We only used GAL4/UAS overexpression when we were generating an artificial boundary of Fz expression with hh-GAL4 to induce repolarisation. The de novo induction system involves Fz::mKate2-sfGFP being expressed directly under an Act5C promoter without use of GAL4/UAS. In response to a comment from Reviewer 1 we have now carried out western blot analysis which shows that Fz::mKate2-sfGFP levels under Act5C are actually lower than endogenous Fz levels. As we achieve normal levels of polarity, similar to what we measure in wild-type conditions when measured using QuantifyPolarity, we assume that therefore Fz levels are not limiting under these conditions. However, we note that lower than normal levels of Fz might sensitise the system to perturbation, which in fact would be advantageous in our study, as it might for instance have been expected to more readily reveal dosage sensitivity of other components.

Changes to manuscript: We now describe the levels of expression achieved using the de novo induction system (Fig.S1C-D) and discuss possible consequences in the relevant Results sections and Discussion.

(2) Fig. 3. The authors use heterozygous mutant backgrounds to test the robustness of de novo PCP establishment towards (partial) depletion in core PCP proteins. The authors conclude that de novo polarization is 'extremely robust to variation in protein level'. Since the authors (presumably) lowered protein levels by 50%, this conclusion appears to be somewhat overstated. The authors should tune down their conclusion. 

Reviewer 1 makes a similar point about whether we can argue that the lack of sensitivity to a 50% reduction in protein levels actually rules out the depletion model. To address the comments of both reviewers we had now added some further narrative and caveats in the text.

We nevertheless believe that the experiments shown effectively make the point that there is no strong dosage sensitivity – and it remains our contention that if protein levels were the key to setting up cell-scale polarity, then a 50% reduction would be expected to show an effect on the rate of polarisation. We further note that as Fz::mKate2-sfGFP levels are lower than endogenous Fz levels, the system might be expected to be sensitised to further dosage reductions, and despite this we fail to see an effect on rate of polarisation.

In a similar vein, Reviewer 2 requested data on whether dosage reduction altered protein levels by the expected amount. We have now added further explanation/references and western blot data to address this.

Changes to manuscript: Added some narrative and caveats regarding whether lowering levels more than 50% would add to our findings in the Discussion. Revised conclusions to be more cautious including altering section title to read ‘Planar polarity establishment is not highly sensitive to variation in protein levels of core complex components.

Also added westerns and text/references showing that for the tested proteins there is a reduction in protein levels upon removal of one gene dosage in Results section ‘Planar polarity establishment is…’ and Fig.S2.

Minor comments 

(1) Page 3. The authors mention and reference that they used the PCA method to quantify cell polarity magnification and magnitude. It would help the unfamiliar reader, if the authors would briefly describe the principle of this method. 

Changes to manuscript: More details have been added in Materials & Methods.

Significance:

The manuscript contributes to our understanding of how planar cell polarity is established. It extends previous work by the authors (Strutt and Strutt, 2002,2007) that already showed that induction of core PCP pathway activity by itself is sufficient to induce de novo PCP. This manuscript further explores the underlying mechanisms. The authors test whether de novo PCP establishment depends on an 'inhibitory signal', as previously postulated (Meinhardt, 2007), but do not find evidence. They also test whether core PCP proteins help to orient microtubules (which could enhance cell intrinsic polarization of core PCP proteins), but, again, do not find evidence, corroborating previous work (Harumoto et al, 2010). The most significant finding of this manuscript, perhaps, is the observation that local de novo PCP establishment does not propagate far through the tissue. A limitation of the study is that the mechanisms establishing intrinsic cell scale polarity remain unknown. The work will likely be of interest to specialists in the field of PCP.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

Summary:

In this manuscript, Liu et al have tried to dissect the neural and molecular mechanisms that C. elegans use to avoid digestion of harmful bacterial food. Liu et al show that C. elegans use the ON-OFF state of AWC olfactory neurons to regulate the digestion of harmful gram-positive bacteria S. saprophyticus (SS). The authors show that when C. elegans are fed on SS food, AWC neurons switch to OFF fate which prevents digestion of S. saprophyticus and this helps C. elegans avoid these harmful bacteria. Using genetic and transcriptional analysis as well as making use of previously published findings, Liu et al implicate the p38 MAPK pathway (in particular, NSY-1, the C. elegans homolog of MAPKKK ASK1) and insulin signaling in this process.

Strengths:

The authors have used multiple approaches to test the hypothesis that they present in this manuscript.

Weaknesses:

Overall, I am not convinced that the authors have provided sufficient evidence to support the various components of their hypothesis. While they present data that loosely align with their hypothesis, they fail to consider alternative explanations and do not use rigorous approaches to strengthen their overall hypothesis. The selective picking of genes from the RNA sequencing data and forcing the data to fit the proposed hypothesis based on previously published findings, without exploring other approaches, indicates a lack of thoroughness and rigor. These critical shortcomings significantly diminish enthusiasm for the manuscript in its totality. In my opinion, this is the biggest weakness in this manuscript.

We appreciate the reviewer’s all the suggestions which help us to improve this paper. We now addressed reviewer’s comments at the section of “Reviewer #1 (Recommendations for the authors)”

Reviewer #2 (Public review):

Summary:

Using C. elegans as a model, the authors present an interesting story demonstrating a new regulatory connection between olfactory neurons and the digestive system.

Mechanistically, they identified key factors (NSY-1, STR-130 et.al) in neurons, as well as critical 'signaling factors' (INS-23, DAF-2) that bridge different cells/tissues to execute the digestive shutdown induced by poor-quality food (Staphylococcus saprophyticus, SS).

Strengths:

The conclusions of this manuscript are mostly well supported by the experimental results shown.

Weaknesses:

Several issues could be addressed and clarified to strengthen their conclusions.

(1) The word "olfactory" should be carefully used and checked in this manuscript. Although AWCs are classic olfactory neurons in C. elegans, no data in this manuscript supports the idea that olfactory signals from SS drive the responses in the digestive system. To validate that it is truly olfaction, the authors may want to check the responses of worms (e.g. AWC, digestive shutdown, INS-23 expression) to odors from SS.

We appreciate the reviewer’s careful attention to terminology. We agree that the term "olfactory" requires direct experimental validation. However, in this paper, we only used "olfactory" to specific define the AWC neurons. As reviewer’s suggestion, we now deleted the word “olfactory”.

(2) In line 113, what does "once the digestive system is activated" mean? The authors need to provide a clearer statement about 'digestive activation' and 'digestive shutdown'.

Previously, we observed that activating larval digestion with heat-killed E. coli or E. coli cell wall peptidoglycan (PGN) enabled the digestion of SS as food (Hao et al., 2024). Additionally, when animals reached the L2 stage by feeding normal OP50 diet, they could utilize SS as a food source to support growth (Figure 1—figure supplement 1D). These findings suggest that once digestion is activated (via E. coli components or L2-stage maturation), worms gain the capacity to process SS as a viable food source, abolishing SS-induced growth impairment (Hao et al., 2024) ( Figure 1—figure supplement 1D).

(3) No control data on OP50. This would affect the conclusions generated from Figures 2A, 2B, 2D, 3B, 3C, 3G, 4D-G, 5D-E, 6B-D.

We appreciate  this point. The central goal of the experiments listed (Figures 2A,B,D; 3B,C,G; 4D-G; 5D-E; 6B-D) was not to compare growth or behavior between SS and OP50 under standard conditions, but rather to understand the genetic basis of the C. elegans response specifically to SS, as identified through our nsy-1 mutant screen.

Our data in Figure 1 clearly establishes the fundamental difference in growth and feeding behavior when larvae encounter SS compared to OP50 (Figures 1A,B). Having established SS as an unfavorable food source that triggers a specific protective response (digestive shutdown), the subsequent experiments focus on deciphering how this response is mediated.

Therefore, within these specific experimental contexts under SS feeding: The primary comparison is between wild-type (N2) and nsy-1 mutant animals. All assays (growth, behavior, survival) are performed under the same SS feeding conditionsfor both genotypes.

This design allows us to directly assess the functional role of NSY-1 in mediating the SS-specific response pathway we are investigating. Including an OP50 control for every figure would not address this core genetic question and could introduce confounding variables given the established difference in how C. elegans treats these two food sources. The critical internal control for these specific experiments is the performance of the wild-type under SS versus the mutant under SS.

(4) Do the authors know which factors are released from AWC neurons to drive the digestive shutdown?

Enrichment analysis revealed that genes related to extracellular functions, such as insulin-related genes, are induced in nsy-1 mutant animals (Figure 5—figure supplement 1A, Supplementary file 4). Further analysis of insulin-related genes from the RNA-seq data showed that ins-23 is predominantly induced in nsy-1 mutant animals (Figure 5—figure supplement 1B), suggesting its potential role in promoting SS digestion. We found that knockdown of ins-23 in nsy-1 mutants inhibited SS digestion (Figure 5D). Given that INS-23 is expressed in AWC neurons (Figure 5—figure supplement 3A, CeNGEN), this suggests increased production and likely enhanced release of INS-23 from AWC neurons in the nsy-1 mutant background, which promotes SS digestion.

The insulin/insulin-like growth factor signaling (IIS) pathway, particularly through the DAF-2 receptor, integrates nutritional signals to regulate various behavioral and physiological responses related to food (Kodama et al., 2006; Ryu et al., 2018). It has been shown that INS-23 acts as an antagonist for the DAF-2 receptor to promote larval diapause (Matsunaga et al., 2018). To test whether ins-23 induction in nsy-1 mutants promotes SS digestion through its receptor, DAF-2, we constructed a nsy-1; daf-2 double mutant. We found that the SS digestion ability of the nsy-1 mutant was inhibited by the daf-2 mutation. This suggests that the nsy-1 mutation induces the insulin peptide ins-23, which promotes SS digestion through its potential receptor, DAF-2.

The data supports a model where AWC neurons regulate digestion via the release of INS-23. Loss of nsy-1 function increases INS-23 release from AWC, activating DAF-2 signaling and promoting digestion. Conversely, in wild-type animals, reduced INS-23 release from AWC contributes to digestive shutdown in response to SS food.

Reviewer #3 (Public review):

Summary:

The study explores a molecular mechanism by which C. elegans detects low-quality food through neuron-digestive crosstalk, offering new insights into food quality control systems. Liu and colleagues demonstrated that NSY-1, expressed in AWC neurons, is a key regulator for sensing Staphylococcus saprophyticus (SS), inducing avoidance behavior and shutting down the digestive system via intestinal BCF-1. They further revealed that INS-23, an insulin peptide, interacts with the DAF-2 receptor in the gut to modulate SS digestion. The study uncovers a food quality control system connecting neural and intestinal responses, enabling C. elegans to adapt to environmental challenges.

Strengths:

The study employs a genetic screening approach to identify nsy-1 as a critical regulator in detecting food quality and initiating adaptive responses in C. elegans. The use of RNA-seq analysis is particularly noteworthy, as it reveals distinct regulatory pathways involved in food sensing (Figure 4) and digestion of Staphylococcus saprophyticus (Figure 5). The strategic application of both positive and negative data mining enhances the depth of analysis. Importantly, the discovery that C. elegans halts digestion in response to harmful food and employs avoidance behavior highlights a physiological adaptation mechanism.

Weaknesses:

Major points:

(1) While NSY-1 positively regulates str-130 expression in AWC neurons and is critical for SS avoidance and survival, the authors should examine whether similar phenotypes are observed in str-130 mutants.

In this study, we mainly focused on how worms sense adverse food sources (SS food) and shutdown digestion (not growth as digestion shutdown readout). We found that nsy-1 in AWC play key roles in response SS food, once nsy-1 mutation, mutant animals cannot detect SS food and digest it, therefore growth under SS food. From RNA-seq, we found that nsy-1 positively regulates several sensory perception related genes (sra-32, str-87, str-112, str-130, str-160, str-230) (Figure 4—figure supplement 1A, Supplementary file 2). After screen, we found that we found that knockdown of str-130 in wild-type animals promoted SS digestion, thereby supporting animal growth (Figure 4D), and the proportion of animals with two AWC^OFF neurons decreased (Figure 4E). Secondly, we found that overexpression of str-130 in nsy-1 mutant animals inhibited SS digestion, thereby slowing animal growth (Figure 4F), and the proportion of animals with two AWC^OFF neurons increased (Figure 4G). These results demonstrate that NSY-1 promotes the AWC^OFF state by inducing str-130 expression, which in turn inhibits SS digestion in C. elegans.

(2) NSY-1 promotes the AWC-OFF state through str-130, inhibiting SS digestion. The authors should investigate whether STR-130 in AWC neurons regulates bcf-1 expression levels in the intestine.

We agree with the reviewer's suggestion regarding the potential role of STR-130 in AWC neurons regulating intestinal bcf-1 expression. To address this, we generated transgenic worms with AWC-specific knockdown of str-130, achieved by rescuing sid-1 cDNA expression under the ceh-36 promoter (AWC-specific) in sid-1(qt9);BCF-1::GFP background worms.

We observed that AWC neuron-specific RNAi of str-130 elevated intestinal BCF-1::GFP expression (Figure 6—figure supplement 1B). This demonstrates that STR-130 functions cell-non-autonomously in AWC neurons to repress BCF-1 expression in the intestine.

(3) The current results rely on str-2 expression levels to indicate the AWC state. Ablating AWC neurons and testing the effects on digestion would provide stronger evidence for their role in digestive regulation.

To confirm the important of AWC state in SS digestion, we performed AWC-specific neuron ablation experiments using previously validated transgenic strain that expresses cleaved caspase under the AWC-specific promoter, ceh-36 (ceh-36p::caspase). Critically, worms with ablated AWC neurons completely failed to digest SS food (Figure 3—figure supplement 4), phenocopying the non-digesting state of wild-type worms on SS when AWC-OFF signaling is impaired. This result directly confirms that functional AWC neurons are essential for initiating SS digestion, aligning with our model where the AWC-OFF state (induced by SS) inhibits digestion while the AWC-ON state promotes it.

Furthermore, we previously study discovered that AWC ablation activates the intestinal mitochondrial unfolded protein response and inhibits food digestion, mechanistically linking neuronal integrity to gut stress responses and digestive inhibition.

Together, these functional ablation studies provide compelling physiological evidence that AWC neurons act as central regulators of food-state sensing and gut function.

(4) The claim that NSY-1 inhibits INS-23 and that INS-23 interacts with DAF-2 to regulate bcf-1 expression (Line 339-340) requires further validation. Neuron-specific disruption of INS-23 and gut-specific rescue of DAF-2 should be tested.

We agree with the reviewer that the proposed NSY-1 ⊣ INS-23 → DAF-2 → BCF-1 signaling axis requires tissue-specific validation. To address this, we conducted compartment-specific functional dissection of INS-23 and DAF-2:

AWC neuronal role of INS-23:

To test whether INS-23 acts in AWC neurons to regulate intestinal BCF-1, we generated AWC-specific knockdown strains which was achieved by rescuing sid-1 cDNA expression under the ceh-36 promoter in a sid-1(qt9);BCF-1::GFP background. We found that AWC-restricted ins-23 knockdown significantly reduced intestinal BCF-1::GFP expression (Figure 6—figure supplement 1A). This confirms that INS-23 functions cell-non-autonomously within AWC sensory neurons to activate intestinal BCF-1, consistent with NSY-1’s upstream inhibition of INS-23 in this neuronal  subtype

Intestinal role of DAF-2 as INS-23 receptor:

To investigate weather DAF-2 acts as the gut-localized receptor for neuronal INS-23 signaling, we performed tissue-specific rescue experiments in the nsy-1(ag3);daf-2(e1370) double mutant. When DAF-2 was re-introduced specifically in the intestine (using the ges-1 promoter), we observed a significant suppression of SS digestion (Figure 5—figure supplement 3B), but not rescue digestive defect. This indicates that INS-23 induction in nsy-1 mutants promotes digestion independently of intestinal DAF-2 function.

(5) Figure Reference Errors: Lines 296-297 mention Figure 6E, which does not exist in the main text. This appears to refer to Figure 5E, which has not been described.

We corrected this.

Reviewer #1 (Recommendations for the authors):

I would like the authors to address the following comments in a resubmission.

(1) The hallmark of the activated p38 MAPK pathway is the phosphorylation of most downstream kinase p38 (PMK-1/PMK2 in C. elegans) of this kinase cascade. Previous work from Bergmann lab showed that the most downstream kinase of this pathway, PMK-1/PMK-2, is not required for AWC asymmetry. I wonder whether that is the case also for the model that Liu et al have presented in this manuscript. Since p38/PMK-1 undergoes activation (phosphorylation) in response to pathogenic bacteria like P. aeruginosa, it is worth testing whether PMK-1 plays a role downstream of NSY-1 in the model that Liu et al present in this manuscript. It would be worth testing whether there is increased phosphorylation of p38 when C. elegans are fed SS and whether that phosphorylation regulates downstream components that Liu et al have identified in this manuscript.

We thank the reviewer for raising this important point regarding PMK-1/p38 MAPK signaling. As established in our prior work (Reference 1), SS exposure triggers phosphorylation of PMK-1 (P-PMK-1) in C. elegans, and pmk-1 mutants exhibit enhanced growth on SS (Figure-1, Figure-2). This confirms that PMK-1-mediated innate immune signaling actively regulates SS responsiveness and digestion.

To address whether PMK-1 functions downstream of NSY-1 within our proposed model, we performed critical epistasis analyses. While we observed that nsy-1 mutation elevates ins-23 (indicating NSY-1 suppression of ins-23), knockdown of pmk-1 did not alter ins-23 expression levels (Figure 5-figure supplement 3C). This demonstrates that PMK-1 does not operate through the ins-23 pathway to regulate SS digestion. Thus, although both pathways respond to SS, the PMK-1-mediated innate immune response and the NSY-1/INS-23 axis constitute distinct regulatory mechanisms governing digestive adaptation.

Reference 1: Geng, S., Li, Q., Zhou, X., Zheng, J., Liu, H., Zeng, J., Yang, R., Fu, H., Hao, F., Feng, Q., & Qi, B. (2022). Gut commensal E. coli outer membrane proteins activate the host food digestive system through neural-immune communication. Cell host & microbe, 30(10), 1401–1416.e8. https://doi.org/10.1016/j.chom.2022.08.004

(2) Since p38 MAPK pathway has a well-established role in host defense in the C. elegans intestine, it is important to show that NSY-1 does not function in the intestine in the model that Liu et al present. I would like the authors to reintroduce nsy-1 in C. elegans intestine in nsy-1 mutant animals and then test whether it has any effect on worm length on SS food (similar to what is done in Figure 3 for AWC-specific nsy-1).

Beyond its  established  role  in  AWC  neurons,  we  detected  NSY-1 expression in the intestine (Figure 3-figure supplement 2A). To assess intestinal NSY-1 function, we performed tissue-specific rescue experiments in nsy-1 mutants using the intestinal-specific vha-1 promoter. Intestinal expression of NSY-1 significantly suppressed the enhanced SS digestion phenotype in nsy-1 mutants (Figure 3-figure supplement 2B), demonstrating functional involvement of gut-localized NSY-1 in regulating digestive responses. We propose intestinal NSY-1 mediates this effect through innate immune signaling, consistent with its known pathway components. As previously established (Reference 1), the canonical PMK-1/p38 MAPK pathway functions downstream of NSY-1, with both sek-1 and pmk-1 knockdown enhancing SS digestion through immune modulation. This indicates intestinal NSY-1 suppresses digestion may act through PMK-1-mediated immune responses. Since neuronal NSY-1's role in digestive control was previously undefined, we prioritized mechanistic analysis of its neuronal function in digestion regulation.

Notably, this immune-mediated mechanism operates independently of NSY-1's neuronal regulation pathway. In AWC neurons, NSY-1 controls digestion exclusively through the neuropeptide signaling axis (INS-23/DAF-2/BCF-1) without engaging innate immune components.

Reference 1: Geng, S., Li, Q., Zhou, X., Zheng, J., Liu, H., Zeng, J., Yang, R., Fu, H., Hao, F., Feng, Q., & Qi, B. (2022). Gut commensal E. coli outer membrane proteins activate the host food digestive system through neural-immune communication. Cell host & microbe, 30(10), 1401–1416.e8. https://doi.org/10.1016/j.chom.2022.08.004

(3) At multiple places, wild-type (WT) controls have been labeled as N2. It is better to label all controls as WT (and not as N2).

Corrected.

(4) In Figure 2B, the aversion response should be scored at multiple time points, like Figure 1C, rather than at just one timepoint.

We thank the reviewer for suggesting multi-timepoint analysis of aversion behavior. In accordance with this recommendation, we have now quantified SS avoidance at multi-timepoint. As shown in the revised Figure 2B, nsy-1 mutants exhibited significantly impaired avoidance responses at both 4h and 6h but not at 8h, confirming that NSY-1 is essential for sustained aversion to SS food in the early response. This data demonstrates that the critical role of NSY-1 in food discrimination at initial sensory responses.

(5) Does the re-introduction of nsy-1 in AWC neurons in nsy-1 mutant background help animals avoid SS in dwelling and food-choice assays? Along the same lines, does the CRISPR-generated AWC-specific mutant of NSY-1 fail to avoid SS in dwelling and food-choice assays similar to the whole-animal mutant? These behavioral data are missing in Figure 3.

We thank the reviewer for prompting behavioral validation of AWC-specific nsy-1 functions. To determine whether NSY-1 in AWC neurons mediates SS sensory perception, we performed dwelling (avoidance) and food-choice assays using AWC-specific nsy-1 knockout and AWC-rescued strains (nsy-1(ag3); Podr-1::nsy-1). In dwelling assays, AWC-specific nsy-1 KO mutants exhibited significantly impaired SS avoidance at 6h (Figure 3-figure supplement 3A), while AWC-rescued strains restored avoidance capacity at 2-6h (Figure 3-figure supplement 3B). Food-choice assays further revealed that AWC nsy-1 KO mutants preferentially migrated toward SS (Figure 3-figure supplement 3C), whereas AWC-rescued showed no preference between SS and HK-E. coli (Figure 3-figure supplement 3D). These data conclusively demonstrate that NSY-1 acts in AWC neurons to mediate SS recognition and aversion behaviors.

(6) In Figure 3E and F, the number of animals that were used for scoring AWC str-2p::GFP expression should be specified.

we added the number of animals in the figure.

(7)  RNA seq analysis identified multiple GPCRs (including STR-130) that are upregulated in an NSY-1-dependent manner when animals are fed with SS bacteria. However, the authors decided to only characterize STR-130 because of previously published findings. It is important to rule out the role of other GPCRs since all are upregulated on SS food as shown in Figure S4 B. I would like the authors to knock down other GPCRs in the same manner as they did for STR-130 and demonstrate that only str-130 knockdown behaves similarly to the nsy-1 mutant (if that is the case) using the assay presented in Figure 4 D.

We appreciate the reviewer’s suggestion to comprehensively evaluate NSY-1-regulated GPCRs. In response, we extended our functional analysis to all six GPCRs (str-130, str-230, str-87, str-112, str-160, and sra-32) identified as NSY-1-dependent and SS-induced in RNA-seq (Figure 4—figure supplement 1).

Using RNAi knockdown and the SS growth assay, we observed that RNAi of str-130, str-230, str-87, or str-112 significantly enhanced SS growth (Figure 4—figure supplement 2A), with str-130 RNAi exhibiting the most robust phenotype—phenocopying nsy-1 mutants. Crucially, none of these GPCR knockdowns further enhanced growth in nsy-1(ag3) mutants (Figure 4—figure supplement 2B), confirming their position downstream of NSY-1. These data establish str-130 as the dominant effector of NSY-1-mediated SS response regulation, while suggesting minor contributions from other GPCRs (str-230, str-87, str-112).

(8) In Figure 4E and G, the number of animals that were used for scoring GFP expression should be specified.

we added the number of animals in the figure.

(9) When comparing Figure 3E and Figure 4E, it appears that the loss of str-130 RNAi does not phenocopy nsy-1 mutant. This raises the question of whether the inefficiency of RNAi targeting str-130 is the cause, or if STR-130 is not the only GPCR regulated by NSY-1 on SS food. I would like the authors to address this discrepancy. If RNAi inefficiency is indeed the cause, using an RNAi-sensitive background, such as an eri- 1 mutant, could help strengthen the data presented in Figure 4E. Conversely, if RNAi inefficiency is not responsible for the discrepancy, I suggest that the authors investigate the roles of other GPCRs that were identified by RNA sequencing.

We appreciate the reviewer’s observation regarding the phenotypic difference between nsy-1 mutants and str-130 (RNAi) animals on SS food (Fig. 3E vs Fig. 4E).

While both genetic perturbations significantly enhance SS growth and increase the proportion of animals exhibiting AWC^ON states compared to wild type (indicating enhanced digestion), the specific AWC^ON neuron configurations differ: nsy-1 mutants predominantly show 2 AWC^ON animals, whereas str-130(RNAi) animals primarily exhibit the 1 AWC^ON /1 AWC^OFF configuration (Fig. 3E vs Fig. 4E).

This difference likely arises because STR-130 is the key GPCR mediating NSY-1's inhibitory effect on SS digestion, but it is not the sole GPCR involved, as evidenced by our RNAi screen identifying several additional NSY-1-regulated GPCRs (str-230, str-87, str-112) whose depletion also enhanced SS growth (Fig. 4A-D).

The robust SS growth enhancement and AWC^ON state increase caused by str-130 (RNAi) (phenocopying the nsy-1 mutant’s functional outcome of enhanced digestion) (Figure 4D, 4E) indicate effective RNAi knockdown for this specific assay. Therefore, the distinct neural configurations reflect the partial redundancy among GPCRs downstream of NSY-1, rather than an inherent inefficiency of the str-130 RNAi.

The nsy-1 mutant phenotype represents the complete loss of all inhibitory GPCR signaling coordinated by NSY-1, while str-130(RNAi) represents the loss of its major component. Investigating the roles of other identified GPCRs (str-230, str-87, str-112) in modulating AWC^ON neuron states is an important direction for future research.

(10) In Figure 4 F and 4 G, the authors show that the overexpression of STR-130 rescues the nsy-1 mutant phenotype suggesting that NSY-1 might function through STR-130 to control digestion on SS food. These data place STR-130 downstream of NSY-1. To further strengthen these epistasis data, authors should knock down str-130 in nsy-1 mutant animals and show that the combined loss of both genes produces the same effect as the loss of either gene alone.

We thank the reviewer for the insightful suggestion to further define the genetic relationship between nsy-1 and str-130. To strengthen our epistasis analysis, we performed RNAi knockdown of str-130 in the nsy-1(ag3) mutant background and assessed development on SS food. Consistent with STR-130 acting downstream of NSY-1, the loss of str-130 via RNAi did not further enhance the developmental capacity (i.e., growth phenotype) of nsy-1(ag3) mutant animals on SS. This lack of enhancement indicates that str-130 and nsy-1 function within the same genetic pathway, with str-130 acting epistatically downstream of nsy-1 (Figure 4—figure supplement 3). This finding reinforces the model proposed from our overexpression data (Fig. 4F-G) – that NSY-1 primarily exerts its inhibitory effect on SS digestion by inducing the expression GPCR STR-130.

(11) In Figure 5C, please mention "ins-23 transcript levels" on the top of the graph so that it is clear what these data represent.

We appreciate the reviewer’s suggestion.

(12) Since all ins genes were upregulated in nsy-1 mutants (though ins-23 was indeed the most highly upregulated gene) on SS food from RNA seq analysis (Figure S5 B), it is important to first phenotypically characterize all of them using "worm length assay". If this analysis shows that ins-23 has the most robust phenotype, it would make more sense to just focus on ins-23.

We agree with the reviewer that initial phenotypic characterization of candidate genes identified through transcriptomic analysis is valuable.Our RNA-seq data revealed that several insulin-like peptide genes, including ins-22, ins-23, ins-24, and ins-27, were significantly upregulated in the nsy-1 mutant on SS food (Figure 5—figure supplement 1B). We prioritized these insulin-like peptide genes for functional validation because they are known to act as neuropeptides capable of mediating non-cell autonomous signaling in previous studies (Shao et al 2016).

To determine if any were functionally responsible for the enhanced SS growth observed in nsy-1 mutants, we performed functional phenotypic screening using the SS growth assay (worm length assay). We individually knocked down each of these candidates (ins-22, ins-23, ins-24, ins-27) in the nsy-1(ag3) mutant background. Among these, only RNAi targeting ins-23 significantly attenuated (i.e., suppressed) the enhanced development of the nsy-1(ag3) mutant on SS (Figure 5—figure supplement 2). This targeted functional screening revealed that ins-23 has the most robust and specific role in mediating the enhanced digestion phenotype downstream of NSY-1 loss, providing the critical justification for our subsequent focus on this particular insulin-like peptide.

Ref:

Shao, L. W., Niu, R., & Liu, Y. (2016). Neuropeptide signals cell non-autonomous mitochondrial unfolded protein response. Cell research, 26(11), 1182–1196. https://doi.org/10.1038/cr.2016.118

Reviewer #2 (Recommendations for the authors):

There are several minor errors and typos in the manuscript

(1) A number of typos in the figures, like "length".

Corrected.

(2) The 'axis labels' are inconsistent from panel to panel, like "relative body length" and "relative worm length".

Corrected.

(3) The fonts are inconsistent from panel to panel.

Corrected.

(4) There is no Ex unique number for transgenic lines.

Corrected.

Reviewer #3 (Recommendations for the authors):

Minor points:

(1)  Figure 3B, 3C, 3G, 4D, 4F, 5D, 5E, and 6C: Replace "lenth" with "length" (consistent with Figure 2A).

Corrected.

(2) Figure 4D: Correct "ctontrol" to "control."

Corrected.

(3) Figure 4G: Update the co-injection marker to Podr-1::GFP instead of Pstr-2::GFP.

Corrected.

(4) Figure 5C: This figure is missing from the Results section.

Corrected.

(5) Figure 6A: Label the graph with Pbcf-1::bcf-1::GFP, as in Figure 6D.

Corrected.

(6) Italicization: Lines 588 and 603-italicize nsy-1.

Corrected.

(7) Supplementary Figure S2A: Correct "Screeng" to "Screening."

Corrected.

(8) Spelling/Proofreading: Ensure consistent spelling and grammar, such as correcting "mutan" to "mutant" in Figure 4A.

Corrected.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public Review):

(1) I suggest that the author's choose a different term in their title, abstract and manuscript to describe the phenotypes associated with ufd-1 and npl-4 knockdown other than an "inflammation-like response." Inflammation is a pathological term with four cardinal signs: redness (rubor), swelling (tumor), warmth (calor) and pain (dolor). These are not symptoms know to occur in C. elegans. The authors could consider using "tolerance" instead, as this term may better describe their findings.

We have changed “inflammation-like response” to “aberrant immune response” throughout the manuscript.

(2) It would help the reader to better understand the novelty of the findings in this study if the authors include a paragraph in their introduction to put their results in context of the published literature that has examined the relationship between immune activation and nematode health and survival. In particular, I suggest that the authors discuss doi:10.7554/eLife.74206 (2022), a study that charcterized a similar observation to what the authors are reporting. This study found that low cholesterol reduces pathogen tolerance and host survival during pathogen infection. Cholesterol scarcity increases p38 PMK-1 phosphorylation, priming immune effector induction in a manner that reduces pathogen accumulation in the intestine during a subsequent infection. I also suggest that the authors highlight in this introductory paragraph that the toxic effects of inappropriate immune activation in C. elegans has been widely catalogued. For example: doi.org/10.1371/journal.ppat.1011120 (2023); doi:10.1186/s12915-016-0320-z (2016).; doi:10.1126/science.1203411 (2011); doi:10.1534/g3.115.025650 (2016).

In this context, the authors could consider re-wording their novelty claim in the abstract and introduction to take into account this previous body of work.

We have added a paragraph to the Discussion section to place our findings in the context of previous research. The revised manuscript now includes the following text (page 11, lines 336–344): “Previous studies have shown that hyperactivation of immune pathways can negatively affect organismal development. For example, sustained activation of the p38 MAPK pathway impairs development in C. elegans (Cheesman et al., 2016; Kim et al., 2016), and excessive activation of the IPR also leads to developmental defects (Lažetić et al., 2023). Similar to our current study, recent work has demonstrated that heightened immune responses can reduce gut pathogen load while paradoxically decreasing host survival during infection (Ghosh and Singh, 2024; Peterson et al., 2022). However, our study uniquely shows that while such heightened immune responses are detrimental to immunocompetent animals, they can be beneficial in the context of immunodeficiency.”

(3) The authors rely on the use of RNAi of ufd-1 and npl-4 to study their effect on P. aeruginosa colonization and pathogen resistance throughout the manuscript. To address the possibility of off-target effects of the RNAi, the authors should consider both (i) showing with qRT-PCR that these genes are indeed targeted during RNAi, and (ii) confirming their phenotypes with an orthologous technique, preferably by studying ufd-1 and npl-4 loss-offunction mutants [both in the wild-type and sek-1(km4) backgrounds]. If mutation of these genes is lethal, the authors could use Auxin Inducible Degron (AID) technology to induce the degradation of these proteins in post-developmental animals.

We attempted several protocols of CRISPR in our laboratory to generate ufd-1 loss-of-function mutants; however, these efforts were unsuccessful. While this does not rule out the possibility of generating ufd-1 mutants, the failure is likely due to technical limitations on our part rather than an inherent inability to disrupt the gene. Nevertheless, to confirm the specificity of our RNAi-based approach, we quantified ufd-1 and npl-4 mRNA levels following RNAi treatment and found that each gene was specifically and effectively downregulated by its respective RNAi. 

Importantly, ufd-1 and npl-4 RNA sequences do not share significant homology, yet knockdown of either gene results in nearly identical phenotypes, including reduced survival on P. aeruginosa, diminished intestinal colonization, and shortened lifespan. These consistent outcomes strongly support the conclusion that the phenotypes are attributable to the disruption of the functional UFD-1-NPL-4 complex. We have added these results in the revised manuscript (pages 4-5, lines 114-125): “To confirm the specificity of the RNAi knockdowns and rule out potential off-target effects, we examined transcript levels of ufd-1 and npl-4 following RNAi treatment. RNAi against ufd-1 significantly reduced ufd-1 mRNA levels without reducing npl-4 expression, while npl-4 RNAi specifically downregulated npl-4 transcripts with no impact on ufd-1 mRNA levels (Figure 1—figure supplement 1A and B). Additionally, alignment of ufd-1 and npl-4 mRNA sequences against the C. elegans transcriptome revealed no significant similarity to other genes, supporting the specificity of the RNAi constructs. Moreover, the ufd-1 and npl-4 RNA sequences do not share significant sequence similarity. Therefore, the highly similar phenotypes observed in ufd-1 and npl-4 knockdown animals, including shortened lifespan, reduced survival on P. aeruginosa, and decreased intestinal colonization with P. aeruginosa, strongly suggest that these outcomes result from the disruption of the functional UFD-1-NPL-4 complex.”

(4) I am confused about the authors explanation regarding their observation that inhibition of the UFD-1/ NPL-4 complex extends the lifespan of sek-1(km25) animals, but not pmk-1(km25) animals, as SEK-1 is the MAPKK that functions immediately upstream of the p38 MAPK PMK-1 to promote pathogen resistance.

I am also confused why their RNA-seq experiment revealed a signature of intracellular pathogen response genes and not PMK-1 targets, which the authors propose is accounting for toxic immune activation. Activation of which immune response leads to toxicity?

We consistently observe that sek-1(km4) mutants are more sensitive to P. aeruginosa infection than pmk-1(km25) mutants, a finding also reported in previous studies (for example, PMID: 33658510). Given that SEK-1 functions upstream of PMK-1 in the MAPK signaling cascade, it is plausible that SEK-1 also regulates additional MAP kinases, such as PMK-2 (PMID: 25671546), which could contribute to the enhanced susceptibility observed in sek-1 mutants.

Our results show that inhibition of the UFD-1-NPL-4 complex improves survival specifically in severely immunocompromised animals, such as sek-1(km4) mutants, but not in pmk1(km25) mutants. To further validate this, we generated the double mutant dbl-1(nk3);pmk1(km25), which exhibits reduced survival on P. aeruginosa compared to either single mutant.

Notably, inhibition of the UFD-1-NPL-4 complex also enhances survival in the dbl1(nk3);pmk-1(km25) background, reinforcing the observation that this response is specific to severely compromised immune states.

We would also like to clarify that the observed phenotypes are independent of the SEK1/PMK-1 pathway, as shown in Figure 3A-3C, Figure 3—figure supplement 1, and Figure 4A-4C. The IPR seems to play a role in the observed phenotypes, as inhibition of some of the protease and pals genes (IPR genes) leads to increased P. aeruginosa colonization in ufd-1 knockdown animals (Figure 6—figure supplement 1). The other immune response pathway that leads to the observed phenotypes is ELT-2, as explained in Figure 6. Finally, we have included in the revised manuscript a note that, in addition, as-yet unidentified pathways are also likely contributing to the phenotypes triggered by disruption of the UFD-1-NPL-4 complex.

(5) The authors did not test alternative explanations for why UFD-1/ NPL-4 complex inhibition compromises survival during pathogen infection, other than exuberant immune activation. For example, it is possible that inhibition of this proteosome complex shortens lifespan by compromising the general health/ normal physiology of nematodes. Immune responses could be activated as a secondary consequence of this stress, and not be a direct cause of early morality. Does sek-1(km4) mutant suppress the lifespan shortened lifespan of ufd-1 and npl-4 knockdown? This experiment should also be done with loss-offunction mutants, as noted in point 3.

We have already included this data in Figure 4D, where we observed that ufd-1 and npl-4 knockdown reduce the lifespan of sek-1(km4) animals. It is possible that immune activation is a secondary consequence of cellular stress induced by inhibition of the UFD-1NPL-4 complex. However, our data strongly suggest that the observed phenotypes, including reduced gut pathogen load and decreased survival on the pathogen, are due to the aberrant immune response activated by the inhibition of the UFD-1-NPL-4 complex. Evidence from sek-1(km4) mutants particularly underscores the role of this dysregulated immune activation. While this aberrant immune response is detrimental to wild-type animals under pathogenic conditions, it appears to be beneficial in severely immunocompromised backgrounds. Specifically, in sek-1(km4) mutants, inhibition of the UFD-1-NPL-4 complex enhances survival during P. aeruginosa infection (Figure 4A). However, under non-infectious conditions, where sek-1(km4) mutants exhibit a normal lifespan, the same immune activation becomes harmful (Figure 4D). Together, these findings demonstrate that the aberrant immune response induced by UFD-1–NPL-4 inhibition is context-dependent: it is advantageous only for immunocompromised animals under infection, but deleterious to healthy animals under infection and to both healthy and immunocompromised animals under non-infectious conditions.

(6) The conclusion of Figure 6 hinges on an experiments that uses double RNAi to knockdown two genes at the same time (Fig. 6D and 6G), an approach that is inherently fraught in C. elegans biology owing the likelihood that the efficiency of RNAi-mediated gene knockdown is compromised and may account for the observed phenotypes. The proper control for double RNAi is not empty vector + ufd-1(RNAi), but rather gfp(RNAi) + ufd1(RNAi), as the introduction of a second hairpin RNA is what may compromise knockdown efficiency. In this context, it is important to confirm that knockdown of both genes occurs as expected (with qRT-PCR) and to confirm this phenotype using available elt-2 loss-of-function mutants.

We thank the reviewer for this helpful suggestion. We have repeated all double

RNAi experiments using gfp RNAi as a control instead of the empty vector (Figure 6 and Figure 6—figure supplement 1). Additionally, we assessed the efficiency of gene knockdown in the double RNAi conditions (Figure 6—figure supplement 2) and found that RNAi efficacy was not compromised by the double RNAi treatment.

(7) A supplementary table with the source data for at least three replications (mean lifespan, n, statistical comparison) for each pathogenesis assay should be included in this manuscript.

The source data is provided for all the data presented in the manuscript.

Reviewer #2 (Public Review):

Summary:

The authors aimed to uncover what role, if any, the UFD1/NPL4 complex might play in the innate immune responses of the nematode C. elegans. The authors find that loss of the complex renders animals more sensitive to both pathogenic and non-pathogenic bacteria. However, there appears to be a complex interplay with known innate immune pathways since the loss of UFD1/NPL4 actually results in increased survival of animals lacking the canonical innate immune pathways.

We thank the reviewer for providing an excellent summary of our work.

Strengths:

The authors perform robust genetic analysis to exclude and include possible mechanisms by which the UFD1/NPL4 pathway acts in the innate immune response.

We thank the reviewer for highlighting the strengths of our work.

Weaknesses:

The argument that the loss of the UFD1/NPL4 complex triggers a response that mimics that of an intracellular pathogen has not been thoroughly investigated. Additionally, the finding of a role of the GATA transcription factor, ELT-2, in this response is suggestive, but experiments showing sufficiency in the context of loss of the UFD1/NPL4 complex need to be explored.

We have investigated the role of IPR genes in the phenotypes observed upon ufd1 knockdown (Figure 6—figure supplement 1), and our results suggest that the IPR may contribute, at least in part, to the phenotypic outcomes of ufd-1 RNAi. In the Discussion section (pages 11–12, lines 345–356), we have included a detailed discussion on the possible mechanisms underlying IPR activation upon inhibition of the UFD-1–NPL-4 complex. We agree that the interaction between the UFD-1–NPL-4 complex and the IPR is intriguing and warrants further investigation. However, we believe that an in-depth exploration of this interaction lies beyond the scope of the current study.

We have incorporated new data on ELT-2 overexpression in the revised manuscript. Overexpression of ELT-2 partially phenocopies the effects of ufd-1 knockdown, supporting the idea that other pathways likely contribute to the full spectrum of phenotypes observed upon UFD-1-NPL-4 complex inhibition. The revised manuscript reads (page 10, lines 311319): “To determine whether ELT-2 activation alone is sufficient to recapitulate the phenotypes observed upon UFD-1-NPL-4 complex inhibition, we analyzed animals overexpressing ELT-2. Similar to ufd-1 knockdown, ELT-2 overexpression led to a significant reduction in the colonization of the gut by P. aeruginosa (Figure 6—figure supplement 3A and 3B). However, overexpression of ELT-2 did not alter the survival of worms on P. aeruginosa (Figure 6—figure supplement 3C). Taken together, these findings suggest that the phenotypes triggered by disruption of the UFD-1-NPL-4 complex are partially mediated by ELT-2. However, additional pathways, yet to be identified, likely cooperate with ELT-2 to regulate both pathogen resistance and host survival.”

Reviewer #1 (Recommendations For The Authors):

The authors could consider avoiding the use of descriptors (e.g., "drastic") when presenting their data.

We have removed the descriptors.

Reviewer #2 (Recommendations For The Authors):

What happens with overexpression of ELT2?

Overexpression of ELT-2 partially recapitulates the phenotypes of ufd-1 knockdowns, indicating that additional pathways are likely involved in controlling the phenotypes observed upon inhibition of the UFD-1-NPL-4 complex. The revised manuscript reads (page 10, lines 311-319): “To determine whether ELT-2 activation alone is sufficient to recapitulate the phenotypes observed upon UFD-1-NPL-4 complex inhibition, we analyzed animals overexpressing ELT-2. Similar to ufd-1 knockdown, ELT-2 overexpression led to a significant reduction in the colonization of the gut by P. aeruginosa (Figure 6—figure supplement 3A and 3B). However, overexpression of ELT-2 did not alter the survival of worms on P. aeruginosa (Figure 6—figure supplement 3C). Taken together, these findings suggest that the phenotypes triggered by disruption of the UFD-1-NPL-4 complex are partially mediated by ELT-2. However, additional pathways, yet to be identified, likely cooperate with ELT-2 to regulate both pathogen resistance and host survival.”

The data with xbp-1 loss of function is very different than that of pek1 and atf-6. Does loss of ufd1/npl4 suppress the increased pathogen survival of xbp-1s overexpressing animals?

We have examined worms overexpressing XBP-1s and found that overexpression of XBP-1s does not rescue the phenotypes caused by ufd-1 knockdown. The revised manuscript reads (page 6, lines 167-174): “To further examine the role of XBP-1 in this context, we assessed the effect of ufd-1 knockdown in animals neuronally overexpressing the constitutively active spliced form of XBP-1 (XBP-1s), which has been previously associated with enhanced longevity (Taylor and Dillin, 2013). Knockdown of ufd-1 resulted in the reduced survival of XBP-1s-overexpressing animals on P. aeruginosa, despite a concurrent decrease in bacterial colonization of the gut (Figure 2—figure supplement 1A-C). This indicated that the XBP-1 pathway was not required for the reduced P. aeruginosa colonization of ufd-1 knockdown animals.” 

Lastly, while the pathogen burden is reduced in ufd1/npl4 loss and pumping rates are marginally affected, have you checked defecation rates? Could they be increased?

We thank the reviewer for this valuable suggestion. We measured defecation rates following ufd-1 and npl-4 knockdown and, unexpectedly, found that inhibition of ufd-1/npl-4 leads to a reduction in defecation frequency. These findings clearly indicate that altered defecation cannot explain the observed decrease in gut colonization. The revised manuscript reads (page 5, lines 138-148): “The clearance of intestinal contents through the defecation motor program (DMP) is known to influence gut colonization by P. aeruginosa in C. elegans (Das et al., 2023). It is therefore conceivable that knockdown of the UFD-1-NPL-4 complex might increase defecation frequency, thereby promoting the physical expulsion of bacteria and resulting in reduced gut colonization. To test this possibility, we measured DMP rates in animals subjected to ufd-1 and npl-4 RNAi. Contrary to this hypothesis, both ufd-1 and npl-4 knockdown animals exhibited a significant reduction in defecation frequency compared to control RNAi-treated animals (Figure 1—figure supplement 2C). This reduction in DMP rate persisted even after 12 hours of exposure to P. aeruginosa (Figure 1—figure supplement 2D). Thus, the change in the DMP rate in ufd-1 and npl-4 knockdown animals is unlikely to be the reason for the reduced gut colonization by P. aeruginosa.”

In summary, we would like to thank the reviewers again for providing constructive and thoughtful feedback. We believe we have fully addressed all the concerns of the reviewers by carrying out several new experiments and modifying the text. The manuscript has undergone substantial revision and has thereby improved significantly. We do hope that the evidence in support of the conclusions is found to be complete in the revised manuscript.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

Summary:

This manuscript presents a compelling study identifying RBMX2 as a novel host factor upregulated during Mycobacterium bovis infection.

The study demonstrates that RBMX2 plays a role in:

(1) Facilitating M. bovis adhesion, invasion, and survival in epithelial cells.

(2) Disrupting tight junctions and promoting EMT.

(3) Contributing to inflammatory responses and possibly predisposing infected tissue to lung cancer development.

By using a combination of CRISPR-Cas9 library screening, multi-omics, coculture models, and bioinformatics, the authors establish a detailed mechanistic link between M. bovis infection and cancer-related EMT through the p65/MMP-9 signaling axis. Identification of RBMX2 as a bridge between TB infection and EMT is novel.

Strengths:

This topic and data are both novel and significant, expanding the understanding of transcriptomic diversity beyond RBM2 in M. bovis responsive functions.

Weaknesses:

(1) The abstract and introduction sometimes suggest RBMX2 has protective anti-TB functions, yet results show it facilitates pathogen adhesion and survival. The authors need to rephrase claims to avoid contradiction.

We sincerely appreciate the reviewer's valuable feedback regarding the need to clarify RBMX2's role throughout the manuscript. We have carefully revised the text to ensure consistent messaging about RBMX2's function in promoting M. bovis infection. Below we detail the specific modifications made:\

(1) Introduction Revisions:

Changed "The objective of this study was to elucidate the correlation between host genes and the susceptibility of M.bovis infection" to "The objective of this study was to identify host factors that promote susceptibility to M.bovis infection"

Revised "RBMX2 polyclonal and monoclonal cell lines exhibited favorable phenotypes" to "RBMX2 knockout cell lines showed reduced bacterial survival"

Replaced "The immune regulatory mechanism of RBMX2" with "The role of RBMX2 in facilitating M.bovis immune evasion"

(2) Results Revisions:

Modified "RBMX2 fails to affect cell morphology and the ability to proliferate and promotes M.bovis infection" to "RBMX2 does not alter cell viability but significantly enhances M.bovis infection"

Strengthened conclusion in Figure 4: "RBMX2 actively disrupts tight junctions to facilitate bacterial invasion"

(3) Discussion Revisions:

Revised screening description: "We screened host factors affecting M.bovis susceptibility and identified RBMX2 as a key promoter of infection"

Strengthened concluding statement: "In summary, RBMX2 drives TB pathogenesis by compromising epithelial barriers and inducing EMT"

These targeted revisions ensure that:

All sections consistently present RBMX2 as promoting infection; the language aligns with our experimental finding; potential protective interpretations have been eliminated. We believe these modifications have successfully addressed the reviewer's concern while maintaining the manuscript's original structure and scientific content. We appreciate the opportunity to improve our manuscript and thank the reviewer for this constructive suggestion.

(2) While p65/MMP-9 is convincingly implicated, the role of MAPK/p38 and JNK is less clearly resolved.

We sincerely appreciate the reviewer's insightful comment regarding the roles of MAPK/p38 and JNK in our study. Our experimental data clearly demonstrated that RBMX2 knockout significantly reduced phosphorylation levels of p65, p38, and JNK (Fig. 5A), indicating potential involvement of all three pathways in RBMX2-mediated regulation.

Through systematic functional validation, we obtained several important findings:

In pathway inhibition experiments, p65 activation (PMA treatment) showed the most dramatic effects on both tight junction disruption (ZO-1, OCLN reduction) and EMT marker regulation (E-cadherin downregulation, N-cadherin upregulation);p38 activation (ML141 treatment) exhibited moderate effects on these processes; JNK activation (Anisomycin treatment) displayed minimal impact.

Most conclusively, siRNA-mediated silencing of p65 alone was sufficient to:

Restore epithelial barrier function

Reverse EMT marker expression

Reduce bacterial adhesion and invasion

These results establish a clear hierarchy in pathway importance: p65 serves as the primary mediator of RBMX2's effects, while p38 plays a secondary role and JNK appears non-essential under our experimental conditions. We have now clarified this relationship in the revised Discussion section to strengthen this conclusion.

This refined understanding of pathway hierarchy provides important mechanistic insights while maintaining consistency with all our experimental data. We thank the reviewer for this valuable suggestion that helped improve our manuscript.

(3) Metabolomics results are interesting but not integrated deeply into the main EMT narrative.

Thank you for this constructive suggestion. In this article, we detected the metabolome of RBMX2 knockout and wild-type cells after Mycobacterium bovis infection, which mainly served as supporting evidence for our EMT model. However, we did not conduct an in-depth discussion of these findings. We have now added a detailed discussion of this section to further support our EMT model.

ADD:Meanwhile, metabolic pathways enriched after RBMX2 deletion, such as nucleotide metabolism, nucleotide sugar synthesis, and pentose interconversion, primarily support cell proliferation and migration during EMT by providing energy precursors, regulating glycosylation modifications, and maintaining redox balance; cofactor synthesis and amino sugar metabolism participate in EMT regulation through influencing metabolic remodeling and extracellular matrix interactions; chemokine and cGMP-PKG signaling pathways may further mediate inflammatory responses and cytoskeletal rearrangements, collectively promoting the EMT process.

(4) A key finding and starting point of this study is the upregulation of RBMX2 upon M. bovis infection. However, the authors have only assessed RBMX2 expression at the mRNA level following infection with M. bovis and BCG. To strengthen this conclusion, it is essential to validate RBMX2 expression at the protein level through techniques such as Western blotting or immunofluorescence. This would significantly enhance the credibility and impact of the study's foundational observation.

Thank you for your comment. We have supplemented the experiments in this part and found that Mycobacterium bovis infection can significantly enhance the expression level of RBMX2 protein.

(5) The manuscript would benefit from a more in-depth discussion of the relationship between tuberculosis (TB) and lung cancer. While the study provides experimental evidence suggesting a link via EMT induction, integrating current literature on the epidemiological and mechanistic connections between chronic TB infection and lung tumorigenesis would provide important context and reinforce the translational relevance of the findings.

We sincerely appreciate the valuable comments from the reviewer. We fully agree with your suggestion to further explore the relationship between tuberculosis (TB) and lung cancer. In the revised manuscript, we will add a new paragraph in the Discussion section to systematically integrate the current literature on the epidemiological and mechanistic links between chronic tuberculosis infection and lung cancer development, including the potential bridging roles of chronic inflammation, tissue damage repair, immune microenvironment remodeling, and the epithelial-mesenchymal transition (EMT) pathway. This addition will help more comprehensively interpret the clinical implications of the observed EMT activation in the context of our study, thereby enhancing the biological plausibility and clinical translational value of our findings.

ADD:There is growing epidemiological evidence suggesting that chronic TB infection represents a potential risk factor for the development of lung cancer. Studies have shown that individuals with a history of TB exhibit a significantly increased risk of lung cancer, particularly in areas of the lung with pre-existing fibrotic scars, indicating that chronic inflammation, tissue repair, and immune microenvironment remodeling may collectively contribute to malignant transformation 74. Moreover, EMT not only endows epithelial cells with mesenchymal features that enhance migratory and invasive capacity but is also associated with the acquisition of cancer stem cell-like properties and therapeutic resistance 75. Therefore, EMT may serve as a crucial molecular link connecting chronic TB infection with the malignant transformation of lung epithelial cells, warranting further investigation in the intersection of infection and tumorigenesis.

Reviewer #2 (Public review):

Summary:

I am not familiar with cancer biology, so my review mainly focuses on the infection part of the manuscript. Wang et al identified an RNA-binding protein RBMX2 that links the Mycobacterium bovis infection to the epithelial-Mesenchymal transition and lung cancer progression. Upon mycobacterium infection, the expression of RBMX2 was moderately increased in multiple bovine and human cell lines, as well as bovine lung and liver tissues. Using global approaches, including RNA-seq and proteomics, the authors identified differential gene expression caused by the RBMX2 knockout during M. bovis infection. Knockout of RBMX2 led to significant upregulations of tight-junction related genes such as CLDN-5, OCLN, ZO-1, whereas M. bovis infection affects the integrity of epithelial cell tight junctions and inflammatory responses. This study establishes that RBMX2 is an important host factor that modulates the infection process of M. bovis.

Strengths:

(1) This study tested multiple types of bovine and human cells, including macrophages, epithelial cells, and clinical tissues at multiple timepoints, and firmly confirmed the induced expression of RBMX2 upon M. bovis infection.

(2) The authors have generated the monoclonal RBMX2 knockout cell lines and comprehensively characterized the RBMX2-dependent gene expression changes using a combination of global omics approaches. The study has validated the impact of RBMX2 knockout on the tight-junction pathway and on the M. bovis infection, establishing RBMX2 as a crucial host factor.

Weaknesses:

(1) The RBMX2 was only moderately induced (less than 2-fold) upon M. bovis infection, arguing its contribution may be small. Its value as a therapeutic target is not justified. How RBMX2 was activated by M. bovis infection was unclear.

Thank you for your valuable and constructive comments. In this study, we primarily utilized the CRISPR whole-genome screening approach to identify key factors involved in bovine tuberculosis infection. Through four rounds of screening using a whole-genome knockout cell line of bovine lung epithelial cells infected with Mycobacterium bovis, we identified RBMX2 as a critical factor.

Although the transcriptional level change of RBMX2 was less than two-fold, following the suggestion of Reviewer 1, we examined its expression at the protein level, where the change was more pronounced, and we have added these results to the manuscript.

Regarding the mechanism by which RBMX2 is activated upon M. bovis infection, we previously screened for interacting proteins using a Mycobacterium tuberculosis secreted and membrane protein library, but unfortunately, we did not identify any direct interacting proteins from M. tuberculosis (https://doi.org/10.1093/nar/gkx1173).

(2) Although multiple time points have been included in the study, most analyses lack temporal resolution. It is difficult to appreciate the impact/consequence of M. bovis infection on the analyzed pathways and processes.

We appreciate the valuable comments from the reviewers. Although our study included multiple time points post-infection, in our experimental design we focused on different biological processes and phenotypes at distinct time points:

During the early phase (e.g., 2 hours post-infection), we focused on barrier phenotypes during the intermediate phase (e.g., 24 hours post-infection), we concentrated more on pathway activation and EMT phenotypes;

And during the later phase (e.g., 48–72 hours post-infection), we focused more on cell death phenotypes, which were validated in another FII article (https://doi.org/10.3389/fimmu.2024.1431207).

We also examined the impact of varying infection durations on RBMX2 knockout EBL cellular lines via GO analysis. At 0 hpi, genes were primarily related to the pathways of cell junctions, extracellular regions, and cell junction organization. At 24 hpi, genes were mainly associated with pathways of the basement membrane, cell adhesion, integrin binding and cell migration By 48 hpi, genes were annotated into epithelial cell differentiation and were negatively regulated during epithelial cell proliferation. This indicated that RBMX2 can regulate cellular connectivity throughout the stages of M. bovis infection.

For KEGG analysis, genes linked to the MAPK signaling pathway, chemical carcinogen-DNA adducts, and chemical carcinogen-receptor activation were observed at 0 hpi. At 24 hpi, significant enrichment was found in the ECM-receptor interaction, PI3K-Akt signaling pathway, and focal adhesion. Upon enrichment analysis at 48 hpi, significant enrichment was noted in the TGF-beta signaling pathway, transcriptional misregulation in cancer, microRNAs in cancer, small cell lung cancer, and p53 signaling pathway.

Reviewer #3 (Public review):

Summary:

This study investigates the role of the host protein RBMX2 in regulating the response to Mycobacterium bovis infection and its connection to epithelial-mesenchymal transition (EMT), a key pathway in cancer progression. Using bovine and human cell models, the authors have wisely shown that RBMX2 expression is upregulated following M. bovis infection and promotes bacterial adhesion, invasion, and survival by disrupting epithelial tight junctions via the p65/MMP-9 signaling pathway. They also demonstrate that RBMX2 facilitates EMT and is overexpressed in human lung cancers, suggesting a potential link between chronic infection and tumor progression. The study highlights RBMX2 as a novel host factor that could serve as a therapeutic target for both TB pathogenesis and infection-related cancer risk.

Strengths:

The major strengths lie in its multi-omics integration (transcriptomics, proteomics, metabolomics) to map RBMX2's impact on host pathways, combined with rigorous functional assays (knockout/knockdown, adhesion/invasion, barrier tests) that establish causality through the p65/MMP-9 axis. Validation across bovine and human cell models and in clinical tissue samples enhances translational relevance. Finally, identifying RBMX2 as a novel regulator linking mycobacterial infection to EMT and cancer progression opens exciting therapeutic avenues.

Weaknesses:

Although it's a solid study, there are a few weaknesses noted below.

(1) In the transcriptomics analysis, the authors performed (GO/KEGG) to explore biological functions. Did they perform the search locally or globally? If the search was performed with a global reference, then I would recommend doing a local search. That would give more relevant results. What is the logic behind highlighting some of the enriched pathways (in red), and how are they relevant to the current study?

We appreciate the reviewer's thoughtful questions regarding our transcriptomic analysis. In this study, we employed a localized enrichment approach focusing specifically on gene expression profiles from our bovine lung epithelial cell system. This cell-type-specific analysis provides more biologically relevant results than global database searches alone.

Regarding the highlighted pathways, these represent:

Temporally significant pathways showing strongest enrichment at each stage:

(1) 0h: Cell junction organization (immediate barrier response)

(2) 24h: ECM-receptor interaction (early EMT initiation)

(3) 48h: TGF-β signaling (chronic remodeling)

Mechanistically linked to our core findings about RBMX2's role in:

(1) Epithelial barrier disruption

(2) Mesenchymal transition

(3) Chronic infection outcomes

We selected these particular pathways because they:

(1) Showed the most statistically significant changes (FDR <0.001)

(2) Formed a coherent biological narrative across infection stages

(3) Were independently validated in our functional assays

This targeted approach allows us to focus on the most infection-relevant pathways while maintaining statistical rigor.

(2) While the authors show that RBMX2 expression correlates with EMT-related gene expression and barrier dysfunction, the evidence for direct association remains limited in this study. How does RBMX2 activate p65? Does it bind directly to p65 or modulate any upstream kinases? Could ChIP-seq or CLIP-seq provide further evidence for direct RNA or DNA targets of RBMX2 that drive EMT or NF-κB signaling?

We sincerely appreciate the reviewer's in-depth questions regarding the mechanisms by which RBMX2 activates p65 and its association with EMT. Although the molecular mechanism remains to be fully elucidated, our study has provided experimental evidence supporting a direct regulatory relationship between RBMX2 and the p65 subunit of the NF-κB pathway. Specifically, we investigated whether the transcription factor p65 could directly bind to the promoter region of RBMX2 using CHIP experiments. The results demonstrated that the transcription factor p65 can physically bind to the RBMX2 region.

Furthermore, dual-luciferase reporter assays were conducted, showing that p65 significantly enhances the transcriptional activity of the RBMX2 promoter, indicating a direct regulatory effect of RBMX2 on p65 expression.

These findings support our hypothesis that RBMX2 activates the NF-κB signaling pathway through direct interaction with the p65 protein, thereby participating in the regulation of EMT progression and barrier function.

In our subsequent work papers, we will also employ experiments such as CLIP to further investigate the specific mechanisms through which RBMX2 exerts its regulatory functions.

ADD and Revise in Results:

To thoroughly verify the regulatory mechanism between RBMX2 and p65, we initiated our investigation by conducting an in-depth analysis of the RBMX2 promoter region to identify potential interactions with the transcription factor p65. Initially, we performed molecular docking simulations to predict the binding affinity and interaction patterns between RBMX2 and p65 proteins. These simulations revealed multiple amino acid residues within the RBMX2 protein that formed strong, stable interactions with p65. The docking analysis yielded a high docking score of 1978.643 (Fig. 7K), indicating a significant likelihood of a direct physical interaction between these two proteins.

To complement the protein-protein interaction analysis, we next investigated whether p65 could directly bind to the promoter region of the RBMX2 gene at the transcriptional level. Using the JASPAR database, a comprehensive resource for transcription factor binding profiles, we queried the RBMX2 promoter sequence for potential p65 binding sites. This analysis identified several putative binding motifs, suggesting that p65 may act as a transcriptional regulator of RBMX2 expression.

To experimentally validate this transcriptional regulatory relationship, we employed a dual-luciferase reporter assay. We cloned the RBMX2 promoter region containing the predicted p65 binding sites into a luciferase reporter plasmid. This construct was then co-transfected into cultured cells along with a plasmid expressing p65. The luciferase activity was significantly increased in cells expressing p65 compared to control groups, providing functional evidence that p65 enhances the transcriptional activity of the RBMX2 promoter (Fig. 7I).

Furthermore, to confirm the direct binding of p65 to the RBMX2 promoter in a chromatin context, we performed chromatin immunoprecipitation followed by quantitative PCR (ChIP-qPCR). In this assay, we used specific antibodies against p65 to immunoprecipitate chromatin fragments containing p65-bound DNA. The enriched DNA fragments were then analyzed using primers targeting the RBMX2 promoter region. Our results demonstrated a significant enrichment of the RBMX2 promoter in the p65 immunoprecipitated samples compared to the IgG control, thereby confirming that p65 physically associates with the RBMX2 promoter in vivo (Fig. 7J). Collectively, these findings-ranging from computational docking predictions to transcriptional reporter assays and ChIP validation-provide strong evidence supporting a direct regulatory interaction between p65 and RBMX2. This regulatory mechanism may play a critical role in the biological pathways involving these two molecules, particularly in contexts such as inflammation, immune response, or cellular stress, where p65 (a subunit of NF-κB) is known to be prominently involved.

(3) The manuscript suggests that RBMX2 enhances adhesion/invasion of several bacterial species (e.g., E. coli, Salmonella), not just M. bovis. This raises questions about the specificity of RBMX2's role in Mycobacterium-specific pathogenesis. Is RBMX2 a general epithelial barrier regulator or does it exhibit preferential effects in mycobacterial infection contexts? How does this generality affect its potential as a TB-specific therapeutic target?

Thank you for your valuable comments. When we initially designed this experiment, we were interested in whether the RBMX2 knockout cell line could confer effective resistance not only against Mycobacterium bovis but also against Gram-negative and Gram-positive bacteria. Surprisingly, we indeed observed resistance to the invasion of these pathogens, albeit weaker compared to that against Mycobacterium bovis.

Nevertheless, we believe these findings merit publication in eLife. Moreover, RBMX2 knockout does not affect the phenotype of epithelial barrier disruption under normal conditions; its significant regulatory effect on barrier function is only evident upon infection with Mycobacterium bovis.

Importantly, during our genome-wide knockout library screening, RBMX2 was not identified in the screening models for Salmonella or Escherichia coli, but was consistently detected across multiple rounds of screening in the Mycobacterium bovis model.

(4) The quality of the figures is very poor. High-resolution images should be provided.

Thank you for your feedback; we provided higher-resolution images.

(5) The methods are not very descriptive, particularly the omics section.

Thank you for your comments; we have revised the description of the sequencing section.

(6) The manuscript is too dense, with extensive multi-omics data (transcriptomics, proteomics, metabolomics) but relatively little mechanistic integration. The authors should have focused on the key mechanistic pathways in the figures. Improving the narratives in the Results and Discussion section could help readers follow the logic of the experimental design and conclusions.

Thank you for your valuable comments. We have streamlined the figures and revised the description of the results section accordingly.

Reviewer #2 (Recommendations for the authors):

(1) The first part of the results and the major conclusions largely overlap with the previous paper by the same authors (Frontiers in Immunology, https://doi.org/10.3389/fimmu.2024.1431207). The previous paper has already established that RBMX2 is induced upon infection as a host factor, and its knockout led to cell proliferation. Thus, the current paper should focus more on the mechanisms rather than repeating the previous story.

We appreciate the reviewer's careful reading and constructive feedback. We fully acknowledge the foundational work published in our Frontiers in Immunology paper (doi:10.3389/fimmu.2024.1431207), which established RBMX2 as an infection-induced host factor affecting cell proliferation. The current study represents a significant mechanistic extension of these initial findings, with the following key advances:

(1) Novel Mechanistic Insights (Current Study Focus):

Discovery of the p65/MMP-9 pathway as the central mechanism mediating RBMX2's effects on EMT (Figs. 4-6)

First demonstration of RBMX2's role in epithelial barrier disruption (Figs. 2-3)

Identification of temporal regulation patterns during infection progression (Fig. 7)

(2) Expanded Biological Scope:

Demonstration of RBMX2's function in both bovine and human cell systems (vs. previous bovine-only data)

Clinical correlation with TB lesions

Therapeutic potential assessment through pathway inhibition

(3) Technical Advancements:

CRISPR-based mechanistic validation (vs. previous siRNA approach)

Multi-omics integration (transcriptomics + metabolomics)

Advanced live-cell imaging

We have now:

Removed redundant proliferation data from Results

Sharpened the Introduction to highlight mechanistic questions

Added explicit discussion comparing both studies

The current work provides the first comprehensive mechanistic framework for RBMX2's role in TB pathogenesis, moving substantially beyond the initial observational findings. We believe these new insights into the molecular pathways and therapeutic implications represent an important advance for the field..

(2) Line 107-110: The CRISPR screening results are not provided. Has it been published, or is it an unpublished dataset? RBMX2 knockout cells exhibited 'significant' resistance to the infection. How significant? Data?

Thank you for your valuable comments. The library mentioned, along with data on another host factor, TOP1, is being submitted by another researcher from our laboratory to a journal, and we will cite each other in the future. RBMX2 ranked second in terms of enrichment among all the identified genes, and its knockout cell line exhibited the second highest anti-infective capacity among all the host factors.

(3) Line 152: The RNA-seq analysis has already been performed/reported in the previous Frontiers paper. Therein, 173 genes were found to be differentially expressed. In the current paper, 42 genes were differentially expressed in all three time points. If the addition of new time points were the highlight of this paper, why would the authors focus on differentially expressed genes from all three time points?

Thank you for your valuable comments.

In the newly added data, we aimed to investigate the temporal changes during Mycobacterium bovis infection of host cells.

Previous study (Frontiers): Single 24h timepoint → 173 DEGs

Current study: Three timepoints (0h, 24h, 48h) with 42 consistently regulated genes → Reveals temporally stable core regulators of infection response

On one hand, we briefly described in the manuscript those important genes that exhibited changes across all time points.

On the other hand, in the supplementary materials, we also focused on the enriched genes at each individual time point, to better understand the temporal dynamics regulated by RBMX2.

(4) Line 153: The '0 h' time point is in fact 2 h post-infection. Why did the authors skip the real 0h time point? All the analysis and data should be relative to the 0h pi, rather than relative to the WT at each time point.

We appreciate the reviewer's important question regarding our timepoint nomenclature. The experimental timeline was designed as follows:

(1) Infection Protocol:

2h to 0h: Bacterial co-culture (MOI 20:1)

0h: Gentamicin (100 μg/ml) added to kill extracellular bacteria

0h+: Monitored intracellular survival

(2) Rationale for "0h" Designation:

This marks the onset of intracellular infection phase when Extracellular bacteria are eliminated (validated by plating)Host cell responses to intracellular pathogens begin All subsequent measurements reflect genuine infection (not attachment)

(3)Technical Validation:

Confirmed complete extracellular killing by:

Culture supernatant plating (0 CFU after gentamycin)

Microscopy ( no surface-associated bacteria)

(4) Comparative Analysis:

All data are presented as:

Fold-change relative to uninfected controls at each timepoint

We have now:

Clarified the timeline in Methods

Specified "0h = post-gentamicin" in all figure legends

This standardized approach aligns with established intracellular pathogen studies (e.g., Cell Microbiol. 2018;20:e12840). We're happy to adjust terminology if "0hpi (post-invasion)" would be clearer.

(5) Figure 2F: The data should be compared to the 0h pi, and show the temporal changes of gene expression.

Thank you for your suggestion. We have added additional information to this section. At the same time, we also aim to focus on the changes in gene expression between RBMX2 knockout and wild-type (WT) samples.

We have now:

Added temporal expression profiles relative to 0hpi baseline (SFig.4C).

Clarified the dual normalization approach in Methods

Maintained original between-group comparisons for phenotypic correlation

(6) Line 207. Not all the proteins were down-regulated post-infection.

Thank you for your comment. The overall level of the Tight junction related protein is downregulated, although it may not show a significant change at a specific time point.

We have revised our description, changing the keyword from "All" to "Most."

(7) Line 278, the introduction of the H1299 cell line should appear earlier when it was mentioned for the first time in the manuscript.

Thank you for your comment. We have provided a description in the abstract and Result1.

ADD:

Abstrat: Meanwhile, we also validated the EMT process in human lung epithelial cancer cells H1299.

Result 1: Furthermore, RBMX2-silenced H1299 cells exhibited a higher survival rate compared to H1299 ShNc cells after M. bovis infection (Fig. 1H).

(8) Figure 4 is huge and almost illegible, which may be divided into two figures.

Thank you for your valuable comments. We have streamlined the figures and revised the description of the results section accordingly.

Reviewer #3 (Recommendations for the authors):

I encountered frequent grammatical and syntactic issues. Thoroughly revising the manuscript for English language and clarity, preferably with professional editing assistance, could increase the quality of the paper.

Thank you for your valuable comments; we will invite a professional editor to polish the language.

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

In the manuscript, Aldridge and colleagues investigate the role of IL-27 in regulating hematopoiesis during T. gondii infection. Using loss-of-function approaches, reporter mice, and the generation of serial chimeric mice, they elegantly demonstrate that IL-27 induction plays a critical role in modulating bone marrow myelopoiesis and monocyte generation to the infection site. The study is well-designed, with clear experimental approaches that effectively adddress the mechanisms by which IL-27 regulates bone marrow myelopoiesis and prevents HSC exhaustion.

Reviewer #2 (Public review):

Summary:

Aldridge et al. aim to demonstrate the role of IL27 in limiting emergency myelopoiesis in response to Toxoplasma gondii infection by acting directly at the level of early haematopoietic progenitors.

They used different mouse genetic models, such as HSC lineage tracing, IL27 and IL27R-deficient mice, to show that:

(1) HSCs actively participate in emergency myelopoiesis during Toxoplasma gondii infection.

(2) The absence of IL27 and IL27R increases monocyte progenitors and monocytes, mainly inflammatory monocytes CCR2hi.

(3) At steady state, loss of IL27 impairs HSC fitness as competitive transplantation shows long-term engraftment deficiency of IL27 BM cells. This impairment is exacerbated after infection.

(4) IL27 is produced by various BM and other tissue cells at steady state, and its expression increases with infection, mainly by increasing the number of monocytes producing it.

Although it is indisputable that IL27 has a role in emergency myelopoiesis by limiting the number of proinflammatory monocytes in response to infection, the authors' claim that it acts only on HSCs and not on more committed progenitors (CMP, GMP, MP) is not supported by the quality of the data presented here, as described below in the weakness section. In addition, this study highlights a role for IL27 during infection, but does not focus on trained immunity, which is the focus of the targeted elife issue.

We thank the reviewer for these comments. We did try (and perhaps failed) to highlight that all cells within the HSPC category, which includes HSCs and MPPs, have the potential to contribute. The lack of IRGM1-RFP reporter expression in CMPs (Supp Fig5C) suggests that only HSCs and MPPs are progenitors that respond to IL-27 within the bone marrow, and thus that IL-27 signaling on these contributes to the effects observed on monopoiesis and peripheral monocyte populations. We have emphasized this in the revised manuscript, particularly in the introduction (line 82) and discussion (lines 469-472). While this manuscript does not focus solely on trained immunity, the impacts of infection regulating HSC differentiation and having a long-term impact on this compartment are a central theme of trained immunity. For example, Figure 6 and the supporting supplemental figures almost exclusively focus on the differentiation potential that is programed into LTHSCs by infection and the role of IL-27 in regulating this programing. Additionally, Figure 7 shows the long-term consequences of such training. The introduction      and discussion have been modified  to emphasize these connections to trained immunity.         

Weakness

(1) In Figure 4, MFI quantification is required. This figure also shows the expression level (FACS and RNA) in progenitors (GMP and CMP, GP, MP), which is quite similar to that of HSC at this level, so it is really surprising that CMP does not respond at all to IL27 (S5C).

As requested, we have included the MFIs, calculated as a fold change over control FMOs, in the revised manuscript. While HSPCs and CMPs show relatively similar RNA expression of Il27ra (Supp. Fig. 5 A), the levels of surface IL-27R expression by CMPs is lower than HSPCs (Fig. 4C, revised). Additional downstream progenitors (including GMPs) show highly reduced RNA expression and a corresponding low expression of the receptor protein. This is now more apparent with the quantified MFIs (Fig 4-5).

(2) Total BM was used to test the direct effect of IL27 on HSC. There could be an indirect effect from other more mature BM cells, even if they show lower receptor expression than HSC. This should be done on a different sorted population to prove the direct effect of IL27 on HSC. The authors need to look more closely at some stat-dependent genes or stat itself in different sorted cell populations, not just irgm1. It is also known that Stat is associated with increased HSC proliferation in response to IFN, which is the opposite of what is observed here.

We thank the reviewer for this question. We have found that the methanol fixation required to detect pSTAT disrupted the ability to stain for HSPCs by flow cytometry. Thus, we used the IRGM1 reporter, which we have found to be a sensitive and high-fidelity reporter of STAT1 activity while preserving epitope markers of HSPCs.

We agree that the use of bulk bone marrow in the in vitro stimulations could allow for the activation of non-HSPC cell types that are IL-27R+. This is now emphasized in the text. However, there are advantages to this bulk approach as it allows simultaneous analysis of all HSPC populations and downstream progenitors in the same cultures, allowing the ability to assess how the small numbers of IL-27R expressing lymphocytes present in these cultures respond (data that are now included, Supp. Fig. 5C). These cultures also allow a direct comparison of our IL-27R expression analysis with responsiveness to IL-27. Only a selection of the populations analyzed are shown in these data; however, all populations in Figure 4A were also analyzed in Supp. Fig. 5C. These data sets directly correlate receptor expression with sensitivity to IL-27. If this effect was indirect (i.e the ability of IL-27 to induce IFN-γ) then we would expect more robust expression of the IRGM1 reporter across other cell populations. However, while IFN-γ stimulates broad expression of IRGM1, the effects of IL-27 are restricted to HSPC and mature lymphocytes (Supp. Fig. 5C). In other words, the cells that express the highest levels of the IL-27R are most responsive to IL-27.

While we do not directly measure HSPC proliferation in these cultures, we agree with the reviewer that the decreased proportions of proliferating HSPCs seen in the absence of IL-27 during infection (Fig. 7A) is a complex data set. The reviewer is also correct that interferons can promote HSC proliferations; however, they can also promote cell stress, DNA damage, and even cell death of HSCs during chronic exposure (reviewed extensively in Demerdash, Y., et al. Exp Hematol. 2021. PMID: 33571568). Thus IFNs, much like IL-27, appear to regulate HSPCs with contextual importance, inducing their proliferation but also death. The activation of STAT1 and STAT3 by IL-27 may be at the core of some of these effects observed in our data, and we point out that IL-10, another activator of STAT1+3, has been shown to limit HSC responses to inflammation (lined 58-62), but we have also presented other possibilities in the discussion.

(3) The decrease in HSC fitness in IL27R KO at steady state could be an indirect effect of the increase in proinflammatory monocytes contributing to high levels of inflammatory cytokines in the BM and thus chronic HSC activation that is enhanced in response to infection. What is the pro-Inflammatory cytokine profile of the BM of IL27 OR IL27R deficient mice and of mixed chimera mice.

We thank the reviewer for this insightful comment. This was part of our stated rationale in generating the mixed WT:IL-27R-/- BM chimeras presented in Figure 2. In this mixed setting, there remained differences between the ability of the IL-27R sufficient and deficient stem cells to generate inflammatory macrophages. These results suggest that differences in the inflammatory environment do not account for the differences observed. This conclusion is further supported by the observation that the infection-induced levels of IFN-γ in the bone marrow are equivalent in the presence or absence of IL-27 (now included in the revised manuscript, Supp. Fig. 1F).

(4) Furthermore, the FACS profile of KI67/brdu of Figure 7 is doubtful, as it is shown in different literature that KSL are not predominantly quiescent as shown here, but about 50% are KI67-. This is also inconsistent with the increase of HSC observed in Figure 1. Quantification of total BruDU+ HSC and other progenitors is also important to quantify all cells that have proliferated during infection. As the repopulation of IL27-deficient BM is also lower in the absence of infection the proliation  of HSC in IL27R KO mice in the absence of infection is also important.

The comment indicates that the reviewer is concerned that our staining for Ki67 is on the low end of reported literature (~10-50% of LSKs, depending on age of the mice and simulation (Thapa R, et al. Stem Cell Res Ther. 2023. PMID: 37280691; Nies KPH, et al. Cytometry A. 2018. PMID: 30176186)). Our stains were performed on cells from infected mice, which does alter the classic markers used to identify HSPCs. For this reason, we are stringent with our gating strategy and may be excluding more HSPCs than are included in other reports. We have included our FMO control in the revised manuscript to indicate our gating approach (Supp. Fig. 9A). While the population of Ki67+ HSPCs is low, these results were consistent between our experiments and provide data sets that are interpretable.

(5) The immunofluorescence in Figure 3 shows a high level of background and it is difficult to see the GFP and tomato positive cells. In this sense, the number of HSCs quantified as Procr+ (more than 8000 on a single BM section) is inconsistent with the total number of HSCs that a BM can contain (i.e., around 6000 per BM as quantified in Figure 1).

We agree with the reviewer and have found that there is a high level of background in these stains. We have thresholded these images, as described in our methods, to minimize this. Additionally, the increased numbers of Procr+ cells in the imaging vs our flow data is expected, and has been reported by others (Steinert, EM, et al. Cell. 2015. PMID: 25957682).

(6) The addition of arrows to the figure will help to visualise positive cells. It is also not clear why the author normalised the GFP+ cells to the tomato+ cells in Figure 3D.

We thank the reviewer for this comment and have added the suggested arrows. We have also included a more detailed explanation for our normalization strategy.

(7) Furthermore, even if monocytes represent a high proportion of IL27-producing cells, they are only 50% of the cells at 5dpi, as shown in Figure 3 and S4. Without other monocyte markers, line 307 is incorrect.

We thank the reviewer for this clarification and have adjusted the text accordingly.

(8) How do the authors explain that in Figure 1, 5-10% of labelled precursors and monocytes can give 100% of monocytes? This would mean that only labelled HSC can differentiate into PEC monocytes. 5

We thank the reviewer for their interest in this result. Monocytes and macrophages are some

Reviewer #1 (Recommendations for the authors):

I have two minor comments that could enhance the conceptual framework of this study:

(1) The authors indirectly show that IL-27R expression on HSPCs is necessary for regulating HSC proliferation and preventing exhaustion. However, given that they have access to IL-27RFlox mice, they could cross these with Fgd5Cre mice to specifically delete IL-27R on long-term HSCs. This would provide direct evidence for the role of IL-27 signaling in LTHSCs during infection.

We appreciate this comment and did attempt this experiment with several HSPC specific Cres, including the Procr-cre (used elsewhere in the manuscript) and the MDS1-cre-ERT2 (Jackson Laboratory Strain #:032863). Unfortunately, validation revealed that deletion efficiency of the IL-27R with these HSCspecific Cre lines was inefficient, and so experiments are ongoing to enhance efficiency of the deletion and test alternative Cre lines (such as the Fgd5-cre).

(2) Since memory T and B cells often home to the bone marrow, it would be interesting to consider the potential cross-talk between these cells, HSPCs, and IL-27 signaling during secondary T. gondii infection. A brief discussion of this possibility would strengthen the study's broader implications.

We thank the reviewer for this opportunity. We have previously investigated the interplay between immune cells in the bone marrow (Glatman Zaretsky A, et al. Cell Rep. 2017. PMID: 28228257) and now include these possibilities in the discussion (line 465-470).

Reviewer #2 (Recommendations for the authors):

Minor points:

(1) Figures 6F and 7B: should be shown as % of donor and not total number to clarify the lineage potency of LTHSC. The fact that the results of transplantation are separated into different figures makes it not easy to follow. To see if the increase in monocyte production by IL27 KO BM is specific, the percent of donorderived cells for other populations, such as lymphoid, but also in MP, and inflammatory monocytes, is necessary to confirm Figure 2.

Perhaps there has been a misunderstanding? In these plots, we are not analyzing mixed chimeras but single transfer chimeras into lethally irradiated hosts. Thus, the % of donor reaches ~80- 90%. However, to measure the actual output of the HSPCs, the cell number was necessary to compare amongst groups. Additional description is provided in the figure legends and in the text of the manuscript (lines 391-392, 434-436, 651-653, and 680-682).

(2) The heavy UMAP description is unnecessary. Responses As requested, we have reduced this description of how the UMAPs were derived.

As requested, we have reduced this description of how the UMAPs were derived

Author response:

The following is the authors’ response to the original reviews.

Reviewer #1 (Public review):

Summary:

The authors were attempting to describe whether trained innate immunity would modulate antibody-dependent cellular phagocytosis (ADCP) and/or efferocytosis.

Strengths:

The use of primary murine macrophages, and not a cell line, is considered a strength. The trained immunity-mediated changes to phagocytosis affected both melanoma and breast cancer cells. The broad effect is consistent with trained immunity.

Weaknesses:

The most significant weakness, also noted by the authors in the discussion, is the lack of in vivo data. Without these data, it is not possible to put the in vitro data in context. It is unknown if the described effects on efferocytosis will be relevant to the in vivo progression of cancer.

We thank the reviewer for these comments. To examine the role of trained immunity on the modulation of macrophage efferocytosis in vivo, we performed immunostaining analysis in sections from B16F10 tumour samples.

Importantly, we found that macrophage efferocytosis of apoptotic tumour cells was significantly decreased in the tumour tissue that was excised from mice treated with β-glucan 7 days prior to tumour inoculation (supplementary Figure 3). These data are consistent with our findings using co-culture assays further strengthening the impact of our key findings in this report.

Reviewer #2 (Public review):

Summary:

The authors follow up their preclinical work on beta-glucan-induced trained immunity in murine tumor models that they published in Cell in 2020. In particular, they focus on the role of trained immunity and efferocytosis of cancer cells

Strengths:

While properly conducted, the work is underwhelming and fully depends on in vitro observations performed with co-cultures of bone marrow derived macrophages from beta-glucantreated mice and tumor cell lines. From these in vitro studies, the authors conclude that trained immunity induction has no effect on antibody-dependent cellular phagocytosis, while it decreases efferocytosis.

Weaknesses:

It would be important to study these phenomena in tumor mouse models in vivo. The authors clearly have the expertise as they have shown in previous studies. Especially because the in vitro observation appears to conflict with the in vivo anti-tumor found in mice prophylactically treated with beta-glucan. Clearly, trained immunity is associated with diverse cellular responses and mechanisms, some of which may promote tumor growth, as the current manuscript suggests, but in the absence of in vivo studies, it is merely a mechanistic exercise of which the relevance is difficult to determine.

We thank the reviewer for raising this important comment. We have followed reviewer’s suggestion and examined the role of trained immunity on the modulation of macrophage efferocytosis in vivo. As mentioned in our response to Reviewer 1, we demonstrate that efferocytosis of apoptotic melanoma cells in situ was attenuated in tumour samples from ‘trained’ mice as compared to those from controltreated mice.

Efferocytosis displays a pro-tumour and immunosuppressive role, therefore both our in vitro co-culture (Figure 1) and in vivo (supplementary Figure 3) findings are consistent with our previously published in vivo data supporting the tumour-suppressive role of prophylactic treatment with β-glucan (Kalafati, Kourtzelis et al, PMID: 33125892). 

Reviewer #3 (Public review):

Summary:

Chatzis et al showed that β-glucan trained macrophages have decreased phagocytic activity of apoptotic tumor cells and that is accompanied by lower levels of secreted IL-1β using a mouse model. Strengths: This finding has a potential impact on designing new cancer immunotherapeutic approaches by targeting macrophage efferocytosis.

Weaknesses:

Whether this finding could be applied to other scenarios is underdetermined.

(1)  Does the decrease of efferocytosis also occur in human monocytes/macrophages after training?

(2)  Both β-glucan and BCG are well-trained innate immunity agents, the authors showed that β-glucan decreased efferocytosis via IL-1 β, so it is interesting to know whether BCG has a similar effect.

We thank the reviewer for these comments. Our data suggest that induction of trained immunity with β-glucan contributes to decreased macrophage efferocytosis of tumour cells based on co-culture and in vivo approaches in a mouse setting.  

We agree with the reviewer that utilisation of a human setting would be important to provide additional validation of our findings.

Induction of trained immunity entails epigenetic and metabolic reprogramming of hematopoietic stem and progenitor cells (HSPCs). As such, the elucidation of mechanisms that modulate trained immunity in human cells would require the establishment of a macrophage differentiation model based on the use of HSPCs rather than the stimulation of monocytes or macrophages with β-glucan.

Additionally, the investigation of the impact of BCG in trained immunity-dependent phagocytosis would require the assessment of all different types of phagocytic cargos (apoptotic melanoma and breast cancer cells, apoptotic neutrophils, microbial bioparticles) as we did in the case of the β-glucan.  The capacity of different molecules to induce trained immunity in the efferocytosis setting requires further investigation that would be beyond the scope of this study. Therefore, we plan to address these very interesting points in a future study.

Additional text was added in the Discussion section to clarify the reviewer's points. In addition, we provide a more specific title that reflects better the specificity of our findings.

Author response:

The following is the authors’ response to the previous reviews

Reviewer #1 (Public review): 

Summary:

This fundamental work employed multidisciplinary approaches and conducted rigorous experiments to study how a specific subset of neurons in the dorsal striatum (i.e., "patchy" striatal neurons) modulates locomotion speed depending on the valence of the naturalistic context. 

Strengths: 

The scientific findings are novel and original and significantly advance our understanding of how the striatal circuit regulates spontaneous movement in various contexts.  Response: We appreciate the reviewer’s positive evaluation.

Weaknesses: 

This is extensive research involving various circuit manipulation approaches. Some of these circuit manipulations are not physiological. A balanced discussion of the technical strengths and limitations of the present work would be helpful and beneficial to the field. Minor issues in data presentation were also noted. 

We have incorporated the recommended discussion of technical limitations and addressed the physiological plausibility of our manipulations on Page 33 of the revised Discussion section. Specifically, we wrote: 

“Judicious interpretation of the present data must consider the technical limitations of the various methods and circuit-level manipulations applied. Patchy neurons are distributed unevenly across the extensive structure of the striatum, and their targeted manipulation is constrained by viral spread in the dorsal striatum. Somatic calcium imaging using single-photon microscopy captures activity from only a subset of patchy neurons within a narrow focal plane beneath each implanted GRIN lens. Similarly, limitations in light diffusion from optical fibers may reduce the effective population of targeted fibers in both photometry and optogenetic experiments. For example, the more modest locomotor slowing observed with optogenetic activation of striatonigral fibers in the SNr compared to the stronger effects seen with Gq-DREADD activation across the dorsal striatum could reflect limited fiber optic coverage in the SNr.Alternatively, it may suggest that non-striatonigral mechanisms also contribute to generalized slowing. Our photometry data do not support a role for striatopallidal projections from patchy neurons in movement suppression. The potential contribution of intrastriatal mechanisms, discussed earlier, remains to be empirically tested. Although the behavioral assays used were naturalistic, many of the circuit-level interventions were not. Broad ablation or widespread activation of patchy neurons and their efferent projections represent non-physiological manipulations. Nonetheless, these perturbation results are interpreted alongside more naturalistic observations, such as in vivo imaging of patchy neuron somata and axon terminals, to form a coherent understanding of their functional role”.

Reviewer #2 (Public review):

Hawes et al. investigated the role of striatal neurons in the patch compartment of the dorsal striatum. Using Sepw1-Cre line, the authors combined a modified version of the light/dark transition box test that allows them to examine locomotor activity in different environmental valence with a variety of approaches, including cell-type-specific ablation, miniscope calcium imaging, fiber photometry, and opto-/chemogenetics. First, they found ablation of patchy striatal neurons resulted in an increase in movement vigor when mice stayed in a safe area or when they moved back from more anxiogenic to safe environments. The following miniscope imaging experiment revealed that a larger fraction of striatal patchy neurons was negatively correlated with movement speed, particularly in an anxiogenic area. Next, the authors investigated differential activity patterns of patchy neurons' axon terminals, focusing on those in GPe, GPi, and SNr, showing that the patchy axons in SNr reflect movement speed/vigor. Chemogenetic and optogenetic activation of these patchy striatal neurons suppressed the locomotor vigor, thus demonstrating their causal role in the modulation of locomotor vigor when exposed to valence differentials. Unlike the activation of striatal patches, such a suppressive effect on locomotion was absent when optogenetically activating matrix neurons by using the Calb1-Cre line, indicating distinctive roles in the control of locomotor vigor by striatal patch and matrix neurons. Together, they have concluded that nigrostriatal neurons within striatal patches negatively regulate movement vigor, dependent on behavioral contexts where motivational valence differs.

We are grateful for the reviewer’s thorough summary of our main findings.

In my view, this study will add to the important literature by demonstrating how patch (striosomal) neurons in the striatum control movement vigor. This study has applied multiple approaches to investigate their functionality in locomotor behavior, and the obtained data largely support their conclusions. Nevertheless, I have some suggestions for improvements in the manuscript and figures regarding their data interpretation, accuracy, and efficacy of data presentation

We appreciate the reviewer’s overall positive assessment and have made substantial improvements to the revised manuscript in response to reviewers’ constructive suggestions.

(1) The authors found that the activation of the striatonigral pathway in the patch compartment suppresses locomotor speed, which contradicts with canonical roles of the direct pathway. It would be great if the authors could provide mechanistic explanations in the Discussion section. One possibility is that striatal D1R patch neurons directly inhibit dopaminergic cells that regulate movement vigor (Nadal et al., Sci. Rep., 2021; Okunomiya et al., J Neurosci., 2025). Providing plausible explanations will help readers infer possible physiological processes and give them ideas for future follow-up studies.

We have added the recommended data interpretation and future perspectives on Page 30 of the revised Discussion section. Specifically, we wrote:

“Potential mechanisms by which striatal patchy neurons reduce locomotion involve the supression of dopamine availability within the striatum. Dopamine, primarily supplied by neurons in the SNc and VTA,broadly facilitates locomotion (Gerfen and Surmeier 2011, Dudman and Krakauer 2016). Recent studies have shown that direct activation of patchy neurons leads to a reduction in striatal dopamine levels, accompanied by decreased walking speed (Nadel, Pawelko et al. 2021, Dong, Wang et al. 2025, Okunomiya, Watanabe et al. 2025). Patchy neuron projections terminate in structures known as “dendron bouquets”, which enwrap SNc dendrites within the SNr and can pause tonic dopamine neuron firing (Crittenden, Tillberg et al. 2016, Evans, Twedell et al. 2020). The present work highlights a role for patchy striatonigral inputs within the SN in decelerating movement, potentially through GABAergic dendron bouquets that limit dopamine release back to the striatum (Dong, Wang et al. 2025). Additionally, intrastriatal collaterals of patch spiny projection neurons (SPNs) have been shown to suppress dopamine release and associated synaptic plasticity via dynorphin-mediated activation of kappa opioid receptors on dopamine terminals (Hawes, Salinas et al. 2017). This intrastriatal mechanism may further contribute to the reduction in striatal dopamine levels and the observed decrease in locomotor speed, representing a compelling avenue for future investigation.”

(2) On page 14, Line 301, the authors stated that "Cre-dependent mCheery signals were colocalized with the patch marker (MOR1) in the dorsal striatum (Fig. 1B)". But I could not find any mCherry on that panel, so please modify it.

We have included representative images of mCherry and MOR1 staining in Supplementary Fig. S1 of the revised manuscript.

(3) From data shown in Figure 1, I've got the impression that mice ablated with striatal patch neurons were generally hyperactive, but this is probably not the case, as two separate experiments using LLbox and DDbox showed no difference in locomotor vigor between control and ablated mice. For the sake of better interpretation, it may be good to add a statement in Lines 365-366 that these experiments suggest the absence of hyperactive locomotion in general by ablating these specific neurons.

As suggested by the reviewer, we have added the following statement on Page 17 of the revised manuscript: “These data also indicate that PA elevates valence-specific speed without inducing general hyperactivity”.

(4) In Line 536, where Figure 5A was cited, the author mentioned that they used inhibitory DREADDs (AAV-DIO-hM4Di-mCherrry), but I could not find associated data on Figure 5. Please cite Figure S3, accordingly.

We have added the citation for the now Fig. S4 on Page 25 of the revised manuscript.

(5) Personally, the Figure panel labels of "Hi" and "ii" were confusing at first glance. It would be better to have alternatives.

As suggested by the reviewer, we have now labeled each figure panel with a distinct single alphabetical letter.

(6) There is a typo on Figure 4A: tdTomata → tdTomato

We have made the correction on the figure.

Reviewer #3 (Public review):

Hawes et al. combined behavioral, optical imaging, and activity manipulation techniques to investigate the role of striatal patch SPNs in locomotion regulation. Using Sepw1-Cre transgenic mice, they found that patch SPNs encode locomotion deceleration in a light-dark box procedure through optical imaging techniques. Moreover, genetic ablation of patch SPNs increased locomotion speed, while chemogenetic activation of these neurons decreased it. The authors concluded that a subtype of patch striatonigral neurons modulates locomotion speed based on external environmental cues. Below are some major concerns:

The study concludes that patch striatonigral neurons regulate locomotion speed. However, unless I missed something, very little evidence is presented to support the idea that it is specifically striatonigral neurons, rather than striatopallidal neurons, that mediate these effects. In fact, the optogenetic experiments shown in Fig. 6 suggest otherwise. What about the behavioral effects of optogenetic stimulation of striatonigral versus striatopallidal neuron somas in Sepw1-Cre mice?

Our photometry data implicate striatonigral neurons in locomotor slowing, as evidenced by a negative cross-correlation with acceleration and a negative lag, indicating that their activity reliably precedes—and may therefore contribute to—deceleration. In contrast, photometry results from striatopallidal neurons showed no clear correlation with speed or acceleration.

Figure 6 demonstrates that optogenetic manipulation within the SNr of Sepw1-Cre⁺ striatonigral axons recapitulated context-dependent locomotor changes seen with Gq-DREADD activation of both striatonigral and striatopallidal Sepw1-Cre⁺ cells in the dorsal striatum but failed to produce the broader locomotor speed change observed when targeting all Sepw1-Cre⁺ cells in the dorsal striatum using either ablation or Gq-DREADD activation. The more subtle speed-restrictive phenotype resulting from ChR activation in the SNr could, as the reviewer suggests, implicate striatopallidal neurons in broad locomotor speed regulation. However, our photometry data indicate that this scenario is unlikely, as activity of striatopallidal Sepw1-Cre⁺ fibers is not correlated with locomotor speed. Another plausible explanation is that the optogenetic approach may have affected fewer striatonigral fibers, potentially due to the limited spatial spread of light from the optical fiber within the SNr. Broad locomotor speed change in LDbox might require the recruitment of a larger number of striatonigral fibers than we were able to manipulate with optogenetics. We have added discussion of these technical limitations to the revised manuscript. Additionally, we now discuss the possibility that intrastriatal collaterals may contribute to reduced local dopamine levels by releasing dynorphin, which acts on kappa opioid receptors located on dopamine fibers (Hawes, Salinas et al. 2017), thereby suppressing dopamine release.

The reviewer also suggests an interesting experiment involving optogenetic stimulation of striatonigral versus striatopallidal somata in Sepw1-Cre mice. While we agree that this approach would yield valuable insights, we have thus far been unable to achieve reliable results using retroviral vectors. Moreover, selectively targeting striatopallidal terminals optogenetically remains technically challenging, as striatonigral fibers also traverse the pallidum, and the broad anatomical distribution of the pallidum complicates precise targeting. This proposed work will need to be pursued in a future study, either with improved retrograde viral tools or the development of additional mouse lines that offer more selective access to these neuronal populations as we documented recently (Dong, Wang et al. 2025).

In the abstract, the authors state that patch SPNs control speed without affecting valence. This claim seems to lack sufficient data to support it. Additionally, speed, velocity, and acceleration are very distinct qualities. It is necessary to clarify precisely what patch neurons encode and control in the current study.

We believe the reviewer’s interpretation pertains to a statement in the Introduction rather than the Abstract: “Our findings reveal that patchy SPNs control the speed at which mice navigate the valence differential between high- and low-anxiety zones, without affecting valence perception itself.” Throughout our study, mice consistently preferred the dark zone in the Light/Dark box, indicating intact perception of the valence differential between illuminated areas. While our manipulations altered locomotor speed, they did not affect time spent in the dark zone, supporting the conclusion that valence perception remained unaltered. We appreciate the reviewer’s insight and agree it is an intriguing possibility that locomotor responses could, over time, influence internal states such as anxiety. We addressed this in the Discussion, noting that while dark preference was robust to our manipulations, future studies are warranted to explore the relationship between anxious locomotor vigor and anxiety itself. We report changes in scalar measures of animal speed across Light/Dark box conditions and under various experimental manipulations. Separately, we show that activity in both patchy neuron somata and striatonigral fibers is negatively correlated with acceleration—indicating a positive correlation with deceleration. Notably, the direction of the cross-correlational lag between striatonigral fiber activity and acceleration suggests that this activity precedes and may causally contribute to mouse deceleration, thereby influencing reductions in speed. To clarify this, we revised a sentence in the Results section:

“Moreover, patchy neuron efferent activity at the SNr may causally contribute to deceleration, asindicated by the negative cross-correlational lag, thereby reducing animal speed.”. We also updated the Discussion to read: “Together, these data specifically implicate patchy striatonigral neurons in slowing locomotion by acting within the SNr to drive deceleration.”

One of the major results relies on chemogenetic manipulation (Figure 5). It would be helpful to demonstrate through slice electrophysiology that hM3Dq and hM4Di indeed cause changes in the activity of dorsal striatal SPNs, as intended by the DREADD system. This would support both the positive (Gq) and negative (Gi) findings, where no effects on behavior were observed.

We were unable to perform this experiment; however, hM3Dq has previously been shown to be effective in striatal neurons (Alcacer, Andreoli et al. 2017). The lack of effect observed in GiDREADD mice serves as an unintended but valuable control, helping to rule out off-target effects of the DREADD agonist JHU37160 and thereby reinforcing the specificity of hM3Dq-mediated activation in our study. We have now included an important caveat regarding the Gi-DREADD results, acknowledging the possibility that they may not have worked effectively in our target cells:

“Potential explanations for the negative results in Gi-DREADD mice include inherently low basal activity among patchy neurons or insufficient expression of GIRK channels in striatal neurons, which may limit the effectiveness of Gicoupling in suppressing neuronal activity (Shan, Fang et al. 2022).”

Finally, could the behavioral effects observed in the current study, resulting from various manipulations of patch SPNs, be due to alterations in nigrostriatal dopamine release within the dorsal striatum?

We agree that this is an important potential implication of our work, especially given that we and others have shown that patchy striatonigral neurons provide strong inhibitory input to dopaminergic neurons involved in locomotor control (Nadel, Pawelko et al. 2021, Lazaridis, Crittenden et al. 2024, Dong, Wang et al. 2025, Okunomiya, Watanabe et al. 2025). Accordingly, we have expanded the discussion section to include potential mechanistic explanations that support and contextualize our main findings.

Reviewer #1 (Recommendations for the authors):

Here are some minor issues for the authors' reference:

(1) This work supports the motor-suppressing effect of patchy SPNs, and >80% of them are direct pathway SPNs. This conclusion is not expected from the traditional basal ganglia direct/indirect pathway model. Most experiments were performed using nonphysiological approaches to suppress (i.e., ablation) or activate (i.e., continuous chemo-optogenetic stimulation). It remains uncertain if the reported observations are relevant to the normal biological function of patchy SPNs under physiological conditions. Particularly, under what circumstances an imbalanced patch/matrix activity may be induced, as proposed in the sections related to the data presented in Figure 6. A thorough discussion and clarification remain needed. Or it should be discussed as a limitation of the present work.

We have added discussion and clarification of physiological limitations in response to reviewer feedback. Additionally, we revised the opening sentence of an original paragraph in the discussion section to emphasize that it interprets our findings in the context of more physiological studies reporting natural shifts in patchy SPN activity due to cognitive conflict, stress, or training. The revised opening sentence now reads: “Together with previous studies of naturally occurring shifts in patchy neuron activation, these data illustrate ethologically relevant roles for a subgroup of genetically defined patchy neurons in behavior.”

(2) Lines 499-500: How striato-nigral cells encode speed and deceleration deserves a thorough discussion and clarification. These striatonigral cells can target both SNr GABAergic neurons and dendrites of the dopaminergic neurons. A discussion of microcircuits formed by the patchy SPNs axons in the SNr GABAergic and SNC DAergic neurons should be presented.

We have added this point at lines 499–500, including a reference to a relevant review of microcircuitry. Additionally, we expanded the discussion section to address microcircuit mechanisms that may underlie our main findings.

(3) Line 70: "BNST" should be spelled out at the first time it is mentioned.

This has been done.

(4) Line 133: only GCaMP6 was listed in the method, but GCaMP8 was also used (Figure 4). Clarification or details are needed.

Thank you for your careful attention to detail. We have corrected the typographical errors in the Methods section. Specifically, in the Stereotaxic Injections section, we corrected “GCaMP83” to “GCaMP8s.” In the Fiber Implant section, we removed the incorrect reference to “GCaMP6s” and clarified that GCaMP8s was used for photometry, and hChR2 was used for optogenetics.

(5) Line 183: Can the authors describe more precisely what "a moment" means in terms of seconds or minutes?

This has been done.

(6) Line 288: typo: missing / in ΔF

Thank you this has been fixed

(7) Line 301-302: the statement of "mCherry and MOR1 colocalization" does not match the images in Figure 1B.

This has been corrected by proving a new Supplementary Figure S1.

(8) Related to the statement between Lines 303-304: Figure 1c data may reflect changes in MOR1 protein or cell loss. Quantification of NeuN+ neurons within the MOR1 area would strengthen the conclusion of 60% of patchy cell loss in Figure 1C

Since the efficacy of AAV-FLEX-taCasp3 in cell ablation has been well established in our previous publications and those of others (Yang, Chiang et al. 2013, Wu, Kung et al. 2019), we do not believe the observed loss of MOR1 staining in Fig. 1C merely reflects reduced MOR1 expression. Moreover, a general neuronal marker such as NeuN may not reliably detect the specific loss of patchy neurons in our ablation model, given the technical limitations of conventional cell-counting methods like MBF’s StereoInvestigator, which typically exhibit a variability margin of 15–20%.

(9) Lines 313-314: "Similarly, PA mice demonstrated greater stay-time in the dark zone (Figure 1E)." Revision is needed to better reflect what is shown in Figure 1E and avoid misunderstandings.

Thank you this has been addressed.

(10) The color code in Figure 2Gi seems inconsistent with the others? Clarifications are needed

Color coding in Figure 2Gi differs from that in 2Eii out of necessity. For example, the "Light" cells depicted in light blue in 2Eii are represented by both light gray and light red dots in 2Gi. Importantly, Figure 2G does not encode specific speed relationships; instead, any association with speed is indicated by a red hue.

(11) Lines 538-539: the statement of "Over half of the patch was covered" was not supported by Figure 5C. Clarification is needed.

Thank you. For clarity, we updated the x-axis labels in Figures 1C and 5C from “% area covered” to “% DS area covered,” and defined “DS” as “dorsal striatal” in the corresponding figure legends. Additionally, we revised the sentence in question to read: “As with ablation, histological examination indicated that a substantial fraction of dorsal patch territories, identified through MOR1 staining, were impacted (Fig. 5C).”

(12) Figure 3: statistical significance in Figure 3 should be labeled in various panels.

We believe the reviewer's concern pertains to the scatter plot in panel F—specifically, whether the data points are significantly different from zero. In panel 3F, the 95% confidence interval clearly overlaps with zero, indicating that the results are not statistically significant.

(13) Figures 6D-E: no difference in the speed of control mice and ChR2 mice under continuous optical stimulation was not expected. It was different from Gq-DRADDS study in Figure 5E-F. Clarifications are needed.

For mice undergoing constant ChR2 activation of Sepw1-Cre+ SNr efferents, overall locomotor speed does not differ from controls. However, the BIL (bright-to-illuminated) effect on zone transitions isdisrupted: activating Sepw1-Cre⁺ fibers in the SNr blunts the typical increase in speed observed when mice flee from the light zone toward the dark zone. This impaired BIL-related speed increase upon exiting the light was similarly observed in the Gq-DREADD cohort. The reviewer is correct that this optogenetic manipulation within the SNr did not produce the more generalized speed reductions seen with broader Gq-DREADD activation of all Sepw1-Cre⁺ cells in the dorsal striatum. A likely explanation is the difference in targeting—ChR2 specifically activates SNr-bound terminals, whereas Gq-DREADD broadly activates entire Sepw1-Cre⁺ cells. Notably, many of the generalized speed profile changes observed with chemogenetic activation are opposite to those resulting from broad ablation of Sepw1-Cre⁺ cells. The more subtle speed-restrictive phenotype observed with ChR2 activation targeted to the SNr may suggest that fewer striatonigral fibers were affected by this technique, possibly due to the limited spread of light from the fiber optic. Broad locomotor speed change in LDbox might require the recruitment of a larger number of striatonigral fibers than we were able to manipulate with an optogenetic approach. Alternatively, it could indicate that non-striatonigral Sepw1-Cre⁺ projections—such as striatopallidal or intrastriatal pathways—play a role in more generalized slowing. If striatopallidal fibers contributed to locomotor slowing, we would expect to see non-zero cross-correlations between neural activity and speed or acceleration, along with negative lag indicating that neural activity precedes the behavioral change. However, our fiber photometry data do not support such a role for Sepw1-Cre⁺ striatopallidal fibers. We have also referenced the possibility that intrastriatal collaterals could suppress striatal dopamine levels, potentially explaining the stronger slowing phenotype observed when the entire striatal population is affected, as opposed to selectively targeting striatonigral terminals. These technical considerations and interpretive nuances have been incorporated and clarified in the revised discussion section.

(14) Lines 632: "compliment": a typo?

Yes, it should be “complement”.

(15) Figure 4 legend: descriptions of panels A and B were swapped

Thank you. This has been corrected.

(16) Friedman (2020) was listed twice in the bibliography (Lines 920-929).

Thank you. This has been corrected.

Reviewer #3 (Recommendations for the authors):

It will be helpful to label and add figure legends below each figure.

Thank you for the suggestion.

Editor's note:

Should you choose to revise your manuscript, if you have not already done so, please include full statistical reporting including exact p-values wherever possible alongside the summary statistics (test statistic and df) and, where appropriate, 95% confidence intervals. These should be reported for all key questions and not only when the p-value is less than 0.05 in the main manuscript. We noted some instances where only p values are reported.

Readers would also benefit from coding individual data points by sex and noting N/sex

We have included detailed statistical information in the revised manuscript. Both male and female mice were used in all experiments in approximately equal numbers. Since no sex-related differences were observed, we did not report the number of animals by sex.

References

Alcacer, C., L. Andreoli, I. Sebastianutto, J. Jakobsson, T. Fieblinger and M. A. Cenci (2017). "Chemogenetic stimulation of striatal projection neurons modulates responses to Parkinson's disease therapy." J Clin Invest 127(2): 720-734.

Crittenden, J. R., P. W. Tillberg, M. H. Riad, Y. Shima, C. R. Gerfen, J. Curry, D. E. Housman, S. B. Nelson, E. S. Boyden and A. M. Graybiel (2016). "Striosome-dendron bouquets highlight a unique striatonigral circuit targeting dopamine-containing neurons." Proc Natl Acad Sci U S A 113(40): 1131811323.

Dong, J., L. Wang, B. T. Sullivan, L. Sun, V. M. Martinez Smith, L. Chang, J. Ding, W. Le, C. R. Gerfen and H. Cai (2025). "Molecularly distinct striatonigral neuron subtypes differentially regulate locomotion." Nat Commun 16(1): 2710.

Dudman, J. T. and J. W. Krakauer (2016). "The basal ganglia: from motor commands to the control of vigor." Curr Opin Neurobiol 37: 158-166.

Evans, R. C., E. L. Twedell, M. Zhu, J. Ascencio, R. Zhang and Z. M. Khaliq (2020). "Functional Dissection of Basal Ganglia Inhibitory Inputs onto Substantia Nigra Dopaminergic Neurons." Cell Rep 32(11): 108156.

Gerfen, C. R. and D. J. Surmeier (2011). "Modulation of striatal projection systems by dopamine." Annual review of neuroscience 34: 441-466.

Hawes, S. L., A. G. Salinas, D. M. Lovinger and K. T. Blackwell (2017). "Long-term plasticity of corticostriatal synapses is modulated by pathway-specific co-release of opioids through kappa-opioid receptors." J Physiol 595(16): 5637-5652.

Lazaridis, I., J. R. Crittenden, G. Ahn, K. Hirokane, T. Yoshida, A. Mahar, V. Skara, K. Meletis, K.Parvataneni, J. T. Ting, E. Hueske, A. Matsushima and A. M. Graybiel (2024). "Striosomes Target Nigral Dopamine-Containing Neurons via Direct-D1 and Indirect-D2 Pathways Paralleling Classic DirectIndirect Basal Ganglia Systems." bioRxiv.

Nadel, J. A., S. S. Pawelko, J. R. Scott, R. McLaughlin, M. Fox, M. Ghanem, R. van der Merwe, N. G. Hollon, E. S. Ramsson and C. D. Howard (2021). "Optogenetic stimulation of striatal patches modifies habit formation and inhibits dopamine release." Sci Rep 11(1): 19847.

Okunomiya, T., D. Watanabe, H. Banno, T. Kondo, K. Imamura, R. Takahashi and H. Inoue (2025).

"Striosome Circuitry Stimulation Inhibits Striatal Dopamine Release and Locomotion." J Neurosci 45(4).

Shan, Q., Q. Fang and Y. Tian (2022). "Evidence that GIRK Channels Mediate the DREADD-hM4Di Receptor Activation-Induced Reduction in Membrane Excitability of Striatal Medium Spiny Neurons." ACS Chem Neurosci 13(14): 2084-2091.

Wu, J., J. Kung, J. Dong, L. Chang, C. Xie, A. Habib, S. Hawes, N. Yang, V. Chen, Z. Liu, R. Evans, B. Liang, L. Sun, J. Ding, J. Yu, S. Saez-Atienzar, B. Tang, Z. Khaliq, D. T. Lin, W. Le and H. Cai (2019). "Distinct Connectivity and Functionality of Aldehyde Dehydrogenase 1a1-Positive Nigrostriatal Dopaminergic Neurons in Motor Learning." Cell Rep 28(5): 1167-1181 e1167.

Wu, J., J. Kung, J. Dong, L. Chang, C. Xie, A. Habib, S. Hawes, N. Yang, V. Chen, Z. Liu, R. Evans, B. Liang, L. Sun, J. Ding, J. Yu, S. Saez-Atienzar, B. Tang, Z. Khaliq, D. T. Lin, W. Le and H. Cai (2019). "Distinct Connectivity and Functionality of Aldehyde Dehydrogenase 1a1-Positive Nigrostriatal Dopaminergic Neurons in Motor Learning." Cell Rep 28(5): 1167-1181 e1167.