Author Response

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

We thank the editor and the reviewers for their valuable and constructive feedback. In the revised manuscript, we have incorporated and addressed the suggestions provided by the reviewers.

Reviewer #1 (Recommendations For The Authors):

The primary recommendation is to provide additional language explaining how KinCytE will be updated.

Response: We appreciate the reviewer’s insightful feedback regarding the KinCytE update. In response, we have included additional details in the “Development and use of KinCyte’ section as follows: “We welcome researchers to actively participate in advancing the development of KinCytE by sharing external screening data, especially data on new secreted factors and cell types that extend beyond macrophages. This collaborative effort promises to enhance our understanding of kinase-focused networks, opening new avenues for cutting-edge therapeutic approaches”. In addition, we explicitly state in the "Data, Software, and Availability" section, "To contribute data, kindly email the corresponding author and refer to Table S2 for guidance on the preferred file format."

Reviewer #2 (Recommendations For The Authors):

Would have been nice to see a validation of the regression models from outside of the training data. I would also consider removing statements like "We anticipate that KinCytE will be highly sought after by biologists... " , it reads like a grant application (and this is not)! Could tone the language down a bit. In the future, you might consider displaying your graphs as "biofabrics", they're much cleaner than "hairballs" (PMID: 23102059). Or potentially, show a hierarchical view where the selected cytokine (or other) is at the root, and you can immediately see what's connected. Anyway, the network display can be expanded. Consider maybe adding the nearest neighbors to the table on the right after selecting the node. Generally, though, I like how it works.

There needs to be a button to download the graph as a .csv file. Maybe the subgraph after selecting a node (or set of nodes). Also, once you're at a graph view, it's hard to guess how to get back to the starting page. Maybe just one button with a "home" on it would fix that. On the Kinases Discovery, why are the gene symbols all lower case? Very cool!

Response:: We greatly value the reviewer's constructive suggestions. To incorporate these, we have made the following changes:

(1) "We anticipate that KinCytE will be highly sought after by biologists... " This sentence is removed.

(2) A ‘SAVE CSV’ button is added to the bottom right of the Cytokine Explorer page, which allows the users to download the graph as a csv file.

(3) A redesigned KinCyte logo now functions as the 'HOME' button, located at the top left of the webpage, ensuring that users can easily return to the homepage at any time.

Author Response

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

eLife assessment

The manuscript describes the synergy among PI3Kbeta activators, providing compelling results concerning the mechanism of their activation. The particular strengths of the work arise to a great extent from the reconstitution system better mimicking the natural environment of the plasma membrane than previous setups have. The study will be a landmark contribution to the signaling field.

Public Reviews:

Reviewer #1 (Public Review):

The manuscript aims to provide mechanistic insight into the activation of PI3Kbeta by its known regulators tyrosine phosphorylated peptides, GTP-loaded Rac1 and G-protein beta-gamma subunits. To achieve this the authors have used supported lipid bilayers, engineered recombinant peptides and proteins (often tagged with fluorophores) and TIRF microscopy to enable bulk (averages of many molecules) and single molecule quantitation. The great strength of this approach is the precision and clarity of mechanistic insight. Although the study does not use "in transfecto" or in vivo models the experiments are performed using "physiologically-based" conditions and provide a powerful insight into core regulatory principles that will be relevant in vivo.

The results are beautiful, high quality, well controlled and internally consistent (and with other published work that overlaps on some points) and as a result are compelling. The primary conclusion is that the primary regulator of PI3Kbeta are tyrosine phosphorylated peptides (and by inference tyrosine phosphorylated receptors/adaptors) and that the other activators can synergise with that input but have relatively weak impacts on their own.

Although the methodology is not easily imported, for reasons of both cost and the experience needed to execute them well, the results have broad importance for the field and reverse an impression that had built in large parts of the broader signalling and PI3K communities that all of the inputs to PI3Kbeta were relatively equivalent, however, these conclusions were based on "in cell" or in vivo studies that were very difficult to interpret clearly.

Reviewer #2 (Public Review):

The manuscript of Duewell et al has made critical observations that help to understand the mechanisms of activation of the class IA PI3Ks. By using single-molecule kinetic measurements, the authors have made outstanding progress toward understanding how PI3Kbeta is uniquely activated by phosphorylated tyrosine kinase receptors, Gbeta/gamma heterodimers and the small G protein Rac1. While previous studies have defined these as activators of PI3Kbeta, the current manuscript makes clear the quantitative limitations of these previous observations. Most previous quantitative in vitro studies of PI3Kbeta activation have used soluble peptides derived from bis-phosphorylated receptors to stimulate the enzyme. These soluble peptides stimulate the enzyme, and even stimulate membrane interaction. Although these previous studies showed that the release of p85-mediated autoinhibition unmasks an intrinsic affinity of the enzyme for lipid membranes, they ignored what would be the consequence of these peptide sequences being present in the context of intrinsic membrane proteins. The current manuscript shows that the effect of membrane-conjugated peptides on the enzyme activity is profound, in terms of recruiting the enzyme to membranes. In this context, the authors show that G proteins associated with the membranes have an important contribution to membrane recruitment, but they also have a profound allosteric effect on the activity on the membrane, These are observations that would not have been possible with bulk measurements, and they do not simply recapitulate observations that were made for other class IA PI3Ks.

An important observation that the authors have made is that Gbeta/gamma heterodimers and RAc1 alone have almost no ability to recruit PI3Kbeta to the membranes that they are using, and this is central to one of the most profoundly novel activation mechanisms offered by the manuscript. The authors propose that the nSH2- and Gbeta/gamma binding sites partially overlap, so that Gbeta/gamma can only bind once the nSH2 domain releases the p110beta subunit. This mechanism would mean that once the nSH2 is engaged by membrane-conjugated pY, the Gbg heterodimer can bind and increase the association of the enzyme with membranes. Indeed, this increased membrane association is observed by the authors. However, the authors also show that this increased recruitment to membranes accounts for relatively little increase in activity, and that the far greater component of activation is due to an allosteric effect of the membrane association on the activity of the enzyme. The proposal for competition between Gbg binding and the nSH2 is consistent with the behavior of an nSH2 mutant that cannot bind to pY and which, consequently, does not vacate the Gbg-binding site. In addition to the outstanding contribution to understanding the kinetics of activation of PI3Kbeta, the authors have offered the first structural interpretation for the kinetics of Gbg activation in synergy with pY activation. The proposal for an overlapping nSH2/Gbg binding site is supported by predictions made by John Burke, using alphafold multimer. Although there is no experimental structure to support this structural model, it is consistent with HDX-MS analyses that were published previously.

Reviewer #1 (Recommendations For The Authors):

1. The approx relative concentrations (surface densities ) of Rac1-GTP, GBetagammas and PY-peptides used in experiments in Fig 1 are not easy to understand and useful to give an intuitive feel for the relative sensitivity of the PI3Kbeta reporter to those inputs.

In our revised manuscript, we provide densities of the individual signaling inputs used to reconstitute Dy647-PI3Kβ membrane recruitment (see Figure legend 1). We provide a more detailed explanation about our quantification method in subsequent figures where the membrane surface density of signaling inputs is varied to modulate the strength of PI3Kβ membrane localization and activity.

Building off the quantification of Rac1-GTP and pY membrane density measurements presented in our initial manuscript submission, we now include an estimate of the GβGγ membrane density. For these new measurements, we recombinantly expressed and purified additional SNAP-GβGγ protein, which we fluorescently labeled with AlexaFluor 555. The membrane surface density of GβGγ was quantified at equilibrium using a combination of AF488-SNAP-GβGγ (bulk signal) and dilute AF555-SNAP-GβGγ (0.0025%), which allowed us to resolve and count the single molecule density (Figure 3A). We calculate the total surface density of GβGγ based on the AF555-SNAP-GβGγ dilution factor. In the methods section titled, “surface density calibration,” we describe our protocol.

1. The estimates of the PIP3 concentrations/densities measured using the BTK reporter seem good but its unclear (to me) how they were derived.

The density of PI(3,4,5)P3 lipids in our supported lipid bilayers was calculated based on the incorporation of a define molar ratio of PI(3,4,5)P3 in our small unilamellar vesicles. Based on the average footprint of 0.72 nm2 for a single lipid, we calculated the density of lipids per µm2. In the methods section titled, “kinetic measurements of PI(3,4,5)P3 lipid production,” we include the following description:

“Assuming an average footprint of 0.72 nm2 for phosphatidylcholine (Carnie et al., 1979; Hansen et al., 2019), we calculated a density of 2.8 × 104 PI(3,4,5)P3 lipids/μm2 for supported membranes that contain an initial concentrations of 2% PI(4,5)P2. We assume that the plateau fluorescence intensity of the AF488-SNAP-Btk sensor following reaction completion in the presence of PI3Kβ represents the production of 2% PI(3,4,5)P3. The bulk membrane intensity of AF488-SNAP-Btk was normalized from 0 to 1, and then multiplied times the total density of PI(3,4,5)P3 lipids to generate kinetic traces that report the kinetics of PI(3,4,5)P3 production.”

Minor points

l164; Rac1(GTP) AND GBeta gammas. In this context it should be OR. Or have I misunderstood?

l1093; kineticS measurementS.

Thank you for pointing out these typos. We made the appropriate edits.

The paper of Suire etal (Suire, S., Lécureuil, C., Anderson, K. E., Damoulakis, G., Niewczas, I., Davidson, K., Guillou, H., Pan, D., Jonathan Clark, Phillip T Hawkins, & Stephens, L. (2012). GPCR activation of Ras and PI3Kc in neutrophils depends on PLCb2/b3 and the RasGEF RasGRP4. The EMBO journal, 31(14), 3118-3129. https://doi.org/10.1038/emboj.2012.167) make the point that in vivo it appears that although Ras-activation is required for full activation of PI3Kgamma (and can activate PI3Kgamma in vitro directly) if you use tools to activate Ras in the absence of receptor and Gbetagamma signalling, it has no affect on PIP3 . This directly supports the authors conclusions.

Thank you for sharing this citation. We incorporated the reviewer’s insight into our discussion section to broaden the significance of our work.

Reviewer #2 (Recommendations For The Authors):

There are only a few relatively minor points that could be addressed to improve the paper:

1. Why is the density still going up after 10 minutes in Figure 1 Figure supplement 2? Doesn't this seem like a very long time? Are we seeing fast on/off combined with fast on/slow off? Are the particles eventually becoming stuck in odd places or are they slowly denaturing?

Our movies do not indicate a slow accumulation of immobilized or stuck Dy647-PI3Kβ particles on the membrane surface. On the long timescale, we believe that a small fraction of Dy647-PI3Kβ molecular do exhibit longer dwell times on membranes containing a high density of pY (>6,000 molecules/µm2). This is likely due to membrane hopping of Dy647-PI3Kβ. In other words, rather than Dy647-PI3Kβ dissociating from the membrane surface directly into the solution, the Dy647-PI3Kβ molecule immediately rebinds to another membrane conjugated pY peptide. This type of behavior of a peripheral membrane binding protein is generally correlated with there being a higher surface density of the binding partner (Yasui et al., 2014). Characterization of potential Dy647-PI3Kβ membrane hopping will require additional experimentation (e.g. PI3Kβ mutants) and quantitative analysis that goes beyond the scope of this study.

1. Lines 188-189. "By quantifying the average number of Alexa488-pY particles per unit area of supported membrane we calculated the absolute density of pY per μm2 (Figure 2D). I think this should be Figure 2C, right hand y-axis.

Thank you for identifying our typo. We’ve corrected the text for clarity.

1. Lines 102-193. "When Dy647-PI3Kβ was flowed over a membrane containing a low density of {less than or equal to} 500 pY/μm2, we observed rapid equilibration kinetics consistent with a 1:1 binding stoichiometry (Figure 2E).” There is no density shown in Fig. 2E. There is only "membrane intensity." Perhaps it was their intent to include a right-hand axis with density (number of particles/area), as they did in Figure 2C. However, they did not, so Figure 2E does not support the text. The value of Intensity/#py/um**2 does not appear to be the same for Figure 2C as for Figure 2E, assuming that the statement in the text is correct. The authors should include the density as a right-hand axis in 2E.

We have reworded this portion of the results section for clarity. In reading the reviewers comment, we recognize that a more convincing way to support our claim of a 1:1 binding stoichiometry would be to show that there are ~500 Dy647-PI3Kβ/μm2 membrane bound complexes when the pY surface density equals ~500 pY/μm2. For us to make this connection, we would need to perform experiments using a Dy647-PI3Kβ concentration that fully saturates all the binding pY binding sites. However, at this elevated Dy647-PI3Kβ solution concentration, individual Dy647-PI3Kβ complexes can start to bind to a single phosphotyrosine of the dually phosphorylated peptide due to competition for pY binding sites. As an alternative to performing the experiment described above, we can infer binding stoichiometry from the shape of the membrane absorption kinetic traces. For example, a simple bimolecular interaction exhibits rapid equilibration kinetics with a hyperbolic shaped kinetic trace. Systems that have more complex binding equilibria, however, generally take longer to equilibrate (due to the change in KOFF) and can often be broken down into 2 or 3 distinct dissociation constants (KD). This type of kinetic analysis has previously been used to describe multivalent membrane binding interactions for the Btk-PI(3,4,5)P3 (Chung et al., 2019) and PI3Kγ-GβGγ (Rathinaswamy et al., 2021) complexes. Considering that there are multiple interpretations of the Dy647-PI3Kβ membrane absorption traces show in Figure 2E, we refrain from saying that our results explicitly reveal a 1:1 binding stoichiometry. Instead, we provide several possible explanations for the results. Ultimately, additional experiments and kinetic modeling of wild type and mutant PI3Kβ is necessary to define the binding stoichiometry under different conditions.

1. Table 1. The authors have analysed the data to extract two dwell times and two diffusion coefficients. The legend should make this clear, referring to D1 as the slow diffusion component and D2 as fast diffusion, similarly, there are short and long dell times. This should be stated in the legend. There are two columns labelled "alpha". This presumably should be alpha1 and alpha2, the fractions of particles with short and long dwell times. The table legend should clarify this.

In our revision, additional text has been added to the figure legends and Table 1.

Text from Table 1: “Alpha (α) equals the fraction of molecules with the characteristic dwell time, τ1 (DT = dwell time). The fraction of molecules with the characteristic dwell time, τ2, equals 1-α. Alpha (αD) equals the fraction of molecules with the characteristic diffusion coefficient, D1. The fraction of molecules with diffusion coefficient, D2, equals 1-αD.”

1. In the legend for Figure 5 figure supplement 1, for part D, the "Cumulative membrane of binding events..." The "of" should be deleted.

Thank you for identifying this typo.

1. Lines 423-426: "We found that PI3Kβ kinase activity is also relatively insensitive to either Rac1(GTP) or GβGγ alone. This is in contrast to previous reports that showed Rho-GTPases (Fritsch et al. 2013) and GβGγ (Katada et al. 1999; Hashem A. Dbouk et al. 2012; Maier, Babich, and Nürnberg 1999) can activate PI3Kβ, albeit modest, compared to synergistic activation with pY peptides plus Rac1(GTP) or GβGγ." It is not clear what this statement means. On the surface, it might be interpreted as saying that these previous studies had some flaw that led the authors to conclude that there is some activation caused by Rac1 or Gbeta/gamma on their own. The current manuscript is an important contribution to understanding the mechanism of synergistic activation, but it is also true that the Hansen and his colleagues have not used the same membranes as were used previously. The authors state that they have used a wide range of membrane compositions, but the only ones that have appeared in the manuscript are nearly pure PC (with 2% PIP2) or PC with 20% PS. Extensive studies with varying membrane compositions are beyond the scope of the current study, since the current manuscript concisely makes important observations regarding mechanism. However, it would be helpful for readers if the authors at least mention the differences in membrane compositions among the studies.

The reviewer raises an important point concerning our interpretation of PI3Kβ activation data in relationship to existing literature. In our original submission, we made conclusions concerning how individual signaling inputs modulate PI3Kβ activity, without showing all our data or providing sufficient explanation. In our revised manuscript, we include PI3Kβ kinase activity measurements performed in the presence of either pY, Rac1(GTP), or GβGγ alone (Figure 5B-5C). These experiments were reconstituted on supported membranes in the absence or presence of 20% PS lipids. We found that increasing the density of anionic lipids increased the overall activity of PI3Kβ in the presence of pY or GβGγ alone. This is consistent with a subtle increase in PI3Kβ membrane affinity due to the negatively charged PS lipids. Mutations that disrupt the direct interaction between PI3Kβ and GβGγ eliminated the observed lipid kinase activity. We were unable to detect PI3Kβ activity in the presence of Rac1(GTP) alone. In conclusion, we’re able to detect some PI3Kβ activity in the presence of GβGγ alone, which is consistent with previous reports (Dbouk et al., 2010; Katada et al., 1999; Maier et al., 2000). In the future, a more comprehensive analysis will be required to map the relationship between PI3Kβ activity, membrane localization, and lipid composition. For example, previous reconstitutions have revealed differential activation of PI3Kα that depends on the most abundant lipid being phosphatidylethanolamine (PE) rather than phosphatidylcholine (PC) (Hon et al., 2012; Ziemba et al., 2016). PE lipids comprise 25-30% of the cellular plasma membrane (Yang et al., 2018) and have been used in previous studies to measure PI3K lipid kinase activity on small unilamellar vesicles (Dbouk et al., 2010; Hon et al., 2012).

In this study, we elected to use a simplified membrane composition that minimized non-specific membrane localization of fluorescently labeled PI3Kβ. This allowed us to more clearly define the strength of individual and combinations of protein-protein interactions that regulate PI3Kβ localization and kinase activity. When reconstituting amphiphilic molecules (i.e. lipids) in aqueous solution a variety of structures, including micelles, inverted micelles, and planar bilayers can form based on the lipid composition (Kulkarni, 2019). The organization of these membrane structures is related to the molecular packing parameter of the individual phospholipids (Israelachvili et al., 1976). The packing parameter (P=v⁄((a•l_c))) depends on the volume of the hydrocarbon (v), area of the lipid head group (a), and the lipid tail length (l_c). When generating supported lipid bilayers on a flat two-dimensional glass surface, we aim to create a fluid lamellar membrane. We find that phosphatidylcholine (PC) lipids are ideal for making supported lipid bilayers because they have a packing parameter of ~1 (Costigan et al., 2000). In other words, PC lipids are cylindrical like a paper towel roll. In contrast, cholesterol and phosphatidylethanolamine (PE) lipids have packing parameters of 1.22 and 1.11, respectively (Angelov et al., 1999; Carnie et al., 1979). This gives cholesterol and PE lipids an inverted truncated cone shape, which prefers to adopt a non-lamellar phase structure. Due to the intrinsic negative curvature of PE lipids, they can spontaneously form inverted micelles (i.e. hexagonal II phase) in aqueous solution when they are the predominant lipid species (Israelachvili et al., 1980; Kobierski et al., 2022; Wnętrzak et al., 2013). In the methods section of our manuscript, we note that from our experience incorporation of PE lipids dramatically reduces the protein-maleimide coupling efficiency, displayed more membrane defects, and resulted in a larger fraction of surface immobilized Dy647-PI3Kβ. This could be related to the intrinsic negative curvature of PE membranes. However, further investigation is needed to decipher these issues.

Angelov B, Ollivon M, Angelova A. 1999. X-ray Diffraction Study of the Effect of the Detergent Octyl Glucoside on the Structure of Lamellar and Nonlamellar Lipid/Water Phases of Use for Membrane Protein Reconstitution. Langmuir 15:8225–8234. doi:10.1021/la9902338

Carnie S, Israelachvili JN, Pailthorpe BA. 1979. Lipid packing and transbilayer asymmetries of mixed lipid vesicles. Biochim Biophys Acta 554:340–357. doi:10.1016/0005-2736(79)90375-4

Chung JK, Nocka LM, Decker A, Wang Q, Kadlecek TA, Weiss A, Kuriyan J, Groves JT. 2019. Switch-like activation of Bruton’s tyrosine kinase by membrane-mediated dimerization. Proc Natl Acad Sci 116:10798–10803. doi:10.1073/pnas.1819309116

Costigan SC, Booth PJ, Templer RH. 2000. Estimations of lipid bilayer geometry in fluid lamellar phases. Biochim Biophys Acta 1468:41–54. doi:10.1016/s0005-2736(00)00220-0

Dbouk HA, Pang H, Fiser A, Backer JM. 2010. A biochemical mechanism for the oncogenic potential of the p110 catalytic subunit of phosphoinositide 3-kinase. Proc Natl Acad Sci 107:19897–19902. doi:10.1073/pnas.1008739107

Hansen SD, Huang WYC, Lee YK, Bieling P, Christensen SM, Groves JT. 2019. Stochastic geometry sensing and polarization in a lipid kinase–phosphatase competitive reaction. Proc Natl Acad Sci 116:15013–15022. doi:10.1073/pnas.1901744116

Hon W-C, Berndt A, Williams RL. 2012. Regulation of lipid binding underlies the activation mechanism of class IA PI3-kinases. Oncogene 31:3655–3666. doi:10.1038/onc.2011.532

Israelachvili JN, Marcelja S, Horn RG. 1980. Physical principles of membrane organization. Q Rev Biophys 13:121–200. doi:10.1017/s0033583500001645

Israelachvili JN, Mitchell DJ, Ninham BW. 1976. Theory of self-assembly of hydrocarbon amphiphiles into micelles and bilayers. J Chem Soc Faraday Trans 2 Mol Chem Phys 72:1525–1568. doi:10.1039/F29767201525

Katada T, Kurosu H, Okada T, Suzuki T, Tsujimoto N, Takasuga S, Kontani K, Hazeki O, Ui M. 1999. Synergistic activation of a family of phosphoinositide 3-kinase via G-protein coupled and tyrosine kinase-related receptors. Chem Phys Lipids 98:79–86. doi:10.1016/S0009-3084(99)00020-1

Kobierski J, Wnętrzak A, Chachaj-Brekiesz A, Dynarowicz-Latka P. 2022. Predicting the packing parameter for lipids in monolayers with the use of molecular dynamics. Colloids Surf B Biointerfaces 211:112298. doi:10.1016/j.colsurfb.2021.112298

Kulkarni CV. 2019. Calculating the “chain splay” of amphiphilic molecules: Towards quantifying the molecular shapes. Chem Phys Lipids 218:16–21. doi:10.1016/j.chemphyslip.2018.11.004

Maier U, Babich A, Macrez N, Leopoldt D, Gierschik P, Illenberger D, Nürnberg B. 2000. Gβ 5 γ 2 Is a Highly Selective Activator of Phospholipid-dependent Enzymes. J Biol Chem 275:13746–13754. doi:10.1074/jbc.275.18.13746

Rathinaswamy MK, Dalwadi U, Fleming KD, Adams C, Stariha JTB, Pardon E, Baek M, Vadas O, DiMaio F, Steyaert J, Hansen SD, Yip CK, Burke JE. 2021. Structure of the phosphoinositide 3-kinase (PI3K) p110γ-p101 complex reveals molecular mechanism of GPCR activation. Sci Adv 7:eabj4282. doi:10.1126/sciadv.abj4282

Wnętrzak A, Lątka K, Dynarowicz-Łątka P. 2013. Interactions of alkylphosphocholines with model membranes-the Langmuir monolayer study. J Membr Biol 246:453–466. doi:10.1007/s00232-013-9557-4

Yang Y, Lee M, Fairn GD. 2018. Phospholipid subcellular localization and dynamics. J Biol Chem 293:6230–6240. doi:10.1074/jbc.R117.000582

Yasui M, Matsuoka S, Ueda M. 2014. PTEN Hopping on the Cell Membrane Is Regulated via a Positively-Charged C2 Domain. PLoS Comput Biol 10:e1003817. doi:10.1371/journal.pcbi.1003817

Ziemba BP, Burke JE, Masson G, Williams RL, Falke JJ. 2016. Regulation of PI3K by PKC and MARCKS: Single-Molecule Analysis of a Reconstituted Signaling Pathway. Biophys J 110:1811–1825. doi:10.1016/j.bpj.2016.03.001

Author Response

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

Point to point response for the editors

We are deeply grateful for the time you have devoted to reviewing this manuscript, and we sincerely thank you. Your insightful feedback has been instrumental in enhancing the quality of our work.

In the revised version of the manuscript, we have carefully addressed each of the concerns you raised. Below, you will find a detailed summary of how your feedback has been incorporated to improve the overall content and clarity of the document.

1. P2RX7 effects: In Figure 2, the vehicle treated P2RX7 knockout (panel M) shows an Ashcroft score of about 1.5 after BLM. Comparing this to the Ashcroft score of 3 after BLM in the wildtype (panel C) suggests that P2RX7 deletion is an effective way to reduce fibrosis by half!.

The argument that HEI3090 also reduces fibrosis by activating P2RX7 is of course very difficult to convey and it seems contradictory that P2RX7 deletion and P2RX7 activation can be both anti-fibrotic. This is an unusual claim and confuses the reviewers as well as the future readers.

This has many important health implications because activating an inflammatory pathway via P2RX7 and IL-18 could be risky in terms of a fibrosis treatment as inflammatory activation can also worsen fibrosis. The authors' own P2RX7 KO data (untreated vehicle groups) indeed confirms that P2RX7 can be pro-fibrotic.

We thank the editors for their comment highlighting the lack of clarity in our message. Indeed, we verified whether the antifibrotic action of HEI3090 depends on the expression of P2RX7 by inducing lung fibrosis in P2RX7 KO mice. In doing so, we initially observed that P2RX7 plays a role in the development of BLM-induced lung fibrosis. This is illustrated by a decrease of 50% in the Ashcroft score, as shown in Figure 2M and Supplemental Figure 2C of the revised manuscript.

To increase the clarity of your message, we added in the text the following paragraph:

"We further verified whether the antifibrotic action of HEI3090 depends on the expression of P2RX7 by inducing lung fibrosis in p2rx7 knockout (KO) mice. In doing so, we initially observed that P2RX7 plays a role in the development of BLM-induced lung fibrosis. This is illustrated by a decrease of 50% in the Ashcroft score, with a mean value of 1.7 in P2RX7 knockout mice compared to 3 in wild-type mice (Figure 2M and Supplemental Figure 2C). It is important to note that p2rx7 -/- mice still exhibit signs of lung fibrosis, such as thickening of the alveolar wall and a reduction in free air space, in comparison to naïve mice that received PBS instead of BLM (see Supplemental Figure 2A). This result confirms a previous report indicating that BLM-induced lung fibrosis partially depends on the activation of the P2RX7/pannexin-1 axis, leading to the production of IL-1β in the lung. Additionally, in contrast to the observations in WT mice, HEI3090 failed to attenuate the remaining lung fibrosis in p2rx7 -/- mice, as measured by the Ashcroft score (Figure 2M), the percentage of lung tissue with fibrotic lesions, or the intensity of collagen fibers (Supplemental Figure 2D). These results show that P2RX7 alone participates in fibrosis and that HEI3090 exerts a specific antifibrotic effect through this receptor (see Supplemental Figure 2C)."

Since we used the HEI3090 compound in this study and to be closer to the results, we have replaced the title of 2 chapters in the results section as followed:

“HEI3090 inhibits the onset of pulmonary fibrosis in the bleomycin mouse model” instead of P2RX7 activation inhibits the onset of pulmonary fibrosis in the bleomycin mouse model and “HEI3090 shapes immune cell infiltration in the lungs" instead of P2RX7 activation shapes immune cell infiltration in the lungs

We concur that the observation of both anti-fibrotic effects following P2RX7 deletion and P2RX7 activation appears contradictory. This specific aspect has been thoroughly addressed and extensively discussed in the revised manuscript.

“A major unmet need in the field of IPF is new treatment to fight this uncurable disease. In this preclinical study, we demonstrate the ability of immune cells to limit lung fibrosis progression. Based on the hypothesis that a local activation of a T cell immune response and upregulation of IFN-γ production has antifibrotic proprieties, we used the HEI3090 positive modulator of the purinergic receptor P2RX7, previously developed in our laboratory (Douguet et al., 2021), to demonstrate that activation of the P2RX7/IL-18 pathway attenuates lung fibrosis in the bleomycin mouse model. We have demonstrated that lung fibrosis progression is inhibited by HEI3090 in the fibrotic phase but also in the acute phase of the BLM fibrosis mouse model, i.e. during the period of inflammation. This lung fibrosis mouse model commonly employed in preclinical investigations, has recently been recognized as the optimal model for studying IPF (Jenkins et al., 2017). In this model, the intrapulmonary administration of BLM induces DNA damage in alveolar epithelial type 1 cells, triggering cellular demise and the release of ATP. The extracellular release of ATP from injured cells activates the P2RX7/pannexin 1 axis, initiating the maturation of IL1β and subsequent induction of inflammation and fibrosis. In line with this, mice lacking P2RX7 exhibited reduced neutrophil counts in their bronchoalveolar fluids and decreased levels of IL1β in their lungs compared to WT mice (Riteau et al., 2010). Based on these findings, Riteau and colleagues postulated that the inhibition of P2RX7 activity may offer a potential strategy for the therapeutic control of fibrosis in lung injury. In the present study we provided strong evidence showing that selective activation of P2RX7 on immune cells, through the use of HEI3090, can dampen inflammation and fibrosis by releasing IL-18. The efficacy of HEI3090 to inhibit lung fibrosis was evaluated histologically on the whole lung’s surface by evaluating the severity of fibrosis using three independent approaches applied to the whole lung, the Ashcroft score, quantification of fibroblasts/myofibroblasts (CD140a) and polarized-light microscopy of Sirius Red staining to quantify collagen fibers. All these methods of fibrosis assessment revealed that HEI3090 exerts an inhibitory effect on lung fibrosis, underscoring the necessity for a thorough pre-clinical assessment of HEI3090's mode of action. Notably, HEI3090 functions as an activator, rather than an inhibitor, of P2RX7, further emphasizing the importance of elucidating its intricate mechanisms.”

We trust that the detailed explanation provided therein will adequately persuade both the reviewers and future readers.

1. The statistical concerns are based on the phrasing of "the experiment was stopped when significantly statistical results were observed". This is different from the power analysis approach that the authors describe in their latest rebuttal. However, it raises the question why the power analysis was performed using "on a one-way ANOVA analysis comparing in each experiment the vehicle and the treated group". The analyses in the manuscript use the Mann-Whitney test for several comparisons which ahs the assumption that the samples do NOT have a normal distribution. An ANOVA and t-tests have the assumption that samples are normally distributed. If the power analysis and "statistical forecasting" assumed a normal distribution and used an ANOVA, then shouldn't all the analyses also use a statistical test appropriate for normally distributed samples such as ANOVA and t-tests?

Several of the data points in the figures seem to be normally distributed and therefore t-test for two group comparisons would be more appropriate. The most rigorous approach would be to check for normal distribution before choosing the correct statistical test and using the t-test/ANOVA in normally distributed data as well as Mann-Whitney for non-normally distributed data.

We described in the Material and Method section of the revised manuscript our approach to determine the size of experimental group.

“The determination of experimental group sizes involved conducting a pilot experiment with four mice in each group. Subsequently, a power analysis, based on the pilot experiment's findings (which revealed a 40% difference with a standard error of 0.9, α risk of 0.05, and power of 0.8), was performed to ascertain the appropriate group size for studying the effects of HEI3090 on BLM-induced lung fibrosis. The results of the pilot experiment and power analysis indicated that a group size of four mice was sufficient to characterize the observed effects. For each full-scale experiment, we initiated the study with 6 to 8 mice per group, ensuring a minimum of 5 mice in each group for robust statistical analysis. Additionally, we systematically employed the ROULT method to identify and subsequently exclude any outliers present in each experiment before conducting statistical analyses”.

We now described in the Material and Method section how we carried out the statistical analyses.

“Quantitative data were described and presented graphically as medians and interquartiles or means and standard deviations. The distribution normality was tested with the Shapiro's test and homoscedasticity with a Bartlett's test. For two categories, statistical comparisons were performed using the Student's t-test or the Mann–Whitney's test. For three and more categories, analysis of variance (ANOVA) or non-parametric data with Kruskal–Wallis was performed to test variables expressed as categories versus continuous variables. If this test was significant, we used the Tukey's test to compare these categories and the Bonferroni’s test to adjust the significant threshold. For the Gene Set Enrichment Analyses (GSEA), bilateral Kolmogorov–Smirnov test, and false discovery rate (FDR) were used. All statistical analyses were performed by biostatistician using Prism8 program from GraphPad software. Tests of significance was two-tailed and considered significant with an alpha level of P < 0.05. (graphically: * for P < 0.05, ** for P < 0.01, *** for P < 0.001).”

We also added in the legend of each figure, the statistical analysis used to determine each p-values.

1. Adoptive transfer: The concerns of the reviewers include an unclear analysis of the effects of adoptive transfer itself and the approaches used to analyze the data independent of the HEI3090 effect. For example, in Figure 4, the adoptive transfer IL18-/- cells (vehicle group) leads to an Ashcroft score of about 1 and among the lowest of the BLM exposed mice. Does that mean that IL18 is pro-fibrotic and that its absence is beneficial? If yes, it would go against the core premise of the study that IL18 is beneficial. Statistical comparisons of the all the vehicle conditions in the adoptive transfer would help clarify whether adoptive transfer of NLRP3-/-, IL18-/- in wild-type and P2RX7-/- mice reduces or increases fibrosis. Such multiple comparisons are necessary to fully understand the adoptive transfer studies and would also require the appropriate statistical test with corrections for multiple comparisons such as Kruskal-Wallis for data without normal distribution and ANOVA with post hoc correction for normal distribution.

We added a new paragraph in the revised version of the manuscript to explain the adoptive transfer approach.

“We wanted to further investigate the mechanism of action of HEI3090 by identifying the cellular compartment and signaling pathway required for its activity. Since the expression of P2RX7 and the P2RX7-dependent release of IL-18 are mostly associated with immune cells (Ferrari et al., 2006), and since HEI3090 shapes the lung immune landscape (Figure 3), we investigated whether immune cells were required for the antifibrotic effect of HEI3090. To do so, we conducted adoptive transfer experiments wherein immune cells from a donor mouse were intravenously injected one day before BLM administration into an acceptor mouse. The intravenous injection route was chosen as it is a standard method for targeting the lungs, as previously documented (Wei and Zhao, 2014). This approach was previously used with success in our laboratory (Douguet et al., 2021). It is noteworthy that this adoptive transfer approach did not influence the response to HEI3090. This was observed consistently in both p2rx7 -/- mice and p2rx7 -/- mice that received splenocytes of the same genetic background. In both cases, HEI3090 failed to mitigate lung fibrosis, as depicted in Figure 2M and Supplemental Figures 2D and 6A and B.”

We added the Supplemental Figure 7 showing that the genetic background does not impact lung fibrosis at steady step levels where p-values were analyzed by one-way ANOVA, with Kruskal-Wallis test for multiple comparisons.

Author response image 1.

Supplemental Figure 7 : The genetic background does not impact lung fibrosis at steady step levels. p2rx7-/- mice were given 3.106 WT, nlrp3-/ , i118-/ or illb -l- splenocytes i_v_ one day prior to BLM delivery (i_n_ 2.5 LJ/kg) p2rx7-/- mice or p2rx7-/- mice adoptively transferred with splenocytes from indicated genetic background were treated daily i.p with mg/kg HE13090 or vehicle for 14 days. Fibrosis score assessed by the Ashcroft method. P-values were analyzed on all treated and non treated groups by one-way ANOVA, with Kruskal-Wallis test for multiple comparisons. The violin plot illustrates the distribution of Ashcroft scores across indicated experimental groups. The width of the violin at each point represents the density of data, and the central line indicates the median expression level. Each point represents one biological replicate. ns, not significant

Author Response

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

Reviewer #1:

We thank the referee for the positive review.

Reviewer #2 (Public review):

We thank the referee for his/her constructive comments

1. The weakness of this work is the lack of clarification on the function of eIF2A in general. The novelty of this study was limited.

We believe our study is valuable in providing strong evidence that eIF2A does not functionally substitute for eIF2 in tRNAi recruitment even when eIF2 function is impaired, and in showing that it does not contribute to translational control by uORFs or IRESs, thus ruling out the most likely possibilities for its function in yeast based on studies of the mammalian factor. We agree that the function of yeast eIF2A remains to be identified; however, we think this should be regarded as a limitation rather than a weakness in experimental design or data obtained in the current study.

1. Related to this, it would be worth investigating common features in mRNAs selectively regulated (surveyed in Figure 3A).

We did not embark on this because only 17 of the 32 transcripts showing TE reductions in Fig. 3A showed a pattern of TE changes consistent with a conditional requirement for eIF2A under conditions of reduced eIF2 function, exhibiting greater TE decreases when both eIF2 function was impaired by phosphorylation and eIF2A was eliminated from cells. Moreover, we could validate this conditional eIF2A dependence by LUC reporter for only a single mRNA, HKR1.

Also, it would be worth analyzing the effect of eIF2A deletion on elongation (ribosome occupancy on each codon and/or global ribosome footprint distribution along CDS) and termination/recycling (footprint reads on stop codon and on 3′ UTR).

We have analyzed the effects of deleting eIF2A on ribosome pausing at individual codons by calculating tri-peptide pause scores from our ribosome profiling data. The results shown in new Fig. 7 reveal that eIF2A plays no discernible role in stimulating the rate of decoding of any three-codon combinations.

1. Regarding Figure 3D, the reporters were designed to include promoter and 5′ UTR of the target genes. Thus, it should be worth noting that reporter design was based on the assumption that eIF2A-dependency in translation regulation was not dependent on 3′ UTR or CDS region. The reason why the effects on ribosome profiling-supported mRNAs could not be recapitulated in reporter assay may originate from this design. This should be also discussed.

We agree and included this stipulation in the DISCUSSION, while at the same time noting that the native mRNAs were examined in the orthogonal assay of polysome distributions.

1. Related to the point above, the authors claimed that eIF2A affects "possibly only one" (HKR1) mRNA. However, this was due to the reporter assay which is technically variable and could not allow some of the constructs to pass the authors' threshold. Alternative wording for this point should be considered.

We agree and revised text in the DISCUSSION to read: “A possible limitation of our LUC reporter analysis in Fig. 3D was the lack of 3’UTR sequences of the cognate transcripts, which might be required to observe eIF2A dependence. Given that native mRNAs were examined in the orthogonal assay of polysome profiling in Fig. 3E, the positive results obtained there for SAG1 and SVL3 in addition to HKR1 should be given greater weight. Nevertheless, our findings indicate a very limited role of yeast eIF2A in providing a back-up mechanism for Met-tRNAi recruitment when eIF2 function is diminished by phosphorylation of its α-subunit.”

1. For Figure 3D, it would be worth considering testing the #-marked genes (in Figure 3C) in this set up.

Actually, we did test 10 of the 17 mRNAs marked with “#”s in the reporter assays of Fig. 3C, which had been noted in the Fig. 3C legend.

1. In box plots, the authors should provide the statistical tests, at least where the authors explained in the main text.

At the first occurrence of a notched box plot (Fig. 2D), we explained in the main text that in all such plots, when the notches of different boxes do not overlap, their median values differ significantly with a 95% confidence level. In cases where overlaps between notches is difficult to assess by eye, we added the results of Mann-Whitney U tests with the p values indicated by asterisks, as explained in the legends. We added results of additional Mann-Whitney U tests to such box plots in Figs. 3B, 6A-C, and 6-supp. 1E & G and mentioned this in the corresponding legends.

Reviewer #2 (Recommendations For The Authors):

The first section of "Yeast eIF2A does not play a prominent role as a functional substitute for eIF2 in the presence or absence of amino acid starvation" can be subdivided into a couple of sections for better readability.

Done.

Although the authors have used SM to induce ISR in yeasts previously, the validation of eIF2alpha phosphorylation in Western blot would be helpful for readers. Also, it should be worth testing whether eIF2alpha phosphorylation was properly induced in eIF2A KO cells.

The translational induction of GCN4 mRNA, which we have documented in WT and eIF2A∆ cells, provides a quantitative read-out of eIF2 functional attenuation superior to determining the proportion of eIF2α that is phosphorylated.

For Figure 2B, the Venn diagram that shows the overlap between TE-changes genes in WT_SM/WT and those in eIF2A∆_SM/eIF2A∆ would be helpful (although a list was provided by the source data).

The Venn diagram has been provided in a new figure, Figure 2-figure supplement 1B.

For Figures 1C and 5A-B, the depiction of the positions of uORFs within the orange gene region would be helpful for readers.

Done.

For Figure 4A-C, the depiction of the IRES regions (if known) within the orange gene region would be helpful for readers.

Done for the URE2 IRES, whose location is known.

For Figures 1C, 4A-C, and 5A-B, the y-axis should have a label/scale.

Added.

For Figure 3C, the definition of #-marked genes should be concretely described (e.g., value range) in the legend.

Added.

For Figure 3D-E, the statistical test has been only shown in a couple of data. A full depiction of the statistical results for all the data sets may be helpful for readers.

We explained that when notches in box plots do not overlap, their medians differ with 95% confidence. In cases where overlaps were difficult to discern, we added p values from Mann-Whitney U tests to the relevant box plots.

For Figure 3E, it would be helpful if the authors could show the UV spectrum of the sucrose density gradient to show the regions isolated for the experiments.

Added for a representative replicate gradient in the new figure, Figure 3-figure supplement 1.

Reviewer #3 (Public Review):

We thank the referee for his/her positive assessment of our study.

Weaknesses:

While no role of eIF2A in translation initiation is apparent, the authors do not determine what function eIF2A does play in yeast. Whether it plays a role in regulating translation in a different stress response is not determined.

We agree that there are many additional possibilities to consider for functions of eIF2A in translation initiation, including different stress situations or mutant backgrounds; however, we regard this as a limitation rather than a weakness in the experimental design and data obtained in the current study in which we examined the most likely possibilities for eIF2A function in yeast based on studies of the mammalian factor.

Reviewer #3 (Recommendations For The Authors):

Curiously, the authors indicate that they could not replicate published results for eIF2A's repressor function for URE2, PAB1, or GIC1 translation. This is a little concerning and one wonders if the yeast strain used in the previous study is different in some way from the authors' strain. Did the authors obtain that strain to test it in their assays?

The same WT and eIF2A∆ strains have been analyzed here and in the two cited studies on yeast IRESs.

The authors do discuss the fact that eIF2A may function to regulate translation in response to different stresses. It would have been a strength to test an alternative stress in the current study. However, I also appreciate that this could be the subject of a future study.

Agreed.

One minor question I have is whether the yeast strains used possess L-A dsRNA virus? While it may not be that this virus would necessarily mask a role of eIF2A-dependent translation, do the authors have any specific thoughts on this? Would different results be obtained if cured strains were used?

According to Ravoityte et al. (doi: 10.3390/jof8040381), the S. cerevisiae strain we employed, BY4741, harbors L-A-1 dsRNA; however, we have not explored whether curing the virus would alter the consequences of eliminating eIF2A.

Author Response

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

Response to reviewers

We thank the two reviewers for their constructive criticism, which helped to significantly improve our manuscript.

During the revision process, we had to realize that the localization pattern reported for H. neptunium LmdCN-mCherry was an artifact caused by bleed-through of the BacA-YFP signal in the mCherry channel. More detailed studies showed that the fusion protein was detectable by Western blot analysis but, for unknown reasons, did not produce any fluorescence signal. Therefore, we have now removed the localization data shown in previous Figure 8B,C and Figure 8—figure supplement 1.

To provide more evidence for a functional interaction between BacA and LmdC in H. neptunium, we have now established an inducible CRISPR interference system for this species and used it successfully to deplete LmdC (new Figure 9A-F). The loss of LmdC causes morphological defects very similar to those observed for the ΔbacA(D) mutant. In line with the physical interaction of BacA with the cytoplasmic region of LmdC observed in vitro, these findings support the hypothesis that the two proteins act in the same pathway. Consistent with the results obtained in H. neptunium, the absence of BacA leads to the delocalization of LmdC in R. rubrum. Moreover, we now provide in vivo evidence for a critical role of the cytoplasmic region of LmdC in the interaction of this protein with BacA in R. rubrum cells (new Figure 11). Together, these new findings strongly support the model that BacA and LmdC form a conserved morphogenetic module involved in the establishment of complex cell shapes in bacteria.

Please see below for a more detailed explanation of our new results and for our response to the issues raised in the first round of review.

Reviewer #1 (Public Review)

In their study, Osorio-Valeriano and colleagues seek to understand how bacterial-specific polymerizing proteins called bactofilins contribute to morphogenesis. They do this primarily in the stalked budding bacterium Hyphomonas neptunium, with supporting work in a spiral-shaped bacterium, Rhodospirillum rubrum. Overall the study incorporates bacterial genetics and physiology, imaging, and biochemistry to explore the function of bactofilins and cell wall hydrolases that are frequently encoded together within an operon. They demonstrate an important, but not essential, function for BacA in morphogenesis of H. neptunium. Using biochemistry and imaging, they show that BacA can polymerize and that its localization in cells is dynamic and cell-cycle regulated. The authors then focus on lmdC, which encodes a putative M23 endopeptidase upstream of bacA in H. neptunium, and find that is essential for viability. The purified LmdC C-terminal domain could cleave E. coli peptidoglycan in vitro suggesting that it is a DD-endopeptidase. LmdC interacts directly with BacA in vitro and co-localizes with BacA in cells. To expand their observations, the authors then explore a related endopeptidase/ bactofilin pair in R. rubrum; those observations support a function for LmdC and BacA in R. rubrum morphogenesis as well.

An overall strength of this study is the breadth and completeness of approaches used to assess bactofilin and endopeptidase function in cells and in vitro. The authors establish a clear function for BacA in morphogenesis in two bacterial systems, and demonstrate a physical relationship between BacA and the cell wall hydrolase LmdC that may be broadly conserved. The eventual model the authors favor for BacA regulation of morphogenesis in H. neptunium is that it serves as a diffusion barrier and limits movement of morphogenetic machinery like the elongasome into the elongating stalk and/or bud. However, there is no data presented here to address that model and the role of LmdC in H. neptunium morphogenesis remains unclear.

We hypothesize that BacA establishes a barrier that prevents the movement of elongasome complexes into the stalk, either directly by sterical hindrance and/or indirectly by promoting the formation of an annular region of high positive inner cell curvature that cannot be passed by the elongasome. To test this model, we have now analyzed the localization dynamics of RodZ, a core structural component of the elongasome complex, in wild-type and ΔbacAD cells. We found that wild-type cells show dynamic YFP-RodZ foci whose movement is limited to the mother cell and the nascent bud, with no signal ob-served in the stalk. In ΔbacAD cells, by contrast, the fusion protein is consistently detected in all regions of the cell, including nascent stalks (new Figure 5). These results support the idea that BacA is required to confine the elongasome to the mother cell and bud regions and, thus, set the limits of the different growth zones in H. neptunium. We also attempted to follow the localization dynamics of other elongasome components, such as PBP2, MreC and MreD, but none of the corresponding fluorescent protein fusions was functional.

In the past, we tried intensively to generate conditional mutants of lmdC, but all attempts to place the expression of this gene under the control of the copper- or zinc-inducible promoters available for H. neptunium were unsuccessful. To clarify the role of LmdC in H. neptunium morphogenesis, we have now established an inducible CRISPR interference system for this species and managed to block the ex-pression of lmdC using an sgRNA directed against the 5' region of its non-coding strand. We observed that cells lacking LmdC show a phenotype very similar to that of the ΔbacA mutant. Together with the finding that the N-terminal cytoplasmic region of LmdC physically interacts with BacA, this result strongly supports the hypothesis that BacA and LmdC act in the same pathway, forming a complex that ensures proper morphogenesis in H. neptunium (new Figure 9).

The data presented illuminate aspects of bacterial morphogenesis and the physical and functional relationship between polymerizing proteins and cell wall enzymes in bacteria, a recurring theme in bacterial cell biology with a variety of underlying mechanisms. Bactofilins in particular are relatively recently discovered and any new insights into their functions and mechanisms of action are valuable. The findings presented here are likely to interest those studying bacterial morphogenesis, peptido-glycan, and cytoskeletal function.

Reviewer #2 (Public Review):

This is an excellent study. It starts with the identification of two bactofilins in H. neptunium, a demonstration of their important role for the determination of cell shape and discovery of an associated endopeptidase to provide a convincing model for how these two classes of proteins interact to control cell shape. This model is backed up by a quantitative characterisation of their properties using high-resolution imaging and image analysis methods.

Overall, all evidence is very convincing and I do not have many recommendations on how to improve the manuscript.

In my opinion, there are only two issues that I have with the paper:

1. The single particle dynamics of BacA is presented as analysed and I would like to give some suggestions how to maybe extract even more information from the already acquired data:

1.1. Presentation: Figure 5A is only showing projections of single particle time-lapse movies. To convince the reader that it was indeed possible to detect single molecules it would be helpful if the authors present individual snapshots and intensity traces. In case of single molecules these will show step wise bleaching.

We have now added a supplementary video that shows both time series and intensity traces of individual BacA-YFP molecules (Figure 6—Video 1). It verifies the step-wise bleaching of the particles observed and thus shows that we observe the mobility of single molecules. Moreover, we have now included a supplementary figure that shows all trajectories identified within representative cells. This visualization provides a more comprehensive view of our data and further supports the notion that our analysis is based on the detection of single molecules.

1.2. Analysis: Figure 5B and Supplement Figure 1 are showing the single particle tracking results, revealing that there are two populations of BacA-YFP in the cell. However, this data does not show if individual BacA particles transition between these two populations or not. A more detailed analysis of the existing data, where one can try to identify confinement events in single particle trajectories could be very revealing and help to understand the behaviour of BacA in more detail.

We agree that an analysis of the single-molecule traces for transitions between the mobile and static states would help to achieve a more detailed understanding of the polymerization behavior of BacA. We believe that the dynamic formation, reorganization and disappearance of BacA-YFP foci observed by time-lapse analysis (Figure 4) indicates that BacA undergoes reversible polymerization in vivo. A deeper investigation of this aspect is beyond the scope of the present study and will be performed at a later point.

1. The title of Fig. 3 says that BacA and BacD copolymerise, however, the data presented to confirm this conclusion is actually rather weak. First, the Alphafold prediction does not show the co-polymer, and second, the in vitro polymerisation experiments were only done with BacA in the absence of BacD. Accordingly, the only evidence that supports this is their colocalization in fluorescence microscopy. I suggest either weakening the statement or changing the title adds more evidence.

To support the idea that BacA and BacD interact with each other, we have now added images of cells producing BacA-YFP or BacD-CFP individually (new Figure 3—figure supplement 1B,C). The results obtained show that Bac-YFP alone still forms filamentous structures, whereas BacD-CFP condenses into tight foci in the absence of its paralog. However, when produced together with BacA-YFP, the two proteins colocalize into filamentous structures, supporting the notion that they interact with each other. However, we agree that it is unclear whether BacA and BacD copolymerize into mixed protofilaments or whether they form distinct protofilaments that then interact laterally to form larger bundles. We have therefore replaced the term “co-polymerize” with “assemble” in the heading of this section.

Finally, did the authors think about biochemical experiments to study the interaction between the cytoplasmic part of LmdC and the bactofilins? These could further support their model.

We show the interaction between the cytoplasmic region of H. neptunium LmdC and BacA in Figure 9G,H (previously Figure 8D,E). For technical reasons, it was not possible to synthesize a peptide com-prising the corresponding region of R. rubrum LmdC, so that our in vitro analysis is limited to the H. neptunium proteins.

To further support the notion that BacA interacts with the cytoplasmic region of LmdC, we have now analyzed the localization behavior of two LmdC variants with amino acid exchanges in the conserved cytoplasmic β-hairpin motif (new Figure 11). Both variants no longer colocalize with BacA and are no longer enriched at the inner cell curve. Interestingly, these exchanges also affect the enrichment of BacA at the inner cell curvature, suggesting that BacA needs to interact with LmdC for proper localization. It is tempting to speculate that BacA polymers have a preferred intrinsic curvature and that the activity of the BacA-LmdC complexes adjusts cell curvature in a manner that facilitates their association with the inner curve.

Reviewer #1 (Recommendations for The Authors):

We have the following specific recommendations for the improvement of the manuscript:

1. Several places would benefit from additional quantitation of data:

a. Figure 1 and supplements: can cell shape be quantified in a more specific way? (e.g. principle component analysis of shape as in https://onlinelibrary.wiley.com/doi/10.1111/mmi.13218). It looks as if BacD production may partially rescue the bacA shape phenotype?

We have made considerable efforts to establish methods to quantify morphological changes and protein localization patterns in Hyphomonas neptunium. Since standard software packages, such as Oufti or MicrobeJ, are not able to reliably detect stalks and, thus, typically identify buds as separate cells, we have developed our own analysis software (BacStalk; Hartmann et al, 2020, Mol Microbiol), that is optimized for the detection of thin cellular extensions. However, while this software works very well with wild-type cells, it also fails to recognize amorphous cells with multiple, ill-defined extensions. Given these problems in cell segmentation, it is currently not possible to use principle component analysis to obtain a robust measure of the morphological defects of bactofilin mutants in H. neptunium.

b. Figures 2-S2b, 7D and 9-S1b - can the area under the peaks be quantified and compared across strains? Visual examination of the spectra makes it difficult to discern differences.

A direct comparison of the peak areas between strains is not possible, because the absolute values depend on the amount of peptidoglycan used in the muropeptide analyses. It is very difficult to precisely quantify peptidoglycan, which makes it challenging to use equal amounts of material from different strains in the reactions. However, the relative proportion of different muropeptide species, as provided in Figure 2—Dataset 1, faithfully reflects the composition of peptidoglycan and can easily compared between strains.

c. Figure 9E,F, 9-S4d - BacA and LmdC localization in R. rubrum is very difficult to assess. It does not look linear/filamentous in most cells and is difficult to tell if it is associated with the inner curvature. Can you quantify the position of the signal along the short axis of the cell to better demonstrate that?

We agree that a better quantification of the distribution of protein along the cell envelope of R. rubrum is required to support the conclusions drawn. To address this issue, we have now used line scans to measure the fluorescence intensities along the inner and outer curve of cells (n=200 per strain) and visualized the data in the form of demographs. The results clearly show an enrichment of BacA and LmdC at the inner curve in wild-type cells and a disruption of this pattern in various mutant backgrounds (new Figures 10F,G,J and 11D,E).

1. Figure 2-S2A. Does ∆bacD grow better than wild-type? It would also be useful to add growth curves of the bacA complemented strains.

In the case of H. neptunium growth curves are often misleading, because cells start to aggregate at the late exponential phase due to abundant EPS formation. The degree of cell aggregation also depends on the morphology of cells, because EPS production is limited to the mother cell body, which makes it challenging to compare morphologically distinct mutant strains. We have now performed growth assays for all H. neptunium deletion and complementation strains used in the study and limited the analysis of doubling times to the early and mid-exponential phase, in which cells do not yet form visible aggregates. The results obtained are now included in the new Figure 1F and Figure 1—figure supplement 2D. They show that the doubling times of the different bactofilin mutants are close to that of the wild-type strain.

1. Figure 4BC: From the demographs provided, BacA and BacD appear to have different localization dynamics. BacD seems to stay at the base of the stalk, nearest the mother cell, whereas BacA migrates towards to bud? Also, "length" is misspelt in the panels.

During the transition to bud formation, we indeed observe that the localization patterns of BacA and BacD are in many cases not fully superimposable, with BacD lagging behind BacA and forming transient additional clusters in the vicinity of the stalk base. Examples are now shown in Figure 4—figure supplement 4). This effect explains the distinct patterns in the demographs. We have now modified the text accordingly. We have also corrected the spelling of “length” in the figure.

1. Can BacD polymerize on its own? It colocalizes with BacA in E. coli but that does not necessarily mean it co-polymerizes.

Please see our response to a similar issue (point 2) raised by Reviewer #1.

1. Lines 263-266. You use E. coli PG as a substrate for LmdC in vitro because "peptidoglycan from H. neptunium shows only a low degree of cross-linkage and hardly any pentapeptides." Does this not have relevance to the physiological significance of the observed activity? Or do you presume that LmdC activity (and/or that of other endopeptidases) is very high in H. neptunium so it is difficult to detect additional activity using HnPG as a substrate? It would be useful to clarify this logic in the text.

DD-crosslinks are formed by all major peptidoglycan biosynthetic complexes, including the elongasome and the divisome, so that their general relevance to cell growth in H. neptunium is beyond doubt. The low degree of crosslinkage observed suggests that H. neptunium contains high endopeptidase activity, which cleaves crosslinks after their formation by DD-transpeptidases. We have now added the explanation “likely due to a high level of autolytic activity” to make this point clearer. Whether LmdC makes a major contribution to the low level of crosslinkage remains to be determined. However, our data suggest that it mostly acts in complex with BacA, so that it may only cleave peptidoglycan locally and not have a global effect global on cell wall composition. It would not possible to detect the DD-endopeptidase activity of LmdC using H. neptunium peptidoglycan as a substrate, because it has a low content of DD-linked peptide chains. To facilitate the in vitro activity assay, we therefore used highly crosslinked peptidoglycan from a mutant E. coli strain.

1. Lines 268-269: Is there some explanation for why monomers do not increase on LmdC treatment? Here quantitation of peaks before and after treatment would allow the reader to more precisely interpret these data.

The absolute peak sizes are not comparable, because there is some variation in the amount of peptido-glycan included in the assays (see also our comments on point 1b raised by Reviewer #1) and the integrated peak areas (which correspond to the amounts of muropeptide species produced) depend on both the height and the width of the peaks, which vary to some degree in different HPLC runs. The relevant measure to compare the muropeptide profiles is therefore the relative content of different muropeptide species in the different conditions. For clarification, we have now added the following sentence to the legend of Figure 8D: “A quantification of the relative abundance of different muropeptide species in each condition, based on a comparison of the relative integrated peak areas, is provided in Figure 8—Dataset 1.” The control reaction lacking LmdC only contains peptidoglycan diluted in buffer and thus provides insight into muropeptide composition of untreated peptidoglycan.

1. Lines 280-283: It would be interesting to know if the transmembrane domain of LmdC is required for its localization since it is dispensable for binding BacA and since LmdC still localizes to foci without BacA.

Given that it is currently not possible to localize LmdC in H. neptunium, we were not able to perform this analysis.

1. Line 296: it is also possible that LmdC localizes with another protein and does not independently assemble into larger complexes.

Since the localization pattern reported for LmdC in the ΔbacAD background is no longer valid, we have not discussed this aspect in the revised version of our manuscript. However, in general, we do not exclude the possibility that LmdC could interact with other peptidoglycan biosynthetic proteins.

1. Line 304-306 and Fig 9: Is the domain organization of RrLmdC the same as for HnLmdC? It would be useful to include its domain organization as well. Also, please add amino acid numbering to Figure 9B.

We have now added a schematic showing the domain organization of LmdC from R. rubrum (new Figure 10B). The protein is highly similar to its homolog from H. neptunium.

1. Line 340-341: "In both cases, they functionally interact with LmdC-type DD-endopeptidases to promote local changes in the pattern of peptidoglycan biosynthesis." This conclusion is not experimentally supported. Since LmdC is essential and you could not make a depletion strain in H. neptunium, it was not shown that the interaction with LmdC is how BacA promotes changes in PG patterning. HADA/FDAA labeling was not performed in R. rubrum, and no global changes in PG chemistry were observed in bacA or lmdC mutants, so you cannot claim BacA or LmdC influences PG patterning there, either. Either soften this statement to a hypothesis or otherwise rephrase.

To further corroborate a functional interaction between BacA and LmdC, we have now established an inducible CRISPRi system to deplete LmdC from H. neptunium cells (see also our comments on the public review of Reviewer #1). We observe that the loss of LmdC leads to a phenotype very similar to that observed for the ΔbacA(D) mutant, supporting the idea that BacA and LmdC act in the same path-way. We have now also performed localization studies of the elongasome component RodZ in H. nep-tunium, which demonstrate that the spatial distribution of elongasome complexes is affected in the absence of the bactofilin cytoskeleton in H. neptunium. Combined with the observation that LmdC is a catalytically active DD-endopeptidase and its absence leads to morphological defects, these results indicate that BacA, together with LmdC, induces local changes in pattern of peptidoglycan biosynthesis, both by affecting elongasome movement and, likely, by reducing peptidoglycan crosslinking in the cell envelope regions it occupies.

1. Figure 9-S4: there is no panel C (change D to C).

Corrected.

1. Lines 344-355: No data is presented here to support the barrier model of bactofilin function. In addition, it is unclear why cells would take on amorphous shapes instead of extended rod shapes/filaments if elongasome function was not constrained on the longitudinal axis. It would be helpful to have more discussion of the potential mechanisms of LmdC function in H. neptunium in this section of the discussion since that is the emphasis of the results section.

To support the barrier model, we have now compared the localization dynamics of the elongasome component RodZ in wild-type and ΔbacAD cells. The results show that RodZ is excluded from the stalk in the wild-type background, whereas it readily enters the stalk in the mutant cells, leading to the expansion of stalks into large, amorphous extensions. Consistent with these findings, HADA labeling is not observed within the stalks in wild-type cells, whereas it is readily observed in the enlarged stalk structures (pseudohyphae) formed in the mutant cells.

The current model of MreB movement suggests that MreB filaments have an intrinsic curvature and thus preferentially align along regions of similar curvature, which is along the circumference of the cell in rod-shaped geometries. However, previous work has shown that MreB starts to move along randomly oriented trajectories as soon as cells lose their rod-shaped morphology and adopt more spherical shapes (Hussain et al, 2018, eLife). In line with these findings, our current and our previous work (Cserti et al, 2017, Mol Microbiol) indicate that the expansion of the ovoid H. neptunium mother cell prior to the onset of stalk biosynthesis as well as bud formation are mediated by the elongasome complex. Thus, the elongasome can clearly also give rise to shapes other than rods. Interestingly, however, the H. neptunium elongasome also appears to drive the formation of the rod-shaped stalk, possibly by moving around the circumference of the stalk base. Thus, species- or growth phase-dependent regulatory mechanisms or, potentially, differences in the spatial arrangement of the glycan strands within the peptido-glycan layer may result in different modes of elongasome movement and, thus, modulate the morphogenetic activity of elongasome complexes.

1. Lines 395-397: It is also possible that LmdC positioning is dependent on cell morphology, rather than directly on BacA, since morphology is so distorted in bacA mutant cells.

We provide several lines of evidence showing that LmdC and BacA functionally and physically interact (see above), making it highly unlikely that the two proteins are not associated with each other. How-ever, our previous (Figure 10I,J) and new (Figure 11) results suggest that the physical interaction with LmdC and/or or the cell shape-modulating activity of the complex are required for the proper localization of BacA at the inner curve of the cell. This finding may indicate the existence of a self-reinforcing cycle, in which the morphological changes induced by BacA-LmdC assemblies stimulate the recruitment of additional assemblies to their site of action.

Author Response

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

eLife assessment

This study presents useful findings regarding the impact of forest cover and fragmentation on the prevalence of malaria in non-human primates. The evidence supporting the claims of the authors is, however, incomplete, as the sampling design cannot adequately address the geospatial issues that this study focuses on.

Public Reviews:

Reviewer #1 (Public Review):

The study as a concept is well designed, although there is still one issue I see in the methodology.

I still have concerns with their attempts to combine the different scales of data. While the use of point data is great, it limits the sample size, and they have included the district to country level data to try and increase the sample size. The problem is that although they try to get an overall estimate at the district/state/country by taking 10 random sample points, which could be a method to get an estimate for the district/state/country. It would be a suitable method if the primates were evenly distributed across the district/state/country. The reality is that the primates are not evenly distributed across the district/state/country therefore the random point sampling is not a reasonable method to get an estimate of the environmental variables in relation to the macaques. For example if you had a mountainous country and you took 10 random points to estimate altitude, you would end up with a large number, but if all the animals of interest lived on the coast, your average altitude is meaningless in relation to the animals of interest as they are all living at low altitude. The fact that the model relies less on highly variable components and places more reliance on less variable components, is really not relevant as the district/state/country measurements have no real meaning in relation to the distribution of masques.

A simple possible way forward could be to run the model without the district/state/country samples and see what the outcome is. If the outcome is similar then the random point method may be viable (but if it gives the same outcome as ignoring those samples then you don't need the district/state/country samples). If you get a totally different outcome then it should raise concerns about using the district/state/country samples.

This paper is a really nice piece of work and is a valuable contribution but the district/state/country sample issue really needs to be addressed.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

A simple possible way forward could be to run the model without the district/state/country samples and see what the outcome is. If the outcome is similar then the random point method may be viable (but if it gives the same outcome as ignoring those samples then you don't need the district/state/country samples). If you get a totally different outcome then it should raise concerns about using the district/state/country samples.

Thank you for your comments, and for the suggestions to address the issues identified in your main commentary by running an analysis on exclusively GPS geolocated data points. This was the original plan for analysis, but the available data identified in the literature review includes only 14 data points (macaque P. knowlesi prevalence surveys) with associated GPS coordinates. This was found to be too limited to obtain meaningful results from a regression analysis, and hence we then explored methods for utilising all available data to identify trends whilst accounting for spatial uncertainty in the analysis. As the point location only represents the location of capture and not the extent of the home range of the NHPs, we additionally feel there is value in exploring methods to encompass the wider surrounding habitat.

We do appreciate the concerns you raise with the random point method being used to represent macaque survey sites when species of interest are not necessarily evenly distributed across an area. To investigate this, we ran sensitivity analysis on a subset of the dataset according to whether the points fall in areas of >50%, >75% or >90% predicted probability of macaque occurrence, with maps derived from published models of macaque suitability in Southeast Asia. For each of these thresholds, points that fall outside these areas were removed – such that, if a random point is located on a mountain range where there is 0 likelihood of macaque occurrence, it is excluded from the analysis. We found that restricting analysis to areas with highly probably macaque habitat still shows a robust effect of forest cover on NHP prevalence, and additionally that for the most conservative (>90%) habitat threshold there remains an effect of forest fragmentation on prevalence (SI Table S17c, Figure S15c). Given that using the full data set increases the uncertainty, as there is more variation in covariates between the replicates, this can be considered a more conservative approach to detecting an effect of environment as reported in the main findings.

Author Response

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

Reviewer #1 (Recommendations For The Authors):

1. A more thorough analysis of transition boundaries between different types of patterns would further strengthen the conclusions.

We agree that the transition between different patterning regimes should be discussed more quantitatively in the manuscript. Specifically, we identified a highly sensitive parameter range where the disorder in the patterns rapidly increases as a function of the VEGF stimulus. We have improved our discussion of the transition between ‘orderedlike’ patterns and ‘disordered-like’ patterns in the main text as follows: “At relatively low VEGF levels, the patterns were mostly ordered, with small deviations from the expected ‘salt and paper’ geometry with a 25%-75% ratio of TipStalk (Fig. 2D). However, as the VEGF input increased, the fraction of Tips grew and the patterns became sharply more disordered over a relatively narrow range of magnitude of the VEGF input, which could be identified as a highly sensitive area separating more ‘ordered-like’ and ‘disordered-like’ patterns. Finally, increasing VEGF stimuli beyond the highly sensitive area further increased the disorder of the patterns, but with a lower VEGF sensitivity, over several more orders of magnitude of VEGF inputs”.

Reviewer #2 (Recommendations For The Authors):

Please refer to the Public Comments above for a broad review. Below, I provide specific concerns that could be addressed.

Main comments

1. Is the salt-and-pepper model observed for the case when there is no VEGF in the experiments? It would be good to confirm the same. If not, the analysis presented in Fig. 3 could be performed for this case and used as a baseline while referring to the data in Fig. 3.

We thank the referee for the interesting suggestion. The pattern predicted by the model is not strictly salt-and-pepper in absence of VEGF, but the disorder quantified in terms of “incorrect” contacts between Tip cells is considerably lower (see for example the disorder quantification in supplementary figure 1C). We have included the Tip-Tip contact statistics for a case of VEGF=1 ng/ml (100-fold lower that the level used in Fig. 3 compare between model and experiment). In this case, there is clearly more spacing between Tip cells, thus demonstrating how high VEGF stimuli increase the probability of contacts between Tip cells. In the main text, we commented: “As a baseline comparison, the mathematical model with a 100-fold reduction of VEGF stimulus (1 ng/ml) exhibited a Tip-Tip distance statistics more closely comparable with the ‘salt-and-pepper’ model”.

1. The authors mention in the Discussion (end of pg. 7) that ...a low level of exogeneous VEGF is essential to induce Delta-NOTCH signalling.. However, in the standard NOTCH signalling (Boareto et al.), we can get the salt-and-pepper pattern without any VEGF. Am I missing something? The authors may want to take a re-look.

We appreciate the referee’s understanding of the mathematical model. The model used here still exhibits a bistable behavior between the low-Delta and high-Delta cell states even in the absence of VEGF input, as seen for example in the cell state distribution of Fig. 2B, and in agreement with the original model by Boareto et al. This behavior is reflective of the more general applicability of the model, as it describes Delta-NOTCH interactions in various systems. For endothelial cells, VEGF is indeed required to trigger this interaction, but this was not the primary focus of the paper, hence the original model was used. In the text referred to by the reviewer, we are discussing the role,of VEGF based in its known biological effects as well as modeling results. We anticipate that the future further adaptation of the model to,endothelial cells will refine its description of of cell interactions in the absence of VEGF.

1. The size of cells (or spacing between cell nuclei) is highly variable (Fig. 3). Since it is known that the size of cell-cell junctions influences signalling, it would good to at least comment on the same, considering that the model in the paper consists of regular static hexagons. Similarly, it seems desirable to comment on expressing the distance between Tip cells (Fig. 3) in cell length units, when the cell lengths are so variable.

We concur with the suggestion that our consideration of the cell-cell contact size in NOTCH signaling should be clarified in the manuscript.

Sprinzak et al. reported in their 2017 article published in Developmental Cell that the cell-cell contact area does influence NOTCH Signaling. In this article, they found that NOTCH trans-endocytosis (TEC) for pairs with a larger contact width (25µm) is up to five times higher than for pairs with a smaller contact (2.5µm), as observed through the two-cell TEC assay. While TEC correlates with contact width across a range from 1 to 40µm, the values fluctuate significantly in the middle range, particularly when excluding extremely low cell-cell contact areas.

In our experiments, we observed that the cell-cell contact area ranges from essentially infinitesimal corner-to-corner contact to roughly 50µm. We excluded the corner contacts, which might correspond to extremely low cell-cell contact areas, from the Tip-Tip distance measurements as depicted in Fig. 3B. We also made the assumption that variations in cell-cell contact size within tens of microns correlate weakly with the strength of NOTCH signaling. This assumption did not impede our effort to compare the overall trends with results from modeling using hexagonal cells, as shown in Figs 6 D&E. We have included this comment and the corresponding reference to elucidate our assumption in the results as follows: In our experiments, the observed cell-cell contact area varied, spanning from very low (cell corner-to-corner contact) up to approximately 50µm. Previous studies(14, 15) have clearly demonstrated the influence of the cell-cell contact area on NOTCH Signaling, but the values get nosy in the middle range, particularly when excluding extremely low cell-cell contact areas. Reflecting these findings, we excluded the corner contacts, which might correspond to extremely low cell-cell contact areas, from the Tip-Tip distance measurements as depicted in Fig. 3B. We also made an assumption that variations in cell-cell contact size within tens of microns correlate weakly with the strength of NOTCH signaling. This assumption did not impede our effort to compare the overall trends with results from modeling using hexagonal cells, as shown in Figs 3 D&E.

1. The results presented in Fig. 6J are quite striking. However, the number of samples N = 10 and N = 11 seem somewhat low. How does one justify that the findings are not influenced by low number fluctuations?

We acknowledge the reviewer's concerns regarding potential biases stemming from a limited number of samples. The analysis presented in Fig. 6J was specifically designed to complement and support the findings in Fig. 6H. In this context, the counts of sprout and mini-sprout dots correspond to the number of instances "including a sprout" and "including a mini-sprout."

While the counts of sprouts and mini-sprouts in Fig. 6H might seem limited as highlighted by the reviewer, the statistical difference between the two groups was found to be significant. Nevertheless, we expanded our regions of interest to encompass neighboring cells, based on the rationale that the local environment might have closely interacting and similar features. The sample sizes in Figure 6J, represented as N=10 and N=11, equate to an examination of 70 cells and 77 cells, respectively. For instance, in the category "including a sprout," five out of ten groups indicated that all seven neighboring cells in a group exhibited fibronectin levels exceeding a given threshold, translating to 35 cells with fibronectin levels above this threshold. Given that the observed trends in distribution were consistently reasonable across the examinations of both 70 and 77 cells, we would like to state that we are confident in our results.

1. It is written towards the end on pg. 5 that ... although all sprouts indeed formed from mini-sprouts, not all .... However, as can be seen from Fig. 4O, Sprouts can also be generated from Stalk cells. This should be corrected.

Thank you for highlighting the discrepancy between our statement on page 5 and the observations in Fig. 4O. While all sprouts undergo a mini-sprout phase, the transition from Stalk to mini-sprout is not always be observed due to the limitations of our observational timeframe. We acknowledge this oversight and adjusted our statement to clarify that sprouts appearing to form directly from Stalks likely passed through an unobserved intermediate mini-sprout stage as follows: We found that all sprouts formed either directly from Stalks or from mini-sprouts, suggesting a non-observed transition from Stalk to mini-sprout due to observational timeframe limitations. Strikingly, however, not all minisprouts persisted and initiated sprout formation.

1. No solid blue bars are shown in Fig. S2A as mentioned in the caption. Kindly correct.

We apologize for the mistake. We have corrected the figure to show the blue bars depicting the experimental measurements for sprout distance probability.

1. How are the high-Delta cells or high-NOTCH cells decided in experiments or simulations? Does it happen that Delta and NOTCH levels are comparable? In that case, what is done? This point could be clarified in the main manuscript or Materials and Methods.

We agree with the reviewer that Tip cell definition should be clarified. In the model, we define a threshold level for cellular Delta to distinguish Tip and Stalk cells, which is now explained in the Methods section “Definition of Tip cells in the model”. As elaborated in the new section, Delta and NOTCH levels are never comparable due to the circuit’s bistable behavior. In experiments, Tip cells based on their key phenotypic characteristic — invasive migration into the surrounding collagen matrix rather than Delta or NOTCH levels. The details can be found in “Precise quantification of Tip cell spatial arrangement suggests disordered patterning in the engineered angiogenesis model” section and Figure 3A.

Minor comments

There are a good number of typos in the paper. The manuscript should be carefully checked and corrected for the same. Below, I provide a few instances.

1. In the abstract towards the end, it should be "understanding" instead of "understating"

2. On pg. 5, just before the beginning of the last paragraph, there is a typo "parodied" which should most likely be "provided"

3. First paragraph on pg. 6 typo "spouts" instead of "Sprouts"

4. Second paragraph on pg. 6, correctly write "testS"

5. Near the beginning of pg. 8, should be "C. elegans" instead of "C. elegance"

6. Figure 1 caption, towards the end, should be "Stalk" instead of "Salk"

We sincerely appreciate your keen attention to detail. we have thoroughly reviewed the manuscript and made the necessary corrections, including those that you have highlighted.

Reviewer #3 (Recommendations For The Authors):

Major concern:

The authors should discuss in more detail how their work can be used for a better understanding of the angiogenesis process in physiological conditions and in pathological conditions such as post-ischemic revascularization or tumor vascularization.

We have included comments and the corresponding references to clarify the aspect the reviewer suggested: The results in this study can further inform our understanding of angiogenesis in physiological and pathophysiological conditions. In particular, in many circumstances, the levels of VEGF is determined by the degree of hypoxia, which can be highly elevated following oxygen supply interruption, e.g., in wound healing or ischemia, or due to progression of neoplastic growth. Our results suggest that in these cases, formation of sprouts can be dysregulated due to higher incidences of co-localizations of prospective Tip cells. In addition, since these conditions are frequently accompanied by altered synthesis of ECM, the sprout density can increase, which may lead to formation of denser and less developed vascular beds frequently observed as a result of tumor angiogenesis(42, 43). Our results thus suggest that the disorder and higher plasticity of the endothelial cell fate speciation at higher VEGF inputs can be a key contributor to some pathological states associated with persistently hypoxic conditions.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

Summary

This article by Zhai et al, investigates sterol transport in bacteria. Synthesis of sterols is rare in bacteria but occurs in some, such as M capsulatus where the sterols are found primarily in the outer membrane. In a previous paper the authors discovered an operon consisting of five genes, with two of these genes encoding demethylases involved in sterol demethylation. In this manuscript, the authors set out to investigate the functions of the other three genes in the operon. Interestingly, through a bioinformatic analysis, they show that they are an inner membrane transporter of the RND family, a periplasmic binding protein, and an outer membrane-associated protein, all potentially involved with lipid transport, so providing a means of transporting the lipids to the outer membrane. These proteins are then extensively investigated through lipid pulldowns, binding analysis on all three, and X-ray crystallography and docking of the latter two.

Strengths

The lipid pulldowns and associated MST binding analysis are convincing, clearly showing that sterols are able to bind to these proteins. The structures of BstB and BstC are high resolution with excellent maps that allow docking studies to be carried out. These structures are distinct from sterol-binding proteins in eukaryotes.

We thank the reviewer for their favorable impression of this work.

Weaknesses

While the docking and molecular dynamics studies are consistent with the binding of sterols to BstB and BstC, this is not backed up particularly well. The MST results of mutants in the binding pocket of BstB have relatively little effect, and while I agree with the authors this may be because of the extensive hydrophobic interactions that the ligand makes with the protein, it is difficult to make any firm conclusions about binding.

We agree with the reviewer that at this point, there is no experimental evidence to define the sterol binding site in BstB. While in the manuscript we allude to the extensive hydrophobic interactions as being especially stabilizing and difficult to eliminate with one or two mutations, we are now also aware that hydrogen-bonding interactions with the polar head of the sterols are quite important (see data on BstC, where disruption of that interaction significantly reduces the equilibrium affinity for sterols). Our MD simulations show that at least 3 protein amino acids can participate in H-bonding with the sterols. Moreover, recent work from our lab show that even ligand site waters can extend an H-bonding network around the polar head of the lipid (Zhai et al., ChemBioChem 2023, 24, e202300156), thereby enabling H-bonding with amino acids that are further away from the ligand site. It is therefore difficult to predict which mutations will sufficiently destabilize the binding. While this question is one we will tackle in future studies focused on obtaining high-resolution substrate-bound structures of BstB or homologs, the findings reported here are still relevant and timely, and we posit will spur the discovery of functional homologs, including some in organisms that are more tractable.

The authors also discuss the possibility of a secondary binding site in BstB based on a slight cavity in domain B next to a flexible loop. This is not backed up in any way and seems unlikely.

The reviewer is correct in that the evidence for this second binding site weak. While the crystallographic structure shows a highly hydrophobic region and the binding studies suggests cooperativity exists in the binding of the 4methylsterol substrate, the docking studies do not strongly support binding at that site. As such, we have clarified in the manuscript that a second hydrophobic cavity is observed, but that its role in ligand interaction remains unexplored.

Reviewer #2 (Public Review):

Summary:

In eukaryotes, sterols are crucial for signaling and regulating membrane fluidity, however, the mechanism governing cholesterol production and transport across the cell membrane in bacteria remains enigmatic. The manuscript by Zhai et al. sheds light on this topic by uncovering three potential cholesterol transport proteins. Through comprehensive bioinformatics analysis, the authors identified three genes bstA, bstB, and bstC encoding proteins which share homology with transporters, periplasmic binding proteins, and periplasmic components superfamily, respectively. Furthermore, the authors confirmed the specific interaction between these three proteins and C-4 methylated sterols and determined the structures of BstB and BstC. Combining these structural insights with molecular dynamics simulation, they postulated several plausible substrate binding sites within each protein.

Strengths:

The authors have identified 3 proteins that seem likely to be involved in sterol transport between the inner and outer membrane. The structures are of high quality, and the sterol binding experiments support a role for these proteins in sterol transport.

We thank the reviewer for this positive view of our work.

Weaknesses:

While the author's model is very plausible, direct evidence for a role of BstABC in transport, or that the 3 proteins function together in a single pathway, is limited.

The reviewer is correct that we were unable to demonstrate that the three proteins work together to transport 4methylsterols. This is not for lack of trying. We first attempted gene deletion studies, and as mentioned in the manuscript (with more details now provided in the experimental section), this appeared to be lethal. We then attempted in vitro exchange experiments, in which the proteins would be used to transfer sterols from sterol-loaded “heavy” liposomes to a sterol-free “light” liposomes – such exchange assays are frequently performed with eukaryotic sterol transporters (see Chung et al., Science 2015, https://doi.org/10.1126/science.aab1370). These assays were not successful because 1) sterols incorporated poorly into liposomes made with E. coli polar lipids and yielded leaky liposomes; 2) use of liposomes prepared with the TLE of M. capsulatus proved more stable, but no appreciable exchange was observed; we reasoned that this might be due to the absence of an energy source for BstA, the RND component for which we have expressed and purified only the soluble periplasmic domain. Given the technical difficulty of these in vitro transport experiments, we will continue to pursue in vivo demonstration of function as new homologs are identified.

Reviewer #3 (Public Review):

Summary:

The work in this manuscript builds on prior efforts by this team to understand how sterols are biosynthesized and utilized in bacteria. The study reports a new function for three genes encoded near sterol biosynthesis enzymes, suggesting the resulting proteins function as a sterol transport system. Biochemical and structural characterization of the two soluble components of the pathway establishes that both proteins can bind sterols, with a preference for 4methylated derivatives. High-resolution x-ray structures of the apoproteins reveal hydrophobic cavities of the appropriate size to accommodate these substrates. Docking and molecular dynamics simulations confirm this observation and provide specific insights into residues involved in substrate binding.

Strengths:

The manuscript is comprehensive and well-written. The annotation of a new function in a set of proteins related to bacterial sterol usage is exciting and likely to enable further study of this phenomenon - which is currently not well understood. The work also has implications for improving our understanding of lipid usage in general among bacterial organisms.

We thank the reviewer for this synopsis of our work.

Weaknesses:

The authors might consider moving some of the bioinformatics figures to the main text, given how much space is devoted to this topic in the results section.

We have taken this advice and moved Figure S1 to the main manuscript.

Reviewer #1 (Recommendations For The Authors):

1. In the analysis of the MST data, the authors quote Hill coefficients. How reliable are these numbers? For BstB, for instance, it seems unlikely that more than one molecule would bind. Can the analysis be done without needing to include Hill coefficients?

We used fits that did and did not invoke cooperativity – see below. We are certain that both BstA and BstB are better fit with cooperativity invoked.

Author response image 1.

1. In looking at the maps associated with the structures, which were included in the review package, I see that two citric acid molecules fit beautifully into the density where currently PEG has been modelled. This needs to be fixed and some comments may be appropriate in the manuscript.

We thank the reviewer for calling our attention to this. Citric acid has now been added to the model, and we reason that these are present in the structure because citric acid was used in the crystallization condition. The revised model is now present in the PDB.

1. It is not necessary to show the two molecules in the asymmetric unit in Figure 4 given that it is not a dimer. This doesn't add anything to the manuscript.

We now show a single molecule of BstC in Figure 4 (now Figure 5).

1. I wouldn't consider the loops shown in Figure S4 as disordered. They have slightly higher B-values but are not completely mobile.

We did not refer to these loops as disordered. In the text, we say they “exhibit poor electron densities, suggesting conformational sampling of more than one state (Fig. S4A).”

Reviewer #2 (Recommendations For The Authors):

pg 7, "hinting at an astounding distinction": I might suggest a word other than astounding that conveys how statistically unlikely, unusual, etc. this result is.

Thank you – we have removed “astounding”.

pg 7, paragraph 2: Here the authors show that in the SSN analysis, BstB proteins cluster separately and suggest this implies a distinction in function. However, they also show that PhnD homologs do not cluster separately (distributed across multiple clusters), yet presumably have similar functions. I am not familiar with SSN, but it seems to me that the second statement about PhnD implies that the first statement about BstB might not be valid, i.e., if PhnD doesn't cluster based on function, on what basis can we conclude that BstB does? On what basis does clustering occur in the SSN analysis? Might it be driven by things other than function? This comment also concerns the final paragraph of this section.

The reviewer is correct in that PhnD homologs occupy separate clusters of the SSN. Many of these homologs were crystallized with phosphate-like compounds, but it is possible that they have non-overlapping substrate scopes and are therefore functionally distinct. As for the basis of clustering, the SSN is fully sequence-based. What has been observed is that proteins with highly similar sequences can have similar functions – but this is not always true.

pg 8, paragraph 1: The authors suggest that BstABC may be essential. This is probably not a critical claim and it might be simplest to just remove it, but if it is mentioned, the authors should probably explain what was attempted that failed, so a reader can assess the strength of the evidence supporting essentiality. For example, I don't see anything in the methods about genetic manipulations of M. capsulatus, so currently, this falls within the realm of "Data not shown".

We have provided additional information about the experimental techniques used to do this. This statement was included so that it is understood that the reason for the experimental failure is unlikely to be technical in nature, as we have successfully deleted some sterol related genes while others remain intractable.

Fig. 2A: It is unclear to me what is being plotted here, perhaps more experimental detail is required in the form of labels and/or legend. Is this a quantification of each sterol in each fraction separated by GC? There are essentially no methods provided for the GC-MS experiments. A reference is provided, but I think providing detailed methods for these specific experiments will provide a higher degree of scientific rigor. I am not sure what is standard for GCMS, but perhaps showing spectra in the supplement that establish the identity of the bound molecules as species I and II would be appropriate?

Additional experimental details have been provided and the figure legend changed to be more clear. Moreover, we now clearly state that the chromatograms shown were used to identify lipids due to retention times for spectra that were previously published in Wei et al., 2016.

pg 10-11, comparison with PhnD structure: Perhaps it is worth mentioning a 3rd possible explanation for the relative opening/closing of the cleft is simply crystal packing? I don't think it necessarily has to imply anything about a difference in function. Also, the focus seems to be on this pairwise comparison, but perhaps more insights could be gleaned from an analysis that included a wider range of homologs, especially if any are thought to bind hydrophobic substrates.

This could be true, and we have included a statement to that effect. We are unaware of homologs shown to bind to large, hydrophobic molecules.

I think that BstB is shown upside-down in sup movies relative to other figures. If it isn't changed, perhaps adding some labels would help orient the reader.

We have rotated the movies to be more consistent with the figures.

Fig. S7: No units are indicated for Kds (uM?).

Thank you – this has been fixed.

pg 11, paragraph 2. "adjacent to three residues: Glu118, Tyr120 and Asn192": The residue number used in the text doesn't seem to match the numbering in the PDB file. I think these residues correspond to Glu98, Tyr100, and Asn172 in the PDB file.

We regret this error. The correct numbering for both structures is now present in the deposited PDB files (7T1M for BstB and 7T1S for BstC).

pg 12, final paragraph: The authors present binding data for BstB variants with mutations in the putative sterol binding pocket identified in the structural and MD analyses. However, these mutants had no effect on binding. The authors rationalize this in terms of the size of the interface and hydrophobic nature (which indeed, may be correct and is very plausible), and it is worth noting that many of their mutations are to Ala and would largely preserve the hydrophobic nature of the cleft. However, these mutants raise questions about where sterols actually bind. No experimental evidence is presented that substrates bind in the cleft, it is only hypothesized based on structural homology, MD simulations, etc. These mutations formally provide evidence against the hypothesis being tested; I think that has to be discussed a bit more directly, alongside the caveats the authors already discuss about hydrophobicity, etc.

This is a valid point by the reviewer, and it is one we have attempted to address with our statement in the manuscript and in our response to reviewer 1. We have modified the relevant text to more clearly state that there is as of yet no experimental evidence for the binding of sterols to the cavity identified via molecular docking.

pg 13: Presumably this is not the full-length lipoprotein, but has been truncated/mutated in some way? Some statement of roughly what was purified/crystallized should be stated.

The SI methods on protein purification states that the genes of BstB and BstC without their respective signal peptides were obtained.

pg 13, last paragraph "TN1 exhibits hybrid hydrophobicity, with the sides horizontal to cavities being hydrophobic while the vertical sides are more hydrophilic". I don't really follow the horizontal vs vertical sides. Perhaps this could be described in a different way.

Noted and changed to “TN1 is closer to the N-terminal face of the structure, while CA1 and CA2 are proximal to the C-terminal face and form two open hydrophobic pockets; TN1 exhibits a mixture of hydrophobic and hydrophilic amino acids (Fig. 4B and Fig. S9B, Table S4).”

pg 15-16, "Comparison to eukaryotic sterol transporters": Perhaps this would be better suited for the discussion section? Could also be streamlined; it is mostly discussing and comparing eukaryotic sterol binding domains to each other, not to BstABC.

Given that BstB and BstC are the first identified proteins (and putative transporters) for bacterial sterol engagement, we thought a careful description of the existing sterol transporters (which are all eukaryotic) was warranted.

Reviewer #3 (Recommendations For The Authors):

I have just two minor suggestions for the authors if they wish to comment on or address them.

1. Do the three proteins (BstA/B/C) form any sort of complex? Perhaps this property was not assessed - but it seemed possible that the B and C components might constitute a shuttle for the membrane-bound transporter?

This is an important observation – the unliganded version of these proteins show no appreciable affinity for each other. However, BstB (which would be expected to engage both with BstA and BstC) belongs to a family of proteins known to undergo significant conformational change upon substrate binding. It is possible that with substrate present, complexes are formed – we have yet to investigate this.

1. In Figure S1, panel C - it appears that the label for the BstC cluster may have migrated away from the intended location. In this figure, it might also be useful to indicate in the caption the meaning of the red coloring of the nodes?

The label is now fixed – thank you for drawing our attention to this.

Author Response

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

We thank the three reviewers and the reviewing editor for their positive evaluation of our manuscript. We particularly appreciate that they unanimously consider our work as “important contributions to the understanding of how the CAF-1 complex works”, “The large amounts of data provided in the paper support the authors' conclusion very well” and “The paper effectively addresses its primary objective and is strong”. We also thank them for a careful reading and useful comments to improve the manuscript. We have built on these comments to provide an improved version of the manuscript, and address them point by point below .

Reviewer #1 (Public Review):

Summary:

This paper makes important contributions to the structural analysis of the DNA replication-linked nucleosome assembly machine termed Chromatin Assembly Factor-1 (CAF-1). The authors focus on the interplay of domains that bind DNA, histones, and replication clamp protein PCNA.

Strengths:

The authors analyze soluble complexes containing full-length versions of all three fission yeast CAF-1 subunits, an important accomplishment given that many previous structural and biophysical studies have focused on truncated complexes. New data here supports previous experiments indicating that the KER domain is a long alpha helix that binds DNA. Via NMR, the authors discover structural changes at the histone binding site, defined here with high resolution. Most strikingly, the experiments here show that for the S. pombe CAF-1 complex, the WHD domain at the C-terminus of the large subunit lacks DNA binding activity observed in the human and budding yeast homologs, indicating a surprising divergence in the evolution of this complex. Together, these are important contributions to the understanding of how the CAF-1 complex works.

Weaknesses:

1. There are some aspects of the experimentation that are incompletely described: <br /> In the SEC data (Fig. S1C) it appears that Pcf1 in the absence of other proteins forms three major peaks. Two are labeled as "1a" (eluting at ~8 mL) and "1b" (~10-11 mL). It appears that Pcf1 alone or in complex with either or both of the other two subunits forms two different high molecular weight complexes (e.g. 4a/4b, 5a/5b, 6a/6b). There is also a third peak in the analysis of Pcf1 alone, which isn't named here, eluting at ~14 mL, overlapping the peaks labeled 2a, 4c, and 5c. The text describing these different macromolecular complexes seems incomplete (p. 3, lines 32-33): "When isolated, both Pcf2 and Pcf3 are monomeric while Pcf1 forms large soluble oligomers". Which of the three Pcf1-alone peaks are oligomers, and how do we know? What is the third peak? The gel analysis across these chromatograms should be shown.

We thank the reviewer for his/her careful reading of the manuscript. Indeed, we plotted two curves in Figure S1C in a color that does not match the legend, leading to confusion. Curve 1, Pcf1 alone, depicted in red, should appear in pink as indicated in the legend and in the SDS-PAGE analysis below. Curve 1 exhibits two peaks, labeled as 1a and 1b. With an elution volume of 8.5mL close to the dead volume of the column, peak 1a corresponds to soluble oligomers, while peak 1b (10.4mL) likely corresponds to monomeric Pcf1. Curve 5 (Pcf1 + Pcf2 mixture) was in pink instead of purple as indicated in the legend. This curve consists of three distinct peaks (5a, 5b, and 5c). The SDS-PAGE analysis revealed the presence of oligomers of Pcf1-Pcf2 (5a, 8.3mL), the Pcf1-Pcf2 complex (5b, 9.8mL), and Pcf2 alone (5c, 13.6 mL).

The color has now been corrected in the revised manuscript.

More importantly, was a particular SEC peak of the three-subunit CAF-1 complex (i.e. 4a or 4b) characterized in the further experimentation, or were the data obtained from the input material prior to the separation of the different peaks? If the latter, how might this have affected the results? Do the forms inter-convert spontaneously?

We conducted all structural analyses and DNA/PCNA interactions Figures (1-4, S1-S4) with freshly SECpurified samples corresponding to the 4b peak (9.7mL). Aliquots were flash-frozen with 50% glycerol for in vitro histone assembly assays (Figure 5).

1. Given the strong structural predication about the roles of residues L359 and F380 (Fig. 2f), these should be mutated to determine effects on histone binding.

We are pleased that our structural predictions are considered as strong. We agree that investigating the role of the L359 and F380 residues will be critical to further refine the binding interface between histone H3-H4 and CAF-1. An in vitro and in vivo analysis of such mutated forms, alongside the current Pcf1-ED mutant characterized in this article and additional potential mutated forms, has the potential to provide a better understanding of the dynamic of histone deposition by CAF-1. However, these additional approaches would require to reach another step in breaking this enigmatic dynamic.

1. Could it be that the apparent lack of histone deposition by the delta-WHD mutant complex occurs because this mutant complex is unstable when added to the Xenopus extract?

We cannot formally exclude this possibility, and this could potentially applies to all mutated forms tested. However, in the absence of available antibodies against the fission yeast CAF-1 complex, we cannot test this hypothesis for technical reasons. Nevertheless, we feel reassured by the fact that the in vitro assays of nucleosome assembly are overall consistent with the in vivo assays. Indeed, all mutated forms tested that abolished or weakened nucleosome assembly also exhibited synthetic lethality/growth defect in the absence of a functional HIRA pathway, including the delta WHD mutated form. This genetic synergy, that reflects a defective histone deposition by CAF-1, is not specific to the fission yeast S. pombe and was previously reported in S. cerevisiae (Kaufman et al. MCB 1998; Krawitz et al. MCB 2002). This further supports the evolutionary conservation based on genetic assay as a read out for defective histone deposition by CAF-1.

Reviewer #1 (Recommendations For The Authors):

• p. 4: "An experimental molecular weight of 179 kDa was calculated using Small Angle X-ray Scattering (SAXS), consistent with a 1:1:1 stoichiometry (Figure S1e). These data are in agreement with a globular complex with a significant flexibility (Figure S1f)." There needs to be more description of the precision of the molecular weight measurement, and what aspects of these data indicate the flexibility.

The molecular weight was estimated using the correlation volume (Vc) defined by (Rambo & Tainer, Nature 2013, 496, 477-481). The estimated error with this method is around 10%. We added this information together with supporting arguments for the existence of flexibility: “An experimental molecular weight of 179 kDa was calculated using Small Angle X-ray Scattering (SAXS). Assuming an accuracy of around 10% with this method (Rambo and Tainer 2013), this value is consistent with a 1:1:1 stoichiometry for the CAF-1 complex (calculated MW 167kDa) (Figure S1e). In addition, the position of the maximum for the dimensionless Kratky plot was slightly shifted to higher values in the y and x axis compared to the position of the expected maximum of the curve for a fully globular protein (Figure S1f).

This shows that the complex was globular with a significant flexibility.”

• p. 6, lines 21-22: "In contrast, a large part of signals (338-396) did not vanish anymore upon addition of a histone complex preformed with two other histone chaperones known to compete with CAF-1 for histone binding..." Given the contrast made later with the 338-351 region which is insensitive to Asf1/Mcm2, it would be clearer for the reader to describe the Asf1/Mcm2-competed regions as residues 325-338 plus 352-396. Note that the numerical scale of residues doesn't line up perfectly with the data points in Figure 2d, and this should be fixed as well.

We thank this reviewer for spotting this typographical error; we intended to write "In contrast, a large part of signals (348-396) did not vanish anymore… “. We modified paragraph as suggested by the reviewer because we agree it is clearer for the reader : “In contrast, only a shorter fragment (338-347) vanished upon addition of Asf1-H3-H4-Mcm2(69-138), a histone complex preformed with two other histone chaperones, Asf1 and Mcm2, known to compete with CAF-1 for histone binding (Sauer et al. 2017) and whose histone binding modes are well established (Figure 2e) (Huang et al. 2015, Richet et al. 2015). This finding underscores a direct competition between residues (325-338) and (349-396) within the ED domain and Asf1/Mcm2 for histone binding.”

The slight shift in the numerical scale Figure 2d was also corrected.

• p. 8. Lines 22-24: "EMSAs with a double-stranded 40bp DNA fragment confirmed the homogeneity of the bound complex. When increasing the SpCAF-1 concentration, additional mobility shifts suggest, a cooperative DNA binding (Figure 3a)." I agree that the migration of the population is further retarded upon the addition of more protein. However, doesn't this negate the first sentence? That is, if multiple CAF-1 complexes can bind each dsDNA molecule, can these complexes be described as homogeneous?

We fully agree with the reviewer's comment and have removed the notion of homogeneity from the first sentence. “EMSAs with a double-stranded 40bp DNA fragment showed the formation of a bound complex.”

• Figure S2b Legend: "1H-15N HSQC spectra of Pcf1_ED (425-496)." The residue numbers should read 325-396.

The typo has been corrected.

• Is the title for Figure 5 correct?: "Figure 5: Rescue using Y340 and W348 in the ED domain, the intact KER DNA binding domain and the C-terminal WHD of Pcf1 in SpCAF-1 mediated nucleosome assembly." I don't see that any point mutation rescue experiments are done here.

The title of figure 5 has been modified for “Efficient nucleosome assembly by SpCAF-1 in vitro requires interactions with H3-H4, DNA and PCNA, and the C-terminal WHD domain”.

• Figure S6C. I assume the top strain lacks the Pcf2-GFP but this should be stated explicitly.

The following sentence “The top strain corresponds to a strain expressing wild-type and untagged Pcf2 as a negative control of GFP fluorescence” is now added to the figure legend. The figure S6C has been modified accordingly to mention “Pcf2 (untagged)” and state more explicitly.

• Regarding point #3 in the public review, a simple initial test of this idea would be to determine if similar amounts of wt and mutant complexes can be immunoprecipitated at the endpoint of the assembly reactions.

In the absence of available antibodies against the fission yeast CAF-1 complex, we cannot test this hypothesis for technical reasons. However, the in vitro assays of nucleosome assembly are overall consistent with the in vivo assays. Indeed, all mutated forms tested that abolished or weakened nucleosome assembly also exhibited synthetic lethality/growth defect in the absence of a functional HIRA pathway, including the delta WHD mutated form. This genetic synergy, reflecting defective histone deposition by CAF-1, is not specific to the fission yeast S. pombe, as it was previously reported in S. cerevisiae (Kaufman et al. MCB 1998; Krawitz et al. MCB 2002), further supporting the evolution conservation in the genetic assay as a read out for defective histone deposition by CAF-1.

• Foundational findings that should be cited: The role of PCNA in CAF-1 activity was first recognized by pioneering studies in the Stillman laboratory (PMID: 10052459, 11089978). The earliest recombinant studies of CAF-1 showed that the large subunit is the binding platform for the other two, showed that the KER and ED domains were required for histone deposition activity, and roughly mapped the p60-binding site on the large subunit (PMID: 7600578). Another early study roughly mapped the binding site for the third subunit and showed that biological effects of impairing the PCNA binding synergized with defects in the HIR pathway (PMID: 11756556), a genetic synergy first demonstrated in budding yeast (PMID: 9671489).

We thank the reviewer for providing these important references that are now cited in the manuscript. PMID: 10052459 and 11089978 are cited page 2 line 18 and 19, PMID: 7600578 page 19 line 5 and PMID: 11756556 and 9671489 page 18 line 2.

Reviewer #2 (Public Review):

Summary:

The authors describe the structure-functional relationship of domains in S. pombe CAF-1, which promotes DNA replication-coupled deposition of histone H3-H4 dimer. The authors nicely showed that the ED domain with an intrinsically disordered structure binds to histone H3-H4, that the KER domain binds to DNA, and that, in addition to a PIP box, the KER domain also contributes to the PCNA binding. The ED and KER domains as well as the WHD domain are essential for nucleosome assembly in vitro. The ED, KER domains, and the PIP box are important for the maintenance of heterochromatin.

Strengths:

The combination of structural analysis using NMR and Alphafold2 modeling with biophysical and biochemical analysis provided strong evidence on the role of the different domain structures of the large subunit of SpCAF-1, spPCF-1 in the binding to histone H3-H4, DNA as well as PCNA. The conclusion was further supported by genetic analysis of the various pcf1 mutants. The large amounts of data provided in the paper support the authors' conclusion very well.

Reviewer #2 (Recommendations For The Authors):

The paper by Ochesenbein describes the structural and functional analysis of S. pombe CAF-1 complex critical for DNA replication-coupled histone H3/H4 deposition. By using structural, biophysical, and biochemical analyses combined with genetic methods, the authors nicely showed that a large subunit of SpCAF1, SpPCF-1, consists of 5 structured domains with four connecting IDR domains. The ED domain with IDR nature binds to histone H3-H4 dimer with the conformational change of the other domain(s). SpCAF-1 binds to dsDNA by using the KER domain, but not the WHD domain. The experiments have been done with great care and a large amount of the data are highly reliable. Moreover, the results are clearly presented and convincingly written. The conclusion in the paper is very solid and will be useful for researchers who work in the field of chromosome biology.

Major points:

1. DNA binding of the KER mutant shown in Figures S3h and S3i, which was measured by the EMSA, looks similar to that of wild-type control in Figure S3f, which is different from the data in Figures 3b and 3e measured by the MST. The authors need a more precise description of the EMSA result of the KER mutant shown in Figures 3 and S3. The quantification of the EMSA result would resolve the point (should be provided).

A proposed by this reviewer, we performed quantification of all EMSA presented in Figure 3 and Figure S3. We quantified the signal of the free DNA band to calculate a percentage of bound DNA in each condition. All EMSA experiments were conducted in duplicate, allowing us to calculate an average value and standard deviation for each interaction. Representative curves and fitted values are reported below in the figure provided for the reviewer (panel a data for Pcf1_KER domain with two fitting models, panel b for the entire CAF-1 complexes and mutants, panel c for the isolated Pcf1_KER domains), all fitted values in panel d. Importantly, as illustrated in panel a, the complete model for a single interaction (complete KD model, dashed line curve) does not adequately fit the data. In contrast, a function incorporating cooperativity (Hill model) better accounts for the measured data (solid line curve). Consistently, we also used the Hill model to fit the binding curves measured with the MST technique. As also specified now in the text, the Hill model allows to determine an EC50 value (concentration of protein resulting in the disappearance of half of the free DNA band intensity) and a Hill coefficient value (representing cooperativity during the interaction) for each curve.

We measure a value of 3.4 ± 0.4 μM for the EC50 of SpCAF-1 WT, which is higher than the value measured by MST (0.7 ± 0.1 μM). Higher values were also calculated for all mutants and isolated Pcf1_KER domains compared to MST. These discrepancies could raise from the fact that the DNA concentration used in the two techniques were very different (20nM for MST experiments and 1μM for EMSA). Unlike the complete KD model, which includes in the calculation the DNA concentration (considered here as the "receptor"), the Hill model is fitted independently of this value. This model assumes that the “receptor” concentration is low compared to the KD. Here we calculate EC50 values on the same order of magnitude as the DNA concentration (low micromolar), The quantification obtained by EMSA is thus challenging to interpret. In contrast, values fitted by the MST measurements are more reliable since this limitation of low “receptor” concentration is correct.

Therefore, although measurements of EC50 and Hill coefficient from EMSA are reproducible, they may be confusing for quantifying apparent affinity values through EC50. Nevertheless, this quantitative analysis of EMSA, requested by the reviewer, has highlighted an interesting characteristic of the KER mutant that is consistent across both methods: even though the EMSA pointed by the reviewer (Figures S3h and S3i compared to the wild-type control in Figure 3d and Figure S3f) show similar EC50 values, the binding cooperativity is different. Binding curves for the KER mutants is no longer cooperative (Hill coefficient ~1), and this is observed for all KER curves (isolated Pcf1_KER domain and the entire SpCAF-1 complex) with both methods, EMSA and MST. We thus decided to emphasize this characteristic of the KER mutant in the text (page 9 line 30-32). “Importantly, this mutant also shows a lower binding cooperativity for DNA binding, as estimated by the Hill coefficient value close to 1, compared to values around 3 for the WT and other mutants.”

Since EMSA quantifications did not show a loss of “affinity” (as measured by the EC50 value) for the KER* mutants, compared to the WT contrary to MST measurements and because the DNA concentration was close to the measured EC50, we consider that EC50 values calculated by EMSA do not represent a KD value. If we add this quantification, we should discuss this point in detail. Thus, for sake of clarity, we prefer to put in the manuscript EMSA measurements as illustrations and qualitative validations of the interaction but not to include the quantification.

Author response image 1.

Quantitative analysis of interaction with DNA by EMSA. a: quantification of the amount of bound DNA for the Pcf1_KER domain (blue points with error bars). The fit with a KD model is shown as a dashed line, and the fit with a Hill model with a solid line. b: Examples of quantifications and fits (Hill model) for reconstituted SpCAF-1 WT and mutants. c: Examples of quantifications and fits (Hill model) for Pcf1_KER domains WT and mutant. d: EC50 values and Hill coefficients obtained for all EMSA experiments presented in Figure 3 and S3.

1. As with the cooperative DNA binding of CAF-1, it is very important to show the stoichiometry of CAF-1 to the DNA or the site size. Given a long alpha-helix of the KER domain with biased charges, it is also interesting to show a model of how the dsDNA binds to the long helix with a cooperative binding property (this is not essential but would be helpful if the authors discuss it).

We agree that having a molecular model for the binding of the KER helix to DNA would be especially interesting, but at this point, considering the accuracy of the tools currently at our disposal for predicting DNA-protein interactions, such a model would remain highly speculative.

1. Figure 5 shows nucleosome assembly by SpCAF-1. SpCAF-1-PIP* mutant produced a product with faster mobility than the control at 2 h incubation. How much amounts of SpCAF-1 was added in the reaction seems to be critical. At least a few different concentrations of proteins should be tested.

The slightly faster migration of the SpCAF-1-PIPis not systematically reproduced and we observed in several experiments that the band corresponding to supercoiled DNA migrated slightly above or below the one for the complementation by the SpCAF-1-WT (see Author response image 2 below). Thus this indicates that after 2 hours incubation the supercoiling assay with the SpCAF-1-PIP mutant compared to those achieved with the SpCAF-1-WT. To further document whether the WT or the PIP mutant are similar or not, we monitored difference of their nucleosome assembly efficiency by testing their ability to produce supercoiled DNA over shorter time, after 45 minute incubation. Under these conditions, we reproducibly detected supercoiled forms at earlier times with SpCAF-1-WT when compared to the SpCAF-1-PIP* (see figure 5 and Author response image 2). These observations indicate that mutation in the PIP motif of Pcf1 affects the rate of supercoiling in a distinct manner when compared to the other mutations that dramatically impair SpCAF-1 capacity to promote supercoiling.

Author response image 2.

Minor points:

1. Page 8, line 26 or Table 1 legend: Please explain what "EC50" is.

The definition of EC50, together with a reference paper for the Hill model have been added in the text page 8 lines 23-26, “The curves were fitted with a Hill model (Tso et al. 2018) with a EC50 value of 0.7± 0.1µM (effective concentration at which a 50% signal is observed) and a cooperativity (Hill coefficient, h) of 2.7 ± 0.2, in line with a cooperative DNA binging of SpCAF-1.”, in the Table 1 figure legend and in the method section (page 26).

1. Page 13, lines 9, 11: "Xenopus" should be italicized.

This is corrected

1. Page 14, second half: In S. pombe, the pcf1 deletion mutant is not lethal. It is helpful to mention the phenotype of the deletion mutant a bit more when the authors described the genetic analysis of various pcf1 mutants.

This point has been added on page 15, line 1.

1. Figure 1d and Figure S2a: Captions and labels on the X and Y axes are overlapped or misplaced.

This is corrected

1. Figure 5: Please add a schematic figure of the assay to explain how one can check the nucleosome assembly by looking at the form I, supercoiled DNAs.

A new panel has been added to Figure 5. This scheme depicts the supercoiling assay where supercoiled DNA (form I) is used as an indication of efficient nucleosome assembly. The figure legend has also been modified accordingly.

Reviewer #3 (Public Review):

Summary:

The study conducted by Ouasti et al. is an elegant investigation of fission yeast CAF-1, employing a diverse array of technologies to dissect its functions and their interdependence. These functions play a critical role in specifying interactions vital for DNA replication, heterochromatin maintenance, and DNA damage repair, and their dynamics involve multiple interactions. The authors have extensively utilized various in vitro and in vivo tools to validate their model and emphasize the dynamic nature of this complex.

Strengths:

Their work is supported by robust experimental data from multiple techniques, including NMR and SAXS, which validate their molecular model. They conducted in vitro interactions using EMSA and isothermal microcalorimetry, in vitro histone deposition using Xenopus high-speed egg extract, and systematically generated and tested various genetic mutants for functionality in in vivo assays. They successfully delineated domain-specific functions using in vitro assays and could validate their roles to large extent using genetic mutants. One significant revelation from this study is the unfolded nature of the acidic domain, observed to fold when binding to histones. Additionally, the authors also elucidated the role of the long KER helix in mediating DNA binding and enhancing the association of CAF-1 with PCNA. The paper effectively addresses its primary objective and is strong.

Weaknesses:

A few relatively minor unresolved aspects persist, which, if clarified or experimentally addressed by the authors, could further bolster the study.

1. The precise function of the WHD domain remains elusive. Its deletion does not result in DNA damage accumulation or defects in heterochromatin maintenance. This raises questions about the biological significance of this domain and whether it is dispensable. While in vitro assays revealed defects in chromatin assembly using this mutant (Figure 5), confirming these phenotypes through in vivo assays would provide additional assurance that the lack of function is not simply due to the in vitro system lacking PTMs or other regulatory factors.

Our work demonstrates that the WHD domain is important CAF-1 function during DNA replication. Indeed, the deletion of this domain lead to a synthetic lethality when combined with mutation of the HIRA complex, as observed for a null pcf1 mutant, indicating a severe loss of function in the absence of the WHD domain. We propose that these genetic interactions, previously reported in S. cerevisiae (Kaufman et al. MCB 1998; Krawitz et al. MCB 2002) are indicative of a defective histone deposition by CAF-1. Moreover, our work establishes that this domain is dispensable to prevent DNA damage accumulation and to maintain silencing at centromeric heterochromatin, indicating that the WHD domain specifies CAF-1 functions. Moreover, our work further demonstrates that, in contrast to the S. cerevisiae and human WHD domain, the S. pombe counterpart exhibits no DNA binding activity. We thus agree that the WHD domain may contribute to nucleosome assembly in vivo via PTMs or interactions with regulatory factors that may potentially lack in in vitro systems. However, addressing these aspects deserves further investigations beyond the scope of this article.

1. The observation of increased Pcf2-gfp foci in pcf1-ED cells, particularly in mono-nucleated (G2phase) and bi-nucleated cells with septum marks (S-phase), might suggest the presence of replication stress. This could imply incomplete replication in specific regions, leading to the persistence of Caf1-ED-PCNA factories throughout the cell cycle. To further confirm this, detecting accumulated single-stranded DNA (ssDNA) regions outside of S-phase using RPA as an ssDNA marker could be informative.

We cannot formally exclude that cells expressing the Pcf1-ED mutated form exhibit incomplete replication in specific regions, an aspect that would require careful investigations. However, the microscopy analysis (Fig. 6c and S6c) of this mutant showed no alteration in the cell morphology, including the absence of elongated cells compared to wild type, a hallmark of checkpoint activation caused by ssDNA (Enoch et al. Gene & Dev 1992). Therefore, investigating the consequences of the interplay between the binding of CAF-1 to PCNA and histones on the dynamic of DNA replication, is of particular interest but out of the scope of the current manuscript.

1. Moreover, considering the authors' strong assertion of histone binding defects in ED through in vitro assays (Figure 2d and S2a), these claims could be further substantiated, especially considering that some degree of histone deposition might still persist in vivo in the ED mutant (Figure 7d, viable though growth defective double ED*+hip1D mutants). For example, the approach, akin to the one employed in Fig. 6a (FLAG-IPs of various Pcf1-FLAG-tagged mutants), could also enable a comparison of the association of different mutants with histones and PCNA, providing a more thorough validation of their findings.

We have provided in the current manuscript data establishing how Pcf1 mutated forms interacted with PCNA (Fig. 6a, 6b). Regarding the interactions with histone H3-H4, the approach based on immunoprecipitation using various Pcf1-FLAG tagged mutants has been unsuccessful in our hands. Indeed, we were unable to obtain robust and reproducible interactions between Pcf1 or its various mutated form with H3-H4. This is likely because Co-IP approaches do not probe for direct interactions. Indirect interactions between Pcf1 and H3-H4 are potentially bridged by additional factors, including the two other subunits of CAF-1, Pcf2 and Pcf3, or Asf1. Therefore, we are not in a position to address in vivo the direct interactions between Pcf1 and histone H3-H4.

1. It would be valuable for the authors to speculate on the necessity of having disordered regions in CAF1. Specifically, exploring the overall distribution of these domains within disordered/unfolded structures could provide insightful perspectives. Additionally, it's intriguing to note that the significant disparities observed among mutants (ED, PIP, and KER*) in in vitro assays seem to become more generic in vivo, except for the indispensability of the WHD-domain. Could these disordered regions potentially play a crucial role in the phase separation of replication factories? Considering these questions could offer valuable insights into the underlying mechanisms at play.

We agree that the potential mechanistic role of partial disorder in CAF-1 is particularly interesting. Disordered regions of human CAF-1 have been reported to form nuclear bodies with liquid-liquid phase separation properties to maintain HIV latency (Ma et al EMBO J. 2021). As suggested, this raises the question of how disordered domains of Pcf1 could promote phase separation for replication factories, if such phenomenon happens in vivo. Moreover, numerous factors of the replisome also harbor disordered regions (Bedina, A. et al, 2013. Intrinsically Disordered Proteins in Replication Process. InTech. doi: 10.5772/51673), adding complexity in disentangling experimentally such questions. We have added these elements at the end of the discussion in the revised manuscript (page 20, lines 23-29). “Such plasticity and cross-talks provided by structurally disordered domains might be key for the multivalent CAF-1 functions. Human CAF-1 has been reported to form nuclear bodies with liquid-liquid phase separation properties to maintain HIV latency (Ma et al. 2021). This raises the question of a potential role of the disordered domains of Pcf1, together with other replisome factor harbouring such disordered regions (Bedina 2013), in promoting phase separation of replication factories, if such phenomenon happens in vivo. Further studies will be needed to tackle these questions.”

Author Response

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

We thank the three reviewers and the reviewing editor for their positive evaluation of our manuscript. We particularly appreciate that they unanimously consider our work as “important contributions to the understanding of how the CAF-1 complex works”, “The large amounts of data provided in the paper support the authors' conclusion very well” and “The paper effectively addresses its primary objective and is strong”. We also thank them for a careful reading and useful comments to improve the manuscript. We have built on these comments to provide an improved version of the manuscript, and address them point by point below .

Reviewer #1 (Public Review):

Summary:

This paper makes important contributions to the structural analysis of the DNA replication-linked nucleosome assembly machine termed Chromatin Assembly Factor-1 (CAF-1). The authors focus on the interplay of domains that bind DNA, histones, and replication clamp protein PCNA.

Strengths:

The authors analyze soluble complexes containing full-length versions of all three fission yeast CAF-1 subunits, an important accomplishment given that many previous structural and biophysical studies have focused on truncated complexes. New data here supports previous experiments indicating that the KER domain is a long alpha helix that binds DNA. Via NMR, the authors discover structural changes at the histone binding site, defined here with high resolution. Most strikingly, the experiments here show that for the S. pombe CAF-1 complex, the WHD domain at the C-terminus of the large subunit lacks DNA binding activity observed in the human and budding yeast homologs, indicating a surprising divergence in the evolution of this complex. Together, these are important contributions to the understanding of how the CAF-1 complex works.

Weaknesses:

1. There are some aspects of the experimentation that are incompletely described: <br /> In the SEC data (Fig. S1C) it appears that Pcf1 in the absence of other proteins forms three major peaks. Two are labeled as "1a" (eluting at ~8 mL) and "1b" (~10-11 mL). It appears that Pcf1 alone or in complex with either or both of the other two subunits forms two different high molecular weight complexes (e.g. 4a/4b, 5a/5b, 6a/6b). There is also a third peak in the analysis of Pcf1 alone, which isn't named here, eluting at ~14 mL, overlapping the peaks labeled 2a, 4c, and 5c. The text describing these different macromolecular complexes seems incomplete (p. 3, lines 32-33): "When isolated, both Pcf2 and Pcf3 are monomeric while Pcf1 forms large soluble oligomers". Which of the three Pcf1-alone peaks are oligomers, and how do we know? What is the third peak? The gel analysis across these chromatograms should be shown.

We thank the reviewer for his/her careful reading of the manuscript. Indeed, we plotted two curves in Figure S1C in a color that does not match the legend, leading to confusion. Curve 1, Pcf1 alone, depicted in red, should appear in pink as indicated in the legend and in the SDS-PAGE analysis below. Curve 1 exhibits two peaks, labeled as 1a and 1b. With an elution volume of 8.5mL close to the dead volume of the column, peak 1a corresponds to soluble oligomers, while peak 1b (10.4mL) likely corresponds to monomeric Pcf1. Curve 5 (Pcf1 + Pcf2 mixture) was in pink instead of purple as indicated in the legend. This curve consists of three distinct peaks (5a, 5b, and 5c). The SDS-PAGE analysis revealed the presence of oligomers of Pcf1-Pcf2 (5a, 8.3mL), the Pcf1-Pcf2 complex (5b, 9.8mL), and Pcf2 alone (5c, 13.6 mL).

The color has now been corrected in the revised manuscript.

More importantly, was a particular SEC peak of the three-subunit CAF-1 complex (i.e. 4a or 4b) characterized in the further experimentation, or were the data obtained from the input material prior to the separation of the different peaks? If the latter, how might this have affected the results? Do the forms inter-convert spontaneously?

We conducted all structural analyses and DNA/PCNA interactions Figures (1-4, S1-S4) with freshly SECpurified samples corresponding to the 4b peak (9.7mL). Aliquots were flash-frozen with 50% glycerol for in vitro histone assembly assays (Figure 5).

1. Given the strong structural predication about the roles of residues L359 and F380 (Fig. 2f), these should be mutated to determine effects on histone binding.

We are pleased that our structural predictions are considered as strong. We agree that investigating the role of the L359 and F380 residues will be critical to further refine the binding interface between histone H3-H4 and CAF-1. An in vitro and in vivo analysis of such mutated forms, alongside the current Pcf1-ED mutant characterized in this article and additional potential mutated forms, has the potential to provide a better understanding of the dynamic of histone deposition by CAF-1. However, these additional approaches would require to reach another step in breaking this enigmatic dynamic.

1. Could it be that the apparent lack of histone deposition by the delta-WHD mutant complex occurs because this mutant complex is unstable when added to the Xenopus extract?

We cannot formally exclude this possibility, and this could potentially applies to all mutated forms tested. However, in the absence of available antibodies against the fission yeast CAF-1 complex, we cannot test this hypothesis for technical reasons. Nevertheless, we feel reassured by the fact that the in vitro assays of nucleosome assembly are overall consistent with the in vivo assays. Indeed, all mutated forms tested that abolished or weakened nucleosome assembly also exhibited synthetic lethality/growth defect in the absence of a functional HIRA pathway, including the delta WHD mutated form. This genetic synergy, that reflects a defective histone deposition by CAF-1, is not specific to the fission yeast S. pombe and was previously reported in S. cerevisiae (Kaufman et al. MCB 1998; Krawitz et al. MCB 2002). This further supports the evolutionary conservation based on genetic assay as a read out for defective histone deposition by CAF-1.

Reviewer #1 (Recommendations For The Authors):

• p. 4: "An experimental molecular weight of 179 kDa was calculated using Small Angle X-ray Scattering (SAXS), consistent with a 1:1:1 stoichiometry (Figure S1e). These data are in agreement with a globular complex with a significant flexibility (Figure S1f)." There needs to be more description of the precision of the molecular weight measurement, and what aspects of these data indicate the flexibility.

The molecular weight was estimated using the correlation volume (Vc) defined by (Rambo & Tainer, Nature 2013, 496, 477-481). The estimated error with this method is around 10%. We added this information together with supporting arguments for the existence of flexibility: “An experimental molecular weight of 179 kDa was calculated using Small Angle X-ray Scattering (SAXS). Assuming an accuracy of around 10% with this method (Rambo and Tainer 2013), this value is consistent with a 1:1:1 stoichiometry for the CAF-1 complex (calculated MW 167kDa) (Figure S1e). In addition, the position of the maximum for the dimensionless Kratky plot was slightly shifted to higher values in the y and x axis compared to the position of the expected maximum of the curve for a fully globular protein (Figure S1f).

This shows that the complex was globular with a significant flexibility.”

• p. 6, lines 21-22: "In contrast, a large part of signals (338-396) did not vanish anymore upon addition of a histone complex preformed with two other histone chaperones known to compete with CAF-1 for histone binding..." Given the contrast made later with the 338-351 region which is insensitive to Asf1/Mcm2, it would be clearer for the reader to describe the Asf1/Mcm2-competed regions as residues 325-338 plus 352-396. Note that the numerical scale of residues doesn't line up perfectly with the data points in Figure 2d, and this should be fixed as well.

We thank this reviewer for spotting this typographical error; we intended to write "In contrast, a large part of signals (348-396) did not vanish anymore… “. We modified paragraph as suggested by the reviewer because we agree it is clearer for the reader : “In contrast, only a shorter fragment (338-347) vanished upon addition of Asf1-H3-H4-Mcm2(69-138), a histone complex preformed with two other histone chaperones, Asf1 and Mcm2, known to compete with CAF-1 for histone binding (Sauer et al. 2017) and whose histone binding modes are well established (Figure 2e) (Huang et al. 2015, Richet et al. 2015). This finding underscores a direct competition between residues (325-338) and (349-396) within the ED domain and Asf1/Mcm2 for histone binding.”

The slight shift in the numerical scale Figure 2d was also corrected.

• p. 8. Lines 22-24: "EMSAs with a double-stranded 40bp DNA fragment confirmed the homogeneity of the bound complex. When increasing the SpCAF-1 concentration, additional mobility shifts suggest, a cooperative DNA binding (Figure 3a)." I agree that the migration of the population is further retarded upon the addition of more protein. However, doesn't this negate the first sentence? That is, if multiple CAF-1 complexes can bind each dsDNA molecule, can these complexes be described as homogeneous?

We fully agree with the reviewer's comment and have removed the notion of homogeneity from the first sentence. “EMSAs with a double-stranded 40bp DNA fragment showed the formation of a bound complex.”

• Figure S2b Legend: "1H-15N HSQC spectra of Pcf1_ED (425-496)." The residue numbers should read 325-396.

The typo has been corrected.

• Is the title for Figure 5 correct?: "Figure 5: Rescue using Y340 and W348 in the ED domain, the intact KER DNA binding domain and the C-terminal WHD of Pcf1 in SpCAF-1 mediated nucleosome assembly." I don't see that any point mutation rescue experiments are done here.

The title of figure 5 has been modified for “Efficient nucleosome assembly by SpCAF-1 in vitro requires interactions with H3-H4, DNA and PCNA, and the C-terminal WHD domain”.

• Figure S6C. I assume the top strain lacks the Pcf2-GFP but this should be stated explicitly.

The following sentence “The top strain corresponds to a strain expressing wild-type and untagged Pcf2 as a negative control of GFP fluorescence” is now added to the figure legend. The figure S6C has been modified accordingly to mention “Pcf2 (untagged)” and state more explicitly.

• Regarding point #3 in the public review, a simple initial test of this idea would be to determine if similar amounts of wt and mutant complexes can be immunoprecipitated at the endpoint of the assembly reactions.

In the absence of available antibodies against the fission yeast CAF-1 complex, we cannot test this hypothesis for technical reasons. However, the in vitro assays of nucleosome assembly are overall consistent with the in vivo assays. Indeed, all mutated forms tested that abolished or weakened nucleosome assembly also exhibited synthetic lethality/growth defect in the absence of a functional HIRA pathway, including the delta WHD mutated form. This genetic synergy, reflecting defective histone deposition by CAF-1, is not specific to the fission yeast S. pombe, as it was previously reported in S. cerevisiae (Kaufman et al. MCB 1998; Krawitz et al. MCB 2002), further supporting the evolution conservation in the genetic assay as a read out for defective histone deposition by CAF-1.

• Foundational findings that should be cited: The role of PCNA in CAF-1 activity was first recognized by pioneering studies in the Stillman laboratory (PMID: 10052459, 11089978). The earliest recombinant studies of CAF-1 showed that the large subunit is the binding platform for the other two, showed that the KER and ED domains were required for histone deposition activity, and roughly mapped the p60-binding site on the large subunit (PMID: 7600578). Another early study roughly mapped the binding site for the third subunit and showed that biological effects of impairing the PCNA binding synergized with defects in the HIR pathway (PMID: 11756556), a genetic synergy first demonstrated in budding yeast (PMID: 9671489).

We thank the reviewer for providing these important references that are now cited in the manuscript. PMID: 10052459 and 11089978 are cited page 2 line 18 and 19, PMID: 7600578 page 19 line 5 and PMID: 11756556 and 9671489 page 18 line 2.

Reviewer #2 (Public Review):

Summary:

The authors describe the structure-functional relationship of domains in S. pombe CAF-1, which promotes DNA replication-coupled deposition of histone H3-H4 dimer. The authors nicely showed that the ED domain with an intrinsically disordered structure binds to histone H3-H4, that the KER domain binds to DNA, and that, in addition to a PIP box, the KER domain also contributes to the PCNA binding. The ED and KER domains as well as the WHD domain are essential for nucleosome assembly in vitro. The ED, KER domains, and the PIP box are important for the maintenance of heterochromatin.

Strengths:

The combination of structural analysis using NMR and Alphafold2 modeling with biophysical and biochemical analysis provided strong evidence on the role of the different domain structures of the large subunit of SpCAF-1, spPCF-1 in the binding to histone H3-H4, DNA as well as PCNA. The conclusion was further supported by genetic analysis of the various pcf1 mutants. The large amounts of data provided in the paper support the authors' conclusion very well.

Reviewer #2 (Recommendations For The Authors):

The paper by Ochesenbein describes the structural and functional analysis of S. pombe CAF-1 complex critical for DNA replication-coupled histone H3/H4 deposition. By using structural, biophysical, and biochemical analyses combined with genetic methods, the authors nicely showed that a large subunit of SpCAF1, SpPCF-1, consists of 5 structured domains with four connecting IDR domains. The ED domain with IDR nature binds to histone H3-H4 dimer with the conformational change of the other domain(s). SpCAF-1 binds to dsDNA by using the KER domain, but not the WHD domain. The experiments have been done with great care and a large amount of the data are highly reliable. Moreover, the results are clearly presented and convincingly written. The conclusion in the paper is very solid and will be useful for researchers who work in the field of chromosome biology.

Major points:

1. DNA binding of the KER mutant shown in Figures S3h and S3i, which was measured by the EMSA, looks similar to that of wild-type control in Figure S3f, which is different from the data in Figures 3b and 3e measured by the MST. The authors need a more precise description of the EMSA result of the KER mutant shown in Figures 3 and S3. The quantification of the EMSA result would resolve the point (should be provided).

A proposed by this reviewer, we performed quantification of all EMSA presented in Figure 3 and Figure S3. We quantified the signal of the free DNA band to calculate a percentage of bound DNA in each condition. All EMSA experiments were conducted in duplicate, allowing us to calculate an average value and standard deviation for each interaction. Representative curves and fitted values are reported below in the figure provided for the reviewer (panel a data for Pcf1_KER domain with two fitting models, panel b for the entire CAF-1 complexes and mutants, panel c for the isolated Pcf1_KER domains), all fitted values in panel d. Importantly, as illustrated in panel a, the complete model for a single interaction (complete KD model, dashed line curve) does not adequately fit the data. In contrast, a function incorporating cooperativity (Hill model) better accounts for the measured data (solid line curve). Consistently, we also used the Hill model to fit the binding curves measured with the MST technique. As also specified now in the text, the Hill model allows to determine an EC50 value (concentration of protein resulting in the disappearance of half of the free DNA band intensity) and a Hill coefficient value (representing cooperativity during the interaction) for each curve.

We measure a value of 3.4 ± 0.4 μM for the EC50 of SpCAF-1 WT, which is higher than the value measured by MST (0.7 ± 0.1 μM). Higher values were also calculated for all mutants and isolated Pcf1_KER domains compared to MST. These discrepancies could raise from the fact that the DNA concentration used in the two techniques were very different (20nM for MST experiments and 1μM for EMSA). Unlike the complete KD model, which includes in the calculation the DNA concentration (considered here as the "receptor"), the Hill model is fitted independently of this value. This model assumes that the “receptor” concentration is low compared to the KD. Here we calculate EC50 values on the same order of magnitude as the DNA concentration (low micromolar), The quantification obtained by EMSA is thus challenging to interpret. In contrast, values fitted by the MST measurements are more reliable since this limitation of low “receptor” concentration is correct.

Therefore, although measurements of EC50 and Hill coefficient from EMSA are reproducible, they may be confusing for quantifying apparent affinity values through EC50. Nevertheless, this quantitative analysis of EMSA, requested by the reviewer, has highlighted an interesting characteristic of the KER mutant that is consistent across both methods: even though the EMSA pointed by the reviewer (Figures S3h and S3i compared to the wild-type control in Figure 3d and Figure S3f) show similar EC50 values, the binding cooperativity is different. Binding curves for the KER mutants is no longer cooperative (Hill coefficient ~1), and this is observed for all KER curves (isolated Pcf1_KER domain and the entire SpCAF-1 complex) with both methods, EMSA and MST. We thus decided to emphasize this characteristic of the KER mutant in the text (page 9 line 30-32). “Importantly, this mutant also shows a lower binding cooperativity for DNA binding, as estimated by the Hill coefficient value close to 1, compared to values around 3 for the WT and other mutants.”

Since EMSA quantifications did not show a loss of “affinity” (as measured by the EC50 value) for the KER* mutants, compared to the WT contrary to MST measurements and because the DNA concentration was close to the measured EC50, we consider that EC50 values calculated by EMSA do not represent a KD value. If we add this quantification, we should discuss this point in detail. Thus, for sake of clarity, we prefer to put in the manuscript EMSA measurements as illustrations and qualitative validations of the interaction but not to include the quantification.

Author response image 1.

Quantitative analysis of interaction with DNA by EMSA. a: quantification of the amount of bound DNA for the Pcf1_KER domain (blue points with error bars). The fit with a KD model is shown as a dashed line, and the fit with a Hill model with a solid line. b: Examples of quantifications and fits (Hill model) for reconstituted SpCAF-1 WT and mutants. c: Examples of quantifications and fits (Hill model) for Pcf1_KER domains WT and mutant. d: EC50 values and Hill coefficients obtained for all EMSA experiments presented in Figure 3 and S3.

1. As with the cooperative DNA binding of CAF-1, it is very important to show the stoichiometry of CAF-1 to the DNA or the site size. Given a long alpha-helix of the KER domain with biased charges, it is also interesting to show a model of how the dsDNA binds to the long helix with a cooperative binding property (this is not essential but would be helpful if the authors discuss it).

We agree that having a molecular model for the binding of the KER helix to DNA would be especially interesting, but at this point, considering the accuracy of the tools currently at our disposal for predicting DNA-protein interactions, such a model would remain highly speculative.

1. Figure 5 shows nucleosome assembly by SpCAF-1. SpCAF-1-PIP* mutant produced a product with faster mobility than the control at 2 h incubation. How much amounts of SpCAF-1 was added in the reaction seems to be critical. At least a few different concentrations of proteins should be tested.

The slightly faster migration of the SpCAF-1-PIPis not systematically reproduced and we observed in several experiments that the band corresponding to supercoiled DNA migrated slightly above or below the one for the complementation by the SpCAF-1-WT (see Author response image 2 below). Thus this indicates that after 2 hours incubation the supercoiling assay with the SpCAF-1-PIP mutant compared to those achieved with the SpCAF-1-WT. To further document whether the WT or the PIP mutant are similar or not, we monitored difference of their nucleosome assembly efficiency by testing their ability to produce supercoiled DNA over shorter time, after 45 minute incubation. Under these conditions, we reproducibly detected supercoiled forms at earlier times with SpCAF-1-WT when compared to the SpCAF-1-PIP* (see figure 5 and Author response image 2). These observations indicate that mutation in the PIP motif of Pcf1 affects the rate of supercoiling in a distinct manner when compared to the other mutations that dramatically impair SpCAF-1 capacity to promote supercoiling.

Author response image 2.

Minor points:

1. Page 8, line 26 or Table 1 legend: Please explain what "EC50" is.

The definition of EC50, together with a reference paper for the Hill model have been added in the text page 8 lines 23-26, “The curves were fitted with a Hill model (Tso et al. 2018) with a EC50 value of 0.7± 0.1µM (effective concentration at which a 50% signal is observed) and a cooperativity (Hill coefficient, h) of 2.7 ± 0.2, in line with a cooperative DNA binging of SpCAF-1.”, in the Table 1 figure legend and in the method section (page 26).

1. Page 13, lines 9, 11: "Xenopus" should be italicized.

This is corrected

1. Page 14, second half: In S. pombe, the pcf1 deletion mutant is not lethal. It is helpful to mention the phenotype of the deletion mutant a bit more when the authors described the genetic analysis of various pcf1 mutants.

This point has been added on page 15, line 1.

1. Figure 1d and Figure S2a: Captions and labels on the X and Y axes are overlapped or misplaced.

This is corrected

1. Figure 5: Please add a schematic figure of the assay to explain how one can check the nucleosome assembly by looking at the form I, supercoiled DNAs.

A new panel has been added to Figure 5. This scheme depicts the supercoiling assay where supercoiled DNA (form I) is used as an indication of efficient nucleosome assembly. The figure legend has also been modified accordingly.

Reviewer #3 (Public Review):

Summary:

The study conducted by Ouasti et al. is an elegant investigation of fission yeast CAF-1, employing a diverse array of technologies to dissect its functions and their interdependence. These functions play a critical role in specifying interactions vital for DNA replication, heterochromatin maintenance, and DNA damage repair, and their dynamics involve multiple interactions. The authors have extensively utilized various in vitro and in vivo tools to validate their model and emphasize the dynamic nature of this complex.

Strengths:

Their work is supported by robust experimental data from multiple techniques, including NMR and SAXS, which validate their molecular model. They conducted in vitro interactions using EMSA and isothermal microcalorimetry, in vitro histone deposition using Xenopus high-speed egg extract, and systematically generated and tested various genetic mutants for functionality in in vivo assays. They successfully delineated domain-specific functions using in vitro assays and could validate their roles to large extent using genetic mutants. One significant revelation from this study is the unfolded nature of the acidic domain, observed to fold when binding to histones. Additionally, the authors also elucidated the role of the long KER helix in mediating DNA binding and enhancing the association of CAF-1 with PCNA. The paper effectively addresses its primary objective and is strong.

Weaknesses:

A few relatively minor unresolved aspects persist, which, if clarified or experimentally addressed by the authors, could further bolster the study.

1. The precise function of the WHD domain remains elusive. Its deletion does not result in DNA damage accumulation or defects in heterochromatin maintenance. This raises questions about the biological significance of this domain and whether it is dispensable. While in vitro assays revealed defects in chromatin assembly using this mutant (Figure 5), confirming these phenotypes through in vivo assays would provide additional assurance that the lack of function is not simply due to the in vitro system lacking PTMs or other regulatory factors.

Our work demonstrates that the WHD domain is important CAF-1 function during DNA replication. Indeed, the deletion of this domain lead to a synthetic lethality when combined with mutation of the HIRA complex, as observed for a null pcf1 mutant, indicating a severe loss of function in the absence of the WHD domain. We propose that these genetic interactions, previously reported in S. cerevisiae (Kaufman et al. MCB 1998; Krawitz et al. MCB 2002) are indicative of a defective histone deposition by CAF-1. Moreover, our work establishes that this domain is dispensable to prevent DNA damage accumulation and to maintain silencing at centromeric heterochromatin, indicating that the WHD domain specifies CAF-1 functions. Moreover, our work further demonstrates that, in contrast to the S. cerevisiae and human WHD domain, the S. pombe counterpart exhibits no DNA binding activity. We thus agree that the WHD domain may contribute to nucleosome assembly in vivo via PTMs or interactions with regulatory factors that may potentially lack in in vitro systems. However, addressing these aspects deserves further investigations beyond the scope of this article.

1. The observation of increased Pcf2-gfp foci in pcf1-ED cells, particularly in mono-nucleated (G2phase) and bi-nucleated cells with septum marks (S-phase), might suggest the presence of replication stress. This could imply incomplete replication in specific regions, leading to the persistence of Caf1-ED-PCNA factories throughout the cell cycle. To further confirm this, detecting accumulated single-stranded DNA (ssDNA) regions outside of S-phase using RPA as an ssDNA marker could be informative.

We cannot formally exclude that cells expressing the Pcf1-ED mutated form exhibit incomplete replication in specific regions, an aspect that would require careful investigations. However, the microscopy analysis (Fig. 6c and S6c) of this mutant showed no alteration in the cell morphology, including the absence of elongated cells compared to wild type, a hallmark of checkpoint activation caused by ssDNA (Enoch et al. Gene & Dev 1992). Therefore, investigating the consequences of the interplay between the binding of CAF-1 to PCNA and histones on the dynamic of DNA replication, is of particular interest but out of the scope of the current manuscript.

1. Moreover, considering the authors' strong assertion of histone binding defects in ED through in vitro assays (Figure 2d and S2a), these claims could be further substantiated, especially considering that some degree of histone deposition might still persist in vivo in the ED mutant (Figure 7d, viable though growth defective double ED*+hip1D mutants). For example, the approach, akin to the one employed in Fig. 6a (FLAG-IPs of various Pcf1-FLAG-tagged mutants), could also enable a comparison of the association of different mutants with histones and PCNA, providing a more thorough validation of their findings.

We have provided in the current manuscript data establishing how Pcf1 mutated forms interacted with PCNA (Fig. 6a, 6b). Regarding the interactions with histone H3-H4, the approach based on immunoprecipitation using various Pcf1-FLAG tagged mutants has been unsuccessful in our hands. Indeed, we were unable to obtain robust and reproducible interactions between Pcf1 or its various mutated form with H3-H4. This is likely because Co-IP approaches do not probe for direct interactions. Indirect interactions between Pcf1 and H3-H4 are potentially bridged by additional factors, including the two other subunits of CAF-1, Pcf2 and Pcf3, or Asf1. Therefore, we are not in a position to address in vivo the direct interactions between Pcf1 and histone H3-H4.

1. It would be valuable for the authors to speculate on the necessity of having disordered regions in CAF1. Specifically, exploring the overall distribution of these domains within disordered/unfolded structures could provide insightful perspectives. Additionally, it's intriguing to note that the significant disparities observed among mutants (ED, PIP, and KER*) in in vitro assays seem to become more generic in vivo, except for the indispensability of the WHD-domain. Could these disordered regions potentially play a crucial role in the phase separation of replication factories? Considering these questions could offer valuable insights into the underlying mechanisms at play.

We agree that the potential mechanistic role of partial disorder in CAF-1 is particularly interesting. Disordered regions of human CAF-1 have been reported to form nuclear bodies with liquid-liquid phase separation properties to maintain HIV latency (Ma et al EMBO J. 2021). As suggested, this raises the question of how disordered domains of Pcf1 could promote phase separation for replication factories, if such phenomenon happens in vivo. Moreover, numerous factors of the replisome also harbor disordered regions (Bedina, A. et al, 2013. Intrinsically Disordered Proteins in Replication Process. InTech. doi: 10.5772/51673), adding complexity in disentangling experimentally such questions. We have added these elements at the end of the discussion in the revised manuscript (page 20, lines 23-29). “Such plasticity and cross-talks provided by structurally disordered domains might be key for the multivalent CAF-1 functions. Human CAF-1 has been reported to form nuclear bodies with liquid-liquid phase separation properties to maintain HIV latency (Ma et al. 2021). This raises the question of a potential role of the disordered domains of Pcf1, together with other replisome factor harbouring such disordered regions (Bedina 2013), in promoting phase separation of replication factories, if such phenomenon happens in vivo. Further studies will be needed to tackle these questions.”

Author Response

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

Public Reviews:

Reviewer #1:

Summary:

Ngoune et al. present compelling evidence that Slender cells are challenged to infect tsetse flies. They explore the experimental context of a recent important paper in the field, Schuster et al., that presents evidence suggesting the proliferative Slender bloodstream T. brucei can infect juvenile tsetse flies. Schuster et al. were disruptive to the widely accepted paradigm that the Stumpy bloodstream-form is solely responsible for tsetse infection and T. brucei transmission potential. Evidence presented here shows that in all cases, Stumpy form parasites are exponentially more capable of infecting tsetse flies. They further show that Slender cells do not infect mature flies.

However, they raise questions of immature tsetse immunological potential and field transmission potential that their experiments do not address. Specifically, they do not show that teneral tsetse flies are immunocompromised, that tsetse flies must be immunocompromised for Slender infection nor that younger teneral tsetse infection is not pertinent to field transmission.

Strengths:

Experimental Design is precise and elegant, outcomes are convincing. Discussion is compelling and important to the field. This is a timely piece that adds important data to a critical discussion of host: parasite interactions, of relevance to all parasite transmission.

Thank you

Weaknesses:

As above, the authors dispute the biological relevance of teneral tsetse infection in the wild, without offering evidence to the contrary. Statements need to be softened for claims regarding immunological competence or relevance to field transmission.

We have modified the revised version to soften these claims (l.156 and l.159). Please, note that the limited immunocompetence of teneral flies has been extensively studied by the labs of S. Aksoy at Yale and M. Lehane at Liverpool. In the discussion, we provide key references from these two labs 18-21. Our comment on the relevance to field transmission is simply based on field observations of the fly biology.

Reviewer #2:

Summary:

Contrary to findings recently reported by Schuster S et al., this short paper shows evidence that the stumpy form of T. brucei is probably the most pre-adapted form to progress with the life cycle of this parasite in the tsetse vector.

Strengths:

One of the most important pieces of experimental evidence is that they conduct all fly infection experiments in the absence of metabolites like GlcNAc or S-glutathione; by doing so, the infection rates in flies infected with slender trypanosomes seem very low or non-existent. This, on its own, is a piece of important experimental evidence that the Schuster S et al findings may need to be revisited.

Thank you

Weaknesses:

I consider that the authors should have included their own experiments demonstrating that the addition of these chemicals enhances the infection rates in flies receiving bloodmeals containing slender trypanosomes.

The main purpose of this study is to assess the intrinsic infectivity of SL Vs. ST in teneral Vs. adult flies, not to reproduce the results obtained by Schuster et al.. We think that the suggested experiment is not necessary as L-Glutathion is well-known to enhance infection rates by reducing the fly immune response efficiency (Ref 24). Most of the experimental infections with procyclic or ST forms (even at low densities) published by our lab and others, especially for studying parasite stages in the salivary glands, were actually performed by complementing the infective meal with L-Glutathion for this reason.

Reviewer #3:

The dogma in the Trypanosome field is that transmission by Tsetse flies is ensured by stumpy forms. This has been recently challenged by the Engstler lab (Schuster et al.), which showed that slender forms can also be transmitted by teneral flies. In this work, the authors aimed to test whether transmission by slender forms is possible and frequent.

For this, the authors repeated Tsetse transmission experiments but with some key critical differences relative to Schuster et al. First, they infected teneral and adult flies. Second, their infective meals lacked two components (N-acetylglucosamine and glutathione), which could have boosted the infection rates in the Schuster et al. work. In these conditions, the authors observed that most stumpy form infections with teneral and adult flies were successful while only 1 out of 24 slender-form infections was successful. Adult flies showed a lower infection rate, which is probably because their immune system is more developed.

Given that in Tsetse-infested areas most transmission is likely ensured by adult flies, the authors conclude that the parasite stage that will have a significant epidemiologic impact on transmission is the stumpy form.

Strengths:

• This work tackles an important question in the field.

• The Rotureau laboratory has well-known expertise in Tsetse fly transmission experiments.

• Experimental setup is robust and data is solid.

• The paper is concise and clearly written.

Thank you

Weaknesses:

• The reason(s) for why this work has lower infection rates with slender forms than Schuster et al. remain unknown. The authors suggested it could be because of the absence of N-acetylglucosamine and/or glutathione, but this was not formally tested. Could another source of variation be the clone of EATRO1125 AnTat1.1 (Paris versus Munich origin)? To reduce the workload, such additional experiments could be done with just one dose of parasites.

Differences between the strain clones, the cell culture conditions and/or the fly colony maintenance conditions could indeed explain the differences in infection rates observed in the two studies. However, the main purpose of this study is to assess the intrinsic infectivity of SL Vs. ST in teneral Vs. adult flies. Our study was designed to stand alone for providing a clear answer to this question, not to reproduce the results obtained by Schuster et al.. Hence, we don’t think that any additional experiments are required here.

• The characterization of what is slender and stumpy is critical. The authors used PAD1 protein expression as the sole reporter. While this is a robust assay to confirm stumpy, an analysis of the cell cycle would have been helpful to confirm that slender forms have not initiated differentiation (Larcombe S et al. 2023, preprint).

In this study, ST are indeed defined by their general morphology and by the expression of PAD1 proteins at the cell membrane as assessed by IFA. This is the simplest and most accurate ST proxy accessible by IFA. We do not think that monitoring in more details the cell cycle would provide key information here. If some SL forms had initiated differentiation in our experiments, then, the low infection rates observed with SL would have reinforced the fact that mostly mature PAD1+ ST are infectious for flies .

• Statistical analysis is missing. Is the difference between adult and teneral infections statistically significant?

An ANOVA statistical analysis was performed and a dedicated section was added to the revised version.

For all conditions, MG infection rate comparisons between adult and teneral flies were statistically significant.

Recommenda8ons for the authors:

Reviewer #1:

While some perceived outcomes pertaining to immunological competence and transmission relevance of teneral flies are overstated, the overall tone of the paper is inappropriately apologe7c. The authors obviously don't want to offend their colleagues but the current wri7ng style obscures meaning, making the paper a bit 'flowery' and difficult to read.

Ngoune et al. have important outcomes that need to be stated more directly.

Words such as 'unequivocally' are not appropriate to Schuster et al's outcomes. As your study shows, their findings are experimentally based, with inherent caveats, and are therefore sugges7ve, not demonstrated or proven.

The word 'unequivocally' has been removed from the revision.

Reviewer #3:

The Engstler lab cul7vates AntTaT1.1 in methylcellulose (Munich clone, if I am not mistaken). The Rotureau lab uses the Paris AntTaT1.1 clone and uses no methylcellulose. Given that methylcellulose helps stumpy forma7on, it seems important to show that the results of this paper are reproducible with the Munich clone grown in the presence of methylcellulose.

Differences between the strain clones and culture conditions could indeed explain the differences in infection rates observed in the two studies. However, the main purpose of this study is to assess the intrinsic infectivity of SL Vs. ST in teneral Vs. adult flies. Our study was designed to stand alone for providing a clear answer to this question, not to reproduce the results obtained by Schuster et al.. Hence, we don’t think that any additional experiments are required here.

Author Response

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

Summary of the reviewers’ discussion:

• The development of MSI-1 as a post-transcriptional regulator of gene expression in Escherichia coli represents a valuable addition to the synthetic biology toolkit. MSI-1 has advantages over transcriptional regulators because it has the potential to target single genes in operons. Allosteric control of MSI-1 by oleic acid increases its versatility.

Authors’ response: We thank the reviewers and editor for this evaluation.

• We recommend that authors add experiments to test the mechanism of regulation by MSI-1 or soften their claims about translational regulation. We also recommend that the authors expand their discussion of other natural and synthetic regulatory systems that target translation.

Authors’ response: In this revision, we have added new experimental results from RT-qPCR, bulk fluorometry, and flow cytometry assays to further support our conclusions. We have also enlarged the Introduction and Discussion.

• Adding an experiment to quantify the effect of oleic acid with the most strongly regulated reporter construct (i.e., flow cytometry with redesign-3) would substantially increase the impact of the work.

Authors’ response: We have done this experimental quantification (see the new Fig. 5d).

Reviewer #1 (Public Review):

The authors develop reporter constructs in E. coli where gene expression, presumably translation, is repressed by MSI-1. This is a potentially useful tool for synthetic biologists, with the advantage over transcriptional regulation that one gene in an operon could be targeted. That being said, an important caveat of translational regulation that is not addressed in the manuscript is the potential for downstream effects on RNA stability and/or transcription termination. The authors' MSI-1-regulated reporter constructs could also be useful for mechanistic studies of MSI-1.

Authors’ response: We thank the reviewer for such appreciation of our work. Regarding the potential effects on RNA stability or transcription termination, we would like to highlight our results with the sfGFP-mScarlet bicistron (Fig. 6c), showing the specific regulation of sfGFP by MSI-1* and not of mScarlet. Anyway, for this revision we have conducted an RT-qPCR experiment to quantify the mRNA level of sfGFP to further support our conclusions (see the new Fig. S2).

The author's initial construct design led to only weak regulation by MSI-1, presumably because the MSI-1 binding sites were not suitably positioned to repress translation initiation. A more rationally designed construct led to considerably greater repression. One weakness of the paper is that the authors did not use their redesigned construct that is more strongly repressed to demonstrate allosteric regulation by oleic acid using a comparable assay (e.g., flow cytometry) to that used in other experiments. The potential for allosteric regulation is a major strength of the MSI-1 system, so this is a significant gap. Similarly, the authors use the weakly regulated constructs to assess the effect of MSI-1 binding site mutations and for their mathematical modeling; these experiments would be better suited to the more strongly regulated construct.

Authors’ response: For this revision, we have performed the flow cytometric quantification of the allosteric regulation by oleic acid in the redesigned-3 system (see the new Fig. 5d). Regarding the kinetic study, we focused on the reporter system with just one recognition motif for simplicity. A reporter system with two recognition motifs, thereby recruiting two different proteins, increases the complexity to distill the effect of point mutations.

Reviewer #1 (Recommendations For The Authors):

1. Figure 5. Panels c-f look at colonies on plates, with numbers from these data being difficult to compare with either the bulk fluorescence or single-cell fluorescence values shown in other figures. Supplementary Figure 8 shows data for single cells; these data would be more appropriate in Figure 5, with the plate-based data moving to the supplement. Moreover, measuring the effect of oleic acid on the redesign-3 reporter using flow cytometry would assess the impact of oleic acid on the most strongly regulated reporter; this would be the most impactful analysis.

Authors’ response: We have redone Fig. 5 to include flow cytometry data (also for the system implemented with the redesign-3 reporter).

1. Paragraph starting line 438. The authors should briefly discuss the potential for translational repression leading to reduced RNA stability, and in the case of rapid repression that impacts transcription-coupled translation, its impact on Rho-dependent transcription termination. These factors could alter the expression of neighboring genes.

Authors’ response: As we have shown with the RT-qPCR experiment, the mRNA level of the target gene does not change in response to protein binding. We agree that mRNA stability could potentially be changed by using other RNA-targeting proteins. But in our view, a reduction of RNA stability is not a regulation of translation. We have added the following sentence in the Discussion: “The additional use of RNA-binding proteins able to alter mRNA stability might lead to the implementation of more complex circuits at the posttranscriptional level.”

1. Figure 1. It would be informative to include a control where cells have an empty plasmid rather than a plasmid expressing MSI-1, to address leakiness of MSI-1 expression.

Authors’ response: We have constructed a void plasmid as suggested and performed new bulk fluorometry assays. The new Fig. S8 shows the tight control of MSI-1* expression with the PLlac promoter. No apparent leakage is observed.

1. Line 132. Where were the two sequences positioned with respect to each other than the start codon? It would be helpful to show the sequence in Figure 1.

Authors’ response: The precise sequence is shown in the inset of Fig. 1b. The motif is placed just after the start codon.

1. Line 135. The authors envisioned repression mechanism isn't clear from the text, specifically the meaning of "block the progression" and "initial phase". As far as I know, there is no precedent for RNA-binding proteins repressing translation in bacteria by preventing translation elongation. Presumably, repression in the context described here would be due to MSI-1 binding over the ribosome-binding site, although the predicted hairpin may also occlude binding of initiating 30S ribosomes in the absence of MSI-1 binding.

Authors’ response: It is difficult to know the exact mode of action. In page 7, we have rewritten a sentence to have: “In this way, MSI-1* can repress translation by blocking the binding of the ribosome, presumably by imposing a steric hindrance for the 30S ribosomal subunit.”

1. Figure 1e is overly complicated and hence is difficult to interpret. The key result is that mScarlet expression is unchanged as a function of lactose concentration. It is sufficient to show the inset graph as a supplementary figure panel and to conclude that regulation of sfGFP is at a post-transcriptional level. Similarly, the inset in Figure 4b is unnecessary.

Authors’ response: The inset of Fig. 1e shows that the growth rate of the cells is almost constant when lactose varies. A change in growth rate will affect protein expression. The use of a two-reporter system, one regulated translationally and the other not, is instrumental to extract from fluorescence data estimates of transcription and translation rates. Of course, showing that mScarlet expression is almost constant when lactose varies would be sufficient, but we believe that performing a fine treatment of the data helps to better understand the regulatory system from a mathematical and mechanistic point of view. Therefore, despite increasing the complexity of the figure, we prefer to keep the representation of the Crick spaces (following Alon’s terminology, see our ref. 32). We have tried to carefully explain Fig. 1e in the text.

1. Figure 1f and Figure 4c would be easier to interpret as two-dimensional plots.

Authors’ response: We decided to use 3D plots to have more compact representations of the data in the main figures. The accompanying insets show the percentage of cells above the threshold, which helps to understand the regulatory effects. In any case, we have provided the corresponding 2D plots in Fig. S10.

1. I don't think Figure 2e is relevant. The key result is shown in Figure 2f, i.e., the effect of mutations on regulation by MSI-1.

Authors’ response: We agree with the reviewer that the key result is shown in panel f. However, we prefer to keep panel e in Fig. 2 because, even if negative, this result may incite further research. In addition, we avoid the rearrangement of the whole figure.

1. Lines 311-313. Without additional evidence that the mutants are toxic, I suggest removing this text.

Authors’ response: As suggested, we have removed that claim.

Reviewer #2 (Public Review):

Summary:

Dolcemascolo and colleagues describe the use of the mammalian RNA-binding protein Musashi-1 (MSI-1) to implement translational regulation systems in E. coli. They perform detailed in vitro studies of MSI-1 and its binding to different RNA sequences. They provide compelling evidence of the effectiveness of the regulatory system in multiple circuits using different mRNA sequence motifs. They harness allosteric inhibition of MSI-1 by omega-9 monounsaturated fatty acids to demonstrate a fatty-acid-responsive circuit in E. coli.

Strengths:

The experimental results are compelling and the characterization of the binding between MSI-1 and different RNA sequences is thorough and performed via multiple complementary techniques. Several new useful circuit components are demonstrated.

Authors’ response: We thank the reviewer for such appreciation of our work.

Weaknesses:

MSI-1 provides 8.6-fold downregulation of sfGFP with an optimized mRNA sequence. In some applications, a larger degree of repression may be required.

Authors’ response: We agree with the reviewer in this point. We expect to conduct further research in the future to optimize the dynamic range of the system. We have added the following sentence in the Discussion: “Further work should be conducted to enhance the fold change of the regulatory module and engineer complex circuits with it.”

Reviewer #2 (Recommendations For The Authors):

Overall, I think this paper is very well done and quite thorough. I only have minor suggestions:

• For Figures 1f and 4c, it is quite hard to interpret the fraction of cells above the threshold with the 3d perspective. It would be clearer to use a more standard 2d plot where the histograms are offset along the y-axis and the threshold is indicated by a vertical line.

Authors’ response: We decided to use 3D plots to have more compact representations of the data in the main figures. The accompanying insets show the percentage of cells above the threshold, which helps to understand the regulatory effects. In any case, we have provided the corresponding 2D plots in Fig. S10.

• For Figure 4b, the highlighting of different sequence regions in red3 appears to be offset by one base (e.g. AAU is highlighted rather than AUG).

Authors’ response: This has been corrected.

• For line 504, it seems that MSI-1 is used for two different proteins. A different name should be assigned to this 200-residue protein to avoid confusion with the other MSI-1.

Authors’ response: We now use the term MSI-1h* for the human version of the protein.

• The note (Page S12) that A_0 + A_R = alpha/delta only applies in steady-state conditions, which should be stated.

Authors’ response: We have specified that.

• It seems that some authors work for the companies that sell some of the instruments/consumables used for the assays, specifically switchSENSE and LigandTracer. This may be something that should be declared under Competing Interests for the paper.

Authors’ response: We are sorry for having missed this point. We have included a Competing Interests section to state that “RAHR and WFV work for Dynamic Biosensors. GPR and JB work for Ridgeview Instruments”.

Reviewer #3 (Public Review):

Summary:

In this work, the authors co-opt the RRM-binding protein Musashi-1 to act as a translational repressor. The novelty of the work is in the adoption of the allosteric RRM protein Musashi-1 into a translational reporter and the demonstration that RRM proteins, which are ubiquitous in eukaryotic systems, but rare in prokaryotic ones, may act effectively as post-translational regulators in E. coli. The extent of repression achieved by the best design presented in this work is not substantially improved compared to other synthetic regulatory schemes developed for E. coli, even those that similarly regulate translation (eg. native PP7 repression is approximately 10-fold, Lim et al. J. Biol. Chem. 2001 276:22507-22513). Furthermore, the mechanism of regulation is not established due to missing key experiments. The work would be of broader interest if the allosteric properties of Musashi-1 were more effective in the context of regulation. Unfortunately, the authors do not demonstrate that fatty acids can completely de-repress expression in the experimental system used for most of their assays, nor do they use this ability in their provided application (NIMPLY gate).

Authors’ response: For this revision, we have performed the flow cytometric quantification of the allosteric regulation by oleic acid in the redesigned-3 system, showing substantial de-repression of the system with the biochemical compound. We have redone Fig. 5 and modified the Results section accordingly. Aligned with the reviewers and editor, we believe that this new result helps to improve our manuscript.

Strengths:

The first major achievement of this work is the demonstration that a eukaryotic RRM protein may be used to posttranscriptionally regulate expression in bacteria. In my limited literature search, this appears to be the first engineering attempt to design an RBP to directly regulate translation in E. coli, although engineered control of translation via other approaches including alterations to RNA structure or via trans-acting sRNAs have been previously described (for review see Vigar and Wieden Biochim Biophys. Acta Gen. Subj. 2017, 1861:3060-3069). Additionally, several viral systems (e.g. MS2 and PP7) have been directly co-opted to work in a similar fashion in the past (utilized recently in Nguyen et al. ACS Synthetic Biol 2022, 11:1710-1718).

Authors’ response: We thank the reviewer for such appreciation of our work.

The second achievement of this work is the demonstration that the allosteric regulation of Musashi-1 binding can be utilized to modulate the regulatory activity. However, the liquid culture demonstration (Suppl. Fig 8) shows that this is not a very effective switch, with de-repressed reporter activity showing substantial change but not approaching un-repressed activity. This effect is stronger when colonies are grown on a solid medium (Fig. 5).

Authors’ response: As we have previously indicated, the flow cytometric quantification of the allosteric regulation by oleic acid in the redesigned-3 system in liquid culture showed substantial de-repression with the biochemical compound. It is now stated in the text the following: “Nevertheless, the system implemented with the redesign-3 reporter displayed a better dynamic behavior in response to lactose and oleic acid. In particular, the percentage of cells in the ON state increased from 0 (with 1 mM lactose) to 71% upon addition of 20 mM oleic acid (Fig. 5d).” This new result helps to improve our manuscript.

Weaknesses:

In this work, the authors codon optimize the mouse Musashi-1 coding sequence for expression in E. coli and demonstrate using an sfGFP reporter that an engineered Musashi-1 binding site near the translational start site is sufficient to enable a modest reduction in reporter gene expression. The authors postulate that the reduction in expression due to inhibition of ribosome translocation along the transcript (lines 134/135), as an expression of a control transcript (mScarlet) driven by the same promoter (Plac) but without the Musashi-1 recognition site does not demonstrate the same repression. However, the situation could be more complex. Other possibilities include inhibition of translation initiation rather than elongation, as well as accelerated mRNA decay of transcripts that are not actively translated. The authors do not present any measurements of sfGFP mRNA levels.

Authors’ response: In page 7, we have rewritten a sentence to have: “In this way, MSI-1* can repress translation by blocking the binding of the ribosome, presumably by imposing a steric hindrance for the 30S ribosomal subunit.” In addition, for this revision we have conducted an RT-qPCR experiment to quantify the mRNA level of sfGFP to further support our conclusions (see the new Fig. S2). As shown, there is no change in the mRNA level upon inducing the system with lactose.

In subsequent sections of the work, the authors create a series of point mutations to assess RNA-protein binding and assess these via both a sfGFP reporter and in vitro binding assays (switchSENSE). Ultimately, it is difficult to fully rationalize and interpret the behavior of these mutants in the context provided. The authors do identify a relationship between equilibrium constant (1/KD) and fold-repression. However, it is not clear from the narrative why this relationship should exist. Fold-repression is one measure of regulator efficacy, but it is an indirect measure determined from unrepressed and repressed expression. It is not clear why unrepressed expression (in the absence of the protein) is expected to be a function of the equilibrium constant.

Authors’ response: A mathematical derivation from mass action kinetics on why the fold change scales with 1/KD is provided in Note S2. It is the ratio between the unrepressed and repressed expression (i.e., fold change) what scales with 1/KD, but not the expression of a particular state. This kind of relationship has been previously established in the case of transcription regulation [see e.g. Garcia & Phillips, PNAS (2011), our ref. 39]. Our mathematical modeling results expand previous work by providing a single picture from which to analyze transcription and translation regulation.

Subsequent rational redesign of the Musashi-1 binding sequence to produce three alternative designs shows that fold-repression may be improved to approximately 8.6-fold. However, the rationalization of why the best design (red3) achieves this increase based on either the extensive modelling or in vitro measured binding constants is not well articulated. Furthermore, this extent of regulation is approximately that which can be achieved from the PP7 system with its native components (Lim et al. J. Biol. Chem. 2001 276:22507-22513).

Authors’ response: In the case of translation control, the regulation is more challenging because the target is quickly degraded, especially in bacteria (in contrast to transcription control, where the target is stable). This is acknowledged in the manuscript. Even though, it is possible to engineer synthetic circuits with sRNAs or RNA-binding proteins with sufficient dynamic range. We expect to conduct further research in the future to optimize the dynamic range of the system. We have added the following sentence in the Discussion: “Further work should be conducted to enhance the fold change of the regulatory module and engineer complex circuits with it.” Regarding the articulation of the results for the mutants and mathematical model, see our responses in the following questions.

The application provided for this regulator (NIMPLY gate), is not an inherently novel regulatory paradigm, and it does not capitalize on the allosteric properties of Musashi-1, but rather treats Musashi-1 as a non-allosteric component of a regulatory circuit.

Authors’ response: The NIMPLY gate refers to lactose and aTC as inputs. Considering oleic acid as an additional input will lead to a more complex logic. In the last Results section, we wanted to show that the post-transcriptional mechanism engineered with Musashi-1 can be useful specifically regulate a gene within an operon, to implement combinatorial regulation (i.e., coupling transcription and translation control), and to reduce protein expression noise. To these ends, the allosteric ability of the Musashi-1 was not so determinant. In this regard, it would be true that such fine regulatory effects might be achieved as well with non-allosteric RNA-binding proteins, such as MS2CP or PP7CP.

Reviewer #3 (Recommendations For The Authors):

1. In the introduction the authors should adequately address the native bacterial mechanisms that allow posttranscriptional regulation in bacteria as well as better discuss previous examples of translational repressors.

Authors’ response: We have added the following paragraph in the Introduction: “Even though bacteria do not appear to exploit proteins to regulate translation in a gene-specific manner, it is worth noting that some bacteriophages do follow this mechanism to modulate their infection cycle. These are the cases, e.g., of the coat proteins of the phages MS2 (infecting Escherichia coli) or PP7 (infecting Pseudomonas aeruginosa), which regulate the expression of the cognate phage replicases through protein-RNA interactions [18]. However, one limitation for synthetic biology developments is that such phage proteins are not allosteric. At the post-transcriptional level, bacteria mostly rely on a large palette of cis- and trans-acting non-coding RNAs to either activate or repress protein expression, resulting in the regulation of translation initiation, mRNA stability, or transcription termination, and even allowing sensing small molecules [1,15]. Thus, there should be efforts to replicate this functional versatility with proteins in bacteria.”

1. Given the location of the Musashi-1 binding site in the sfGFP reporter, it may be blocking translation initiation, rather than blocking the progression of the ribosome once attached (line 134/135). The schematic in Fig 1a. is also not overly clear in describing the differences in mechanisms between eukaryotic and prokaryotic systems described in the text.

Authors’ response: In page 7, we have rewritten a sentence to have: “In this way, MSI-1 can repress translation by blocking the binding of the ribosome, presumably by imposing a steric hindrance for the 30S ribosomal subunit.” In page 14, we have added the following sentence: “In this way, MSI-1 can also block the RNA component of the 30S ribosomal subunit.”

1. The authors did not directly examine mRNA levels of their reporter to establish translational regulation. In many cases, inhibition of translation is accompanied by an increased degradation rate in bacterial systems. The authors do not seem to recognize this as a possible amplifier in their system, relying exclusively on normalization via another transcript produced from the same promoter (mScarlet).

Authors’ response: For this revision we have conducted an RT-qPCR experiment to quantify the mRNA level of sfGFP to further support our conclusions (see the new Fig. S2). As shown, there is no change in the mRNA level upon inducing the system with lactose.

1. The results presented for mutations 1-5 are not consistent with the author's models for what is occurring. In particular, mutant 1 displays a reduction in reporter production in the absence of Musashi-1, but the production in the presence does not change from the unaltered sequence. The claim that mutation 1 (in the UAG binding site) results in less binding and ultimately in less regulation is not substantiated since this loss of regulation is due to a reduction in unrepressed expression rather than an increase in expression when Musashi-1 is present.

Authors’ response: We respectfully disagree with this appreciation. In the case of mutant 1, if the Musashi protein recognized the target mRNA with the same affinity as in the original scenario, the red bar would be much lower. Because the Musashi protein hardly recognizes the mutant-1 mRNA, the blue and red bars are quite similar. To clarify this point, we have added the following text in the manuscript: “Despite that mutation substantially reduced sfGFP expression in absence of MSI-1*, the presumed repressed state upon addition of lactose did not change much, suggesting the difficulty of the protein for targeting the mutated mRNA.”

1. Given point 5 above, it is not clear to me why one would expect the 1/KD to be predictive fold-repression in the presence and absence of the repressor. I would rather see the relationship described as predictive in Fig. 2f (fold change vs. 1/KD) rather than the non-linear relationship. It is difficult to qualitatively evaluate the fit quality with the way the data are currently presented.

Authors’ response: Note S2 provides a mathematical derivation from mass action kinetics on why the fold change scales with 1/KD. The R2 value that we provide for the fitting corresponds to the linear regression between fold and 1/KD, as specified in the figure legend. However, we think that the representation of fold vs. KD in log scale is more illustrative in this case.

1. It is not clear what conclusion is determined from the computational modeling, or how this work contributes to the narrative presented. It does not seem like what is learned from these experiments is utilized for novel designs. Furthermore, several of the assumptions within the model may be problematic including the high rate of "elongation leakage" described and the lack of justification for RNA degradation rates utilized.

Authors’ response: The mathematical modeling was performed to rationalize our experimental data. Our idea was more to recapitulate the observed dynamics than to guide the design of new systems. Our model might be exploited to this end in further research, as the reviewer suggests. Besides, elongation leakage is a concept that applies to both transcription and translation regulation systems, and it is not more than the ability of the RNA polymerase or ribosome to elongate even if there is a protein bound to the nucleic acid. This parameter can be set to 0 in the model if appropriate. Moreover, we cite the paper by Bernstein et al., PNAS (2002), our ref. 38, to justify that in E. coli the average mRNA half-life is about 5 min (i.e., degradation rate of 0.14 min-1).

1. The data presented in Figure 4 are not presented in a consistent way. While it would be somewhat redundant, including the 0 and 1 mM lactose data for red3 in Figure 4a would be helpful for comparison purposes.

Authors’ response: We have added the requested bar plot in Fig. 4a.

1. The presence of additional Musashi-1 sites upstream of the start codon in red3, and their impact on impact on the fold-repression may support an inhibition of the translation initiation model rather than an inhibition of elongation.

Authors’ response: In page 7, we have rewritten a sentence to have: “In this way, MSI-1 can repress translation by blocking the binding of the ribosome, presumably by imposing a steric hindrance for the 30S ribosomal subunit.” In page 14, we have added the following sentence: “In this way, MSI-1 can also block the RNA component of the 30S ribosomal subunit.”

Author Response

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The authors aim to address a critical challenge in the field of bioinformatics: the accurate and efficient identification of protein binding sites from sequences. Their work seeks to overcome the limitations of current methods, which largely depend on multiple sequence alignments or experimental protein structures, by introducing GPSite, a multi-task network designed to predict binding residues of various molecules on proteins using ESMFold.

Strengths:

1. Benchmarking. The authors provide a comprehensive benchmark against multiple methods, showcasing the performances of a large number of methods in various scenarios.

2. Accessibility and Ease of Use. GPSite is highlighted as a freely accessible tool with user-friendly features on its website, enhancing its potential for widespread adoption in the research community.

We thank the reviewer for acknowledging the contributions and strengths of our work! Weaknesses:

1. Lack of Novelty. The method primarily combines existing approaches and lacks significant technical innovation. This raises concerns about the original contribution of the work in terms of methodological development. Moreover, the paper reproduces results and analyses already presented in previous literature, without providing novel analysis or interpretation. This further diminishes the contribution of this paper to advancing knowledge in the field.

The novelty of this work is primarily manifested in four key aspects. Firstly, although we agree with the reviewer that we did employ several existing tools such as ProtTrans and ESMFold to extract sequence features and predict protein conformations, these techniques were hardly explored in the field of binding site prediction. We have successfully demonstrated the feasibility of substituting multiple sequence alignments with language model embeddings and training with “less accurate” predictive structures, providing a new solution to overcome the limitations of current methods for genome-wide applications. Secondly, though a few methods tend to capture geometric information based on protein surfaces or atom graphs, surface calculation and property mapping are usually time-consuming, while massage passing on full atom graphs is memory-consuming and thus challenging to process long sequences. Besides, these methods are sensitive towards details and errors in the predictive structures. To facilitate large-scale annotations, we have innovatively applied geometric deep learning to protein residue graphs for comprehensively capturing backbone and sidechain geometric contexts in an efficient and effective manner (Figure 1). Thirdly, we have not only exploited multi-task learning to integrate diverse ligands and enhance performance, but also shown its capability to easily extend to the binding site prediction of other unseen ligands (Figure 4 D-E). Last but not least, as a Tools and Resources article, we have provided a fast, accurate and user-friendly webserver, as well as constructed a large annotation database for the sequences in Swiss-Prot. Leveraging this database, we have conducted extensive analyses on the associations between binding sites and molecular functions, biological processes, and disease-causing mutations (Figure 5), indicating the potential of our tool to unveil unexplored biology underlying genomic data.

1. Benchmark Discrepancies. The variation in benchmark results, especially between initial comparisons and those with PeSTo. GPSite achieves a PR AUC of 0.484 on the global benchmark but a PR AUC of 0.61 on the benchmark against PeSTo. For consistency, PeSTo should be included in the benchmark against all other methods. It suggests potential issues with the benchmark set or the stability of the method. This inconsistency needs to be addressed to validate the reliability of the results.

We thank the reviewer for the constructive comments. Since our performance comparison experiments involved numerous competitive methods whose training sets were disparate, it was difficult to compare or rank all these methods fairly using a single test set. As described in the “GPSite outperforms state-of-the-art methods” section, 358 out of 375 proteins in our protein-protein binding site test set share >30% sequence identity with the training sequences of PeSTo. To address this, we meticulously re-split our entire protein-protein binding site dataset to generate a new test set that avoids any overlap with the training sets of both GPSite and PeSTo and performed a separate evaluation. This is quite common in this field. For instance, in the study of PeSTo [Nat Commun 2023], the comparisons of PeSTo with MaSIF-site, SPPIDER, and PSIVER were conducted using one test set, while the comparison with ScanNet was performed on a separate test set. Based on the reviewer’s suggestion, in the revised version of the manuscript, we intend to include other comparative methods alongside PeSTo on the new test set or retrain our model directly on PeSTo's training set for comparison, which should enhance the completeness of our results.

1. Interface Definition Ambiguity. There is a lack of clarity in defining the interface for the binding site predictions. Different methods are trained using varying criteria (surfaces in MaSIF-site, distance thresholds in ScanNet). The authors do not adequately address how GPSite's definition aligns with or differs from these standards and how this issue was addressed. It could indicate that the comparison of those methods is unreliable and unfair.

We thank the reviewer for the comments. The precise definition of ligand-binding sites is elucidated in the “Benchmark datasets” section. Specifically, the datasets of DNA, RNA, peptide, ATP, HEM and metal ions used to train GPSite were collected from the widely acknowledged BioLiP database [PMID: 23087378]. In BioLiP, a binding residue is defined if the smallest atomic distance between the target residue and the ligand is <0.5 Å plus the sum of the Van der Waal’s radius of the two nearest atoms. In the meanwhile, most comparative methods regarding these ligands were also trained on data from BioLiP, thereby ensuring fair comparisons.

However, since BioLiP does not include data on protein-protein binding sites, studies for protein-protein binding site prediction may adopt slightly distinct label definitions, as the reviewer suggested. Here, we employed protein-protein binding site data from our previous study [PMID: 34498061], where a protein-binding residue was defined as a surface residue (relative solvent accessibility > 5%) that lost more than 1 Å2 absolute solvent accessibility after protein-protein complex formation. This definition was initially introduced in PSIVER [PMID: 20529890] and widely applied in various studies (e.g., PMID: 31593229, PMID: 32840562). SPPIDER [PMID: 17152079] and MaSIF-site [PMID: 31819266] have also adopted similar surface-based definitions as PSIVER. On the other hand, ScanNet [PMID: 35637310] employed an atom distance threshold of 4 Å to define contacts while PeSTo [PMID: 37072397] used a threshold of 5 Å. However, it is noteworthy that current methods in this field including ScanNet [Nat Methods 2022] and PeSTo [Nat Commun 2023] directly compared methods using different label definitions without any alignment in their benchmark studies, likely due to the subtle distinctions among these definitions. For instance, the study of PeSTo directly performed comparisons with ScanNet, MaSIF-site, SPPIDER, and PSIVER. Therefore, we followed these previous works, directly comparing GPSite with other protein-protein binding site predictors. In our revised manuscript, we will provide more details for the binding site definitions to avoid any potential ambiguity.

While GPSite demonstrates the potential to surpass state-of-the-art methods in protein binding site prediction, the evidence supporting these claims seems incomplete. The lack of methodological novelty and the unresolved questions in benchmark consistency and interface definition somewhat undermine the confidence in the results. Therefore, it's not entirely clear if the authors have fully achieved their aims as outlined.

The work is useful for the field, especially in disease mechanism elucidation and novel drug design. The availability of genome-scale binding residue annotations GPSite offers is a significant advancement. However, the utility of this tool could be hampered by the aforementioned weaknesses unless they are adequately addressed.

We thank the reviewer for acknowledging the advancement and value of our work, as well as pointing out areas where improvements can be made. As discussed above, we will carry out the corresponding revisions in the next version of the manuscript to enhance the completeness and clearness of our work.

Reviewer #2 (Public Review):

Summary:

This work provides a new framework, "GPsite" to predict DNA, RNA, peptide, protein, ATP, HEM, and metal ions binding sites on proteins. This framework comes with a webserver and a database of annotations. The core of the model is a Geometric featurizer neural network that predicts the binding sites of a protein. One major contribution of the authors is the fact that they feed this neural network with predicted structure from ESMFold for training and prediction (instead of native structure in similar works) and a high-quality protein Language Model representation. The other major contribution is that it provides the public with a new light framework to predict protein-ligand interactions for a broad range of ligands.

The authors have demonstrated the interest of their framework with mostly two techniques: ablation and benchmark.

Strengths:

The performance of this framework as well as the provided dataset and web server make it useful to conduct studies.

The ablations of some core elements of the method, such as the protein Language Model part, or the input structure are very insightful and can help convince the reader that every part of the framework is necessary. This could also guide further developments in the field. As such, the presentation of this part of the work can hold a more critical place in this work.

We thank the reviewer for recognizing the contributions of our work and for noting that our experiments are thorough.

Weaknesses:

Overall, we can acknowledge the important effort of the authors to compare their work to other similar frameworks. Yet, the lack of homogeneity of training methods and data from one work to the other makes the comparison slightly unconvincing, as the authors pointed out. Overall, the paper puts significant effort into convincing the reader that the method is beating the state of the art. Maybe, there are other aspects that could be more interesting to insist on (usability, interest in protein engineering, and theoretical works).

We sincerely appreciate the reviewer for the constructive and insightful comments. As to the concern of training data heterogeneity raised by the reviewer, it is noteworthy that current studies in this field, such as ScanNet [Nat Methods 2022] and PeSTo [Nat Commun 2023], tend to directly compare methods trained on different datasets in their benchmark experiments. Therefore, we have adhered to the paradigm in these previous works. According to the detailed recommendations by the reviewer, we will improve our manuscript by incorporating additional ablation studies regarding the effects of predicted structures and language model representations. Besides, we will refine the Discussion section to focus more on the achievements of this work and its potential applications including protein engineering. A comprehensive point-by-point response to the reviewer’s recommendations will be provided alongside the revised manuscript. This will ensure that all concerns and suggestions are adequately addressed.

Reviewer #3 (Public Review):

Summary

The authors of this work aim to address the challenge of accurately and efficiently identifying protein binding sites from sequences. They recognize that the limitations of current methods, including reliance on multiple sequence alignments or experimental protein structure, and the under-explored geometry of the structure, which limit the performance and genome-scale applications. The authors have developed a multi-task network called GPSite that predicts binding residues for a range of biologically relevant molecules, including DNA, RNA, peptides, proteins, ATP, HEM, and metal ions, using a combination of sequence embeddings from protein language models and ESMFold-predicted structures. Their approach attempts to extract residual and relational geometric contexts in an end-to-end manner, surpassing current sequence-based and structure-based methods.

Strengths

1. The GPSite model's ability to predict binding sites for a wide variety of molecules, including DNA, RNA, peptides, and various metal ions.

2. Based on the presented results, GPSite outperforms state-of-the-art methods in several benchmark datasets.

3. GPSite adopts predicted structures instead of native structures as input, enabling the model to be applied to a wider range of scenarios where native structures are rare.

4. The authors emphasize the low computational cost of GPSite, which enables rapid genome-scale binding residue annotations, indicating the model's potential for large-scale applications.

We thank the reviewer for recognizing the significance and value of our work!

Weaknesses

1. One major advantage of GPSite, as claimed by the authors, is its efficiency. Although the manuscript mentioned that the inference takes about 5 hours for all datasets, it remains unclear how much improvement GPSite can offer compared with existing methods. A more detailed benchmark comparison of running time against other methods is recommended (including the running time of different components, since some methods like GPSite use predicted structures while some use native structures).

We thank the reviewer for the valuable suggestion. Empirically, it takes about 30 min for existing MSA-based methods to make predictions for a protein with 500 residues, while it only takes less than 1 min for GPSite (including structure prediction). However, it is worth noting that some predictors in our benchmark study are solely available as webservers, and it is challenging to compare the runtime between a standalone program and a webserver due to the disparity in hardware configurations. Therefore, we will include comprehensive runtime comparisons between the GPSite webserver and other existing servers in the revision to illustrate the practicality and efficiency of our method.

1. Since the model uses predicted protein structure, the authors have conducted some studies on the effect of the predicted structure's quality. However, only the 0.7 threshold was used. A more comprehensive analysis with several different thresholds is recommended.

We thank the reviewer for the comment. We assessed the effect of the predicted structure's quality by evaluating GPSite’s performance on high-quality (TM-score > 0.7) and low-quality (TM-score ≤ 0.7) predicted structures. We did not employ multiple thresholds (e.g., 0.3, 0.5, and 0.7), as the majority of proteins in the test sets were accurately predicted by ESMFold. Specifically, as shown in Figure 3B, Appendix 3-figure 2 and Appendix 2-table 5, the numbers of proteins with TM-score ≤ 0.7 are small in most datasets. Consequently, there is insufficient data available for analysis with lower thresholds, except for the RNA test set. Notably, Figure 3C presents a detailed inspection of the proteins with TM-score < 0.5 in the RNA test set. Within this subset, GPSite consistently outperforms the state-of-the-art structure-based method GraphBind with predicted structures as input, regardless of the prediction quality of ESMFold. Only in cases where structures are predicted with extremely low quality (TM-score < 0.3) does GPSite fall behind GraphBind input with native structures. This result further demonstrates the robustness of GPSite.

1. To demonstrate the robustness of GPSite, the authors performed a case study on human GR containing two zinc fingers, where the predicted structure is not perfect. The analysis could benefit from more a detailed explanation of why the model can still infer the binding site correctly even though the input structural information is slightly off.

We thank the reviewer for the comment. We have actually explained the potential reason for the robustness of GPSite in the second paragraph of the “GPSite is robust for low-quality predicted structures” section. In summary, although the whole structure of this protein is not perfectly predicted, the binding domains of peptide, DNA and Zn2+ are actually predicted accurately as evidenced by the superpositions of the native and predicted structures in Figure 3D and 3E. Therefore, GPSite can still make reliable predictions.

1. To analyze the relatively low AUC value for protein-protein interactions, the authors claimed that it is "due to the fact that protein-protein interactions are ubiquitous in living organisms while the Swiss-Prot function annotations are incomplete", which is unjustified. It is highly recommended to support this claim by showing at least one example where GPSite's prediction is a valid binding site that is not present in the current Swiss-Prot database or via other approaches.

We thank the reviewer for the valuable recommendation. We will perform such analysis in the revised manuscript.

1. The authors reported that many GPSite-predicted binding sites are associated with known biological functions. Notably, for RNA-binding sites, there is a significantly higher proportion of translation-related binding sites. The analysis could benefit from a further investigation into this observation, such as the analyzing the percentage of such interactions in the training site. In addition, if there is sufficient data, it would also be interesting to see the cross-interaction-type performance of the proposed model, e.g., train the model on a dataset excluding specific binding sites and test its performance on that class of interactions.

We thank the reviewer for the suggestion. We would like to clarify that the analysis in Figure 5C was conducted at “protein-level” instead of “residue-level”. As described in the second paragraph of the “Large-scale binding site annotation for Swiss-Prot” section, a protein-level ligand-binding score was assigned to a protein by averaging the top k residue-level predictive binding scores. This protein-level score indicates the overall binding propensity of the protein to a specific ligand. We gathered the top 20,000 proteins with the highest protein-level binding scores for each ligand and found that their biological process annotations from Swiss-Prot were consistent with existing knowledge.

As for the cross-interaction-type performance raised by the reviewer, we will include such analysis in the revised manuscript.

Author Response

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

Response to reviewers

We would like to thank the reviewers for their feedback. Below we address their comments and have indicated the associated changes in our point-by-point response (blue: answers, red: changes in manuscript).

Reviewer #1:

Overall, the hypotheses and results are clearly presented and supported by high quality figures. The study is presented in a didactic way, making it easy for a broad audience to understand the significance of the results. The study does present some weaknesses that could easily be addressed by the authors.

We thank the reviewer for appreciating our work and providing useful suggestions for improvement.

1) First, there are some anatomical inaccuracies: line 129 and fig1C, the authors omit m.dial septum projections to area CA1 (in addition to the entorhinal cortex). Moreover, in addition to CA1, CA3 also provides monosynaptic feedback projections to the medial septum CA3. Finally, an indirect projection from CA1/3 excitatory neurons to the lateral septum, which in turn sends inhibitory projections to the medial septum could be included or mentioned by the authors. This could be of particular relevance to support claims related to effects of neurostimulations, whereby minutious implementation of anatomical data could be key.

If not updating their model, the authors could add this point to their limitation section, where they already do a good job of mentioning some limitations of using the EC as a sole oscillatory input to CA1.

We acknowledge that our current model strongly simplifies the interconnections between the medial septum and the hippocampal formation, but including more anatomical details is beyond the scope of this manuscript and would be a topic for future work. Nevertheless, we followed the reviewer’s advice to stress this point in our manuscript. First, we moved a paragraph that was initially in the “methods” section to the “results” section (L.141-150 of the revised manuscript):

“Biologically, GABAergic neurons from the medial septum project to the EC, CA3, and CA1 fields of the hippocampus (Toth et al., 1993; Hajós et al., 2004; Manseau et al., 2008; Hangya et al., 2009; Unal et al., 2015; Müller and Remy, 2018). Although the respective roles of these different projections are not fully understood, previous computational studies have suggested that the direct projection from the medial septum to CA1 is not essential for the production of theta in CA1 microcircuits (Mysin et al., 2019). Since our modeling of the medial septum is only used to generate a dynamic theta rhythm, we opted for a simplified representation where the medial septum projects only to the EC, which in turn drives the different fields of the hippocampus. In our model, Kuramoto oscillators are therefore connected to the EC neurons and they receive projections from CA1 neurons (see methods for more details).”

Second, we expanded the corresponding paragraph in the limitation section to discuss this point further (L.398-415 of the revised manuscript):

“We decided to model septal pacemaker neurons projecting to the EC as the main source of hippocampal theta as reported in multiple experimental studies (Buzsáki, 2002; Buzsáki et al., 2003; Hangya et al., 2009). However, experimental findings and previous models have also proposed that direct septal inputs are not essential for theta generation (Wang, 2002; Colgin et al., 2013; Mysin et al., 2019), but play an important role in phase synchronization of hippocampal neurons. Furthermore, the model does not account for the connections between the lateral and medial septum and the hippocampus (Takeuchi et al., 2021). These connections include the inhibitory projections from the lateral to the medial septum and the monosynaptic projections from the hippocampal CA3 field to the lateral septum. An experimental study has highlighted the importance of the lateral septum in regulating the hippocampal theta rhythm (Bender et al., 2015), an area that has not been included in the model. Specifically, theta-rhythmic optogenetic stimulation of the axonal projections from the lateral septum to the hippocampus was shown to entrain theta oscillations and lead to behavioral changes during exploration in transgenic mice. To account for these discrepancies, our model could be extended by considering more realistic connectivity patterns between the medial / lateral septum and the hippocampal formation, including glutamatergic, cholinergic, and GABAergic reciprocal connections (Müller and Remy, 2018), or by considering multiple sets of oscillators each representing one theta generator.”

1. The authors test conditions of low theta inputs, which they liken to pathological states (line 112). It is not clear what pathology the authors are referring to, especially since a large amount of 'oscillopathies' in the septohippocampal system are associated with decreased gamma/PAC, but not theta oscillations (e.g. Alzheimer's disease conditions).

In the manuscript, we referred to “oscillopathies” in a broad sense way as we did not want to overstate the biological implications of the model or the way we modeled pathological states. To our knowledge, several studies have yielded inconsistent results regarding the specific changes in theta or gamma power in Alzheimer’s disease, and the most convincing alteration seems to be the theta-gamma phase-amplitude coupling (PAC) (for review see e.g., Kitchigina, V. F. Alterations of Coherent Theta and Gamma Network Oscillations as an Early Biomarker of Temporal Lobe Epilepsy and Alzheimer’s Disease. Front Integr Neurosci 12, 36 (2018)), as also mentioned by the reviewer.

In this study, the most straightforward way to reduce theta-gamma PAC was to reduce the amplitude of the oscillators’ gain, which affected theta power, gamma power, and theta-gamma PAC (Figure 5 of the revised manuscript). Affecting their synchronization level (i.e., the order parameter) did not affect any of these variables (Figure 5 – Figure Supplement 4).

In order to alter theta-gamma PAC without affecting theta or gamma power, we believe that more complex changes should be performed in the model, likely at the level of individual neurons in the hippocampal formation. For example, cholinergic deprivation has been previously used in a multi-compartment model of the hippocampal CA3 to mimic Alzheimer’s disease and to draw functional implications on the slowing of theta oscillations and the storage of new information (Menschik, E. D. & Finkel, L. H. Neuromodulatory control of hippocampal function: towards a model of Alzheimer’s disease. Artif Intell Med 13, 99–121 (1998)).

This has now been added to the limitations section (L.458-465 of the revised manuscript):

“Finally, we likened conditions of low theta input to pathological states characteristic of oscillopathies such as Alzheimer’s disease, as these conditions disrupted all aspects of theta-gamma oscillations in our model: theta power, gamma power, and theta-gamma PAC (Figure 5). However, it should be noted that changes in theta or gamma power in these pathologies are often unclear, and that the most consistent alteration that has been reported in Alzheimer’s disease is a reduction of theta-gamma PAC (for review, see Kitchigina, 2018). Future work should explore the effects of cellular alterations intrinsic to the hippocampal formation and their impact on theta-gamma oscillations.”

1. While relevant for the clinical field, there is overall a missed opportunity to explain many experimental accounts with this novel model. Although to this day, clinical use of DBS is mostly restricted to electrical (and thus cell-type agnostic) stimulation, recent studies focusing on mechanisms of neurostimulations have manipulated specific subtypes in the medial septum and observed effects on hippocampal oscillations (e.g. see Muller & Remy, 2017 for review). Focusing stimulations in CA1 is of course relevant for clinical studies but testing mechanistic hypotheses by focusing stimulation on specific cell types could be highly informative. For instance, could the author reproduce recent optogenetic studies (e.g. Bender et al. 2015 for stimulation of fornix fibers; Etter et al., 2019 & Zutshi et al. 2018 for stimulation of septal inhibitory neurons)? Cell specific manipulations should at least be discussed by the authors.

We acknowledge the importance of cell-type-specific manipulation in the septo-hippocampal circuitry. However, our model was designed to study neurostimulation protocols that affect the hippocampal formation, not the medial septum, which is why only the hippocampal formation is composed of biophysically realistic (i.e., conductance-based) neuronal models. To replicate the various studies mentioned by the reviewer (which are all very relevant), we would need to implement a biophysical model of the medial septum, which would be an entirely new project.

Nevertheless, we can use the existing model to replicate optogenetic studies that induced gamma oscillations in excitatory-inhibitory circuits, using either ramped photostimulation targeting excitatory neurons (Adesnik et al., 2010; Akam et al., 2012; Lu et al., 2015), or pulsed stimulation driving inhibitory cells in the gamma range (Cardin et al., 2009; Iaccarino et al., 2016). In fact, such approaches have been demonstrated not just in the hippocampus but also in the neocortex, and represent a hallmark of local excitatory-inhibitory circuits. To account for these experimental results and replicate them, we have added 4 new figures (Figure 2 and its 3 figure supplements) and an extensive section in the results part (L.151-217 of the revised manuscript):

“From a conceptual point of view, our model is thus composed of excitatory-inhibitory (E-I) circuits connected in series, with a feedback loop going through a population of coupled phase oscillators. In the next sections, we first describe the generation of gamma oscillations by individual E-I circuits (Figure 2), and illustrate their behavior when driven by an oscillatory input such as theta oscillations (Figure 3). We then present a thorough characterization of the effects of theta input and stimulation amplitude on theta-nested gamma oscillations (Figure 4 and Figure 5). Finally, we present some results on the effects of neurostimulation protocols for restoring theta-nested gamma oscillations in pathological states (Figure 6 and Figure 7).

Generation of gamma oscillations by E-I circuits

It is well-established that a network of interconnected pyramidal neurons and interneurons can give rise to oscillations in the gamma range, a mechanism termed pyramidal-interneuronal network gamma (PING) (Traub et al., 2004; Onslow et al., 2014; Segneri et al., 2020;). This mechanism has been observed in several optogenetic studies with gradually increasing light intensity (i.e., under a ramp input) affecting multiple different circuits, such as layer 2-3 pyramidal neurons of the mouse somatosensory cortex (Adesnik et al., 2010), the CA3 field of the hippocampus in rat in vitro slices (Akam et al., 2012), and in the non-human primate motor cortex (Lu et al., 2015). In all cases, gamma oscillations emerged above a certain threshold in terms of photostimulation intensity, and the frequency of these oscillations was either stable or slightly increased when increasing the intensity further. We sought to replicate these findings with our elementary E-I circuits composed of single-compartment conductance-based neurons driven by a ramping input current (Figure 2 and Figure S2). As an example, all the results in this section will be shown for an E-I circuit that has similar connectivity parameters as the CA1 field of the hippocampus in our complete model (see section “Hippocampal formation: inputs and connectivity” in the methods).

For low input currents provided to both neuronal populations, only the highly-excitable interneurons were activated (Figure 2A). For a sufficiently high input current (i.e., a strong input that could overcome the inhibition from the fast-spiking interneurons), the pyramidal neurons started spiking as well. As the amplitude of the input increased, the activity of the both neuronal populations became synchronized in the gamma range, asymptotically reaching a frequency of about 60 Hz (Figure 2A bottom panel). Decoupling the populations led to the abolition of gamma oscillations (Figure 2B), as neuronal activity was determined solely by the intrinsic properties of each cell. Interestingly, when the ramp input was provided solely to the excitatory population, we observed that the activity of the pyramidal neurons preceded the activity of the inhibitory neurons, while still preserving the emergence of gamma oscillations (Figure S2 A). As expected, decoupling the populations also abolished gamma oscillations, with the excitatory neurons spiking a frequency determined by their intrinsic properties and the inhibitory population remaining silent (Figure S2B).

To further characterize the intrinsic properties of individual inhibitory and excitatory neurons, we derived their input-frequency (I-F) curves, which represent the firing rate of individual neurons in response to a tonic input (Figure S3A). We observed that for certain input amplitudes, the firing rates of both types of neurons was within the gamma range. Interestingly, in the absence of noise, each population could generate by itself gamma oscillations that were purely driven by the input and determined by the intrinsic properties of the neurons (Figure S3B). Adding stochastic Gaussian noise in the membrane potential disrupted these artificial oscillations in decoupled populations (Figure S3C). All subsequent simulations were run with similar noise levels to prevent the emergence of artificial gamma oscillations.

Another potent way to induce gamma oscillations is to drive fast-spiking inhibitory neurons using pulsed optogenetic stimulation at gamma frequencies, a strategy that has been used both in the neocortex (Cardin et al., 2009) and hippocampal CA1 (Iaccarino et al., 2016). In particular, Cardin and colleagues systematically investigated the effect of driving either excitatory or fast-spiking inhibitory neocortical neurons at frequencies between 10 and 200 Hz (Cardin et al., 2009). They showed that fast-spiking interneurons are preferentially entrained around 40-50 Hz, while excitatory neurons respond better to lower frequencies. To verify the behavior of our model against these experimental data, we simulated pulsed optogenetic stimulation as an intracellular current provided to our reduced model of a single E-I circuit. Stimulation was applied at frequencies between 10 and 200 Hz to excitatory cells only, to inhibitory cells only, or to both at the same time (Figure S4). The population firing rates were used as a proxy for the local field potentials (LFP), and we computed the relative power in a 10-Hz band centered around the stimulation frequency, similarly to the method proposed in (Cardin et al., 2009). When presented with continuous stimulation across a range of frequencies in the gamma range, interneurons showed the greatest degree of gamma power modulation (Figure S4). Furthermore, when the stimulation was delivered to the excitatory population, the relative power around the stimulation frequency dropped significantly in frequencies above 10 Hz, similar to the reported experimental data (Cardin et al., 2009). The main difference between our simulation results and these experimental data is the specific frequencies at which fast-spiking interneurons showed resonance, which was slow gamma around 40 Hz in the mouse barrel cortex and fast gamma around 90 Hz in our model. This could be attributed to several factors, such as differences in the cellular properties between cortical and hippocampal fast-spiking interneurons, or the differences between the size of the populations and their relevant connectivity in the cortex and the hippocampus.”

Author response image 1.

Figure 2. Emergence of gamma oscillations in coupled excitatory-inhibitory populations under ramping input to both populations. A. Two coupled populations of excitatory pyramidal neurons (NE = 1000) and inhibitory interneurons (NI = 100) are driven by a ramping current input (0 nA to 1 nA) for 5 s. As the input becomes stronger, oscillations start to emerge (shaded green area), driven by the interactions between excitatory and inhibitory populations. The green inset shows the raster plot (neuronal spikes across time) of the two populations during the green shaded period (red for inhibitory; blue for excitatory). When the input becomes sufficiently strong (shaded magenta area), the populations become highly synchronized and produce oscillations in the gamma range (at approximately 50 Hz). The spectrogram (bottom panel) shows the power of the instantaneous firing rate of the pyramidal population as a function of time and frequency. It reveals the presence of gamma oscillations that emerge around 2s and increase in frequency until 4 s, when they settle at approximately 60 Hz. B. Similar depiction as in panel A. with the pyramidal-interneuronal populations decoupled. The absence of coupling leads to the abolition of gamma oscillations, each cell spiking activity being driven by its own inputs and intrinsic properties.

Author response image 2.

Figure S2 (Figure 2 – Figure Supplement 1). Emergence of gamma oscillations in coupled excitatoryinhibitory populations under ramping input to the excitatory population. Similar representation as in Figure 2, but with the input provided only to the excitatory population. All conclusions remain the same. In addition, the inhibitory population does not show any spiking activity in the decoupled case.

Author response image 3.

Figure S3 (Figure 2 – Figure Supplement 2). Cell-intrinsic spiking activity in decoupled excitatory and inhibitory populations under ramping input. A. Input-Frequency (I-F) curves for excitatory cells (left panel; pyramidal neurons with ICAN) and inhibitory cells (right panel; interneurons, fast-spiking) used in the model. Above a certain tonic input (around 0.35 nA for excitatory and 0.1 nA for inhibitory neurons), neurons can spike in the gamma range. B. Raster plot showing the spiking activity of excitatory (blue, NE = 1000) and inhibitory (red, NI = 100) neurons in decoupled populations under ramping input (top trace) and in the absence of noise in the membrane potential. Despite random initial conditions across neurons, oscillations emerge in both populations due to the intrinsic properties of the cells, with a frequency that is predicted by the respective I-F curves (panel A.). C. Similar representation as panel B. but with the addition of stochastic noise in the membrane potential of each neuron. The presence of noise disrupts the emergence of oscillations in these decoupled populations.

Author response image 4.

Figure S3 (Figure 2 – Figure Supplement 2). Cell-intrinsic spiking activity in decoupled excitatory and inhibitory populations under ramping input. A. Input-Frequency (I-F) curves for excitatory cells (left panel; pyramidal neurons with ICAN) and inhibitory cells (right panel; interneurons, fast-spiking) used in the model. Above a certain tonic input (around 0.35 nA for excitatory and 0.1 nA for inhibitory neurons), neurons can spike in the gamma range. B. Raster plot showing the spiking activity of excitatory (blue, NE = 1000) and inhibitory (red, NI = 100) neurons in decoupled populations under ramping input (top trace) and in the absence of noise in the membrane potential. Despite random initial conditions across neurons, oscillations emerge in both populations due to the intrinsic properties of the cells, with a frequency that is predicted by the respective I-F curves (panel A.). C. Similar representation as panel B. but with the addition of stochastic noise in the membrane potential of each neuron. The presence of noise disrupts the emergence of oscillations in these decoupled populations.

Beyond these weaknesses, this study has a strong utility for researchers wanting to explore hypotheses in the field of neurostimulations. In particular, I see value in such models for exploring more intricate, phase specific effects of continuous, as well as close loop stimulations which are on the rise in systems neuroscience.

We thank the reviewer for this appreciation of our work and its future perspectives.

Recommendations For The Authors:

Line 144, the authors mention that their MI values are erroneous in absence of additive noise - could this be due to the non-sinusoidal nature of the phase signal recorded, and be fixed by upscaling model size?

We thank the reviewer for this question and suggestion. The main reason behind the errors in the computation of the MI lies in the complete absence of oscillations at specific frequencies. Filtered signals within specific bands produced a power of 0 (or extremely low values), as seen in the power spectral densities. In such cases, the phase signal was not mathematically defined, but the toolbox we used to compute it still returned a numerical result that was inaccurate (for more details on the computation of the MI see Tort et al., 2010). To mitigate this numerical artefact, we decided to add uniform noise in the computed firing rates. This strategy is illustrated on Figure S6 (Figure 3 – Figure Supplement 2), which we have copied below for reference. Alternative approaches could probably have been used, such as increasing the noise in the membrane potential so that neurons would start spiking with firing rates that show more realistic power spectra, even in the absence of external inputs.

Author response image 5.

Figure S6 (Figure 3 – Figure Supplement 2). Quantification of PAC with and without noise. A. Quantifying PAC in the absence of noise produced inaccurate identification of the coupled frequency bands, due to the complete absence of oscillations at some frequencies. All analyses are based on the CA1 firing rates (top traces) during a representative simulation. Power spectral densities of these firing rates (left) indicate that some frequencies have a power of 0. PAC of the excitatory population was assessed using two graphical representations, the polar plot (middle) and comodulogram (right), and quantified using the MI. The comodulogram was calculated by computing the MI across 80% overlapping 1-Hz frequency bands in the theta range and across 90% overlapping 10-Hz frequency bands in the gamma range and subsequently plotted as a heat map. In the absence of noise, a slow theta frequency centered around 5 Hz is found to modulate a broad range of gamma frequencies between 40 and 100 Hz. The value indicated on the comodulogram indicates the average MI in the 3-9 Hz theta range and 40-80 Hz gamma range. As in Figure 2, the polar plot represents the amplitude of gamma oscillations (averaged across all theta cycles) at each phase of theta (theta range: 3-9 Hz, phase indicated as angular coordinate) and for different gamma frequencies (radial coordinate, binned in 1-Hz ranges). B. Adding uniform noise to the firing rate (with an amplitude ranging between 15 and 25% of the maximum firing rate) improved the identification of the coupled frequency bands. In this case, the slower theta frequency centered around 5 Hz modulates a gamma band located between 45 and 75 Hz.

Reviewer #2:

The main strength of this model is its use of a fairly physiologically detailed model of the hippocampus. The cells are single-compartment models but do include multiple ion channels and are spatially arranged in accordance with the hippocampal structure. This allows the understanding of how ion channels (possibly modifiable by pharmacological agents) interact with system-level oscillations and neurostimulation. The model also includes all the main hippocampal subfields. The other strength is its attention to an important topic, which may be relevant for dementia treatment or prevention, which few modeling studies have addressed. The work has several weaknesses.

We thank the reviewer for appreciating our detailed description of the hippocampal formation and the focus on neurostimulation applications that aim at treating oscillopathies, especially dementia.

1. First, while investigations of hippocampal neurostimulation are important there are few experimental studies from which one could judge the validity of the model findings. All its findings are therefore predictions. It would be much more convincing to first show the model is able to reproduce some measured empirical neurostimulation effect before proceeding to make predictions.

We acknowledge that the results presented in Figures 4-7 of the revised manuscript cannot be compared to existing experimental data, and are therefore purely predictive. Future experimental work is needed to verify these predictions.

Yet, we would also like to stress that the motivation behind this project was the inadequacy of previous models of theta-nested gamma oscillations (Onslow et al., 2014; Aussel et al., 2018; Segneri et al., 2020) to account for the mechanism of theta phase reset that occurs during electrical stimulation of the fornix or perforant path (Williams and Givens, 2003). Since we could not use these previous models to study the effects of neurostimulation on theta-nested gamma oscillations, we had to modify them to account for a dynamical theta input, which is the main methodological novelty that is reported in our manuscript (Figures 1 and 3 of the revised manuscript).

Despite the scarcity of experimental studies that could confirm the full model, we sought to replicate a few experimental findings that employed optogenetic stimulation to induce gamma oscillations in individual excitatory-inhibitory circuits. Although not specific to the hippocampus, these studies have shown that gamma oscillations can be induced using either ramped photostimulation targeting excitatory neurons (Adesnik et al., 2010; Akam et al., 2012; Lu et al., 2015), or pulsed stimulation driving inhibitory cells in the gamma range (Cardin et al., 2009; Iaccarino et al., 2016). To account for these experimental results and replicate them, we have added 4 new figures (Figure 2 and its 3 figure supplements) and an extensive section in the results part (L.141-217 of the revised manuscript). The added section and related figures are indicated in our response to reviewer 1, comment 3 (p 2-7).

2.1. Second, the model is very specific. Or if its behavior is to be considered general it has not been explained why.

Although the spatial organization and cellular details of the model are indeed very specific, its general behavior, i.e., the production of theta-nested gamma oscillations and theta phase reset, are common to any excitatory-inhibitory circuit interconnected with Kuramoto oscillators. To illustrate this point, we have generalized our approach to the neural mass model developed by Onslow and colleagues (Onslow ACE, Jones MW, Bogacz R. A Canonical Circuit for Generating Phase-Amplitude Coupling. PLoS ONE. 2014 Aug; 9(8):e102591). These results are represented in a new supplementary figure (Figure3 – Figure Supplement 4), and briefly described in a new paragraph of the results section (L.262-268 of the revised manuscript):

“Importantly, our approach is generalizable and can be applied to other models producing theta-nested gamma oscillations. For instance, we adapted the neural mass model by Onslow and colleagues (Onslow et al., 2014), replaced the fixed theta input by a set of Kuramoto oscillators, and demonstrated that it could also generate theta phase reset in response to single-pulse stimulation (Figure S8). These results illustrate that the general behavior of our model is not specific to the tuning of individual parameters in the conductancebased neurons, but follows general rules that are captured by the level of abstraction of the Kuramoto formalism.”

Author response image 6.

Figure S8 (Figure 3 – Figure Supplement 4). A neural mass model of coupled excitatory and inhibitory neurons driven by Kuramoto oscillators generates theta-nested gamma oscillations and theta phase reset. A. Two coupled neural masses (one excitatory and one inhibitory) driven by Kuramoto oscillators, which represent a dynamical oscillatory drive in the theta range, were used to implement a neural mass equivalent to our conductance-based model represented in Figure 1. Neural masses were modeled using the WilsonCowan formalism, with parameters adapted from Onslow et al. (2014) (𝑊𝐸𝐸 = 4.8, 𝑊𝐸𝐼 = 𝑊𝐼𝐸 = 4, 𝑊𝐼𝐼 = 0). B. The normalized population firing rates exhibit theta-nested gamma oscillations (middle and bottom panels) in response to the dynamic theta rhythm (top panel). A stimulation pulse delivered at the descending phase of the rhythm to both populations (marked by the inverted red triangle) produces a robust theta phase reset, similarly to Figure 3A.

This simplified model is described in more details in the methods (L.694-710 of the revised manuscript). Additionally, the generation of gamma oscillations by individual excitatory-inhibitory circuits is now described in details in the added section “Generation of gamma oscillations by E-I circuits” (L.159-217 of the revised manuscript), which has already been discussed in our response to reviewer 1, comment 3 (p 2-7).

2.2. For example, the model shows bistability between quiescence and TNGO, however what aspect of the model underlies this, be it some particular network structure or particular ion channel, for example, is not addressed.

We thank the reviewer for mentioning this point, which we have now addressed. The “bistable” behavior that we reported occurs for values of the theta input that are just below the threshold to induce selfsustained theta-gamma oscillations (Figure 5 of the revised manuscript, point B). Moreover, the presence of the Calcium-Activated-Nonspecific (CAN) cationic channel, which is expressed by pyramidal neurons in the entorhinal cortex, CA3, and CA1 fields of the hippocampus, is necessary for this behavior to occur. Indeed, abolishing CAN channels in all areas of the model suppresses this behavior. We have now addressed this point in a new supplementary figure (Figure 5 – Figure Supplement 4) and a short description in the text (L.287-303 of the revised manuscript).

“In the presence of dynamic theta input, the effects of single-pulse stimulation depended both on theta input amplitude and stimulation amplitude, highlighting different regimes of network activity (Figure 5 and Figure S9, Figure S10, Figure S11). For low theta input, theta-nested gamma oscillations were initially absent and could not be induced by stimulation (Figure 5A). At most, the stimulation could only elicit a few bursts of spiking activity that faded away after approximately 250 ms, similar to the rebound of activity seen in the absence of theta drive. For increasing theta input, the network switched to an intermediate regime: upon initialization at a state with no spiking activity, it could be kicked to a state with self-sustained theta-nested gamma oscillations by a single stimulation pulse of sufficiently high amplitude (Figure 5B). This regime existed for a range of septal theta inputs located just below the threshold to induce self-sustained theta-gamma oscillations without additional stimulation, as characterized by the post-stimulation theta power, gamma power, and theta-gamma PAC (Figure 5D). Removing CAN currents from all areas of the model abolished this behavior (Figure S12), which is interesting given the role of this current in the multistability of EC neurons (Egorov et al., 2002; Fransen et al., 2006) and in the intrinsic ability of the hippocampus to generate thetanested gamma oscillations (Giovannini et al., 2017). For the highest theta input, the network became able to spontaneously generate theta-nested gamma oscillations, even when initialized at a state with no spiking activity and without additional neurostimulation (Figure 5C).”

Author response image 7.

Figure S12 (Figure 5 – Figure Supplement 4). CAN currents are necessary for the production of selfsustained theta-gamma oscillations in response to single-pulse stimulation. A. Same as Figure 5B. B. Similar simulation as panel A., but without the presence of CAN currents in the EC, CA3 and CA1 fields of the hippocampus. Removing CAN currents from the model abolishes self-sustained theta-nested gamma oscillations in response to a single stimulation pulse (for the parameters represented in Figure 5, point B).

Furthermore, we realized that the terminology “bistable” may not be justified as we could not perform a systematic bifurcation analysis, which is typically carried out in simpler neural mass models (e.g., Onslow et al., 2014; Segneri et al., 2020). Therefore, we decided to rephrase the sentences about “bistability” to keep a more general terminology. The following sentences were revised:

L.20-23: “We showed that, for theta inputs just below the threshold to induce self-sustained theta-nested gamma oscillations, a single stimulation pulse could switch the network behavior from non-oscillatory to a state producing sustained oscillations.”

L.305-309: “Based on the above analyses, we considered two pathological states: one with a moderate theta input (i.e., moderately weak projections from the medial septum to the EC) that allowed the initiation of selfsustained oscillations by single stimulation pulses (Figure 5, point B), and one with a weaker theta input characterized by the complete absence of self-sustained oscillations even following transient stimulation (Figure 5, point A).”

L.316-317: “In the case of a moderate theta input and in the presence of phase reset, delivering a pulse at either the peak or trough of theta could induce theta-nested gamma oscillations (Figure 6A and 6C).”

L.353-357: “A very interesting finding concerns the behavior of the model in response to single-pulse stimulation for certain values of the theta amplitude (Figure5). For low theta amplitudes, a single stimulation pulse was capable of switching the network behavior from a state with no spiking activity to one with prominent theta-nested gamma oscillations. Whether such an effect can be induced in vivo in the context of memory processes remains an open question.”

2.3. Similarly for the various phase reset behaviors that are found.

We would like to clarify the fact that the observed phase reset curves (reported in Figure 3D) are a direct consequence of the choice of an appropriate phase response function for the Kuramoto oscillators representing the medial septum. This choice is inspired by experimentally measured phase response curves from CA3 neurons. These aspects are described briefly in the introduction and in more details in the methods, as indicated below:

L.101: “This new hybrid dynamical model could generate both theta-nested gamma oscillations and theta phase reset, following a particular phase response curve (PRC) inspired by experimental literature (Lengyel et al., 2005; Akam et al., 2012; Torben-Nielsen et al., 2010).”

L.528-537: “Hereafter, we call the term 𝑍(𝜃) the phase response function, to distinguish it from the PRC obtained from experimental data or simulations (see section below "Data Analysis", "Phase Response Curve"). Briefly, the PRC of an oscillatory system indicates the phase delay or advancement that follows a single pulse, as a function of the phase at which this input is delivered. The phase response function 𝑍(𝜃) was chosen to mimic as well as possible experimental PRCs reported in the literature (Lengyel et al., 2005; Kwag and Paulsen, 2009; Akam et al., 2012). These PRCs appear biphasic and show a phase advancement (respectively delay) for stimuli delivered in the ascending (respectively descending) slope of theta. To accurately model this behavior, we used the following equation for the phase response function, where 𝜃𝑝𝑒𝑎𝑘 represents the phase at which the theta rhythm reaches its maximum and the parameter 𝜙𝑜𝑓𝑓𝑠𝑒𝑡 controls the desired phase offset from the peak:

Author response image 8.

On the figure below, we illustrate the phase response curves of CA3 neurons measured by Lengyel et al., 2005 (panel A.), and compare it with our simulated phase response curves (panel B.). Note that the conventions for phase advance and phase delay are reversed between the two panels.

Finally, we would like to acknowledge that the model “is not derived from experimental phase response curves of septal neurons of which there is no direct measurement”, as mentioned by the reviewer in their comment 4 below. Despite the lack of experimental data specific to medial septum neurons, we argue that this phase response function is the only one that mathematically supports the generation of self-sustained theta-nested gamma oscillations in our current model. This statement is illustrated by Figure S7 (Figure 3 – Figure Supplement 3) and is mentioned in the results (L.249-261 of the revised manuscript):

We modeled this behavior by a specific term (which we called the phase response function) in the general equation of the Kuramoto oscillators (see methods, Equation 1). Importantly, introducing a phase offset in the phase response function disrupted theta-nested gamma oscillations (Figure S7), which suggests that the septohippocampal circuitry must be critically tuned to be able to generate such oscillations. The strength of phase reset could also be adjusted by a gain that was manually tuned. In the presence of the physiological phase response function and of a sufficiently high reset gain, a single stimulation pulse delivered to all excitatory and inhibitory CA1 neurons could reset the phase of theta to a value close to its peaks (Figure 3A). We computed the PRC of our simulated data for different stimulation amplitudes and validated that our neuronal network behaved according to the phase response function set in our Kuramoto oscillators (Figure 3D). It should be noted that including this phase reset mechanism affected the generated theta rhythm even in the absence of stimulation, extending the duration of the theta peak and thereby slowing down the frequency of the generated theta rhythm.

Author response image 9.

Figure S7 (Figure 3 – Figure Supplement 3). Network behavior generated by Kuramoto oscillators with nonphysiological phase response functions. Each panel is similar to Figure 3A, but with a different offset added to the phase response function of the Kuramoto oscillators (see methods, Equation 4). The center frequency was set to 6 Hz in all of these simulations. Overall, theta oscillations in these cases are less sinusoidal and show more abrupt phase changes than in the physiological case. A. A phase offset of −𝜋∕2 leads to an overall theta oscillation of 4 Hz, with a second peak following the main theta peak. B. A phase offset of +𝜋∕2 reduces the peak of theta, resetting the rhythm to the middle of the ascending phase. C. A phase offset of 𝜋 or -𝜋 leads to the CA1 output resetting the theta rhythm to the trough of theta.

2.4. We may wonder whether a different hippocampal model of TNGO, of which there are many published (for example [1-6]) would show the same effect under neurostimulation. This seems very unlikely […]

[1] Hyafil A, Giraud AL, Fontolan L, Gutkin B. Neural cross-frequency coupling: connecting architectures, mechanisms, and functions. Trends in neurosciences. 2015 Nov 1;38(11):725-40.

[2] Tort AB, Rotstein HG, Dugladze T, Gloveli T, Kopell NJ. On the formation of gamma-coherent cell assemblies by oriens lacunosum-moleculare interneurons in the hippocampus. Proceedings of the National Academy of Sciences. 2007 Aug 14;104(33):13490-5.

[3] Neymotin SA, Lazarewicz MT, Sherif M, Contreras D, Finkel LH, Lytton WW. Ketamine disrupts theta modulation of gamma in a computer model of hippocampus. Journal of Neuroscience. 2011 Aug 10;31(32):11733-43.

[4] Ponzi A, Dura-Bernal S, Migliore M. Theta-gamma phase-amplitude coupling in a hippocampal CA1 microcircuit. PLOS Computational Biology. 2023 Mar 23;19(3):e1010942.

[5] Bezaire MJ, Raikov I, Burk K, Vyas D, Soltesz I. Interneuronal mechanisms of hippocampal theta oscillations in a full-scale model of the rodent CA1 circuit. Elife. 2016 Dec 23;5:e18566.

[6] Chatzikalymniou AP, Gumus M, Skinner FK. Linking minimal and detailed models of CA1 microcircuits reveals how theta rhythms emerge and their frequencies controlled. Hippocampus. 2021 Sep;31(9):982-1002.

The highlighted publications, while very important in their findings regarding theta-gamma phase-amplitude coupling, focused on specific subfields of the hippocampus. In our work, we aimed to develop a model that includes the different anatomical divisions of the hippocampal formation, while still exhibiting theta-nested gamma oscillations, which is why we decided to expand the model by Aussel et al. (2018). Exploring the behavior of all these different hippocampal models under neurostimulation is beyond the scope of the current manuscript.

Nevertheless, we have added a new figure (Figure 3 – Figure Supplement 4) showing an adaptation of our modeling approach to a generic neural mass model of theta-nested gamma oscillations (Onslow et al., 2014), which illustrates the generalizability of our findings and is described in details in our response to comment 2.1. Moreover, we have further addressed the comments of the reviewers regarding bistability and phase response curves in our responses to comments 2.2 and 2.3.

Furthermore, we have added references to all 6 of these publications in the revised version of the manuscript:

L.43-50: Moreover, the modulation of gamma oscillations by the phase of theta oscillations in hippocampal circuits, a phenomenon termed theta-gamma phase-amplitude coupling (PAC), correlates with the efficacy of memory encoding and retrieval (Jensen and Colgin, 2007; Tort et al., 2009; Canolty and Knight, 2010; Axmacher et al., 2010; Fell and Axmacher, 2011; Lisman and Jensen, 2013; Lega et al., 2016). Experimental and computational work on the coupling between oscillatory rhythms has indicated that it originates from different neural architectures and correlates with a range of behavioral and cognitive functions, enabling the long-range synchronization of cortical areas and facilitating multi-item encoding in the context of memory (Hyafil et al., 2015)."

L.415-426: “In terms of neuronal cell types, we also made an important simplification by considering only basket cells as the main class of inhibitory interneuron in the whole hippocampal formation. However, it should be noted that many other types of interneurons exist in the hippocampus and have been modeled in various works with higher computational complexity (e.g., Bezaire et al., 2016; Chatzikalymniou et al., 2021). Among these various interneurons, oriens-lacunosum moleculare (OLM) neurons in the CA1 field have been shown to play a crucial role in synchronizing the activity of pyramidal neurons at gamma frequencies (Tort et al., 2007), and in generating theta-gamma PAC (e.g., Neymotin et al., 2011; Ponzi et al., 2023). Additionally, these cells may contribute to the formation of specific phase relationships within CA1 neuronal populations, through the integration between inputs from the medial septum, the EC, and CA3 (Mysin et al., 2019). Future work is needed to include more diverse cell types and detailed morphologies modeled through multiple compartments.”

2.5. […] and indeed the quiescent state itself shown by this model seems quite artificial.

We would like to clarify the fact that the “quiescent state” mentioned by the reviewer is a simply a state where the theta input is too low to induce theta-nested gamma oscillations. In this regime, neurons are active only due to the noise term in the membrane potential, which was adjusted based on Figure S3 (Figure 2 – Figure Supplement 2, shown below), at the minimal level needed to disrupt artificial synchronization in decoupled populations. For an input of 0 nA, we acknowledge that this network is indeed fully quiescent (i.e., does not show any spiking activity). However, as soon as the input increases, spontaneous spiking activity starts to appear with an average firing rate that depends on the input amplitude and is characterized by the input-frequency curves (panel A.). Please note that adding more noise could eliminate the observed quiescence in the absence of any input, but that it would not affect qualitatively the reported results.

Author response image 10.

Figure S3 (Figure 2 – Supplement 2). Cell-intrinsic spiking activity in decoupled excitatory and inhibitory populations under ramping input. A. Input-Frequency (I-F) curves for excitatory cells (left panel; pyramidal neurons with ICAN) and inhibitory cells (right panel; interneurons, fast-spiking) used in the model. Above a certain tonic input (around 0.35 nA for excitatory and 0.1 nA for inhibitory neurons), neurons can spike in the gamma range. B. Raster plot showing the spiking activity of excitatory (blue, NE = 1000) and inhibitory (red, NI = 100) neurons in decoupled populations under ramping input (top trace) and in the absence of noise in the membrane potential. Despite random initial conditions across neurons, oscillations emerge in both populations due to the intrinsic properties of the cells, with a frequency that is predicted by the respective IF curves (panel A.). C. Similar representation as panel B. but with the addition of stochastic noise in the membrane potential of each neuron. The presence of noise disrupts the emergence of oscillations in these decoupled populations.

2.6. Some indication that particular ion channels, CAN and M are relevant is briefly provided and the work would be much improved by examining this aspect in more detail.

We thank the reviewer for acknowledging the importance of these ion channels. We have now added a new supplementary figure (Figure 5 – Figure Supplement 4), which is described in more details in our response to comment 2.2 and illustrates the role of the CAN current in the generation of theta-nested gamma oscillations following a single stimulation pulse. Moreover, we would like to stress that the impact of CAN currents in the ability of the hippocampus to generate theta-nested gamma oscillations intrinsically, i.e., in the absence of persistent external input, has already been investigated in details by a previous computational study cited in our manuscript (Giovannini F, Knauer B, Yoshida M, Buhry L. The CAN-In network: A biologically inspired model for self-sustained theta oscillations and memory maintenance in the hippocampus. Hippocampus. 2017 Apr;809 27(4):450–463).

2.7. In summary, the work would benefit from an intuitive analysis of the basic model ingredients underlying its neurostimulation response properties.

We thank the reviewer for this suggestion. By addressing the reviewer’s previous comments (reviewer 2, comments 2.1 and 2.2), which overlap partly with the first reviewer (reviewer 1, comment 3), we believe we have improved the manuscript and have provided key information related to the way the model responds to neurostimulation.

3..) Third, while the model is fairly realistic, considerable important factors are not included and in fact, there are much more detailed hippocampal models out there (for example [5,6]). In particular, it includes only excitatory cells and a single type of inhibitory cell. This is particularly important since there are many models and experimental studies where specific cell types, for example, OLM and VIP cells, are strongly implicated in TNGO.

[5] Bezaire MJ, Raikov I, Burk K, Vyas D, Soltesz I. Interneuronal mechanisms of hippocampal theta oscillations in a full-scale model of the rodent CA1 circuit. Elife. 2016 Dec 23;5:e18566.

[6] Chatzikalymniou AP, Gumus M, Skinner FK. Linking minimal and detailed models of CA1 microcircuits reveals how theta rhythms emerge and their frequencies controlled. Hippocampus. 2021 Sep;31(9):982-1002.

We thank the reviewer for pointing out these interesting avenues for future studies. As indicated in previous responses (reviewer 1, comment 1; reviewer 2, comment 2.4), we have added several paragraphs to discuss these limitations, the rationale behind our simplifications, and potential improvements. In particular, we have added the following paragraphs to discuss our simplifications in terms of connectivity and cell types:

Anatomical connectivity:

L.141-150: “Biologically, GABAergic neurons from the medial septum project to the EC, CA3, and CA1 fields of the hippocampus (Toth et al., 1993; Hajós et al., 2004; Manseau et al., 2008; Hangya et al., 2009; Unal et al., 2015; Müller and Remy, 2018). Although the respective roles of these different projections are not fully understood, previous computational studies have suggested that the direct projection from the medial septum to CA1 is not essential for the production of theta in CA1 microcircuits (Mysin et al., 2019). Since our modeling of the medial septum is only used to generate a dynamic theta rhythm, we opted for a simplified representation where the medial septum projects only to the EC, which in turn drives the different subfields of the hippocampus. In our model, Kuramoto oscillators are therefore connected to the EC neurons and they receive projections from CA1 neurons (see methods for more details).”

Cell types:

L.415-426: “In terms of neuronal cell types, we also made an important simplification by considering only basket cells as the main class of inhibitory interneuron in the whole hippocampal formation. However, it should be noted that many other types of interneurons exist in the hippocampus and have been modeled in various works with higher computational complexity (e.g., Bezaire et al., 2016; Chatzikalymniou et al., 2021). Among these various interneurons, oriens-lacunosum moleculare (OLM) neurons in the CA1 field have been shown to play a crucial role in synchronizing the activity of pyramidal neurons at gamma frequencies (Tort et al., 2007), and in generating theta-gamma PAC (e.g., Neymotin et al., 2011; Ponzi et al., 2023). Additionally, these cells may contribute to the formation of specific phase relationships within CA1 neuronal populations, through the integration between inputs from the medial septum, the EC, and CA3 (Mysin et al., 2019). Future work is needed to include more diverse cell types and detailed morphologies modeled through multiple compartments.”

3.2. Other missing ingredients one may think might have a strong impact on model response to neurostimulation (in particular stimulation trains) include the well-known short-term plasticity between different hippocampal cell types and active dendritic properties.

We agree with the reviewer that plasticity mechanisms are important to include in future work, which we had already mentioned in the limitations section of the manuscript:

L.436-443: “Importantly, we did not consider learning through synaptic plasticity, even though such mechanisms could drastically modify synaptic conduction for the whole network (Borges et al., 2017). Even more interestingly, the inclusion of spike-timing-dependent plasticity would enable the investigation of stimulation protocols aimed at promoting LTP, such as theta-burst stimulation (Larson et al., 2015). This aspect would be of uttermost importance to make a link with memory encoding and retrieval processes (Axmacher et al., 2006; Tsanov et al., 2009; Jutras et al., 2013) and with neurostimulation studies for memory improvement (Titiz et al., 2017; Solomon et al., 2021).”

1. Fourth the MS model seems somewhat unsupported. It is modeled as a set of coupled oscillators that synchronize. However, there is also a phase reset mechanism included. This mechanism is important because it underlies several of the phase reset behaviors shown by the full model. However, it is not derived from experimental phase response curves of septal neurons of which there is no direct measurement. The work would benefit from the use of a more biologically validated MS model.

We would like to confirm that the phase reset mechanism is indeed at the core of using Kuramoto oscillators to model a particular system. For more details about our choice of a phase response function and the obtained results in terms of phase response curves, we refer the reader to our response to comment 2.3.

Generally speaking, we chose to use Kuramoto oscillators as it is the simplest model that can provide an oscillatory input to another system while including a phase reset mechanism. This set of oscillators was used to replace the fixed sinusoidal wave that represented theta inputs in previous models (Onslow et al., 2014; Aussel et al., 2018; Segneri et al., 2020). Kuramoto oscillators are a well-established model of synchronization in various fields of physics. They have also been used in neuroscience to model the phase reset of collective rhythms (Levnajić et al. 2010), and the effects of DBS on the basal ganglia network in Parkinson’s disease (Tass et al. 2003, Ebert et al. 2014, Weerasinghe et al. 2019).

More detailed models of the medial septum exist in the literature (e.g., Wang et al. 2002, Hajós et al. 2004) and model the GABAergic effects of the septal projections onto the hippocampal formation. However, it is not trivial to infer the connectivity parameters and the degree of innervation between the hippocampus and the medial septum. Furthermore, the claims made in our study do not necessarily depend on the nature of the projections between the two areas. Therefore, we decided to represent the medial septum in a conceptual way and focus mostly on the effects of these projections rather than replicating them in detail.

Aussel, Amélie, Laure Buhry, Louise Tyvaert, and Radu Ranta. “A Detailed Anatomical and Mathematical Model of the Hippocampal Formation for the Generation of Sharp-Wave Ripples and Theta-Nested Gamma Oscillations.” Journal of Computational Neuroscience 45, no. 3 (December 2018): 207–21. https://doi.org/10.1007/s10827-018-0704-x.

Ebert, Martin, Christian Hauptmann, and Peter A. Tass. “Coordinated Reset Stimulation in a Large-Scale Model of the STN-GPe Circuit.” Frontiers in Computational Neuroscience 8 (2014): 154. https://doi.org/10.3389/fncom.2014.00154.

Hajós, M., W.E. Hoffmann, G. Orbán, T. Kiss, and P. Érdi. “Modulation of Septo-Hippocampal θ Activity by GABAA Receptors: An Experimental and Computational Approach.” Neuroscience 126, no. 3 (January 2004): 599–610. https://doi.org/10.1016/j.neuroscience.2004.03.043.

Levnajić, Zoran, and Arkady Pikovsky. “Phase Resetting of Collective Rhythm in Ensembles of Oscillators.” Physical Review E 82, no. 5 (November 3, 2010): 056202. https://doi.org/10.1103/PhysRevE.82.056202.

Onslow, Angela C. E., Matthew W. Jones, and Rafal Bogacz. “A Canonical Circuit for Generating PhaseAmplitude Coupling.” Edited by Adriano B. L. Tort. PLoS ONE 9, no. 8 (August 19, 2014): e102591. https://doi.org/10.1371/journal.pone.0102591.

Segneri, Marco, Hongjie Bi, Simona Olmi, and Alessandro Torcini. “Theta-Nested Gamma Oscillations in Next Generation Neural Mass Models.” Frontiers in Computational Neuroscience 14 (2020). https://doi.org/10.3389/fncom.2020.00047. T ass, Peter A. “A Model of Desynchronizing Deep Brain Stimulation with a Demand-Controlled Coordinated Reset of Neural Subpopulations.” Biological Cybernetics 89, no. 2 (August 1, 2003): 81–88. https://doi.org/10.1007/s00422-003-0425-7.

Wang, Xiao-Jing. “Pacemaker Neurons for the Theta Rhythm and Their Synchronization in the Septohippocampal Reciprocal Loop.” Journal of Neurophysiology 87, no. 2 (February 1, 2002): 889–900. https://doi.org/10.1152/jn.00135.2001.

Weerasinghe, Gihan, Benoit Duchet, Hayriye Cagnan, Peter Brown, Christian Bick, and Rafal Bogacz. “Predicting the Effects of Deep Brain Stimulation Using a Reduced Coupled Oscillator Model.” PLoS Computational Biology 15, no. 8 (August 8, 2019): e1006575. https://doi.org/10.1371/journal.pcbi.1006575.

Author Response

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

Reviewer #1 (Public Review):

The manuscript by Wagstyl et al. describes an extensive analysis of gene expression in the human cerebral cortex and the association with a large variety of maps capturing many of its microscopic and macroscopic properties. The core methodological contribution is the computation of continuous maps of gene expression for >20k genes, which are being shared with the community. The manuscript is a demonstration of several ways in which these maps can be used to relate gene expression with histological features of the human cortex, cytoarchitecture, folding, function, development and disease risk. The main scientific contribution is to provide data and tools to help substantiate the idea of the genetic regulation of multi-scale aspects of the organisation of the human brain. The manuscript is dense, but clearly written and beautifully illustrated.

Main comments

The starting point for the manuscript is the construction of continuous maps of gene expression for most human genes. These maps are based on the microarray data from 6 left human brain hemispheres made available by the Allen Brain Institute. By technological necessity, the microarray data is very sparse: only 1304 samples to map all the cortex after all subjects were combined (a single individual's hemisphere has ~400 samples). Sampling is also inhomogeneous due to the coronal slicing of the tissue. To obtain continuous maps on a mesh, the authors filled the gaps using nearest-neighbour interpolation followed by strong smoothing. This may have two potentially important consequences that the authors may want to discuss further: (a) the intrinsic geometry of the mesh used for smoothing will introduce structure in the expression map, and (b) strong smoothing will produce substantial, spatially heterogeneous, autocorrelations in the signal, which are known to lead to a significant increase in the false positive rate (FPR) in the spin tests they used.

Many thanks to the reviewer for their considered feedback. We have addressed these primary concerns into point-by-point responses below. The key conclusions from our new analyses are: (i) while the intrinsic geometry of the mesh had not originally been accounted for in sufficient detail, the findings presented in this manuscript paper are not driven by mesh-induced structure, (ii) that the spin test null models used in this manuscript [(including a modified version introduced in response to (i)] are currently the most appropriate way to mitigate against inflated false positive rates when making statistical inferences on smooth, surface-based data.

a. Structured smoothing

A brain surface has intrinsic curvature (Gaussian curvature, which cannot be flattened away without tearing). The size of the neighbourhood around each surface vertex will be determined by this curvature. During surface smoothing, this will make that the weight of each vertex will be also modulated by the local curvature, i.e., by large geometric structures such as poles, fissures and folds. The article by Ciantar et al (2022, https://doi.org/10.1007/s00429-022-02536-4) provides a clear illustration of this effect: even the mapping of a volume of pure noise into a brain mesh will produce a pattern over the surface strikingly similar to that obtained by mapping resting state functional data or functional data related to a motor task.

Comment 1

It may be important to make the readers aware of this possible limitation, which is in large part a consequence of the sparsity of the microarray sampling and the necessity to map that to a mesh. This may confound the assessments of reproducibility (results, p4). Reproducibility was assessed by comparing pairs of subgroups split from the total 6. But if the mesh is introducing structure into the data, and if the same mesh was used for both groups, then what's being reproduced could be a combination of signal from the expression data and signal induced by the mesh structure.

Response 1

The reviewer raises an important question regarding the potential for interpolation and smoothing on a cortical mesh to induce a common/correlated signal due to the intrinsic mesh structure. We have now generated a new null model to test this idea which indicates that intrinsic mesh structure is not inflating reproducibility in interpolated expression maps. This new null model spins the original samples prior to interpolation, smoothing and comparison between triplet splits of the six donors, with independent spins shared across the triplet. For computational tractability we took one pair of triplets and regenerated the dataset for each triplet using 10 independent spins. We used these to estimate gene-gene null reproducibility for 90 independent pairwise combinations of these 10 spins. Across these 90 permutations, the average median gene-gene correlation was R=0.03, whereas in the unspun triplet comparisons this was R=0.36. These results indicate that the primary source of the gene-level triplet reproducibility is the underlying shared gene expression pattern rather than interpolation-induced structure.

In Methods 2a: "An additional null dataset was generated to test whether intrinsic geometry of the cortical mesh and its impact on interpolation for benchmarking analyses of DEMs and gradients (Fig S1d, Fig S2d, Fig S3c). In these analyses, the original samples were rotated on the spherical surface prior to subsequent interpolation, smoothing and gradient calculation. Due to computational constraints the full dataset was recreated only for 10 independent spins. These are referred to as the “spun+interpolated null”.

Author response image 1.

Figure S1d, Gene predictability was higher across all triplet-triplet pairs than when compared to spun+interpolated null.

Comment 2

It's also possible that mesh-induced structure is responsible in part for the "signal boost" observed when comparing raw expression data and interpolated data (fig S1a). How do you explain the signal boost of the smooth data compared with the raw data otherwise?

Response 2

We thank the reviewer for highlighting this issue of mesh-induced structure. We first sought to quantify the impact of mesh-induced structure through the new null model, in which the data are spun prior to interpolation. New figure S1d, S2d and S3c all show that the main findings are not driven by interpolation over a common mesh structure, but rather originate in the underlying expression data.

Specifically, for the original Figure S1a, the reviewer highlights a limitation that we compared intersubject predictability of raw-sample to raw-sample and interpolated-to-interpolated. In this original formulation improved prediction scores for interpolated-to-interpolated (the “signal boost”) could be driven by mesh-induced structure being applied to both the input and predicted maps. We have updated this so that we are now comparing raw-to-raw and interpolated-to-raw, i.e. whether interpolated values are better estimations of the measured expression values. The new Fig S1a&b (see below) shows a signal boost in gene-level and vertex level prediction scores (delta R = +0.05) and we attribute this to the minimisation of location and measurement noise in the raw data, improving the intersubject predictability of expression levels.

In Methods 2b: "To assess the effect of data interpolation in DEM generation we compared gene-level and vertex-level reproducibility of DEMs against a “ground truth” estimate of these reproducibility metrics based on uninterpolated expression data. To achieve a strict comparison of gene expression values between different individuals at identical spatial locations we focused these analyses on the subset of AHBA samples where a sample from one subject was within 3 mm geodesic distance of another. This resulted in 1097 instances (spatial locations) with measures of raw gene expression of one donor, and predicted values from the second donor’s un-interpolated AHBA expression data and interpolated DEM. We computed gene-level and vertex-level reproducibility of expression using the paired donor data at each of these sample points for both DEM and uninterpolated AHBA expression values. By comparing DEM reproducibility estimates with those for uninterpolated AHBA expression data, we were able to quantify the combined effect of interpolation and smoothing steps in DEM generation. We used gene-level reproducibility values from DEMs and uninterpolated AHBA expression data to compute a gene-level difference in reproducibility, and we then visualized the distribution of these difference values across genes (Fig S1a). We used gene-rank correlation to compare vertex-level reproducibility values between DEMs and uninterpolated AHBA expression data (Fig S1b)."

Author response image 2.

Figure S1. Reproducibility of Dense Expression Maps (DEMs) interpolated from spatially sparse postmortem measures of cortical gene expression. a, Signal boost in the interpolated DEM dataset vs. spatially sparse expression data. Restricting to samples taken from approximately the same cortical location in pairs of individuals (within 3mm geodesic distance), there was an overall improvement in intersubject spatial predictability in the interpolated maps. Furthermore, genes with lower predictability in the interpolated maps were less predictable in the raw dataset, suggesting these regions exhibit higher underlying biological variability rather than methodologically introduced bias. b, Similarly at the paired sample locations, gene-rank predictability was generally improved in DEMs vs. sparse expression data (median change in R from sparse samples to interpolated for each pair of subjects, +0.5).

1. How do you explain that despite the difference in absolute value the combined expression maps of genes with and without cortical expression look similar? (fig S1e: in both cases there's high values in the dorsal part of the central sulcus, in the occipital pole, in the temporal pole, and low values in the precuneus and close to the angular gyrus). Could this also reflect mesh-smoothing-induced structure?

Response 3

As with comment 1, this is an interesting perspective that we had not fully considered. We would first like to clarify that non-cortical expression is defined from the independent datasets including the “cortex” tissue class of the human protein atlas and genes identified as markers for cortical layers or cortical cells in previous studies. This is still likely an underestimate of true cortically expressed genes as some of these “non-cortical genes” had high intersubject reproducibility scores. Nevertheless we think it appropriate to use a measure of brain expression independent of anything included in other analyses for this paper. These considerations are part of the reason we provide all gene maps with accompanying uncertainty scores for user discretion rather than simply filtering them out.

In terms of the spatially consistent pattern of the gene ranks of Fig S1f, this consistent spatial pattern mirrors Transcriptomic Distinctiveness (r=0.52 for non-cortical genes, r=0.75 for cortical genes), so we think that as the differences in expression signatures become more extreme, the relative ranks of genes in that region are more reproducible/easier to predict.

To assess whether mesh-smoothing-induced structure is playing a role, we carried out an additional the new null model introduced in response to comment 1, and asked if the per-vertex gene rank reproducibility of independently spun subgroup triplets showed a similar structure to that in our original analyses. Across the 90 permutations, the median correlation between vertex reproducibility and TD was R=0.10. We also recalculated the TD maps for the 10 spun datasets and the mean correlation with the original TD did not significantly differ from zero (mean R = 0.01, p=0.2, nspins =10). These results indicate that folding morphology is not the major driver of local or large scale patterning in the dataset. We have included this as a new Figure S3c.

We have updated the text as follows:

In Methods 3a: "Third, to assess whether the covariance in spatial patterning across genes could be a result of mesh-associated structure introduced through interpolation and smoothing, TD maps were recomputed for the spun+interpolated null datasets and compared to the original TD map (Fig S3c)."

In Results: "The TD map observed from the full DEMs library was highly stable between all disjoint triplets of donors (Methods, Fig S3a, median cross-vertex correlation in TD scores between triplets r=0.77) and across library subsets at all deciles of DEM reproducibility (Methods, Fig S3b, cross-vertex correlation in TD scores r>0.8 for the 3rd-10th deciles), but was not recapitulated in spun null datasets (Fig S3c)."

Author response image 3.

Figure S3c, Correlations between TD and TD maps regenerated on datasets spun using two independent nulls, one where the rotation is applied prior to interpolation and smoothing (spun+interpolated) and one where it is applied to the already-created DEMs. In each null, the same rotation matrix is applied to all genes.

Comment 4

Could you provide more information about the way in which the nearest-neighbours were identified (results p4). Were they nearest in Euclidean space? Geodesic? If geodesic, geodesic over the native brain surface? over the spherically deformed brain? (Methods cite Moresi & Mather's Stripy toolbox, which seems to be meant to be used on spheres). If the distance was geodesic over the sphere, could the distortions introduced by mapping (due to brain anatomy) influence the geometry of the expression maps?

Response 4

We have clarified in the Methods that the mapping is to nearest neighbors on the spherically-inflated surface.

The new null model we have introduced in response to comments 1 & 3 preserves any mesh-induced structure alongside any smoothing-induced spatial autocorrelations, and the additional analyses above indicate that main results are not induced by systematic mesh-related interpolation signal. In response to an additional suggestion from the reviewer (Comment 13), we also assessed whether local distortions due to the mesh could be creating apparent border effects in the data, for instance at the V1-V2 boundary. At the V1-V2 border, which coincides anatomically with the calcarine sulcus, we computed the 10 genes with the highest expression gradient along this boundary in the actual dataset and the spun-interpolated null. The median test expression gradients along this border was higher than in any of the spun datasets, indicating that these boundary effects are not explained by the interpolation and cortical geometry effects on the data (new Fig S2d). The text has been updated as follows:

In Methods 1: "For cortical vertices with no directly sampled expression, expression values were interpolated from their nearest sampled neighbor vertex on the spherical surface (Moresi and Mather, 2019) (Fig 1b)."

In Methods 2: "We used the spun+interpolated null to test whether high gene gradients could be driven by non-uniform interpolation across cortical folds. We quantified the average gradient for all genes along the V1-V2 border in the atlas, as well as for 10 iterations of the atlas where the samples were spun prior to interpolation. We computed the median gradient magnitude for the 20 top-ranked genes for each (Fig S2d)."

Author response image 4.

Figure S2d Mean of gradient magnitudes for 20 genes with largest gradients along V1-V2 border, compared to values along the same boundary on the spun+interpolated null atlas. Gradients were higher in the actual dataset than in all spun version indicating this high gradient feature is not primarily due to the effects of calcarine sulcus morphology on interpolation

Comment 5

Could you provide more information about the smoothing algorithm? Volumetric, geodesic over the native mesh, geodesic over the sphere, averaging of values in neighbouring vertices, cotangent-weighted laplacian smoothing, something else?

Response 5

We are using surface-based geodesic over the white surface smoothing described in Glasser et al., 2013 and used in the HCP workbench toolbox (https://www.humanconnectome.org/software/connectome-workbench). We have updated the methods to clarify this.

In Methods 1: "Surface expression maps were smoothed using the Connectome Workbench toolbox (Glasser et al. 2013) with a 20mm full-width at half maximum Gaussian kernel , selected to be consistent with this sampling density (Fig 1c)."

Comment 6

Could you provide more information about the method used for computing the gradient of the expression maps (p6)? The gradient and the laplacian operator are related (the laplacian is the divergence of the gradient), which could also be responsible in part for the relationships observed between expression transitions and brain geometry.

Response 6

We are using Connectome Workbench’s metric gradient command for this Glasser et al., 2013 and used in the HCP workbench pipeline. The source code for gradient calculation can be found here: https://github.com/Washington-University/workbench/blob/131e84f7b885d82af76e be21adf2fa97795e2484/src/Algorithms/AlgorithmMetricGradient.cxx

In Methods 2: >For each of the resulting 20,781 gene-level expression maps, the orientation and magnitude of gene expression change at each vertex (i.e. the gradient) was calculated for folded, inflated, spherical and flattened mesh representations of the cortical sheet using Connectome Workbench’s metric gradient command (Glasser et al. 2013).

b. Potentially inflated FPR for spin tests on autocorrelated data."

Spin tests are extensively used in this work and it would be useful to make the readers aware of their limitations, which may confound some of the results presented. Spin tests aim at establishing if two brain maps are similar by comparing a measure of their similarity over a spherical deformation of the brains against a distribution of similarities obtained by randomly spinning one of the spheres. It is not clear which specific variety of spin test was used, but the original spin test has well known limitations, such as the violation of the assumption of spatial stationarity of the covariance structure (not all positions of the spinning sphere are equivalent, some are contracted, some are expanded), or the treatment of the medial wall (a big hole with no data is introduced when hemispheres are isolated).

Another important limitation results from the comparison of maps showing autocorrelation. This problem has been extensively described by Markello & Misic (2021). The strong smoothing used to make a continuous map out of just ~1300 samples introduces large, geometry dependent autocorrelations. Indeed, the expression maps presented in the manuscript look similar to those with the highest degree of autocorrelation studied by Markello & Misic (alpha=3). In this case, naive permutations should lead to a false positive rate ~46% when comparing pairs of random maps, and even most sophisticated methods have FPR>10%.

Comment 7 There's currently several researchers working on testing spatial similarity, and the readers would benefit from being made aware of the problem of the spin test and potential solutions. There's also packages providing alternative implementations of spin tests, such as BrainSMASH and BrainSpace, which could be mentioned.

Response 7

We thank the reviewer for raising the issue of null models. First, with reference to the false positive rate of 46% when maps exhibit spatial autocorrelation, we absolutely agree that this is an issue that must be accounted for and we address this using the spin test. We acknowledge there has been other work on nulls such as BrainSMASH and BrainSpace. Nevertheless in the Markello and Misic paper to which the reviewer refers, the BrainSmash null models perform worse with smoother maps (with false positive rates approaching 30% in panel e below), whereas the spin test maintains false positives rates below 10%.

Author response image 5.

We have added a brief description of the challenge and our use of the spin test.

In Methods 2a: "Cortical maps exhibit spatial autocorrelation that can inflate the False Positive Rate, for which a number of methods have been proposed(Alexander-Bloch et al. 2018; Burt et al. 2020; Vos de Wael et al. 2020). At higher degrees of spatial smoothness, this high False Positive Rate is most effectively mitigated using the spin test(Alexander-Bloch et al. 2018; Markello and Misic 2021; Vos de Wael et al. 2020). In the following analyses when generating a test statistic comparing two spatial maps, to generate a null distribution, we computed 1000 independent spins of the cortical surface using https://netneurotools.readthedocs.io, and applied it to the first map whilst keeping the second map unchanged. The test statistic was then recomputed 1000 times to generate a null distribution for values one might observe by chance if the maps shared no common organizational features. This is referred to throughout as the “spin test” and the derived p-values as pspin."

Comment 8

Could it be possible to measure the degree of spatial autocorrelation?

Response 8

We agree this could be a useful metric to generate for spatial cortical maps. However, there are multiple potential metrics to choose from and each of the DEMs would have their own value. To address this properly would require the creation of a set of validated tools and it is not clear how we could summarize this variety of potential metrics for 20k genes. Moreover, as discussed above the spin method is an adequate null across a range of spatial autocorrelation degrees, thus while we agree that in general estimation of spatial smoothness could be a useful imaging metric to report, we consider that it is beyond the scope of the current manuscript.

Comment 9

Could you clarify which version of the spin test was used? Does the implementation come from a package or was it coded from scratch?

Response 9

As Markello & Misic note, at the vertex level, the various implementations of the spin test become roughly equivalent to the ‘original’ Alexander-Bloch et al., implementation. We used took the code for the ‘original’ version implemented in python here: https://netneurotools.readthedocs.io/en/latest/_modules/netneurotools/stats.html# gen_spinsamples.

This has been updated in the methods (see Response 7).

Comment 10

Cortex and non-cortex vertex-level gene rank predictability maps (fig S1e) are strikingly similar. Would the spin test come up statistically significant? What would be the meaning of that, if the cortical map of genes not expressed in the cortex appeared to be statistically significantly similar to that of genes expressed in the cortex?

Response 10

Please see response to comment 3, which also addresses this observation.

Reviewer #2 (Public Review):

The authors convert the AHBA dataset into a dense cortical map and conduct an impressively large number of analyses demonstrating the value of having such data.

I only have comments on the methodology.

Comment 1

First, the authors create dense maps by simply using nearest neighbour interpolation followed by smoothing. Since one of the main points of the paper is the use of a dense map, I find it quite light in assessing the validity of this dense map. The reproducibility values they calculate by taking subsets of subjects are hugely under-powered, given that there are only 6 brains, and they don't inform on local, vertex-wise uncertainties). I wonder if the authors would consider using Gaussian process interpolation. It is really tailored to this kind of problem and can give local estimates of uncertainty in the interpolated values. For hyperparameter tuning, they could use leave-one-brain-out for that.

I know it is a lot to ask to change the base method, as that means re-doing all the analyses. But I think it would strengthen the paper if the authors put as much effort in the dense mapping as they did in their downstream analyses of the data.

Response 1

We thank the reviewer for the suggestion to explore Gaussian process interpolation. We have implemented this for our dataset and attempted to compare this with our original method with the 3 following tests: i) intertriplet reproducibility of individual gene maps, ii) microscale validations: area markers, iii) macroscale validations: bio patterns.

Overall, compared to our original nearest-neighbor interpolation method, GP regression (i) did not substantially improve gene-level reproducibility of expression maps (median correlation increase of R=0.07 which was greater for genes without documented protein expression in cortex): ii) substantially worsened performance in predicting areal marker genes and iii) showed similar but slightly worse performance at predicting macroscale patterns from Figure 1.

Given the significantly poorer performance on one of our key tests (ii) we have opted not to replace our original database, but we do now include code for the alternative GP regression methodology in the github repository so others can reproduce/further develop these methods.

Author response image 6.

ii) Genes ranked by mean expression gradient from current DEMs (left) and Gaussian process-derived interpolation maps (right). Established Human and macaque markers are consistently higher-ranked in DEM maps. iii) Figure 1 Interpolated vs GP regression

Author response table 1.

Comment 2

It is nice that the authors share some code and a notebook, but I think it is rather light. It would be good if the code was better documented, and if the user could have access to the non-smoothed data, in case they was to produce their own dense maps. I was only wondering why the authors didn't share the code that reproduces the many analyses/results in the paper.

Response 2

We thank the reviewer for this suggestion. In response we have updated the shared github repository (https://github.com/kwagstyl/magicc). This now includes code and notebooks to reproduce the main analyses and figures.

Reviewer #1 (Recommendations For The Authors):

Minor comments

Comment 11

p4 mentions Fig S1h, but the supp figures only goes from S1a to S1g

Response 11

We thank the reviewer for capturing this error. It was in fact referring to what is now Fig S1h and has been updated.

Comment 12

It would be important that the authors share all the code used to produce the results in the paper in addition to the maps. The core methodological contribution of the work is a series of continuous maps of gene expression, which could become an important tool for annotation in neuroimaging research. Many arbitrary (reasonable) decisions were made, it would be important to enable users to evaluate their influence on the results.

Response 12

We thank both reviewers for this suggestion. We have updated the github to be able to reproduce the dense maps and key figures with our methods.

Comment 13

p5: Could the sharp border reflect the effect of the geometry of the calcarine sulcus on map smoothing? More generally, could there be an effect of folds on TD?

Response 13

Please see our response to Reviewer 1, Comment 1 above, where we introduce the new null models now analyzed to test for effects of mesh geometry on our findings. These new null models - where original source data were spun prior to interpolation suggest that neither the sharp V1/2 border or the TD map are effects of mesh geometry. Specifically: (i) , the magnitudes of gradients along the V1/2 boundary from null models were notably smaller than those in our original analyses (see new figure S2d), and (ii) TD maps computed from the new null models showed no correlation with TD maps from ur original analyses (new Figure S3c, mean R = 0.01, p=0.2, nspins =10).

Comment 14

p5: Similar for the matching with the areas in Glasser's parcellation: the definition of these areas involves alignment through folds (based on freesurfer 'sulc' map, see Glasser et al 2016). If folds influence the geometry of TDs, could that influence the match?

Response 14

We note that Fig S3c provided evidence that folding was not the primary driver of the TD patterning. However, it is true that Glasser et al. use both neuroanatomy (folding, thickness and myelin) and fMRI-derived maps to delineate their cortical areas. As such Figure 2 f & g aren’t fully independent assessments. Nevertheless the reason that these features are used is that many of the sulci in question have been shown to reliably delineate cytoarchitectonic boundaries (Fischl et al., 2008).

In Results: "A similar alignment was seen when comparing gradients of transcriptional change with the spatial orientation of putative cortical areas defined by multimodal functional and structural in vivo neuroimaging(Glasser et al., 2016) (expression change running perpendicular to area long-axis, pspin<0.01, Fig 2g, Methods)."

Comment 15

p6: TD peaks are said to overlap with functionally-specialised regions. A comment on why audition is not there, nor language, but ba 9-46d is? Would that suggest a lesser genetic regulation of those functions?

Response 15

The reviewer raises a valid point and this was a result that we were also surprised by. The finding that the auditory cortex is not as microstructurally distinctive as, say V1, is consistent with other studies applying dimensionality-reduction techniques to multimodal microstructural receptor data (e.g. Zilles et al., 2017, Goulas et al., 2020). These studies found that the auditory microstructure is not as extreme as either visual and somatomotor areas. From a methodological view point, the primary auditory cortex is significantly smaller than both visual and somatomotor areas, and therefore is captured by fewer independent samples, which could reduce the detail in which its structure is being mapped in our dataset.

For the frontal areas, we would note that i) the frontal peak is the smallest of all peaks found and was more strongly characterised by low z-score genes than high z-score. ii) the anatomical areas in the frontal cortex are much more highly variable with respect to folding morphology (e.g. Rajkowska 1995). The anatomical label of ba9-46d (and indeed all other labels) were automatically generated as localisers rather than strict area labels. We have clarified this in the text as follows:

In Methods 3a: "Automated labels to localize TD peaks were generated based on their intersection with a reference multimodal neuroimaging parcellation of the human cortex(Glasser et al., 2016). Each TD was given the label of the multimodal parcel that showed greatest overlap (Fig 2b)."

Comment 16.

p7: The proposition that "there is a tendency for cortical sulci to run perpendicular to the direction of fastest transcriptional change", could also be "there is a tendency for the direction of fastest transcriptional change to run perpendicular to cortical sulci"? More pragmatically, this result from the geometry of transcriptional maps being influenced by sulcal geometry in their construction.

Response 16

Please see our response to Reviewer 1, Comment 1 above, where we introduce the new null models now analyzed to test for effects of mesh geometry on our findings. These models indicate that the topography of interpolated gene expression maps do not reflect influences of sulcal geometry on their construction.

Comment 17

p7: TD transitions are indicated to precede folding. This is based on a consideration of folding development based on the article by Chi et al 1977, which is quite an old reference. In that paper, the authors estimated the tempo of human folding development based on the inspection of photographs, which may not be sufficient for detecting the first changes in curvature leading to folds. The work of the Developing Human Connectome consortium may provide a more recent indication for timing. In their data, by PCW 21 there's already central sulcus, pre-central, post-central, intra-parietal, superior temporal, superior frontal which can be detected by computing the mean curvature of the pial surface (I can only provide a tweet for reference: https://twitter.com/R3RT0/status/1617119196617261056). Even by PCW 9-13 the callosal sulcus, sylvian fissure, parieto-occipital fissure, olfactory sulcus, cingulate sulcus and calcarine fissure have been reported to be present (Kostovic & Vasung 2009).

Response 17

Our field lacks the data necessary to provide a comprehensive empirical test for the temporal ordering of regional transcriptional profiles and emergence of folding. Our results show that transcriptional identities of V1 and TGd are - at least - present at the very earliest stages of sulcation in these regions. In response to the reviewers comment we have updated with a similar fetal mapping project which similarly shows evidence of the folds between weeks 17-21 and made the language around directionality more cautious.

In Results: "The observed distribution of these angles across vertices was significantly skewed relative to a null based on random alignment between angles (pspin<0.01, Fig 2f, Methods) - indicating that there is indeed a tendency for cortical sulci and the direction of fastest transcriptional change to run perpendicular to each other (pspin<0.01, Fig 2f).

As a preliminary probe for causality, we examined the developmental ordering of regional folding and regional transcriptional identity. Mapping the expression of high-ranking TD genes in fetal cortical laser dissection microarray data(Miller et al., 2014) from 21 PCW (Post Conception Weeks) (Methods) showed that the localized transcriptional identity of V1 and TGd regions in adulthood is apparent during the fetal periods when folding topology begins to emerge (Chi et al. 1977; Xu et al. 2022) (Fig " S2d).

In Discussion: "By establishing that some of these cortical zones are evident at the time of cortical folding, we lend support to a “protomap”(Rakic 1988; O'Leary 1989; O'Leary et al. 2007; Rakic et al. 2009) like model where the placement of some cortical folds is set-up by rapid tangential changes in cyto-laminar composition of the developing cortex(Ronan et al., 2014; Toro and Burnod, 2005; Van Essen, 2020). The DEMs are derived from fully folded adult donors, and therefore some of the measured genetic-folding alignment might also be induced by mechanical distortion of the tissue during folding(Llinares-Benadero and Borrell 2019; Heuer and Toro 2019). However, no data currently exist to conclusively assess the directionality of this gene-folding relationship."

Comment 18

p7: In my supplemental figures (obtained from biorxiv, because I didn't find them among the files submitted to eLife) there's no S2j (only S2a-S2i).

Response 18

We apologize, this figure refers to S3k (formerly S3j), rather than S2j. We have updated the main text.

Comment 19 p7: It is not clear from the methods (section 3b) how the adult and fetal brains were compared. Maybe using MSM (Robinson et al 2014)?

Response 19

We have now clarified this in Methods text as reproduced below.

In Methods 3b: "We averaged scaled regional gene expression values between donors per gene, and filtered for genes in the fetal LDM dataset that were also represented in the adult DEM dataset - yielding a single final 20,476*235 gene-by-sample matrix of expression values for the human cortex at 21 PCW. Each TD peak region was then paired with the closest matching cortical label within the fetal regions. This matrix was then used to test if each TD expression signature discovered in the adult DEM dataset (Fig 2, Table 3) was already present in similar cortical regions at 21 PCW."

Comment 20

p7: WGCNA is used prominently, could you provide a brief introduction to its objectives? The gene coexpression networks are produced after adjusting the weight of the network edges to follow a scale-free topology, which is meant to reflect the nature of protein-protein interactions. Soft thresholding increases contrast, but doesn't this decrease a potential role of infinitesimal regulatory signals?

Response 20

We agree with the reviewer that the introduction to WGCNA needed additional details and have amended the Results (see below). One limitation of WGCNA-derived associations is that it will downweigh the role of smaller relationships including potentially important regulatory signals. WGCNA methods have been titrated to capture strong relationships. This is an inherent limitation of all co-expression driven methods which lead to an incomplete characterisation of the molecular biology. Nevertheless we feel these stronger relationships are still worth capturing and interrogating. We have updated the text to introduce WGCNA and acknowledge this potential weakness in the approach.

In Results: "Briefly, WGCNA constructs a constructs a connectivity matrix by quantifying pairwise co-expression between genes, raising the correlations to a power (here 6) to emphasize strong correlations while penalizing weaker ones, and creating a Topological Overlap Matrix (TOM) to capture both pairwise similarities expression and connectivity. Modules of highly interconnected genes are identified through hierarchical clustering. The resultant WGCNA modules enable topographic and genetic integration because they each exist as both (i) a single expression map (eigenmap) for spatial comparison with neuroimaging data (Fig 3a,b, Methods) and, (ii) a unique gene set for enrichment analysis against marker genes systematically capturing multiple scales of cortical organization, namely: cortical layers, cell types, cell compartments, protein-protein interactions (PPI) and GO terms (Methods, Table S2 and S4)."

Comment 21

WGCNA modules look even more smooth than the gene expression maps. Are these maps comparable to low frequency eigenvectors? Autocorrelation in that case should be very strong?

Response 21

These modules are smooth as they are indeed eigenvectors which likely smooth out some of the more detailed but less common features seen in individual gene maps. These do exhibit high degrees of autocorrelation, nevertheless we are applying the spin test which is currently the appropriate null model for spatially autocorrelated cortical maps (Response 7).

Comment 22

If the WGCNA modules provide an orthogonal basis for surface data, is it completely unexpected that some of them will correlate with low-frequency patterns? What would happen if random low frequency patterns were generated? Would they also show correlations with some of the 16 WGCNA modules?

Response 22

We agree with the reviewer that if we used a generative model like BrainSMASH, we would likely see similar low frequency patterns. However, the inserted figure in Response 7 from Makello & Misic provide evidence that is not as conservative a null as the spin test when data exhibit high spatial autocorrelation. The spatial enrichment tests carried out on the WGCNA modules are all carried out using the spin test.

Comment 23

In part (a) I commented on the possibility that brain anatomy may introduce artifactual structure into the data that's being mapped. But what if the relationship between brain geometry and brain organisation were deeper than just the introduction of artefacts? The work of Lefebre et al (2014, https://doi.org/10.1109/ICPR.2014.107; 2018, https://doi.org/10.3389/fnins.2018.00354) shows that clustering based on the 3 lowest frequency eigenvectors of the Laplacian of a brain hemisphere mesh produce an almost perfect parcellation into lobes, with remarkable coincidences between parcel boundaries and primary folds and fissures. The work of Pang et al (https://doi.org/10.1101/2022.10.04.510897) suggests that the geometry of the brain plays a critical role in constraining its dynamics: they analyse >10k task-evoked brain maps and show that the eigenvectors of the brain laplacian parsimoniously explain the activity patterns. Could brain anatomy have a downward effect on brain organisation?

Response 23

The reviewer raises a fascinating extension of our work identifying spatial modes of gene expression. We agree that these are low frequency in nature, but would first like to note that the newly introduced null model indicates that the overlaps with salient neuroanatomical features are inherent in the expression data and not purely driven by anatomy in a methodological sense.

Nevertheless we absolutely agree there is likely to be a complex multidirectional interplay between genetic expression patterns through development, developing morphology and the “final” adult topography of expression, neuroanatomical and functional patterns.

We think that the current manuscript currently contains a lot of in depth analyses of these expression data, but agree that a more extensive modeling analysis of how expression might pattern or explain functional activation would be a fascinating follow on, especially in light of these studies from Pang and Lefebre. Nevertheless we think that this must be left for a future modeling paper integrating these modes of microscale, macroscale and functional anatomy.

In Discussion: "Indeed, future work might find direct links between these module eigenvectors and similar low-frequency eigenvectors of cortical geometry have been used as basis functions to segment the cortex (Lefèvre et al. 2018) and explain complex functional activation patterns(Pang et al. 2023)."

Comment 24

On p11: ASD related to rare, deleterious mutations of strong effect is often associated with intellectual disability (where the social interaction component of ASD is more challenging to assess). Was there some indication of a relationship with that type of cognitive phenotype?

Response 24

Across the two ABIDE cohorts, the total number of those with ASD and IQ <70, which is the clinical threshold for intellectual disability was n=10, which unfortunately did not allow us to conduct a meaningful test of whether ID impacts the relationship between imaging changes in ASD and the expression maps of genes implicated in ASD by rare variants.

Comment 25

Could you clarify if the 6 donors were aligned using the folding-based method in freesurfer?

Response 25

The 6 donors were aligned using MSMsulc (Robinson et al., 2014), which is a folding based method from the HCP group. This is now clarified in the methods.

In Methods 1: "Cortical surfaces were reconstructed for each AHBA donor MRI using FreeSurfer(Fischl, 2012), and coregistered between donors using surface matching of individuals’ folding morphology (MSMSulc) (Robinson et al., 2018)."

Comment 26

The authors make available a rich resource and a series of tools to facilitate their use. They have paid attention to encode their data in standard formats, and their code was made in Python using freely accessible packages instead of proprietary alternatives such as matlab. All this should greatly facilitate the adoption of the approach. I think it would be important to state more explicitly the conceptual assumptions that the methodology brings. In the same way that a GWAS approach relies on a Mendelian idea that individual alleles encode for phenotypes, what is the idea about the organisation of the brain implied by the orthogonal gene expression modules? Is it that phenotypes - micro and macro - are encoded by linear combinations of a reduced number of gene expression patterns? What would be the role of the environment? The role of non-genic regulatory regions? Some modalities of functional organisation do not seem to be encoded by the expression of any module. Is it just for lack of data or should this be seen as the sign for a different organisational principle? Likewise, what about the aspects of disorders that are not captured by expression modules? Would that hint, for example, to stronger environmental effects? What about linear combinations of modules? Nonlinear? Overall, the authors adopt implicitly, en passant, a gene-centric conceptual standpoint, which would benefit from being more clearly identified and articulated. There are citations to Rakic's protomap idea (I would also cite the original 1988 paper, and O'Leary's 1989 "protocortex" paper stressing the role of plasticity), which proposes that a basic version of brain cytoarchitecture is genetically determined and transposed from the proliferative ventricular zone regions to the cortical plate through radial migration. In p13 the authors indicate that their results support Rakic's protomap. Additionally, in p7 the authors suggest that their results support a causal arrow going from gene expression to sulcal anatomy. The reviews by O'leary et al (2007), Ronan & Fletcher (2014, already cited), Llinares-Benadero & Borrell (2019) could be considered, which also advocate for a similar perspective. For nuances on the idea that molecular signals provide positional information for brain development, the article by Sharpe (2019, DOI: 10.1242/dev.185967) is interesting. For nuances on the gene-centric approach of the paper the articles by Rockmann (2012, DOI: 10.1111/j.1558-5646.2011.01486.x) but also from the ENCODE consortium showing the importance of non-genic regions of the genome ("Perspectives on ENCODE" 2020 DOI: 10.1038/s41586-021-04213-8) could be considered. I wouldn't ask to cite ideas from the extended evolutionary synthesis about different inheritance systems (as reviewed by Jablonka & Lamb, DOI: 10.1017/9781108685412) or the idea of inherency (Newman 2017, DOI: 10.1007/978-3-319-33038-9_78-1), but the authors may find them interesting. Same goes for our own work on mechanical morphogenesis which expands on the idea of a downward causality (Heuer and Toro 2019, DOI: 10.1016/j.plrev.2019.01.012)

Response 26

We thank the reviewer for recommending these papers, which we enjoyed reading and have deepened our thinking on the topic. In addition to toning down some of the language with respect to causality that our data cannot directly address, we have included additional discussion and references as follows:

In Discussion: "By establishing that some of these cortical zones are evident at the time of cortical folding, we lend support to a “protomap”(Rakic 1988; O'Leary 1989; O'Leary et al. 2007; Rakic et al. 2009) like model where the placement of some cortical folds is set-up by rapid tangential changes in cyto-laminar composition of the developing cortex(Ronan et al., 2014; Toro and Burnod, 2005; Van Essen, 2020). The DEMs are derived from fully folded adult donors, and therefore some of the measured genetic-folding alignment might also be induced by mechanical distortion of the tissue during folding(Llinares-Benadero and Borrell 2019; Heuer and Toro 2019). However, no data currently exist to conclusively assess the directionality of this gene-folding relationship.

Overall, the manuscript is very interesting and a great contribution. The amount of work involved is impressive, and the presentation of the results very clear. My comments indicate some aspects that could be made more clear, for example, providing additional methodological information in the supplemental material. Also, making aware the readers and future users of MAGICC of the methodological and conceptual challenges that remain to be addressed in the future for this field of research.

Reviewer #2 (Recommendations For The Authors):

Comment 1

The supplementary figures seem to be missing from the eLife submission (although I was able to find them on europepmc)

Response 1

We apologize that these were not included in the documents sent to reviewers. The up-to-date supplementary figures are included in this resubmission and again on biorxiv.

Author Response

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

eLife assessment

This important study combines genetically barcoded rabies viruses with spatial transcriptomics in vivo in the mouse brain to decode connectivity of neural circuits. The data generated by the combination of these approaches in this new way is mostly convincing as the authors provide validation and proof-of-concept that the approach can be successful. While this new combination of established techniques has promise for elucidating brain connectivity, there are still some nuances and caveats to the interpretations of the results that are lacking especially with regards to noting unexpected barcodes either due to unexpected/novel connections or unexpected rabies spread.

In this revised manuscript, we added a new control experiment and additional analyses to address two main questions from the reviewers: (1) How the threshold of glycoprotein transcript counts used to identify source cells was determined, and (2) whether the limited long-range labeling was expected in the trans-synaptic experiment. The new experiments and analyses validated the distribution of source cells and presynaptic cells observed in the original barcoded transsynaptic tracing experiment and validated the choice of the threshold of glycoprotein transcripts. As the reviewers suggested, we also included additional discussion on how future experiments can improve upon this study, including strategies to improve source cell survival and minimizing viral infection caused by leaky expression of TVA. We also provided additional clarification on the analyses for both the retrograde labeling experiment and the trans-synaptic tracing experiment. We modified the Results and Discussion sections on the trans-synaptic tracing experiment to improve clarity to general readers. Detailed changes to address specific comments by reviewers are included below.

Public Reviews:

Reviewer #1 (Public Review):

In this preprint, Zhang et al. describe a new tool for mapping the connectivity of mouse neurons. Essentially, the tool leverages the known peculiar infection capabilities of Rabies virus: once injected into a specific site in the brain, this virus has the capability to "walk upstream" the neural circuits, both within cells and across cells: on one hand, the virus can enter from a nerve terminal and infect retrogradely the cell body of the same cell (retrograde transport). On the other hand, the virus can also spread to the presynaptic partners of the initial target cells, via retrograde viral transmission.

Similarly to previously published approaches with other viruses, the authors engineer a complex library of viral variants, each carrying a unique sequence ('barcode'), so they can uniquely label and distinguish independent infection events and their specific presynaptic connections, and show that it is possible to read these barcodes in-situ, producing spatial connectivity maps. They also show that it is possible to read these barcodes together with endogenous mRNAs, and that this allows spatial mapping of cell types together with anatomical connectivity.

The main novelty of this work lies in the combined use of rabies virus for retrograde labeling together with barcoding and in-situ readout. Previous studies had used rabies virus for retrograde labeling, albeit with low multiplexing capabilities, so only a handful of circuits could be traced at the same time. Other studies had instead used barcoded viral libraries for connectivity mapping, but mostly focused on the use of different viruses for labeling individual projections (anterograde tracing) and never used a retrograde-infective virus.

The authors creatively merge these two bits of technology into a powerful genetic tool, and extensively and convincingly validate its performance against known anatomical knowledge. The authors also do a very good job at highlighting and discussing potential points of failure in the methods.

We thank the reviewer for the enthusiastic comments.

Unresolved questions, which more broadly affect also other viral-labeling methods, are for example how to deal with uneven tropism (ie. if the virus is unable or inefficient in infecting some specific parts of the brain), or how to prevent the cytotoxicity induced by the high levels of viral replication and expression, which will tend to produce "no source networks", neural circuits whose initial cell can't be identified because it's dead. This last point is particularly relevant for in-situ based approaches: while high expression levels are desirable for the particular barcode detection chemistry the authors chose to use (gap-filling), they are also potentially detrimental for cell survival, and risk producing extensive cell death (which indeed the authors single out as a detectable pitfall in their analysis). This is likely to be one of the major optimisation challenges for future implementations of these types of barcoding approaches.

As the reviewer suggested, we included additional discussion about tropism and cytotoxicity in the revised Discussion. Our sensitivity for barcode detection is sufficient, since we estimated (based on manual proofreading) that most barcoded neurons had more than ten counts of a barcode in the trans-synaptic tracing experiment. The high sensitivity may potentially allow us to adapt next-generation rabies virus with low replication, such as the third generation ΔL rabies virus (Jin et al, 2022, biorxiv) in future optimizations.

Overall the paper is well balanced, the data are well presented and the conclusions are strongly supported by the data. Impact-wise, the method is definitely going to be useful for the neurobiology research community.

We thank the reviewer for her/his enthusiasm.

Reviewer #2 (Public Review):

Although the trans-synaptic tracing method mediated by the rabies virus (RV) has been widely utilized to infer input connectivity across the brain to a genetically defined population in mice, the analysis of labeled pre-synaptic neurons in terms of cell-type has been primarily reliant on classical low-throughput histochemical techniques. In this study, the authors made a significant advance toward high-throughput transcriptomic (TC) cell typing by both dissociated single-cell RNAseq and the spatial TC method known as BARseq to decode a vast array of molecularly labeled ("barcoded") RV vector library. First, they demonstrated that a barcoded-RV vector can be employed as a simple retrograde tracer akin to AAVretro. Second, they provided a theoretical classification of neural networks at the single-cell resolution that can be attained through barcoded-RV and concluded that the identification of the vast majority (ideally 100%) of starter cells (the origin of RV-based trans-synaptic tracing) is essential for the inference of single-cell resolution neural connectivity. Taking this into consideration, the authors opted for the BARseq-based spatial TC that could, in principle, capture all the starter cells. Finally, they demonstrated the proof-of-concept in the somatosensory cortex, including infrared connectivity from 381 putative pre-synaptic partners to 31 uniquely barcoded-starter cells, as well as many insightful estimations of input convergence at the cell-type resolution in vivo. While the manuscript encompasses significant technical and theoretical advances, it may be challenging for the general readers of eLife to comprehend. The following comments are offered to enhance the manuscript's clarity and readability.

We modified the Results and Discussion sections on the trans-synaptic tracing experiment to improve clarity to general readers. We separated out the theoretical discussion about barcode sharing networks as a separate subsection, explicitly stated the rationale of how different barcode sharing networks are distinguished in the in situ trans-synaptic tracing experiment, and added additional discussion on future optimizations. Detailed descriptions are provided below.

Major points:

1. I find it difficult to comprehend the rationale behind labeling inhibitory neurons in the VISp through long-distance retrograde labeling from the VISal or Thalamus (Fig. 2F, I and Fig. S3) since long-distance projectors in the cortex are nearly 100% excitatory neurons. It is also unclear why such a large number of inhibitory neurons was labeled at a long distance through RV vector injections into the RSP/SC or VISal (Fig. 3K). Furthermore, a significant number of inhibitory starter cells in the somatosensory cortex was generated based on their projection to the striatum (Fig. 5H), which is unexpected given our current understanding of the cortico-striatum projections.

The labeling of inhibitory neurons can be explained by several factors in the three different experiments.

(1) In the scRNAseq-based retrograde labeling experiment (Fig. 2 and Fig. S3), the injection site VISal is adjacent to VISp. Because we dissected VISp for single-cell RNAseq, we may find labeled inhibitory neurons at the VISp border that extend short axons into VISal. We explained this in the revised Results.

(2) In the in situ sequencing-based retrograde labeling experiment (Fig. 3,4), the proximity between the two injection sites VISal and RSP/SC, and the sequenced areas (which included not only VISp but also RSP) could also contribute to labeling through local axons of inhibitory neurons. Furthermore, because we also sequenced midbrain regions, inhibitory neurons in the superior colliculus could pick up the barcodes through local axons. We included an explanation of this in the revised Results.

(3) In the trans-synaptic tracing experiment, we speculate that low level leaky expression from the TREtight promoter led to non-Cre-dependent expression in many neurons. To test this hypothesis, we first performed a control injection in which we saw that the fluorescent protein expression were indeed restricted to layer 5, as expected from corticostriatal labeling. Based on the labeling pattern, we estimated that about 12 copies of the glycoprotein transcript per cell would likely be needed to achieve fluorescent protein expression. Since many source cells in our experiment were below this threshold, these results support the hypothesis that the majority of source cells with low level expression of the glycoprotein were likely Cre-independent. Because these cells could still contribute to barcode sharing networks, we could not exclude them as in a conventional bulk trans-synaptic tracing experiment. In future experiments, we can potentially reduce this population by improving the helper AAV viruses used to express TVA and the glycoprotein. We included this explanation in Results and more detailed analysis in Supplementary Note 2, and discussed potential future optimizations in the Discussion. This new analysis in Supplementary Note 2 is also related to the Reviewer’s question regarding the threshold used for determining source cells (see below).

1. It is unclear as to why the authors did not perform an analysis of the barcodes in Fig. 2. Given that the primary objective of this manuscript is to evaluate the effectiveness of multiplexing barcoded technology in RV vectors, I would strongly recommend that the authors provide a detailed description of the barcode data here, including any technical difficulties or limitations encountered, which will be of great value in the future design of RV-barcode technologies. In case the barcode data are not included in Fig. 2, I would suggest that the authors consider excluding Fig. 2 and Fig. S1-S3 in their entirety from the manuscript to enhance its readability for general readers.

In the single-cell RNAseq-based retrograde tracing, all barcodes recovered matched to known barcodes in the corresponding library. We included a short description of these results in the revised manuscript.

1. Regarding the trans-synaptic tracing utilizing a barcoded RV vector in conjunction with BARseq decoding (Fig. 5), which is the core of this manuscript, I have a few specific questions/comments. First, the rationale behind defining cells with only two rolonies counts of rabies glycoprotein (RG) as starter cells is unclear. Why did the authors not analyze the sample based on the colocalization of GFP (from the AAV) and mCherry (from the RV) proteins, which is a conventional method to define starter cells? If this approach is technically difficult, the authors could provide an independent histochemical assessment of the detection stringency of GFP positive cells based on two or more colonies of RG.

In situ sequencing does not preserve fluorescent protein signals, so we used transcript counts to determine which cells expressed the glycoprotein. We have added new analyses in the Results and in Supplementary Note 2 to determine the transcript counts that were equivalent to cells that had detectable BFP expression. We found that BFP expression is equivalent to ~12 counts of the glycoprotein transcript per cell, which is much higher than the threshold we used. However, we could not solely rely on this estimate to define the source cells, because cells that had lower expression of the glycoprotein (possibly from leaky Cre-independent expression) may still pass the barcodes to presynaptic cells. This can lead to an underestimation of double-labeled and connected-source networks and an overestimation of single-source networks and can obscure synaptic connectivity at the cellular resolution. We thus used a very conservative threshold of two transcripts in the analysis. This conservative threshold will likely overestimate the number of source cells that shared barcodes and underestimate the number of single-source networks. Since this is a first study of barcoded transsynaptic tracing in vivo, we chose to err on the conservative side to make sure that the subsequent analysis has single-cell resolution. Future characterization and optimization may lead to a better threshold to fully utilize data.

Second, it is difficult to interpret the proportion of the 2,914 barcoded cells that were linked to barcoded starter cells (single-source, double-labeled, or connected-source) and those that remained orphan (no-source or lost-source). A simple table or bar graph representation would be helpful. The abundance of the no-source network (resulting from Cre-independent initial infection of the RV vector) can be estimated in independent negative control experiments that omit either Cre injection or AAV-RG injection. The latter, if combined with BARseq decoding, can provide an experimental prediction of the frequency of double-labeled events since connected-source networks are not labeled in the absence of RG.

We have added Table 2, which breaks down the 2,914 barcoded cells based on whether they are presynaptic or source cells, and which type of network they belong to. We agree with the reviewer that the additional Cre- or RG- control experiments in parallel would allow an independent estimate of the double labeled networks and the no-source networks. We have included added a discussion of possible controls to further optimize the trans-synaptic tracing approach in future studies in the Discussion.

Third, I would appreciate more quantitative data on the putative single-source network (Fig. 5I and S6) in terms of the distribution of pre- and post-synaptic TC cell types. The majority of labeling appeared to occur locally, with only two thalamic neurons observed in sample 25311842 (Fig. S6). How many instances of long-distance labeling (for example, > 500 microns away from the injection site) were observed in total? Is this low efficiency of long-distance labeling expected based on the utilized combinations of AAVs and RV vectors? A simple independent RV tracing solely detecting mCherry would be useful for evaluating the labeling efficiency of the method. I have experienced similar "less jump" RV tracing when RV particles were prepared in a single step, as this study did, rather than multiple rounds of amplification in traditional protocols, such as Osakada F et al Nat Protocol 2013.

We imaged an animal that was injected in parallel to assess labeling (now included in Supplementary Note 2 and Supp. Fig. S5). The labeling pattern in the newly imaged animal was largely consistent with the results from the barcoded experiment: most labeled neurons were seen in the vicinity of the injection site, and sparser labeling was seen in other cortical areas and the thalamus. We further found that most neurons that were labeled in the thalamus were about 1 mm posterior to the center of the injection site, and thus would not have been sequenced in the in situ sequencing experiment (in which we sequenced about 640 µm of tissue spanning the injection site).

In addition, we found that the bulk of the cells that expressed mCherry from the rabies virus only partially overlapped with the area that contained cells co-expressing BFP with the rabies glycoprotein. Moreover, very few cells co-expressed mCherry and BFP, which would be considered source cells in a conventional mono-synaptic tracing experiment. The small numbers of source cells likely also contributed to the sparseness of long-range labeling in the barcoded experiment.

These interpretations and comparisons to the barcoded experiment are now included in Supplementary Note 2.

Reviewer #3 (Public Review):

The manuscript by Zhang and colleagues attempts to combine genetically barcoded rabies viruses with spatial transcriptomics in order to genetically identify connected pairs. The major shortcoming with the application of a barcoded rabies virus, as reported by 2 groups prior, is that with the high dropout rate inherent in single cell procedures, it is difficult to definitively identify connected pairs. By combining the two methods, they are able to establish a platform for doing that, and provide insight into connectivity, as well as pros and cons of their method, which is well thought out and balanced.

Overall the manuscript is well-done, but I have a few minor considerations about tone and accuracy of statements, as well as some limitations in how experiments were done. First, the idea of using rabies to obtain broader tropism than AAVs isn't really accurate - each virus has its own set of tropisms, and it isn't clear that rabies is broader (or can be made to be broader).

As the reviewer suggested, we toned down this claim and stated that rabies virus has different tropism to complement AAV.

Second, rabies does not label all neurons that project to a target site - it labels some fraction of them.

We meant to say that retrograde labeling is not restricted to labeling neurons from a certain brain region. We have clarified in the text.

Third, the high rate of rabies virus mutation should be considered - if it is, or is not a problem in detecting barcodes with high fidelity, this should be noted.

Our analysis showed that sequencing 15 bases was sufficient to tolerate a small number of mismatches in the barcode sequences and could distinguish real barcodes from random sequences (Fig. 4A). Thus, we can tolerate mutations in the barcode sequence. We have clarified this in the text.

Fourth, there are a number of implicit assumptions in this manuscript, not all of which are equally backed up by data. For example, it is not clear that all rabies virus transmission is synaptic specific; in fact, quite a few studies argue that it is not (e.g., detection of rabies transcripts in glial cells). Thus, arguments about lost-source networks and the idea that if a cell is lost from the network, that will stop synaptic transmission, is not clear. There is also the very real propensity that, the sicker a starter cell gets, the more non-specific spread of virus (e.g., via necrosis) occurs.

We agree with the reviewer that how strictly virus transmission is restricted to synapses remains a hotly debated question in the field, and this question is relevant not only to techniques based on barcoded rabies tracing, but to all trans-synaptic tracing experiments. A barcoding-based approach can generate single-cell data that enable direct comparison to other data modalities that measure synaptic connectivity, such as multi-patch and EM. These future experiments may provide additional insights into the questions that the reviewer raised. We have included additional discussion about how non-synaptic transmission of barcodes because of the necrosis of source cells may affect the analysis in the Discussion.

Regarding the scenario in which the source cell dies, we agree with the reviewer and have clarified in the revised manuscript.

Fifth, in the experiments performed in Figure 5, the authors used a FLEx-TVA expressed via a retrograde Cre, and followed this by injection of their rabies virus library. The issue here is that there will be many (potentially thousands) of local infection events near the injection site that TVA-mediated but are Cre-dependent (=off-target expression of TVA in the absence of Cre). This is a major confound in interpreting the labeling of these cells. They may express very low levels of TVA, but still have infection be mediated by TVA. The authors did not clearly explore how expression of TVA related to rabies virus infection of cells near the rabies injection site. A modified version of TVA, such as 66T, should have been used to mitigate this issue. Otherwise, it is impossible to determine connectivity locally. The authors do not go to great lengths to interpret the findings of these observations, so I am not sure this is a critical issue, but it should be pointed out by the authors as a caveat to their dataset.

We agree with the reviewer that this type of infection could potentially be a major contributor to no-source networks, which were abundant in our experiment. Because small no-source networks were excluded from our analyses, and large no-source networks were only included for barcodes with low frequency (i.e., it would be nearly impossible statistically to generate such large no-source networks from independent infections), we believe that the effect of independent infections on our analyses were minimized. We have added a control experiment in Fig S5 and Supplementary Note 2, which further supported the hypothesis that there were many independent infections. We also included additional discussion about how this can be assessed and optimized in future studies in the Discussion.

Sixth, the authors are making estimates of rabies spread by comparison to a set of experiments that was performed quite differently. In the two studies cited (Liu et al., done the standard way, and Wertz et al., tracing from a single cell), the authors were likely infecting with a rabies virus using a high multiplicity of infection, which likely yields higher rates of viral expression in these starter cells and higher levels of input labeling. However, in these experiments, the authors need to infect with a low MOI, and explicitly exclude cells with >1 barcode. Having only a single virion trigger infection of starter cells will likely reduce the #s of inputs relative to starter neurons. Thus, the stringent criteria for excluding small networks may not be entirely warranted. If the authors wish to only explore larger networks, this caveat should be explicitly noted.

In the trans-synaptic labeling experiment, we actually used high rabies titer (200 nL, 7.6e10 iu/mL) that was comparable to conventional rabies tracing experiments. We did not exclude cells with multiple barcodes (as opposed to barcodes in multiple source cells), because we could resolve multiple barcodes in the same cell and indeed found many cells with multiple barcodes. We have clarified this in the text.

Overall, if the caveats above are noted and more nuance is added to some of the interpretation and discussion of results, this would greatly help the manuscript, as readers will be looking to the authors as the authority on how to use this technology.

In addition to addressing the specific concerns of the reviewer as described above, we modified the Results and Discussion sections on the trans-synaptic tracing experiment to improve clarity to general readers and expanded the discussion on future optimizations.

Reviewer #1 (Recommendations For The Authors):

The scientific problem is clearly stated and well laid out, the data are clearly presented, and the experiments well justified and nicely discussed. It was overall a very enjoyable read. The figures are generally nice and clear, however, I find the legends excessively concise. A bit too often, they just sort of introduce the title of the panel rather than a proper explanation of what it is depicted. A clear case is for example visible in Fig 2, where the description of the panels is minimal, but this is a general trend of the manuscript. This makes the figures a bit hard to follow as self-contained entities, without having to continuously go back to the main text. I think this could be improved with longer and more helpful descriptions.

We have revised all figure legends to make them more descriptive.

Other minor things:

In the cDNA synthesis step for in-situ sequencing, I believe the authors might have forgotten one detail: the addition of aminoallyl dUTP to the RT reaction. If I recall correctly this is done in BARseq. The fact that the authors crosslink with BS-PEG on day 2, makes me suspect they spike in these nucleotides during the RT but this is not specified in the relevant step. Perhaps this is a mistake that needs correction.

The RT primers we used have an amine group at 5’, which directly allows crosslinking. Thus, we did not need to spike in aminoallyl dUTP in the RT reaction. We have clarified this in the Methods.

Reviewer #2 (Recommendations For The Authors):

Throughout the manuscript, there are frequent references to the "Methods" section for important details. However, it can be challenging to determine which specific section of the Methods the authors are referring to, and in some cases, a thorough examination of the entire Methods section fails to locate the exact information needed to support the authors' claims. Below are a few specific examples of this issue. The authors are encouraged to be more precise in their references to the Methods section.

In the revised manuscript, we numbered each subsection of Methods and updated pointers and associated hyperlinks in the main text to the subsection numbers.

• On page 7, line 14, it is unclear how the authors compared the cell marker gene expression with the marker gene expression in the reference cell type.

We have clarified in the revised manuscript.

• On page 7, line 33, the authors note that some barcodes may have been missed during the sequencing of the rabies virus libraries, but the Methods section lacked a convincing explanation on this issue (see my point 2 above).

We included a separate subsection on the sequencing of rabies libraries and the analysis of the sequencing depth in the Methods. In this new subsection, we further clarified our reasoning for identifying the lack of sequencing depth as a reason for missing barcodes, especially in comparison to sequencing depth required for establishing exact molecule counts used in established MAPseq and BARseq techniques with Sindbis libraries.

• On page 9, line 44, the authors state that they considered a barcode to be associated with a cell if they found at least six molecules of that barcode in a cell, as detailed in the Methods section. However, the rationale behind this level of stringency is not provided in the Methods.

We initially chose this threshold based on visual inspection of the sequencing images of the barcoded cells. Because the labeled cell types were consistent with our expectations (Fig. 4E-G), we did not further optimize the threshold for detecting retrogradely labeled barcoded cells.

• I have noticed that some important explanations of figure panels are missing in the legends, making it challenging to understand the figures. Below are typical examples of this issue.

In addition to the examples that the reviewer mentioned below, we also revised many other figure panels to make them clear to the readers.

• In Fig. 2, "RV into SC" in panel C does not make sense, as RV was injected into the thalamus. There is no explanation of the images in this panel C.

We have corrected the typo in the revision.

• In Fig. 3, information on the endogenous gene panel for cell type classification (Table S3) could be mentioned in the legend or corresponding text.

We now cite Table S3 both in Fig 3 legend and in the main text. We also included a list of the 104 cell type marker genes we used in Table S3.

• In panel J, it is unclear why the total number of BC cells is 2,752, and not 4,130 as mentioned in the text.

This is a typo. We have corrected this in the revision. The correct number (3,746) refers to the number of cells that did not belong to either of the two categories at the bottom of the panel, and not the total number of neurons. To make this clear, we now also include the total number of barcoded cells at the top of the panel.

• In Fig. 4, the definitions of "+" and "−" symbols in panels K and L are unclear. Also, it seems that the second left column of panel K should read "T −."

We corrected the typo in K, further clarified the “Area” labels, and changed the “S” label in 4K to “−”. This change does not change the original meaning of the figure: when considering the variance explained in L4/5 IT neurons, considering the subclass compositional profile is equivalent to not using the compositional profiles of cell types, because L4/5 IT neurons all belong to the same subclass (L4/5 IT subclass). Although operationally we simply considered subclass-level compositional profiles when calculating the variance explained, we think that changing this to “−” is clearer for the readers.

• In Fig. 5, panel E is uninterpretable.

We revised the main text and the figure to clarify how we manually proofread cells to determine the QC thresholds for barcoded cells. These plots showed a summary of the proofreading. We also revised the figures to indicate that they showed the fraction of barcoded cells that were considered real after proofreading. In the revised version, we moved these plots to Fig. S5.

• In Fig. S1, I do not understand the identity of the six samples on the X-axis of panel A (given that only two animals were described in the main text) and what panel B shows, including the definition of map_cluster_conf and map_cluster_corr.

In the revised Fig. S1, we made it more explicit that the six animals include both animals used for retrograde tracing (2 animals) and those used for trans-synaptic tracing (4 animals). We updated the y axis labels to be more readable and cited the relevant Methods section for definitions.

• In Fig. S2, please provide the definitions of blue and red dots and values in panel A, as well as the color codes and size of the circles in panel B. My overall impression from panel B is that there is no significant difference between RV-infected and non-infected cells. The authors should provide more quantitative and statistical support for the claim that "RV-infected cells had higher expression of immune response-related genes."

We toned down the statement to “Consistent with previous studies […], some immune response related genes were up-regulated in virus-infected cells compared to non-infected cells.” Because the main point of the single-cell RNAseq analysis was that rabies did not affect the ability to distinguish transcriptomic types, the change in immune response-related genes was not essential to the main conclusions. We clarified the red and blue dots in panel A and changed panel B to show the top up-regulated immune response-related genes in the revised manuscript.

• In Fig. S3, the definitions of the color code and circle size are missing.

We have added the legends in Fig. S3.

Author Response

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

We appreciate the reviewers for their insightful feedback, which has substantially improved our manuscript. Following the suggestions of the reviewers, we have undertaken the following major revisions:

a. Concerning data transformation, we have adjusted the methodology in Figures 2 and 3. Instead of normalizing c-Fos density to the whole brain c-Fos density as initially described, we now normalize to the c-Fos density of the corresponding brain region in the control group. b. We have substituted the PCA approach with hierarchical clustering in Figures 2 and 3.

c. In the discussion section, we added a subsection on study limitations, focusing on the variations in drug administration routes and anesthesia depth.

Enclosed are our detailed responses to each of the reviewer's comments.

Reviewer #1:

1a. The addition of the EEG/EMG is useful, however, this information is not discussed. For instance, there are differences in EEG/EMG between the two groups (only Ket significantly increased delta/theta power, and only ISO decreased EMG power). These results should be discussed as well as the limitation of not having physiological measures of anesthesia to control for the anesthesia depth.

1b. The possibility that the differences in fos observed may be due to the doses used should be discussed.

1c. The possibility that the differences in fos observed may be due kinetic of anesthetic used should be discussed.

Thank you for your suggestions. We have now discussed EEG/EMG result, limitation of not having physiological measures of anesthesia to control for the anesthesia depth, The possibility that the differences in fos observed may be due to the doses, The possibility that the differences in Fos observed may be due kinetic of anesthetic in the revised manuscript (Lines 308-331, also shown below).

Lines 308-331: "...Our findings indicate that c-Fos expression in the KET group is significantly elevated compared to the ISO group, and the saline group exhibits notably higher c-Fos expression than the home cage group, as seen in Supplementary Figures 2 and 3. Intraperitoneal saline injections in the saline group, despite pre-experiment acclimation with handling and injections for four days, may still evoke pain and stress responses in mice. Subtle yet measurable variations in brain states between the home cage and saline groups were observed, characterized by changes in normalized EEG delta/theta power (home cage: 0.05±0.09; saline: -0.03±0.11) and EMG power (home cage: -0.37±0.34; saline: 0.04±0.13), as shown in Supplementary Figure 1. These changes suggest a relative increase in overall brain activity in the saline group compared to the home cage group, potentially contributing to the higher c-Fos expression. Although the difference in EEG power between the ISO group and the home cage control was not significant, the increase in EEG power observed in the ISO group was similar to that of KET (0.47 ± 0.07 vs 0.59 ± 0.10), suggesting that both agents may induce loss of consciousness in mice. Regarding EMG power, ISO showed a significant decrease in EMG power compared to its control group. In contrast, the KET group showed a lesser reduction in EMG power (ISO: -1.815± 0.10; KET: -0.96 ± 0.21), which may partly explain the higher overall c-Fos expression levels in the KET group. This is consistent with previous studies where ketamine doses up to 150 mg/kg increase delta power while eliciting a wakefulness-like pattern of c-Fos expression across the brain [1]. Furthermore, the observed differences in c-Fos expression may arise in part from the dosages, routes of administration, and their distinct pharmacokinetic profiles. This variation is compounded by the lack of detailed physiological monitoring, such as blood pressure, heart rate, and respiration, affecting our ability to precisely assess anesthesia depth. Future studies incorporating comprehensive physiological monitoring and controlled dosing regimens are essential to further elucidate these relationships and refine our understanding of the effects of anesthetics on brain activity"

1. Lu J, Nelson LE, Franks N, Maze M, Chamberlin NL, Saper CB: Role of endogenous sleep-wake and analgesic systems in anesthesia. J Comp Neurol 2008, 508(4):648-662.

2b. I am confused because Fig 2C seems to show significant decrease in %fos in the hypothalamus, midbrain and cerebellum after KET, while the author responded that " in our analysis, we did not detect regions with significant downregulation when comparing anesthetized mice with controls." Moreover the new figure in the rebuttal in response to reviewer 2 suggests that Ket increases Fos in almost every single region (green vs blue) which is not the conclusion of the paper.

Your concern regarding the apparent discrepancy is well-founded. The inconsistency arose due to an inappropriate data transformation, which affected the interpretation. We have now rectified this by adjusting the data transformation in Figures 2 and 3. Specifically, we have recalculated the log relative c-Fos density values relative to the control group for each brain region. This revision has resolved the issue, confirming that our analysis did not detect any regions with significant downregulation in the anesthetized mice compared to controls. We have also updated the results, discussion, and methods sections of Figures 2 and 3 to accurately reflect these changes and ensure consistency with our findings.

Author response image 1.

Figure 2. Whole-brain distributions of c-Fos+ cells induced by ISO and KET. (A) Hierarchical clustering was performed on the log relative c-Fos density data for ISO and KET using the complete linkage method based on the Euclidean distance matrix, with clusters identified by a dendrogram cut-off ratio of 0.5. Numerical labels correspond to distinct clusters within the dendrogram. (B) Silhouette values plotted against the ratio of tree height for ISO and KET, indicating relatively higher Silhouette values at 0.5 (dashed line), which is associated with optimal clustering. (C) The number of clusters identified in each treatment condition at different ratios of the dendrogram tree height, with a cut-off level of 0.5 corresponding to 4 clusters for both ISO and KET (indicated by the dashed line). (D) The bar graph depicts Z scores for clusters in ISO and KET conditions, represented with mean values and standard errors. One-way ANOVA with Tukey's post hoc multiple comparisons. ns: no significance; ***P < 0.001. (E) Z-scored log relative density of c-Fos expression in the clustered brain regions. The order and abbreviations of the brain regions and the numerical labels correspond to those in Figure 2A. The red box denotes the cluster with the highest mean Z score in comparison to other clusters. CTX: cortex; TH: thalamus; HY: hypothalamus; MB: midbrain; HB: hindbrain.

Author response image 2.

Figure 3. Similarities and differences in ISO and KET activated c-Fos brain areas. (A) Hierarchical clustering was performed on the log-transformed relative c-Fos density data for ISO and KET using the complete linkage method based on the Euclidean distance matrix, with clusters identified by a dendrogram cut-off ratio of 0.5. (B) Silhouette values are plotted against the ratio of tree height from the hierarchical clustered dendrogram in Figure 3A. (C) The relationship between the number of clusters and the tree height ratio of the dendrogram for ISO and KET, with a cut-off ratio of 0.5 resulting in 3 clusters for ISO and 5 for KET (indicated by the dashed line). (D) The bar graph depicts Z scores for clusters in ISO and KET conditions, represented with mean values and standard errors. One-way ANOVA with Tukey's post hoc multiple comparisons. ns: no significance; ***P < 0.001. (E) Z-scored log relative density of c-Fos expression within the identified brain region clusters. The arrangement, abbreviations of the brain regions, and the numerical labels are in accordance with Figure 3A. The red boxes highlight brain regions that rank within the top 10 percent of Z score values. The white boxes denote brain regions with an Z score less than -2.

1. There are still critical misinterpretations of the PCA analysis. For instance, it is mentioned that " KET is associated with the activation of cortical regions (as evidenced by positive PC1 coefficients in MOB, AON, MO, ACA, and ORB) and the inhibition of subcortical areas (indicated by negative coefficients) " as well as " KET displays cortical activation and subcortical inhibition, whereas ISO shows a contrasting preference, activating the cerebral nucleus (CNU) and the hypothalamus while inhibiting cortical areas. To reduce inter-individual variability." These interpretations are in complete contradiction with the answer 2b above that there was no region that had decreased Fos by either anesthetic.

Thank you for bringing this to our attention. In response to your concerns, we have made significant revisions to our data analysis. We have updated our input data to incorporate log-transformed relative c-Fos density values, normalized against the control group for each brain region, as illustrated in Figures 2 and 3. Instead of PCA, we have applied this updated data to hierarchical clustering analysis. The results of these analyses are consistent with our original observation that neither anesthetic led to a decrease in Fos expression in any region.

1. I still do not understand the rationale for the use of that metric. The use of a % of total Fos makes the data for each region dependent on the data of the other regions which wrongly leads to the conclusion that some regions are inhibited while they are not when looking at the raw data. Moreover, the interdependence of the variable (relative density) may affect the covariance structure which the PCA relies upon. Why not using the PCA on the logarithm of the raw data or on a relative density compared to the control group on a region-per-region basis instead of the whole brain?

Thank you for your insightful suggestion. Following your advice, we have revised our approach and now utilize the logarithm of the relative density compared to the control group on a region-by-region basis. We attempted PCA analyses using the logarithm of the raw data, the logarithm of the Z-score, and the logarithm of the relative density compared to control, but none yielded distinct clusters.

Author response image 3.

As a result, we employed hierarchical cluster analysis. We then examined the Z-scores of the log-transformed relative c-Fos densities (Figures 2E and 3E) to assess expression levels across clusters. Our analysis revealed that neither ISO nor KET treatments led to a significant suppression of c-Fos expression in the 53 brain regions examined. In the ISO group alone, there were 10 regions that demonstrated relative suppression (Z-score < -2, indicated by white boxes) as shown in Figure 3.

Fig. 2B: it's unclear to me why the regions are connected by a line. Such representation is normally used for time series/within-subject series. What is the rationale for the order of the regions and the use of the line? The line connecting randomly organized regions is meaningless and confusing.

Thank you for your suggestion. We have discontinued the use of PCA calculations and have removed this figure.

Fig 6A. The correlation matrices are difficult to interpret because of the low resolution and arbitrary order of brain regions. I recommend using hierarchical clustering and/or a combination of hierarchical clustering and anatomical organization (e.g. PMID: 31937658). While it is difficult to add the name of the regions on the graph I recommend providing supplementary figures with large high-resolution figures with the name of each brain region so the reader can actually identify the correlation between specific brain regions and the whole brain, Rationale for Metric Choice: Note that I do not dispute the choice of the log which is appropriate, it is the choice of using the relative density that I am questioning.

Thank you for your constructive feedback. In line with your suggestion, we have implemented hierarchical clustering combined with anatomical organization as per the referenced literature. Additionally, we have updated the vector diagrams in Figure 6A to present them with greater clarity.

Furthermore, we have revised our network modular division method based on cited literature recommendations. We used hierarchical clustering with correlation coefficients to segment the network into modules, illustrated in Figure 6—figure supplement 1. Due to the singular module structure of the KET network and the sparsity of intermodular connections in the home cage and saline networks, the assessment of network hub nodes did not employ within-module degree Z-score and participation coefficients, as these measures predominantly underscore the importance of connections within and between modules. Instead, we used degree, betweenness centrality, and eigenvector centrality to detect the hub nodes, as detailed in Figure 6—figure supplement 2. With this new approach, the hub node for the KET condition changed from SS to TeA. Corresponding updates have been made to the results section for Figure 6, as well as to the related discussions and the abstract of our paper.

Author response image 4.

Figure 6. Generation of anesthetics-induced networks and identification of hub regions. (A) Heatmaps display the correlations of log c-Fos densities within brain regions (CTX, CNU, TH, HY, MB, and HB) for various states (home cage, ISO, saline, KET). Correlations are color-coded according to Pearson's coefficients. The brain regions within each anatomical category are organized by hierarchical clustering of their correlation coefficients. (B) Network diagrams illustrate significant positive correlations (P < 0.05) between regions, with Pearson’s r exceeding 0.82. Edge thickness indicates correlation magnitude, and node size reflects the number of connections (degree). Node color denotes betweenness centrality, with a spectrum ranging from dark blue (lowest) to dark red (highest). The networks are organized into modules consistent with the clustering depicted in Supplementary Figure 8. Figure 6—figure supplement 1

Author response image 5.

Figure 6—figure supplement 1. Hierarchical clustering of brain regions under various conditions: home cage, ISO, saline, and KET. (A) Heatmaps show the relative distances among brain regions assessed in naive mice. Modules were identified by sectioning each dendrogram at a 0.7 threshold. (B) Silhouette scores plotted against the dendrogram tree height ratio for each condition, with optimal cluster definition indicated by a dashed line at a 0.7 ratio. (C) The number of clusters formed at different cutoff levels. At a ratio of 0.7, ISO and saline treatments result in three clusters, whereas home cage and KET conditions yield two clusters. (D) The mean Pearson's correlation coefficient (r) was computed from interregional correlations displayed in Figure 6A. Data were analyzed using one-way ANOVA with Tukey’s post hoc test, ***P < 0.001.

Author response image 6.

Figure 6—figure supplement 2. Hub region characterization across different conditions: home cage (A), ISO (B), saline (C), and KET (D) treatments. Brain regions are sorted by degree, betweenness centrality, and eigenvector centrality, with each metric presented in separate bar graphs. Bars to the left of the dashed line indicate the top 20% of regions by rank, highlighting the most central nodes within the network. Red bars signify regions that consistently appear within the top rankings for both degree and betweenness centrality across the metrics.

1. I am still having difficulties understanding Fig. 3.

Panel A: The lack of identification for the dots in panel A makes it impossible to understand which regions are relevant.

Panel B: what is the metric that the up/down arrow summarizes? Fos density? Relative density? PC1/2?

Panel C: it's unclear to me why the regions are connected by a line. Such representation is normally used for time series/within-subject series. What is the rationale for the order of the regions?

Thank you for your patience and for reiterating your concerns regarding Figure 3.

a. In Panel A, we have substituted the original content with a display of hierarchical clustering results, which now clearly marks each brain region. This change aids readers in identifying regions with similar expression patterns and facilitates a more intuitive understanding of the data.

a. Acknowledging that our analysis did not reveal any significantly inhibited brain regions, we have decided to remove the previous version of Panel B from the figure.

b. We have discontinued the use of PCA calculations and have removed this figure to avoid any confusion it may have caused. Our revised analysis focuses on hierarchical clustering, which are presented in the updated figures.

Reviewer #2:

1. Aside from issues with their data transformation (see below), (a) I think they have some interesting Fos counts data in Figures 4B and 5B that indicate shared and distinct activation patterns after KET vs. ISO based anesthesia. These data are far closer to the raw data than PC analyses and need to be described and analyzed in the first figures long before figures with the more abstracted PC analyses. In other words, you need to show the concrete raw data before describing the highly transformed and abstracted PC analyses. (b) This gets to the main point that when selecting brain areas for follow up analyses, these should be chosen based on the concrete Fos counts data, not the highly transformed and abstracted PC analyses.

Thank you for your suggestions.

a. We have added the original c-Fos cell density distribution maps for Figures 2, 3, 4, and 5 in Supplementary Figures 2 and 3 (also shown below). To maintain consistency across the document, we have updated both the y-axis label and the corresponding data in Figures 4B and 5B from 'c-Fos cell count' to 'c-Fos density'.

b. The analyses in Figures 2 and 3 include all brain regions. Figures 4 and 5 present the brain regions with significant differences as shown in Figure 3—figure supplement 1.

Author response image 7.

Figure 2—figure supplement 1. The c-Fos density in 53 brain areas for different conditions. (home cage, n = 6; ISO, n = 6 mice; saline, n = 8; KET, n = 6). Each point represents the c-Fos density in a specific brain region, denoted on the y-axis with both abbreviations and full names. Data are shown as mean ± SEM. Brain regions are categorized into 12 brain structures, as indicated on the right side of the graph.

Author response image 8.

Figure 3—figure supplement 1. c-Fos density visualization across 201 distinct brain regions under various conditions. The graph depicts the c-Fos density levels for each condition, with data presented as mean and standard error. Brain regions with statistically significant differences are featured in Figures 4 and 5. Brain regions are organized into major anatomical subdivisions, as indicated on the left side of the graph.

1. Now, the choice of data transformation for Fos counts is the most significant problem. First, the authors show in the response letter that not using this transformation (region density/brain density) leads to no clustering. However, they also showed the region-densities without transformation (which we appreciate) and it looks like overall Fos levels in the control group Home (ISO) are a magnitude (~10-fold) higher than those in the control group Saline (KET) across all regions shown. This large difference seems unlikely to be due to a biologically driven effect and seems more likely to be due to a technical issue, such as differences in staining or imaging between experiments. Was the Homecage-ISO experiment or at least the Fos labeling and imaging performed at the same time as for the Saline-Ketamine experiment? Please state the answer to this question in the Results section one way or the other.

a. “Home (ISO) are a magnitude (~10-fold) higher than those in the control group saline (KET) across all regions shown.” We believe you might be indicating that compared to the home cage group (gray), the saline group (blue) shows a 10-fold higher expression (Supplementary Figure 2/3). Indeed, we observed that the total number of c-Fos cells in the home cage group is significantly lower than in the saline group. This difference may be due to reduced sleep during the light-on period (ZT 6- ZT 7.5) in the saline mice or the pain and stress response caused by intraperitoneal injection of saline. We have explained this discrepancy in the discussion section.Line 308-317（also see below）

“…Our findings indicate that c-Fos expression in the KET group is significantly elevated compared to the ISO group, and the saline group exhibits notably higher c-Fos expression than the home cage group, as seen in Supplementary Figures 2 and 3. Intraperitoneal saline injections in the saline group, despite pre-experiment acclimation with handling and injections for four days, may still evoke pain and stress responses in mice. Subtle yet measurable variations in brain states between the home cage and saline groups were observed, characterized by changes in normalized EEG delta/theta power (home cage: 0.05±0.09; saline: -0.03±0.11) and EMG power (home cage: -0.37±0.34; saline: 0.04±0.13), as shown in Figure 1—figure supplement 1. These changes suggest a relative increase in overall brain activity in the saline group compared to the home cage group, potentially contributing to the higher c-Fos expression…”

b. Drug administration and tissue collection for both Homecage-ISO and Saline-Ketamine groups were consistently scheduled at 13:00 and 14:30, respectively. Four mice were administered drugs and had tissues collected each day, with two from the experimental group and two from the control group, to ensure consistent sampling. The 4% PFA fixation time, sucrose dehydration time, primary and secondary antibody concentrations and incubation times, staining, and imaging parameters and equipment (exposure time for VS120 imaging was fixed at 100ms) were all conducted according to a unified protocol.

We have included the following statement in the results section: Line 81-83, “Sample collection for all mice was uniformly conducted at 14:30 (ZT7.5), and the c-Fos labeling and imaging were performed using consistent parameters throughout all experiments. ”

1. Second, they need to deal with this large difference in overall staining or imaging for these two (Home/ISO and Saline/KET) experiments more directly; their current normalization choice does not really account for the large overall differences in mean values and variability in Fos counts (e.g. due to labeling and imaging differences).

3a. I think one option (not perfect but I think better than the current normalization choice) could be z-scoring each treatment to its respective control. They can analyze these z-scored data first, and then in later figures show PC analyses of these data and assess whether the two treatments separate on PC1/2. And if they don't separate, then they don't separate, and you have to go with these results.

3b. Alternatively, they need to figure out the overall intensity distributions from the different runs (if that the main reason of markedly different counts) and adjust their thresholds for Fos-positive cell detection based on this. I would expect that the saline and HC groups should have similar levels of activation, so they could use these as the 'control' group to determine a Fos-positive intensity threshold that gets applied to the corresponding 'treatment' group.

3c. If neither 3a nor 3b is an option then they need to show the outcomes of their analysis when using the untransformed data in the main figures (the untransformed data plots in their responses to reviewer are currently not in the main or supplementary figs) and discuss these as well.

a. Thank you very much for your valuable suggestion. We conducted PCA analysis on the ISO and KET data after Z-scoring them with their respective control groups and did not find any significant separation.

Author response image 9.

As mentioned in our response to reviewer #1, we have reprocessed the raw data. Firstly, we divided the ISO and KET data by their respective control brain regions and then performed a logarithmic transformation to obtain the log relative c-Fos density. The purpose of this is to eliminate the impact of baseline differences and reduce variability. We then performed hierarchical clustering, and finally, we Z-scored the log relative c-Fos density data. The aim is to facilitate comparison of ISO and KET on the same data dimension (Figure 2 and 3).

b. We appreciate your concerns regarding the detection thresholds for Fos-positive cells. The enclosed images, extracted from supplementary figures for Figures 4 and 5, demonstrate notable differences in c-Fos expression between saline and home cage groups in specific brain regions. These regions exhibit a discernible difference in staining intensity, with the saline group showing enhanced c-Fos expression in the PVH and PVT regions compared to the home cage group. An examination of supplementary figures for Figures 4 and 5 shows that c-Fos expression in the home cage group is consistently lower than in the saline group. This comparative analysis confirms that the discrepancies in c-Fos levels are not due to varying detection thresholds.

Author response image 10.

b. We have added the corresponding original data graphs to Supplementary Figures 2 and 3, and discussed the potential reasons for the significant differences between these groups in the discussion section (also shown below).

Lines 308-317: "...Our findings indicate that c-Fos expression in the KET group is significantly elevated compared to the ISO group, and the saline group exhibits notably higher c-Fos expression than the home cage group, as seen in Supplementary Figures 2 and 3. Intraperitoneal saline injections in the saline group, despite pre-experiment acclimation with handling and injections for four days, may still evoke pain and stress responses in mice. Subtle yet measurable variations in brain states between the home cage and saline groups were observed, characterized by changes in normalized EEG delta/theta power (home cage: 0.05±0.09; saline: -0.03±0.11) and EMG power (home cage: -0.37±0.34; saline: 0.04±0.13), as shown in Figure 3—figure supplement 1. These changes suggest a relative increase in overall brain activity in the saline group compared to the home cage group, potentially contributing to the higher c-Fos expression.…”

Author Response

We thank the reviewers for their detailed and constructive criticisms of our work. They raise many important questions (such as the issue of defining context) that we have also been thinking about extensively and they provide new and insightful avenues that have the potential to meaningfully improve the manuscript. We also appreciate that they commented on the novelty and importance of this work. Going forward, we will address the methodological concerns raised as best as we can and thereby hope to make the evidence for our conclusion more compelling

Author Response

eLife assessment

This study provides direct evidence showing that Kv1.8 channels underly several potassium currents in the two types of sensory hair cells found in the mouse vestibular system. This is an important finding because the nature of the channels underpinning the unusual potassium conductance gK,L in type I hair cells has been under scrutiny for many years. Although most of the experimental evidence is compelling and the analysis is rigorous, the evidence supporting some of the claims related to Kv1.4 channels is incomplete. The study will be of interest to cell and molecular biologists and auditory neuroscientists.

We are thankful to the editor and reviewers for their thorough assessment of our work and insightful feedback. Our responses to the comments and suggestions are below.

Reviewer #1 (Public Review):

Summary:

In this paper, the authors provide a thorough demonstration of the role that one particular type of voltage-gated potassium channel, Kv1.8, plays in a low voltage-activated conductance found in type I vestibular hair cells. Along the way, they find that this same channel protein appears to function in type II vestibular hair cells as well, contributing to other macroscopic conductances. Overall, Kv1.8 may provide especially low input resistance and short time constants to facilitate encoding of more rapid head movements in animals that have necks. Combination with other channel proteins, in different ratios, may contribute to the diversified excitability of vestibular hair cells.

Strengths:

The experiments are comprehensive and clearly described, both in the text and in the figures. Statistical analyses are provided throughout.

Weaknesses:

None.

Reviewer #2 (Public Review):

The focus of this manuscript was to investigate whether Kv1.8 channels, which have previously been suggested to be expressed in type I hair cells of the mammalian vestibular system, are responsible for the potassium conductance gK,L. This is an important study because gK,L is known to be crucial for the function of type I hair cells, but the channel identity has been a matter of debate for the past 20 years. The authors have addressed this research topic by primarily investigating the electrophysiological properties of the vestibular hair cells from Kv1.8 knockout mice. Interestingly, gK,L was completely abolished in Kv1.8-deficient mice, in agreement with the hypothesis put forward by the authors based on the literature. The surprising observation was that in the absence of Kv1.8 potassium channels, the outward potassium current in type II hair cells was also largely reduced. Type II hair cells express the largely inactivating potassium conductance gK,A, but not gK,L. The authors concluded that heteromultimerization of non-inactivating Kv1.8 and the inactivating Kv1.4 subunits could be responsible for the inactivating gK,A. Overall, the manuscript is very well written and most of the conclusions are supported by the experimental work. The figures are well described, and the statistical analysis is robust.

My only comment relates to the statement regarding the results providing "evidence" that Kv1.4 form heteromultimers with Kv1.8 channels (see Discussion). The only data I can see from the results is that Kv1.4 channels are expressed in the membrane of type II hair cells, which is not sufficient evidence for the above claim. Is the distribution of Kv1.8 and Kv1.4 overlapping in type II hair cells? Have the authors attempted to perform some pharmacological studies on Kv1.4? For example, would gK,A be completely blocked by a Kv1.4 antagonist? Addressing at least some of these questions would strengthen your argument.

Author response: With respect to the “evidence” for heteromultimerization of Kv1.4 and Kv1.8: We agree that there is not conclusive evidence but have pulled together reasons to suggest that the fast inactivation of Kv1.8-dependent gA in type II hair cells reflects a contribution from Kv1.4 subunits. The reasons we note are mostly from other sources: 1) Kv1.4 subunits are the only Kv1 alpha subunits known to make channels with intrinsic rapid inactivation (Bertoli et al., 1994); 2) Kv1.4 is highly expressed in type II hair cells, but not type I hair cells, in mouse utricle (McInturff et al., Biol. Open., 2018; Jan et al., Cell Reports, 2021; Orvis et al., Nat. Methods, 2021); 3) previous work from M. Correia and colleagues suggested Kv1.4 as the likely source of A-current in pigeon vestibular hair cells; 4) some rat type II hair cells show comparatively strong Kv1.4-like immunoreactivity (our Fig. 5). While we consider heteromultimerization of Kv1.4 and Kv1.8 alpha subunits a plausible explanation consistent with available data from different sources, we agree that the question is not at all settled, and indeed raise the possibility that KV beta subunits, which are also differentially expressed by type I and II hair cells, play a role. Experiments to definitively advance or refute this hypothesis are beyond the scope of this paper.

Reviewer #3 (Public Review):

Summary:

This paper by Martin et al. describes the contribution of a Kv channel subunit (Kv1.8, KCNA10) to voltage-dependent K+ conductances and membrane properties of type I and type II hair cells of the mouse utricle. Previous work has documented striking differences in K+ conductances between vestibular hair cell types. In particular, amniote type I hair cells are known to express a non-typical low-voltage-activated K+ conductance (GK,L) whose molecular identity has been elusive. K+ conductances in hair cells from 3 different mouse genotypes (wildtype, Kv1.8 homozygous knockouts, and heterozygotes) are examined here and whole-cell patch-clamp recordings indicate a prominent role for Kv1.8 subunits in generating GK,L. Results also interestingly support a role for Kv1.8 subunits in type II hair cell K+ conductances; inactivating conductances in null mice are reduced in type II hair cells from striola and extrastriola regions of the utricle. Kv1.8 is therefore proposed to contribute as a pore-forming subunit for 3 different K+ conductances in vestibular hair cells. The impact of these conductances on membrane responses to current steps is studied in the current clamp. Pharmacological experiments use XE991 to block some residual Kv7-mediated current in both hair cell types, but no other pharmacological blockers are used. In addition, immunostaining data are presented and raise some questions about Kv7 and Kv1.8 channel localization. Overall, the data present compelling evidence that the removal of Kv1.8 produces profound changes in hair cell membrane conductances and sensory capabilities. These changes at hair cell level suggest vestibular function would be compromised and further assessment in terms of balance behavior in the different mice would be interesting.

Strengths:

This study provides strong evidence that Kv1.8 subunits are major contributors to the unusual K+ conductance in type I hair cells of the utricle. It also indicates that Kv1.8 subunits are important for type II hair cell K+ conductances because Kv1.8-/- mice lacked an inactivating A conductance and had reduced delayed rectifier conductance compared to controls. A comprehensive and careful analysis of biophysical profiles is presented of expressed K+ conductances in 3 different mouse genotypes. Voltage-dependent K+ currents are rigorously characterized at a range of different ages and their impact on membrane voltage responses to current input is studied. Some pharmacological experiments are performed in addition to immunostaining to bolster the conclusions from the biophysical studies. The paper has a significant impact in showing the role of Kv1.8 in determining utricular hair cell electrophysiological phenotypes.

Weaknesses:

1. From previous work it is known that GK,L in type I hair cells has unusual ion permeation and pharmacological properties that differ greatly from type II hair cell conductances. Notably GK,L is highly permeable to Cs+ as well as K+ ions and is slightly permeable to Na+. It is blocked by 4-aminopyridine and divalent cations (Ba2+, Ca2+, Ni2+), enhanced by external K+, and modulated by cyclic GMP. The question arises, if Kv1.8 is a major player and pore-forming subunit in type I and type II cells (and cochlear inner hair cells as shown by Dierich et al. 2020) how are subunits modified to produce channels with very different properties? A role for Kv1.4 channels (gA) is proposed in type II hair cells based on previous findings in bird hair cells and immunostaining for Kv1.4 channels in rat utricle presented here in Fig. 6. However, hair cell-specific partner interactions with Kv1.8 that result in GK,L in type I hair cells and Cs+ impermeable, inactivating currents in type II hair cells remain for the most part unexplored.

Author response: Our results raise the question of how Kv1.8/Kcna10 is regulated to produce gK,L in type I hair cells, which has different properties from the Kv1.8 conductance expressed heterologously (Lang et al., Am. J. Physiol. Renal Physiol., 2000; Ranjan et al., Front. Cell. Neurosci., 2019; Dierich et al., Cell Reports, 2020) and the Kv1.8 conductance inferred in inner hair cells (Dierich et al., 2020). We lay out several possibilities in the Discussion, but testing these suggestions is beyond the scope of the present paper.

The relatively high Cs+ permeability of gK,L (0.31 pCs/pK, Rüsch & Eatock, J. Neurophysiol., 1996; Rennie & Correia, J. Membr. Biol., 2000) suggests there is something different about the selectivity filter and pore region of gK,L relative to most Kv1 family members. Although the intrinsic Cs+ permeability of heterologously expressed Kv1.8 is not reported. While we note that the pore region in Kv1.8 differs from other Kv1 subunits by a single amino acid (a glycine instead of alanine at position 411 – placed by AlphaFold in the pore helix of hKCNA10, Jumper et al., Nature, 2021), the effect of this difference is not known. A separate study is needed to determine why gK,L has a high Cs+ permeability relative to other Kv channels.

For type II hair cells, the Cs+ permeability of Kv currents has not been fully characterized. Internal Cs+ does appear to reduce outward current more effectively in type II hair cells (Lang & Correia, J. Neurophysiol., 1989; Sokolowski et al., Dev. Biol., 1993) than in type I hair cells (Rüsch & Eatock, J. Neurophysiol., 1996; Rennie & Correia, J. Membr. Biol., 2000).

With respect to cochlear inner hair cells, note that the assignment of Kv1.8 by Dierich et al. (2021) to a delayed rectifier in cochlear inner hair cells (IHCs) was based on inference – that is, existing inner ear expression databases show that Kv1.8 is expressed in IHCs, and heterologous Kv1.8 channels have a current resembling that observed in IHCs after block of multiple other K channels. We agree with Dierich et al. that Kv1.8 is an attractive candidate for the residual conductance in cochlear IHCs based on comparison with its properties in heterologous expression data. Together their study and our study suggest that Kv1.8 takes on quite different voltage dependence depending on the hair cell environment, and it will be an interesting challenge to sort out the reasons.

1. Data from patch-clamp and immunocytochemistry experiments are not in close alignment. XE991 (Kv7 channel blocker) decreases remaining K+ conductance in type I and type II hair cells from null mice supporting the presence of Kv7 channels in hair cells (Fig. 7). Also, Holt et al. (2007) previously showed inhibition of GK,L in type I hair cells (but not delayed rectifier conductance in type II hair cells) using a dominant negative construct of Kv7.4 channels. However, immunolabelling indicates Kv7.4 channels on the inner face of calyx terminals adjacent to hair cells (Fig. 5). Some reconciliation of these findings is needed.

Author response: Our pharmacology with XE991 suggests a small but significant population of Kv7 channels in type I and II hair cells (Fig 7). With the immunogold technique, Kharkovets et al. (PNAS, 2000) and Hurley et al. (J. Neurosci., 2006) counted significant Kv7.4 particles in type I hair cells, although the particles occurred at much greater density in the postsynaptic calyx membrane facing the hair cell. These results lead us to propose that the Kv7 channel we identified pharmacologically includes the Kv7.4 subunit, possibly in combination with other Kv7 subunits (Lysakowski et al., J. Neurosci., 2011). By this argument, the absence of clear hair cell staining in the confocal images of Fig. 5A is likely to reflect differences in methods, which include the use of different mouse strains, different sensitivities of immunogold vs. confocal imaging, and different antibodies.

Holt et al. (J. Neurosci., 2007) indeed saw inhibition of gK,L in hair cells grown in organotypic cultures of the neonatal mouse utricle after viral expression of a dominant negative Kv7.4 construct. However, other studies show that Kv7 antagonists do not block gK,L (Hurley et al., J. Neurosci., 2006), and the Jentsch group, which first proposed Kv7.4 as a likely candidate for gK,L (Kharkovets et al., PNAS, 2000), ultimately showed that knocking out Kv7.4 and Kv7.5 expression failed to eliminate gK,L (Spitzmaul et al., J. Biol. Chem., 2013). Together, these results suggest that in Holt et al. (2007), the inhibition of gK,L by transfection with the dominant negative KCNQ4 construct may have occurred through unintended interactions with native gK,L channels. The young age of the neonatal cultured and transfected utricles raises the possibility of a developmental effect – that functional Kv7 channels are needed for the developmental transition to a Kv1.8 conductance. Consistent with this idea is the observation that Kv7 current is present in neonatal hair cells, where it is a relatively large proportion of Kv current in type I HCs before they acquire gK,L (Hurley et al., J. Neurosci., 2006). Alternatively, the overexpression of nonfunctional Kv7.4 channels in virally-transfected hair cells may have inhibited or delayed gK,L acquisition through a more general effect on membrane proteins.

1. Strong immunosignal appears in the cuticle plates of hair cells in addition to signal in basal regions of hair cells and supporting cells. Please provide a possible explanation for this.

Author response: There is significant non-specific staining of apical cell surfaces and supporting cell membranes in addition to specific staining of hair cell basolateral membranes. We infer non-specific staining when immunolabeling is present in the knockout tissue, as it is for the apical surfaces and supporting cell membranes—compare Fig. 5B.3 (control tissue) with Fig. 5B.4 (Kv1.8 null mutant). Non-specific immunostaining can occur with polyclonal antibodies (specific to several epitopes) if the antibodies are not affinity-purified, but we used an affinity-purified antibody. The apical surfaces are reputed to be “sticky” (susceptible to non-specific staining) but the non-specific labeling in the basal parts of supporting cells is more puzzling. One possibility is that the Kv1.8 antibody weakly recognized closely related Kv1.1 channels, which are more strongly expressed in supporting cells than hair cells (Scheffer et al., J. Neurosci., 2015).

1. A previous paper reported that a vestibular evoked potential was abnormal in Kv1.8-/- mice (Lee et al. 2013) as briefly mentioned (lines 94-95). It would be very interesting to know if any vestibular-associated behaviors and/or hearing loss were observed in the mice populations. If responses are compromised at the sensory hair cell level across different zones, degradation of balance function would be anticipated and should be elucidated.

Author response: We agree; some of these questions are the subject of another paper in preparation.

Author Response

Reviewer 1:

Comment 1.1: The distinction of PIGS from nearby OPA, which has also been implied in navigation and ego-motion, is not as clear as it could be.

Response1.1: The main functional distinction between TOS/OPA and PIGS is that TOS/OPA responds preferentially to moving vs. stationary stimuli (even concentric rings), likely due to its overlap with the retinotopic motion-selective visual area V3A, for which this is a defining functional property (e.g. Tootell et al., 1997, J Neurosci). In comparison, PIGS does not show such a motion-selectivity. Instead, PIGS responds preferentially to more complex forms of motion within scenes. In this revision, we tried to better highlight this point in the Discussion (see also the response to the first comment from Reviewer #2).

Reviewer 2:

Comment 2.1: First, the scene-selective region identified appears to overlap with regions that have previously been identified in terms of their retinotopic properties. In particular, it is unclear whether this region overlaps with V7/IPS0 and/or IPS1. This is particularly important since prior work has shown that OPA often overlaps with v7/IPS0 (Silson et al, 2016, Journal of Vision). The findings would be much stronger if the authors could show how the location of PIGS relates to retinotopic areas (other than V6, which they do currently consider). I wonder if the authors have retinotopic mapping data for any of the participants included in this study. If not, the authors could always show atlas-based definitions of these areas (e.g. Wang et al, 2015, Cerebral Cortex).

Response 2.1: We thank the reviewers for reminding us to more clearly delineate this issue of possible overlap, including the information provided by Silson et al, 2016. The issue of possible overlap between area TOS/OPA and the retinotopic visual areas, both in humans and non-human primates, was also clarified by our team in 2011 (Nasr et al., 2011). As you can see in the enclosed figure, and consistent with those previous studies, TOS/OPA overlaps with visual areas V3A/B and V7. Whereas PIGS is located more dorsally close to IPS2-4. As shown here, there is no overlap between PIGS and TOS/OPA and there is no overlap between PIGS and areas V3A/B and V7. To more directly address the reviewer’s concern, in the next revision, we will show the relative position of PIGS and the retinotopic areas (at least) in one individual subject.

Author response image 1.

The relative location of PIGS, TOS/OPA and the retinotopic visual areas. The left panel showed the result of high-resolution (7T; voxel size = 1 mm; no spatial smoothing) polar angle mapping in one individual. The right panel shows the location of scene-selective areas PIGS and TOS/OPA in the same subject (7T; voxel size = 1 mm; no spatial smoothing). While area TOS/OPA shows some overlap with the retinotopic visual areas V3A/B and V7, PIGS shows partial overlap with area IPS2-4. In both panels, the activity maps are overlaid on the subjects’ own reconstructed brain surface.

Comment 2.2: Second, recent studies have reported a region anterior to OPA that seems to be involved in scene memory (Steel et al, 2021, Nature Communications; Steel et al, 2023, The Journal of Neuroscience; Steel et al, 2023, biorXiv). Is this region distinct from PIGS? Based on the figures in those papers, the scene memory-related region is inferior to V7/IPS0, so characterizing the location of PIGS to V7/IPS0 as suggested above would be very helpful here as well. If PIGS overlaps with either of V7/IPS0 or the scene memory-related area described by Steel and colleagues, then arguably it is not a newly defined region (although the characterization provided here still provides new information).

Response 2.2: The lateral-place memory area (LPMA) is located on the lateral brain surface, anterior relative to the IPS (see Figure 1 from Steel et al., 2021 and Figure 3 from Steel et al., 2023). In contrast, PIGS is located on the posterior brain surface, also posterior relative to the IPS. In other words, they are located on two different sides of a major brain sulcus. In this revision we have clarified this point, including the citations by Steel and colleagues.

Comments 2.3: Another reason that it would be helpful to relate PIGS to this scene memory area is that this scene memory area has been shown to have activity related to the amount of visuospatial context (Steel et al, 2023, The Journal of Neuroscience). The conditions used to show the sensitivity of PIGS to ego-motion also differ in the visuospatial context that can be accessed from the stimuli. Even if PIGS appears distinct from the scene memory area, the degree of visuospatial context is an alternative account of what might be represented in PIGS.

Response 2.3: The reviewer raises an interesting point. One minor confusion is that we may be inadvertently referring to two slightly different types of “visuospatial context”. Specifically, the stimuli used in the ego-motion experiment here (i.e. coherently vs. incoherently changing scenes) represent the same scenes, and the only difference between the two conditions is the sequence of images across the experimental blocks. In that sense, the two experimental conditions may be considered to have the same visuospatial context. However, it could be also argued that the coherently changing scenes provide more information about the environmental layout. In that case, considering the previous reports that PPA/TPA and RSC/MPA may also be involved in layout encoding (Epstein and Kanwisher 1998; Wolbers et al. 2011), we expected to see more activity within those regions in response to coherently compared incoherently changing scenes. These issues are now more explicitly discussed in the revised article.

Reviewer 3:

Comment 3.1: There are few weaknesses in this work. If pressed, I might say that the stimuli depicting ego-motion do not, strictly speaking, depict motion, but only apparent motion between 2s apart photographs. However, this choice was made to equate frame rates and motion contrast between the 'ego-motion' and a control condition, which is a useful and valid approach to the problem. Some choices for visualization of the results might be made differently; for example, outlines of the regions might be shown in more plots for easier comparison of activation locations, but this is a minor issue.

Response 3.1: We thank the reviewer for these constructive suggestions, and we agree with their comment that the ego-motion stimuli are not smooth, even though they were refreshed every 100 ms. However, the stimuli were nevertheless coherent enough to activate areas V6 and MT, two major areas known to respond preferentially to coherent compared to incoherent motion.

Epstein, R., and N. Kanwisher. 1998. 'A cortical representation of the local visual environment', Nature, 392: 598-601.

Wolbers, T., R. L. Klatzky, J. M. Loomis, M. G. Wutte, and N. A. Giudice. 2011. 'Modality-independent coding of spatial layout in the human brain', Curr Biol, 21: 984-9.

Author Response

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

eLife assessment

This study presents valuable findings about synaptic connectivity among subsets of unipolar brush cells (UBCs), a specialized interneuron primarily located in the vestibular lobules of the cerebellar cortex. The evidence supporting the claims are interesting although incomplete in some areas. The work will be of interest to cerebellar neuroscientists as well as those focussed on synaptic properties and mechanisms. Although several compelling pieces of data were presented, substantial work remains to be conducted in order for the hypothesis and predictions of the manuscript to confirm how these factors play out in the actual brain circuit and how it would impact the processing of feedback or feedforward activity that would be required to promote behavior.

Public Reviews:

Reviewer #1 (Public Review):

The manuscript by Hariani et al. presents experiments designed to improve our understanding of the connectivity and computational role of Unipolar Brush Cells (UBCs) within the cerebellar cortex, primarily lobes IX and X. The authors develop and cross several genetic lines of mice that express distinct fluorophores in subsets of UBCs, combined with immunocytochemistry that also distinguishes subtypes of UBCs, and they use confocal microscopy and electrophysiology to characterize the electrical and synaptic properties of subsets of so-labelled cells, and their synaptic connectivity within the cerebellar cortex. The authors then generate a computer model to test possible computational functions of such interconnected UBCs.

Using these approaches, the authors report that:

1. GRP-driven TDtomato is expressed exclusively in a subset (20%) of ON-UBCs, defined electrophysiologically (excited by mossy fiber afferent stimulation via activation of UBC AMPA and mGluR1 receptors) and immunocytochemically by their expression of mGluR1.

2. UBCs ID'd/tagged by mCitrine expression in Brainbow mouse line P079 is expressed in a similar minority subset of OFF-UBCs defined electrophysiologically (inhibited by mossy fiber afferent stimulation via activation of UBC mGluR2 receptors) and immunocytochemically by their expression of Calretinin. However, such mCitrine expression was also detected in some mGluR1 positive UBCs, which may not have shown up electrophysiologically because of the weaker fluorophore expression without antibody amplification.

3. Confocal analysis of crossed lines of mice (GRP X P079) stained with antibodies to mGluR1 and calretinin documented the existence of all possible permutations of interconnectivity between cells (ON-ON, ON-OFF, OFF-OFF, OFF-ON), but their overall abundance was low, and neither their absolute or relative abundance was quantified.

4. A computational model (NEURON ) indicated that the presence of an intermediary UBC (in a polysynaptic circuit from MF to UBC to UBC) could prolong bursts (MF-ON-ON), prolong pauses (MF-ON-OFF), cause a delayed burst (MF-OFF-OFF), cause a delayed pause (MF-OFF-ON) relative to solely MF to UBC synapses which would simply exhibit long bursts (MF-ON) or long pauses (MF-OFF).

The authors thus conclude that the pattern of interconnected UBCs provides an extended and more nuanced pattern of firing within the cerebellar cortex that could mediate longer lasting sensorimotor responses.

The cerebellum's long known role in motor skills and reflexes, and associated disorders, combined with our nascent understanding of its role in cognitive, emotional, and appetitive processing, makes understanding its circuitry and processing functions of broad interest to the neuroscience and biomedical community. The focus on UBCs, which are largely restricted to vestibular lobes of the cerebellum reduces the breadth of likely interest somewhat. The overall design of specific experiments is rigorous and the use of fluorophore expressing mouse lines is creative. The data that is presented and the writing are clear. However, despite some additional analysis in response to the initial review, the overall experimental design still has issues that reduce overall interpretation (please see specific issues for details), which combined with a lack of thorough analysis of the experimental outcomes undermines the value of the NEURON model results and the advance in our understanding of cerebellar processing in situ (again, please see specific issues for details).

Specific issues:

1. All data gathered with inhibition blocked. All of the UBC response data (Fig. 1) was gathered in the presence of GABAAR and Glycine R blockers. While such an approach is appropriate generally for isolating glutamatergic synaptic currents, and specifically for examining and characterizing monosynaptic responses to single stimuli, it becomes problematic in the context of assaying synaptic and action potential response durations for long lasting responses, and in particular for trains of stimuli, when feed-forward and feed-back inhibition modulates responses to afferent stimulation. I.e. even for single MF stimuli, given the >500ms duration of UBC synaptic currents, there is plenty of time for feedback inhibition from Golgi cells (or feedforward, from MF to Golgi cell excitation) to interrupt AP firing driven by the direct glutamatergic synaptic excitation. This issue is compounded further for all of the experiments examining trains of MF stimuli. Beyond the impact of feedback inhibition on the AP firing of any given UBC, it would also obviously reduce/alter/interrupt that UBC's synaptic drive of downstream UBCs. This issue fundamentally undermines our ability to interpret the simulation data of Vm and AP firing of both the modeled intermediate and downstream UBC, in terms of applying it to possible cerebellar cortical processing in situ.

The goal of Figure 1 was to determine the cell types of labeled UBCs in transgenic mouse lines, which is determined entirely by their synaptic responses to glutamate (Borges-Merjane and Trussell, 2015). Thus, blocking inhibition was essential to produce clear results in the characterization of GRP and P079 UBCs. While GABAergic/glycinergic feedforward and feedback inhibition is certainly important in the intact circuit, it was not our intention, nor was it possible, to study its contribution in the present study. Leaving inhibition unblocked does not lead to a physiologically realistic stimulation pattern in acute brain slices, because electrical stimulation produces synchronous excitation and inhibition by directly exciting Golgi cells, rather than their synaptic inputs. The main inhibition that UBCs receive that are crucial to determining burst or pause durations is not via GABA/glycine, but instead through mGluR2, which lasts for 100-1000s of milliseconds. The main excitation that drives UBC firing is mGluR1 and AMPA, which both last 100-1000s of milliseconds. Thus, these large conductances are unlikely to be significantly shaped by 1-10 ms IPSCs from feedforward and feedback GABA/glycine inhibition. Recent studies that examined the duration of bursting or pausing in UBCs had inhibition blocked in their experiments, presumably for the reasons outlined above (Guo et al., 2021; Huson et al., 2023).

Below is an example showing the synaptic currents and firing patterns in an ON UBC before and after blocking inhibition. The GABA/glycinergic inhibition is fast, occurs soon after the stimuli and has little to no effect on the slow inward current that develops after the end of stimulation, which is what drives firing for 100s of milliseconds.

Author response image 1.

Example showing small effect of GABAergic and glycinergic inhibition on excitatory currents and burst duration. A) Excitatory postsynaptic currents in response to train of 10 presynaptic stimuli at 50 Hz before (black) and after (Grey) blocking GABA and glycine receptors. The slow inward current that occurs at the end of stimulation is little affected. B) Expanded view of the synaptic currents evoked during the train of stimuli. GABA/glycine receptors mediate the fast outward currents that occur immediately after the first couple stimuli. C) Three examples of the bursts caused by the 50 Hz stimulation in the same cell without blocking GABA and glycine receptors. D) Three examples in the same cell after blocking GABA and glycine receptors.

The authors' response to the initial concern is (to paraphrase), "its not possible to do and its not important", neither of which are soundly justified.

As stated in the original review, it is fully understandable and appropriate to use GABAAR/GlycineR antagonists to isolate glutamatergic currents, to characterize their conductance kinetics. That was not the issue raised. The issue raised was that then using only such information to generate a model of in situ behavior becomes problematic, given that feedback and lateral inhibition will sculpt action potential output, which of course will then fundamentally shape their synaptic drive of secondary UBCs, which will be further sculpted by their own inhibitory inputs. This issue undermines interpretation of the NEURON model.

The argument that taking inhibition into account is not possible because of assumed or possible direct electrical excitation of Golgi cells is confusing for two interacting reasons. First, one can certainly stimulate the mossy fiber bundle to get afferent excitation of UBCs (and polysynaptic feedback/lateral inhibitory inputs) without directly stimulating the Golgi cells that innervate any recorded UBC. Yes, one might be stimulating some Golgi cells near the stimulating electrode, but one can position the stimulating electrode far enough down the white matter track (away from the recorded UBC), such that mossy fiber inputs to the recorded UBC can be stimulated without affecting Golgi cells near or synaptically connected to the recorded UBC. Moreover, if the argument were true, then presumably the stimulation protocol would be just as likely to directly stimulate neighboring UBCs, which then drove the recorded UBC's responses. Thus, it is both doable and should be ensured that stimulation of the white matter is distant enough to not be directly activating relevant, connected neurons within the granule cell layer.

Finally, the authors present three examples of UBC recordings with and without inhibitory inputs blocked, and state "Thus, these large conductances are unlikely to be significantly shaped by 1-10 ms IPSCs from feedforward and feedback GABA/glycine inhibition" and "GABA/glycinergic inhibition...has little to no effect on the slow inward current that develops after the end of stimulation". This response reflects on original concerns about lack of quantification or consideration of important parameters. In particular, while the traces with and without inhibition are qualitatively similar, quantitative considerations indicate otherwise. First, unquantified examples are not adequate to drive conclusions. Regardless, the main issue (how inhibition affects actual responses in situ) is actually highlighted by the authors current clamp recordings of UBC responses, before and after blocking inhibition. The output response is dramatically different, both at early and late time points, when inhibition is blocked. Again, a lack of quantification (of adequate n's) makes it hard to know exactly how important, but quick "eye ball" estimates of impact include: 1) a switch from only low frequency APs initially (without inhibition blocked) to immediate burst of high frequency APs (high enough to not discern individual APs with given figure resolution) when inhibition is blocked, 2) Slow rising to a peak EPSP, followed by symmetrical return to baseline (without inhibition blocked) versus immediate rise to peak, followed by prolonged decay to baseline (with inhibition blocked), 3) substantially shorter duration (~34% shorter) secondary high frequency burst (individual APs not discernible) of APs (with inhibition blocked versus without inhibition blocked), and 4) substantial reduction in number of long delayed APs (with inhibition blocked versus without inhibition blocked). Thus, clearly, feedback/lateral inhibition is actually sculpting AP output at all phases of the UBC response to trains of afferent stimulations. Importantly, the single voltage clamp trace showing little impact of transient IPSCs on the slow EPSC do not take into account likely IPSC influences on voltage-activated conductances that would not occur in voltage-clamp recordings but would be free to manifest in current clamp, and thereby influence AP output, as observed.

So again, our ability to understand how interconnected UBCs behave in the intact system is undermined by the lack of consideration and quantification of the impact of inhibition, and it not being incorporated into the model. At the very least a strong proviso about lack of inclusion of such information, given the authors' data showing its importance in the few examples shown, should be added to the discussion.

Thank you for this substantive explanation. Your points are well described and we agree that the single experiment shown is not strong evidence for a lack of importance of Golgi cell inhibition, especially on the temporal dynamics of spiking. Previous work has clearly shown that Golgi cells have several important roles in shaping the activity of the granular layer, including affecting the temporal dynamics of granule cell spikes. However, the work presented here focuses on the feedforward circuitry of UBCs and the large inward and large outward glutamatergic currents that drive spiking or pausing for 100s of milliseconds. Our model does not focus on the aspects that are most sensitive to Golgi cell inhibition, including timing of the first spikes in the UBC’s response. Nor does our model focus on short term plasticity, which we thought was reasonable because the slow currents in UBCs are quite insensitive to the temporal characteristics of glutamate release (See the example in the previous rebuttal). Our model does not include long term plasticity, which is also affected by Golgi cells. For these reasons we agree that the model presented does not explain how feedforward UBC circuits might “play out in the actual brain circuit and how it would impact the processing of feedback or feedforward activity that would be required to promote behavior.” We have included a new paragraph in the discussion clarifying the limitations of this study and the model, reproduced below.

"Limitations of the model

Here we addressed how feedforward glutamatergic excitation and inhibition is transformed from one UBC to the next depending on their subtype. The model focuses on AMPA receptor mediated excitation and mGluR2 mediated inhibition. One limitation of the model is that it does not consider feedforward and lateral inhibition from Golgi cells, which shape the spiking of UBCs in response to afferent stimulation. Golgi cells receive mossy fiber input and inhibit UBCs through their corelease of GABA and glycine (Dugue et al., 2005; Rousseau et al., 2012). Golgi cells control the temporal dynamics of the firing of granule cells as well as their gain (Rossi et al., 2003; Kanichay and Silver, 2008) and are critical to larger scale dynamics of the cerebellar cortical network (D‘Angelo, 2008). Purkinje cells provide additional inhibition to ON UBCs that could influence how UBC circuits transform signals (Guo et al., 2016). A more complex model that implements Golgi cells and other critical circuit elements will be needed to investigate the role of feedforward UBC circuits in cerebellar network dynamics and motor behaviors in vivo."

1. No consideration for involvement of polysynaptic UBCs driving UBC responses to MF stimulation in electrophysiology experiments. Given the established existence (in this manuscript and Dino et al. 2000 Neurosci, Dino et al. 2000 ProgBrainRes, Nunzi and Mugnaini 2000 JCompNeurol, Nunzi et al. 2001 JCompNeurol) of polysynaptic connections from MFs to UBCs to UBCs, the MF evoked UBC responses established in this manuscript, especially responses to trains of stimuli could be mediated by direct MF inputs, or to polysynaptic UBC inputs, or possibly both (to my awareness not established either way). Thus the response durations could already include extension of duration by polysynaptic inputs, and so would overestimate the duration of monosynaptic inputs, and thus polysynaptic amplification/modulation, observed in the NEURON model.

We are confident that the synaptic responses shown are monosynaptic for several reasons. UBCs receive a single mossy fiber input on their dendritic brush, and thus if our stimulation produces a reliable, short-latency response consistent with a monosynaptic input, then there is not likely to be a disynaptic input, because the main input is accounted for by the monosynaptic response. In all cells included in our data set, the fast AMPA receptor-mediated currents always occurred with short latency (1.24 ± 0.29 ms; mean ± SD; n = 13), high reliability (no failures to produce an EPSC in any of the 13 GRP UBCs in this data set), and low jitter (SD of latency; 0.074 ± 0.046 ms; mean ± SD; n = 13). These measurements have been added to the results section.

In some rare cases, we did observe disynaptic currents, which were easily distinguishable because a single electrical stimulation produced a burst of EPSCs at variable latencies. Please see example below. These cases of disynaptic input, which have been reported by others (Diño et al., 2000; Nunzi and Mugnaini, 2000; van Dorp and De Zeeuw, 2015) support the conclusion that UBCs receive input from other UBCs.

Author response image 2.

Example of GRP UBC with disynaptic input. Three examples of the effect of a single presynaptic stimulus (triangle) in a GRP UBC with presumed disynaptic input. Note the variable latency of the first evoked EPSC, bursts of EPSCs, and spontaneous EPSCs.

Author response: "UBCs receive a single mossy fiber input on their dendritic brush, and thus if our stimulation produces a reliable, short-latency response consistent with a monosynaptic input, then there is not likely to be a disynaptic input."

This statement is not congruent with the literature, with early work by Mugnaini and colleagues (Mugnaini et al. 1994 Synapse; Mugnaini and Flores 1994 J. Comp. Neurol.) indicating that UBCs are innervated by 1-2 mossy fibers, which are as likely other UBC terminals as MFs. This leaves open the possibility that so called monosynaptic responses do, as originally suggested, already include polysynaptic feedforward amplification of duration. While the authors also indicate that isolated disynaptic currents can be observed when they occur in isolation, a careful examination and objective documentation of "monosynaptic" responses would address this issue. Presumably, if potential disynaptic UBC inputs occur during a monosynaptic MF response, it would be detected as an abrupt biphasic inward/outward current, due to additional AMPA receptor activation but further desensitization of those already active (as observed by Kinney et al. 1997 J. Neurophysiol: "The delivery of a second MF stimulus at the peak of the slow EPSC evoked a fast EPSC of reduced amplitude followed by an undershoot of the subsequent slow current"). If such polysynaptic inputs are truly absent and are "rare" in isolation, some estimation of how common or not such synaptic amplification is, would improve our understanding of the overall significance of these inputs.

We are confident that these currents are monosynaptic, because, as suggested, we carefully analyzed the latency, jitter and reliability, which was added to the previous revision. The latency and jitter are strong (quantitative) evidence that the first EPSC evoked was monosynaptic. While some UBCs have been reported to have multiple brushes, or brushes that branch and may contact multiple mossy fibers, or receive synaptic input onto their somas, these cases are rare in our experience in this age of mouse and there is no evidence for them in this dataset. For every trace we made a careful examination and documented that no delayed EPSCs were present. The presence of delayed EPSCs (or ‘abrupt biphasic inward/outward currents’ as described in Kinney et al 1997) would indeed suggest the presence of disynaptic activity or multiple inputs to the UBC, but these would be easily identified, even during a stimulation train. For these reasons we feel that we have established that polysynaptic feedforward amplification of duration is not present

We agree that the monosynaptic responses could be due to the stimulation of UBC axons. However, the absence of delayed EPSCs again suggests that if stimulation of a presynaptic UBC axon was producing the currents in the recorded UBC, then the axon was severed from the soma and AIS, because this region is necessary for the cell to produce more than a single spike per stimulation. We added a sentence describing the potential for the monosynaptic EPSCs to be due to the stimulation of presynaptic UBC axons.

Your point is well taken that a discussion of how common or rare these UBC to UBC connections is necessary to more clearly explain how we interpret their significance and we have expanded the paragraph in the discussion that does so. Thank you for this suggestion.

1. Lack of quantification of subtypes of UBC interconnectivity. Given that it is already established that UBCs synapse onto other UBCs (see refs above), the main potential advance of this manuscript in terms of connectivity is the establishment and quantification of ON-ON, ON-OFF, OFF-ON, and OFF-OFF subtypes of UBC interconnections. But, the authors only establish that each type exists, showing specific examples, but no quantification of the absolute or relative density was provided, and the authors' unquantified wording explicitly or implicitly states that they are not common. This lack of quantification and likely small number makes it difficult to know how important or what impact such synapses have on cerebellar processing, in the model and in situ.

As noted by the reviewer, the connections between UBCs were rare to observe. We decided against attempting to quantify the absolute or relative density of connections for several reasons. A major reason for rare observations of anatomical connections between UBCs is likely due to the sparse labeling. First, the GRP mouse line only labels 20% of ON UBCs and we are unable to test whether postsynaptic connectivity of GRP ON UBCs is the same as that of the rest of the population of ON UBCs that are not labeled in the GRP mouse line. Second, the Brainbow reporter mouse only labels a small population of Cre expressing cells for unknown reasons. Third, the Brainbow reporter expression was so low that antibody amplification was necessary, which then limited the labeled cells to those close to the surface of the brain slices, because of known antibody penetration difficulties. Therefore, we refrained from estimating the density of these connections, because each of these variables reduced the labeling to unknown degrees and we reasoned that extrapolating our rare observations to the total population would be inaccurate.

A paper that investigated UBC connectivity using organotypic slice cultures from P8 mice suggests that 2/3 of the UBC population receives UBC input, based on the observation that 2/3 of the mossy fibers did not degenerate as would be expected after 2 days in vitro if they were severed from a distant cell body (Nunzi and Mugnaini, 2000). It remains to be seen if this high proportion is due to the young age of these mice or is also the case in adult mice. Even if these connections are indeed rare, they are expected to have profound effects on the circuit, as each UBC has multiple mossy fiber terminals (Berthie and Axelrad, 1994), and mossy fiber terminals are estimated to contact 40 granule cells each (Jakab and Hamori, 1988). We have added a comment regarding this point to the discussion.

To address this issue, the authors added the following text to the discussion section: "We did not estimate the density of these UBC to UBC connections, because the sparseness of labeling using these approaches made an accurate calculation impossible. Previous work using organotypic slice cultures from P8 mice estimated that 2/3 of the UBC population receives input from other UBCs (Nunzi & Mugnaini, 2000), although it is unclear whether this is the case in older mice."

While accurate, the addition doesn't really address the situation, which is that apparently the reported connections are rare. Adding the information about 2/3 of UBCs having UBC inputs in culture, implies the opposite might be true (i.e. that they might be quite common), which is in contrast to the authors' data, so should be reworded for clarity, which should also incorporate the considerations covered in point #2 above. I.e. if the authors do establish that none of their recordings have polysynaptic inputs, and if they determine that the number of cells that showed isolated di-synaptic inputs is indeed rare, then it suggests that these specific polysynaptic connections are in fact rare.

Thank you for pointing this out. We agree that adding this information is somewhat contradictory to our results and we have added more to this section in the discussion, provided below.

Anatomically identifiable connections between UBCs were not present in all brain slices and finding them required a careful search. UBC labeling was sparse due to the highly specific genetic labeling techniques and further sparsification by the Brainbow reporter, which made it impossible to estimate the density of these UBC to UBC connections. Electrophysiological evidences suggest that UBC to UBC connections are not common, because spontaneous EPSCs that would indicate a spontaneously firing presynaptic UBC are only rarely observed in UBCs recorded in acute brain slices. In an analysis of feedforward excitation of granule layer neurons, only 4 out of 140 UBCs had this indirect evidence of a firing presynaptic UBC (van Dorp and De Zeeuw, 2015), which suggests that UBC to UBC connections may be rare. On the other hand, previous work using organotypic slice cultures from P8 mice estimated that 2/3 of the UBC population receives input from other UBCs (Nunzi & Mugnaini, 2000). This suggests a much higher density of UBC to UBC connections, but could be due to the young age of the brains used, which is before UBCs have matured (Morin et al., 2001), and also due to increased collateral sprouting that can occur in culture (Jaeger et al., 1988). Another study imaged 2-4 week old rat cerebellar slices at an electron microscopic level and found that 4 out of 14 UBC axon terminals contacted UBC brushes (Diño et al., 2000). Future work is necessary to accurately estimate the density and impact of these feedforward UBC circuits.

1. Lack of critical parameters in NEURON model.

A) The model uses # of molecules of glutamate released as the presumed quantal content, and this factor is constant.

However, no consideration of changes in # of vesicles released from single versus trains of APs from MFs or UBCs is included. At most simple synapses, two sequential APs alters release probability, either up or down, and release probability changes dynamically with trains of APs. It is therefore reasonable to imagine UBC axon release probability is at least as complicated, and given the large surface area of contact between two UBCs, the number of vesicles released for any given AP is also likely more complex.

B) the model does not include desensitization of AMPA receptors, which in the case of UBCs can paradoxically reduce response magnitude as vesicle release and consequent glutamate concentration in the cleft increases (Linney et al. 1997 JNeurophysiol, Lu et al. 2017 Neuron, Balmer et al. 2021 eLIFE), as would occur with trains of stimuli at MF to ON-UBCs.

A) The model produces synaptic AMPA and mGluR2 currents that reproduce those we recorded in vitro. We did not find it necessary to implement changes in glutamate release during a train as the model was fit to UBC data with the assumption that the glutamate transient did not change during the train. If there is a change in neurotransmitter release during a train, it is therefore built into the model, which has the advantage of reducing its complexity. UBCs are a special case where the postsynaptic currents are mediated mostly by the total amount of transmitter released. Most of the evoked current occurs tens to hundreds of milliseconds after neurotransmitter release and is therefore much more sensitive to total release and less sensitive to how it is released during the train. The figure below shows the effect of reducing the amount of glutamate released by 10% on each stimulus in the model. Despite a significant change in the pattern of neurotransmitter release, as well as a reduction in the total amount of glutamate, the slow EPSC still decays over the course of hundreds of milliseconds.

B) The detailed kinetic AMPA receptor model used here accurately reproduces desensitization, which in fact mediates that the slow ON UBC current. This AMPA receptor is a 13-state model, including 4 open states with 1-4 glutamates bound, 4 closed states with 1-4 glutamates bound, 4 desensitized states with 1-4 glutamates bound, and 5 closed states with 0-4 glutamates bound. The forward and reverse rates between different states in the model were fit to AMPA receptor currents recorded from dissociated UBCs and they accurately reproduced the ON UBC currents evoked by synaptic stimulation in our previous work (Balmer et al., 2021).

Author response image 3.

Effect of short-term depression of neurotransmitter release. A) The top trace shows the glutamate transient that drives the AMPA receptor model used in our study. No change in release is implemented, although the slow tail of the transient summates during the train. The bottom trace shows the modeled AMPA receptor mediated current. B) In this model the amount of glutamate released on each stimulus is reduced by 10%. The duration of the slow AMPA current is similar, despite a profound change in the pattern of neurotransmitter exposure.

While the authors have not added the suggested additional parameters, their clarifications regarding the implications of existing parameters, and demonstration of reasonable fits to experimental data, and lack of substantial effect of simulating reduced vesicle release probability,

1. Lack of quantification of various electrophysiological responses. UBCs are defined (ON or OFF) based on inward or outward synaptic response, but no information is provided about the range of the key parameter of duration across cells, which seems most critical to the current considerations. There is a similar lack of quantification across cells of AP duration in response to stimulation or current injections, or during baseline. The latter lack is particularly problematic because in agreement with previous publications, the raw data in Fig. 1 shows ON UBCs as quiescent until MF stimulation and OFF UBCs firing spontaneously until MF stimulation, but, for example, at least one ON UBC in the NEURON model is firing spontaneously until synaptically activated by an OFF UBC (Fig. 11A), and an OFF UBC is silent until stimulated by a presynaptic OFF UBC (Fig. 11C). This may be expected/explainable theoretically, but then such cells should be observed in the raw data.

To address this reasonable concern of a general lack of quantification of electrophysiological responses we have added data characterizing the slow inward and outward currents evoked by synaptic stimulation in GRP and P079 UBCs in the results section and in new panels in Figure 1. We report the action potential pause lengths in P079 UBCs and burst lengths in ON UBCs in the results section. However, we favor the duration of the currents to the length of burst and pause, because the currents do not depend on a stable resting membrane potential, which is itself difficult to determine in intracellular recordings of these small cells. In a series of recent publications that focused on UBC firing, the authors argue that cell-attached recordings are necessary to determine accurately the burst and pause lengths, as well as spontaneous firing rates (Guo et al., 2021; Huson et al., 2023). (The trade-off of these extracellular recordings is that the monosynaptic nature of the input is nearly impossible to confirm.) Spontaneous firing rates were variable within both GRP and P079 UBCs from silent to firing regularly or in bursts, as previously reported (Kim et al., 2012; van Dorp and De Zeeuw, 2015). For clarity, we chose to model the GRP UBCs as silent unless receiving synaptic input and P079 UBCs as active unless receiving synaptic input. As the reviewer suggests, we have observed UBCs firing in the patterns similar to those shown in the model UBCs having input from spontaneous presynaptic UBCs. Below are some examples of spontaneous EPSCs and IPSCs in UBCs that suggest the presence of a presynaptic UBC.

Author response image 4.

Examples of UBCs that receive spontaneous input. A) Three ON UBCs that had spontaneous EPSCs, suggesting the presence of an active presynaptic UBC. B) Two OFF UBCs that had spontaneous outward currents.

The authors have added additional analysis and discussion, which adequately addresses this concern.

Reviewer #2 (Public Review):

In this paper, the authors presented a compelling rationale for investigating the role of UBCs in prolonging and diversifying signals. Based on the two types of UBCs known as ON and OFF UBC subtypes, they have highlighted the existing gaps in understanding UBCs connectivity and the need to investigate whether UBCs target UBCs of the same subtype, different subtypes, or both. The importance of this knowledge is for understanding how sensory signals are extended and diversified in the granule cell layer.

The authors designed very interesting approaches to study UBCs connectivity by utilizing transgenic mice expressing GFP and RFP in UBCs, Brainbow approach, immunohistochemical and electrophysiological analysis, and computational models to understand how the feed-forward circuits of interconnected UBCs transform their inputs.

This study provided evidence for the existence of distinct ON and OFF UBC subtypes based on their electrophysiological properties, anatomical characteristics, and expression patterns of mGluR1 and calretinin in the cerebellum. The findings support the classification of GRP UBCs as ON UBCs and P079 UBCs as OFF UBCs and suggest the presence of synaptic connections between the ON and OFF UBC subtypes. In addition, they found that GRP and P079 UBCs form parallel and convergent pathways and have different membrane capacitance and excitability. Furthermore, they showed that UBCs of the same subtype provide input to one another and modify the input to granule cells, which could provide a circuit mechanism to diversify and extend the pattern of spiking produced by mossy fiber input. Accordingly, they suggested that these transformations could provide a circuit mechanism for maintaining a sensory representation of movement for seconds.

Overall, the article is well written in a sound detailed format, very interesting with excellent discovery and suggested model.

I believe the authors have provided appropriate responses and have consequently revised the manuscript in a convincing manner. Although I am not an expert in physiology, I find the explanations and clarifications to be acceptable.

Author Response

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

Reviewer 1 (Public Review):

1. The name of the new method "inter-haplotype distance" is more confusing than helpful, as the haplotype information is not critical for implementing this method. First, the mutation spectrum is aggregated genome-wide regardless of the haplotypes where the mutations are found. Second, the only critical haplotype information is that at the focal site (i.e., the locus that is tested for association): individuals are aggregated together when they belong to the same "haplotype group" at the focal site. However, for the classification step, haplotype information is not really necessary: individuals can be grouped based on their genotypes at the given locus (e.g., AA vs AB). As the authors mentioned, this method can be potentially applied to other mutation datasets, where haplotype information may well be unavailable. I hope the authors can reconsider the name and remove the term "haplotype" (perhaps something like "inter-genotype distance"?) to avoid giving the wrong impression that haplotype information is critical for applying this method.

We appreciate the reviewer's concern about the name of our method. The reviewer is correct that haplotype information is not critical for our method to work, and as a result we've decided to simply rename the approach to "aggregate mutation spectrum distance" (abbreviated AMSD). For simplicity, we refer to the method as IHD throughout our responses to reviewers, but the revised manuscript now refers to AMSD.

1. The biggest advantage of the IHD method over QTL mapping is alleviation of the multiple testing burden, as one comparison tests for any changes in the mutation spectrum, including simultaneous, small changes in the relative abundance of multiple mutation types. Based on this, the authors claim that IHD is more powerful to detect a mutator allele that affects multiple mutation types. Although logically plausible, it is unclear under what quantitative conditions IHD can actually have greater power over QTL. It will be helpful to support this claim by providing some simulation results.

This comment prompted us to do a more detailed comparison of IHD vs. QTL power under conditions that are more similar to those observed in the BXD cohort. While preparing the original manuscript, we assumed that IHD might have greater power than QTL mapping in a population like the BXDs because some recombinant inbred lines have accumulated many more germline mutations than others (see Figure 1 in Sasani et al. 2022, Nature). In a quantitative trait locus scan (say, for the fraction of C>A mutations in each line) each BXD's mutation data would be weighted equally, even if a variable number of mutations was used to generate the phenotype point estimate in each line.

To address this, we performed a new series of simulations in which the average number of mutations per haplotype was allowed to vary. At the low end, some BXDs accumulated as few as 100 total germline mutations, while others have accumulated as many as 2,000. Thus, instead of simulating a mean number of mutations on each simulated haplotype, we allowed the mean number of mutations per haplotype to vary from N to 20N. By simulating a variable count of mutations on each haplotype, we could more easily test the benefits of comparing aggregate, rather than individual, mutation spectra between BXDs.

In these updated simulations, we find that IHD routinely outperforms QTL mapping under a range of parameter choices (see Author Response image 1). Since IHD aggregates the mutation spectra of all haplotypes with either B or D alleles at each locus in the genome, the method is much less sensitive to individual haplotypes with low mutation counts. We include a mention of these updated simulations on lines 135-138 and describe the updated simulations in greater detail in the Materials and Methods (lines 705-715).

Author response image 1.

Power of IHD and QTL mapping on simulated haplotypes with variable counts of mutations. We simulated germline mutations on the specified number of haplotypes (as described in the manuscript) but allowed the total number of mutations per haplotype to vary by a factor of 20.

1. The flip side of this advantage of IHD is that, when a significant association is detected, it is not immediately clear which mutation type is driving the signal. Related to this, it is unclear how the authors reached the point that "...the C>A mutator phenotype associated with the locus on chromosome 6", when they only detected significant IHD signal at rs46276051 (on Chr6), when conditioning on D genotypes at the rs27509845 (on Chr4) and no significant signal for any 1-mer mutation type by traditional mapping. The authors need to explain how they deduced that C>A mutation is the major source of the signal. In addition, beyond C>A mutations, can mutation types other than C>A contribute to the IHD signal at rs46276051? More generally, I hope the authors can provide some guidelines on how to narrow a significant IHD signal to specific candidate mutation type(s) affected, which will make the method more useful to other researchers.

We thank the reviewer for pointing out this gap in our logic. We omitted specific instructions for narrowing down an IHD signal to specific mutation type(s) for a few reasons. First, this can be addressed using mutational signature analysis methods that are in widespread use. For example, upon identifying one or more candidate mutator loci, we can enter the mutation spectra of samples with each possible mutator genotype into a program (e.g., SigProfilerExtractor) to determine which combinations of mutation types occur proportionally more often in the genomes that harbor mutators (see Figure 3c in our manuscript). A second approach for narrowing down an IHD signal, highlighted in Figure 3a (and now described in the text of the Results section at lines 256-261), is to simply test which mutation type proportion(s) differ significantly between groups of samples with and without a candidate mutator (for example, with a Chi-square test of independence for each mutation type).

Although this second approach incurs a multiple testing burden, the burden is offset somewhat by using IHD to identify mutator loci, rather than performing association tests for every possible mutation type to begin with. Although Figure 3a only shows the significant difference in C>A fraction among BXDs with different mutator locus genotypes, Figure 3-figure supplement 1 shows the complete set of 1-mer spectrum comparisons. It is possible that this second approach would not prove very useful in the case of a mutator with a “flat” signature (i.e., a mutator that slightly perturbs the rates of many different mutation types), but in our case it clearly shows which mutation type is affected.

1. To account for differential relatedness between the inbred lines, the authors regressed the cosine distance between the two aggregate mutation spectra on the genome-wide genetic similarity and took the residual as the adjusted test metric. What is the value of the slope from this regression? If significantly non-zero, this would support a polygenic architecture of the mutation spectrum phenotype, which could be interesting. If not, is this adjustment really necessary? In addition, is the intercept assumed to be zero for this regression, and does such an assumption matter? I would appreciate seeing a supplemental figure on this regression.

The reviewer raises a good question. We find that the slope of the "distance vs. genetic similarity" regression is significantly non-zero, though the slope estimate itself is small. A plot of cosine distance vs. genome-wide genetic similarity (using all BXDs) is shown below in Author response image 2:

Author response image 2.

Relationship between cosine distance and genetic similarity in the BXDs. As described in the Materials and Methods, we computed two values at each marker in the BXDs: 1) the cosine distance between the aggregate mutation spectra of BXDs with either B or D genotypes at the marker, and 2) the correlation between genome-wide D allele frequencies in BXDs with either B or D genotypes at the marker. We then regressed these two values across all genome-wide markers.

This result indicates that if two groups of BXDs (one with D genotypes and one with B genotypes at a given locus) are more genetically similar, their mutation spectra are also more similar. Since the regression slope estimate is significantly non-zero (p < 2.2e-16), we believe that it's still worth using residuals as opposed to raw cosine distance values. This result also suggests that there may be a polygenic effect on the mutation spectrum in the BXDs.

We have also generated a plot showing the cosine distance between the mutation spectra of every possible pair of BXDs, regressed against the genetic similarity between each of those pairs (Author Response image 3). Here, the potential polygenic effects on mutation spectra similarity are perhaps more obvious.

Author response image 3.

Pairwise cosine distance between BXD mutation spectra as a function of genetic similarity. We computed two values for every possible pair of n = 117 BXDs: 1) the cosine distance between the samples' individual 1-mer mutation spectra and 2) the correlation coefficient between the samples' genome-wide counts of D alleles.

Private Comments

1. It will also be useful to see how the power of IHD and QTL mapping depend on the allele frequency of the mutator allele and the sample size, as mutator alleles are likely rare or semi-rare in natural populations (such as the human de novo mutation dataset that the authors mentioned).

This is another good suggestion. In general, we'd expect the power of both IHD and QTL mapping to decrease as a function of mutator allele frequency. At the same time, we note that the power of these scans should mostly depend on the absolute number of carriers of the mutator allele and less on its frequency. In the BXD mouse study design, we observe high frequency mutators but also a relatively small sample size of just over 100 individuals. In natural human populations, mutator frequencies might be orders of magnitude smaller, but sample sizes may be orders of magnitude larger, especially as new cohorts of human genomes are routinely being sequenced. So, we expect to have similar power to detect a mutator segregating at, say, 0.5% frequency in a cohort of 20,000 individuals, as we would to detect a mutator segregating at 50% frequency in a dataset of 200 individuals.

To more formally address the reviewer's concern, we performed a series of simulations in which we simulated a population of 100 haplotypes. We assigned the same average number of mutations to each haplotype but allowed the allele frequency of the mutator allele to vary between 0.1, 0.25, and 0.5. The results of these simulations are shown in Author response image 4 and reveal that AMSD tends to have greater power than QTL mapping at lower mutator allele frequencies. We now mention these simulations in the text at lines 135-138 and include the simulation results in Figure 1-figure supplement 4.

Author response image 4.

Power of AMSD and QTL mapping on simulated haplotypes with variable marker allele frequencies. We simulated germline mutations on the specified number of haplotypes (as described in the manuscript), but simulated genotypes at the mutator allele such that "A" alleles were at the specified allele frequency.

1. In the Methods section of "testing for epistasis between the two mutator loci", it will be helpful to explicitly lay out the model and assumptions in mathematical formulae, in addition to the R scripts. For example, are the two loci considered independent when their effects on mutation rate is multiplicative or additive? Given the R scripts provided, it seems that the two loci are assumed to have multiplicative effects on the mutation rate, and that the mutation count follows a Poisson distribution with mean being the mutation rate times ADJ_AGE (i.e., the mutation opportunity times the number of generations of an inbred line). However, this is not easily understandable for readers who are not familiar with R language. In addition, I hope the authors can be more specific when discussing the epistatic interaction between the two loci by explicitly saying "synergistic effects beyond multiplicative effects on the C>A mutation rate".

The reviewer raises a good point about the clarity of our descriptions of tests for epistasis. We have now added a more detailed description of these tests in the section of the Materials and Methods beginning at line 875. We have also added a statement to the text at lines 289-291: “the combined effects of D genotypes at both loci exceed the sum of marginal effects of D genotypes at either locus alone.” We hope that this will help clarify the results of our tests for statistical epistasis.

Reviewer 2 (Public Review):

1. The main limitation of the approach is that it is difficult to see how it might be applied beyond the context of mutation accumulation experiments using recombinant inbred lines. This is because the signal it detects, and hence its power, is based on the number of extra accumulated mutations linked to (i.e. on the same chromosome as) the mutator allele. In germline mutation studies of wild populations the number of generations involved (and hence the total number of mutations) is typically small, or else the mutator allele becomes unlinked from the mutations it has caused (due to recombination), or is lost from the population altogether (due to chance or perhaps selection against its deleterious consequences).

The reviewer is correct that as it currently exists, IHD is mostly limited to applications in recombinant inbred lines (RILs) like the BXDs. This is due to the fact that IHD assumes that each diploid sample harbors one of two possible genotypes at a particular locus and ignores the possibility of heterozygous genotypes for simplicity. In natural, outbreeding populations, this assumption will obviously not hold. However, as we plan to further iterate on and improve the IHD method, we hope that it will be applicable to a wider variety of experimental systems in the future. We have added additional caveats about the applicability of our method to other systems in the text at lines 545-550.

Private Comments

1. On p. 8, perhaps I've misunderstood but it's not clear in what way the SVs identified were relevant to the samples used in this dataset - were the founder strains assembled? Is there any chance that additional SVs were present, e.g. de novo early in the accumulation line?

Our description of this structural variation resource could have been clearer. The referenced SVs were identified in Ferraj et al. (2023) by generating high-quality long read assemblies of inbred laboratory mice. Both DBA/2J and C57BL/6J (the founder strains for the BXD resource) were included in the Ferraj et al. SV callset. We have clarified our description of the callset at lines 247-248.

It is certainly possible that individual BXD lines have accumulated de novo structural variants during inbreeding. However, these "private" SVs are unlikely to produce a strong IHD association signal (via linkage to one of the ~7,000 markers) at either the chromosome 4 or chromosome 6 locus, since we only tested markers that were at approximately 50% D allele frequency among the BXDs.

1. On p. 13, comparing the IHD and QTL approaches, regarding the advantage of the former in that it detects the combined effect of multiple k-mer mutation types, would it not be straightforward to aggregate counts for different types in a QTL setting as well?

The mutation spectrum is a multi-dimensional phenotype (6-dimensional if using the 1-mer spectrum, 96-dimensional if using the 3-mer spectrum, etc.). Most QTL mapping methods use linear models to test for associations between genotypes and a 1-dimensional phenotype (e.g., body weight, litter size). In the past, we used QTL mapping to test for associations between genotypes and a single element of the mutation spectrum (e.g., the rate of C>A mutations), but there isn't a straightforward way to aggregate or collapse the mutation spectrum into a 1dimensional phenotype that retains the information contained within the full 1-mer or 3-mer spectrum. For that reason, we developed the "aggregate mutation spectrum" approach, as it preserves information about the complete mutation spectrum in each group of strains.

The reviewer is correct that we could also aggregate counts of different mutation types to, say, perform a QTL scan for the load of a specific mutational signature. For example, we could first perform standard mutational signature analysis on our dataset and then test for QTLs associated with each signature that is discovered. However, this approach would not solve the second problem that our method is designed to solve: the appropriate weighting of samples based on how many mutations they contain.

1. pp. 15-16: In the discussion of how you account for relatedness between strains, I found the second explanation (on p. 16) much clearer. It would be interesting to know how much variance was typically accounted for by this regression?

As shown in the response to Reviewer 1, genotype similarity between genotype groups (i.e., those with either D or B genotypes at a marker) generally explains a small amount of variance in the cosine distance between those groups (R2 ~= 0.007). However, since the slope term in that regression is significantly non-zero, correcting for this relationship should still improve our power relative to using raw cosine distance values that are slightly confounded by this relationship.

1. Similarly, in the section on Applying the IHD method to the BXDs (pp. 18-19), I think this description was very useful, and some or all of this description of the experiment (and how the DNMs in it arise) could profitably be moved to the introduction.

We appreciate the reviewer’s feedback about the details of the BXD cohort. Overall, we feel the description of the BXDs in the Introduction (at lines 65-73) is sufficient to introduce the cohort, though we now add some additional detail about variability in BXD inbreeding duration (at lines 89-93) to the Introduction as well, since it is quite relevant to some of the new simulation results presented in the manuscript.

1. A really minor one, not sure if this is for the journal or the authors, but it would be much better to include both page and line numbers in any version of an article for review. My pdf had neither!

We apologize for the lack of page/line numbers in the submitted PDF. We have now added line numbers to the revised version of the manuscript.

Reviewer 3 (Public Review):

1. Under simulated scenarios, the authors' new IHD method is not appreciably more powerful than conventional QTL mapping methods. While this does not diminish the rigor or novelty of the authors findings, it does temper enthusiasm for the IHD method's potential to uncover new mutators in other populations or datasets. Further, adaptation of this methodology to other datasets, including human trios or multigenerational families, will require some modification, which could present a barrier to broader community uptake. Notably, BXD mice are (mostly) inbred, justifying the authors consideration of just two genotype states at each locus, but this decision prevents out-of-the-box application to outbred populations and human genomic datasets. Lastly, some details of the IHD method are not clearly spelled out in the paper. In particular, it is unclear whether differences in BXD strain relatedness due to the breeding epoch structure are fully accounted for in permutations. The method's name - inter-haplotype distance - is also somewhat misleading, as it seems to imply that de novo mutations are aggregated at the scale of sub-chromosomal haplotype blocks, rather than across the whole genome.

The reviewer raises very fair concerns. As mentioned in response to a question from Reviewer 1, we performed additional simulation experiments that demonstrate the improved power of IHD (as compared to QTL mapping) in situations where mutation counts are variable across haplotypes or when mutator alleles are present at allele frequencies <50% (see Author response image 2 and 3, as well as new supplements to Figure 1 in the manuscript). However, the reviewer is correct that the IHD method is not applicable to collections of outbred individuals (that is, individuals with both heterozygous and homozygous genotypes), which will limit its current applications to datasets other than recombinant inbred lines. We have added a mention of these limitations to the Results at lines 138-141 and the Discussion at lines 545-550, but plan to iterate on the IHD method and introduce new features that enable its application to other datasets. We have also explicitly stated that we account for breeding epochs in our permutation tests in the Materials and Methods at lines 670-671. Both Reviewer 1 and Reviewer 3 raised concerns about the name of our method, and we have therefore changed “inter-haplotype distance” to “aggregate mutation spectrum distance” throughout the manuscript.

1. Nominating candidates within the chr6 mutator locus requires an approach for defining a credible interval and excluding/including specific genes within that interval as candidates. Sasani et al. delimit their focal window to 5Mb on either side of the SNP with the most extreme P-value in their IHD scan. This strategy suffers from several weaknesses. First, no justification for using 10 Mb window, as opposed to, e.g., a 5 Mb window or a window size delimited by a specific threshold of P-value drop, is given, rendering the approach rather ad hoc. Second, within their focal 10Mb window, the authors prioritize genes with annotated functions in DNA repair that harbor protein coding variants between the B6 and D2 founder strains. While the logic for focusing on known DNA repair genes is sensible, this locus also houses an appreciable number of genes that are not functionally annotated, but could, conceivably, perform relevant biological roles. These genes should not be excluded outright, especially if they are expressed in the germline. Further, the vast majority of functional SNPs are non-coding, (including the likely causal variant at the chr4 mutator previously identified in the BXD population). Thus, the author's decision to focus most heavily on coding variants is not well-justified. Sasani et al. dedicate considerable speculation in the manuscript to the likely identity of the causal variant, ultimately favoring the conclusion that the causal variant is a predicted deleterious missense variant in Mbd4. However, using a 5Mb window centered on the peak IHD scan SNP, rather than a 10Mb window, Mbd4 would be excluded. Further, SNP functional prediction accuracy is modest [e.g., PMID 28511696], and exclusion of the missense variant in Ogg1 due its benign prediction is potentially premature, especially given the wealth of functional data implicating Ogg1 in C>A mutations in house mice. Finally, the DNA repair gene closest to the peak IHD SNP is Rad18, which the authors largely exclude as a candidate.

We agree that the use of a 10 Mb window, rather than an empirically derived confidence interval, is a bit arbitrary and ad hoc. To address this concern, we have implemented a bootstrap resampling approach (Visscher et al. 1996, Genetics) to define confidence intervals surrounding IHD peaks. We have added a description of the approach to the Materials and Methods at lines 609-622, but a brief description follows. In each of N trials (here, N = 10,000), we take a bootstrap sample of the BXD phenotype and genotype data with replacement. We then perform an IHD scan on the chromosome of interest using the bootstrap sample and record the position of the marker with the largest cosine distance value (i.e., the "peak" marker). After N trials, we calculate the 90% confidence interval of bootstrapped peak marker locations; in other words, we identify the locations of two genotyped markers, between which 90% of all bootstrap trials produced an IHD peak. We note that bootstrap confidence intervals can exhibit poor "coverage" (a measure of how often the confidence intervals include the "true" QTL location) in QTL mapping studies (see Manichaikul et al. 2006, Genetics), but feel that the bootstrap is more reasonable than simply defining an ad hoc interval around an IHD peak.

The new 90% confidence interval surrounding the IHD peak on chromosome 6 is larger than the original (ad hoc) 10 Mbp window, now extending from around 95 Mbp to 114 Mbp. Notably, the new empirical confidence interval excludes Mbd4. We have accordingly updated our Results and Discussion sections to acknowledge the fact that Mbd4 no longer resides within the confidence interval surrounding the IHD peak on chromosome 6 and have added additional descriptions of genes that are now implicated by the 90% confidence interval. Given the uncertainties associated with using bootstrap confidence intervals, we have retained a brief discussion of the evidence supporting Mbd4 in the Discussion but focus primarily on Ogg1 as the most plausible candidate.

The reviewer raises a valid concern about our treatment of non-DNA repair genes within the interval surrounding the peak on chromosome 6. We have added more careful language to the text at lines 219-223 to acknowledge the fact that non-annotated genes in the confidence interval surrounding the chromosome 6 peak may play a role in the epistatic interaction we observed.

The reviewer also raises a reasonable concern about our discussions of both Mbd4 and Ogg1 as candidate genes in the Discussion. Since Mbd4 does not reside within the new empirical bootstrap confidence interval on chromosome 6 and given the strong prior evidence that Ogg1 is involved in C>A mutator phenotypes (and is in the same gene network as Mutyh), we have reframed the Discussion to focus on Ogg1 as the most plausible candidate gene (see lines 357360).

Using the GeneNetwork resource, we also more carefully explored the potential effects of noncoding variants on the C>A mutator phenotype we observed on chromosome 6. We have updated the Results at lines 240-246 and the Discussion at line 439-447 to provide more evidence for regulatory variants that may contribute to the C>A mutator phenotype. Specifically, we discovered a number of strong-effect cis-eQTLs for Ogg1 in a number of tissues, at which D genotypes are associated with decreased Ogg1 expression. Given new evidence that the original mutator locus we discovered on chromosome 4 harbors an intronic mobile element insertion that significantly affects Mutyh expression (see Ferraj et al. 2023, Cell Genomics), it is certainly possible that the mutator phenotype associated with genotypes on chromosome 6 may also be mediated by regulatory, rather than coding, variation.

1. Additionally, some claims in the paper are not well-supported by the author's data. For example, in the Discussion, the authors assert that "multiple mutator alleles have spontaneously arisen during the evolutionary history of inbred laboratory mice" and that "... mutational pressure can cause mutation rates to rise in just a few generations of relaxed selection in captivity". However, these statements are undercut by data in this paper and the authors' prior publication demonstrating that a number of candidate variants are segregating in natural mouse populations. These variants almost certainly did not emerge de novo in laboratory colonies, but were inherited from their wild mouse ancestors. Further, the wild mouse population genomic dataset used by the authors falls far short of comprehensively sampling wild mouse diversity; variants in laboratory populations could derive from unsampled wild populations.

The reviewer raises a good point. In our previous publication (Sasani et al. 2022, Nature), we hypothesized that Mutyh mutator alleles had arisen in wild, outbreeding populations of Mus musculus, and later became fixed in inbred strains like DBA/2J and C57BL/6J. However, in the current manuscript, we included a statement about mutator alleles "spontaneously arising during the evolutionary history of inbred laboratory mice" to reflect new evidence (from Ferraj et al. 2023, Cell Genomics) that the mutator allele we originally identified in Mutyh may not be wild derived after all. Instead, Ferraj et al. suggest that the C>A mutator phenotype we originally identified is caused by an intronic mobile element insertion (MEI) that is present in DBA/2J and a handful of other inbred laboratory strains. Although this MEI may have originally occurred in a wild population of mice, we wanted to acknowledge the possibility that both the original Mutyh mutator allele, as well as the new mutator allele(s) we discovered in this manuscript, could have arisen during the production and inbreeding of inbred laboratory lines. We have also added language to the Discussion at lines 325-327 to acknowledge that the 67 wild mice we analyzed do not comprise a comprehensive picture of the genetic diversity present in wild-derived samples.

We have added additional language to the Discussion at lines 349-357 in which we acknowledge that the chromosome 6 mutator allele might have originated in either laboratory or wild mice and elaborate on the possibility that mutator alleles with deleterious fitness consequences may be more likely to persist in inbred laboratory colonies.

1. Finally, the implications of a discovering a mutator whose expression is potentially conditional on the genotype at a second locus are not raised in the Discussion. While not a weakness per se, this omission is perceived to be a missed opportunity to emphasize what, to this reviewer, is one of the most exciting impacts of this work. The potential background dependence of mutator expression could partially shelter it from the action of selection, allowing the allele persist in populations. This finding bears on theoretical models of mutation rate evolution and may have important implications for efforts to map additional mutator loci. It seems unfortunate to not elevate these points.

We agree and have added additional discussion of the possibility that the C>A mutator phenotypes in the BXDs are a result of interactions between the expression of two DNA repair genes in the same base-excision network to the Discussion section at lines 447-449.

Private comments

1. The criteria used to determine or specify haplotype size are not specified in the manuscript. I mention this above but reiterate here as this was a big point of confusion for me when reading the paper. Haplotype length is important consideration for overall power and for proper extension of this method to other systems/populations.

We may not have been clear enough in our description of our method, and as suggested by Reviewer 1, the name "inter-haplotype distance" may also have been a source of confusion. At a given marker, we compute the aggregate mutation spectrum in BXDs with either B or D genotypes using all genome-wide de novo mutations observed in those BXDs. Since the BXDs were inbred for many generations, we expect that almost all de novo germline mutations observed in an RIL are in near-perfect linkage with the informative genotypes used for distance scans. Thus, the "haplotypes" used in the inter-haplotype distance scans are essentially the lengths of entire genomes.

1. Results, first paragraph, final sentence. I found the language here confusing. I don't understand how one can compute the cosine distance at single markers, as stated. I'm assuming cosine distance is computed from variants residing on haplotypes delimited by some defined window surrounding the focal marker?

As discussed above, we aggregate all genome-wide de novo mutations in each group of BXDs at a given marker, rather than only considering DNMs within a particular window surrounding the marker. The approach is discussed in greater detail in the caption of Figure 1.

1. Nominating candidates for the chr6 locus, Table 1. It would be worth confirming that the three prioritized candidates (Setmar, Ogg1, and Mbd4) all show germline expression.

Using the Mouse Genome Informatics online resource, we confirmed that all prioritized candidate genes (now including Setmar and Ogg1, but not Mbd4) are expressed in the male and female gonads, and mention this in the Results at lines 228 and 233-234.

1. Does the chr6 peak on the C>A LOD plot (Figure 2- figure supplement 1) overlap the same peak identified in the IHD scan? And, does this peak rise to significance when using alpha = 0.05? Given that the goal of these QTL scans is to identify loci that interact with the C>A mutator on chr4, it is reasonable to hypothesize that the mutation impact of epistatic loci will also be restricted to C>A mutations. Therefore, I am not fully convinced that the conservative alpha = 0.05/7 threshold is necessary.

The chromosome 6 peak in Figure 2-figure supplement 1 does, in fact, overlap the peak marker we identified on chromosome 6 using IHD. One reason we decided to use a more conservative alpha of (0.05 / 7) is that we wanted these results to be analogous to the ones we performed in a previous paper (Sasani et al. 2022, Nature), in which we first identified the mutator locus on chromosome 4. However, the C>A peak does not rise to genome-wide significance if we use a less conservative alpha value of 0.05 (see Author response image 5). As discussed in our response to Reviewer 1, we find that QTL mapping is not as powerful as IHD when haplotypes have accumulated variable numbers of germline mutations (as in the BXDs), which likely explains the fact that the peak on chromosome 6 is not genome-wide significant using QTL mapping.

Author response image 5.

QTL scan for the fraction of C>A mutations in BXDs harboring D alleles at the locus near Myth QTL scan was performed at a genome-wide significance alpha of 0.05, rather than 0.05/7.

1. Is there significant LD between the IHD peaks on chr6 and chr4 across the BXD? If so, it could suggest that the signal is driven by cryptic population structure that is not fully accounted for in the author's regression based approach. If not, this point may merit an explicit mention in the text as an additional validation for the authenticity of the chr6 mutator finding.

This is a good question. We used the scikit-allel Python package to calculate linkage disequilibrium (LD) between all pairs of genotyped markers in the BXD cohort, and found that the two peak loci (on chromosomes 4 and 6) exhibit weak LD (r2 = 4e-5). We have added a mention of this to the main text of the Results at lines 212-213. That being said, we do not think the chromosome 6 mutator association (or the apparent epistasis between the alleles on chromosomes 4 and 6) could be driven by cryptic population structure. Unlike in human GWAS and other association studies in natural populations, there is no heterogeneity in the environmental exposures experienced by different BXD subpopulations. In humans, population structure can create spurious associations (e.g., between height and variants that are in LD and are most common in Northern Europe), but this requires the existence of a phenotypic gradient caused by genetic or environmental heterogeneity that is not likely to exist in the context of inbred laboratory mice that are all the progeny of the same two founder strains.

1. Discussion, last sentence of the "Possible causal alleles..." section: I don't understand how the absence of the Mariner-family domain leads the authors to this conclusion. Setmar is involved in NHEJ, which to my knowledge is not a repair process that is expected to have a specific C>A mutation bias. I think this is grounds enough for ruling out its potential contributions, in favor of focusing on other candidates, (e.g., Mbd4 and Ogg1).

The reviewer raises a good point. Our main reason for mentioning the absence of the Marinerfamily domain is that even if NHEJ were responsible for the C>A mutator phenotype, it likely wouldn't be possible for Setmar to participate in NHEJ without the domain. However, the reviewer is correct that NHEJ is not expected to cause a C>A mutation bias, and we have added a mention of this to the text as well at lines 379-382.

1. Discussion, second to last paragraph of section "Mbd4 may buffer...": The authors speculate that reduced activity of Mbd4 could modulate rates of apoptosis in response to DNA damage. This leads to the prediction that mice with mutator alleles at both Mutyh and Mbd4 should exhibit higher overall mutation rates compared to mice with other genotypes. This possibility could be tested with the authors' data.

The reviewer raises a good question. As mentioned above, however, we implemented a new approach to calculate confidence intervals surrounding distance peaks and found that this empirical approach (rather than the ad hoc 10-Mbp window approach we used previously) excluded Mbd4 from the credible interval. Although we still mention Mbd4 as a possible candidate (since it still resides within the 10 Mbp window), we have refactored the Discussion section to focus primarily on the evidence for Ogg1 as a candidate gene on chromosome 6.

In any case, we do not observe that mice with mutator alleles at both the chromosome 4 and chromosome 6 loci have higher overall mutation rates compared to mice with other genotype combinations. This may not be terribly surprising, however, since C>A mutations only comprise about 10% of all possible mutations. Thus, given the variance in other 1-mer mutation counts, even a substantial increase in the C>A mutation rate might not have a detectable effect on the overall mutation rate. Indeed, in our original paper describing the Mutyh mutator allele (Sasani et al. 2022, Nature), we did not identify any QTL for the overall mutation rate in the BXDs and found that mice with the chromosome 4 mutator allele only exhibited a 1.11X increase in their overall mutation rates relative to mice without the mutator allele.

1. Methods, "Accounting for BXD population structure": An "epoch-aware" permutation strategy is described here, but it is not clear when (and whether) this strategy is used to determine significance of IHD P-values.

We have added a more explicit mention of this to the Methods section at lines 670-671, as we do, in fact, use the epoch-aware permutation strategy when calculating empirical distance thresholds.

1. The simulation scheme employed for power calculations is highly specific to the BXD population. This is not a weakness, and perfectly appropriate to the study population used here. However, it does limit the transferability of the power analyses presented in this manuscript to other populations. This limitation may merit an explicit cautionary mention to readers who may aspire to port the IHD method over to their study system.

This is true. Our simulation strategy is relatively simple and makes a number of assumptions about the simulated population of haplotypes (allele frequencies normally distributed around 0.5, expected rates of each mutation type, etc.). In response to concerns from Reviewer 1, we performed an updated series of simulations in which we varied some of these parameters (mutator allele frequencies, mean numbers of mutations on haplotypes, etc.). However, we have added a mention of the simulation approach's limitations and specificity to the BXDs to the text at lines 545-550.

Author Response

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

Author response:

Reviewer #1:

The main objective of this study is to achieve the development of a synthetic autotroph using adaptive laboratory evolution. To accomplish this, the authors conducted chemostat cultivation of engineered E. coli strains under xylose-limiting conditions and identified autotrophic growth and the causative mutations. Additionally, the mutational mechanisms underlying these causative mutations were also explored with drill down assays. Overall, the authors demonstrated that only a small number of genetic changes were sufficient (i.e., 3) to construct an autotrophic E. coli when additional heterologous genes were added. While natural autotrophic microorganisms typically exhibit low genetic tractability, numerous studies have focused on constructing synthetic autotrophs using platform microorganisms such as E. coli. Consequently, this research will be of interest to synthetic biologists and systems biologists working on the development of synthetic autotrophic microorganisms. The conclusions of this paper are mostly well supported by appropriate experimental methods and logical reasoning. However, further experimental validation of the mutational mechanisms involving rpoB and crp would enhance readers' understanding and provide clearer insights, despite acknowledgement that these genes impact a broad set of additional genes. Additionally, a similar study, 10.1371/journal.pgen.1001186, where pgi was deleted from the E. coli genome and evolved to reveal an rpoB mutation is relevant to this work and should be placed in the context of the presented findings.

We thank the reviewer for pointing this study out. It is very interesting that a mutation in a similar region in RpoB was observed in a related context of Pgi loss of activity. We have added a reference to this study in our text (Page 11, line 21).

he authors addressed rpoB and crp as one unit and performed validation. They cultivated the mutant strain and wild type in a minimal xylose medium with or without formate, comparing their growth and NADH levels. The authors argued that the increased NADH level in the mutant strain might facilitate autotrophic growth. Although these phenotypes appear to be closely related, their relationship cannot be definitively concluded based on the findings presented in this paper alone. Therefore, one recommendation is to explore investigating transcriptomic changes induced by the rpoB and crp mutations. Otherwise, conducting experimental verification to determine whether the NADH level directly causes autotrophic growth would provide further support for the authors' claim.

We appreciate the valuable comment and agree that the work was lacking such an analysis. Due to various reasons we have opted to use a proteomic approach which we feel fulfills the same purpose as the transcriptomics suggestion. We found interesting evidence in up-regulation of the fdoGH operon (comprising the native formate dehydrogenase O enzyme complex) which could indicate why there is an increase in NADH/NAD+ levels. We also hypothesize that this upregulation might be important more generally by drawing comparisons to natural chemo-autotrophs.

Further experimental work (which we were not able to include in the current study) could help validate this link by deleting fdoGH and observing a loss of phenotype and, on the flip side, directly overexpressing the fdoGH operon and observing an increase in the NADH/NAD+ ratio. Indeed, if this overexpression were to prove sufficient for achieving an autotrophic phenotype without the mutations in the global transcription regulators, it would be a much more transparent design.

We have added a section titled "Proteomic analysis reveals up-regulation of rPP cycle and formate-associated genes alongside down-regulation of catabolic genes" to the Results based on this analysis.

• It would be beneficial to provide a more detailed explanation of the genetic background before the evolution stage, specifically regarding the ∆pfk and ∆zwf mutations. Furthermore, it is suggested to include a figure that provides a comprehensive depiction of the reductive pentose phosphate pathway and the bypass pathway. These will help readers grasp the concept of the "metabolic scaffold" as proposed by the authors.

We agree with the reviewer that this could be helpful and we added a reference to the original paper Gleizer et al. 2019 that reported this design and also includes the relevant figure. We feel that the figure should not be added to the current manuscript as we continue to show that this design is not relevant in the context of the three reported mutations and such a figure could distract the attention of the reader from the main takeaways of the current study.

• Despite the essentiality of the rpoB mutation (A1245V) to the autotrophic phenotype in the final strain, the inclusion of this mutation in step C1 does not appear to be justified. According to line 37 on page 3, the authors chose to retain the unintended mutation in rpoB based on its essentiality to the phenotype observed in other evolved strains. However, it should be noted that the mutations found in the evolved strain I, II, and III (P552T or D866E) were entirely different from the unintended mutation (A1245V) during genetic engineering. This aspect should be revised to avoid confusion among readers.

Thank you for pointing this issue out, we added a clarification in the text (page 4 line 7) to avoid such confusion. We believe this point is much clearer now.

The rpoB mutation which was shown to be essential in the study is indeed known to be common in ALE experiments in E. coli. Thus, I searched the different rpoB mutations in ALEdb in E. coli and I was able to find a similar mutation in a study where pgi was knocked out and then evolved. https://doi.org/10.1371/journal.pgen.1001186 This study seems very relevant given that pgi was a key mutation in the compact set of this work and the section "Modulation of a metabolic branch-point activity increased the concentration of rPP metabolites" informs that loss of function mutations in pgi were also found. The findings of this study should thus be put in the context of the previous related ALE study. I would recommend a similar analysis of crp mutations from studies in ALEdb to see if there are similar mutations in this gene as well or if this a unique mutation.

We thank the reviewer for bringing this publication to our attention. We have addressed this observation in the main text (page 11 , line 21). We agree that it could have some connection to the pgi mutation yet we would not want to overspeculate about this role, as we also found the exact same mutation (A1245V) as an adaptation to higher temperature in another E. coli study (Tenaillon et al. 2012). We would like to bring forward the fact that the two reported rpoB mutations are always accompanied by another mutation with pleiotropic effects, either in the transcription factor Crp or in another RNA polymerase subunit (e.g RpoC). As such many epistatic effects could occur, one of which we also report here in page 13, line 18. In conclusion, although there could be a connection between the rpoB and pgi mutations, it could be a mere coincidence and the two mutations could exhibit two distinct roles in two distinct phenotypes.

We also would like to thank the reviewer for suggesting a similar analysis for crp and found another mutation at a nearby residue with strong adaptive effects and mentioned it in our main text.

Can the typical number of mutations found in a given ALE experiment be directly compared to those found in this study? It seems like a retrospective analysis of other ALE studies to show how many mutations typically occur in an ALE study and sets which were found to be causal to reproduce the phenotype of interest (through similar reverse engineering in the starting strain) should be presented. Again, the authors cite ALEdb which should provide direct numbers of mutations found in similar ALE studies with E. coli and one could then examine them to find sets of clearly causal mutations which recreate phenotypes of interest. Such an analysis would go a long way in supporting the main finding of "small number" of mutations.

Discussion, page 12, line 42. "This could serve as a promising strategy for achieving minimally perturbed genotypes in future metabolic engineering attempts". There is an entire body of work around growth-coupled production which can be predicted and evolved with a genome-scale metabolic model and ALE. Thus, if this statement is going to be made, relevant studies should be cited and placed in context.

The reviewer raises an important point which could indeed yield an interesting perspective. However, it would be difficult to perform this comparison in practice since many of the studies published on ALEdb have not isolated essential mutations from other mutation incidents nor have they determined the role of each mutation in the reported phenotypes. For example, many ALE trajectories include a hypermutator that greatly increases the number of irrelevant mutations and it is nearly impossible to sieve through them to find an essential set.

Moreover, it is hard to compare the “level of difficulty” of achieving one phenotype over another and therefore feel that even though such an analysis would be insightful, it requires an amount of work which is outside the scope of this study.

Finally, we would like to highlight our approach of using the iterative approach, isolating the relevant consensus mutations and repeating this process until no evolution process is required, we are not aware of prior studies that used this approach.

We now clarified what we mean by "promising strategy" in the discussion in order to avoid any false claims about novelty (page 16 line 32): "Using metabolic growth-coupling as a temporary 'metabolic scaffold' that can be removed, could serve as a promising strategy for achieving minimally perturbed genotypes in future metabolic engineering attempts."

Reviewer #2:

Synthetic autotrophy of biotechnologically relevant microorganisms offers exciting chances for CO2 neutral or even CO2 negative production of goods. The authors' lab has recently published an engineered and evolved Escherichia coli strain that can grow on CO2 as its only carbon source. Lab evolution was necessary to achieve growth. Evolved strains displayed tens of mutations, of which likely not all are necessary for the desired phenotype.

In the present paper the authors identify the mutations that are necessary and sufficient to enable autotrophic growth of engineered E. coli. Three mutations were identified, and their phenotypic role in enhancing growth via the introduced Calvin-Benson-Bassham cycle were characterized. It was demonstrated that these mutations allow autotrophic growth of E. coli with the introduced CBB cycle without any further metabolic intervention. Autotrophic growth is demonstrated by 13C labelling with 13C CO2, measured in proteinogenic amino acids. In Figures 2B and S1, the labeling data are shown, with an interval of the "predicted range under 13CO2".

Here, the authors should describe how this interval was derived.

The methodology is clearly described and appropriate.

The present results will allow other labs to engineer E. coli and other microorganisms further to assimilate CO2 efficiently into biomass and metabolic products. The importance is evident in the opportunity to employ such strain in CO2 based biotech processes for the production of food and feed protein or chemicals, to reduce atmospheric CO2 levels and the consumption of fossil resources.

Please describe in the methodology how the interval of the predicted range of 13C labeling was derived for Figures 2B and S1. Was it calculated by the dilution factor during 4 generations, or did you predict the label incorporation individually with a metabolic model?

The text needs careful editing, some sentences are incomplete and there are frequent inconsistencies in writing metabolites and enzymes.

P2L6: unclear sentence (incomplete?)

P2L19: pastoris with lower case "p"

P2L40: incomplete sentence

P2L42: here, and at many other places, the writing of RuBisCO needs to be aligned. It is an abbreviation and should begin with a capital letter. Most commonly it is written as RuBisCO which I would suggest - please unify throughout the text.

P3L3: formate dehydrogenase ... metabolites and enzymes with lower case letter. And, no hyphen here.

P5L4: delete the : after unintentionally

P6L16: carboxylation of RuBP (it is not CO2 that is carboxylated - if any, CO2 is carboxylating)

P7L25: phosphoglucoisomerase (lower case)

P8L5: in line

P8L9: part of glycolysis/ ...

P10L4: pentose phosphates (lower case, no hyphen).

P10L4: all metabolites lower case

P12L28: incomplete sentence

P18L4: Escherichia coli in italics P18L15: Pseudomonas sp. in italics P18L16: ... promoter and with a strong ...

P20, chapter Metabolomics: put the numbers of 12C and 13C in superscript P23L9: pentose phosphates ; all metabolites in lower case (as above) P23: all 12C and 13C with superscript numbers.

Response to reviewer #2:

We thank the reviewer for their comments, and for pointing out the need to clarify how we derived the predicted range of 13C labeling. We edited the text accordingly, and added the relevant calculation to the methods section (under the “13C Isotopic labeling experiment”). We would like to also thank the reviewer for the required text improvements, which were implemented. 

Reviewer #3:

The authors previously showed that expressing formate dehydrogenase, rubisco, carbonic anhydrase, and phosphoribulokinase in Escherichia coli, followed by experimental evolution, led to the generation of strains that can metabolise CO2. Using two rounds of experimental evolution, the authors identify mutations in three genes - pgi, rpoB, and crp - that allow cells to metabolise CO2 in their engineered strain background. The authors make a strong case that mutations in pgi are loss-of-function mutations that prevent metabolic efflux from the reductive pentose phosphate autocatalytic cycle. The authors also argue that mutations in crp and rpoB lead to an increase in the NADH/NAD+ ratio, which would increase the concentration of the electron donor for carbon fixation. While this may explain the role of the crp and rpoB mutations, there is good reason to think that the two mutations have independent effects, and that the change in NADH/NAD+ ratio may not be the major reason for their importance in the CO2-metabolising strain.

We thank the reviewer for their comments and constructive feedback.

We agree that there is probably a broader effect caused by the rpoB and crp mutations, besides the change in the NADH/NAD+ ratio. Hence, we performed a proteomics analysis, comparing the rpoB and crp mutations on a WT background to an autotrophic E.coli, searching for a mutual change in both strains compared to their "ancestors". We found up-regulation of rPP cycle and formate-associated genes, and a down-regulation of catabolic genes. We added a section dedicated to this matter under the title "Proteomic analysis reveals up-regulation of rPP cycle and formate-associated genes alongside down-regulation of catabolic genes".

Specific comments:

1. Deleting pgi rather than using a point mutation would allow the authors to more rigorously test whether loss-off-function mutants are being selected for in their experimental evolution pipeline. The same argument applies to crp.

We appreciate this recommendation and indeed tried to delete pgi, but the genetic manipulation caused a knockout of other genes along with pgi (pepE, rluF, yjbD, lysC) so in the time available to us we cannot confidently determine whether the deletion alone is sufficient and can replace the mutation.

Regarding crp, we do not think there is a reason to believe the mutation is a loss-of-function. In any case, the proteomics-based characterization of the crp mutation is now included in the SI.

1. Page 10, lines 10-11, the authors state "Since Crp and RpoB are known to physically interact in the cell (26-28), we address them as one unit, as it is hard to decouple the effect of one from the other". CRP and RpoB are connected, but the authors' description of them is misleading. CRP activates transcription by interacting with RNA polymerase holoenzyme, of which the Beta subunit (encoded by rpoB) is a part. The specific interaction of CRP is with a different RNA polymerase subunit. The functions of CRP and RpoB, while both related to transcription, are otherwise very different. The mutations in crp and rpoB are unlikely to be directly functionally connected. Hence, they should be considered separately.

Indeed, the fact that the proteins are interacting in the cell does not necessarily mean that the mutations are functionally connected. We therefore added as further justification in the new section:

"As far as we know, the mutations in the Crp and RpoB genes affect the binding of the RNA polymerase complex to DNA and/or its transcription rates. Depending on the transcribed gene target, the effect of the two mutations might be additive, antagonistic, or synergistic. Since each one of these mutations individually (in combination with the pgi mutation) is not sufficient to achieve autotrophic growth, it is reasonable to assume that only the target genes whose levels of expression change significantly in the double-mutant are the ones relevant for the autotrophic phenotype”.

In our proteomics analysis we considered each mutation separately. We found that in some cases the two mutations together have an additive effect, but in other cases we found that the two mutations together affect differently on the proteome, compared to the effect of each mutation alone. Since both mutations are essential to the phenotype, we decided to go with the approach of addressing the two mutations as one unit for the physiological and metabolic experiments.

1. A Beta-galactosidase assay would provide a very simple test of CRP H22N activity. There are also simple in vivo and in vitro assays for transcription activation (two different modes of activation) and DNA-binding. H22 is not near the DNA-binding domain, but may impact overall protein structure.

The mutation is located in “Activating Region 2”, interacting with RNA polymerase. We tried an in-vivo assay to determine the CRP H22N activity and got inconclusive results, we believe the proteomics analysis serves as a good method for understanding the global effect of the mutation.

1. There are many high-resolution structures of both CRP and RpoB (in the context of RNA polymerase). The authors should compare the position of the sites of mutation of these proteins to known functional regions, assuming H22N is not a loss-of-function mutation in crp.

We added a supplementary figure regarding the structural location of the two mutations, where it is demonstrated that crp H22N is located in a region interacting with the RNA polymerase and rpoB A1245V is located in proximity to regions interacting with the DNA.

1. RNA-seq would provide a simple assay for the effects of the crp and rpoB mutations. While the precise effect of the rpoB mutation on RNA polymerase function may be hard to discern, the overall impact on gene expression would likely be informative.

Indeed we agree that an omics approach to infer the global effect of these mutations is beneficial, we opted to use a proteomics approach and think it serves the purpose of clarifying the final, down-stream, effect on the cell.

1. Page 2, lines 40-45, the authors should more clearly explain that the deletion of pfkA, pfkB and zwf was part of the experimental evolution strategy in their earlier work (Gleizer et al., 2019), and not a new strategy in the current study.

We thank you for pointing this out, and edited the text accordingly.

1. Page 3, line 27. Why did the authors compare the newly acquired mutants to only two mutants from the earlier work, not all 6?

The 6 clones that were isolated in Gleizer et al., had 2 distinct mutation profiles. During the isolation process the lineage split into two groups. Three out of the 6 clones (clones 1,2,6) came from the same ancestor, and the other three (clones 3,4,5) came from another ancestor. Hence, these two groups shared almost all of their mutations (see Venn diagram). We decided to use for our comparison the representative with the highest number of mutations from each group (clones 5 and 6).

Author response image 1.

Author Response

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

Reviewer #1:

Continuous attractor networks endowed with some sort of adaptation in the dynamics, whether that be through synaptic depression or firing rate adaptation, are fast becoming the leading candidate models to explain many aspects of hippocampal place cell dynamics, from hippocampal replay during immobility to theta sequences during run. Here, the authors show that a continuous attractor network endowed with spike frequency adaptation and subject to feedforward external inputs is able to account for several previously unaccounted aspects of theta sequences, including (1) sequences that move both forwards and backwards, (2) sequences that alternate between two arms of a T-maze, (3) speed modulation of place cell firing frequency, and (4) the persistence of phase information across hippocampal inactivations. I think the main result of the paper (findings (1) and (2)) are likely to be of interest to the hippocampal community, as well as to the wider community interested in mechanisms of neural sequences. In addition, the manuscript is generally well written, and the analytics are impressive. However, several issues should be addressed, which I outline below.

Major comments:

1. In real data, population firing rate is strongly modulated by theta (i.e., cells collectively prefer a certain phase of theta - see review paper Buzsaki, 2002) and largely oscillates at theta frequency during run. With respect to this cyclical firing rate, theta sweeps resemble "Nike" check marks, with the sweep backwards preceding the sweep forwards within each cycle before the activity is quenched at the end of the cycle. I am concerned that (1) the summed population firing rate of the model does not oscillate at theta frequency, and (2) as the authors state, the oscillatory tracking state must begin with a forward sweep. With regards to (1), can the authors show theta phase spike preference plots for the population to see if they match data? With regards to (2), can the authors show what happens if the bump is made to sweep backwards first, as it appears to do within each cycle?

Thank you for raising these two important points. As the reviewer mentioned, experimental data does show that the population activity (e.g., calculated from the multiunit activity of tetrode recording) is strongly modulated by theta. While we mainly focused on sweeps of bump position, the populational activity also shows cyclical firing at the theta frequency (we added Fig. S7 to reflect this). This is also reflected in Fig. 4d where the bump height (representing the overall activity) oscillates at individual theta cycles. The underlying mechanism of cyclical population activity is as follows: the bump height is determined by the amount of input the neuron received (which located at the center of the bump). While the activity bump sweeps away from the external input, the center neuron receives less input from the external input, and hence the bump height is smaller. Therefore, not only the position sweeps around the external input, also the populational activity sweeps accordingly at the same frequency.

For the “Nike” check marks: we first clarify that the reason for we observed a forward sweep preceding a backward sweep is that we always force the artificial animal runs from left to right on the track where we treated “right” as “forward”. At the beginning of simulation, the external input to the network moves towards right, and therefore the activity bump starts from a position behind the animals and sweeps towards right (forward). In general, this means that the bump will never do a backward sweep first in our model. However, this does not mean that the forward sweeps precede the backward sweeps in each theta cycle. Experimentally, to determine the “0” phase of theta cycles, the LFP signal in CA1 was first bandpass filtered and then Hilbert transformed to get the phase at each time point. Then, a phase histogram of multiunit activity in CA1 was calculated across locomotor periods; the phase of maximal CA1 firing on the histogram was then defined to be “0” phase. Since we didn’t model LFP oscillation in the attractor model, we cannot obtain a “0” phase reference like the experimental procedure. Instead, we define the “0” phase using the “population activity quenched time”, where phase “0” is defined as the minimum population activity during oscillation cycles, which happens when the activity bump is farthest from the animal position. In this way, we observed a “Nike” pattern where the activity bump begins with a backward sweep towards the external input and then followed up with a forward sweep. This was showed in Fig. 3b in the main text.

1. I could not find the width of the external input mentioned anywhere in the text or in the table of parameters. The implication is that it is unclear to me whether, during the oscillatory tracking state, the external input is large compared to the size of the bump, so that the bump lives within a window circumscribed by the external input and so bounces off the interior walls of the input during the oscillatory tracking phase, or whether the bump is continuously pulled back and forth by the external input, in which case it could be comparable to the size of the bump. My guess based on Fig 2c is that it is the latter. Please clarify and comment.

Thank you for your comment. We added the width of the external input to the text and table (see table 1). The bump is continuously pulled back and forth by the external input, as guessed by the reviewer. Experimentally, theta sweeps live roughly in the window of place field size. This is also true in our model, where theta sweep length depends on the strength of recurrent connections which determines the place field size. However, it also depends on the adaptation strength where large adaptation (more intrinsic mobility) leads to large sweep length. We presume that the reason for the reviewer had the guess that the bump may live within a window bounded by the external input is that we also set the width of external input comparable to the place field size (in fact, we don’t know how wide the external location input to the hippocampal circuits is in the biological brain, but it might be reasonable to set the external input width as comparable to the place field size, otherwise the location information conveyed to the hippocampus might be too dispersed). We added a plot in the SI (see Fig. S1) to show that when choosing a smaller external input width, but increasing the adaptation strength, the activity bump lives in a window exceeding the external input.

We clarified this point by adding the following text to line 159

“... It is noteworthy that the activity bump does not live within a window circumscribed by the external input bump (bouncing off the interior walls of the input during the oscillatory tracking state), but instead is continuously pulled back and forth by the external input (see Fig. S1)...”

1. I would argue that the "constant cycling" of theta sweeps down the arms of a T-maze was roughly predicted by Romani & Tsodyks, 2015, Figure 7. While their cycling spans several theta cycles, it nonetheless alternates by a similar mechanism, in that adaptation (in this case synaptic depression) prevents the subsequent sweep of activity from taking the same arm as the previous sweep. I believe the authors should cite this model in this context and consider the fact that both synaptic depression and spike frequency adaptation are both possible mechanisms for this phenomenon. But I certainly give the authors credit for showing how this constant cycling can occur across individual theta cycles.

Thank you for raising this point. We added the citation of Romani & Tsodyks’ model in the context (line 304). As the reviewer pointed out, STD can also act as a potential mechanism for this phenomenon. We also gave the Romani & Tsodyks’ model credit for showing how this “cycling spanning several theta cycles” can account for the phenomenon of slow (~1Hz) and deliberative behaviors, namely, head scanning (Johson and Redish, 2007). We commented this in line 302

“... As the external input approaches the choice point, the network bump starts to sweep onto left and right arms alternatively in successive theta cycles (Fig. 5b and video 4; see also Romani and Tsodyks (2015) for a similar model of cyclical sweeps spanning several theta cycles) ...”

1. The authors make an unsubstantiated claim in the paragraph beginning with line 413 that the Tsodyks and Romani (2015) model could not account for forwards and backwards sweeps. Both the firing rate adaptation and synaptic depression are symmetry breaking models that should in theory be able to push sweeps of activity in both directions, so it is far from obvious to me that both forward and backward sweeps are not possible in the Tsodyks and Romani model. The authors should either prove that this is the case (with theory or simulation) or excise this statement from the manuscript.

Thank you for your comment. Our claim about the Tsodyks and Romani (2015) model's inability to account for both forward and backward sweeps was inappropriate. We made this claim based on our own implementation of the Tsodyks and Romani (2015) model and didn’t find a parameter region where the bump oscillation shows both forward and backward sweeps. It might be due to the limited parameter range we searched from. Additionally, we also note some difference in these two models, where the Romani & Tsodyks’ model has an external theta input to the attractor network which prevent the bump to move further. This termination may also prevent the activity bump to move backward as well. We didn’t consider external theta input in our model, and the bump oscillation is based on internal dynamics. We have deleted that claim from line 424 in the revised paper, and revised that portion of the manuscript by adding the following text to line 424:

“…Different from these two models, our model considers firing rate adaptation to implement symmetry breaking and hence generates activity propagation. To prevent the activity bump from spreading away, their model considers an external theta input to reset the bump location at the end of each theta cycle, whereas our model generates an internal oscillatory state, where the activity bump travels back due to the attraction of external location input once it spreads too far away. Moreover, theoretical analysis of our model reveals how the adaptation strength affect the direction of theta sweeps, as well as offers a more detailed understanding of theta cycling in complex environments…”

1. The section on the speed dependence of theta (starting with line 327) was very hard to understand. Can the authors show a more graphical explanation of the phenomenon? Perhaps a version of Fig 2f for slow and fast speeds, and point out that cells in the latter case fire with higher frequency than in the former?

Thank you for raising this valuable point. There are two different frequencies showed in Fig. 6 a,c &d. One is the bump oscillation frequency, the other is the firing frequency of single cell. To help understanding, we included experimental results (from Geisler et al, 2007) in Fig. 6a. It showed that when the animal increases its running speed, the LFP theta only increases a bit (compare the blue curve and the green curve), while the single cell firing rate oscillation frequency increases more. In our model, we first demonstrated this result using unimodal cells which have only significant phase precession (Fig. 6c). While the animal runs through the firing field of a place cell, the firing phase will always precess for half a cycle in total. Therefore, faster running speed means that the half cycle will be accomplished faster, and hence single cell oscillation frequency will be higher. We also predicted the results on bimodal cells (Fig. 6d). To make this point clearer, we modified Fig. 6 by including experimental results, and rewrote the paragraph as follows (line 337):

“…As we see from Fig. 3d and Fig. 4a&b, when the animal runs through the firing field of a place cell, its firing rate oscillates, since the activity bump sweeps around the firing field center of the cell. Therefore, the firing frequency of a place cell has a baseline theta frequency, which is the same as the bump oscillation frequency. Furthermore, due to phase precession, there will be a half cycle more than the baseline theta cycles as the animal runs over the firing field, and hence single cell oscillatory frequency will be higher than the baseline theta frequency (Fig. 6c). The faster the animal runs, the faster the extra half cycle is accomplished. Consequently, the firing frequency of single cells will increase more (a steeper slope in Fig. 6c red dots) than the baseline frequency.…”

1. I had a hard time understanding how the Zugaro et al., (2005) hippocampal inactivation experiment was accounted for by the model. My intuition is that while the bump position is determined partially by the location of the external input, it is also determined by the immediate history of the bump dynamics as computed via the local dynamics within the hippocampus (recurrent dynamics and spike rate adaptation). So that if the hippocampus is inactivated for an arbitrary length of time, there is nothing to keep track of where the bump should be when the activity comes back online. Can the authors please explain more how the model accounts for this?

Thank you for the comments. The easiest way to understand how the model account for the experimental result from Zugaro et al., (2005) is from Eq. 8:

This equation says that the firing phase of a place cell is determined by the time the animal traveled through the place field, i.e., the location of the animal in the place field (with d0,c0 and vext all constant, and tf the only variable). No matter how long the hippocampus is inactivated (for an arbitrary length of time), once the external input is on, the new phase will continue from the new location of the animal in the place field. In other words, the peak firing phase keeps tracking the location of the animal. To make this point clearer, we modified Fig. 6 by including experimental results from Zugaro et al., (2005), and updated the description from line 356:

“…Based on the theoretical analysis (Eq. 8), we see that the firing phase is determined by the location of the animal in the place field, i.e., vext tf. This means that the firing phase keeps tracking the animal's physical location. No matter how long the network is inactivated, the new firing phase will only be determined by the new location of the animal in the place field. Therefore, the firing phase in the first bump oscillation cycle after the network perturbation is more advanced than the firing phase in the last bump oscillation cycle right before the perturbation, and the amount of precession is similar to that in the case without perturbation (Fig. 6e) …”

1. Can the authors comment on why the sweep lengths oscillate in the bottom panel of Fig 5b during starting at time 0.5 seconds before crossing the choice point of the T-maze? Is this oscillation in sweep length another prediction of the model? If so, it should definitely be remarked upon and included in the discussion section.

We appreciate the reviewer’s valuable attention of this phenomenon. We thought it was a simulation artifact due to the parameter setting. However, we found that this phenomenon is quite robust to different parameter settings. While we haven’t found a theoretical explanation, we provide a qualitative explanation for it: this length oscillation frequency may be coupled with the time constant of the firing rate adaptation. Specifically, for a longer sweep, the neurons at the end of the sweep are adapted (inhibited), and hence the activity bump cannot travel that long in the next round. Therefore, the sweep length is shorter compared to the previous one. In the next round, the bump will sweep longer again because those neurons have recovered from the previous adaptation effect. We think this length oscillation is quite interesting and will check that in the experimental data in future works. We added this point in the main text as a prediction in line 321:

“…We also note that there is a cyclical effect in the sweep lengths across oscillation cycles before the animal enters the left or right arm (see Fig. 5b lower panel), which may be interesting to check in the experimental data in future work (see Discussion for more details) …”

And line 466:

“…Our model of the T-maze environment showed an expected phenomenon that as the animal runs towards the decision point, the theta sweep length also shows cyclical patterns (Fig. 5b lower panel). An intuitive explanation is that, due to the slow dynamics in firing rate adaptation (with a large time constant compared to neural firing), a long sweep leads to an adaptation effect on the neurons at the end of the sweep path. Consequently, the activity bump cannot travel as far due to the adaptation effect on those neurons, resulting in a shorter sweep length compared to the previous one. In the next round, the activity bump exhibits a longer sweep again because those neurons have recovered from the previous adaptation effect. We plan to test this phenomenon in future experiments...”

1. Perhaps I missed this, but I'm curious whether the authors have considered what factors might modulate the adaptation strength. In particular, might rat speed modulate adaptation strength? If so, would have interesting predictions for theta sequences at low vs high speeds.

Thank you for raising up this important point. As we pointed out in line 279: “…the experimental data (Fernandez et al, 2017) has indicated that there is a laminar difference between unimodal cells and bimodal cells, with bimodal cells correlating more with the firing patterns of deep CA1 neurons and unimodal cells with the firing patterns of superficial CA1 neurons. Our model suggests that this difference may come from the different adaptation strengths in the two layers…”. Our guess is that the adaptation strength might reflect some physiological differences of place cells in difference pyramidal layers in the hippocampus. For example, place cells in superficial layer and deep layer receive different amount of input from MEC and sensory cortex, and such difference may contribute to a different effect of adaptation of the two populations of place cells.

Our intuition is that animal’s running speed may not directly modulate the adaptation strength. Note that the effect of adaptation and adaptation strength are different. As the animal rapidly runs across the firing field, the place cell experiences a dense firing (in time), therefore the adaptation effect is large; as the animal slowly runs across the field, the place cell experiences sparse firing (in time), and hence the adaptation effect is small. In these two situations, the adaption strength is fixed, but the difference is due to the spike intervals.

From Eq. 45-47, our theoretical analysis shows several predictions of theta sequences regarding to the parameters in the network. For example, how the sweep length varies when the running speed changes in the network. We simulated the network in both low running speed and high running speed (while kept all other parameters fixed), and found that the sweep length at low speed is larger than that at high speed. This is different from previously data, where they showed that the sweep length increases as the animal runs faster (Maurer et al, 2012). However, we are not sure how other parameters are changed in the biological brain as the animal runs faster, e.g., the external input strength and the place field width might also vary as confounds. We will explore this more in the future and investigate how the adaptation strength is modulated in the brain.

1. I think the paper has a number of predictions that would be especially interesting to experimentalists but are sort of scattered throughout the manuscript. It would be beneficial to have them listed more prominently in a separate section in the discussion. This should include (1) a prediction that the bump height in the forward direction should be higher than in the backward direction, (2) predictions about bimodal and unimodal cells starting with line 366, (3) prediction of another possible kind of theta cycling, this time in the form of sweep length (see comment above), etc.

Thank you for pointing this out. We updated the manuscript by including a paragraph in Discussion summarizing the prediction we made throughout the manuscript (from line 459):

‘’…Our model has several predictions which can be tested in future experiments. For instance, the height of the activity bump in the forward sweep window is higher than that in the backward sweep window (Fig. 4c) due to the asymmetric suppression effect from the adaptation. For bimodal cells, they will have two peaks in their firing frequency as the animal runs across the firing fields, with one corresponding to phase precession and the other corresponding to phase procession. Similar to unimodal cells, both the phase precession and procession of a bimodal cell after transient intrahippocampal perturbation will continue from the new location of the animal (Fig. S5). Interestingly, our model of the T-maze environment showed an expected phenomenon that as the animal runs towards the decision point, the theta sweep length also shows cyclical patterns (Fig. 5b lower panel). An intuitive explanation is that, due to the slow dynamics in firing rate adaptation (with a large time constant compared to neural firing), a long sweep leads to an adaptation effect on the neurons at the end of the sweep path. Consequently, the activity bump cannot travel as far due to the adaptation effect on those neurons, resulting in a shorter sweep length compared to the previous one. In the next round, the activity bump exhibits a longer sweep again because those neurons have recovered from the previous adaptation effect. We plan to test this phenomenon in future experiments…’

Reviewer #2:

In this work, the authors elaborate on an analytically tractable, continuous-attractor model to study an idealized neural network with realistic spiking phase precession/procession. The key ingredient of this analysis is the inclusion of a mechanism for slow firing-rate adaptation in addition to the otherwise fast continuous-attractor dynamics. The latter which continuous-attractor dynamics classically arises from a combination of translation invariance and nonlinear rate normalization. For strong adaptation/weak external input, the network naturally exhibits an internally generated, travelling-wave dynamics along the attractor with some characteristic speed. For small adaptation/strong external stimulus, the network recovers the classical externally driven continuous-attractor dynamics. Crucially, when both adaptation and external input are moderate, there is a competition with the internally generated and externally generated mechanism leading to oscillatory tracking regime. In this tracking regime, the population firing profile oscillates around the neural field tracking the position of the stimulus. The authors demonstrate by a combination of analytical and computational arguments that oscillatory tracking corresponds to realistic phase precession/procession. In particular the authors can account for the emergence of a unimodal and bimodal cells, as well as some other experimental observations with respect the dependence of phase precession/procession on the animal's locomotion. The strengths of this work are at least three-fold: 1) Given its simplicity, the proposed model has a surprisingly large explanatory power of the various experimental observations. 2) The mechanism responsible for the emergence of precession/procession can be understood as a simple yet rather illuminating competition between internally driven and externally driven dynamical trends. 3) Amazingly, and under some adequate simplifying assumptions, a great deal of analysis can be treated exactly, which allows for a detailed understanding of all parametric dependencies. This exact treatment culminates with a full characterization of the phase space of the network dynamics, as well as the computation of various quantities of interest, including characteristic speeds and oscillating frequencies.

1. As mentioned by the authors themselves, the main limitation of this work is that it deals with a very idealized model and it remains to see how the proposed dynamical behaviors would persist in more realistic models. For example, the model is based on a continuous attractor model that assumes perfect translation-invariance of the network connectivity pattern. Would the oscillating tracking behavior persist in the presence of connection heterogeneities?

Thank you for raising up this important point. Continuous attractor models have been widely used in modeling hippocampal neural circuits (see McNaughton et al, 2006 for a review), and researchers often assumed that there is a translation-invariance structure in these network models. The theta sweep state we presented in the current work is based on the property of the continuous attractor state. We do agree with the reviewer that the place cell circuit might not be a perfect continuous attractor network. For a simpler case where the connection weights are sampled from a Gaussian distribution around J_0, the theta sweep state still exhibit in the network (see Fig. S8 for an example). We also believe that the model can be extended to more complex cases where there exist over-representations of the “home” location and decision points in the real environment, i.e., the heterogeneity is not random, but has stronger connections near those locations, then the theta sweeps will be more biased to those location. However, if the heterogeneity breaks the continuous attractor state, the theta sweep state may not be presented in the network.

1. Can the oscillating tracking behavior be observed in purely spiking models as opposed to rate models as considered in this work?

Thank you for pointing this out. The short answer is yes. If the translation-invariance of the network connectivity pattern hold in the network, i.e., the spiking network is still a continuous attractor network (see the work from Tsodyks et al, 1996; and from Yu et al. "Spiking continuous attractor neural networks with spike frequency adaptation for anticipative tracking"), then the adaptation, which has the mathematical form of spike frequency adaptation (instead of firing rate adaptation), will still generate sweep state of the activity bump. We here chose the rate-based model because it is analytically tractable, which gives us a better understanding of the underlying dynamics. Many of the continuous attractor model related to spatial tuning cell populations are rate-based (see examples Zhang 1996; Burak & Fiete 2009). However, extending to spike-based model would be straightforward.

1. Another important limitation is that the system needs to be tuned to exhibit oscillation within the theta range and that this tuning involves a priori variable parameters such as the external input strength. Is the oscillating-tracking behavior overtly sensitive to input strength variations?

Thank you for pointing this out. In rodent studies, theta sequences are thought to result from the integration of both external inputs conveying sensory-motor information, and intrinsic network dynamics possibly related to memory processes (see Drieu and Zugaro 2019; Drieu at al, 2018). We clarified here that, in our modeling work, the generation of theta sweeps also depends on both the external input and the intrinsic dynamics (induced by the firing rate adaptation). Therefore, we don’t think the dependence of theta sweeps on the prior parameter – the external input strength – is a limitation here. We agreed with the reviewer that the system needs to be tuned to exhibit oscillation within the theta range. However, the parameter range of inducing oscillatory state is relatively large (see Fig. 2g in the main text). It will be interesting to investigate (and find experimental evidence) how the biological system adjusts the network configuration to implement the sweep state in network dynamics.

1. The author mentioned that an external pacemaker can serve to drive oscillation within the desired theta band but there is no evidence presented supporting this.

Thank you for pointing this out. We made this argument based on our initial simulation before but didn’t go into the details of that. We have deleted that argument in the discussion and rewrote that part. We will carry out more simulations in the future to verify if this is true. See our changes from line 418 to line 431:

“... A representative model relying on neuronal recurrent interactions is the activation spreading model. This model produces phase precession via the propagation of neural activity along the movement direction, which relies on asymmetric synaptic connections. A later version of this model considers short-term synaptic plasticity (short-term depression) to implicitly implement asymmetric connections between place cells, and reproduces many other interesting phenomena, such as phase precession in different environments. Different from these two models, our model considers firing rate adaptation to implement symmetry breaking and hence generates activity propagation. To prevent the activity bump from spreading away, their model considers an external theta input to reset the bump location at the end of each theta cycle, whereas our model generates an internal oscillatory state, where the activity bump travels back due to the attraction of external location input once it spreads too far away. Moreover, theoretical analysis of our model reveals how the adaptation strength affect the direction of theta sweeps, as well as offers a more detailed understanding of theta cycling in complex environments...”

1. A final and perhaps secondary limitation has to do with the choice of parameter, namely the time constant of neural firing which is chosen around 3ms. This seems rather short given that the fast time scale of rate models (excluding synaptic processes) is usually given by the membrane time constant, which is typically about 15ms. I suspect this latter point can easily be addressed.

Thank you for pointing this out. The time constant we currently chose is relatively short as used in other studies. We conducted additional simulation by adjusting the time constant to 10ms, and the results reported in this paper remain consistent. Please refer to Fig S9 for the results obtained with a time constant of 10 ms.

Reviewer #3:

With a soft-spoken, matter-of-fact attitude and almost unwittingly, this brilliant study chisels away one of the pillars of hippocampal neuroscience: the special role(s) ascribed to theta oscillations. These oscillations are salient during specific behaviors in rodents but are often taken to be part of the intimate endowment of the hippocampus across all mammalian species, and to be a fundamental ingredient of its computations. The gradual anticipation or precession of the spikes of a cell as it traverses its place field, relative to the theta phase, is seen as enabling the prediction of the future - the short-term future position of the animal at least, possibly the future in a wider cognitive sense as well, in particular with humans. The present study shows that, under suitable conditions, place cell population activity "sweeps" to encode future positions, and sometimes past ones as well, even in the absence of theta, as a result of the interplay between firing rate adaptation and precise place coding in the afferent inputs, which tracks the real position of the animal. The core strength of the paper is the clarity afforded by the simple, elegant model. It allows the derivation (in a certain limit) of an analytical formula for the frequency of the sweeps, as a function of the various model parameters, such as the time constants for neuronal integration and for firing rate adaptation. The sweep frequency turns out to be inversely proportional to their geometric average. The authors note that, if theta oscillations are added to the model, they can entrain the sweeps, which thus may superficially appear to have been generated by the oscillations.

1. The main weakness of the study is the other side of the simplicity coin. In its simple and neat formulation, the model envisages stereotyped single unit behavior regulated by a few parameters, like the two time constants above, or the "adaptation strength", the "width of the field" or the "input strength", which are all assumed to be constant across cells. In reality, not only assigning homogeneous values to those parameters seems implausible, but also describing e.g. adaptation with the simple equation included in the model may be an oversimplification. Therefore, it remains important to understand to what extent the mechanism envisaged in the model is robust to variability in the parameters or to eg less carefully tuned afferent inputs.

Thank you for pointing out this important question. As the reviewer pointed out, there is an oversimplification in our model compared to the real hippocampal circuits (also see Q1 and Q3 from reviewer2). We also pointed out that in the main text line 504:

“…Nevertheless, it is important to note that the CANN we adopt in the current study is an idealized model for the place cell population, where many biological details are missed. For instance, we have assumed that neuronal synaptic connections are translation-invariant in the space...”

To investigate model robustness to parameter setting, we divided all the parameters into two groups. The first group of parameters determines the bump state, i.e., width of the field a, neuronal density ρ, global inhibition strength k, and connection strength J_0. The second group of parameters determines the bump sweep state (which based on the existence of the bump state), i.e., the input strength α and the adaptation strength m. For the first group of parameters, we refer the reviewer to the Method part: stability analysis of the bump state. This analysis tells us the condition when the continuous attractor state holds in the network (see Eq. 20, which guides us to perform parameter selection). For the second group of parameters, we refer the reviewer to Fig. 2g, which tells us when the bump sweep state occurs regarding to input strength and adaptation strength. When the input strength is small, the range of adaptation strength is also small (to get the bump sweep state). However, as the input strength increases, we can see from Fig. 2g that the range of adaptation strength (to get the bump sweep state) also linearly increases. Although there exists other two state in the network when the two parameters are set out of the colored area in Fig. 2g, the parameter range of getting sweep state is also large, especially when the input strength value is large, which is usually the case when the animal actively runs in the environment.

To demonstrate how the variability affect the results, we added variability to the connection weights by sampling the connection weights from a Gaussian distribution around J_0 (this introduces heterogeneity in the connection structure). We found that the bump sweep state still holds in this condition (see Fig. S8 as well as Q1 from reviewer2). For the variability in other parameter values, the results will be similar. Although adding variability to these parameters will not bring us difficulty in numerical simulation, it will make the theoretical analysis much more difficult.

1. The weak adaptation regime, when firing rate adaptation effectively moves the position encoded by population activity slightly ahead of the animal, is not novel - I discussed it, among others, in trying to understand the significance of the CA3-CA1 differentiation (2004). What is novel here, as far as I know, is the strong adaptation regime, when the adaptation strength m is at least larger than the ratio of time constants. Then population activity literally runs away, ahead of the animal, and oscillations set in, independent of any oscillatory inputs. Can this really occur in physiological conditions? A careful comparison with available experimental measures would greatly strengthen the significance of this study.

Thank you for raising up this interesting question.

Re: “…firing rate adaptation effectively moves the position encoded by population activity slightly ahead of the animal, is not novel…”, We added Treves, A (2004) as a citation when we introduce the firing rate adaptation in line 116

To test if the case of “…the adaptation strength m is at least larger than the ratio of time constants…” could occur in physiological conditions, it requires a measure of the adaptation strength as well as the time constant of both neuron firing and adaptation effect. The most straightforward way would be in vivo patch clamp recording of hippocampal pyramidal neurons when the animal is navigating an environment. This will give us a direct measure of all these values. However, we don’t have these data to verify this hypothesis yet. Another possible way of measure these values is through a state-space model. Specifically, we can build a state space model (considering adaptation effect in spike release) by taking animal’s position as latent dynamics, and recorded spikes as observation, then infer the parameters such as adaptation strength and time constant in the slow dynamics. Previous work of state-space models (without firing rate adaptation) in analyzing theta sweeps and replay dynamics have been explored by Denovellis et al. (2021), as well as Krause and Drugowitsch (2022). We think it might be doable to infer the adaptation strength and adaptation time constant in a similar paradigm in future work. We thank the reviewer for pointing out that and hope our replies have clarified the concerns of the reviewer.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

The authors focused on genetic variability in relation to insulin resistance. They used genetically different lines of mice and exposed them to the same diet. They found that genetic predisposition impacts the overall outcome of metabolic disturbances. This work provides a fundamental novel view on the role of genetics and insulin resistance.

Reviewer #2 (Public Review):

Summary:

In the present study, van Gerwen et al. perform deep phosphoproteomics on muscle from saline or insulin-injected mice from 5 distinct strains fed a chow or HF/HS diet. The authors follow these data by defining a variety of intriguing genetic, dietary, or gene-by-diet phosphor-sites that respond to insulin accomplished through the application of correlation analyses, linear mixed models, and a module-based approach (WGCNA). These findings are supported by validation experiments by intersecting results with a previous profile of insulin-responsive sites (Humphrey et al, 2013) and importantly, mechanistic validation of Pfkfb3 where overexpression in L6 myotubes was sufficient to alter fatty acid-induced impairments in insulin-stimulated glucose uptake. To my knowledge, this resource provides the most comprehensive quantification of muscle phospho-proteins which occur as a result of diet in strains of mice where genetic and dietary effects can be quantifiably attributed in an accurate manner. Utilization of this resource is strongly supported by the analyses provided highlighting the complexity of insulin signaling in muscle, exemplified by contrasts to the "classically-used" C57BL6/J strain. As it stands, I view this exceptional resource as comprehensive with compelling strength of evidence behind the mechanism explored. Therefore, most of my comments stem from curiosity about pathways within this resource, many of which are likely well beyond the scope of incorporation in the current manuscript. These include the integration of previous studies investigating these strains for changes in transcriptional or proteomic profiles and intersections with available human phospho-protein data, many of which have been generated by this group.

Strengths:

Generation of a novel resource to explore genetic and dietary interactions influencing the phospho-proteome in muscle. This is accompanied by the elegant application of in silico tools to highlight the utility.

Weaknesses:

Some specific aspects of integration with other data among the same fixed strains could be strengthened and/or discussed.

Reviewer #3 (Public Review):

Summary:

The authors aimed to investigate how genetic and environmental factors influence the muscle insulin signaling network and its impact on metabolism. They utilized mass spectrometry-based phosphoproteomics to quantify phosphosites in the skeletal muscle of genetically distinct mouse strains in different dietary environments, with and without insulin stimulation. The results showed that genetic background and diet both affected insulin signaling, with almost half of the insulin-regulated phosphoproteome being modified by genetic background on an ordinary diet, and high-fat high-sugar feeding affecting insulin signaling in a strain-dependent manner.

Strengths:

The study uses state-of-the-art phosphoproteomics workflow allowing quantification of a large number of phosphosites in skeletal muscle, providing a comprehensive view of the muscle insulin signaling network. The study examined five genetically distinct mouse strains in two dietary environments, allowing for the investigation of the impact of genetic and environmental factors on insulin signaling. The identification of coregulated subnetworks within the insulin signaling pathway expanded our understanding of its organization and provided insights into potential regulatory mechanisms. The study associated diverse signaling responses with insulin-stimulated glucose uptake, uncovering regulators of muscle insulin responsiveness.

Weaknesses:

Different mouse strains have huge differences in body weight on normal and high-fat high-sugar diets, which makes comparison between the models challenging. The proteome of muscle across different strains is bound to be different but the changes in protein abundance on phosphosite changes were not assessed. Authors do get around this by calculating 'insulin response' because short insulin treatment should not affect protein abundance. The limitations acknowledged by the authors, such as the need for larger cohorts and the inclusion of female mice, suggest that further research is needed to validate and expand upon the findings.

Reviewer #1 (Recommendations For The Authors):

I would suggest further discussion of the potential differences between males and females of the various strains.

In the revised manuscript we have included a more detailed discussion of the potential differences between male and female mice in the "Limitations of this study" section on lines 455-459. In particular, a landmark study of HFD-fed inbred mouse strains found that insulin sensitivity, as inferred from the proxy HOMA-IR, was affected by interactions between sex and strain despite generally being greater in female mice (10.1016/j.cmet.2015.01.002). Furthermore, a recent phosphoproteomics study of human induced pluripotent stem-cell derived myoblasts identified groups of insulin-regulated phosphosites affected by donor sex, and by interactions between sex and donor insulin sensitivity (10.1172/JCI151818). Based on these results, we anticipate that both soleus insulin sensitivity and phoshoproteomic insulin responses would differ between male and female mice through interactions with strain and diet, adding yet another layer of complexity to what we observed in this study. This will be an important avenue for future research to explore.

Reviewer #2 (Recommendations For The Authors):

The following are comments to authors - many, if not all are suggestions for extended discussion and beyond the scope of the current elegant study.

In the discussion section (line 428) the authors make a key point in that the genetic, dietary, and interacting patterns of variation of Phospho-sites could be due to changes in total protein and/or transcript levels across strains. For example, given the increased expression of Pfkfb3 was sufficient to impact glucose uptake, suggesting that the transcript levels of the gene might also show a similar correlation with insulin responsiveness as in Fig 6b. Undoubtedly, phospho-proteomics analyses will provide unique information on top of more classical omics layers and uncover what would be an important future direction. Therefore, I would suggest adding to the discussion some guidance on performing similar applications to datasets from, at least some, of the strains used where RNA-seq and proteomics are available.

We thank the reviewer for this suggestion. To address this, we mined recently published total proteomics data collected from soleus muscles of seven CHOW or HFD-fed inbred mouse strains, three of which were in common with our study (C57Bl6J, BXH9, BXD34; 10.1016/j.cmet.2021.12.013). In this study ex vivo soleus glucose uptake was measured and correlation analysis was performed, so we directly extracted the resulting glucose uptake-protein associations and compared them to the glucose uptake-phosphoprotein associations identified in our study. Indeed, we found that only a minority of proteins correlated at both the phosphosite and total protein levels, highlighting the utility of phosphoproteomics to provide orthogonal information to more classical omics layers. We have included this analysis in lines 303-311.

Relevant to this, the authors might want to consider depositing scripts to analyze some aspects of the data (ex. WGCNA on P-protein data or insulin-regulated anova) in a repository such as github so that these can be applied easily to other datasets.

We refer the reviewer to the section "Code availability" on lines 511-513, where we deposited all code used to analyse the data on github.

In contrast to the points above, I feel that the short time-course of insulin stimulation was one important aspect of the experimental design that was not emphasized enough as a strength. It was mentioned as a limitation in that other time points could provide more info, yes. But given that the total abundance of proteins and transcripts likely doesn't shift tremendously in this time frame, this provides an important appeal to the analysis of phosphor-proteomic data. I would suggest highlighting the insulin-stimulated response analysis here as something that leverages the unique nature of phosphoproteomics.

We are grateful for the reviewer's positivity regarding this aspect of our experimental design. We have reiterated the value of the 10min insulin stimulation - that it temporally segregates phosphoproteomic and total proteomic changes - in the "Limitations of this study" section on lines 477-481.

While I recognize the WGCNA analysis as an instrumental way to highlight global patterns of phospo-peptide abundance co-regulation, the analysis currently seems somewhat underdeveloped. For example, Fig 5f-h shows a lot of overlap between kinase substrates and pathways among modules. Clearly, there are informative differences based on the intersection with Humphries 2013 and the correlation with Pfkbp3. To highlight the specific membership of these modules, most people rank-order module members by correlation with eigen-gene (or P-peptide) and then perform pathway enrichments on these. Alternatively, it looks like all data was used to generate modules across conditions. One consideration would be to perform WGCNA on relevant comparison data separately (ex. chow mice only and HFHS only) and then compare modules whose membership is retained or shift between the two. Or even look at module representation for genes that show large correlations with insulin-responsiveness. This might also be a good opportunity to suggest readers intersect module members with muscle eQTLs which colocalize to glucose or insulin to prioritize some potential key drivers.

We thank the reviewer for their helpful suggestions, which we feel have substantially improved the WGCNA analysis. To probe specific functional differences between subnetworks, we performed rank-based enrichment using phosphopeptide module membership scores. Interestingly, this did reveal pathways that were enriched only in certain modules. However, we found that after p-value adjustment, virtually all enriched pathways lost statistical significance, hence we interpret these results as suggestive only. We have made this analysis available to readers in Fig S4b-d and lines 263-265: "To further probe functional differences we analysed phosphopeptide subnetwork membership scores, which revealed additional pathways enriched in individual subnetworks. However, these results were not significant after p-value adjustment and hence are suggestive only (Fig. S4b-d)". We also visualised module representation for glucose-uptake correlated phosphopeptides. This agreed with our existing analyis in Fig. 6f, where the eigenpeptides of modules V and I were correlated with glucose uptake (Fig. 6f). We have incorporated this new analysis in Fig. S6b-c and lines 324-325: "Examining the subnetwork membership scores for glucose-uptake correlated phosphopeptides also revealed a preference for clusters V and I, supporting this analysis (Fig. S6b-c)." Finally, in the discussion we have presented the integration of genetic data, such as muscle-specific eQTLs, as a future direction (lines 398-401): "Alternatively, one could overlap subnetworks with genetic information, such as genes associated with glucose homeostasis and other metabolic traits in human GWAS studies, or muscle-specific eQTLs or pQTLs genetically colocalised with similar traits, to further prioritise subnetwork-associated phenotypes and identify potential drivers within subnetworks."

Have the authors considered using their heritability and GxE estimated for module eigenpeptides? To my knowledge, this has never been performed and might provide some informative information as the co-regulated P-protein structure occurs as a result of relevant contexts.

In the revised manuscript we have now analysed eigenpeptides with the same statistical tests used to identify Strain and Diet effects in insulin-regulated phosphopeptides. We have displayed the statistical results in Fig. S4a, and have explicitly mentioned examples of StrainxDiet effects on lines 245-247: "For example, HFD-feeding attenuated the insulin response of subnetwork I in CAST and C57Bl6J strains (t-test adjusted p = 0.0256, 0.0365), while subnetwork II was affected by HFD-feeding only in CAST and NOD (Fig. 5e, Fig. S4a, t-test adjusted p = 0.00258, 0.0256)."

The integration of modules with adipocyte phosphoproteomic data from the authors 2013 Cell metab paper seems like an important way to highlight the integration of this resource to define critical cellular signaling mechanisms. To assess the conservation of signaling mechanisms and relationships to additional key contexts (ex. exercise), the intersection of the insulin-stimulated P-peptides with human datasets generated by this group (ex. cell metab 2015, nature biotech 2022) seems like an obvious future direction to prioritize targets. Figure S3B shows a starting point for these types of integrations.

To demonstrate the value of integrating our results with related phosphoproteomics data, we have incorporated the reviewer's advice of comparing insulin-regulated phosphosites to exercise-regulated phosphosites from Needham et. Nature Biotech 2022 and Hoffman et al. Cell Metabolism 2015. We identified a small subset of commonly regulated phosphosites (8 across all three studies). Given insulin and exercise both promote GLUT4 translocation, these sites may represent conserved regulatory mechanisms. This analysis is presented in Fig. S3d, Table S2, and lines 129-135: "In addition to insulin, exercise also promotes GLUT4 translocation in skeletal muscle. We identified a small subset of phosphosites regulated by insulin in this study that were also regulated by exercise in two separate human phosphoproteomics studies (Fig. S3d, Table S2, phosphosites: Eef2 T57 and T59, Mff S129 and S131, Larp1 S498, Tbc1d4 S324, Svil S300, Gys1 S645), providing a starting point for exploring conserved signalling regulators of GLUT4 translocation."

For the Pfkfb3 overexpression system, are there specific P-peptides that are increased/decreased upon insulin stimulation? This might be an interesting future direction to mention in order to link signaling mechanisms.

We assessed whether canonical insulin signalling was affected by Pfkfb3 overexpression by immunoblotting. Insulin-stimulated phosphorylation of Akt S473, Akt T308, Gsk3a/b S21/S9, and PRAS40 T246 differed little across conditions, with only a weak, statistically insignificant trend towards increased pT308 Akt, pS21/S9 Gsk3a/b, and pT246 PRAS40 in palmitate-treated Pfkfb3-overexpressing cells. Hence, as the reviewer has suggested, an interesting future direction will be to perform phosphoproteomics to characterise more deeply the effects of palmitate and Pfkfb3 overexpression on insulin signalling. We have modified the manuscript to reflect these findings and suggested future directions on lines 362-365: "immunoblotting of canonical insulin-responsive phosphosites on Akt and its substrates GSK3α/β and PRAS40 revealed minimal effect of palmitate treatment and Pfkfb3 overexpression (Fig. S7e-f), hence more detailed phosphoproteomics studies are needed to clarify whether Pfkfb3 overexpression restored insulin action by modulating insulin signalling."

Reviewer #3 (Recommendations For The Authors):

This remarkable contribution by the esteemed research group has significantly enriched the field of metabolism. The extensive dataset, intertwined with a sophisticated research design, promises to serve as an invaluable resource for the scientific community. I offer a series of suggestions aimed at potentially elevating the manuscript to an even higher standard.

Mouse Weight Variation and Correlation Analysis: The pronounced variances in mouse body weights pose a challenge to meaningful comparisons (Fig S1). Could the disparities in the phosphoproteome between basal and insulin-stimulated conditions be attributed to differences in body weight? Consider performing a correlation analysis. Furthermore, does the phosphoproteome of these mouse strains evolve comparably over time? Do these mice age similarly? Kindly incorporate this information.

We thank the reviewer for the suggested analysis. We found there was a significant correlation between the phosphopeptide insulin response and mouse body weight, either in CHOW-fed mice (Strain effects) or across both diets (Diet effects), for ~ 25% of phosphopeptides that exhibited a Strain or Diet effect. Hence, while there is a clear effect of body weight on insulin signalling, this influences only a small proportion of the entire insulin-responsive phosphoproteome. Notably, insulin was dosed according to mouse lean mass to ensure equivalent dosage received by the soleus muscle, hence any insulin signalling differences associated with body weight are unlikely due to differences in dosing. As the reviewer also alludes to, different strains could have different lifespans. This may result in mice having different biological ages at the time of experimentation, and this in turn could influence insulin signalling. This possibility is challenging to assess in a quantitative manner because lifespan data is not available for most strains used. However, it is worth noting that female CAST mice live 77% as long as C57Bl6J mice (median age of 671 vs 866 (10.1073/pnas.1121113109); data is not available for male mice nor the other three strains), and substantial differences in insulin signalling were observed between these two strains. Ultimately, regardless of whether body weight and/or lifespan altered insulin signalling, such differences would still have arisen solely from the distinct genetic backgrounds and diets of the mice, hence we believe they are meaningful results that should not be dismissed. We have added this analysis to the revised manuscript in the "Limitations of this study" section on lines 471-477: "We were also unable to determine the extent to which signalling changes arose from muscle-intrinsic or extrinsic factors. For instance, body weight varied substantially across mice and correlated significantly with 25% of Strain and Diet-affected phosphopeptides (Fig. S8c), suggesting obesity-related systemic factors likely impact a subset of the muscle insulin signalling network. Furthermore, genetic differences in lifespan could alter the “biological age” of different strains and their phosphoproteomes, though we could not assess this possibility since lifespan data are not available for most strains used. "

Soleus Muscle Data and Bias Considerations: Were measurements taken for lean mass and soleus muscle weight? If so, please present the corresponding data.

Measurements for lean mass and the mass of soleus muscle after grinding have been including in Supplementary Figure S1 (panels c-d)

As outlined in the methods section, the variation in protein yield from the soleus muscle across each strain is substantial. Notably, the distinct peptide input for phospho enrichment introduces biases, given that muscles with lower input may exhibit reduced identification (Fig S2). This bias might also manifest in the PCA plot (S2C). Ideally, adopting a uniform protein/peptide input would have been advantageous. Address this concern and contemplate moving the PCA plot to the main figure. It's prudent to reconsider the sentence stating, "Samples from animals of the same strain and diet were highly correlated and generally clustered together, implying the data are highly reproducible (Fig. S2b-d)," particularly if the input and total IDs were not matched.

The reviewer highlights an important point. As the reviewer comments, it would have been our preference to use the same amount of protein material for all samples. However, as there was a wide range in the mass of the soleus muscle across mouse strains (in particular much lower in CAST mice), it was not appropriate to use the same amount of material for all strains. This is indeed evident in the PCA plot (Figure S2c), whereby samples cluster in the second component (PC2) based on the amount of protein material. However, this clustering is not observed in the hierarchical clustering (Figure S2d), and nor are the number of phosphopeptides quantified in each sample substantially impacted by these differences (Figure S2a) as implied by the reviewer. Indeed, the number of phosphopeptides quantified did not noticeably vary when comparing BXH9/BXD34 to C57Bl6J/NOD despite 32.3% less material used, and there were only 12.4% fewer phosphopeptides (average #13891.56 vs 15851.29) in CAST compared to C57Bl6J/NOD strains, despite 51.8% less material used. To further emphasise the minimal effect that input material had on phosphopeptide quantification, we have additionally plotted the number of phosphopeptides quantified in each sample following the filtering steps we employed prior to statistical analysis of the dataset (i.e. ANOVA). This plot (Author response image 1) shows that there is even less variation in the number of quantified phosphopeptides between strains, with only 9.12% fewer phosphopeptides quantified and filtered on average in CAST compared to C57Bl6J/NOD (average #9026.722 vs 9932.711). From a quantitative perspective, in both the PCA (Principal Component 1) and hierarchical clustering analyses, samples are additionally clustered by individual strains, and in the latter they also cluster generally by diet, implying that biological variation between samples remains the primary variation captured in our data. We have modified the manuscript so that these observations are forefront (lines 103-106): "Furthermore, while different strains clustered by the amount of protein material used in the second component of the PCA (Figure S2c), samples from animals of the same strain and diet were highly correlated and generally clustered together, indicating that our data are highly reproducible". To ensure that readers are aware of our decision to alter protein starting material and its implications, we have moved the description of this from the methods to the results, and we have highlighted the impact on phosphopeptide quantification in CAST mice (lines 99-103): "Due to the range in soleus mass across strains (Fig. S1D) we altered the protein material used for EasyPhos (C57Bl6J and NOD: 755 µg, BXH9 and BXD34: 511 µg, CAST: 364 µg), though phosphopeptide quantification was minimally affected, with only 12.4% fewer phosphopeptides quantified on average in CAST compared to the C57lB6J/NOD (average 13891.56 vs 15851.29 Fig. S2a)."

Author response image 1.

Phosphopeptide quantification following filtering. a) The number of phosphopeptides quantified in each sample after filtering prior to statistical analysis.

Phosphosite Quantification Filtering: The quantified phosphosites have been dropped from 23,000 to 10,000. Could you elucidate the criteria employed for filtering and provide a concise explanation in the main text?

We thank the reviewer for drawing this ambiguity to our attention. Before testing for insulin regulation, we performed a filtering step requiring phosphopeptides to be quantified well enough for comparisons across strains and diets. Specifically, phosphopeptides were retained if they were quantified well enough to assess the effect of insulin in more than eight strain-diet combinations (≥ 3 insulin-stimulated values and ≥ 3 unstimulated values in each combination). We have now included this explanation of the filtering in the main text on lines 108-114.

ANOVA Choice Clarification: In Figure 4, there's a transition from one-way ANOVA in B to two-way ANOVA in C. Could you expound on the rationale for selecting these distinct methods?

In panel B, we first focussed on kinase regulation differences between strains in the absence of a dietary perturbation. Hence, we performed one-way ANOVAs only within the CHOW-fed mice. In panel C, we then consider the effect of perturbation with the HFD. We perform two-way ANOVAs, allowing us to identify effects of the HFD that are uniform across strains (Diet main effect) or variable across strains (Strain-by-diet interaction).

Cell Line Selection for Functional Experiments: Could you elucidate the rationale behind opting for L6 cells of rat origin over C2C12 mouse cells for functional experiments?

We acknowledge that C2C12 cells have the benefit of being of mouse origin, which aligns with our mouse-derived phosphoproteomics data. However, they are unsuitable for glucose uptake experiments as they lack an insulin-responsive vesicular compartment even upon GLUT4 overexpression, and undergo spontaneous contraction when differentiated resulting in confounding non-insulin dependent glucose uptake (10.1152/ajpendo.00092.2002, 10.1007/s11626-999-0030-8). In contrast, L6 cells readily express insulin-responsive GLUT4, and cannot contract (doi.org/10.1113/JP281352, 10.1007/s11626-999-0030-8). Therefore they are a superior model for studying insulin-dependent glucose transport. We have added a justification of L6 cells over C2C12 cells in the revised manuscript, on lines 352-354: "While L6 cells are of rat origin, they are preferable to the popular C2C12 mouse cell line since the latter lack an insulin-responsive vesicular compartment and undergo spontaneous contraction, resulting in confounding non-insulin dependent glucose uptake."

It's intriguing that while a phosphosite was modulated on Pfkfb2, functional assays were conducted on a different isoform (Pfkfb3) wherein the phosphosite was not detected.

The correlation between Pfkfb2 S469 phosphorylation and insulin-stimulated glucose uptake suggests that F2,6BP production, and subsequent glycolytic activation, positively regulate insulin responsiveness. There are several ways of testing this: 1) Knock down endogenous Pfkfb2, and re-express either wild-type protein or a S469A phosphomutant. If S469 phosphorylation positively regulates insulin responsiveness, then knockdown should decrease insulin responsiveness and re-expression of wild-type Pfkfb2, but not S469A, should restore it. 2) Induce insulin resistance (e.g. through palmitate treatment), and overexpress phosphomimetic S469D or S469E Pfkfb2 to enhance F2,6BP production. Under our hypothesis, this should reverse insulin resistance. 3) There is some evidence that dual phosphorylation of S469 and S486, another activating phosphosite on Pfkfb2, enhances F2,6BP production through 14-3-3 binding (10.1093/emboj/cdg363). Hence, we may expect that introduction of an R18 sequence into Pfkfb2, which causes constitutive 14-3-3 binding (10.1074/jbc.M603274200), would increase Pfkfb2-driven F2,6BP production, and under our hypothesis this should reverse insulin resistance. 4) The paralog Pfkfb3 lacks Akt regulatory sites and has substantially higher basal activity than Pfkfb2. Thus, overexpression of Pfkfb3 should mimic the effect of phosphorylated Pfkfb2, and hence reverse insulin resistance under our hypothesis. While approaches 1), 2), and 3) directly target Pfkfb2, they have drawbacks. For example, 1) may not work if Pfkfb2 knockdown is compensated for by other Pfkfb isoforms, 2) may not work since D/E phosphomimetics often do not recapitulate the molecular effects of S/T phosphorylation (10.1091/mbc.E12-09-0677), and 3) may not work if S469 phosphorylation does not operate through 14-3-3 binding. Hence we performed 4) as it seemed to be the most robust and cleanest experiment to test our hypothesis. We have revised the manuscript to further clarify the challenges of directly targeting Pfkfb2 and the benefits of targeting Pfkfb3 on lines 342-349: "Since Pfkfb2 requires phosphorylation by Akt to produce F2,6BP substantially, increasing F2,6BP production via Pfkfb2 would require enhanced activating site phosphorylation, which is difficult to achieve in a targeted fashion, or phosphomimetic mutation of activating sites to aspartate/glutamate, which often does not recapitulate the molecular effects of serine/threonine phosphorylation. By contrast, the paralog Pfkfb3 has high basal production rates and lacks an Akt motif at the corresponding phosphosites. We therefore rationalised that overexpressing Pfkfb3 would robustly increase F2,6BP production and enhance glycolysis regardless of insulin stimulation and Akt signalling."

Insulin-Independent Action of Pfkfb3: The functionality of Pfkfb3 unfolds in an insulin-independent manner, yet it restores insulin action (Fig 6h). Could you shed light on the mechanism underpinning this phenomenon? Consider measuring F2,6BP concentrations or assessing kinase activity upon overexpression.

Pfkfb3 overexpression increased the glycolytic capacity of L6 myotubes in the absence of insulin stimulation, as inferred by extracellular acidification rate (Fig. S7c). This is indeed consistent with Pfkfb3 enhancing glycolysis through increased F2,6BP concentration in an insulin-independent manner. To shed light on the mechanism connecting this to insulin action, we performed immunoblotting experiments to assess the kinase activity of Akt, a master regulator of the insulin response. Indeed, this experimental direction has precedent as we previously observed that Pfkfb3 overexpression enhanced insulin-stimulated Akt signalling in HEK293 cells, while small-molecule inhibition of Pfkfb kinase activity reduced Akt signalling in 3T3-L1 adipocytes (10.1074/jbc.M115.658815). However, insulin-stimulated phosphorylation of Akt S473, Akt T308, Gsk3a/b S21/S9, and PRAS40 T246 differed little across conditions, with only a weak, statistically insignificant trend towards increased pT308 Akt, pS21/S9 Gsk3a/b, and pT246 PRAS40 in palmitate-treated Pfkfb3-overexpressing cells. Hence, a more detailed phosphoproteomics study will be needed to assess whether Pfkfb3 restores insulin action by modulating insulin signalling. We have described these immunoblotting experiments in lines 361-365 and Fig. S7e-f. We also discussed potential mechanisms through which Pfkfb3-enhanced glycolysis could connect to insulin action in the discussion (lines 427-434).

Figure 6h Statistical Analysis: For the 2DG uptake in Figure 6h, a conventional two-way ANOVA might be more appropriate than a repeated measures ANOVA.

On reflection, we agree that a conventional ANOVA is more appropriate. Furthermore, for simplicity and conciseness we have decided to analyse and present only insulin-stimulated/unstimulated 2DG uptake fold change values in Figure 6h. We have presented all unstimulated and insulin-stimulated values in Figure S7d.

Author Response

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

Thank you for overseeing the assessment of our manuscript, “Comprehensive mutagenesis maps the effect of all single codon mutations in the AAV2 rep gene on AAV production". We would also like to thank the reviewers for their feedback. We have carried out the suggested experiments that we feel are most central to our conclusions and summarized the revisions to the manuscript below.

We appreciate the reviewers’ suggestion with regards to testing different rAAV genomes. We have measured the effect of Rep variants on the production of rAAV containing three additional genomes: a 4.4 kb single-stranded genome, a 3.9 kb single-stranded genome, and a 2.1 kb self-complementary genome (Figures 5C and 5D). The DNase-resistant particles titers - reported as a percent of wild-type Rep titers - are relatively consistent across these three constructs as well as the 5.0 kb single-stranded genome previously tested.

We agree with the reviewers that measurement of the relative transduction efficiency of rAAV produced with different Rep variants is an important experiment to conduct. To address this, we transduced HEK293T cells with rAAVs, containing a luciferase genome, which were produced using two different Rep variants. When a constant volume of purified rAAV was used for transduction, we observed that the rAAV produced with the S110R Rep variant resulted in higher transduction than rAAV produced with wild-type Rep (as measured by luciferase signal). While we tested only a small number of variants, these results indicate that at least one of the Rep variants we identified can increase not only the viral genome titer but also the titer of transducing particles.

To generate this transduction data, we produced additional rAAV preps using S110R and Q439T Rep variants. In the previous version of this manuscript, we used the Q439T variant to produce rAAV and noted a 10% increase in the ratio of viral genomes: capsids as determined by comparison of qPCR and capsid ELISA titers. However, a similar increase was not observed in the more recent experiment discussed above. We attribute this discrepancy to changes in the plasmid quantification methods used for transfection. Previously, we quantified plasmids using a fluorometric assay (Qubit); in our more recent experiments, we used qPCR to quantify plasmids for transfection. qPCR provides a more accurate measurement of plasmid concentration due to the specific nature of the primers and probes used, which may account for the subtle shift in quantification. While outside the scope of the current work, it will also be interesting to further investigate the proportion of full capsids using additional Rep variants and more direct methods, such as cryoEM or analytical ultracentrifugation.

We agree with the reviewers’ observation that there are differences in the production fitness values for synonymous variants. However, the variation in production fitness values between synonymous variants is smaller than that between non-synonymous variants. We conducted the following analysis to clarify this point. We calculated two mean centered fitness values for each codon variant in the WT AAV2 library. The “positional mean centered fitness value” was determined using the production fitness values of all variants at a given amino acid position and describes how far a given fitness value diverges from the mean fitness value for that position. The “synonymous codon mean centered fitness value” was determined using the production fitness values of all synonymous variants at a given position and describes how far a given fitness value diverges from the mean fitness value for all its synonymous codon variants. We then plotted both mean centered fitness values versus amino acid position (Figure S8).

The distribution of mean centered selection values is narrower when calculated at the synonymous codon level as opposed to the position level. This indicates that, in general, synonymous variants have more tightly distributed production fitness values than non-synonymous variants. This observation precludes us from conducting a more thorough analysis of the effects of synonymous codons on AAV production. (Although, there is at least one instance where clear differences between synonymous codons can be observed (Figure S9C and Figure S9D).) We agree with the reviewers that synonymous variants almost certainly influence aspects of AAV production, such as genome replication, transcriptional regulation, mRNA stability, and protein expression. However, our assay measures the aggregate effect of rep variants on all steps in the AAV production process and is likely unable to detect the effects of synonymous variants on specific steps in this process if those steps are not rate-limiting. We have updated the discussion section to include an explanation of the above.

The X-axes in Figures 5B and 5D have been updated to plot s’ instead of percent WT titer. We have also added asterisks to indicate significance in Figures 5A and 5C. Thank you for these suggestions.

We agree with Reviewer 3 that it would be interesting to sequence barcodes from the mRNA pool. The 20 bp barcodes are located upstream of the polyA site and should be present in mRNA transcripts. Something to consider is that AAV2 transcripts expressed from all three promoters (p5, p19, and p40) are polyadenylated at the same site (Stutika et al., 2016). As such, in our WT AAV2 library, barcode representation in the mRNA pool would indicate the aggregate effect of a rep variant on the levels of all AAV2 transcripts. In the pCMV-Rep78/68 library, only two AAV2 transcripts are generated - a spliced and unspliced version of the p5 product. Sequencing of barcodes present in the mRNA pool could be informative regarding the effect of rep variants on combined Rep78/68 expression levels. However, we feel that this experiment is outside the scope of the current work.

We were also surprised at the number of novel functional Rep variants that were identified in our library. As the reviewer pointed out, optimal rAAV production likely does not equate to optimal fitness of naturally occurring AAV in the endogenous host. Naturally occurring AAV has both a latent and a lytic cycle and the Rep proteins play a role in both these processes (Pereira et al., 1997; Surosky et al., 1997). rAAV production, however, is primarily analogous to the lytic cycle of naturally occurring AAV. In their endogenous hosts, AAV must balance the effect of any mutations on fitness in both the lytic and latent contexts while we assay specifically for production fitness. We additionally attribute this finding to the relatively small number of AAV serotypes, for which rep sequences are available. We have added a discussion of the above to the manuscript.

Finally, in response to feedback from other researchers, we determined which amino acid substitutions resulted in production fitness values that were significantly different from that of wild-type (Figure S4). These results further emphasized the importance of the origin-binding domain; most statistically significant beneficial substitutions clustered here. Additionally, we noted that the majority of substitutions in the zinc-finger domain resulted in production fitness changes that were not significant. This lines up with previous work indicating that the zinc-finger domain is dispensable for rAAV production. We have added a discussion of these results to the main text.

We again thank the reviewers for their suggestions; we feel that incorporation of their suggestions has strengthened support for our conclusions and enhanced the utility of this work for others in the field.

References Pereira, D. J., McCarty, D. M., & Muzyczka, N. (1997). The adeno-associated virus (AAV) Rep protein acts as both a repressor and an activator to regulate AAV transcription during a productive infection. Journal of Virology, 71(2), 1079–1088. https://doi.org/10.1128/jvi.71.2.1079-1088.1997

Stutika, C., Gogol-Döring, A., Botschen, L., Mietzsch, M., Weger, S., Feldkamp, M., Chen, W., & Heilbronn, R. (2016). A Comprehensive RNA Sequencing Analysis of the Adeno-Associated Virus (AAV) Type 2 Transcriptome Reveals Novel AAV Transcripts, Splice Variants, and Derived Proteins. Journal of Virology, 90(3), 1278–1289. https://doi.org/10.1128/JVI.02750-15

Surosky, R. T., Urabe, M., Godwin, S. G., McQuiston, S. A., Kurtzman, G. J., Ozawa, K., & Natsoulis, G. (1997). Adeno-associated virus Rep proteins target DNA sequences to a unique locus in the human genome. Journal of Virology, 71(10), 7951–7959. https://doi.org/10.1128/jvi.71.10.7951-7959.1997

Author Response

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

eLife assessment

The authors' finding that PARG hydrolase removal of polyADP-ribose (PAR) protein adducts generated in response to the presence of unligated Okazaki fragments is important for S-phase progression is potentially valuable, but the evidence is incomplete, and identification of relevant PARylated PARG substrates in S-phase is needed to understand the role of PARylation and dePARylation in S-phase progression. Their observation that human ovarian cancer cells with low levels of PARG are more sensitive to a PARG inhibitor, presumably due to the accumulation of high levels of protein PARylation, suggests that low PARG protein levels could serve as a criterion to select ovarian cancer patients for treatment with a PARG inhibitor drug.

Thank you for the assessment and summary. Please see below for details as we have now addressed the deficiencies pointed out by the reviewers.

We believe that PARP1 is one of the major relevant PARG substrates in S phase cells. Previous studies reported that PARP1 recognizes unligated Okazaki fragments and induces S phase PARylation, which recruits single-strand break repair proteins such as XRCC1 and LIG3 that acts as a backup pathway for Okazaki fragment maturation (Hanzlikova et al., 2018; Kumamoto et al., 2021). In this study, we revealed that accumulation of PARP1/2-dependent S phase PARylation eventually led to cell death (Fig. 2). Furthermore, we found that chromatin-bound PARP1 as well as PARylated PARP1 increased in PARG KO cells (Fig. S4A and Fig. 4A), suggesting that PARP1 is one of the key substrates of PARG in S phase cells. Of course, PARG may have additional substrates besides PARP1 which are required for its roles in S phase progression, as PARG is known to be recruited to DNA damage sites through pADPr- and PCNA-dependent mechanisms (Mortusewicz et al., 2011). Precisely how PARG regulates S phase progression warrants further investigation.

Public Reviews:

Reviewer #1 (Public Review):

I have a major conceptual problem with this manuscript: How can the full deletion of a gene (PARG) sensitize a cell to further inhibition by its chemical inhibitor (PARGi) since the target protein is fully absent?

Please see below for details about this point. Briefly, we found that PARG is an essential gene (Fig. 7). There was residual PARG activity in our PARG KO cells, although the loss of full-length PARG was confirmed by Western blotting and DNA sequencing (Fig. S9). The residual PARG activity in these cells can be further inhibited by PARG inhibitor, which eventually lead to cell death.

The authors state in the discussion section: "The residual PARG dePARylation activity observed in PARG KO cells likely supports cell growth, which can be further inhibited by PARGi". What does this statement mean? Is the authors' conclusion that their PARG KOs are not true KOs but partial hypomorphic knockdowns? Were the authors working with KO clones or CRISPR deletion in populations of cells?

The reviewer is correct that our PARG KOs are not true KOs. We were working with CRISPR edited KO clones. As shown in this manuscript, we validated our KO clones by Western blotting, DNA sequencing and MMS-induced PARylation. Despite these efforts and our inability to detect full-length PARG in our KO clones, we suspect that our PARG KO cells may still express one or more active fragments of PARG due to alternative splicing and/or alternative ATG usage.

As shown in Fig. 7, we believe that PARG is essential for proliferation. Our initial KO cell lines are not complete PARG KO cells and residual PARG activity in these cells could support cell proliferation. Unfortunately, due to lack of appropriate reagents we could not draw solid conclusions regarding the isoforms or the truncated PARG expressed in these cells (Please see Western blots below).

Are there splice variants of PARG that were not knocked down? Are there PARP paralogues that can complement the biochemical activity of PARG in the PARG KOs? The authors do not discuss these critical issues nor engage with this problem.

There are five reviewed or potential PARG isoforms identified in the Uniprot database. The two sgRNAs (#1 and #2) used to generate initial PARG KO cells in this manuscript target all three catalytically active isoforms (isoforms 1, 2 and 3), and sgRNA#2 used in HeLa cells also targets isoforms 4 and 5, but these isoforms are considered catalytically inactive according to the Uniprot database. However, it is likely that sgRNA-mediated genome editing may lead to the creation of new alternatively spliced PARG mRNAs or the use of alternative ATG, which can produce catalytically active forms of PARG. Instead of searching for these putative spliced PARG RNAs, we used two independent antibodies that recognize the C-terminus of PARG for WB as shown below. Unfortunately, besides full-length PARG, these antibodies also recognized several other bands, some of them were reduced or absent in PARG KO cells, others were not. Thus, we could not draw a clear conclusion which functional isoform was expressed in our PARG KO cells. Nevertheless, we directly measured PARG activity in PARG KO cells (Fig. S9) and showed that we were still able to detect residual PARG activity in these PARG KO cells. These data clearly indicate that residual PARG activity are present and detected in our KO cells, but the precise nature of these truncated forms of PARG remains elusive.

Author response image 1.

These issues have to be dealt with upfront in the manuscript for the reader to make sense of their work.

We thank this reviewer for his/her constructive comments and suggestions. We will include the data above and additional discussion upfront in our revised manuscript to avoid any further confusion by our readers.

Reviewer #2 (Public Review):

Summary:

In this manuscript, Nie et al investigate the effect of PARG KO and PARG inhibition (PARGi) on pADPR, DNA damage, cell viability, and synthetic lethal interactions in HEK293A and Hela cells. Surprisingly, the authors report that PARG KO cells are sensitive to PARGi and show higher pADPR levels than PARG KO cells, which are abrogated upon deletion or inhibition of PARP1/PARP2. The authors explain the sensitivity of PARG KO to PARGi through incomplete PARG depletion and demonstrate complete loss of PARG activity when incomplete PARG KO cells are transfected with additional gRNAs in the presence of PARPi. Furthermore, the authors show that the sensitivity of PARG KO cells to PARGi is not caused by NAD depletion but by S-phase accumulation of pADPR on chromatin coming from unligated Okazaki fragments, which are recognized and bound by PARP1. Consistently, PARG KO or PARG inhibition shows synthetic lethality with Pol beta, which is required for Okazaki fragment maturation. PARG expression levels in ovarian cancer cell lines correlate negatively with their sensitivity to PARGi.

Thank you for your nice comments. The complete loss of PARG activity was observed in PARG complete/conditional KO (cKO) cells. These cKO clones were generated using wild-type cells transfected with sgRNAs targeting the catalytic domain of PARG in the presence of PARP inhibitor.

Strengths:

The authors show that PARG is essential for removing ADP-ribosylation in S-phase.

Thanks!

Weaknesses:

1. This begs the question as to the relevant substrates of PARG in S-phase, which could be addressed, for example, by analysing PARylated proteins associated with replication forks in PARG-depleted cells (EdU pulldown and Af1521 enrichment followed by mass spectrometry).

We believe that PARP1 is one of the major relevant PARG substrates in S phase cells. Previous studies reported that PARP1 recognizes unligated Okazaki fragments and induces S phase PARylation, which recruits single-strand break repair proteins such as XRCC1 and LIG3 that acts as a backup pathway for Okazaki fragment maturation (Hanzlikova et al., 2018; Kumamoto et al., 2021). In this study, we revealed that accumulation of PARP1/2-dependent S phase PARylation eventually led to cell death (Fig. 2). Furthermore, we found that chromatin-bound PARP1 as well as PARylated PARP1 increased in PARG KO cells (Fig. S4A and Fig. 4A), suggesting that PARP1 is one of the key substrates of PARG in S phase cells. Of course, PARG may have additional substrates besides PARP1 which are required for its roles in S phase progression, as PARG is known to be recruited to DNA damage sites through pADPr- and PCNA-dependent mechanisms (Mortusewicz et al., 2011). Precisely how PARG regulates S phase progression warrants further investigation.

1. The results showing the generation of a full PARG KO should be moved to the beginning of the Results section, right after the first Results chapter (PARG depletion leads to drastic sensitivity to PARGi), otherwise, the reader is left to wonder how PARG KO cells can be sensitive to PARGi when there should be presumably no PARG present.

Thank you for your suggestion! However, we would like to keep the complete PARG KO result at the end of the Results section, since this was how this project evolved. Initially, we did not know that PARG is an essential gene. Thus, we speculated that PARGi may target not only PARG but also a second target, which only becomes essential in the absence of PARG. To test this possibility, we performed FACS-based and cell survival-based whole-genome CRISPR screens (Fig. 5). However, this putative second target was not revealed by our CRISPR screening data (Fig. 5). We then tested the possibility that these cells may have residual PARG expression or activity and only cells with very low PARG expression are sensitive to PARGi, which turned out to be the case for ovarian cancer cells. Equipped with PARP inhibitor and sgRNAs targeting the catalytic domain of PARG, we finally generated cells with complete loss of PARG activity to prove that PARG is an essential gene (Fig. 7). This series of experiments underscore the challenge of validating any KO cell lines, i.e. the identification of frame-shift mutations, absence of full-length proteins, and phenotypic changes may still not be sufficient to validate KO clones. This is an important lesson we learned and we would like to share it with the scientific community.

To avoid further misunderstanding, we will include additional statements/comments at the end of “PARG depletion leads to drastic sensitivity to PARGi” section and at the beginning of “CRISPR screens reveal genes responsible for regulating pADPr signaling and/or cell lethality in WT and PARG KO cells”. Hope that our revised manuscript will make it clear.

1. Please indicate in the first figure which isoforms were targeted with gRNAs, given that there are 5 PARG isoforms. You should also highlight that the PARG antibody only recognizes the largest isoform, which is clearly absent in your PARG KO, but other isoforms may still be produced, depending on where the cleavage sites were located.

The two sgRNAs (#1 and #2) used to generate initial PARG KO cells in this manuscript target all three catalytically active isoforms (isoforms 1, 2 and 3), and sgRNA#2 used in HeLa cells also targets isoforms 4 and 5, but these isoforms are considered catalytically inactive according to the Uniprot database. As suggested, we will modify Fig. S1D and the figure legends.

The manufacturer instruction states that the Anti-PARG antibody (66564S) can only recognize isoform 1, this antibody could recognize isoforms 2 and 3 albeit weakly based on Western blot results with lysates prepared from PARG cKO cells reconstituted with different PARG isoforms, as shown below. As suggested, we will add a statement in the revised manuscript and provide the Western blotting data below.

Author response image 2.

To test whether other isoforms were expressed in 293A and/or HeLa cells, we used two independent antibodies that recognize the C-terminus of PARG for WB as shown below. Unfortunately, besides full-length PARG, these antibodies also recognized several other bands, some of them were reduced or absent in PARG KO cells, others were not. Thus, we could not draw a clear conclusion which functional isoforms or truncated forms were expressed in our PARG KO cells.

Author response image 3.

1. FACS data need to be quantified. Scatter plots can be moved to Supplementary while quantification histograms with statistical analysis should be placed in the main figures.

We agree with this reviewer that quantification of FACS data may provide straightforward results in some of our data. However, it is challenging to quantify positive S phase pADPr signaling in some panels, for example in Fig. 3A and Fig. 4C. In both panels, pADPr signaling was detected throughout the cell cycle and therefore it is difficult to know the percentage of S phase pADPr signaling in these samples. Thus, we decide to keep the scatter plots to demonstrate the dramatic and S phase-specific pADPr signaling in PARG KO cells treated with PARGi. We hope that these data are clear and convincing even without any quantification.

1. All colony formation assays should be quantified and sensitivity plots should be shown next to example plates.

As suggested, we will include the sensitivity plot next to Fig. 3D. However, other colony formation assays in this study were performed with a single concentration of inhibitor and therefore we will not provide sensitivity plots for these experiments. Nevertheless, the results of these experiments are straightforward and easy to interpret.

1. Please indicate how many times each experiment was performed independently and include statistical analysis.

As suggested, we will add this information in the revised manuscript.

Reviewer #3 (Public Review):

Here the authors carried out a CRISPR/sgRNA screen with a DDR gene-targeted mini-library in HEK293A cells looking for genes whose loss increased sensitivity to treatment with the PARG inhibitor, PDD00017273 (PARGi). Surprisingly they found that PARG itself, which encodes the cellular poly(ADP-ribose) glycohydrolase (dePARylation) enzyme, was a major hit. Targeted PARG KO in 293A and HeLa cells also caused high sensitivity to PARGi. When PARG KO cells were reconstituted with catalytically-dead PARG, MMS treatment caused an increase in PARylation, not observed when cells were reconstituted with WT PARG or when the PARG KO was combined with PARP1/2 DKO, suggesting that loss of PARG leads to a strong PARP1/2-dependent increase in protein PARylation. The decrease in intracellular NADH+, the substrate for PARP-driven PARylation, observed in PARG KO cells was reversed by treatment with NMN or NAM, and this treatment partially rescued the PARG KO cell lethality. However, since NAD+ depletion with the FK868 nicotinamide phosphoribosyltransferase (NAMPT) inhibitor did not induce a similar lethality the authors concluded that NAD+ depletion/reduction was only partially responsible for the PARGi toxicity. Interestingly, PARylation was also observed in untreated PARG KO cells, specifically in S phase, without a significant rise in γH2AX signals. Using cells synchronized at G1/S by double thymidine blockade and release, they showed that entry into S phase was necessary for PARGi to induce PARylation in PARG KO cells. They found an increased association of PARP1 with a chromatin fraction in PARG KO cells independent of PARGi treatment, and suggested that PARP1 trapping on chromatin might account in part for the increased PARGi sensitivity. They also showed that prolonged PARGi treatment of PARG KO cells caused S phase accumulation of pADPr eventually leading to DNA damage, as evidenced by increased anti-γH2AX antibody signals and alkaline comet assays. Based on the use of emetine, they deduced that this response could be caused by unligated Okazaki fragments. Next, they carried out FACS-based CRISPR screens to identify genes that might be involved in cell lethality in WT and PARG KO cells, finding that loss of base excision repair (BER) and DNA repair genes led to increased PARylation and PARGi sensitivity, whereas loss of PARP1 had the opposite effects. They also found that BER pathway disruption exhibited synthetic lethality with PARGi treatment in both PARG KO cells and WT cells, and that loss of genes involved in Okazaki fragment ligation induced S phase pADPr signaling. In a panel of human ovarian cancer cell lines, PARGi sensitivity was found to correlate with low levels of PARG mRNA, and they showed that the PARGi sensitivity of cells could be reduced by PARPi treatment. Finally, they addressed the conundrum of why PARG KO cells should be sensitive to a specific PARG inhibitor if there is no PARG to inhibit and found that the PARG KO cells had significant residual PARG activity when measured in a lysate activity assay, which could be inhibited by PARGi, although the inhabited PARG activity levels remained higher than those of PARG cKO cells (see below). This led them to generate new, more complete PARG KO cells they called complete/conditional KO (cKO), whose survival required the inclusion of the olaparib PARPi in the growth medium. These PARG cKO cells exhibited extremely low levels of PARG activity in vitro, consistent with a true PARG KO phenotype.

We thank this reviewer for his/her constructive comments and suggestions.

The finding that human ovarian cancer cells with low levels of PARG are more sensitive to inhibition with a small molecule PARG inhibitor, presumably due to the accumulation of high levels of protein PARylation (pADPr) that are toxic to cells is quite interesting, and this could be useful in the future as a diagnostic marker for preselection of ovarian cancer patients for treatment with a PARG inhibitor drug. The finding that loss of base excision repair (BER) and DNA repair genes led to increased PARylation and PARGi sensitivity is in keeping with the conclusion that PARG activity is essential for cell fitness, because it prevents excessive protein PARylation. The observation that increased PARylation can be detected in an unperturbed S phase in PARG KO cells is also of interest. However, the functional importance of protein PARylation at the replication fork in the normal cell cycle was not fully investigated, and none of the key PARylation targets for PARG required for S phase progression were identified. Overall, there are some interesting findings in the paper, but their impact is significantly lessened by the confusing way in which the paper has been organized and written, and this needs to be rectified.

We believe that PARP1 is one of the major relevant PARG substrates in S phase cells. Previous studies reported that PARP1 recognizes unligated Okazaki fragments and induces S phase PARylation, which recruits single-strand break repair proteins such as XRCC1 and LIG3 that acts as a backup pathway for Okazaki fragment maturation (Hanzlikova et al., 2018; Kumamoto et al., 2021). In this study, we revealed that accumulation of PARP1/2-dependent S phase PARylation eventually led to cell death (Fig. 2). Furthermore, we found that chromatin-bound PARP1 as well as PARylated PARP1 increased in PARG KO cells (Fig. S4A and Fig. 4A), suggesting that PARP1 is one of the key substrates of PARG in S phase cells. Of course, PARG may have additional substrates besides PARP1 which are required for its roles in S phase progression, as PARG is known to be recruited to DNA damage sites through pADPr- and PCNA-dependent mechanisms (Mortusewicz et al., 2011). Precisely how PARG regulates S phase progression warrants further investigation.

As suggested, we will revise our manuscript accordingly and provide additional explanation/statement upfront to avoid any misunderstandings.  

Reviewer #1 (Recommendations For The Authors):

1. Figure 1c. Why does the viability of PARG KO cells improve at higher doses of PARGi? How do the authors explain this paradox?

This phenomenon was observed in 293A PARG KO cells and happened in CellTiter-Glo assay, especially with the top three PARGi concentrations (100 µM, 33.33 µM and 11.11 µM). This may due to the low solubility of this PARGi in the medium, since we sometimes observed precipitation at high concentrations when PARGi stock was diluted in medium.

1. Figure 2d. The authors show that PARGi reduced NAD+ level by 20%. This reduction in NAD+ probably does not explain the cell death phenotype observed by parthanatos cell death. What pathway is activated by PARGi to induce cell death?

Since PARG KO cells treated with PARGi led to uncontrolled pADPr accumulation, it is possible that some of these cells may die due to parthanotos. However, we did not observe a dramatic reduction in NAD+ level. A previous study showed that Parg(-/-) mouse ES cells predominantly underwent caspase-dependent apoptosis (Shirai et al., 2013). Indeed, PARP1 cleavage was detected in PARG KO cells with prolonged PARGi treatment, indicating that at least some of these cells die due to apoptosis (Fig. 2A). Cytotoxicity of PARGi in PARG KO cells may due to several mechanisms including apoptosis, parthanatos and NAD+ reduction.

1. The authors refer to FK866 in the text without explaining what this agent is. FK866 is a noncompetitive inhibitor of nicotinamide phosphoribosyltransferase (NAPRT), a key enzyme in the regulation of NAD+ biosynthesis from the natural precursor nicotinamide. The authors should explain experimental tools in the text as they use them for clarity to the reader.

Thanks for the suggestion! We will include additional citations and discuss how FK866 works in our revised manuscript.

1. In addition to these issues, there are significant formatting and textual problems, such that there are multiple gaps in the body of the text that make coherent reading of the manuscript impossible. Examples are: Page 3 line 10. Page 6 line 5 and line 15, Page 7 line 2, 3, and line 8. Page 8, line 1, and line 3 from bottom. Page 9 line 1, line 7 from bottom and line 9 from the bottom, Page 18 of the results in several places, etc. etc. etc. These formatting errors convey the impression that the submitting authors did not adequately review the manuscript for technical problems prior to submission. The authors need to correct these errors.

Sorry, we will edit the text and remove these gaps as suggested.

Reviewer #3 (Recommendations For The Authors):

1. The major problem with this paper is conceptual - namely, how could PARG knockout cells be hypersensitive to a selective PARG small molecular inhibitor. The evidence in Figure 7 that there is measurable residual PARG activity in the so-called PARG KO 293A and HeLa cells provides a partial explanation for why PARG inhibitor treatment might be deleterious to the PARG KO cells, i.e., because PARGi blocks this residual PARG activity. However, although the authors characterized the PARG alleles in the 293A PARG KO cells by sequencing, the molecular origin of the significant level of residual PARG activity remains unclear (see points 7-9).

Yes, in our study we showed that PARGi treatment inhibited the residual PARG activity in PARG KO cells, which mimics complete loss of PARG as PARG is an essential gene. These data agree with a previous study using Parg(-/-) mouse cells (Koh et al., 2004).We attempted to define the molecular origin of the residual PARG activity, unfortunately this was challenging (please see below for additional discussions). Nevertheless, we showed that residual PARG activity could be detected in PARG KO cells and more importantly cells with reduced PARG expression or activity are sensitive to PARGi. These results indicate that PARG expression and/or activity may be used as a biomarker for PARGi-based therapy.

1. Although the most obvious explanation for the PARGi sensitivity data presented in Figures 1-4 is that the PARG KO cells have residual PARG activity, the authors wait until the discussion on page 26 to raise the possibility that the PARG KO cells might have residual PARG activity that renders them sensitive to PARGi. It would be more logical to move the PARG activity data in Figure 7 earlier in the paper as a supplementary figure, so that the reader is not left wondering how a PARG KO cell remains sensitive to a PARG inhibitor. For this reason, it is recommended that the whole paper be reorganized and rewritten to provide a more logical flow that allows the reader to understand what was done, and why it is hard to generate complete PARG KO cells because the accumulation of pADPR adducts is toxic to the cell.

Thank you for your suggestion! However, we would like to keep the complete PARG KO result at the end of the Results section, since this was how this project evolved. Initially, we did not know that PARG is an essential gene. Thus, we speculated that PARGi may target not only PARG but also a second target, which only becomes essential in the absence of PARG. To test this possibility, we performed FACS-based and cell survival-based whole-genome CRISPR screens (Fig. 5). However, this putative second target was not revealed by our CRISPR screening data (Fig. 5). We then tested the possibility that these cells may have residual PARG expression or activity and only cells with very low PARG expression are sensitive to PARGi, which turned out to be the case for ovarian cancer cells. Equipped with PARP inhibitor and sgRNAs targeting the catalytic domain of PARG, we finally generated cells with complete loss of PARG activity to prove that PARG is an essential gene (Fig. 7). This series of experiments underscore the challenge of validating any KO cell lines, i.e. the identification of frame-shift mutations, absence of full-length proteins, and phenotypic changes may still not be sufficient to validate KO clones. This is an important lesson we learned and we would like to share it with the scientific community.

To avoid further misunderstanding, we will include additional statements/comments at the end of “PARG depletion leads to drastic sensitivity to PARGi” section and at the beginning of “CRISPR screens reveal genes responsible for regulating pADPr signaling and/or cell lethality in WT and PARG KO cells”. Hope that our revised manuscript will make it clear.

1. Exactly how PARG activity would be coordinated with PARP1/2 activity during normal S phase to ensure that PARylation can serve its required function, whatever that may be, and is then removed by PARG is unclear - how would this be orchestrated at the level of a replication fork?

PARG is known to be recruited to sites of DNA damage through pADPr- and PCNA-dependent mechanisms (Mortusewicz et al., 2011). Our current hypothesis is that PARP1 is one of the major PARG substrates in S phase cells. Previous studies reported that PARP1 recognizes unligated Okazaki fragments and induces S phase PARylation, which recruits single-strand break repair proteins such as XRCC1 and LIG3 that acts as a backup pathway for Okazaki fragment maturation (Hanzlikova et al., 2018; Kumamoto et al., 2021). In this study, we revealed that accumulation of PARP1/2-dependent S phase PARylation eventually led to cell death (Fig. 2). Furthermore, we found that chromatin-bound PARP1 as well as PARylated PARP1 increased in PARG KO cells (Fig. S4A and Fig. 4A), suggesting that PARP1 is one of the key substrates of PARG in S phase cells. Of course, PARG may have additional substrates besides PARP1 which are required for its roles in S phase progression. Precisely how PARG regulates S phase progression warrants further investigation.

1. Figure 2B: What gRNAs were used to generate the 293A and HeLa PARG knock clones, i.e., where are they located in the PARG gene? If they are not in the catalytic domain it might be possible to generate PARG proteins with N-terminal deletions that are still active (see points 8-10 below).

The two sgRNAs (#1 and #2) used to generate initial PARG KO cells in this manuscript target all three catalytically active isoforms (isoforms 1, 2 and 3), and sgRNA#2 used in HeLa cells also targets isoforms 4 and 5, but these isoforms are considered catalytically inactive according to the Uniprot database. As suggested, we will modify Fig. S1D and the figure legends to show the localization of gRNAs.

We agree with this reviewer that truncated but active forms of PARG exist in these KO cells. We attempted to identify these trunated forms of PARG by using two independent antibodies that recognize the C-terminus of PARG for WB as shown below. Unfortunately, besides full-length PARG, these antibodies also recognized several other bands, some of them were reduced or absent in PARG KO cells, others were not. Thus, we could not draw a clear conclusion which functional isoform/truncated form was expressed in our PARG KO cells. Nevertheless, we directly measured PARG activity in PARG KO cells (Fig. S9) and showed that we were still able to detect residual PARG activity in these PARG KO cells. Based on these results, we stated that the residual PARG activity was detected in our KO cells, but we were not able to specify the truncated variants of PARG in these cells.

Author response image 4.

1. Figure 3B/page 19: The authors state that "emetine, which diminishes Okazaki fragments, greatly inhibited S phase pADPr signaling in PARG KO cells", and from this deduced that Okazaki fragments on the lagging strand activate PARylation. However, emetine is not a specific lagging strand synthesis inhibitor, as implied here, but rather a protein synthesis inhibitor, which inhibits Okazaki fragment formation indirectly (see PMID: 36260751). The authors need to rewrite this section to explain how emetine works in this context.

As suggested, we will cite this reference and discuss how emetine inhibits Okazaki fragment maturation in our revised manuscript. Additionally, we used three different POLA1 inhibitors to diminish Okazaki fragments. As shown in Fig. S3B, all three POLA1 inhibitors significantly abolished S-phase pADPr induced by PARGi in PARG KO cells. Furthermore, POLA1 inhibitors, adarotene and CD437, were able to rescue cell lethality caused by PARGi in PARG KO cells (Fig. 3E).

1. Figure 7: It is not clear why these cells are called PARG complete/conditional KO cells (cKO). Generally, "conditional knockout" refers to a cell or animal in which a gene can be conditionally knocked out by inducible expression of Cre. Here, it appears that "conditional" refers to the fact that the PARG KO cells only grow in the presence of olaparib - is this the case?

Yes, we used the name to separate these cells from our initial PARG KO cells. Moreover, we were only able to obtain and maintain these PARG cKO clones with complete loss of PARG activity in the presence of PARP inhibitor. Therefore, we called them PARG complete/conditional KO (cKO) cells.

1. Figure 7B and D: The level of full-length PARG protein was much lower in the 293A and HeLa cKO cells compared to WT cells consistent with cKO cells representing a more complete PARG KO. The level of PARG protein in the 293A PARG cKO cells was apparently also lower than in the original PARG KO cells, but the KO and cKO samples should be run side by side to demonstrate this conclusively, and the bands need to be quantified. In panel B, it is not clear from the legend what cKO_3 and cKO_4 are, but presumably, they are different clones, and this should be stated.

Full-length PARG was not detected in either PARG KO or PARG cKO cells by WB. The apparent lower level of endogenous PARG in Fig. 7D was due to the fact that reconstituted cells had high exogenous PARG expression and therefore we had to reduce exposure time for WB.

As for cKO_3 and cKO_4 in Fig.7, they are different clones created by different sgRNAs. As suggested, we will include additional information in figure legends to clearly state which sgRNA was used to generate the respective KO and cKO clones.

1. Figure S8: There is not enough information here or in the text to allow the reader to interpret these PARG allele sequences obtained from the PARG KO cells. From the Methods section, it appears that the PARG KO cells were clonal, with sequence data from one clone of each of the 293A and HeLa cell PARG KO cells being shown. If this is right, then in both cell types one out of four PARG alleles is wild type, and therefore one would expect the PARG protein signal to be ~25% of that in WT cells. However, based on the 293A PARG KO cells PARG immunoblot in Figure 2B the PARG protein signal is clearly much lower than 25% (these bands need to be quantified), and this discrepancy needs to be explained. What is the level of PARG protein in the PARG KO HeLa cells? If different PARG KO cell clones are analyzed by sequencing, do they all have an apparently intact PARG allele? Four different gRNA target sites in the PARG gene are shown in panel A in Figure 7, but the description in the text regarding how the four gRNAs were used is totally inadequate - were all four used simultaneously or only the two in the catalytic domain? Were pairs of gRNAs used in an attempt to generate a large intervening deletion - some Southern blots of the PARG gene region in the PARG cKO cells are needed to figure this out. The gRNAs are given numbers in Figure 7A, but it is unclear from the sequences shown in Figures S8 and S9 which gRNA sites are shown. All of this has to be clarified, so that the reader can understand the nature of the KO/cKO cells knockout alleles, and what PARG-related products, if any, they can express.

Yes, all KO and cKO cells used in this study are single clones. As suggested, we will revise figure legends in Fig.7, S8 and S9 to include detailed information. To avoid any further misunderstanding, we will label the allele “WT” to “WT (reference)” in Fig. S8 and S9. We did not detect intact/wild-type PARG sequence in any single KO/cKO clone by DNA sequencing. Sequencing of single KO/cKO clones was performed by using TOP TA Cloning kit. Briefly, genomic DNA was extracted from each single KO/cKO clone. Approximately 300bp surrounding the sgRNA targeting sequence was amplified by PCR. The PCR product was cloned into the vector and approximately 10-15 bacteria clones were extracted and sent for sequencing. If any intact/wild-type PARG sequence was detected in these 10-15 bacteria clones, this KO/cKO clone was considered heterozygous clone and discarded.

HEK293A and HeLa cells are not diploid cells and have complex karyotypes. PARG gene is located on chromosome 10. Karyotyping by M-FISH shows that HeLa cells have 3 copies of chromosome 10 (Landry et al., 2013). HEK293 cells predominantly have 3 copies of chromosome 10 and sometimes 4 copies can be detected by G-banding (Binz et al., 2019). Therefore, it is anticipated that 1 to 4 mutant alleles would be detected in each KO/cKO clone by sequencing.

Only one sgRNA was transfected into cells for the selection of single clones. We did not use paired or multiple sgRNAs in any of these experiments. As shown in Fig. S1D and Fig. 7A, HEK293A derived and HeLa derived PARG KO single clones were generated with the use of different sgRNAs. In addition, the two PARG cKO single clones from HEK293A and HeLa cells were also generated by the use of two different sgRNAs, as shown in Fig. 7A-B. We will include all the information above in the revised manuscript, i.e. in Methods section as well as in figure legends.

1. Figure S9A: The sequences of the 293A PARG alleles in the cKO cells suggest that these cells also have one intact PARG allele, which again does not fit with the very low level of intact PARG protein shown in Figure 7B. How do the authors explain this?

Sorry, this is a misunderstanding. The allele “WT” in Fig. S8 and S9 is the reference sequence. We will change it to “Reference sequence” to avoid further confusion. As mentioned above, we did not detect any intact/wild-type PARG sequence in any of our single KO/cKO clones by sequencing.

1. Figure S9B: These critical lysate activity data show that the PARG KO cells have ~50% of the PARG activity detected in WT cells. However, this is not consistent with the PARG protein level detected in PARG immunoblot in Figure 1B, which appears to be less than 5% of the PARG protein level in WT cells (with one intact PARG allele in these cells one would theoretically expect~ 25%, although this depends on whether all four alleles are expressed equally). One possibility is that active PARG fragments are generated from one or more of the PARG KO alleles in the PARG KO cells. Targeted sequencing of PARG mRNAs might reveal whether there are shorter RNAs that could encode a protein containing the C-terminal catalytic domain (aa 570-910). In addition, the authors need to show the entire immunoblot to determine if there are smaller proteins recognized by the anti-PARG antibodies that might represent shorter PARG gene products (for this we need to know where the epitope against which the PARG antibodies are directed are located within the PARG protein - ideally they authors need to use an antibody directed against an epitope near the C-terminus).

As stated in the Methods section, we incubated cell lysates with substrates overnight to evaluate the maximum level of pADPr hydrolysis, i.e. PARG activity, we were able to detect in this assay. It is very likely that the PARG activity in PARG KO cells was much lower than 50%, due to saturation of signals for lysates isolated from wild-type cells. Thus, the data presented in our manuscript probably underestimate the reduction of PARG activity in PARG KO cells. Nevertheless, these data indicate that residual PARG activity was detected in PARG KO cells, however this activity was absent in PARG cKO cells.

As aforementioned, we used two independent antibodies that recognize the C-terminus of PARG for WB. Unfortunately, we could not draw a clear conclusion which functional isoforms or truncated proteins were expressed in our PARG KO cells. The dePARylation assay used here may be the best way to test the residual PARG activity in our KO and cKO cells.

1. Figure 7D: In this experiment, the level of re-expressed WT PARG protein was much higher than that of the endogenous PARG protein (quantification is needed) - how might this affect the interpretation of these experiments (N.B., WT and catalytically-dead PARG were also re-expressed for the experiments shown in Figure 1, but there are no PARG immunoblots to demonstrate how much the exogenous proteins were overexpressed, or activity measurements). If regulated pADPr signaling is important for a normal S phase, then one would have thought that expressing a very high level of active PARG would create problems.

In Fig. S1E, we blotted endogenous PARG level in control cells and exogenous PARG level in reconstituted cells. The reviewer is correct that exogenous PARG expression was much higher (~10-fold) than that of endogenous PARG in WT control cells. Nevertheless, we did not observe any obvious phenotypes in PARG KO/cKO cells reconstituted with high level of exogeneous PARG, which may reflect excess PARG level/activity in wild-type control cells.

References:

Binz, R. L., Tian, E., Sadhukhan, R., Zhou, D., Hauer-Jensen, M., and Pathak, R. (2019). Identification of novel breakpoints for locus- and region-specific translocations in 293 cells by molecular cytogenetics before and after irradiation. Sci Rep 9, 10554.

Hanzlikova, H., Kalasova, I., Demin, A. A., Pennicott, L. E., Cihlarova, Z., and Caldecott, K. W. (2018). The Importance of Poly(ADP-Ribose) Polymerase as a Sensor of Unligated Okazaki Fragments during DNA Replication. Mol Cell 71, 319-331 e313.

Koh, D. W., Lawler, A. M., Poitras, M. F., Sasaki, M., Wattler, S., Nehls, M. C., Stoger, T., Poirier, G. G., Dawson, V. L., and Dawson, T. M. (2004). Failure to degrade poly(ADP-ribose) causes increased sensitivity to cytotoxicity and early embryonic lethality. Proc Natl Acad Sci U S A 101, 17699-17704.

Kumamoto, S., Nishiyama, A., Chiba, Y., Miyashita, R., Konishi, C., Azuma, Y., and Nakanishi, M. (2021). HPF1-dependent PARP activation promotes LIG3-XRCC1-mediated backup pathway of Okazaki fragment ligation. Nucleic Acids Res 49, 5003-5016.

Landry, J. J., Pyl, P. T., Rausch, T., Zichner, T., Tekkedil, M. M., Stutz, A. M., Jauch, A., Aiyar, R. S., Pau, G., Delhomme, N., et al. (2013). The genomic and transcriptomic landscape of a HeLa cell line. G3 (Bethesda) 3, 1213-1224.

Mortusewicz, O., Fouquerel, E., Ame, J. C., Leonhardt, H., and Schreiber, V. (2011). PARG is recruited to DNA damage sites through poly(ADP-ribose)- and PCNA-dependent mechanisms. Nucleic Acids Res 39, 5045-5056.

Shirai, H., Fujimori, H., Gunji, A., Maeda, D., Hirai, T., Poetsch, A. R., Harada, H., Yoshida, T., Sasai, K., Okayasu, R., and Masutani, M. (2013). Parg deficiency confers radio-sensitization through enhanced cell death in mouse ES cells exposed to various forms of ionizing radiation. Biochem Biophys Res Commun 435, 100-106.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

The authors were trying to investigate whether viral IBs are involved in antagonizing IFN-I production during EBOV trVLPs infection. They found that IRF3 is hijacked and sequestered into EBOV IBs after viral infection, thereby leading to the spatial isolation of IRF3 with TBK1 and IKKε. In such a progress, the activity of IRF3 is suppressed and downstream IFN-I induction is inhibited. The authors designed many experiments, such as the PLA that examined the colocalization, to support their conclusions. However, necessary negative controls were missed in several assays. More key index is needed to be examined in several assays.

The paper is well organized and most data in this paper could support the conclusions, while there are several issues that need to be further solved.

1. In Figure 2-4, authors should examine the expression of downstream IFNs as well as the phosphorylation and nuclear localization of IRF3 to further prove the suppression of IRF3 activity by infecting with trVLPs.

Response: The inhibitory effect of trVLPs infection on the phosphorylation of IRF3 S396 and SeV-induced IRF3 nuclear localization was determined by immunoprecipitation (Figure 3D) and immunofluorescence (Figure 4A and 4B), respectively. In addition, we demonstrated that IFN-β transcription was inhibited more potently by EBOV viral inclusion bodies compared with VP35 alone (Figure 7B and 7C).

Moreover, EBOV viral inclusion bodies were demonstrated to inhibit the transcription of IFN downstream genes (e.g., CXCL10, ISG15 and ISG56) more potently than VP35 alone (new Figure 7D-F).

1. In Figure 5, to better prove the conclusion that EBOV NP and VP35 play an important role in sequestering IRF3 in IBS, authors should add the "NP+VP35+VP30" and "NP+VP35+VP24" groups to reperform the assay.

Response: According to the reviewer’s suggestion, VP24 or VP30 was added to the “VP35+NP” group, and the results showed that the “NP+VP35+VP24” and “NP+VP35+VP30” groups exhibited little, if any, effect on the distribution of IRF3 compared with the “NP+VP35” group (new Figure 5 - figure supplement 2A-B).

1. In Figure 6f, the expression of STING should be examined by immunostaining to show the knockdown efficiency in trVLPs-infected cells.

Response: As suggested by the reviewer, immunostaining was performed to visually detect the effect of STING knockdown on the IRF3 distribution during trVLPs infection (new Figure 6F).

Reviewer #2 (Public Review):

The manuscript by Zhu et al explored molecular mechanisms by which Ebola virus (EBOV) evades host innate immune response. EBOV has a number of means to shut down the type I interferon induction (by viral VP35 protein) and block type I interferon action (by viral VP24 protein). This study reported a new mechanism that inclusion body (IB) used for viral replication sequesters IRF3, a key transcription factor involved in the interferon signaling, resulting in blockade of downstream type I interferon gene transcription. This finding is potentially interesting and may provide a new insight into EBOV's evasion of innate immunity. However, there are some flaws in the experimentations and analyses that need to be addressed.

1. Most of experiments were performed by transfection of trVLP plasmids, which is very different from virus infection. The conclusions should be examined and verified in the context of virus infection.

Response: As suggested by the reviewer, the effects of IRF3 depletion on live Ebola virus replication were examined as described in the revised manuscript. Consistent with the results obtained after trVLPs infection, IRF3 depletion exerted little, if any, effect on viral replication (new Figure 7H), which supports the notion that, upon EBOV infection and the formation of inclusion bodies, IRF3 has little, if any, transcription activation activity after sequestration by inclusion bodies.

1. Fig 1 - VP35 displayed a classical IB staining only in Panel A, while much less so in Panel C and not in panel B. It seemed that the VP35 staining images were chosen in a way towards the authors' favor. The statistical analysis of co-localization of VP35 and IRF3, TBK1 or IKKe should be performed to draw the conclusion. Another concern is that IKKe is normally lowly expressed under a rest condition and becomes induced only when the interferon signaling is activated. It seemed to be expressed at a high level even when the interferon signaling is blocked in Panel C. The authors should comment on this discrepancy.

Response: Ebola virus inclusion bodies show variations in both shape and size. According to the reviewer’s suggestion, the colocalization of TBK1 or IKKε and VP35 is shown in new figures (new Figure 1C and 1E), and quantitatively analyzed by the fluorescence intensity using ImageJ software (new Figure 1B, 1D and 1F).

1. Fig 2 - Was this experiment done by transfection or infection? The description of result is not consistent with the figure legend. The labeling was also not consistent between panel A and B. I would suggest performing Western blot to analyze the expression level of IRF3.

Response: We apologize for the incorrect description of the data. Ebola virus trVLPs were initially produced based on transfection but also involved the viral infection process. The use of “transfection” in the figure and figure legends has been changed to “infection” in the revised manuscript. As suggested by the reviewer, Western blotting was performed to analyze the IRF3 expression levels at different time points after trVLPs infection (new Figure 2D).

1. Fig 3 and 4 - As VP35 is well known for its highly efficient blockade of type I interferon activation, how would the authors differentiate the effect of VP35 alone from the sequestration of IRF3 in IBs in these experiments?

Response: Previous studies have found that VP35, rather than NP, inhibits the expression of interferon, and the “VP35+NP” treatment, which induces IRF3 sequestration, showed inhibited IFN-β luciferase activity much more potently than VP35 expression alone (Figure 7B).

1. Fig 3 - PolyIC can activate both RLR and TLR signaling pathways. Can the author comment on which pathway it activates in this experiment?

Response: In this study, the effect of poly(I:C) was consistent with the results observed with SeV, which indicated that poly(I:C) may mainly activate the RLR signaling pathway. A discussion was added to the revised manuscript.

1. The authors demonstrated that VP35 interacts with STING and recruit the latter to IBs. How would this affect the function of STING given that STING plays essential roles in cGAS/cGAMP pathway?

Response: This study unexpectedly showed that VP35 can recruit IRF3 into viral inclusion bodies through STING, but whether it regulates the cGAS-STING pathway remains to be further investigated. Related discussion was added to the revised manuscript.

1. It is difficult to follow the logics of Fig 7. The expression level of each viral protein should be determined. Ideally, a mutation in VP35 that disrupts its ability to antagonize the interferon signaling but still allows for the IB formation can be used to assess the relative contribution of IB sequestering IRF3.

Response: As suggested by the reviewer, a series of VP35 mutants were constructed, but we failed to obtain a VP35 mutant that contains a mutation that disrupts the ability of the protein to antagonize interferon signaling but still allows IB formation. Instead, coexpression of “NP+VP35+VP30+L”, which induces IBs formation, inhibited IFN-I more potently than the expression of VP35 alone (Figure 7B). IRF3 knockout inhibited poly(I:C)-induced IFN-I production but had little, if any, effect on poly(I:C)-induced IFN-I production in the “NP+VP35+VP30+L” group (Figure 7C). IRF3 knockout in the cells did not significantly affect viral replication, but overexpression of activated IRF3 (IRF3/5D), instead of wild-type IRF3, inhibited viral replication (new Figure 7G-H). These results collectively suggested that almost all IRF3 in cells was hijacked and sequestered into IBs in the Ebola virus-infected cells.

Author Response

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

RESPONSE TO REVIEWERS:

Reviewer #1 (Recommendations For The Authors):

I think the manuscript of this excellent work can be improved, especially in writing (including a suggestion in the title) and presentation (Figure 6); Also some additional specific experiments and analyses could be important, as I suggest below,

1. For the title, perhaps a shorter "The acetylase activity of Cdu1 protects Chlamydia effectors from degradation" would be better to convey the major significance of this work. Of course, Cdu1 must regulate the function of InaC, IpaM and CTL0480. But perhaps it is speculative to think that egress is the major function of these effectors as their activity on other host cell processes during the cycle could eventually impact the extrusion process indirectly.

Although we concur with the insights provided by reviewer 1, we wish to underscore that a significant breakthrough presented in our study revolves around the regulation of Chlamydia exit by Cdu1. Consequently, we believe that this noteworthy discovery should be incorporated into the title.

1. For the writing:

a. The description of ubiquitination and DUBs could be synthesized to the essential, so that space is gained to explain things that then come a bit out of the blue in the results (what are Incs, the specific functions of InaC, IpaM, and CTL0480 - at least place the citations in lines 110-112 next to the corresponding Incs -, Cdu2, etc - see specifics below)

In lines 182-196 of the revised manuscript, we have incorporated additional contextual information concerning the roles of Incs, along with descriptions of the functions of InaC, IpaM, and CTL0480.

b. In the Results, there is a lot of Chlamydia- and maybe lab-specific jargon that could be significantly simplified for the more general reader. I detail some suggestions below in the specific issues.

We have improved the readability of our manuscript for a general audience by removing Chlamydia-specific terminology from the entire text and figures.

1. For the figures:

a. Figure 6, this figure could be reorganized: why two graphs in panel D? If detailed quantifications were done, perhaps in panel B just zoom on the examples of Golgi distributed/compacted? And again the labelling Rif-R L2, L2 pBOMB, M407 p2TK2, etc, simplify?

Figure 6 has undergone restructuring. The representative images have been relocated to Supplemental Figures 5 and 6, while we have introduced sample images demonstrating F-actin assembly and Golgi repositioning. Furthermore, the quantification of Golgi dispersal has been streamlined into a single panel. Additionally, we have simplified the labeling of the strains utilized in the study.

b. Figure 3, in the labelling, WT, inaC null, cdu1::GII wouldn't be enough? Leave the details to the legend and/or M&M.

We have simplified the labeling of Ct strains in Figure 3.

c. Figure 3C, these arrowheads should not be so symmetric (small arrows instead?) and it is unclear that the indicated cells do not show CTL0480.

We have substituted arrowheads with small arrow symbols and have also revised the Figure to incorporate a new representative image that prominently illustrates the absence of CTL0480 at the inclusion membrane of some cdu1::GII inclusions within infected Hela cells at 36 hpi.

1. Experiments:

a. In Figure 7, at least extrusion should be analysed also with the Cdu1-deficient strain expressing Ac-deficient Cdu1 and the inaC and ipaM phenotypes should be complemented.

We have conducted additional experiments to analyze extrusion production in Hela cells infected with a cdu1 null strain expressing the acetylase-deficient Cdu1 variant. We have incorporated the relevant data into revised Figure 7, where the impact of this strain on extrusion production and size is presented. Additionally, we updated Supplemental Figure 8 to include data illustrating the number of inclusions produced by this strain. We have also addressed these new results in the revised manuscript (lines 424-432). We are currently complementing inaC and ipaM mutant strains with various InaC and IpaM constructs that will be used in a follow up manuscript.

b. Does overexpression of InaC, IpaM, or CTL0480 in a cdu1-null background prevent the degradation of these Incs and suppress the defects of cells infected by the cdu1 mutant (F-actin, Golgi, MYPT1)? This would show that the multiple phenotypes displayed by cells infected by the cdu1 null mutant are indeed related to the decreased levels of InaC, IpaM and CTL0480.

We opted not to include data from the overexpression of these effectors in a cdu1-null background due to an unexpected decrease in shuttle plasmid load during overexpression. This development prompted concerns regarding the potential detrimental effects of overexpressing these effectors in the absence of Cdu1. Data supporting this observation are not included in this report.

c. Figures 3A and 3B should be quantified (it says it is from 3 independent experiments). It would be important to have a relative perspective of how much Cdu1 protects these Incs over time (for InaC, it would also be nice to have the 36 and 48 hpi time-point). This is in contrast with the microscopy data in Figure 5, which illustrates very clear effects, and the quantification is a bit redundant.

In Figure 3, we have incorporated a new Western Blot image showing endogenous InaC protein levels in Hela cells following infection with both WT Ct and cdu1::GII strains at 24, 36, and 48 hours post-infection (hpi). Additionally, we have quantified the Western Blot signals for both InaC and IpaM, and these results are also presented in Figure 3. The quantification of MYPT1 recruitment has been relocated to a supplementary figure. We have also included details regarding the methodology employed for the quantification of Western Blot signals in the Materials and Methods section.

d. What is the subcellular localization of InaC, IpaM, CTL0480 and Cdu1 when analysed by transfection? Does Cdu1 bind to of InaC, IpaM, CTL0480 in infected cells? If this was attempted and unsuccessful it should be mentioned.

In transfected HEK cells, InaC, IpaM, CTL0480, and Cdu1 all exhibit cytoplasmic localization with a diffuse pattern (data not shown). Despite our efforts, we encountered challenges in observing co-immunoprecipitation of Cdu1 with all three Incs in infected Hela cells at 24 hpi, We have duly acknowledged this limitation in our findings, as reflected in line 221-226 of the revised manuscript.

1. Specific issues:

2. Line 87, "propagule" is really needed to describe the EB?

The EB is the infectious form of Chlamydia species that spreads within the host to renew its life cycle; thus, "propagule" is a suitable term to characterize the EB.

• Exocytosis implies fusion with the plasma membrane so "inclusion is exocytosed" (line 91) is not entirely correct.

In line 91 of the revised manuscript, we referred to extrusion as the exit of an intact inclusion from the host cell and omitted the use of "exocytosed" to describe this process.

• Line 126, "a Ct L2 (LGV L2 434 Bu) background". Maybe "a Ct cdu1-null strain" would be enough and leave the detail for Materials and Methods.

In line 128 of the revised manuscript, we omitted "(LGV L2 434 Bu)" to avoid using jargon that may be unfamiliar to readers not well-versed in Chlamydia terminology.

• Line 138, in the previous Pruneda et al, Nature Microbiol 2018, the title of figure 4 is "ChlaDUB deubiquitinase activity is required for C. trachomatis Golgi fragmentation", so why raise this hypothesis? And why in the end is the acetylation activity of Cdu1 that promotes Golgi distribution? I think this related with infection vs transfection experiments but it deserved to be briefly explained/discussed.

In lines 140-142 of the revised manuscript, we provide clarification that the DUB activity of Cdu1 is required for Golgi fragmentation in transfected cells. This observation supports our initial hypothesis suggesting that the DUB activity of Cdu1 is also required for Golgi distribution in infected cells, and our rationale for identifying targets of its DUB activity.

• Lines 147-155, what is the relevance of this non-ubiquitinated proteins that come along? Couldn't this be synthesized?

We have included a discussion on non-ubiquitinated proteins, as they could potentially encompass proteins that interact with those protected by Cdu1. This perspective provides supplementary insights into the roles of proteins targeted for ubiquitination in the absence of Cdu1. The results of this analysis have been succinctly summarized in a single paragraph within the initial manuscript (lines 151-159 of the revised manuscript).

• Line 170, I think it is the first time that "Type 3 secretion"; perhaps explain in the introduction.

Type 3 secretion systems have been extensively characterized and discussed in the literature, and we anticipate that the majority of our readers are well-acquainted with this secretory mechanism.

• Line 184, I think it is the first time "microdomains" are mentioned; perhaps mention in the introduction.

The definition of "microdomains" has been provided in line 191 of the revised manuscript.

• Figure 2, as it stands the analysis with truncated Cdu1 proteins adds little to the work. Binding to the Incs seems to be affected when the TM domain is not present, but it still binds. And this is in a transfection context.

The results depicted in Figure 2, involving truncated Cdu1 proteins, illustrates that Cdu1 is capable of interacting with InaC, IpaM, and CTL0480 even in the absence of infection. This finding serves as evidence suggesting that all three Incs could potentially serve as direct targets for Cdu1 activity. As a result, we prefer to keep these findings in the manuscript.

• Line 219, "late stages of infection", this is shown (albeit not completely quantified) for IpaM and CTL0480, but not for InaC.

In the revised Figure 3, we show InaC protein levels at 24, 36, and 48 hours post-infection, and we have incorporated quantitative data for both InaC and IpaM protein levels in the context of Hela cells infected with both WT L2 and cdu1::GII strains. This updated figure serves to emphasize the pivotal role of Cdu1 in safeguarding all three Incs during the late stages of infection.

• Line 233, "pBOMB-MCI backbone" - is this needed in the Results section? And this refers to Figure 4 while pBOMB appear already in Fig. 3.

We have removed “pBOMB-MCI backbone” in the revised manuscript.

• Line 236, should be cdu1 endogenous promoter.

In line 265 of the revised manuscript we have replaced Cdu1 with cdu1 (italicized).

• Line 263, WT.

In line 293 of the revised manuscript we replaced “wild type” with “WT”.

• Line 277, IncA instead of "the Inc protein IncA".

In the manuscript we wanted to emphasize that IncA is also an inclusion membrane protein, therefore we have included “the Inc protein IncA” in the revised manuscript to avoid any confusion.

• How does the data in Figure 5 relates to the relatively few proteins ubiquitinated in cells infected with cdu1-mutant Ct? These Ub-labelling corresponds to ubiquitinated InaC, IpaM and CTL0480?

The findings presented in Figure 5 demonstrate that the acetylase activity of Cdu1 plays a crucial role in enabling Ct to block all ubiquitination events taking place on or in proximity to the periphery of the inclusion membrane. This encompasses Cdu1 targets that might not have been identified through our proteomic analysis.

• Lines 299-301, "M923 inclusions", there is certainly a clear way to write this.

In lines 326-327 and 332-332 of the revised manuscript, we have clarified that “M923” is an incA null strain to provide clarification.

• Line 309, is "peripheries" correct?

We have changed “peripheries” with “periphery” in the revised manuscript (line 360).

• Line 312, "Rif-R L2" and "M407" - can this be simplified?

In the revised manuscript, "Rif-R L2" was substituted with "WT L2" in lines 363 and 382, while "M407" was exchanged with "an inaC null strain" in lines 311, 367, and 368. These same replacements were applied to the Figures and their corresponding legends for consistency.

• Lines 308-321, and 326-335, these % are all approximate figures and this should be made clear.

In lines 364-395 of the revised manuscript we have stated that all percentages are approximate values.

• Fig. S1, kb and not k.b; what's the "+ control"; and is not really possible to have a PCR that works for the *? 3 kb is not that long.

In the updated Figure S1, we have corrected "k.b" to "kb". In the legend of Figure S1, we have clarified that the + control corresponds to the cdu2 locus. Moreover, we could not cleanly amplify a 3 kb PCR product from bacteria in whole cell lysates of infected mammalian cells (Vero cells).

• Fig. S2, kb and not k.b, bp and not b.p

In the updated Figure S2, we have corrected “k.b” with “kb” and “b.p” with “bp”.

Reviewer #2 (Recommendations For The Authors):

Figure 1 describes an affinity-based purification and mass spectrometric identification of differentially ubiquitinated proteins (host and chlamydial). Through different permutations of combinations of infection (mock, wild type, and Cdu1 mutant), three effectors, IpaM, InaC, and CTL0480, were identified as putative targets of Cdu1. The authors used a high-stringency cutoff, which could explain identification of only three targets. Having said this, the localization of Cdu1 to the inclusion membrane would be expected to also narrow down the number of targets. Interestingly, Cdu2, another deubiquitinase remained active in these experiments, which could have affected identification of Cdu1 targets. The authors addressed this issue by referring to previously reported structural studies. A somewhat glaring omission is the lack of reference to NF-kB as a substrate of ChlaDub1/Cdu1. In experiments by Le Negrate et al., ChlaDub1 ectopic overexpression in cells led to the deubiquitination of IkB-alpha, thus inhibiting the nuclear translation of NF-kB. Based on the inclusion membrane localization of Cdu1 during infection, is the identification of IkB an artifact of overexpression of Cdu1, or is it still a bona fide Cdu1 target?

We conducted experiments using our cdu1 null strain to investigate whether IκBα could be a target of Cdu1 activity. While our findings are intriguing and relevant, it is not feasible to determine, at this stage, whether our findings result from a direct or indirect consequence of Cdu1 localizing to the inclusion membrane. Consequently, these findings extend beyond the scope of the current manuscript. We plan to explore the implications of our observations more deeply in a subsequent manuscript, where we intend to provide a more comprehensive and mechanistic analysis based on these preliminary findings. Additionally, we have referenced the potential targeting of IκBα by Cdu1 in lines 100-101 and 166-171 of the revised manuscript.

Figure 2 demonstrates the individual interaction of the identified effectors with Cdu1. Interaction at the inclusion membrane is inferred from colocalization studies, while protein-protein interaction is monitored using ectopic overexpression of tagged versions of Cdu1 and the individual effectors. This is somewhat of a weakness of the manuscript because the mechanism of action of Cdu1 towards its target hinges on protein-protein interaction.

Despite our efforts, we encountered challenges in co-immunoprecipitating endogenous Cdu1 with all three Incs in infected Hela cells at 24 hpi. There are multiple technical reasons as to why these interactions, which are predicted to be transient, will not be captured by bulk affinity approaches such as immunoprecipitations, especially when the starting materials are present in very low abundance. We acknowledged these limitations in our findings, as reflected in lines 221-226 of the revised manuscript.

Figure 3 provides the first evidence in this paper of the importance of the inferred interaction of Cdu1 with the three effectors. The authors show that the loss of cdu1 has stability consequences on the three effectors. This figure would benefit from quantifying InaC- or IpaM-positive inclusions in the same manner done with CTL0480. The timepoint-dependent effect of Cdu1 loss of function is intriguing. Do InaC and IpaM retention at the inclusion show the same timepoint-dependent characteristic?

In the revised Figure 3, we have incorporated InaC protein levels at 24, 36, and 48 hours post-infection. Additionally, we have included quantitative data representing both InaC and IpaM protein levels in HeLa cells infected with both WT L2 and cdu1::GII strains. The quantification of CTL0480 localization to cdu1::GII inclusions has been moved to a supplementary figure.

This updated figure illustrates that the absence of Cdu1 has a time-dependent impact on both InaC and IpaM. However, it is noteworthy that the kinetics of degradation for these two proteins diverge significantly.

For Figure 7, the authors should consider monitoring timing of inclusion extrusion to gain additional insight into the functional interactions between the effectors. For example, the loss of CTL0480 leads to increased extrusion, implying a role in delaying or suppressing extrusion. In a time-course experiment, a CTL0480 mutant could exhibit an earlier occurrence of inclusion extrusion.

One of the principal discoveries of this study is that Cdu1, InaC, IpaM, and CTL0480 collaborate to facilitate optimal extrusion of Ct from host cells. These findings represent a significant contribution to our understanding of how Chlamydia controls its exit from infected cells. We are currently in the process of expanding on these results. A forthcoming follow-up manuscript will provide more detailed and comprehensive exploration of these findings.

Reviewer #3 (Recommendations For The Authors):

Specific comments.

a. I have some concerns related to the time point chosen for mass spec analysis and potential caveats and alternative interpretations. This work was done relatively early (24 hours) compared to the most convincing Cdu1 functions that occur later, thus this may limit the authors global understanding of protein changes. For example, the known substrate of Cdu1, Mcl-1 was not identified but this is altered relatively late during infection. Thus, the surprise that minimal host proteins are altered in ubiquitination may be partially driven by the timing of the assay. This should be more clearly discussed as a caveat.

In the revised manuscript (lines 166-171), we have acknowledged that there might be additional targets of Cdu1 that remain unidentified, primarily due to the specific time point we utilized in our study.

b. Another caveat to these studies is while the loss of Cdu1 alters different effectors stability and function and extrusion size, these changes do not modulate bacterial growth in cells. The authors speculate that regulating extrusion size may alter interactions with innate cells to drive dissemination. However, a previous study found defects in an animal model using a Cdu1 transposon mutant found decreased bacterial load in the genital tract. It is also possible that redundancy of effectors may mask importance in growth of Cdu1, but the authors strongly argue against redundancy of Cdu1 and Cdu2 so this weakens the authors argument here. These concepts and published data should be more directly discussed in the context of the authors proposed extrusion model and the role in driving Chlamydia growth and pathogenesis.

In our revised manuscript (lines 460-466) we propose that while we do not observe any growth impairments during Ct growth in the absence of Cdu1 in HeLa cells, the reduction in bacterial loads observed in murine models of infection with an independent cdu1 mutant strain (cdu1::Tn) may potentially be linked to defects in extrusion production or alterations in Cdu1-dependent regulation of extrusion size.

c. Recent studies have found that IFNg activation can result in dramatic changes in ubiquitination to pathogen containing vacuoles. While some of these are blocked by the newly found GarD, it seems possible that Cdu1 may also play a role (and perhaps use its deubiquinating activity) to further protect the inclusion. In light of published results showing that Cdu1 mutants have lower IFU burst size only in IFNg activated cells, this may be an important caveat in the current studies. This should be more directly addressed in the current manuscript.

We have incorporated two experimental findings indicating that the presence of Cdu1 is not required for Ct to defend itself against IFN cellular immunity in human cells. These recent discoveries are now presented in the updated Figure 5 and detailed in lines 338-355 of the revised manuscript.

d. On lines 433-434 the authors claim that Cdu1 is atypical since it is not encoded with the metaeffector/target pairs. However, this is an oversimplification of what is known about metaeffectors. For example, there are meta-effector/effector pairs that are not encoded together in Legionella (see table 1 DOI: https://doi.org/10.3390/pathogens10020108). Thus, the discussion should be adjusted. It seems Cdu1 is the first meta-effector found in Chlamydia, and maybe this should be highlighted more strongly rather than its uniqueness in this aspect of meta-effector/effector functions.

In lines 488-489 of the revised manuscript, we have removed the assertion that Cdu1 functions as an atypical metaeffector and emphasized that it represents the initial discovery of a metaeffector within Ct.

Author Response

eLife assessment

This important work describes the first high-resolution structure of HGSNAT, a lysosomal membrane protein required for the degradation of heparan sulfate (HS). Through careful structural analysis, this work proposes potential reasons why certain mutations in HGSNAT lead to lysosomal storage disorders and outlines the enzyme's catalytic mechanism. The experimental evidence presented provides incomplete support for the proposed molecular mechanism of the HS acetylation reaction and the impact of disease-causing mutations.

We thank the editors and reviewers for taking the time to provide a critical assessment of our manuscript. We appreciate the input and suggestions to improve the analysis. Included here are only our provisional responses. We will address the concerns raised in more detail and incorporate them in the revised version of the manuscript.

Reviewer #1 (Public Review):

This article by Navratna et al. reports the first structure of human HGSNAT in an acetyl-CoAbound state. Through careful structural analysis, the authors propose potential reasons why certain human mutations lead to lysosomal storage disorders and outline a catalytic mechanism. The structural data are of good quality, and the manuscript is clearly written. This study represents an important step toward understanding the mechanism of HGSNAT and is valuable to the field. I have the following suggestions:

We thank the reviewer for their encouraging and positive overall assessment of our work.

1. The authors should characterize whether the purified protein is active. Otherwise, how does one know if the detergent used maintains the protein in a biologically relevant state? The authors should at least attempt to do so. If these prove to be challenging, at the very least, the authors should try a cell-based assay to demonstrate that the GFP tag does not interfere with the function.

Thank you for highlighting this concern. The cryo-EM sample was prepared without the exogenous addition of ligand, as noted in the manuscript; the acetyl-CoA that we see in the structure was intrinsically bound to the protein, indicating the ability of GFP-tagged HGSNAT protein to bind the ligand. We purified the protein at a pH optimal for acetyl-CoA binding, as suggested by Bame, K. J. and Rome, L. H. (1985) and Meikle, P. J. et al., (1995). Because we see acetyl-CoA in a structure obtained using a GFP fusion, we argue that GFP does not interfere with protein stability and ability to bind to the co-substrate. As demonstrated by existing literature HGSNAT catalyzed reaction is compartmentalized spatially and conditionally. The binding of acetyl-CoA happens towards the cytosol and is optimal at pH 7-0.8.0, while the transfer of the acetyl group to heparan sulfate occurs towards the luminal side and is optimal at pH 5.0-6.0. We are working on establishing a robust assay to study this complicated and compartmentalized acetyl transfer assay.

1. In Figure 5, the authors present a detailed schematic of the catalytic cycle, which I find to be too speculative. There is no evidence to suggest that this enzyme undergoes isomerization, like a transporter, between open-to-lumen and open-to-cytosol states. Could it not simply involve some movements of side chains to complete the acetyl transfer?

The acetyl-CoA bound structure presented in the paper does not conclusively support a potential for isomerization and conformational dynamics. We agree with the reviewer that the reaction schematic presented in Figure 5 is speculative. We acknowledge in the discussion that our structure represents only a single step of the reaction, and defining the precise mechanism of acetyl transfer needs additional work. However, we will reword the discussion and change Figure 5 to address this concern raised by multiple reviewers.

Reviewer #2 (Public Review):

Summary:

This work describes the structure of Heparan-alpha-glucosaminide N-acetyltransferase (HGSNAT), a lysosomal membrane protein that catalyzes the acetylation reaction of the terminal alpha-D-glucosamine group required for the degradation of heparan sulfate (HS). HS degradation takes place during the degradation of the extracellular matrix, a process required for restructuring tissue architecture, regulation of cellular function, and differentiation. During this process, HS is degraded into monosaccharides and free sulfate in lysosomes.

HGSNAT catalyzes the transfer of the acetyl group from acetyl-CoA to the terminal non-reducing amino group of alpha-D-glucosamine. The molecular mechanism by which this process occurs has not been described so far. One of the main reasons to study the mechanism of HGSNAT is that multiple mutations spanning the entire sequence of the protein, such as nonsense mutations, splicesite variants, and missense mutations lead to dysfunction that causes abnormal accumulation of HS within the lysosomes. This accumulation is a cause of mucopolysaccharidosis IIIC (MPS IIIC), an autosomal recessive neurodegenerative lysosomal storage disorder, for which there are no approved drugs or treatment strategies.

This paper provides a 3.26A structure of HGSNAT, determined by single-particle cryo-EM. The structure reveals that HGSNAT is a dimer in detergent micelles and a density assigned to acetylCoA. The authors speculate about the molecular mechanism of the acetylation reaction, map the mutations known to cause MPS IIIC on the structure and speculate about the nature of the HGSNAT disfunction caused by such mutations.

Strengths:

The description of the architecture of HGSNAT is the highlight of the paper since this corresponds to the first description of the structure of a member of the transmembrane acyl transferase (TmAT) superfamily. The high resolution of an HGSNAT bound to acetyl-CoA is an important leap in our understanding of the HGSNAT mechanism. The density map is of high quality, except for the luminal domain. The location of the acetyl-CoA allows speculation about the mechanistic role of multiple residues surrounding this molecule. The authors thoroughly describe the architecture of HGSNAT and map the mutations leading to MPS IIIC. The description of the dimeric interphase is a novel result, and future studies are left to confirm the importance of oligomerization for function.

We thank the reviewer for their time and for highlighting both the quality and novelty of the structure presented in this work.

Weaknesses:

Apart from the cryo-EM structure, the article does not provide any other experimental evidence to support or explain a molecular mechanism. Due to the complete absence of functional assays, mutagenesis analysis, or other structures such as a ternary complex or an acetylated enzyme intermediate, the mechanistic model depicted in Figure 5 should be taken with caution.

Thank you for pointing out this concern. The proposed mechanistic model in Figure 5 is a hypothesis based on previously reported biochemical characterization of HGSNAT by Rome & Crain (1981), Rome et al, (1983), Miekle et al., (1995) and Fan et al., (2011). However, we agree with the reviewer that this schematic is not experimentally proven and is speculative at best. Especially because our structure presents only a single step of the reaction, which does not conclusively support either ping-pong or random-order bi-substrate reactions. We will rephrase this section of our discussion and edit Figure 5 to address this concern.

The authors discuss that H269 is an essential residue that participates in the acetylation reaction, possibly becoming acetylated during the process. However, there is no solid experimental evidence, e.g. mutagenesis analysis or structural analysis, in this or previous articles, that demonstrates this to be the case.

H269, as a crucial catalytic residue, was suggested by monitoring the effect of chemical modifications of amino acids on acetylation of HGSNAT membranes by Bame, K. J. and Rome, L. H. (1986). We agree that mutagenesis, catalysis, and structural evidence for the same are not currently available. We are pursuing a more thorough exploration of the role of both H269 (previous studies) and N258 (from this study) on the stability and function of HGSNAT.

In the discussion part, the authors mention previous studies in which it was postulated that the catalytic reaction can be described by a random order mechanistic model or a Ping Pong Bi Bi model. However, the authors leave open the question of which of these mechanisms best describes the acetylation reaction. The structure presented here does not provide evidence that could support one mechanism or the other.

We agree with the reviewer’s observation that the structure doesn’t indeed support one reaction mechanism or another. We are pursuing the structural and kinetic characterization of HGSNAT in the presence of other co-substrates and multiple pHs that are required to address this concern thoroughly.

Although the authors map the mutations leading to MPS IIIC on the structure and use FoldX software to predict the impact of these mutations on folding and fold stability, there is no experimental evidence to support FoldX's predictions.

We are working on assessing the impact of specific mutations on the stability of HGSNAT and will add them to the revised version of the manuscript. We thank the reviewer for this suggestion.

Reviewer #3 (Public Review):

Summary:

Navratna et al. have solved the first structure of a transmembrane N-acetyltransferase (TNAT), resolving the architecture of human heparan-alpha-glucosaminide N-acetyltransferase (HGSNAT) in the acetyl-CoA bound state using single particle cryo-electron microscopy (cryoEM). They show that the protein is a dimer and define the architecture of the alpha- and beta- GSNAT fragments, as well as convincingly characterizing the binding site of acetyl-CoA.

Strengths:

This is the first structure of any member of the transmembrane acyl transferase superfamily, and as such it provides important insights into the architecture and acetyl-CoA binding site of this class of enzymes.

The structural data is of a high quality, with an isotropic cryoEM density map at 3.3Å facilitating the building of a high-confidence atomic model. Importantly, the density of the acetyl-CoA ligand is particularly well-defined, as are the contacting residues within the transmembrane domain.

The open-to-lumen structure of HSGNAT presented here will undoubtedly lay the groundwork for future structural and functional characterization of the reaction cycle of this class of enzymes.

We thank the reviewer for their positive assessment of the data presented in this work. We really appreciate and agree with the reviewer's comment that the “structure of HSGNAT presented here will undoubtedly lay the groundwork for future structural and functional studies.”

Weaknesses:

While the structural data for the open-to-lumen state presented in this work is very convincing, and clearly defines the binding site of acetyl-CoA, to get a complete picture of the enzymatic mechanism of this family, additional structures of other states will be required.

We agree with the reviewers’ assessment and are heavily invested in pursuing the structures of all the steps of acetyl transfer by HGSNAT.

A potentially significant weakness of the study is the lack of functional validation. The enzymatic activity of the enzyme characterized was not measured, and the enzyme lacks native proteolytic processing, so it is a little unclear whether the structure represents an active enzyme.

We thank the reviewer for this comment. While the proteolytic cleavage of the protein remains debated, we find no evidence of such an event in our purification (SDS-PAGE and SEC). Studies like Durand et al., (2010) and Fan et al., (2011) suggest that even the ER retained monomeric HGSNAT is active. Because we see acetyl-CoA (co-substrate) bound to the protein in our structure, we surmise that proteolysis is not necessary for function, at least not for substrate binding. However, we are working towards the structural and kinetic characterization of recombinant α- and β-HGSNAT construct to explore the role of proteolysis on HGSNAT stability and function.

Author Response

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

eLife assessment

This valuable paper examines the Bithorax complex in several butterfly species, in which the complex is contiguous and not split, as it is in the well-studied fruit fly Drosophila. Based on genetic screens and genetic manipulations of a boundary element involved in segment-specific regulation of Ubx, the authors provide solid evidence for their conclusions, which could be further strengthened by additional data and analyses. The data presented are relevant for those interested in the evolution and function of Hox genes and of gene regulation in general.

We are deeply grateful to the eLife editorial team and the two reviewers for their thoughtful and constructive feedback. We have used this feedback to improve our manuscript and have provided a point-by-point response below.

Public Reviews:

Reviewer #1 (Public Review):

In their article, "Cis-regulatory modes of Ultrabithorax inactivation in butterfly forewings," Tendolkar and colleagues explore Ubx regulation in butterflies. The authors investigated how Ubx expression is restricted to the hindwing in butterflies through a series of genomic analyses and genetic perturbations. The authors provide evidence that a Topologically Associated Domain (TAD) maintains a hindwing-enriched profile of chromatin around Ubx, largely through an apparent boundary element. CRISPR mutations of this boundary element led to ectopic Ubx expression in forewings, resulting in homeotic transformation in the wings. The authors also explore the results of the mutation in two non-coding RNA regions as well as a possible enhancer module. Each of these induces homeotic phenotypes. Finally, the authors describe a number of homeotic phenotypes in butterflies, which they relate to their work.

Together, this was an interesting paper with compelling initial data. That said, I have several items that I feel would warrant further discussion, presentation, or data.

First, I would not state, "Little is known about how Hox genes are regulated outside of flies." They should add "in insects" since so much in known in vertebrates

Corrected

For Figure 1, it would aid the readers if the authors could show the number of RNAseq reads across the locus. This would allow the readership to evaluate the frequency of the lncRNAs, splice variants, etc.

We have found it useful in the past to feature “Sashimi Plots”, as they provide a good overview of transcript splicing junctions and read support. Here we could not accommodate this in our Fig. 1A as this would require compiling the RNAseq reads from many tissues and stages to be meaningful, and we would lose the resolution on forewing vs hindwing tissues that is important in this article (only the Kallima inachus dataset allows this comparison, and was used in Fig 1B). More specifically, the wing transcriptomes available for J. coenia and V. cardui are not deep enough to provide a good visualization of Antp alternative promoter usage or on AS5’ transcription.

How common are boundary elements within introns? Typically, boundary elements are outside gene bodies, so this could be explored further. This seems like an interesting bit of biology which, following from the above point, it would be interesting to, at a minimum, discuss, but also relate to how transcription occurs through a possible boundary element (are there splice variants, for example?).

We do not see evidence of alternative splicing, and prefer to avoid speculating on transcriptional effects, but we agree that the intragenicity of the TAD boundary is interesting. We briefly highlighted this point in the revised Discussion:

"Lastly, it is worth noting that the Antp/Ubx TAD boundary we identified is intragenic, within the last intron of Ubx. It is unclear if this feature affects Ubx transcription, but this configuration might be analogue to the Notch locus in Drosophila, which includes a functional TAD boundary in an intronic position (Arzate-Mejía et al. 2020)."

The CRISPR experiments led to compelling phenotypes. However, as a Drosophila biologist, I found it hard to interpret the data from mosaic experiments. For example, in control experiments, how often do butterflies die? Are there offsite effects? It's striking that single-guide RNAs led to such strong effects. Is this common outside of this system? Is it possible to explore the function effects at the boundary element - are these generating large deletions (for example, like Mazo-Vargas et al., 2022)? For the mosaic experiments, how frequent are these effects in nature or captive stocks? Would it be possible to resequence these types of effects? At the moment, this data, while compelling, was hard to put into the context of the experiments above without understanding how common the effects are. Ideally, there would be resequencing of these tissues, which could be targeted, but it was not clear to me the general rates of these variants.

We agree with this assessment completely: mosaics complicate the proper interpretation of CRISPR based perturbation assays in regulatory regions. Here, unlike in Mazo-Vargas et al. (2022), we were unable to breed homeotic effects to a G1 generation, possibly because the phenotypes are dominant and lethal at the embryonic stage (see also our reply to Reviewer 2). This means that mosaic mutants are often survivors with clones of restricted size in the wing, and they are probably rare, but we are unable to meaningfully measure a mutation spectrum frequency (e.g. how often large deletions are generated). As mentioned in the first paragraph of our Discussion, we think that many of the phenotypes we observed (besides the Ubx GOF effects from the BE targeting) were confounded by alleles that could include large SVs. We aim to address these questions in an upcoming manuscript, at a locus where regulatory perturbation does not impact survival, including using germline mutants and unbiased genotyping (whole genome resequencing).

We elaborated on this issue in our Discussion:

"It is crucial here to highlight the limitations of the method, in order to derive proper insights about the functionality of the regulatory regions we tested. In essence, butterfly CRISPR experiments generate random mutations by non-homologous end joining repair, that are usually deletions (Connahs et al. 2019; Mazo-Vargas et al. 2022; Van Belleghem et al. 2023). Ideally, regulatory CRISPR-induced alleles require genotyping in a second (G1) generation to be properly matched to a phenotype (Mazo-Vargas et al. 2022). Possibly because of lethal effects, we failed to pass G0 mutations to a G1 generation for genotyping, and were thus limited here to mosaic analysis. As adult wings have lost scale building cells that may underlie a given phenotype, we circumvented this issue by genotyping a pupal forewing displaying an homeotic phenotype in the more efficient Antp-Ubx_BE perturbation experiment (Fig. S4). In this case, PCR amplification of a 600 bp fragment followed by Sanger sequencing recovered signatures of indel variants, with mixed chromatograms starting at the targeted sites. But in all other experiments (CRM11, IT1, and AS5’ targets), we did not genotype mutant tissues, as they were only detected in adult stages and generally with small clone sizes. Some of these clones may have been the results of large structural variants, as data from other organisms suggests that Cas9 nuclease targeting can generate larger than expected mutations that evade common genotyping techniques (Shin et al. 2017; Adikusuma et al. 2018; Kosicki et al. 2018; Cullot et al. 2019; Owens et al. 2019). Even under the assumption that such mutations are relatively rare in butterfly embryos, the fact we injected >100 embryos in each experiment makes their occurrence likely (Fig. 9), and we are unable to assign a specific genotype to the homeotic effects we obtained in CRM11, IT1 and AS5’ perturbation assays."

Our revision also includes a new Fig. S4 that features the mosaic genotyping of a G0 Antp-Ubx_BE mutant tissue. While this does not fully address the reviewer questions, it provides reasonable validation that the frequent GOF effects we observed upon perturbation at this target site are generated by on-target indels from DNA repair.

Author response image 1.

Validation of CRISPR-induced DNA Lesions in an Antp-Ubx_BE crispant pupat forewing. (A-A') Pupal forewing cuticle phenotype of an Antp-Ubx_BE J. coenia crispant, as in Fig. S3. (B-B") Aspect of the same forewing under trans-illumination following dissection out of the pupal case. Regions from mutant clones have a more transparent appearance. (C). Sanger sequencing of an amplicon targeting the Antp-Ubx_BE region in the mutant tissue shown in panel B", compared to a control wing tissue, showing mixed chromatogram around the expected CRISPR cutting site due to indel mutations from non-homologous end-joining.

In sum, I enjoyed the extensive mosaic perturbations. However, I feel that more molecular descriptions would elevate the work and make a larger impact on the field.

Reviewer #2 (Public Review):

Summary:

The existence of hox gene complexes conserved in animals with bilateral symmetry and in which the genes are arranged along the chromosome in the same order as the structures they specify along the anteroposterior axis of organisms is one of the most spectacular discoveries of recent developmental biology. In brief, homeotic mutations lead to the transformation of a given body segment of the fly into a copy of the next adjacent segment. For the sake of understanding the main observation of this work, it is important to know that in loss-of-function (LOF) alleles, a given segment develops like a copy of the segment immediately anterior to it, and in gain-of-function mutations (GOF), the affected segment develops like a copy of the immediately posterior segment. Over the last 30 years the molecular lesions associated with GOF alleles led to a model where the sequential activation of the hox genes along the chromosome result from the sequential opening of chromosomal domains. Most of these GOF alleles turned out to be deletions of boundary elements (BE) that define the extent of the segment-specific regulatory domains. The fruit fly Drosophila is a highly specialized insect with a very rapid mode of segmentation. Furthermore, the hox clusters in this lineage have split. Given these specificities it is legitimate to question whether the regulatory landscape of the BX-C we know of in D.melanogaster is the result of very high specialization in this lineage, or whether it reflects a more ancestral organization. In this article, the authors address this question by analyzing the continuous hox cluster in butterflies. They focus on the intergenic region between the Antennapedia and the Ubx gene, where the split occurred in D.melanogaster. Hi-C and ATAC-seq data suggest the existence of a boundary element between 2 Topologically-Associated-Domain (TAD) which is also characterized by the presence of CTCF binding sites. Butterflies have 2 pairs of wings originating from T2 (forewing) specified by Antp and T3 specified by Ubx (hindwing). Remarkably, CRISPR mutational perturbation of this boundary leads to the hatching of butterflies with homeotic clones of cells with hindwings identities in the forewing (a posteriorly oriented homeotic transformation). In agreement with this phenotype, the authors observe ectopic expression of Ubx in these clones of cells. In other words, CRISPR mutagenesis of this BE region identified by molecular tool give rise to homeotic transformations directed towards more posterior segment as the boundary mutations that had been 1st identified on the basis of their posterior oriented homeotic transformation in Drosophila. None of the mutant clones they observed affect the hindwing, indicating that their scheme did not affect the nearby Ubx transcription unit. This is reassuring and important first evidence that some of the regulatory paradigms that have been proposed in fruit flies are also at work in the common ancestor to Drosophilae and Lepidoptera.

Given the large size of the Ubx transcription unit and its associated regulatory regions it is not surprising that the authors have identified ncRNA that are conserved in 4 species of Nymphalinae butterflies, some of which also present in D.melanogaster. Attempts to target the promoters by CRISPR give rise to clones of cells in both forewings and hindwings, suggesting the generation of regulatory mutations associated with both LOF and GOF transformations. The presence of clones with dual homeosis suggests the targeting of Ubx activator and repression CRMs. Unfortunately, these experiments do not allow us to make further conclusions on the role of these ncRNA or in the identification of specific regulatory elements. To the opinion of this reviewer, some recent papers addressing the role that these ncRNA may play in boundary function should be taken with caution, and evidence that ncRNA(s) regulate boundaries in the BX-C in a WT context is still lacking.

Strengths:

The convincing GOF phenotype resulting from the targeting of the Antp-Ubx_BE.

Weaknesses:

The lack of comparisons with the equivalent phenotypes obtained in D.melanogaster with for example the Fub mutation.

We are grateful for this excellent contextualization of our findings and have incorporated some of the historical elements into our revision, as detailed below.

Reviewer #2 (Recommendations For The Authors):

In the whole paper, the authors bring the notion of boundaries through the angle of the existence of TADs and ignore almost entirely to explain the characteristics of boundary mutation in the BX-C. To my knowledge examples where targeted boundary deletions between TADs result in misregulation of the neighboring genes, and/or a phenotype, are extremely sparse (especially in the context of the mouse hox genes). Given the extensive litterature describing the boundary mutations and their associated GOF phenotypes, the paper would certainly gain strength if the authors justify their approach through this wealth of information. I must admit that this referee is surprised by the absence of any references to the founding work of the Karch and Bender laboratories on this topic. As a matter of fact, one of the founding members of the boundary class of regulatory elements was already brought in 1993 with the Fab-7 and Mcp elements of the BX-C. Based on gain-of-function homeotic phenotypes, additional Fab boundaries were added to the list. Finally, in 2013, Bender and Lucas (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3606092/) identified the Fub boundary element that delimits the Ubx and abd-A domains in the BX-C. Fub fulfills the criterium of lying at the border of 2 neighboring TADs. Significantly, a deletion of Fub leads to a very penetrant and strong homeotic gain-of-function phenotype in which the flies hatch with a 1st abdominal segment transformed into the 2nd. In agreement with this, abd-A is expressed one parasegment too anterior in embryos. This is exactly the observation gathered from the targeted mutations in the Antp-Ubx_BE; a dominant transformation of anterior to posterior wing accompanied by an ectopic expression of Ubx in the forming primordia of the forwing where it is normally silenced. I believe the paper would gain credibility if the results were reported with the knowledge of the similarities with Fub.

Line 53, I am not aware of the existence of TADs for each of the 9 regulatory domains. The insulators delimit the extent of the regulatory domains but certainly not of TADs.

We thank the reviewer for these suggestions, as well as for the correction – we agree our previous text suggested that all BX-C boundaries are TAD boundaries, which was incorrect. We added a new introduction paragraph that combines classic literature on GOF mutations at boundary elements with recent evidence these are TAD insulators, including Fub (as suggested), and adding Fab-7 for breadth of scope.

"For instance, the deletion of a small region situated between Ubx and abd-A produces the Front-ultraabdominal phenotype (Fub) where the first abdominal segment (A1) is transformed into a copy of the second abdominal segment A2, due to a gain-of-expression of abd-A in A1 where it is normally repressed (Bender and Lucas 2013). At the molecular level, the Fub boundary is enforced by insulating factors that separate Topologically Associating Domains (TADs) of open-chromatin, while also allowing interactions of Ubx and abd-A enhancers with their target promoters (Postika et al. 2018; Srinivasan and Mishra 2020). Likewise, the Fab-7 deletion, which removes a TAD boundary insulating abd-A and Abd–B (Moniot-Perron et al. 2023), transforms parasegment 11 into parasegment 12 due to an anterior gain-of-expression of Abd-B (Gyurkovics et al. 1990). By extrapolation, one may expect that if the Drosophila Hox locus was not dislocated into two complexes, Antp and Ubx 3D contact domains would be separated by a Boundary Element (BE), and that deletions similar with Fub and Fab-7 mutations would result in gain-of-function mutations of Ubx that could effectively transform T2 regions into T3 identities."

A reference to the 1978 Nature article of Lewis should be added after line 42 of introduction.

Added

Line 56-57; the BX-C encoded miRNAs are known to regulate Ubx and abd-A, but not Abd-B.

Corrected

From lines 57 to 61, the authors mention reports aimed at demonstrating a role of ncRNA into Ubx regulation. To my eyes, these gathered evidences are rather weak. A reference to the work of Pease et al in Genetics in 2013 should be mentioned (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3832271/).

Added. Our paragraph includes qualifier language about the functionality of the Ubx-related ncRNAs (“are thought to”, “appears to”), and updated references regarding bxd (Petruk et al. 2006; Ibragimov et al. 2023).

Line 62 authors, should write "Little is known about how Hox genes are regulated outside of Drosophila" and not flies.

Corrected

Lines 110-112 could lncRNA:Ubx-IT1 correspond to PS4 antisense reported by Pease et al in 2023 (see URL above)? Lines 115-117, could lncRNA:UbxAS5' correspond to bxd antisense of Pease et al in 2023 (see above)?

As we could not detect sequence similarities, we preferred to avoid drawing homology, and we intentionally avoided reference to the fly transcripts when we named IT1 and AS5’. This said, we agree it is important to clarify that further studies are needed to clarify this relationship. We elaborated on this point in our discussion:

"Of note, a systematic in-situ survey (Pease et al. 2013) showed that Drosophila embryos express an antisense transcripts in its 5’ region (lncRNA:bxd), as well as within its first intron (lncRNA:PS4). It is thought that Drosophila bxd regulates Ubx, possibly by transcriptional interference or by facilitation of the Fub-1 boundary effect (Petruk et al. 2006; Ibragimov et al. 2023), while the possible regulatory roles of PS4 remain debated (Hermann et al. 2022). While these dipteran non-coding transcripts lack detectable sequence similarity with the lepidopteran IT1 and AS5’ transcripts, further comparative genomics analyses of the Ubx region across the holometabolan insect phylogeny should clarify the extent to which Hox cluster lncRNAs have been conserved or independently evolved."

Lines 154-155: "This concordance between Hi-C profiling and CTCF motif prediction thus indicates that Antp-Ubx_BE region functions as an insulator between regulatory domains of Antp and Ubx ». This is only correlative, I would write "suggests" instead of "indicates" and add a "might function".

Corrected as suggested.

Line 254, I assume the authors wish to write Ubx-IT1 in V. cardui instead of Ubx-T1.

Typo corrected

Line 255 : Fig.5 is absent from the pdf file and replaced by table 1. I did not find a legend for Table 1.

Corrected, with our sincere apologies for the loss of this image in our first submission.

Line 293 "Individual with hindwing clones 2.75 times more common than...." "are" is missing?

Corrected

Lines 303-313, it is not entirely clear how many guide RNAs were injected. Would be useful to indicate the sites targeted in Fig.S8.

We specify in the revised text : using a single guide RNA (Ubx11b9)

Lines 323-337: it is not entirely clear to this referee (a drosophilist) if those spontaneous mutations can be inbred or whether these individuals are occasional mosaics. In general, did anyone try to derive lines from those mosaic animals? Is it possible to hit the germline at the syncitial stages at which the guides are injected? Are the individuals with wing phenotype fertile? Given the fact that the Antp-Ubx_BE mutations should be dominant, I wonder if this characteristic would not help in identifying germline transmission. Similar remark for the discussion where the authors explain at line 360, that genotyping can only be done in the progeny of the Go. I do not have the impression that the authors have performed this genotyping and if I am right, I do not understand why.

We improved our discussion section on this topic (new text in orange):

"It is crucial here to highlight the limitations of the method, in order to derive proper insights about the functionality of the regulatory regions we tested. In essence, butterfly CRISPR experiments generate random mutations by non-homologous end joining repair, that are usually deletions (Connahs et al. 2019; Mazo-Vargas et al. 2022; Van Belleghem et al. 2023). Ideally, regulatory CRISPR-induced alleles require genotyping in a second (G1) generation to be properly matched to a phenotype (Mazo-Vargas et al. 2022). Possibly because of lethal effects, we failed to pass G0 mutations to a G1 generation for genotyping, and were thus limited here to mosaic analysis. As adult wings have lost scale building cells that may underlie a given phenotype, we circumvented this issue by genotyping a pupal forewing displaying an homeotic phenotype in the more efficient Antp-Ubx_BE perturbation experiment (Fig. S4). In this case, PCR amplification of a 600 bp fragment followed by Sanger sequencing recovered signatures of indel variants, with mixed chromatograms starting at the targeted sites. But in all other experiments (CRM11, IT1, and AS5’ targets), we did not genotype mutant tissues, as they were only detected in adult stages and generally with small clone sizes. Some of these clones may have been the results of large structural variants, as data from other organisms suggests that Cas9 nuclease targeting can generate larger than expected mutations that evade common genotyping techniques (Shin et al. 2017; Adikusuma et al. 2018; Kosicki et al. 2018; Cullot et al. 2019; Owens et al. 2019). Even under the assumption that such mutations are relatively rare in butterfly embryos, the fact we injected >100 embryos in each experiment makes their occurrence likely (Fig. 9), and we are unable to assign a specific genotype to the homeotic effects we obtained in CRM11, IT1 and AS5’ perturbation assays."

We agree that the work we conducted with mosaics has important caveats. So far, our attempts at breeding homeotic G0 mutants have not been fruitful at this locus, while less deleterious loci can yield viable alleles into further generations, such as WntA (published) and cortex (in prep.). We prefer to stay vague about negative data here, as it is difficult to disentangle if they were due to real mutational effects (e.g. the alleles can be dominant and lethal in the G1 generation) to failure to germline carriers of mutations as founders, or to health issues that are often amplified by inbreeding depression (including a possible iflavirus in our V. cardui cultures).

We concur with the prediction that Antp-Ubx_BE mutations are probably dominant, and intend to follow up with similar GOF experiments in the Plodia pantry moth, a laboratory model for lepidopteran functional genomics that is more amenable than butterflies to inbreeding and long-term studies in mutant lines. In our experience (https://www.frontiersin.org/articles/10.3389/fevo.2021.643661/full), Ubx coding knock-out can be more extensive in Plodia than in butterflies, so we think these animals will also be more resilient to the deleterious effects of the GOF phenotype.

Line 423, 425, I am not a fan of the term "de-insulating!!!!!

We replaced this neologism by Similar deletion alleles resulting in a TAD fusion and misexpression effect (see below).

Line 425, why bring the work on Notch while there are so many examples in the BX-C itself....

Our revised sentence makes it more clear we are referring here to documented examples of deletion-mediated TAD fusion (ie. featuring a conformation capture assay such as HiC/micro-C):

This suggests a possible loss of the TAD boundary in the crispant clones, resulting in a TAD fusion or in a long-range interaction between a T2-specific enhancer and Ubx promoter. Similar deletion alleles resulting in a TAD fusion and misexpression effect have been described at the Notch locus in Drosophila (Arzate-Mejía et al. 2020), in digit-patterning mutants in mice and humans (Lupiáñez et al. 2015; Anania et al. 2022), or at murine and fly Hox loci depleted of CTCF-mediated regulatory blocking (Narendra et al. 2015; Gambetta and Furlong 2018; Kyrchanova et al. 2020).

Our revision also includes more emphasis on the Drosophila BX-C boundary elements Fub and Fab-7 (see above).

Author Response

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

Reviewer #1 (Recommendations For The Authors):

The manuscript is very well written, the data are clearly presented and the methodology is robust. I only have suggestions to improve the manuscript, to make the study more appealing or to discuss in more detail some questions raised by the work.

1. In the study as it stands, PFG seems to come out of the blue. The authors apparently selected this protein based on sequence conservation between species but this is unlikely to be sufficient to identify novel TFs. Explaining in more detail the reasoning that led to PFG would make the story more appealing. Perhaps PFG was identified through a large reverse genetics screening?

Response: Thank you for your suggestion. We identified this gene solely by the strategy we described in the manuscript. We decided on this strategy based on the findings of our previous study on AP2-Family TFs, whose DNA binding domains are highly conserved among Plasmodium orthologues. Using this screening strategy, we identified a novel AP2 family TF AP2-Z. The results of the present study demonstrated that this strategy is applicable to TFs other than those belonging to the AP2 family. We are aware that this strategy is not all-encompassing. In fact, we failed to identify HDP1 as a candidate TF when it was also in the target list of AP2-G. However, at present, this is our primary strategy for identifying novel TFs in the targetome.

1. The authors propose that PFG and AP2-FG form a complex, but this is actually not shown. Did they try to document a physical interaction between the two proteins, for example using co-IP?

Response: Even when the two molecules were identified to be at the same position by ChIPseq, it cannot be concluded that they form a physical complex because it is possible that they competitively occupy the region. However, in this study, we performed ChIP-seq in the absence of PFG and demonstrated that the cAP2-FG peaks disappeared while those of sAP2-FG remained. This result can only be explained by the two proteins forming a complex at this region, which excludes the possibility that AP2-FG binds the region independently.

1. It is unclear how PFG can bind to DNA in the absence of DNA-binding domain. Did the authors search for unconventional domains in the protein? This should be at least discussed in the manuscript.

Response: We speculate that the two highly conserved regions, region 1 and region 2, function as DNA-binding domains in PFG. However, this domain is not similar to any DNA binding domains reported thus far. A straightforward way to demonstrate this would be to perform in vitro binding assays using a recombinant protein. However, thus far, we have not succeeded in obtaining soluble recombinant proteins for these regions. We have added the following sentences to the results section.

“At present, we speculate that PFG directly interacts with genomic DNA through two highly conserved regions; region 1 and region 2. However, these regions are not similar to any DNA binding domains reported thus far. In other apicomplexan orthologues, these two domains are located adjacent to one another in the protein (Fig. 1A). Therefore, these two regions may be separated by a long interval region but constitute a DNA binding domain of PFG as a result of protein folding.”

1. How do the authors explain that PFG is still expressed in the absence of AP2-FG? Is AP2G alone sufficient to express sufficient levels of the protein? Is PFG down-regulated in the absence of AP2-FG?

Response: Our previous ChIP-seq data indicate that PFG is a target of AP2-G. According to the study by Kent et al. (2018), this gene is up-regulated in the early period following conditional AP2-G induction. The results of the present study showed that PFG is capable of autoactivation through a transcriptional positive feed-back loop. These results suggest that PFG can maintain its expression to a certain level once activated by AP2-G, even in the absence of AP2-FG. In our previous microarray analysis, significant decreases in PFG expression were not observed in AP2-FG-diaruptedparasites.

1. How do AP2-FG regulated genes (based on RNAseq) compare with the predicted cAP2FG/sAP2-FG predicted genes (based on ChIPseq)? Are the two subsets included in the genes that are actually down-regulated in AP2-FG(-)?

Response: Disruption of the AP2-FG gene impairs gametocyte development. We considered that the direct effect of this disruption would be difficult to analyze in gametocyte-enriched blood, in which gametocytes are pooled during sulfadiazine treatment to deplete asexual stages. Therefore, in our previous paper, we performed microarray analysis between WT and KO parasites to detect the direct effect of AP2-FG disruption on target gene expression, using mice which were synchronously infected with parasites. According to our results, 206 genes were down-regulated in AP2-FG-disrupted parasites. Of these genes, 40 and 117 were targets of sAP2-FG and cAP2-FG, respectively. However, it is still possible that a significant proportion of genes were indirectly down-regulated by AP2-FG disruption, which may impair gametocyte development. Moreover, based on the results of the present study, expression of a significant proportion of AP2-FG target genes could be complemented by PFG transcription. We believe that it would be difficult to compare the direct effects of these TFs on gene expression via transcriptome analysis (therefore, targetome analysis is important). In this study, we compared the expression of target genes of sAP2-FG and cAP2FG between PFG(-) and WT parasites. We expected that down-regulation of PFG (cAP2FG) targets would be complemented with transcription by sAP2-FG.

1. Minor points

-Page 5 Line 10, remove "as"

Response: We have corrected this.

-Page 7 Lines 4-13: is it possible to perform the assay in PFG(-) parasites?

Response: Thank you for your question. Even when the marker gene expression was decreased in PFG(-) parasites, we cannot conclude the reason to be a direct effect of the mutation. To determine the function of the motif, it is necessary to perform the assay using wild-type parasites.

-Page 7 Line 45: Fig6C instead of 5C

Response: Thank you for pointing this out. We have corrected this.

-Page 8 Line 27: "decreases"

Response: Thank you for pointing this out. We have corrected this.

-Page 8 Line 36: PFG instead of PGP

Response: We have corrected this.

-Page 8 Line 39: remove "the fact"

Response: We have removed this word.

-Page 8 Line 42: Fig6G instead of 5G

Response: We have corrected this.

-Page 8 Line 43: PFG instead of PGP

Response: We have corrected this.

-Page 9 Line 23: "electroporation"

Response: We have corrected this.

-Page 9 Line 32: "BamHI"

Response: We have corrected this.

-Fig 2E: in the crosses did the authors check oocyst formation in the mosquito?

Response: We did not check oocyst formation because abnormalities in males may not affect oocyst formation.

-Page 17, legend Fig3, Line 14, there is probably an inversion between left and right for PFG versus AP2-FG (either in the legend or in the figure)

Response: Thank you for pointing this out. PFG peaks are located in the center in both heat maps. The description “AP2-FG peaks” over the arrowhead in the left map was incorrect. We have corrected this to “PFG peaks”. The peaks in the left heat map must be located in the center; thus, this figure might be redundant.

Reviewer #2 (Recommendations for the Authors):

• Could the authors please state in the results section that PFG stands for partner of AP2FG.

Response: Thank you for the comment. We have added the following to the results section:

“Through this screening, a gene encoding a 2709 amino acid protein with two regions highly conserved among Plasmodium was identified (PBANKA0902300, designated as a partner of AP2-FG (PFG; Fig. 1A).”

• Given that the transcriptional program is so dynamic, the timing of the ChIP-seq experiments is crucial. Could the authors clarify the timings of the different ChIP-seq experiments (AP2-FG, PFG, PFG in AP2-FG-, AP2-FG in PFG-, ...)

Response: Thank you for the comment. To deplete any parasites in the asexual stages, all ChIP-seq experiments in this study were performed using blood from mice treated with sulfadiazine, namely, gametocyte-enriched blood. As the reviewer points out, timing is important, and samples from the period when TFs are maximally expressed are optimal for ChIP-seq. However, when parasites in the asexual stages are present, the background becomes higher. Thus we usually use gametocyte-enriched blood for ChIP-seq when expression of the TF is observed in mature gametocytes. The exception was our ChIP-seq analysis of AP2-G, because is not present in mature gametocytes.

• Fig 4c is an example of great overlap of peaks, but it would be helpful if the authors could quantify the overlaps between experiments (and describe the overlap parameters used).

Response: According to the comment, we have created a Venn diagram of overlapping peaks (attached below). However, the peaks used for this Venn diagram were selected after peakcalling via fold-enrichment values. Thus, even if the counterpart of a peak is absent in these selected peaks (non-overlapping peaks in the Venn diagram), it does not indicate that it is absent in the original read map. We believe the overlap of peaks would be estimated more correctly in the heat maps.

Author response image 1.

Legged: The Venn diagram shows the number of common peaks between these ChIP seq experiments (distance of peak summits < 150

• Additionally, how were the promoter coordinates used for each gene when they associate ChIP peaks to a gene target. Did the authors choose 1-2kb? Or use a TSS/5utr dataset such as Adjalley 2016 or Chappell 2020?

Response: We selected a 1.2 Kbp region for target prediction based on our previous studies. As the reviewer pointed out, target prediction using TSS information may be more accurate. However, reliable TSS information is not available for P. berghei to the best of our knowledge.

The two papers are studies on P. falciparum.

• In the absence of evidence of physical interaction, it remains unclear if AP2-FG and PFG actually interact directly or as part of the same complex. A more detailed characterisation with IPs/co-IPs followed by mass spectrometry of the GFP-tagged version of PFG in the presence and absence of AP2-FG would be highly informative.

Response: Thank you for the comment. Even when these two TFs occupy the same genomic region, it cannot be conclusively said that they exist at the same time in the region: they might competitively occupy the region. However, we showed that the cAP2-FG peaks disappear from the region when PFG was disrupted, while sAP2-FG peaks remain. We believe that this is evidence that the two TFs physically interact with each other.

• It was not clear if the assessment of motif binding using cytometry was performed using all the required controls and compensation. This section should be clarified.

Response: Thank you for the comment. Condensation was performed using parasites expressing a single fluorescent protein. The results are attached below. The histogram of mCherry using control parasites expressing GFP under the control of the HSP70 promoter is also attached.

Author response image 2.

However, we found that descriptions of the filters for detecting red signals were not correct. This assay was performed using parasites which expressed GFP constitutively and mCherry under the control of the p28 promoter. These two fluorescent proteins were excited by independent lasers (488 and 561, respectively), and the emission spectra were detected using independent detectors (through 530/30 and 610/20 filters, respectively). We have revised the description regarding our FACS protocols as follows:

“Flow cytometric analysis was performed using an LSR-II flow cytometer (BD Biosciences). In experiments using 820 parasites, the tail blood from infected mice was selected via gating with forward scatter and staining with Hoechst 33342 (excitation =355 nm, emission = 450/50). The gated population was then analyzed for GFP fluorescence (excitation = 488 nm, emission = 530/30) and RFP fluorescence (excitation = 561 nm, emission = 610/20). In the promoter assay (using parasites transfected with a centromere plasmid), the tail blood from infected mice was selected via gating with forward scatter and staining with Hoechst 33342 (excitation =355 nm, emission = 450/50), followed by GFP fluorescence (excitation = 488 nm, emission = 530/30). The gated population was analyzed for mCherry fluorescence (excitation = 561 nm, emission = 610/20). Analysis was performed using the DIVER program (BD Biosciences).”

Minor points:

• Page 4, line 37: The authors should specify the timing of expression of AP2-FG on the text.

Response: We have added the following description to the text.

“The timing of the expression was approximately four hours later than that of AP2-FG, which started at 16 hpi (9).” .

• Ref 9 and 17 are repeated

Response: Thank you for pointing this out. We have corrected this.

• Fig 1D and 1F do not have scale bars

Response: We have added scale bars to Fig. 1D.

We have not changed Fig. 1F, because we believe that the scales can be estimated from the size of the erythrocyte.

• Page 5, line 29-30. Could the authors specify how many and which of the de-regulated genes have a PFG in their promoter.

Response: Thank you for the comment, As described in a later section (page 7; Impact of PFG disruption on the expression of AP2-FG target genes), among the 279 genes significantly downregulated in PFG(-) parasites, 165 genes were targets for PFG (unique for PFG or common for sAP2-FG and PFG). In contrast, only four genes were targets unique to sAP2-FG. Therefore, 165 genes harbor the upstream peaks of PFG. These genes are shown in Table S1.

• Fig 5F. in the methods associated with this figure there seems to be a mixup with the description of the lasers. In addition, given the spillover of the red and green signal between detectors this experiment needs compensation parameters. The authors should provide the gating strategy before and after compensation as this is critical for the correct calculation of the number of red parasites. Indeed, the lowest red cloud on the gate shown could be green signal spill over.

Response: Thank you for the comment. As described above, there were some incorrect descriptions about the conditions of our FACS protocols in the methods section. We have revised them.

-Page 7, line 19. Could the authors explicitly say in the text that the 810 genes are those with 1 (or more?) PFG peaks in their promoter (out of a total of 1029) to best guide the reader. Additionally, it is important to define the maximum distance allowed between a peak and CDS for it to be associated with said CDS.

Response: We have revised Table S2 by adding the nearest genes. The revised table shows the relationship between a PFG peak and its nearest genes, together with their distances.

• Page 7, line 45: fig 6c, not 5c

Response: Thank you for the comment. We have corrected this.

• Page 7 last paragraph: This section is very hard to follow. For instance, on line 50 do the authors mean that the sAP2-FG unique targets are LESS de-regulated? On line 51: do the authors mean unique targets of cAP2-FG or unique targets of PFG? Line 53: do the authors mean that genes expressed in the "common" category are LESS de-regulated than the PFG unique targets?

Response: We are sorry for the lack of clarity; after reviewing the manuscript, it appears to be unclear what the fold change means in this section. Here, fold change means the ratio of PFG(-)/wild type. Thus “High log2(fold change) value” means that the genes were less downregulated. We have revised the description as follows:

“The log2 distribution (fold change = PFG(-)/wild type) in the three groups of target genes showed that the average value was significantly higher (i.e., less down-regulated) in targets unique to sAP2-FG than in the other two groups (targets unique to cAP2-FG or common targets for both), with p-values of 1.3 × 10-10 and 1.4 × 10-5, respectively, by two-tailed Student’s t-test (Fig. 6F). In addition, the average log2 (fold change) value of the common target genes was relatively higher (i.e., less down-regulated) than that of targets unique to PFG, suggesting that transcriptional activation by sAP2-FG partly complements the impact of PFG disruption on these common targets.”

• Page 8, line 42: Fig 6G, not 5G

Response: Thank you for pointing this out. We have corrected this.

Reviewer #3 (Recommendations For The Authors):

1. The gene at the center of this study (PBANKA_0902300) was identified in an earlier genetic screen by Russell et al. as being a female specific gene with essential role in transmission and named Fd2 (for female-defective 2). Since this name entered the literature first and is equally descriptive, the Fd2 name should be used instead of PFG to maintain clarity and avoid unnecessary confusion. Surprisingly, this study is neither cited nor acknowledged despite a preprint having been available since August of 2021. This should be remedied.

Response: Thank you for the comment. We have added the paper by Russell et al. accordingly and mentioned the name FD2 in the revised manuscript. However, we have retained the use of PFG throughout the paper. We believe that this usage of PFG shouldn’t be confusing, as FD2 has only been used in one previous paper. We have added the following:

“Through this screening, a gene encoding a 2709 amino acid protein with two regions highly conserved among Plasmodium was identified (PBANKA0902300, designated as a partner of AP2-FG (PFG; Fig. 1A). This gene is one of the P. berghei genes that were previously identified as genes involved in female gametocyte development (named FD2), based on mass screening combined with single cell RNA-seq (ref).”

1. While it isn't really important how the authors came to arrive at studying the function of Fd2, the rationale/approach given in the first paragraph of the result section seems far too broad to lead to Fd2, given that it lacks identifiable domains and many other ortholog sets exist across these species.

Response: We selected this gene from the list of AP2-G targets as a candidate for a sequence-specific TF based on the hypothesis that the amino acid sequences of DNAbinding domains are highly conserved. We successfully identified two TFs (including PFG) using this method. However, there may be TFs that do not fit this hypothesis which are also targets of AP2-G. In fact, we were unable to identify HDP1 as a TF candidate, despite being a AP2-G target.

1. Fig. 1A-C: Gene IDs for the orthologs should be provided, as well as the methodology for generating the alignments.

Response; We have added the gene IDs and method for alignment in the legend as follows:

(A) Schematic diagram of PFG from P. berghei and its homologs in apicomplexan parasites. Regions homologous to Regions 1 and 2, which are highly conserved among Plasmodium species, are shown as yellow and blue rectangles, respectively. Nuclear localization signals were predicted using the cNLS mapper (http://nls-10 mapper.iab.keio.ac.jp/cgibin/NLS_Mapper_form.cgi). The gene IDs of P. berghei PFG, P. falciparum PFG, and their homologs in Toxoplasma gondii, Eimeria tenella and Vitrella brassicaformis are PBANKA_0902300, PF3D7_1146800, TGGT1_239670, ETH2_1252400, and Vbra_10234, respectively.

(C) The amino acid sequences of Regions 1 and 2 from P. berghei PFG and its homologs from other apicomplexan parasites in (A) were aligned using the ClustalW program in MEGA X. The positions at which all these sequences have identical amino acids are indicated by two asterisks, and positions with amino acid residues possessing the same properties are indicated by one asterisk.

1. Figure 2: The Phenotype of Fd2 knockout should be characterized more comprehensively.

It remains unclear whether ∆Fd2 parasite generate the same number of females but these are defective upon fertilization or whether there is also a decrease in the number of female gametocytes. Is the defect just post-fertilization and zygotes lyse or are there fewer fertilization events? If so is activation of female GCs effected?

The number of male and female gametocytes should be quantified using sex-specific markers not affected by Fd2 knockout rather than providing a single image of each. The ability of ∆Fd2 GCs should also be evaluated.

This is also important for the interpretation of Fig 2G. Is the down-regulation of the genes due to fewer female GCs or are the down-regulated genes only a subset of female-specific genes.

Response: In PFG(-) parasites, the rate of conversion into zygotes of female gametocytes decreased, and zygotes had lost capacity for developing into ookinetes. This indicates that gametocyte development (i.e., the ability to egress the erythrocyte and to fertilize) and zygote development were both impaired. This phenotype is consistent with the observation that genes expressed in female gametocytes are broadly downregulated. PFG is a TF, and its disruption led to decreased expression of hundreds of female genes. Thus, the observed phenotype may be derived from combined decreased expression of these genes. We believe further detailed phenotypic analyses will not generate much novel information on this TF. Instead, RNA-seq data in PFG(-) parasites and the targetome have promise in helping to characterize the functions of this TF.

1. Figure 3: what fraction of down-regulated genes have the Fd2 10mer motif?

Response: Thank you for the question. We investigated the upstream binding motifs of these genes. Of the 279 significantly down-regulated genes (containing 165 targets), 161 genes harbor the motif (including nine-base motifs that lack one lateral base which is likely not essential for binding) in their upstream regions (within 1,200 bp from the first methionine codon). However, this result has not been described in the revised manuscript because it is more important whether these regions harbor PFG peaks (upstream motifs can exist without being involved in the binding of PFG).

1. sAP2-FG (single) vs cAP2-FG (complex) nomenclature is confusing and possibly misleading since few TFs function in isolation and sAP2-FG likely functions in a complex that doesn't contain Fd2, possibly with another DNA binding protein that binds the TGCACA hexamer. The name for the distinct peaks should refer to the presence or absence of Fd2 in the complex, or maybe simply refer to them as complex A & B.

Response: As shown in the DIP-seq analysis results, AP2-FG can bind the motif by itself. In contrast, AP2-FG must form a complex with PFG to bind to the ten-base motif. The complex and single forms are named according to this difference (the presence or absence of PFG) and used solely in its relation with PFG. We wrote “In the following, we refer to the form with PFG as cAP2-FG or the complex form, and the form without PFG as sAP2-FG or the single form.” We believe that the nomenclature has sufficient clarity. However, we have partially (underlined) revised certain sentences in the discussion section as follows.

“As the expression of PFG increases via this mechanism, AP2-FG recruited by PFG (cAP2FG) increases and eventually becomes predominant in the transcriptional regulation of female gametocytes.”

“This suggests that the promoter of the CCP2 gene, which is a target of PFG only, is still active in AP2-FG(-)820 parasites.”

We recently reported that the TGCACA motif is a cis-activation motif in early gametocytes and important for both male and female gametocyte development. Thus we speculate that sAP2-FG is not involved in cis-activation by the TGCACA motif. The p-value of the six-base motif is indeed comparable to that of the five-base motif. However, the pvalue (calculated by Fisher’s exact test) in six-base motifs tend to be lower than that calculated in five-base motifs, because the population is much large. We speculate that there is a sequence-specific TF that may be expressed in early gametocytes and bind this motif, independently of AP2-FG.

1. I compared the overlap of peaks in the 4 ChIP-seq data sets:

90% of the Fd2 peaks are shared with AP2-FG (binding 24% of shared peaks is lost in ∆AP2FG)

10% are bound by Fd2 alone (binding at 35% of Fd2 is lost in ∆AP2-FG)

75% of Fd2 peaks are bound independently of AP2-FG

47% of AP2-FG peaks shared with Fd2 (binding at 71% of shared peaks is lost in ∆Fd2) 53% of AP2-FG peaks are bound only by AP2-FG (but binding at 82% of AP2-FG only peaks is still lost in the ∆Fd2)

Binding at 78% of all AP2-FG peaks is lost in ∆Fd2

This indicates that much of AP2-FG binding in regions even in regions devoid of Fd2 still depends on Fd2. What are possible explanations for this?

https://elife-rp.msubmit.net/eliferp_files/2023/04/03/00117573/00/117573_0_attach_10_17936_convrt.pdf

Response: In the ChIP-seq of AP2-FG in the absence of PFG, 441 peaks are still called. This means that at least 441 binding sites for AP2-FG independent of PFG exist. This is a straightforward conclusion from our ChIP-seq data. On the other hand, simple deduction of peaks between two ChIP-seq experiments (AP2-FG peaks minus PFG peaks) is not a precise method for determining sAP2-FG. Peak-calling is independently performed in each ChIP-seq experiment. Thus, peaks remaining after the deduction between two experiments can still contain peaks that are actually common, but which are differentially picked up through the process of peak calling. Even when using data obtained by the same ChIP-seq experiment, markedly different numbers of peaks are called according to the conditions for peak calling (in contrast, common peaks between two independent experiments increase the reliability of the data). If wanting to identify sAP2-FG peaks via comparisons between AP2-FG peaks and PFG peaks, the reviewer has to increase the number of PFG peaks by reducing the peak-calling threshold until the number of overlapping peaks between AP2-FG and PFG are saturated, and then deduce the overlapping peaks from the AP2-FG peaks. However, as described above, for the purposes of estimating the number of sAP2-FG, it would be better to perform ChIP-seq of AP2-FG in the absence of PFG.

1. Possible explanations of why recombinant Fd2 doesn't bind the TGCACA hexamer. It would also be good to note that the GCTCA AP2-FG motif found in Fig4G is now perfect match for the motif identified by protein binding microarray in Campbell et al.

Response: It is not known what sequence recombinant PFG binds. The TGCACA motif is not enriched in PFG peaks. If the reviewer is referring to AP2-FG, our findings that the recombinant AP2 domain binds the five-base motif strongly suggests that other TFs recognize this motif. As described in our response to comment 9, we recently reported that TGCACA is a cis-activating sequence important for the normal development of both male and female gametocytes. Therefore, we currently speculate that this motif is a binding motif of other TFs and is independent of AP2-FG.

We have mentioned the protein binding microarray data in the Results section as follows.

“The most enriched motif matched well with the binding sequence of the AP2 domain of P. falciparum AP2-FG, which was reported by Campbell et al.”

1. What might explain the strong enrichment for TGCACA in ChIPseq but when pulled down by AP2-FG DBD: another binding partner? requires more of AP2-DF than just DBD?

Response: As described above in our response to comment 6, we have recently submitted a preprint studying the roles of the remodeler subunit PbARID in gametocyte development. We reported that the remodeler subunit is recruited to the six-base motif and that the motif is a novel cis-activation element for early gametocyte development. We speculate that a proportion of AP2-FG targets are also targets of a TF that recognizes this motif and recruits the remodeler subunit. These two TFs may be involved in the regulation of early gametocyte genes but function independently.

1. Calling DNA pulldown with recombinant AP2-FG DNA-binding domain DNAImmunoprecipitation sequencing (DIP-seq) is confusing since there are no antibodies involved. Describing it directly as a pulldown of fragmented DNA will be clearer to the reader.

Response: Thank you for the comment. We have also recognized this discrepancy. However we called the method DIP-seq because the original paper reporting this method used this name, wherein it did not use antibodies to capture the MBP-fusion recombinant protein. Our experiment was performed using essentially the same methods, and thus we retained the name.

1. The legends and methods are very sparse and should include substantially more detail.

Response: Thank you for the comment. We have revised the description of the FACS experimental method for clarity.

1. BigWig files for all ChIPseq enrichment used for analysis in this study need to be provided.

(two replicates each of : Fd2 in WT, Fd2 in ∆AP2-GF, AP2-FG in WT, AP2-FG in ∆Fd2)

Response: We have deposited the BigWig files to GEO (GSE.226028 and GSE114096).

1. Tables of ChIP data need to have both summits and peaks and need to list nearest gene. Also the ChIPseq peaks for Fd2 are surprisingly broad (ChIP peaks are very large, e.g. 68% of Fd2 peaks (dataset2) are greater than 1000kb) give its specificity for a long motif. Why is this?

Response: We have revised Table S2 to include the nearest genes. We are unsure why peaks in the over 1000-bp peak region exist in such high proportions. However, this proportion was also high in our previous ChIP-seq data. Therefore, we speculate that this is a tendency of peak-calling by MACS2. We did not use these values in this paper. For example, targets were predicted using peak summits, and binding motifs were calculated using the 100-base regions around peak summits.

1. Figure 5E: The positions of the 10mer and 5mer motifs in the promoter should be indicated as well as the length of the promoter. Moreover, mutation of just the 5bp motifs would be valuable to understand if 10mer is sufficient for expression of the reporter.

Response: Thank you for the comment. We have revised the figure accordingly. The majority of female-specific promoters only harbor ten-base motifs. Thus the ten-base motif is sufficient for evaluating reporter activity (i.e., it would function without five-base motifs).

1. How is AP2-FG expression affected in ∆Fd2 and vice versa?

Response: According to our previous microarray data, PFG expression was not significantly downregulated by disruption of AP2-FG. This may be because PFG transcriptionally activates itself through a positive feedback loop after being induced by AP2-G. Similarly, according to our present study, AP2-FG expression was not downregulated by PFG disruption. This may be because AP2-FG is transcriptionally activated by AP2-G.

1. The single cell data in Russell et al. could easily be used to indicate the order of expression.

Response: Determining the expression order of gametocyte TFs via the single cell RNA-seq data from Russel et al. is difficult, because only a small number of parasite cells were considered to be in the early gametocyte stage in this study. This is because the parasites were cultured for 24h before the analysis. The analysis suggested by the reviewer may be possible via single cell RNA-seq, but the experiments must be performed with more focus on the early gametocyte stage.

1. A discussion of the implication of P. falciparum transmission would be appreciated.

Response: Thank you for the comment. We have added the following to the Discussion section:

“P. falciparum gametocytes require 9-12 days to mature, which is much longer than that of P. berghei. Meanwhile, it has been reported that the ten-base motif is highly enriched in the upstream regions of female-specific genes also in P. falciparum. Thus, despite the difference in maturation periods, PFG is likely to play an important role in the transcriptional regulation of female P. falciparum gametocyte development."

1. The lack of identifiable DNA binding domains in Fd2 is intriguing given the strong sequence-specificity. Do the authors think they have identified a new DNA-binding fold ?

Alphafold of the orthologs with contiguous regions 1&2 might offer insight.

Response: We speculate that these regions function as DNA binding domains. We performed analysis using Alfafold2 according to the comment. However, the predicted structure of the region was not similar to any other canonical DNA-binding domains. Thus, it may be a novel DNA-binding fold as the reviewer mentioned. Further studies such as binding assays using recombinant proteins would be necessary to confirm this, but thus far we have not successfully obtained the soluble proteins of these regions.

Author Response

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

Thank you and the reviewers for further providing constructive comments and suggestions on our manuscript. On behalf of all the co-authors, I have enclosed a revised version of the above referenced paper. Below, I have merged similar public reviews and recommendations (if applicable) from each reviewer and provided point-by-point responses.

Reviewer #1:

People can perform a wide variety of different tasks, and a long-standing question in cognitive neuroscience is how the properties of different tasks are represented in the brain. The authors develop an interesting task that mixes two different sources of difficulty, and find that the brain appears to represent this mixture on a continuum, in the prefrontal areas involved in resolving task difficulty. While these results are interesting and in several ways compelling, they overlap with previous findings and rely on novel statistical analyses that may require further validation.

Strengths

1. The authors present an interesting and novel task for combining the contributions of stimulus-stimulus and stimulus-response conflict. While this mixture has been measured in the multi-source interference task (MSIT), this task provides a more graded mixture between these two sources of difficulty.

2. The authors do a good job triangulating regions that encoding conflict similarity, looking for the conjunction across several different measures of conflict encoding. These conflict measures use several best-practice approaches towards estimating representational similarity.

3. The authors quantify several salient alternative hypothesis and systematically distinguish their core results from these alternatives.

4. The question that the authors tackle is important to cognitive control, and they make a solid contribution.

The authors have addressed several of my concerns. I appreciate the authors implementing best practices in their neuroimaging stats.

I think that the concerns that remain in my public review reflect the inherent limitations of the current work. The authors have done a good job working with the dataset they've collected.

Response: We would like to thank the reviewer for the positive evaluation of our manuscript and the constructive comments and suggestions. In response to your suggestions and concerns, we have removed the Stroop/Simon-only and the Stroop+Simon models, revised our conclusion and modified the misleading phrases.

We have provided detailed responses to your comments below.

1. The evidence from this previous work for mixtures between different conflict sources makes the framing of 'infinite possible types of conflict' feel like a strawman. The authors cite classic work (e.g., Kornblum et al., 1990) that develops a typology for conflict which is far from infinite. I think few people would argue that every possible source and level of difficulty will have to be learned separately. This work provides confirmatory evidence that task difficulty is represented parametrically (e.g., consistent with the n-back, MOT, and random dot motion literature).

notes for my public concerns.

In their response, the authors say:

'If each combination of the Stroop-Simon combination is regarded as a conflict condition, there would be infinite combinations, and it is our major goal to investigate how these infinite conflict conditions are represented effectively in a space with finite dimensions.'

I do think that this is a strawman. The paper doesn't make a strong case that this position ('infinite combinations') is widely held in the field. There is previous work (e.g., n-back, multiple object tracking, MSIT, dot motion) that has already shown parametric encoding of task difficulty. This paper provides confirmatory evidence, using an interesting new task, that demand are parametric, but does not provide a major theoretical advance.

Response: We agree that the previous expression may have seemed somewhat exaggerative. While it is not “infinite”, recent research indeed suggests that the cognitive control shows domain-specificity across various “domains”, including conflict types (Egner, 2008), sensory modalities (Yang et al., 2017), task-irrelevant stimuli (Spape et al., 2008), and task sets (Hazeltine et al., 2011), to name a few.

These findings collectively support the notion that cognitive control is contextspecific (Bream et al., 2014). That is, cognitive control can be tuned and associated with different (and potentially large numbers of) contexts. Recently, Kikumoto and Mayr (2020) demonstrated that combinations of stimulus, rule and response in the same task formed separatable, conjunctive representations. They further showed that these conjunctive representations facilitate performance. This is in line with the idea that each stimulus-location combination in the present task may be represented separately in a domain-specific manner. Moreover, domain-general task representation can also become domain-specific with learning, which further increases the number of domain-specific conjunctive representations (Mill et al., 2023). In line with the domain-specific account of cognitive control, we referred to the “infinite combinations” in our previous response to emphasize the extreme case of domainspecificity. However, recognizing that the term “infinite” may lead to ambiguity, we have replaced it with phrases such as “a large number of”, “hugely varied”, in our revised manuscript.

We appreciate the reviewer for highlighting the potential connection of our work to existing literature that showed the parametric encoding of task difficulty (e.g., Dagher et al., 1999; Ritz & Shenhav, 2023). For instance, in Ritz et al.’s (2023) study, they parametrically manipulated target difficulty based on consistent ratios of dot color, and found that the difficulty was encoded in the caudal part of dorsal anterior cingulate cortex. Analogically, in our study, the “difficulty” pertains to the behavioral congruency effect that we modulated within the spatial Stroop and Simon dimensions. Notably, we did identify univariate effects in the right dmPFC and IPS associated with the difficulty in the Simon dimension. This parametric effect may lend support to our cognitive space hypothesis, although we exercised caution in interpreting their significance due to the absence of a clear brain-behavioral relevance in these regions. We have added the connection of our work to prior literature in the discussion. The parametric encoding of conflict also mirrors prior research showing the parametric encoding of task demands (Dagher et al., 1999; Ritz & Shenhav, 2023).

However, our analyses extend beyond solely testing the parametric encoding of difficulty. Instead, we focused on the multivariate representation of different conflict types, which we believe is independent from the univariate parametric encoding. Unlike the univariate encoding that relies on the strength within one dimension, the multivariate representation of conflict types incorporates both the spatial Stroop and Simon dimensions. Furthermore, we found that similar difficulty levels did not yield similar conflict representation, as indicated by the low similarity between the spatial Stroop and Simon conditions, despite both showing a similar level of congruency effect (Fig. S1). Additionally, we also observed an interaction between conflict similarity and difficulty (i.e., congruency, Fig. 4B/D), such that the conflict similarity effect was more pronounced when conflict was present. Therefore, we believe that our findings make contribution to the literature beyond the difficulty effect.

Reference:

Egner, T. (2008). Multiple conflict-driven control mechanisms in the human brain. Trends in Cognitive Sciences, 12(10), 374-380. https://doi.org/10.1016/j.tics.2008.07.001

Yang, G., Nan, W., Zheng, Y., Wu, H., Li, Q., & Liu, X. (2017). Distinct cognitive control mechanisms as revealed by modality-specific conflict adaptation effects. Journal of Experimental Psychology: Human Perception and Performance, 43(4), 807-818. https://doi.org/10.1037/xhp0000351

Spapé MM, Hommel B (2008). He said, she said: episodic retrieval induces conflict adaptation in an auditory Stroop task. Psychonomic Bulletin Review,15(6):1117-21. https://doi.org/10.3758/PBR.15.6.1117

Hazeltine E, Lightman E, Schwarb H, Schumacher EH (2011). The boundaries of sequential modulations: evidence for set-level control. Journal of Experimental Psychology: Human Perception & Performance. 2011 Dec;37(6):1898-914. https://doi.org/10.1037/a0024662

Braem, S., Abrahamse, E. L., Duthoo, W., & Notebaert, W. (2014). What determines the specificity of conflict adaptation? A review, critical analysis, and proposed synthesis. Frontiers in Psychology, 5, 1134. https://doi.org/10.3389/fpsyg.2014.01134

Kikumoto A, Mayr U. (2020). Conjunctive representations that integrate stimuli, responses, and rules are critical for action selection. Proceedings of the National Academy of Sciences, 117(19):10603-10608. https://doi.org/10.1073/pnas.1922166117.

Mill, R. D., & Cole, M. W. (2023). Neural representation dynamics reveal computational principles of cognitive task learning. bioRxiv. https://doi.org/10.1101/2023.06.27.546751

Dagher, A., Owen, A. M., Boecker, H., & Brooks, D. J. (1999). Mapping the network for planning: a correlational PET activation study with the Tower of London task. Brain, 122 ( Pt 10), 1973-1987. https://doi.org/10.1093/brain/122.10.1973

Ritz, H., & Shenhav, A. (2023). Orthogonal neural encoding of targets and distractors supports multivariate cognitive control. https://doi.org/10.1101/2022.12.01.518771

1. (Public Reviews) The degree of Stroop vs Simon conflict is perfectly negatively correlated across conditions. This limits their interpretation of an integrated cognitive space, as they cannot separately measure Stroop and Simon effects. The author's control analyses have limited ability to overcome this task limitation. While these results are consistent with parametric encoding, they cannot adjudicate between combined vs separated representations.

(Recommendations) I think that it is still an issue that the task's two features (stroop and simon conflict) are perfectly correlated. This fundamentally limits their ability to measure the similarity in these features. The authors provide several control analyses, but I think these are limited.

Response: We need to acknowledge that the spatial Stroop and Simon components in the five conflict conditions were not “perfectly” correlated, with r = –0.89. This leaves some room for the preliminary model comparison to adjudicate between these models. However, it’s essential to note that conclusions based on these results must be tempered. In line with the reviewer’s observation, we agree that the high correlation between the two conflict sources posed a potential limitation on our ability to independently investigate the contribution of spatial Stroop and Simon conflicts. Therefore, in addition to the limitation we have previously acknowledged, we have now further revised our conclusion and adjusted our expressions accordingly.

Specifically, we now regard the parametric encoding of cognitive control not as direct evidence of the cognitive space view but as preliminary evidence that led us to propose this hypothesis, which requires further testing. Notably, we have also modified the title from “Conflicts are represented in a cognitive space to reconcile domain-general and domain-specific cognitive control” to “Conflicts are parametrically encoded: initial evidence for a cognitive space view to reconcile the debate of domain-general and domain-specific cognitive control”. Also, we revised the conclusion as: In sum, we showed that the cognitive control can be parametrically encoded in the right dlPFC and guides cognitive control to adjust goal-directed behavior. This finding suggests that different cognitive control states may be encoded in an abstract cognitive space, which reconciles the long-standing debate between the domain-general and domain-specific views of cognitive control and provides a parsimonious and more broadly applicable framework for understanding how our brains efficiently and flexibly represents multiple task settings.

From Recommendations The authors perform control analyses that test stroop-only and simon-only models. However, these analyses use a totally different similarity metric, that's based on set intersection rather than geometry. This metric had limited justification or explanation, and it's not clear whether these models fit worse because of the similarity metric. Even here, Simon-only model fit better than Stroop+Simon model. The dimensionality analyses may reflect the 1d manipulation by the authors (i.e. perfectly corrected stroop and simon effects).

Response: The Jaccard measure is the most suitable method we can conceive of for assessing the similarity between two conflicts when establishing the Stroop-only and Simon-only models, achieved by projecting them onto the vertical or horizontal axes, respectively (Author response image 1A). This approach offers two advantages. First, the Jaccard similarity combines both similarity (as reflected by the numerator) and distance (reflected by the difference between denominator and numerator) without bias towards either. Second, the Jaccard similarity in our design is equivalent to the cosine similarity because the denominator in the cosine similarity is identical to the denominator in the Jaccard similarity (both are the radius of the circle, Author response image 1B).

Author response image 1.

Definition of Jaccard similarity. A) Two conflicts (1 and 2) are projected onto the spatial Stroop/Simon axis in the Stroop/Simon-only model, respectively. The Jaccard similarity for Stroop-only and Simon-only model are and respectively. Letters a-d are the projected vectors from the two conflicts to the two axes. Blue and red colors indicate the conflict conditions. Shorter vectors are the intersection and longer vectors are the union. B) According to the cosine similarity model, the similarity is defined as , where e is the projected vector from conflict 1 to conflict 2, and g is the vector of conflict 1. The Jaccard similarity for this case is defined by , where f is the projector vector from conflict 2 to itself. Because f = g in our design, the Jaccard similarity is equivalent to the cosine similarity.

Therefore, we believe that the model comparisons between cosine similarity model and the Stroop/Simon-Only models were equitable. However, we acknowledge the reviewer’s and other reviewers’ concerns about the correlation between spatial Stroop and Simon conflicts, which reduces the space to one dimension (1d) and limits our ability to distinguish between the Stroop-only and Simon-only models, as well as between Stroop+Simon and cosine similarity models. While these distinctions are undoubtedly important for understanding the geometry of the cognitive space, we recognize that they go beyond the major objective of this study, that is, to differentiate the cosine similarity model from domain-general/specific models. Therefore, we have chosen to exclude the Stroop-only, Simon-only and Stroop+Simon models in our revised manuscript.

Something that raised additional concerns are the RSMs in the key region of interest (Fig S5). The pure stroop task appears to be represented very differently from all of the conditions that include simon conflict.

Together, I think these limitations reflect the structure of the task and research goals, not the statistical approach (which has been meaningfully improved).

Response: We appreciate the reviewer for pointing this out. It is essential to clarify that our conclusions were based on the significant similarity modulation effect identified in our statistical analysis using the cosine similarity model, where we did not distinguish between the within-Stroop condition and the other four within-conflict conditions (Fig. 7A, now Fig. 8A). This means that the representation of conflict type was not biased by the seemingly disparities in the values shown here. Moreover, to specifically test the differences between the within-Stroop condition and the other within-conflict conditions, we conducted a mixed-effect model analysis only including trial pairs from the same conflict type. In this analysis, the primary predictor was the cross-condition difference (0 for within-Stroop condition and 1 for other within-conflict conditions). The results showed no significant cross-condition difference in either the incongruent (t = 1.22, p = .23) or the congruent (t = 1.06, p = .29) trials. Thus, we believe the evidence for different similarities is inconclusive in our data and decided not to interpret this numerical difference. We have added this note in the revised figure caption for Figure S5.

Author response image 2.

Fig. S5. The stronger conflict type similarity effect in incongruent versus congruent conditions. (A) Summary representational similarity matrices for the right 8C region in incongruent (left) and congruent (right) conditions, respectively. Each cell represents the averaged Pearson correlation of cells with the same conflict type and congruency in the 1400×1400 matrix. Note that the seemingly disparities in the values of Stroop and other within-conflict cells (i.e., the diagonal) did not reach significance for either incongruent (t = 1.22, p = .23) or congruent (t = 1.06, p = .29) trials. (2) Scatter plot showing the averaged neural similarity (Pearson correlation) as a function of conflict type similarity in both conditions. The values in both A and B are calculated from raw Pearson correlation values, in contrast to the z-scored values in Fig. 4D.

Minor:

• In the analysis of similarity_orientation, the df is very large (~14000). Here, and throughout, the df should be reflective of the population of subjects (ie be less than the sample size).

Response: The large degrees of freedom (df) in our analysis stem from the fact that we utilized a mixed-effect linear model, incorporating all data points (a total of 400×35=14000). In mixed-effect models, the df is determined by subtracting the number of fixed effects (in our case, 7) from the total number of observations. Notably, we are in line with the literature that have reported the df in this manner (e.g., Iravani et al., 2021; Schmidt & Weissman, 2015; Natraj et al., 2022).

Reference:

Iravani B, Schaefer M, Wilson DA, Arshamian A, Lundström JN. The human olfactory bulb processes odor valence representation and cues motor avoidance behavior. Proc Natl Acad Sci U S A. 2021 Oct 19;118(42):e2101209118. https://doi.org/10.1073/pnas.2101209118.

Schmidt, J.R., Weissman, D.H. Congruency sequence effects and previous response times: conflict adaptation or temporal learning?. Psychological Research 80, 590–607 (2016). https://doi.org/10.1007/s00426-015-0681-x.

Natraj, N., Silversmith, D. B., Chang, E. F., & Ganguly, K. (2022). Compartmentalized dynamics within a common multi-area mesoscale manifold represent a repertoire of human hand movements. Neuron, 110(1), 154-174. https://doi.org/10.1016/j.neuron.2021.10.002.

• it would improve the readability if there was more didactic justification for why analyses are done a certain way (eg justifying the jaccard metric). This will help less technically-savvy readers.

Response: We appreciate the reviewer’s suggestion. However, considering the Stroop/Simon-only models in our design may not be a valid approach for distinguishing the contributions of the Stroop/Simon components, we have decided not to include the Jaccard metrics in our revised manuscript.

Besides, to improve the readability, we have moved Figure S4 to the main text (labeled as Figure 7), and added the domain-general/domain-specific schematics in Figure 8.

Author response image 3.

Figure 8. Schematic of key RSMs. (A) and (B) show the orthogonality between conflict similarity and orientation RSMs. The within-subject RSMs (e.g., Group1-Group1) for conflict similarity and orientation are all the same, but the cross-group correlations (e.g., Group2-Group1) are different. Therefore, we can separate the contribution of these two effects when including them as different regressors in the same linear regression model. (C) and (D) show the two alternative models. Like the cosine model (A), within-group trial pairs resemble between-group trial pairs in these two models. The domain-specific model is an identity matrix. The domain-general model is estimated from the absolute difference of behavioral congruency effect, but scaled to 0(lowest similarity)-1(highest similarity) to aid comparison. The plotted matrices here include only one subject each from Group 1 and Group 2. Numbers 1-5 indicate the conflict type conditions, for spatial Stroop, StHSmL, StMSmM, StLSmH, and Simon, respectively. The thin lines separate four different sub-conditions, i.e., target arrow (up, down) × congruency (incongruent, congruent), within each conflict type.

Reviewer #2:

This study examines the construct of "cognitive spaces" as they relate to neural coding schemes present in response conflict tasks. The authors use a novel experimental design in which different types of response conflict (spatial Stroop, Simon) are parametrically manipulated. These conflict types are hypothesized to be encoded jointly, within an abstract "cognitive space", in which distances between task conditions depend only on the similarity of conflict types (i.e., where conditions with similar relative proportions of spatial-Stroop versus Simon conflicts are represented with similar activity patterns). Authors contrast such a representational scheme for conflict with several other conceptually distinct schemes, including a domain-general, domain-specific, and two task-specific schemes. The authors conduct a behavioral and fMRI study to test which of these coding schemes is used by prefrontal cortex. Replicating the authors' prior work, this study demonstrates that sequential behavioral adjustments (the congruency sequence effect) are modulated as a function of the similarity between conflict types. In fMRI data, univariate analyses identified activation in left prefrontal and dorsomedial frontal cortex that was modulated by the amount of Stroop or Simon conflict present, and representational similarity analyses (RSA) that identified coding of conflict similarity, as predicted under the cognitive space model, in right lateral prefrontal cortex.

This study tackles an important question regarding how distinct types of conflict might be encoded in the brain within a computationally efficient representational format. The ideas postulated by the authors are interesting ones and the statistical methods are generally rigorous.

Response: We would like to express our sincere appreciation for the reviewer’s positive evaluation of our manuscript and the constructive comments and suggestions. In response to your suggestions and concerns, we excluded the StroopOnly, SimonOnly and Stroop+Simon models, and added the schematic of domain-general/specific model RSMs. We have provided detailed responses to your comments below.

The evidence supporting the authors claims, however, is limited by confounds in the experimental design and by lack of clarity in reporting the testing of alternative hypotheses within the method and results.

1. Model comparison

The authors commendably performed a model comparison within their study, in which they formalized alternative hypotheses to their cognitive space hypothesis. We greatly appreciate the motivation for this idea and think that it strengthened the manuscript. Nevertheless, some details of this model comparison were difficult for us to understand, which in turn has limited our understanding of the strength of the findings.

The text indicates the domain-general model was computed by taking the difference in congruency effects per conflict condition. Does this refer to the "absolute difference" between congruency effects? In the rest of this review, we assume that the absolute difference was indeed used, as using a signed difference would not make sense in this setting. Nevertheless, it may help readers to add this information to the text.

Response: We apologize for any confusion. The “difference” here indeed refers to the “absolute difference” between congruency effects. We have now clarified this by adding the word “absolute” accordingly.

"Therefore, we defined the domain-general matrix as the absolute difference in their congruency effects indexed by the group-averaged RT in Experiment 2."

Regarding the Stroop-Only and Simon-Only models, the motivation for using the Jaccard metric was unclear. From our reading, it seems that all of the other models --- the cognitive space model, the domain-general model, and the domain-specific model --- effectively use a Euclidean distance metric. (Although the cognitive space model is parameterized with cosine similarities, these similarity values are proportional to Euclidean distances because the points all lie on a circle. And, although the domain-general model is parameterized with absolute differences, the absolute difference is equivalent to Euclidean distance in 1D.) Given these considerations, the use of Jaccard seems to differ from the other models, in terms of parameterization, and thus potentially also in terms of underlying assumptions. Could authors help us understand why this distance metric was used instead of Euclidean distance? Additionally, if Jaccard must be used because this metric seems to be non-standard in the use of RSA, it would likely be helpful for many readers to give a little more explanation about how it was calculated.

Response: We believe that the Jaccard similarity measure is consistent with the Cosine similarity measure. The Jaccard similarity is calculated as the intersection divided by the union. To define the similarity of two conflicts in the Stroop-only and Simon-only models, we first project them onto the vertical or horizontal axes, respectively (as shown in Author response image 1A). The Jaccard similarity in our design is equivalent to the cosine similarity because the denominator in the Jaccard similarity is identical to the denominator in the cosine similarity (both are the radius of the circle, Author response image 1B).

However, it is important to note that a cosine similarity cannot be defined when conflicts are projected onto spatial Stroop or Simon axis simultaneously. Therefore, we used the Jaccard similarity in the previous version of our manuscript.

Author response image 4.

Definition of Jaccard similarity. A) Two conflicts (1 and 2) are projected onto the spatial Stroop/Simon axis in the Stroop/Simon-only model, respectively. The Jaccard similarity for Stroop-only and Simon-only model are and respectively. Letters a-d are the projected vectors from the two conflicts to the two axes. Blue and red colors indicate the conflict conditions. Shorter vectors are the intersection and longer vectors are the union. B) According to the cosine similarity model, the similarity is defined as , where e is the projected vector from conflict 1 to conflict 2, and g is the vector of conflict 1. The Jaccard similarity for this case is defined by , where f is the projector vector from conflict 2 to itself. Because f = g in our design, the Jaccard similarity is equivalent to the cosine similarity.

However, we agree with the reviewer’s and other reviewers’ concern that the correlation between spatial Stroop and Simon conflicts makes it less likely to distinguish the Stroop+Simon from cosine similarity models. While distinguishing them is essential to understand the detailed geometry of the cognitive space, it is beyond our major purpose, that is, to distinguish the cosine similarity model with the domain-general/specific models. Therefore, we have chosen to exclude the Stroop-only, Simon-only and Stroop+Simon models from our revised manuscript.

When considering parameterizing the Stroop-Only and Simon-Only models with Euclidean distances, one concern we had is that the joint inclusion of these models might render the cognitive space model unidentifiable due to collinearity (i.e., the sum of the Stroop-Only and Simon-Only models could be collinear with the cognitive space model). Could the authors determine whether this is the case? This issue seems to be important, as the presence of such collinearity would suggest to us that the design is incapable of discriminating those hypotheses as parameterized.

Response: We acknowledge that our design does not allow for a complete differentiation between the parallel encoding (StroopOnly+SimonOnly) model and the cognitive space model, given their high correlation (r = 0.85). However, it is important to note that the StroopOnly+SimonOnly model introduces more free parameters, making the model fitting poorer than the cognitive space model.

Additionally, the cognitive space model also shows high correlations with the StroopOnly and SimonOnly models (both rs = 0.66). It is crucial to emphasize that our study’s primary goal does not involve testing the parallel encoding hypothesis (through the StroopOnly+SimonOnly model). As a result, we have chosen to remove the model comparison results with the StroopOnly, SimonOnly and StroopOnly+SimonOnly models. Instead, the cognitive space model shows lower correlation with the purely domain-general (r = −0.16) and domain-specific (r = 0.46) models.

1. Issue of uniquely identifying conflict coding

We certainly appreciate the efforts that authors have taken to address potential confounders for encoding of conflict in their original submission. We broach this question not because we wish authors to conduct additional control analyses, but because this issue seems to be central to the thesis of the manuscript and we would value reading the authors' thoughts on this issue in the discussion.

To summarize our concerns, conflict seems to be a difficult variable to isolate within aggregate neural activity, at least relative to other variables typically studied in cognitive control, such as task-set or rule coding. This is because it seems reasonable to expect that many more nuisance factors covary with conflict -- such as univariate activation, level of cortical recruitment, performance measures, arousal --- than in comparison with, for example, a well-designed rule manipulation. Controlling for some of these factors post-hoc through regression is commendable (as authors have done here), but such a method will likely be incomplete and can provide no guarantees on the false positive rate.

Relatedly, the neural correlates of conflict coding in fMRI and other aggregate measures of neural activity are likely of heterogeneous provenance, potentially including rate coding (Fu et al., 2022), temporal coding (Smith et al., 2019), modulation of coding of other more concrete variables (Ebitz et al., 2020, 10.1101/2020.03.14.991745; see also discussion and reviews of Tang et al., 2016, 10.7554/eLife.12352), or neuromodulatory effects (e.g., Aston-Jones & Cohen, 2005). Some of these origins would seem to be consistent with "explicit" coding of conflict (conflict as a representation), but others would seem to be more consistent with epiphenomenal coding of conflict (i.e., conflict as an emergent process). Again, these concerns could apply to many variables as measured via fMRI, but at the same time, they seem to be more pernicious in the case of conflict. So, if authors consider these issues to be germane, perhaps they could explicitly state in the discussion whether adopting their cognitive space perspective implies a particular stance on these issues, how they interpret their results with respect to these issues, and if relevant, qualify their conclusions with uncertainty on these issues.

Response: We appreciate the reviewer’s insightful comments regarding the representation and process of conflict.

First, we agree that the conflict is not simply a pure feature like a stimulus but often arises from the interaction (e.g., dimension overlap) between two or more aspects. For example, in the manual Stroop, conflict emerges from the inconsistent semantic information between color naming and word reading. Similarly, other higher-order cognitive processes such as task-set also underlie the relationship between concrete aspects. For instance, in a face/house categorization task, the taskset is the association between face/house and the responses. When studying these higher-order processes, it is often impossible to completely isolate them from bottomup features. Therefore, methods like the representational similarity analysis and regression models are among the limited tools available to attempt to dissociate these concrete factors from conflict representation. While not perfect, this approach has been suggested and utilized in practice (Freund et al., 2021).

Second, we agree that conflict can be both a representation and an emerging process. These two perspectives are not necessarily contradictory. According to David Marr’s influential three-level theory (Marr, 1982), representation is the algorithm of the process to achieve a goal based on the input. Therefore, a representation can refer to not only a static stimulus (e.g., the visual representation of an image), but also a dynamic process. Building on this perspective, we posit that the representation of cognitive control consists of an array of dynamic representations embedded within the overall process. A similar idea has been proposed that the abstract task profiles can be progressively constructed as a representation in our brain (Kikumoto & Mayr, 2020).

We have incorporated this discussion into the manuscript:

"Recently an interesting debate has arisen concerning whether cognitive control should be considered as a process or a representation (Freund, Etzel, et al., 2021). Traditionally, cognitive control has been predominantly viewed as a process. However, the study of its representation has gained more and more attention. While it may not be as straightforward as the visual representation (e.g., creating a mental image from a real image in the visual area), cognitive control can have its own form of representation. An influential theory, Marr’s (1982) three-level model proposed that representation serves as the algorithm of the process to achieve a goal based on the input. In other words, representation can encompass a dynamic process rather than being limited to static stimuli. Building on this perspective, we posit that the representation of cognitive control consists of an array of dynamic representations embedded within the overall process. A similar idea has been proposed that the representation of task profiles can be progressively constructed with time in the brain (Kikumoto & Mayr, 2020)."

Reference:

Freund, M. C., Etzel, J. A., & Braver, T. S. (2021). Neural Coding of Cognitive Control: The Representational Similarity Analysis Approach. Trends in Cognitive Sciences, 25(7), 622-638. https://doi.org/10.1016/j.tics.2021.03.011

Marr, D. C. (1982). Vision: A computational investigation into human representation and information processing. New York: W.H. Freeman.

Kikumoto A, Mayr U. (2020). Conjunctive representations that integrate stimuli, responses, and rules are critical for action selection. Proceedings of the National Academy of Sciences, 117(19):10603-10608. https://doi.org/10.1073/pnas.1922166117.

1. Interpretation of measured geometry in 8C

We appreciate the inclusion of the measured similarity matrices of area 8C, the key area the results focus on, to the supplemental, as this allows for a relatively model-agnostic look at a portion of the data. Interestingly, the measured similarity matrix seems to mismatch the cognitive space model in a potentially substantive way. Although the model predicts that the "pure" Stroop and Simon conditions will have maximal self-similarity (i.e., the Stroop-Stroop and Simon-Simon cells on the diagonal), these correlations actually seem to be the lowest, by what appears to be a substantial margin (particularly the Stroop-Stroop similarities). What should readers make of this apparent mismatch? Perhaps authors could offer their interpretation on how this mismatch could fit with their conclusions.

Response: We appreciate the reviewer for bringing this to our attention. It is essential to clarify that our conclusions were based on the significant similarity modulation effect observed in our statistical analysis using the cosine similarity model, where we did not distinguish between the within-Stroop condition and the other four withinconflict conditions (Fig. 7A). This means that the representation of conflict type was not biased by the seemingly disparities in the values shown here. Moreover, to specifically address the potential differences between the within-Stroop condition and the other within-conflict conditions, we conducted a mixed-effect model. In this analysis, the primary predictor was the cross-condition difference (0 for within-Stroop condition and 1 for other within-conflict conditions). The results showed no significant cross-condition difference in either the incongruent trials (t = 1.22, p = .23) or the congruent (t = 1.06, p = .29) trials. Thus, we believe the evidence for different similarities is inconclusive in our data and decided not to interpret this numerical difference.

We have added this note in the revised figure caption for Figure S5.

Author response image 5.

Fig. S5. The stronger conflict type similarity effect in incongruent versus congruent conditions. (A) Summary representational similarity matrices for the right 8C region in incongruent (left) and congruent (right) conditions, respectively. Each cell represents the averaged Pearson correlation of cells with the same conflict type and congruency in the 1400×1400 matrix. Note that the seemingly disparities in the values of Stroop and other within-conflict cells (i.e., the diagonal) did not reach significance for either incongruent (t = 1.22, p = .23) or congruent (t = 1.06, p = .29) trials. (2) Scatter plot showing the averaged neural similarity (Pearson correlation) as a function of conflict type similarity in both conditions. The values in both A and B are calculated from raw Pearson correlation values, in contrast to the z-scored values in Fig. 4D.

1. It would likely improve clarity if all of the competing models were displayed as summarized RSA matrices in a single figure, similar to (or perhaps combined with) Figure 7.

Response: We appreciate the reviewer’s suggestion. We now have incorporated the domain-general and domain-specific models into the Figure 7 (now Figure 8).

Author response image 6.

Figure 8. Schematic of key RSMs. (A) and (B) show the orthogonality between conflict similarity and orientation RSMs. The within-subject RSMs (e.g., Group1-Group1) for conflict similarity and orientation are all the same, but the cross-group correlations (e.g., Group2-Group1) are different. Therefore, we can separate the contribution of these two effects when including them as different regressors in the same linear regression model. (C) and (D) show the two alternative models. Like the cosine model (A), within-group trial pairs resemble between-group trial pairs in these two models. The domain-specific model is an identity matrix. The domain-general model is estimated from the absolute difference of behavioral congruency effect, but scaled to 0(lowest similarity)-1(highest similarity) to aid comparison. The plotted matrices here include only one subject each from Group 1 and Group 2. Numbers 1-5 indicate the conflict type conditions, for spatial Stroop, StHSmL, StMSmM, StLSmH, and Simon, respectively. The thin lines separate four different sub-conditions, i.e., target arrow (up, down) × congruency (incongruent, congruent), within each conflict type.

1. Because this model comparison is key to the main inferences in the study, it might also be helpful for most readers to move all of these RSA model matrices to the main text, instead of in the supplemental.

Response: We thank the reviewer for this suggestion. We have moved the Fig. S4 to the main text, labeled as the new Figure 7.

1. It may be worthwhile to check how robust the observed brain-behavior association (Fig 4C) is to the exclusion of the two datapoints with the lowest neural representation strength measure, as these points look like they have high leverage.

Response: We calculated the Pearson correlation after excluding the two points and found it does not affect the results too much, with the r = 0.50, p = .003 (compared to the original r = 0.52, p = .001).

Additionally, we found the two axes were mistakenly shifted in Fig 4C. Therefore, we corrected this error in the revised manuscript. The correlation results would not be influenced.

Author response image 7.

Fig. 4. The conflict type effect. (A) Brain regions surviving the Bonferroni correction (p < 0.0001) across the regions (criterion 1). Labeled regions are those meeting the criterion 2. (B) Different encoding of conflict type in the incongruent with congruent conditions. * Bonferroni corrected p < .05. (C) The brain-behavior correlation of the right 8C (criterion 3). The x-axis shows the beta coefficient of the conflict type effect from the RSA, and the y-axis shows the beta coefficient obtained from the behavioral linear model using the conflict similarity to predict the CSE in Experiment 2. (D) Illustration of the different encoding strength of conflict type similarity in incongruent versus congruent conditions of right 8C. The y-axis is derived from the z-scored Pearson correlation coefficient, consistent with the RSA methodology. See Fig. S4B for a plot with the raw Pearson correlation measurement. l = left; r = right.

Reviewer #3:

Yang and colleagues investigated whether information on two task-irrelevant features that induce response conflict is represented in a common cognitive space. To test this, the authors used a task that combines the spatial Stroop conflict and the Simon effect. This task reliably produces a beautiful graded congruency sequence effect (CSE), where the cost of congruency is reduced after incongruent trials. The authors measured fMRI to identify brain regions that represent the graded similarity of conflict types, the congruency of responses, and the visual features that induce conflicts. They applied univariate, multivariate, and connectivity analyses to fMRI data to identify brain regions that represent the graded similarity of conflict types, the congruency of responses, and the visual features that induce conflicts. They further directly assessed the dimensionality of represented conflict space.

The authors identified the right dlPFC (right 8C), which shows 1) stronger encoding of graded similarity of conflicts in incongruent trials and 2) a positive correlation between the strength of conflict similarity type and the CSE on behavior. The dlPFC has been shown to be important for cognitive control tasks. As the dlPFC did not show a univariate parametric modulation based on the higher or lower component of one type of conflict (e.g., having more spatial Stroop conflict or less Simon conflict), it implies that dissimilarity of conflicts is represented by a linear increase or decrease of neural responses. Therefore, the similarity of conflict is represented in multivariate neural responses that combine two sources of conflict.

The strength of the current approach lies in the clear effect of parametric modulation of conflict similarity across different conflict types. The authors employed a clever cross-subject RSA that counterbalanced and isolated the targeted effect of conflict similarity, decorrelating orientation similarity of stimulus positions that would otherwise be correlated with conflict similarity. A pattern of neural response seems to exist that maps different types of conflict, where each type is defined by the parametric gradation of the yoked spatial Stroop conflict and the Simon conflict on a similarity scale. The similarity of patterns increases in incongruent trials and is correlated with CSE modulation of behavior.

The main significance of the paper lies in the evidence supporting the use of an organized "cognitive space" to represent conflict information as a general control strategy. The authors thoroughly test this idea using multiple approaches and provide convincing support for their findings. However, the universality of this cognitive strategy remains an open question.

(Public Reviews) Taken together, this study presents an exciting possibility that information requiring high levels of cognitive control could be flexibly mapped into cognitive map-like representations that both benefit and bias our behavior. Further characterization of the representational geometry and generalization of the current results look promising ways to understand representations for cognitive control.

Response: We would like to thank the reviewer for the positive evaluation of our manuscript and for providing constructive comments. In response to your suggestions, we have acknowledged the potential limitation of the design and the cross-subject RSA approach, and incorporated the open questions to the discussions. Please find our detailed responses below.

The task presented in the study involved two sources of conflict information through a single salient visual input, which might have encouraged the utilization of a common space.

Response: We agree that the unified visual input in our design may have facilitated the utilization of a common space. However, we believe the stimuli are not necessarily unified in the construction of the common space. To further test the potential interaction between the concrete stimulus setting and the cognitive space representation, it is necessary to use varied stimuli in future research. We have left this as an open question in the discussion:

Can we effectively map any sources of conflict with completely different stimuli into a single space?

The similarity space was analyzed at the level of between-individuals (i.e., crosssubject RSA) to mitigate potential confounds in the design, such as congruency and the orientation of stimulus positions. This approach makes it challenging to establish a direct link between the quality of conflict space representation and the patterns of behavioral adaptations within individuals.

Response: By setting the variables as random effects at the subject level, we have extracted the individual effects that incorporate both the group-level fixed effects and individual-level random effects. We believe this approach yields results that are as reliable, if not more, than effects calculated from individual data only. First, the mixed effect linear (LME) model has included all the individual data, forming the basis for establishing random effects. Therefore, the individual effects derived from this approach inherently reflect the individual-specific effects. To support this notion, we have included a simulation script (accessible in the online file “simulation_LME.mlx” at https://osf.io/rcq8w) to demonstrate the strong consistency between the two approaches (see Author response image 8). In this simulation, we generated random data (Y) for 35 subjects, each containing 20 repeated measurements across 5 conditions. To streamline the simulation, we only included one predictor (X), which was treated as both fixed and random effects at the subject level. We applied two methods to calculate the individual beta coefficient. The first involved extracting individual beta coefficients from the LME model by summing the fixed effect with the subject-specific random effect. The second method was entailed conducting a regression analysis using data from each subject to obtain the slope. We tested their consistency by calculating the Pearson correlation between the derived beta coefficients. This simulation was repeated 100 times.

Author response image 8.

The consistent individual beta coefficients between the mixed effect model and the individual regression analysis. A) The distribution of Pearson correlation between the two methods for 100 times. B) An example from the simulation showing the highly correlated results from the two methods. Each data point indicates a subject (n=35).

Second, the potential difference between the two methods lies in that the LME model have also taken the group-level variance into account, such as the dissociable variances of the conflict similarity and orientation across subject groups. This enabled us to extract relatively cleaner conflict similarity effects for each subject, which we believe can be better linked to the individual behavioral adaptations. Moreover, we have extracted the behavioral adaptations scores (i.e., the similarity modulation effect on CSE) using a similar LME approach. Conducting behavioral analysis solely using individual data would have been less reliable, given the limited sample size of individual data (~32 points per subject). This also motivated us to maintain consistency by extracting individual neural effects using LME models.

Furthermore, it remains unclear at which cognitive stages during response selection such a unified space is recruited. Can we effectively map any sources of conflict into a single scale? Is this unified space adaptively adjusted within the same brain region? Additionally, does the amount of conflict solely define the dimensions of this unified space across many conflict-inducing tasks? These questions remain open for future studies to address.

Response: We appreciate the reviewer’s constructive open questions. We respond to each of them based on our current understanding.

1) It remains unclear at which cognitive stages during response selection such a unified space is recruited.

We anticipate that the cognitive space is recruited to guide the transference of behavioral CSE at two critical stages. The first stage involves the evaluation of control demands, where the representational distance/similarity between previous and current trials influences the adjustment of cognitive control. The second stage pertains to is control execution, where the switch from one control state to another follows a path within the cognitive space. It is worth noting that future studies aiming to address this question may benefit from methodologies with higher temporal resolutions, such as EEG and MEG, to provide more precise insights into the temporal dynamics of the process of cognitive space recruitment.

2) Can we effectively map any sources of conflict into a single scale?

It is possible that various sources of conflict can be mapped onto the same space based on their similarity, even if finding such an operational defined similarity may be challenging. However, our results may offer an approach to infer the similarity between two conflicts. One way is to examine their congruency sequence effect (CSE), with a stronger CSE suggesting greater similarity. The other way is to test their representational similarity within the dorsolateral prefrontal cortex.

3) Is this unified space adaptively adjusted within the same brain region? We do not have an answer to this question. We showed that the cognitive space does not change with time (Note. S3). What have adjusted is the control demand to resolve the quickly changing conflict conditions from trial to trial. Though, it is an interesting question whether the cognitive space may be altered, for example, when the mental state changes significantly. And if yes, we can further test whether the change of cognitive space is also within the right dlPFC.

4) Additionally, does the amount of conflict solely define the dimensions of this unified space across many conflict-inducing tasks?

Our understanding of this comment is that the amount of conflict refers to the number of conflict sources. Based on our current finding, the dimensions of the space are indeed defined by how many different conflict sources are included. However, this would require the different conflict sources are orthogonal. If some sources share some aspects, the cognitive space may collapse to a lower dimension. We have incorporated the first question into the discussion:

Moreover, we anticipate that the representation of cognitive space is most prominently involved at two critical stages to guide the transference of behavioral CSE. The first stage involves the evaluation of control demands, where the representational distance/similarity between previous and current trials influences the adjustment of cognitive control. The second stage pertains to control execution, where the switch from one control state to another follows a path within the cognitive space. However, we were unable to fully distinguish between these two stages due to the low temporal resolution of fMRI signals in our study. Future research seeking to delve deeper into this question may benefit from methodologies with higher temporal resolutions, such as EEG and MEG.

We have included the other questions into the manuscript as open questions, calling for future research.

Several interesting questions remains to be answered. For example, is the dimension of the unified space across conflict-inducing tasks solely determined by the number of conflict sources? Can we effectively map any sources of conflict with completely different stimuli into a single space? Is the cognitive space geometry modulated by the mental state? If yes, what brain regions mediate the change of cognitive space?

Minor comments:

• The original comment about out-of-sample predictions to examine the continuity of the space was a suggestion for testing neural representations, not behavior (I apologize for the lack of clarity). Given the low dimensionality of the conflict space shown by the participation ratio, we expect that linear separability exists only among specific combinations of conditions. For example, the pair of conflicts 1 and 5 together is not linearly separable from conflicts 2 and 3. But combined with other results, this is already implied.

Response: We apologize for the misunderstanding. In fact, performing a prediction analysis using the extensive RSM in our study does presents certain challenges, primarily due to its substantial size (1400x1400) and the intricate nature of the mixed-effect linear model. In our efforts to simplify the prediction process by excluding random effects, we did observe a correlation between the predicted and original values, albeit a relatively small Pearson correlation coefficient of r = 0.024, p < .001. This small correlation can be attributed to two key factors. First, the exclusion of data points impacts not only the conflict similarity regressor but also other regressors within the model, thereby diminishing the predictive power. Secondly, the large amount of data points in the model heightens the risk of overfitting, subsequently reducing the model’s capacity for generalization and increasing the likelihood of unreliable predictions. Given these potential problems, we have opted not to include this prediction in the revised manuscript.

Author Response

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

eLife assessment:

This important study advances the understanding of physiological mechanisms in deep-sea Planctomycetes bacteria, revealing unique characteristics such as the only known Phycisphaerae using a budding mode of division, extensive involvement in nitrate assimilation and release phage particles without cell death. The study uses convincing evidence, based on experiments using growth assays, phylogenetics, transcriptomics, and gene expression data. The work will be of interest to bacteriologists and microbiologists in general.

Response: Thanks for the Editor’s and Reviewers’ positive comments, which help us improve the quality of our manuscript entitled “Physiological and metabolic insights into the first cultured anaerobic representative of deep-sea Planctomycetes bacteria” (paper#eLife-RP-RA-2023-89874). The comments are all valuable, and we have studied the comments carefully and have made corresponding revisions according to the suggestions. Revised portions are marked in blue in the modified manuscript.

Please find the detailed responses as following.

Public Reviews:

Reviewer #1 (Public Review):

The authors of the manuscript cultivated a Planctomycetes strain affiliated with Phycisphaerae. The strain was one of the few Planctomycetes from deep-sea environments and demonstrated several unique characteristics, such as being the only known Phycisphaerae using a budding mode of division, extensive involvement in nitrate assimilation, and being able to release phage particles without cell death. The manuscript is generally well-written. However, a few issues need to be more clearly addressed, especially regarding the identification and characterization of the phage.

Response: Thanks for your positive comments. Please find the detailed responses as following.

Reviewer #1 (Recommendations For The Authors):

• Line 75-77, add a reference for this statement.

Response: Thanks for your suggestion. We have added a reference (Fuerst and Sagulenko, 2011) for this statement in the revised manuscript (Line 77).

References related to this response:

Fuerst, J.A., and Sagulenko, E. Beyond the bacterium: planctomycetes challenge our concepts of microbial structure and function. Nat Rev Microbiol. 2011;9:403-413.

• Line 124-134, add key statistics (such as ANI) of strain ZRK32 and KS4 to this section.

Response: Thanks for your suggestion. We added the key statistics of strain ZRK32 and KS4, and described as “Based on the 16S rRNA sequence of strain ZRK32, a sequence similarity calculation using the NCBI server indicated that the closest relatives of strain ZRK32 were Poriferisphaera corsica KS4T (98.06%), Algisphaera agarilytica 06SJR6-2T (88.04%), Phycisphaera mikurensis NBRC 102666T (85.28%), and Tepidisphaera mucosa 2842T (82.94%). Recently, the taxonomic threshold for species based on 16S rRNA gene sequence identity value was 98.65% (Kim et al., 2014). Based on these criteria, we proposed that strain ZRK32 might be a novel representative of the genus Poriferisphaera. In addition, to clarify the phylogenetic position of strain ZRK32, the genome relatedness values were calculated by the average nucleotide identity (ANI), the tetranucleotide signatures (Tetra), and in silico DNA-DNA similarity (isDDH), against the genomes of strains ZRK32 and KS4. The ANIb, ANIm, Tetra, and isDDH values were 72.89%, 85.34%, 0.97385, and 20.90%, respectively (Table S1). These results together demonstrated the strain ZRK32 genome to be obviously below established ‘cut-off’ values (ANIb: 95%, ANIm: 95%, Tetra: 0.99, isDDH: 70%) for defining bacterial species, suggesting strain ZRK32 represents a novel strain within the genus Poriferisphaera.” in the revised manuscript (Lines 124-139).

• Fig. 2A missing description for figure key.

Response: Thanks for your comments. We modified the Figure 2A, shown as below:

Author response image 1.

Figure. 2. Growth assay and transcriptomic analysis of P. heterotrophicis ZRK32 strains cultivated in basal medium and rich medium.

• Regarding the page released, could this be a membrane vesicle-engulfed phage? I would recommend checking "Spontaneous Prophage Induction Contributes to the Production of Membrane Vesicles by the Gram-Positive Bacterium Lacticaseibacillus casei BL23" and "Chronic Release of Tailless Phage Particles from Lactococcus lactis" for further references.

Response: Thanks for your valuable comments. We carefully read these two papers and found that phage ZRK32 is most likely a membrane vesicle-engulfed phage. We added the corresponding description as “Moreover, it has recently been reported that the tailless Caudoviricetes phage particles are enclosed in lipid membrane and are released from the host cells by a nonlytic mechanism (Liu et al., 2022), and the prophage induction contributes to the production of membrane vesicles by Lacticaseibacillus casei BL23 during cell growth (da Silva Barreira et al., 2022). Considering that strain ZRK32 has a large number of membrane vesicles during cell growth (Figure S9), we speculated that Phage-ZRK32 might be a membrane vesicle-engulfed phage and its release should be related to membrane vesicles.” in the revised manuscript (Lines 381-388).

References related to this response:

Liu Y, Alexeeva S, Bachmann H, Guerra Martníez J.A, Yeremenko N, Abee T et al. Chronic release of tailless phage particles from Lactococcus lactis. Appl Environ Microbiol. 2022; 88: e0148321.

Silva Barreira, D., Lapaquette, P., Novion Ducassou, J., Couté, Y., Guzzo, J., and Rieu, A. Spontaneous prophage induction contributes to the production of membrane vesicles by the gram-positive bacterium Lacticaseibacillus casei BL23. mBio. 2022;13:e0237522.

• How were the reference sequences for Fig. S10-S13 retrieved, was it by blasting the phage gene against the entire NCBI database, or only the virus sequence within the NCBI? Please clarify this.

Response: Thanks for your comments. The reference sequences for Fig. S10-S13 were retrieved by blasting the phage gene against the entire NCBI database. We clarified this as “The reference sequences of four AMGs encoding amidoligase, glutamine amidotransferase, gamma-glutamylcyclotransferase, and glutathione synthase were retrieved by blasting the phage gene against the entire NCBI database, respectively.” in the revised manuscript (Lines 444-447).

Reviewer #2 (Public Review):

Summary:

Planctomycetes encompass a group of bacteria with unique biological traits, the compartmentalized cells make them appear to be organisms in between prokaryotes and eukaryotes. However, only a few of the Planctomycetes bacteria are cultured thus far, and this hampers insight into the biological traits of these evolutionarily important organisms. This work reports the methodology details of how to isolate the deep-sea bacteria that could be recalcitrant to laboratory cultivation, and further reveals the distinct characteristics of the new species of a deep-sea Planctomycetes bacterium, such as the chronic phage release without breaking the host and promote the host and related bacteria in nitrogen utilization. Therefore, the finding of this work is of importance in extending our knowledge of bacteria.

Response: Thanks for your positive comments.

Strengths:

Through the combination of microscopic, physiological, genomics, and molecular biological approaches, this reports the isolation and comprehensive investigation of the first anaerobic representative of the deep-sea Planctomycetes bacterium, in particular in that of the budding division, and release phage without lysis of the cells. Most of the results and conclusions are supported by the experimental evidence.

Response: Thanks for your positive comments.

Weaknesses:

1. While EMP glycolysis is predicted to be involved in energy conservation, no experimental evidence indicated any sugar utilization by the bacterium.

Response: Thanks for your comments. We have previously tested the sugar utilization of strain ZRK32, and now added this description as “Consistent with the presence of EMP glycolysis pathway in strain ZRK32, we found that it could use a variety of sugars including glucose, maltose, fructose, isomaltose, galactose, D-mannose, and rhamnose (Table S2).” in the revised manuscript (Lines 281-284).

1. "anaerobic representative" is indicated in the Title, the contrary, TCA in energy metabolism is predicted by the bacterium.

Response: Thanks for your valuable comments. Currently, anaerobic microorganisms can use other alternative electron acceptors (such as sulfate reducers, nitrate reducers, iron reducers, etc) in place of oxygen for the TCA cycle. For example, Proteus mirabilis uses the whole oxidative TCA cycle without using oxygen as the final electron acceptor when it performs multicellular swarming (Alteri et al., 2012). In this study, all the genes involved in the TCA cycle were present in anaerobic strain ZRK32 and most of them are upregulated, thus we speculate that it might function through the complete TCA metabolic pathway to obtain energy. We added the related description as “Notably, when growing in the rich medium, the expressions of most genes involved in the TCA cycle and EMP glycolysis pathway in strain ZRK32 were upregulated (Figure 2B-D, Figure S5B and Figure S6), suggesting that strain ZRK32 might function through the complete TCA metabolic pathway and EMP glycolysis pathway to obtain energy for growth (Figure S8) (Zheng et al., 2021b). Consistent with the presence of EMP glycolysis pathway in strain ZRK32, we found that it could use a variety of sugars including glucose, maltose, fructose, isomaltose, galactose, D-mannose, and rhamnose (Table S2). As for the presence of TCA cycle in the anaerobic strain ZRK32, we propose that it might use other alternative electron acceptors (such as sulfate reducers, nitrate reducers, iron reducers, etc) in place of oxygen for the TCA cycle, as shown in other anaerobic bacteria (Alteri et al., 2012).” in the revised manuscript (Lines 277-287).

References related to this response:

Alteri CJ, Himpsl SD, Engstrom MD, Mobley HL. Anaerobic respiration using a complete oxidative TCA cycle drives multicellular swarming in Proteus mirabilis. mBio. 2012; 3(6): e00365-12.

1. The possible mechanisms of the chronic phage release without breaking the host are not discussed.

Response: Thanks for your valuable comments. The possible mechanism of the chronic phage release without breaking the host might be that it was enclosed in lipid membrane and released from the host cells by a nonlytic mechanism. We added the corresponding description as “Moreover, it has recently been reported that the tailless Caudoviricetes phage particles are enclosed in lipid membrane and are released from the host cells by a nonlytic mechanism (Liu et al., 2022), and the prophage induction contributes to the production of membrane vesicles by Lacticaseibacillus casei BL23 during cell growth (da Silva Barreira et al., 2022). Considering that strain ZRK32 has a large number of membrane vesicles during cell growth (Figure S9), we speculated that Phage-ZRK32 might be a membrane vesicle-engulfed phage and its release should be related to membrane vesicles.” in the revised manuscript (Lines 381-388).

References related to this response:

Liu Y, Alexeeva S, Bachmann H, Guerra Martníez J.A, Yeremenko N, Abee T et al. Chronic release of tailless phage particles from Lactococcus lactis. Appl Environ Microbiol. 2022; 88: e0148321. da Silva Barreira, D., Lapaquette, P., Novion Ducassou, J., Couté, Y., Guzzo, J., and Rieu, A. Spontaneous prophage induction contributes to the production of membrane vesicles by the gram-positive bacterium Lacticaseibacillus casei BL23. mBio. 2022;13:e0237522.

Reviewer #2 (Recommendations For The Authors):

• Have you tested whether strain ZRK32 uses any sugars? If not, why it uses EMP pathway to obtain energy?

Response: Thanks for your comments. We have previously tested the sugar utilization of strain ZRK32, and now added this description as “Consistent with the presence of EMP glycolysis pathway in strain ZRK32, we found that it could use a variety of sugars including glucose, maltose, fructose, isomaltose, galactose, D-mannose, and rhamnose (Table S2).” in the revised manuscript (Lines 281-284).

• Further discussion on possible mechanisms of the chronic phage release without breaking the host is expected.

Response: Thanks for your valuable comments. The possible mechanism of the chronic phage release without breaking the host might be that it was enclosed in lipid membrane and released from the host cells by a nonlytic mechanism. We added the corresponding description as “Moreover, it has recently been reported that the tailless Caudoviricetes phage particles are enclosed in lipid membrane and are released from the host cells by a nonlytic mechanism (Liu et al., 2022), and the prophage induction contributes to the production of membrane vesicles by Lacticaseibacillus casei BL23 during cell growth (da Silva Barreira et al., 2022). Considering that strain ZRK32 has a large number of membrane vesicles during cell growth (Figure S9), we speculated that Phage-ZRK32 might be a membrane vesicle-engulfed phage and its release should be related to membrane vesicles.” in the revised manuscript (Lines 381-388).

References related to this response:

Liu Y, Alexeeva S, Bachmann H, Guerra Martníez J.A, Yeremenko N, Abee T et al. Chronic release of tailless phage particles from Lactococcus lactis. Appl Environ Microbiol. 2022; 88: e0148321.

da Silva Barreira, D., Lapaquette, P., Novion Ducassou, J., Couté, Y., Guzzo, J., and Rieu, A. Spontaneous prophage induction contributes to the production of membrane vesicles by the gram-positive bacterium Lacticaseibacillus casei BL23. mBio. 2022;13:e0237522.

• It is recommended that the writing is improved, including presentation style and grammar.

Response: Thanks for your comments. We have invited an English native speaker (Dr. Diana Walsh from Life Science Editors, USA) to revise our manuscript, which we hope to meet your approval.

Author Response

We are delighted that eLife has assessed our study as a valuable contribution as well as appreciating the importance of working on asymptomatic reservoirs of P. falciparum in high transmission where not just children, but adolescents and adults harbor multiclonal infections. The constructive public reviews will serve to improve our manuscript.

Detailed responses to referees’ comments and a revised manuscript are forthcoming. Here we make a provisional response to three key areas addressed by the referees:

(1) census population size

Referee 1 raises important questions although we respectfully disagree on the terminology we have adopted (of “census”) and on the unclear utility of the proposed quantity.

We consider the quantity a census in that it is a total enumeration or count of the infections in a given population sample and over a given time period. In this sense, it gives us a tangible notion of the size of the parasite population, in an ecological sense, distinct from the formal effective population size used in population genetics. Given the low overlap between var repertoires of parasites (as observed in monoclonal infections), the population size we have calculated translates to a diversity of strains or repertoires. But our focus here is in a measure of population size itself. The distinction between population size in terms of infection counts and effective population size from population genetics has been made before for pathogens (see for example Bedford et al. 2011 for the seasonal influenza virus and for the measles virus) and is a clear one in the ecological literature for non-pathogen populations (Palstra et al. 2012).

Both referees 1 and 2 point out that census population size will be sensitive to sample size. We completely agree with the dependence of our quantity on sample size. We used it for comparisons across time of samples of the same depth, to describe the large population size characteristic of high transmission, and persistent across the IRS intervention. Of course, one would like to be able to use this notion across studies that differ in sampling depth.

Here, referee 1 makes an insightful and useful suggestion. It is true that we can use mean MOI, and indeed there is a simple map between our population size and mean MOI (as we just need to divide or multiply by sample size). We can do even more, as with mean MOI we can presumably extrapolate to the full sample size of the host population, or the population size of another sample in another location. What is needed for this purpose is a stable mean MOI relative to sample size. We can show that indeed in our study mean MOI is stable in that way, by subsampling to different depths of our original sample. We will include in the revision discussion of this point and result, which allows an extrapolation of the census population size to the whole population of hosts in the local area. We’ll also clarify the time denominator, as given the typical duration of infections, we expect our population size to be representative of a per-generation measure.

Referee 2 suggests we adopt the term “census count” but as a census in our mind is a count we prefer to use “census”.

Referee 3 considers the genetic data tracking parasite MOI and census changes gives the same result as prevalence which tracks infected hosts. Respectfully, we disagree and will provide an expanded response.

(2) the importance of lineages (in response to referee 2)

We do not think that lineages moving exclusively through a given type of host or “patch” is a requirement for enumerating the size of the total infections in such a subset. It is true that what we have is a single parasite population, but we are enumerating for the season the respective size in host classes (children and adults). This is akin to enumerating subsets of a population in ecological settings.

We are also not clear on the concept of lineage for these highly recombinant parasites as we struggle to find highly related repertoires. In fact, we see the use of the var fingerprinting methodology as a means to capture changes in strain or var repertoires dynamics as a result of changing transmission conditions.

(3) var methodology

Comments and queries were made by all three referees about aspects of var methodology, including the Bayesian approach. These will be addressed in our full response.

Here we respond to a very good point made by referee 2: “Thinking about the applicability of this approach to other studies, I would be interested in a larger treatment of how overlapping DBLa repertoires would impact MOIvar estimates. Is there a definable upper bound above which the method is unreliable? Alternatively, can repertoire overlap be incorporated into the MOI estimator?”

There is no predefined threshold one can present a priori. Intuitively, the approach to estimate MOI would appear to breakdown as overlap moves away from extremely low, and therefore, for locations with lower transmission intensity. Interestingly, we have observed that this is not the case in our paper by Labbé et al. 2023 where we used model simulations in a gradient of three transmission intensities, from high to low. The original varcoding method performed well across the gradient. This may arise from a nonlinear and fast transition from low overlap to high overlap that is accompanied by the MOI transitioning quickly from primarily multiclonal (MOI > 1) to monoclonal (MOI = 1). This issue needs to be investigated further, including ways to extend the estimation to explicitly include the distribution of DBL repertoire overlap.

References: Bedford T, Cobey S, Pascual, M. 2011. Strength and tempo of selection revealed in viral gene genealogies. BMC Evol Biol 11, 220. https://doi.org/10.1186/1471-2148-11-220

Labbé F, He Q, Zhan Q, Tiedje KE, Argyropoulos DC, Tan MH, Ghansah A, Day KP, Pascual M. 2023. Neutral vs . non-neutral genetic footprints of Plasmodium falciparum multiclonal infections. PLoS Comput Biol 19 :e1010816. doi:doi.org/10.1101/2022.06.27.49780

Palstra FP, Fraser DJ. 2012. Effective/census population size ratio estimation: a compendium and appraisal. Ecol Evol. Sep;2(9):2357-65. doi:10.1002/ece3.329.

Author Response

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

eLife assessment

This research advance arctile describes a valuable image analysis method to identify individual cells (neurons) within a population of fluorescently labeled cells in the nematode C. elegans. The findings are solid and the method succeeds to identify cells with high precision. The method will be valuable to the C. elegans research community.

Public Reviews:

Reviewer #1 (Public Review):

In this paper, the authors developed an image analysis pipeline to automatically identify individual neurons within a population of fluorescently tagged neurons. This application is optimized to deal with multi-cell analysis and builds on a previous software version, developed by the same team, to resolve individual neurons from whole-brain imaging stacks. Using advanced statistical approaches and several heuristics tailored for C. elegans anatomy, the method successfully identifies individual neurons with a fairly high accuracy. Thus, while specific to C. elegans, this method can become instrumental for a variety of research directions such as in-vivo single-cell gene expression analysis and calcium-based neural activity studies.

The analysis procedure depends on the availability of an accurate atlas that serves as a reference map for neural positions. Thus, when imaging a new reporter line without fair prior knowledge of the tagged cells, such an atlas may be very difficult to construct. Moreover, usage of available reference atlases, constructed based on other databases, is not very helpful (as shown by the authors in Fig 3), so for each new reporter line a de-novo atlas needs to be constructed.

We thank the reviewer for pointing out a place where we can use some clarification. While in principle that every new reporter line would need fair prior knowledge, atlases are either already available or not difficult to construct. If one can make the assumption that the anatomy of a particular line is similar to existing atlases (Yemini 2021,Nejatbakhsh 2023,Toyoshima 2020), the cell ID can be immediately performed. Even in the case that one suspects the anatomy may have changes from existing atlases (e.g. in the case of examining mutants), existing atlases can serve as a starting point to provide a draft ID, which facilitates manual annotation. Once manual annotations on ~5 animals are available as we have shown in this work (which is a manageable number in practice), this new dataset can be used to build an updated atlas that can be used for future inferences. We have added this discussion in the manuscript: “If one determines that the anatomy of a particular animal strain is substantially different from existing atlases, new atlases can be easily constructed using existing atlases as starting points.” (page 18).

I have a few comments that may help to better understand the potential of the tool to become handy.

1. I wonder the degree by which strain mosaicism affects the analysis (Figs 1-4) as it was performed on a non-integrated reporter strain. As stated, for constructing the reference atlas, the authors used worms in which they could identify the complete set of tagged neurons. But how senstiive is the analysis when assaying worms with different levels of mosaicism? Are the results shown in the paper stem from animals with a full neural set expression? Could the authors add results for which the assayed worms show partial expression where only 80%, 70%, 50% of the cells population are observed, and how this will affect idenfication accuracy? This may be important as many non-integrated reporter lines show high mosaic patterns and may therefore not be suitable for using this analytic method. In that sense, could the authors describe the mosaic degree of their line used for validating the method.

We appreciate the reviewer for this comment. We want to clarify that most of the worms used in the construction of the atlas are indeed affected by mosaicism and thus do not express the full set of candidate neurons. We have added such a plot as requested (Figure 3 – figure supplement 2, copied below). Our data show that there is no correlation between the fraction of cells expressed in a worm and neuron ID correspondence. We agree with the reviewer this additional insight may be helpful; we have modified the text to include this discussion: “Note that we observed no correlation between the degree of mosaicism and neuron ID correspondence (Figure 3- figure supplement 2).” (page 10).

Author response image 1.

No correlation between the degree of mosaicism (fraction of cells expressed in the worm) and neuron ID correspondence.

1. For the gene expression analysis (Fig 5), where was the intensity of the GFP extracted from? As it has no nuclear tag, the protein should be cytoplasmic (as seen in Fig 5a), but in Fig 5c it is shown as if the region of interest to extract fluorescence was nuclear. If fluorescence was indeed extracted from the cytoplasm, then it will be helpful to include in the software and in the results description how this was done, as a huge hurdle in dissecting such multi-cell images is avoiding crossreads between adjacent/intersecting neurons.

For this work, we used nuclear-localized RFP co-expressed in the animal, and the GFP intensities were extracted from the same region RFP intensities were extracted. If cytosolic reporters are used, one would imagine a membrane label would be necessary to discern the border of the cells. We clarified our reagents and approach in the text: “The segmentation was done on the nuclear-localized mCherry signals, and GFP intensities were extracted from the same region.” (page21).

1. In the same mater: In the methods, it is specified that the strain expressing GCAMP was also used in the gene expression analysis shown in Figure 5. But the calcium indicator may show transient intensities depending on spontaneous neural activity during the imaging. This will introduce a significant variability that may affect the expression correlation analysis as depicted in Figure 5.

We apologize for the error in text. The strain used in the gene expression analysis did not express GCaMP. We did not analyze GCaMP expression in figure 5. We have corrected the error in the methods.

Reviewer #2 (Public Review):

The authors succeed in generalizing the pre-alignment procedure for their cell idenfication method to allow it to work effectively on data with only small subsets of cells labeled. They convincingly show that their extension accurately identifies head angle, based on finding auto fluorescent tissue and looking for a symmetric l/r axis. They demonstrate that the method works to identify known subsets of neurons with varying accuracy depending on the nature of underlying atlas data. Their approach should be a useful one for researchers wishing to identify subsets of head neurons in C. elegans, for example in whole brain recording, and the ideas might be useful elsewhere.

The authors also strive to give some general insights on what makes a good atlas. It is interesting and valuable to see (at least for this specific set of neurons) that 5-10 ideal examples are sufficient. However, some critical details would help in understanding how far their insights generalize. I believe the set of neurons in each atlas version are matched to the known set of cells in the sparse neuronal marker, however this critical detail isn't explicitly stated anywhere I can see.

This is an important point. We have made text modifications to make it clear to the readers that for all atlases, the number of entities (candidate list) was kept consistent as listed in the methods. In the results section under “CRF_ID 2.0 for automatic cell annotation in multi-cell images,” we added the following sentence: “Note that a truncated candidate list can be used for subse-tspecific cell ID if the neuronal expression is known” (page 3). In the methods section, we added the following sentence: “For multi-cell neuron predictions on the glr-1 strain, a truncated atlas containing only the above 37 neurons was used to exclude neuron candidates that are irrelevant for prediction” (Page 20).

In addition, it is stated that some neuron positions are missing in the neuropal data and replaced with the (single) position available from the open worm atlas. It should be stated how many neurons are missing and replaced in this way (providing weaker information).

We modified the text in the result section as follows: “Eight out of 37 candidate neurons are missing in the neuroPAL atlas, which means 40% of the pairwise relationships of neurons expressing the glr-1p::NLS-mcherry transgene were not augmented with the NeuroPAL data but were assigned the default values from the OpenWorm atlas” (page 10).

It also is not explicitly stated that the putative identities for the uncertain cells (designated with Greek letters) are used to sample the neuropal data. Large numbers of openworm single positions or if uncertain cells are misidentified forcing alignment against the positions of nearby but different cells would both handicap the neuropal atlas relative to the matched florescence atlas. This is an important question since sufficient performance from an ideal neuropal atlas (subsampled) would avoid the need for building custom atlases per strain.

The putative identities are not used to sample the NeuroPAL data. They were used in the glr-1 multi-cell case to indicate low confidence in manual identification/annotation. For all steps of manual annotation and CRF_ID predictions, we used real neuron labels, and the Greek labels were used for reporting purposes only. It is true that the OpenWorm values (40% of the atlas) would be a handicap for the neuroPAL atlas. This is mainly due to the difficulty of obtaining NeuroPAL data as it requires 3-color fluorescence microscopy and significant time and labor to annotate the large set of neurons. This is one reason to take a complementary approach as we do in this paper.

Reviewer #1 (Recommendations For The Authors):

1. Figure 3, there is a confusion in the legend relating to panels c-e (e.g. panel c is neuron ID accuracy but it is described per panel e in the legend.

We made the necessary changes.

1. Figure 3, were statistical tests performed for panels d-e? if so, and the outcome was not significant, then it might be good to indicate this in the legend.

We have added results of statistical tests in the legend as the following sentence: “All distributions in panel d and e had a p-value of less than 0.0001 for one sample t-test against zero.” One sample t-tests were performed because what is plotted already represents each atlas’ differences to the glr-1 25 dataset atlas, we didn’t think the statistical analyses between the other atlases would add significant value.

1. Figure 4, no asterisks are shown in the figure so it is possible to remove the sentence in the legend describing what the asterisk stands for.

Thank you. We made the necessary changes.

Reviewer #2 (Recommendations For The Authors):

Comparison with deep learning approaches could be more nuanced and structured, the authors (prior) approach extended here combines a specific set of comparative relationship measurements with a general optimization approach for matching based on comparative expectations. Other measurements could be used whether explicit (like neighbor expectations) or learned differences in embeddings. These alternate measurements would both need to be extensively re-calibrated for different sets of cells but might provide significant performance gains. In addition deep learning approaches don't solve the optimization part of the matching problem, so the authors approach seems to bring something strong to the table even if one is committed to learned methods (necessary I suspect for human level performance in denser cell sets than the relatively small number here). A more complete discussion of these themes might better frame the impact of the work and help readers think about the advantages and disadvantages or different methods for their own data.

We thank the reviewer for bringing up this point. We apologize perhaps not making the point clearer in the original submission. This extension of the original work (Chaudhary et al) is not changing the CRF-based framework, but only augmenting the approach with a better defined set of axes (solely because in multicell and not whole-brain datasets, the sparsity of neurons degrades the axis definition and consequently the neuron ID predictions). We are not fundamentally changing the framework, and therefore all the advantages (over registration-based approaches for example) also apply here. The other purpose of this paper is to demonstrate a couple of use-cases for gene expression analysis, which is common in studies in C. elegans (and other organisms). We hope that by showing a use-case others can see how this approach is useful for their own applications.

We have clarified these points in the paper (page 18). “The fundamental framework has not been changed from CRF_ID 1.0, and therefore the advantages of CRF_ID outlined in the original work apply for CRF_ID 2.0 as well.”

The atribution of anatomical differences to strain is interesting, but seems purely speculative, and somewhat unlikely. I would suspect the fundamentally more difficult nature of aligning N items to M>>N items in an atlas accounts for the differences in using the neuroPAL vs custom atlas here. If this is what is meant, it could be stated more clearly.

It is important to note that the same neuron candidate list (listed in methods) was used for all atlases, so there is no difference among the atlases in terms of the number of cells in the query vs. candidate list. In other words, the same values for M and for N are used regardless of the reference atlas used.

We have preliminary data indicating differences between the NeuroPAL and custom atlas. For instance, the NeuroPAL atlas scales smaller than the custom glr-1 atlas. Since direct comparisons of the different atlases are beyond the scope of this paper, we will leave the exact comparisons for future work. We suspect that the differences are from a combination of differences in anatomy and imaging conditions. While NeuroPAL atlas may not be exactly fitting for the custom dataset, it can serve as a good starting point for guesses when no custom atlases are available, as we have discussed earlier (response to Public Comments from Reviewer 1 Point 1). As explained earlier, we have added these discussions in the paper (see page 18).

I was also left wondering if the random removal of landmarks had to be adjusted in this work given it is (potentially) helping cope with not just occasional weak cells but the systematic loss of most of the cells in the atlas. If the parameters of this part of the algorithm don't influence the success for N to M>>N alignment (here when the neuroPAL or OpenWorm atlas is used) this seems interesting in itself and worth discussing. Conversely, if these parameters were opitmized for the matched atlas and used for the others, this would seem to bias performance results.

We may have failed to make this clear in the main text. As we have stated in our responses in the public review section, we do systematically limit the neuron labels in the candidate list to neurons that are known to be expressed by the promotor. The candidate list, which is kept consistent for all atlases, has more neurons than cells in the query, so it is always an N-to-M matching where M>N. We did not use landmarks, but such usage is possible and will only improve the matching.

We have attempted to clarify these points in the manuscript. In the results section under “CRF_ID 2.0 for automatic cell annotation in multi-cell images,” we added the following sentence: “Note that a truncated candidate list can be used for subset-specific cell ID if the neuronal expression is known” (page 3). In the methods section, we added the following sentence: “For multi-cell neuron predictions on the glr-1 strain, a truncated atlas containing only the above 37 neurons was used to exclude neuron candidates that are irrelevant for prediction” (Page 20).

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

In this study, the authors examined the putative functions of hypothalamic groups identifiable through Foxb1 expression, namely the parvofox Foxb1 of the LHA and the PMd Foxb1, with emphasis on innate defensive responses. First, they reported that chemogenetic activation of Foxb1hypothalamic cell groups led to tachypnea. The authors tend to attribute this effect to the activation of hM3Dq expressed in the parvofox Foxb1 but did not rule out the participation of the PMd Foxb1 cell group which may as well have expressed hM3Dq, particularly considering the large volume (200 nl) of the viral construct injected. It is also noteworthy that the activation of the Foxb1hypothalamic cell groups in this experiment did not alter the gross locomotor activity, such as time spent immobile state. Thus, contrasts with the authors finding on the optogenetic activation of the Foxb1hypothalamic fibers projecting to the dorsolateral PAG. In the second experiment, the authors applied optogenetic ChR2-mediated excitation of the Foxb1+ cell bodies' axonal endings in the dlPAG leading to freezing and, in a few cases, bradycardia as well. The effective site to evoke freezing was the rostral PAGdl, and fibers positioned either ventral or caudal to this target had no response. Considering the pattern of Foxb1hypothalamic cell groups projection to the PAG, the fibers projecting to the rostral PAGdl are likely to arise from the PMd Foxb1 cell group, and not from the parvofox Foxb1 of the LHA. Here it is important to consider that optogenetic ChR2-mediated excitation of the axonal endings is likely to have activated the cell bodies originating these fibers, and one cannot ascertain whether the behavioral effects are related to the activation of the terminals in the PAGdl or the cell bodies originating the projection.

Authors’ reply: We acknowledge and agree about the possibility of backpropagation in ChR2mediated terminal stimulation experiments. We have introduced a paragraph in the discussion section discussing this issue. In short, the observation of an opposing phenotype in ArchT3.0 animals indicates, that the ChR2-mediated phenotype is indeed Foxb1PAG projection specific. This is due to the fact, that the use of light-activated proton pumps for terminal stimulation can not induce backpropagation of an inhibitory effect to the soma. Potential downsides of the use of proton pumps in small compartments as e.g. in the axon are also discussed.

Moreover, activation of PMd CCK cell group, which consists of around 90% of the PMd cells, evokes escape, and not freezing. According to the present findings, a specific population of PMd Foxb1 cells may be involved in producing freezing. In addition, only a small number of the animals with correct fiber placement presented sudden onset of bradycardia in response to the photostimulation. Considering the authors' findings, the Foxb1+ hypothalamic groups are likely to mediate behavioral responses related to innate defensive responses, where the parvofox Foxb1 of the LHA would be involved in promoting tachypnea and the PMd Foxb1group in mediating freezing and bradycardia. These findings are very interesting, and, at this point, they need to be tested in a scenario of real exposure to a natural predator.

Authors’ reply: We fully agree with the proposed experiments. Due to the previously mentioned retirement of Prof. Celio and the concomitant expiration of licenses for animal experimentation we are prevented from conducting these experiments on our own. We have integrated a statement in the discussion, regarding these potential future experiments.

Reviewer #2 (Public Review):

The authors aimed to examine the role of a group of neurons expressing Foxb1 in behaviors through projections to the dlPAG. Standard chemogenetic activation or inhibition and optogentic terminal activation or inhibition at local PAG were used and results suggested that, while activation led to reduced locomotion and breathing, inhibition led to a small degree of increased locomotion.

The observed effects on breathing are evident and dramatic. However, this study needs significant improvements in terms of data analysis and presentation and some of studies seem incomplete; and therefore the data may not yet support the conclusion.

1. Fig.1 has no experimental data and needs to be replaced with detailed pictures from the viral injected mice showing the projections diagrammed.

Authors’ reply: We believe that this graphic illustration is helpful to the reader to comprehend the spatial relationship between the parvafoxFoxb1 nucleus, the mammillary nuclei, and the PAG. In a previous study we have characterized the projections of the parvafoxFoxb1 nucleus in detail (using the same Foxb1-Cre mouse line as in the present study) and, in this regard, would like to refer Reviewer #2 to this publication (https://onlinelibrary.wiley.com/doi/10.1002/cne.24057).

1. Fig. 3 needs control pictures and statistical comparison with different conditions in c-Fos. Also expression in other nearby regions needs to be presented to demonstrate the specificity of the expression.

Authors’ reply: We have modified the original Fig. 3 with more pictures across all three conditions used in the chemogenetic experiments. Since the new figure now takes up a whole page, and because the data in this figure is for validation purpose of the DREADD experiments, we have decided to rather put it into the supplementary files. The figure is now labelled as “Supplementary File S1”. All figure and file numberings throughout the text have been adjusted accordingly.

1. Fig. 5, a great effort has been made to illustrate the point that CCK and Foxb1 are differentially expressed. Why not just perform a double in situ experiment to directly illustrate the point?

Authors’ reply: We have addressed this comment in the initial release of the eLife manuscript. In short, we agree that a double ISH experiment would have been an alternative approach, but would like to state that scRNAseq is a well established and valid method for this purpose.

1. Fig. 7 data on optogenetic stimulation on immobility and breathing, since not all mice showed the same phenotype, what is the criterion for allocating these mice to hit or no hit groups? Given the dramatically reduced breathing and locomotion, what is the temperature response? More data needs to be gathered to support that this is a defense behavior.

Authors’ reply: The criteria for allocation of animals to the experimental groups is described in section “Optogenetic modulation of Foxb1 terminals in the dlPAG induces immobility” and is based on the stereotaxic coordinates of the tips of the glass fiber implants. We did not perform any experiments, in which we recorded body temperatures or temperature preferences in optogenetic animals. Such experiments were outside the scope of the study. As mentioned in a previous comment above, we have added an additional paragraph to the discussion section regarding future investigations of these hypothalamic Foxb1 neurons during exposure to natural predators. Such experiments would certainly allow more insight into the defensive nature of the described phenotype.

1. The authors claim to target dlPAG. However, in the picture shown in Fig. 8C, almost all PAG contains ChR2 fibers and it is likely all the fibers will be activated by light. Thus, as presented, the data does not support the claim of the specificity on dlPAG. Also c-Fos data needs to be presented on the degree of activation of downstream PAG neurons after light exposure.

Authors’ reply: We attach the original image 8c, without arrows and indications, in which the localization of ChR2-positive fibers in the dlPAG is better visible. They are located exactly under the tip of the fiberoptic fiber. We do not know the functional characteristics of the post-synaptic PAG neurons and have not determined experimentally their downstream targets. Investigating the downstream target was outside the scope of the current publication.

Author response image 1.

1. Fig. 9 only showed one case. A statistical comparison needs to be presented.

Authors’ reply: Our cardiovascular experiments are of exploratory and descriptive nature (i.e. pilot experiments). It was a conscious decision to not perform hypothesis tests on these experiments. We did not have enough mice to perform statistical tests with sufficient statistical power. Providing results from hypothesis tests on these data would lead to statistically unjustified conclusions. To clarify this issue, we have added a paragraph to the relevant results section.

1. Optogentic terminal activation in the PAG will likely elicit back-propagation and subsequent activation of additional downstream brain sites of Foxb1 neurons. More experiments need to be done to assess this and as presented, the data does not support the role of PAG necessarily.

Authors’ reply: Please see our answer to Reviewer #1 regarding the same issue.

1. The authors claim negative data from PVH-Cre mice. More data need to be presented to make this case.

Authors’ reply: We would like to refer to our answer to point 6) that was raised by Reviewer #2

The conclusion, even as presented, adds to the known evidence of the PAG in the defense behavior.

Reviewer #1 (Recommendations For The Authors):

In the pharmacogenetic experiments, the authors need to clarify which Foxb1hypothalamic presented the activation of hM3Dq. It is important to know whether this activation-producing tachypnea was restricted to the parvofoxFoxb1 or also included the PMd Foxb1 group. It would be important to isolate the effect of the pharmacogenetic activation of each one of these Foxb1 hypothalamic cell groups.

After determining which cell group would be involved in mediating this respiratory effect, it would be nice to discuss the possible pathways involved in this effect.<br /> In the optogenetic experiments, the authors should differentiate between the effects of the PAG projecting fibers from the PMd and those from the parvofox groups. As it stands, it seems that the freezing and bradycardia depend on projection from the PMd Foxb1 group to the rostral PAGdl. However, considering the large volume (200 nl) of the viral construct injected, both groups were likely to express channelrhodopsin, and it would be important if the authors could restrict the viral injections to each one of the Foxb1 hypothalamic cell groups.

Authors’ reply: We fully agree with the suggestion, but due to the recent retirement of Prof. Celio we unfortunately not allowed to conduct any further animal experiments.

The authors also reported that photoactivation ventral to the PAGdl, possibly in the PAGl did not yield any clear behavioral response. However, as pointed out in the discussion, a recent publication found that the parvofox Foxb1 projection to the lateral PAG drives social avoidance, and we were wondering whether there was any avoidance behavior during the photoactivation of the PAGl fibers.

Authors’ reply: We did not conduct any social avoidance experiments ourselves. However, we did perform ultrasonic vocalization experiments (unpublished data) in which we optogenetically stimulated Foxb1+ terminals in the PAG. Due to experimental issues related to the age of the tested mice, we did not obtain conclusive results regarding the ultrasonic vocalizations. By a purely observational account, we did not observe any active avoidance during optogenetic stimulation, but rather a cessation of interaction. We are unable to judge whether this was more pronounced in the PAGl targeted mice or not.

Another important point is that optogenetic ChR2-mediated excitation of the axonal endings is likely to activate the cell bodies originating these fibers, and one cannot ascertain whether the behavioral effects depend on the activation of the terminals in the PAGdl or the activation of the cell bodies originating these terminals. Note, in the present case, PMd cell bodies may also project elsewhere, such as the cuneiform nucleus, known to mediate freezing responses. To circumvent this problem, during photoactivation of the PAGdl terminals, the authors should inhibit the cell bodies originating these terminals.

Authors’ reply: We would like to refer to the answer we provided above regarding the issue of backpropagation or ChR2-mediated phenotypes and projection-specificity.

Another important issue is related to the fact that around 90% of the PMd express CCK (Wang et al., 2021), and previous work showed that activation of these cells yielded escape and not freezing (Wang et al., 2021). Although the authors claim that the single-cell RNA sequencing dataset reveals distinct Foxb1 expression in the PMd, these results derive from tissues collected in the posterior hypothalamus, not exactly restricted to the PMd. Therefore, it would be desirable if the authors could show CCK and Foxb1doulbe labeled PMd sections to evaluate the exact percentage of cells expressing either one of these peptides.

Authors’ reply: The tissues for the scRNAseq data were obtained from hypothalamic tissues between stereotaxic coordinates of AP-2.54 to AP-3.16 (please see Fig. 1b in Mickelsen et al. 2020) and not purely from the posterior hypothalamic nucleus. These tissues hence include a large proportion of the PMd neurons. We would like to point out that the expression profile of the PMd cluster matches well with the ISH data from the Allen Brain Atlas that we have put together in "Supplementary File S6” (originally “Supplementary File S5”)

The authors should also explain why only a small number of animals that received PAGdl photoactivation presented bradycardia. Moreover, they should also discuss the possible pathways mediating this effect. Here, it is important to point out that the cuneiform nucleus, as suggested by the authors as one possible way to mediate this effect, promotes sympathetic vasomotor activity (Verbene, 1995).

We have added the sentence: “The projections of the cuneiform nucleus to the rostral ventrolateral medulla promote sympathetic vasomotor activity (Verberne 1995).” to the Discussion section.

Reviewer #2 (Recommendations For The Authors):

In this reviewer's view, this study needs substantial improvement:

1. The writing is very sloppy and difficult to follow. There is no clear logic flow in the main text and the figures need substantial realigning for panels, additions of labelling etc.

We have added the sentence.

1. Fig. 6 the hot plate data is out of place and should be placed in supplementary or removed completely.

Authors’ reply: We and others have previously shown that the parvalbumin+ population of the Parvafox nucleus is involved in nociceptive behavior. Hence, we believe it is of interest to show, that we do not see the same phenotype with the stimulation of the Foxb+ population of the parvafox nucleus. This data shows that the nociceptive component of the parvafox nucleus is confined to its parvalbumin+ population.

1. The authors discussed social behavior data in the Discussion, but no such data is presented, which is very confusing.

Authors’ reply: Indeed we did not perform any experiments to investigate social behavior. However, we address that the observed locomotive phenotype of optogenetic Foxb1+-terminals could have lead to a bias in the interpretation of the social behavior experiments published elsewhere by others.

1. The authors discussed a great deal on potential differences between parvafox and PMd Foxb1 neurons, however, no clear data was presented to show a functional difference between them, which is also confusing.

Authors’ reply: Even though investigations on the functional differences of parvafox and PMd Foxb1 neurons would be highly interesting, it was outside the scope of the current study. Due to the recent retirement of Prof. Celio, we are not allowed to perform any additional animal experiments.

Author Response

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

eLife assessment

This is an important study that leverages a human-chimpanzee tetraploid iPSC model to test whether cis-regulatory divergence between species tends to be cell type-specific. The evidence supporting the study's primary conclusion--that species differences in gene regulation are enriched in cell type-specific genes and regulatory elements--is compelling, although attention to biases introduced by sequence conservation is merited, and the case that is made for cell type-specific changes reflecting adaptive evolution is incomplete. This work will be of broad interest in evolutionary and functional genomics.

Public Reviews:

Reviewer #1 (Public Review):

This study aims to identify gene expression differences exclusively caused by cis-regulatory genetic changes by utilizing hybrid cell lines derived from human and chimpanzee. While previous attempts have focused on specific tissues, this study expands the comparison to six different tissues to investigate tissue specificity and derive insights into the evolution of gene expression.

One notable strength of this work lies in the use of composite cell lines, enabling a comparison of gene expression between human and chimpanzee within the same nucleus and shared trans factors environment. However, a potential weakness of the methodology is the use of bulk RNA-seq in diverse tissues, which limits the ability to determine cell-type-specific gene expression and chromatin accessibility regions.

We agree that profiling single cells could lead to additional exciting discoveries. Although heterogeneity in cell types within samples will indeed reduce our power to detect cell-type-specific divergence, thankfully any heterogeneity will not introduce false positives, since our use of interspecies hybrids controls for differences in cell-type abundance. As a result, we think that the molecular differences we identified in this study represent a subset of the true cell-type specific cis-regulatory differences that would be identified with deep single-cell profiling. We have included a new paragraph in the discussion on future directions, highlighting the utility of single-cell profiling as an exciting future direction (lines 482-490): “In addition to following up on our findings on GAD1 and FABP7, there are other exciting future directions for this work. First, additional bulk assays such as those that measure methylation, chromatin conformation, and translation rate could lead to a better understanding of what molecular features ultimately lead to cell type-specific changes in gene expression. Furthermore, the use of deep single cell profiling of hybrid lines derived from iPSCs from multiple individuals of each species during differentiation could enable the identification of many more highly context-specific changes in gene expression and chromatin accessibility such as the differences in GAD1 we highlighted here. Finally, integration with data from massively parallel reporter assays and deep learning models will help us link specific variants to the molecular differences we identified in this study.”

Another concern is the use of two replicates derived from the same pair of individuals. While the authors produced cell lines from two pairs of individuals in a previous study (Agloglia et al., 2021), I wonder why only one pair was used in this study. Incorporating interindividual variation would enhance the robustness of the species differences identified here.

We agree that additional replicates, especially from lines from other individuals, would have improved the robustness of the species differences we identified. In our experience with these hybrid cells (as well as related work from many other labs), inter-species differences typically have much larger magnitudes than intra-species differences, so we expect that the vast majority of differences we identified would be validated with data from additional individuals. Unfortunately, differentiating additional cells and generating these data for this study would be cost-prohibitive. We now mention the use of additional replicates in lines 485-488 of the discussion: “Furthermore, the use of deep single cell profiling of hybrid lines derived from iPSCs from multiple individuals of each species during differentiation could enable the identification of many more highly context-specific changes in gene expression and chromatin accessibility such as the differences in GAD1 we highlighted here.”

Furthermore, the study offers the opportunity to relate inter-species differences to trends in molecular evolution. The authors discovered that expression variance and haploinsufficiency score do not fully account for the enrichment of divergence in cell-type-specific genes. The reviewer suggests exploring this further by incorporating external datasets that bin genes based on interindividual transcriptomics variation as a measure of extant transcriptomics constraint (e.g., GTEx reanalysis by Garcia-Perez et al., 2023 - PMID: 36777183). Additionally, stratifying sequence conservation on ASCA regions, which exhibit similar enrichment of cell-type-specific features, using the Zoonomia data mentioned also in the text (Andrews et al., 2023 -- PMID: 37104580) could provide valuable insights.

To address this, we used PhastCons scores computed from a 470-way alignment of mammals as we could not find publicly available PhastCons data from Zoonomia. When stratifying by the median PhastCons score of all sites in a peak, we observe very similar results to those obtained when stratifying by the constraint metric from the gnomAD consortium (see below). The one potential difference is that peaks in the top two bins have slightly weaker enrichment relative to the other bins when using PhastCons, but this is not the case when using gnomAD’s metric. We have elected to include this in the public review but not the manuscript as we are reluctant to add to the complexity of what is already complex analysis.

Author response image 1.

Finally, we think that comparisons of the properties of gene expression variance computed from ASE (as done by Starr et al.) and total expression (as done by Garcia-Perez et al.) is a very interesting, potentially complex question that is beyond the scope of this paper but an exciting direction for future work.

Another potential strength of this study is the identification of specific cases of paired allele-specific expression (ASE) and allele-specific chromatin accessibility (ASCA) with biological significance.

Prioritizing specific variants remains a challenge, and the authors apply a machine-learning approach to identify potential causative variants that disrupt binding sites in two examples (FABP7 and GAD1 in motor neurons). However, additional work is needed to convincingly demonstrate the functionality of these selected variants. Strengthening this section with additional validation of ASE, ASCA, and the specific putative causal variants identified would enhance the overall robustness of the paper.

We strongly agree with the reviewer that additional work validating our results would be of considerable interest. We hope to perform follow-up experiments in the future. For now, we have been careful to present these variants only as candidate causal variants.

Additionally, the authors support the selected ASE-ASCA pairs by examining external datasets of adult brain comparative genomics (Ma et al., 2022) and organoids (Kanton et al., 2019). While these resources are valuable for comparing observed species biases, the analysis is not systematic, even for the two selected genes. For example, it would be beneficial to investigate if FABP7 exhibits species bias in any cell type in Kanton et al.'s organoids or if GAD1 is species-biased in adult primate brains from Ma et al. Comparing these datasets with the present study, along with the Agoglia et al. reference, would provide a more comprehensive perspective.

We agree with the reviewer’s suggestion that investigating GAD1 and FABP7 expression in other datasets is worthwhile. Unfortunately, the difference in human vs. chimpanzee organoid maturation rates and effects of culture conditions in Kanton et al. makes it unsuitable for plotting the expression of FABP7 as its expression is highly dependent on neuronal maturation. We therefore plotted bulk RNAseq data from multiple cortical regions from Sousa et al. 2017 (see below). This corroborates our claim that FABP7 has human-biased expression in adult humans compared to chimpanzees and rhesus macaques. We also investigated expression of GAD1 in the Ma et al. data as the reviewer suggested.

Author response image 2.

While there are differences in GAD1 expression between adult humans and chimpanzees, they are unlikely to be linked to the HAR we highlight as it is likely a transiently active cis-regulatory element (see below). In addition, some cell types seem to have chimpanzee-derived changes in GAD1 expression (e.g. SST positive neurons) whereas others seem to have human-derived changes in GAD1 expression (e.g. LAMP5 positive neurons).

Author response image 3.

While these are potentially interesting observations, we think that their inclusion in the manuscript might distract from our emphasis on the cell type-specific and developmental stage-specific of the changes in FABP7 and GAD1 expression we observe so we have not included them in the manuscript.

The use of the term "human-derived" in ASE and ASCA should be avoided since there is no outgroup in the analysis to provide a reference for the observed changes.

We agree with the reviewer that the term human-derived should be used with care and have changed the phrasing of line 230 to “human-chimpanzee differences in expression”. With regard to FABP7 we think that our analysis of the Ma et al. data—which includes data from rhesus macaques as an outgroup—justifies our use of “human-derived” in lines 360 and 457. As chimpanzee and macaque expression of FABP7 are similar but human expression is quite different, the most parsimonious explanation for our observations is that FABP7 upregulation occurred in the human lineage.

Finally, throughout the paper, the authors refer to "hybrid cell lines." It has been suggested to use the term "composite cell lines" instead to address potential societal concerns associated with the term "hybrid," which some may associate with reproductive relationships (Pavlovic et al., 2022 -- PMID: 35082442). It would be interesting to know the authors' perspective on these concerns and recommendations presented in Pavlovic et al., given their position as pioneers in this field.

We appreciate this question. Whether to refer to our fused cells as “hybrids” or not was indeed a question we considered at great length, starting from the very beginning of this project in 2015. From consultations with multiple bioethicists-- both formal and informal-- we have long been aware of the possibility of misunderstanding based on the word “hybrid”. However, we felt this possibility was outweighed by the long and well-established history of other scientists referring to interspecies fused cells as hybrids. This convention-- which is based on hundreds of papers about heterokaryons, somatic cell hybrids, and radiation hybrids-- goes back over 50 years (e.g. Bolund et al, Exp Cell Res 1969). Soon after the establishment of this nomenclature, cell fusion became widespread and ever since then it has become commonplace to generate interspecies hybrid cells from animals, plants and fungi.

It is also important to note that in over two years since we published the first two papers on humanchimpanzee fused cells, we have been unable to find any misunderstanding of our use of the term “hybrid”. We have searched blogs, media articles, and social media, all with no evidence of misunderstanding. Therefore, in the current manuscript, rather than creating confusion by renaming a well-established approach, we have opted to clearly and prominently define hybrid cells: in the abstract of our paper we introduce the hybrid cells as “the product of fusing induced pluripotent stem (iPS) cells of each species in vitro.”

Reviewer #2 (Public Review):

In this paper, Wang and colleagues build on previous technical and analytical achievements in establishing tetraploid human-chimpanzee hybrid iPSCs to investigate the cell type-specificity of allelespecific expression and allele-specific chromatin accessibility across six differentiated cell types (here, "allele-specific" indicates species differences with a cis-regulatory basis). The combined body of work is remarkable in its creativity and ambition and has real potential for overcoming major challenges in understanding the evolutionary genetics of between-species differences. The present paper contributes to these efforts by showing how differentiated cells can be used to test a long-standing hypothesis in evolutionary genetics: that cis-regulatory changes may be particularly important in divergence because of their potential for modularity.

In my view, the paper succeeds in making this case: allele (species)-specific expression (ASE) and allelespecific chromatin accessibility (ASCA) are enriched in genes asymmetrically expressed in one cell type, and many cases of ASE/ASCA are cell type-specific. The authors do an excellent job showing that these results are robust across a set of possible analysis decisions. It is somewhat less clear whether these enrichments are primarily a product of relaxed constraint on cell type-specific genes or primarily result from positive selection in the human or chimp lineage. While the authors attempt to control for constraint using several variables (variance in ASE in humans and the sequence-based probability of haploinsufficiency score, pHI), these are imperfect proxies for constraint. For the pHI scores, enrichments for ASE also appear to be strongest in the least constrained genes. Overall, the relative role of relaxation of constraint versus positive selection is unresolved, although the manuscript's language leans in favor of an important role for selection.

We agree with the reviewer and apologize for the wording that indeed focused more on positive selection than relaxed constraint. We have added language clarifying that our stance is that our analyses suggest some role for positive selection, but that we do not claim that positive selection plays a larger role than reduced constraint (lines 432-437): “Overall, this suggests that broad changes in expression in cell type-specifically expressed genes may be an important substrate for evolution but it remains unclear whether positive selection or lower constraint plays a larger role in driving the faster evolution of more cell type-specifically expressed genes. Future work will be required to more precisely quantify the relative roles of positive selection and evolutionary constraint in driving changes in gene expression.”

The remainder of the manuscript draws on the cell type-specific ASE/ASCA data to nominate candidate genes and pathways that may have been important in differentiating humans and chimpanzees. Several approaches are used here, including comparing human-chimp ASE to the distribution of ASE observed in humans and investigating biases in the direction of ASE for genes in the same pathway. The authors also identify interesting candidate genes based on their role in development or their proximity to human accelerated regions (where many changes have arisen on the human lineage in otherwise deeply conserved sequence) and use a deep neural network to identify sequence changes that might be causally responsible for ASE/ASCA. These analyses have value and highlight potential strategies for using ASE/ASCA and hybrid cell line data as a hypothesis-generating tool. Of course, the functional follow-up that experimentally tested these hypotheses or linked sequence/expression changes in the candidate pathways to organismal phenotype would have strengthened the paper further- but this is a lot to ask in an already technically and analytically challenging piece of work.

We thank the reviewer for the kind words and strongly agree that follow-up experiments and orthogonal analyses will be key in validating our results and establishing links to human-specific phenotypes.

As a minor critique, the present paper is very closely integrated with other manuscripts that have used the hybrid human-chimp cell lines for biological insight or methods development. Although its contributions make it a strong stand-alone contribution, some aspects of the methods are not described in sufficient detail for readers to understand (even on a general conceptual level) without referencing that work, which may somewhat limit reader understanding.

We agree with the points the reviewer raises regarding the clarity of our methods. We have amended several sections to provide more conceptual information while pointing the reader to other publications for the technical details. For convenience, we include the text here as well as in the new draft.

Lines 207-214 now provide more intuition for the method used to detect lineage-specific selection: “Next, we sought to use our RNA-seq data to identify instances of lineage-specific selection. In the absence of positive selection, one would expect that an approximately equal number of genes in a pathway would have human-biased vs. chimpanzee-biased ASE. Significant deviation from this expectation (as determined by the binomial test) rejects the null hypothesis of neutral evolution, instead providing evidence of lineage-specific selection on this pathway. Using our previously published modification of this test that incorporates a tissue-specific measure of constraint on gene expression, we detected several signals of lineage-specific selection, some of which were cell type-specific (Starr et al., 2023, Additional file 2).” This is also reflected in the Methods in lines 729-731: “Positive selection on a gene set is only inferred if there is statistically significant human- or chimpanzee-biased ASE in that gene set (using an FDR-corrected p-value from the binomial test).”

Reviewer #3 (Public Review):

The authors utilize chimpanzee-human hybrid cell lines to assess cis-regulatory evolution. These hybrid cell lines offer a well-controlled environment, enabling clear differentiation between cis-regulatory effects and environmental or other trans effects.

In their research, Wang et al. expand the range of chimpanzee-human hybrid cell lines to encompass six new developmental cell types derived from all three germ layers. This expansion allows them to discern cell type-specific cis-regulatory changes between species from more pleiotropic ones. Although the study investigates only two iPSC clones, the RNA- and ATAC-seq data produced for this paper is a valuable resource.

The authors begin their analysis by examining the relationship between allele-specific expression (ASE) as a measure of species divergence and cell type specificity. They find that cell-type-specific genes exhibit more divergent expression. By integrating this data with measures of constraint within human populations, the authors conclude that the increased divergence of tissue-specific genes is, at least in part, attributable to positive selection. A similar pattern emerges when assessing allele-specific chromatin accessibility (ASCA) as a measure of divergence of cis-regulatory elements (CREs) in the same cell lines.

By correlating these two measures, the authors identify 95 CRE-gene pairs where tissue-specific ASE aligns with tissue-specific ASCA. Among these pairs, the authors select two genes of interest for further investigation. Notably, the authors employ an intriguing machine-learning approach in which they compare the inferred chromatin state of the human sequence with that of the chimpanzee sequence to pinpoint putatively causal variants.

Overall, this study delves into the examination of gene expression and chromatin accessibility within hybrid cell lines, showcasing how this data can be leveraged to identify potential causal sequence differences underlying between-species expression changes.

We appreciate this assessment.

I have three major concerns regarding this study:

1. The only evidence that the cells are indeed differentiated in the right direction is the expression of one prominent marker gene per cell type. Especially for the comparison of conservation between the differentiated cell types, it would be beneficial to describe the cell type diversity and the differentiation success in more detail.

We appreciate this assessment. We agree that evidence beyond a single marker gene is necessary to demonstrate that the differentiations were successful and that a discussion of the limitations of these differentiations in the manuscript is worthwhile. We included figures showing additional marker genes and a thorough discussion of the differentiations in the supplement. For convenience, we have copied the supplemental figure and text here:

“Before continuing with the analysis, we tested whether the differentiations were successful and contained primarily our target cell types. The very low expression of NANOG, a marker for pluripotency, across all differentiations indicates that the samples contain very few iPSCs (Agoglia et al., 2021). For cardiomyocytes (CM), NKX2-5, MYBPC3, and TNNT2 definitively distinguish CM from other heart cell types and their high expression indicates successful differentiations (Burridge et al., 2014). For motor neurons, the high expression of ELAVL2, a pan-neuronal marker, indicates a high abundance of neurons in the sample (Mickelsen et al., 2019). The expression of ISL1 and OLIG2 further demonstrates that these are motor neurons and not other types of neurons (Maury et al., 2015). For retinal pigment epithelium (RPE), the combined expression of MITF, PAX6, and TYRP1 provides strong evidence that the differentiations were successful in producing RPE cells (Sharma et al., 2019). For skeletal muscle, the very high expression of MYL1, MYLPF, and MYOG indicates that these samples contain a high proportion of skeletal muscle cells (Chal et al., 2016). In general, all these populations of cells contain some proportion of progenitors as there is detectable expression of MKI67 in all samples.

The low expression of ALB (a marker for mature hepatocytes) and the high expression of TTR and GPC3 (markers for hepatocyte progenitors) combined with the high expression of HNF1B indicate that the bulk of the cells in the HP samples are hepatocyte progenitors rather than mature hepatocytes or endoderm cells, although there are likely some endoderm cells and immature hepatocytes in the sample (Hay et al., 2008; Mallanna & Duncan, 2013). Similarly, the combined expression of PDX1 and NKX6-1 and the low expression of NEUROG3 (a marker of endocrine progenitors which differentiate from pancreatic progenitors) in the PP samples indicates that these primarily contain pancreatic progenitors but likely contain some endocrine progenitors and endoderm cells (Cogger et al., 2017; Korytnikov & Nostro, 2016).

Notably, HP and PP are closely related cell types that are derived from the same lineage. Indeed, heterogeneous multipotent progenitors can contribute to both the adult liver and adult pancreas in mice (Willnow et al., 2021). Progenitors that express PDX1 (often used as a marker for the pancreatic lineage) can differentiate into hepatocytes (Willnow et al., 2021). As a result, some overlap in the transcriptomic signature of both cell types is expected and we cannot rule out that the HP samples contain cells that could differentiate into pancreatic cells or that the PP samples contain cells that could differentiate into hepatocytes. However, the expression of NKX6-1 and GP2, markers for pancreatic progenitors, in the PP samples but not the HP samples indicates that these two populations of cells are distinct. Overall, the similarity of PP and HP likely explains the lower number of cell type-specific genes and genes showing cell type-specific ASE for these cell types. This similarity does not alter the conclusions presented in the main text.”

Author response image 4.

Author response image 5.

Marker gene expression in different cell types. In order, the panels show: a marker for pluripotency, a marker gene for dividing cells, marker genes for cardiomyocytes, marker genes for hepatocytes and hepatocyte progenitors, marker genes for motor neurons, marker genes for pancreatic progenitors and more mature pancreatic cell types, marker genes for retinal pigment epithelial cells, and marker genes for skeletal myocytes. Hepatocyte progenitors and pancreatic progenitors generally show similar gene expression profiles. TPM: transcript per million.

1. Check for a potential confounding effect of sequence similarity on the power to detect ASE or ASCA.

We agree that checking for confounding by power to detect ASE or ASCA would increase confidence in our results. We have added supplementary figures 29-33 to show the results as well as a discussion of these figures in the text (lines 318-326):

“Finally, it is possible that CREs and genes that are less conserved will have more SNPs, and therefore more power to call ASCA and ASE, leading to systematically biased estimates. There is a weak positive correlation between the number of SNPs and the -log10(FDR) for ASE and a weak negative or no correlation for ASCA (Supp Fig. 29). Similarly, we observe a weak relationship between the number of SNPs in CREs or genes and absolute log fold-change estimates (Supp Fig. 30). Although the relationship between the number of SNPs and ASE/ASCA is weak, we confirmed that cell type-specific genes and peaks are still strongly enriched for ASE and ASCA when stratifying by number of SNPs (Supp Fig. 31-32). Overall, our analysis suggests that the result that more cell type-specific genes and CREs are more evolutionarily diverged is robust to a variety of possible confounders.”

Author response image 6.

Relationship between number of SNPs and -log10(FDR) in a) ASE and -log10(pvalue) b) ASCA. These scatter plots show the relationship between the number of SNPs in a gene or peak and the -log10(FDR) for ASE or ASCA. Genes with significant ASE (FDR < 0.05) and peaks with significant ASCA (binomial p-value < 0.05) were annotated as blue dots, and all other genes and peaks were annotated as gray dots. All genes in each cell type in RNA-seq are shown. For clarity, the few outlier peaks with more than 200 SNPs are excluded from these plots.

Author response image 7.

Relationship between number of SNPs and absolute log2 fold-change in a) ASE and b) ASCA. These scatter plots show the relationship between the number of SNPs in a gene or peak and the estimated absolute log2 fold-change for ASE or ASCA. Genes with significant ASE (FDR < 0.05) and peaks with significant ASCA (binomial p-value < 0.05) were annotated as blue dots, and all other genes and peaks were annotated as gray dots. All genes in each cell type in RNA-seq are shown. For clarity, the few outlier peaks with more than 200 SNPs are excluded from these plots.

Author response image 8.

Cell type-specifically expressed genes are enriched for genes with ASE when stratifying by the number of SNPs per gene. a) Results when SKM is included. Genes were put into five bins with an equal number of genes in each bin. Genes with the fewest SNPs are in the 0-20% bin and genes with the most SNPs are in the 80-100% bin. Significance (using the Wald test) is indicated by asterisks where *** indicates p < 0.005, ** indicates p < 0.01, and * indicates p < 0.05. b) The same as in (a) but excluding SKM.

Author response image 9.

Cell type-specific peaks are enriched for ASCA when stratifying by the number of SNPs per peak. a) Peaks with an absolute log2 fold-change greater than or equal to 0.5 were called as having ASCA. Peaks were put into five bins with an equal number of peaks in each bin. Peaks with the fewest SNPs are in the 0-20% bin and genes with the most SNPs are in the 80-100% bin. Significance (using the Wald test) is indicated by asterisks where *** indicates p < 0.005, ** indicates p < 0.01, and * indicates p < 0.05. b) The same as in (a) but peaks with a binomial p-value less than or equal to 0.05 were called as having ASCA.

1. In the last part the authors showcase 2 examples for which the log2 fold changes in chromatin state scores as inferred by the machine learning model Sei are used. This is an interesting and creative approach, however, more sanity checks on this application are necessary.

We agree with the reviewer about the importance of sanity checks and apologize for omitting these from the manuscript. Below we highlight several such checks from previous publications:

In the original Sei paper (Chen et al. 2022), the authors included several tests of their model’s ability to predict the effects on individual genetic variants. Using eQTL data from GTEx, they found that variants predicted to increase enhancer activity were more likely to be up-regulating eQTLs, and those predicted to increase polycomb repression had the expected repressive effect. These relationships became stronger when restricting the analysis only to fine-mapped eQTLs with >95% posterior probabilities of causality. Chen et al. also found that previously known disease-causing noncoding variants from the Human Gene Mutation Database were far more likely to reduce predicted enhancer/promoter activity than matched variants not linked to any disease.

In addition, we note that a similar approach to ours was recently used to analyze all HARs and included considerable efforts to validate the utility of the Sei predictions in identifying causal variants (Whalen et al. 2023 in Neuron). For example, Whalen et al. found that the Sei output correlated with the effects of genetic variants on expression in a massively parallel reporter assay. They also found that the effect sizes predicted by Sei were much higher for variants in HARs than polymorphic variants in the human population, which is consistent with the idea that variants in HARs lie in highly conserved bases that are more likely to disrupt cis-regulatory elements. Finally, Whalen et al. found that effects on chromatin state predicted by Sei were generally highly correlated across tissues, supporting our approach that leverages all Sei outputs regardless of which cell type or tissue they correspond to. Overall, we think that Sei is a potentially powerful way to prioritize causal variants and that improved machine learning models trained on more extensive and context-specific data will be even more powerful.

Author Response

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

Reviewer #1 (Public Review):

The study isolated extracellular vesicles (EV) from healthy controls (HCs) and Parkinson patients (PwP), using plasma from the venous blood of non-fasting people. Such EVs were characterized and validated by the presence of markers, their size, and their morphology. The main aim of the manuscript is to correlate the presence of synaptic proteins, namely SNAP-25, GAP-43, and SYNAPTOTAGMIN-1, normalized with HSP70, with the clinical progression of PwP. Changes in synaptic proteins have been documented in the CSF of Alzheimer's and Parkinson's patients. The demographics of participants are adequately presented.

• One important limiting, as well as puzzling aspect, is the fact that authors did not find differences between groups at the beginning of the study nor after one year, after age and sex adjustment.

Response: Thanks for your comments. We acknowledge your observation that the absence of a discernible difference in plasma EV synaptic protein levels between the PD and control subjects constitutes a significant limitation of our study. This outcome could be attributed to the fact that the controls were recruited from the neurology outpatient clinic, representing a group that could be considered "sub-healthy." Moreover, these individuals are not exempt from aging-related neurodegenerative processes. Considering that our PD subjects are in the early stages of the disease (with a mean disease duration of less than 3 years) and that synaptic dysfunction is a broader indicator rather than specific to PD, these factors could collectively contribute to the lack of distinction between the PD and control groups.

However, our primary intention was also to explore the potential of plasma EV synaptic proteins as predictive markers for disease progression in PD. In this regard, we have identified their applicability within the current PD cohort. We are committed to conducting further follow-up with these study subjects over an extended duration to delve deeper into these findings.

We revised the following statement in the discussion part to address this issue as following “Additionally, synaptic dysfunction is a frequently observed phenomenon in several neurological diseases, and it is not exclusive to PD. Consequently, the HC group in our current study may have included individuals with coexisting neurological conditions, potentially explaining the lack of a significant difference between the PD group and the HCs. However, this approach also illuminates the significance of synaptic dysfunction in the advancement of PD. This insight can be invaluable for monitoring disease progression, particularly in the context of clinical trials focused on disease modification.”

• Tables in general are hard to follow. Specifically, Table 2 does not convey a clear message nor in the text of the Table itself, and the per 100% of change needs to be explained in the corresponding legend.

Response: Thanks for your comment. In Table 2, our aim was to demonstrate the association between the change of plasma EV synaptic proteins with the change of clinical severity, and presented as coefficient (p value). We apologize for any prior ambiguity in the main text's description of these results and have since made revisions to enhance clarity.

Regarding the "per 100% change," this is due to the quantification of plasma EV synaptic proteins being based on a semi-quantitative Western blot method. Each measurement was normalized by the average baseline plasma synaptic protein levels of healthy controls (HCs). The term "per 100% change" denotes the increase or decrease in plasma EV synaptic protein abundance relative to the average baseline levels observed in healthy controls. We apologize for any confusion caused and removed this term. In addition, we rephrased the statement to ensure better understanding and readability in the Table legend of revised manuscript as following “The association between the change of plasma EV synaptic proteins abundance (between baseline and follow-up) with the change of clinical severity in motor and cognitive domains (between baseline and follow-up) in people with Parkinson’s disease. A generalized linear model was employed and the data was presented as coefficient (p value).”

• It is only when PwP were classified as a first quartile that a significantly greater deterioration was found. However, in the case of tremor, the top 25% had values going from 0.46-0.47 to 0.32-0.35, whereas the lower three quarters went from 0.33-0.34 to 0.27-0.28 depending on the protein analyzed. This needs to be clarified in the text.

Response: Thanks for your comments. As per the unified Parkinson's disease rating score (UPDRS), a higher score indicates greater severity of symptoms. Regarding tremor, we observed a general trend of improvement in both groups. PwP with elevated baseline plasma EV proteins had a trendy of worse tremor score at baseline, and the improvement was significantly better than the rest of PwP. This improvement seems to contradict the progressive nature of PD, and one possible explanation could be the alleviation of symptoms due to medication usage. The assessment of motor symptoms took place within the hospital setting, where we refrained from requesting patients to withhold their anti-PD medications due to concerns about safety issues such as falls. Consequently, certain motor symptoms might have been effectively controlled by the anti-PD medication. Traditionally, symptoms like tremor and rigidity (as reflected by the akinetic rigidity score) respond well to medications, while postural instability and gait disturbance (PIGD) are less responsive. In our cohort, we noted an improvement in tremor scores and stability in akinetic rigidity (AR) scores. Conversely, PD patients with higher baseline plasma EV synaptic protein levels exhibited notable progression in PIGD scores. These findings have been documented in the results section and discussed comprehensively within the revised manuscript as following “On the other hand, the evaluation of motor symptoms occurred in a hospital setting where we did not ask patients to stop taking their anti- PD medications due to safety concerns like the risk of falls. As a result, specific motor symptoms, particularly tremor and AR, which are more sensitive to medication compared to PIGD, may have been effectively managed by the anti-PD medications. This could potentially explain the improvement in tremor observed between the baseline and one-year follow-up, especially among PwP with elevated baseline plasma EV synaptic proteins.”

• Table 3 is hard to read and some of the values seem repetitive, especially for tremor, AR, and PIGD. It looks as if Figure 2 represents the same information as Table 3.

Response: Thanks for your information. We have ensured the accuracy of the results presented in Table 2. While some of the entries may appear similar, they do indeed possess distinct differences.

To enhance readability, we streamlined the information in Table 3 by removing the p-values from the intra-group comparisons between baseline and the 1-year follow-up within each domain. We retained the original p-values for trend related to the inter-group comparisons for changes. Detailed information has been relocated to the supplementary section of the revised manuscript. In Figure 2, we illustrated the relationship between baseline plasma extracellular vesicle (EV) synaptic protein levels and the clinical assessment parameters during follow-up in patients with Parkinson's disease (PwP). This portrayal is distinct from the information depicted in Table 3.

If you had concerns about the resemblance between Table 3 and Figure 3, please note that the values in Table 3 represent raw scores, while the values in Figure 3, namely the estimated marginal means, are the "adjusted" scores for UPDRS-II and PIGD at baseline and follow-up. These adjustments encompass age, sex, and disease duration. We sincerely apologize for any lack of clarity in our previous description and have since revised it accordingly.

• The text and figure legends are not helpful in guiding the reader to understand the presented information.

Response: Thanks for your comments and we apologized for the unclear statement. We revised the figure legend and the main text for better understanding of the readers.

Reviewer #2 (Public Review):

Hong and collaborators investigated variations in the amount of synaptic proteins in plasma extracellular vesicles (EV) in Parkinson's Disease (PD) patients on one-year follow-up. Their findings suggest that plasma EV synaptic proteins may be used as clinical biomarkers of PD progression.

• It is a preliminary study using semi-quantitative analysis of synaptic proteins.

Response: Thanks for your comments. The present study represents the initial phase of our investigation into the role of plasma EV synaptic proteins within our PD cohort. Our findings have revealed the potential predictive significance of these synaptic proteins in relation to PD progression. We are committed to conducting further follow-up with these study subjects over an extended period.

Furthermore, it's important to acknowledge that the semi-quantitative approach employed to assess protein abundance was a limitation of this study. This limitation stems from the low concentration of plasma EV synaptic proteins, which restricts the feasibility of utilizing techniques such as ELISA or other quantitative methods for protein assessment. We have duly acknowledged this limitation within the scope of the present study as following “Semiquantitative assessment of plasma EV synaptic protein (SNAP-25, GAP-43, and synaptotagmin-1) levels was performed using western blot analysis. The lack of absolute values limits further clinical application.”

Moving forward, we intend to adopt alternative EV isolation methods that enable the extraction of a larger abundance of plasma EV proteins, facilitating more accurate quantitative assessments. In addition, a longer longitudinal follow-up is warranted to clearly assess the prognostic efficacy of plasma EV synaptic proteins in PwP, which we had mentioned in the manuscript.

• The authors have a cohort of PD patients with clinical examination and a know-how on EV purification. Regarding this latter part, they may improve their description of EV purification. EV may be broken into smaller size EV after freezing. Does it explain the relatively small size in their EV preparation? Do the authors refer to the MISEV guidelines for EV purity?

Response: Thanks for your comments. In the previous manuscript, we provided a relatively detailed account of the procedures related to EV isolation and validation (https://doi.org/10.1096/fj.202100787R). In the revised manuscript, we added some information about the principle of the EV isolation kit, and the validation antibody as following “Plasma EVs were isolated from 1 mL of plasma by exoEasy Maxi Kit (Qiagen, Valencia, CA, USA), a membrane-based affinity binding step to isolate exosomes and other EVs without relying on a particular epitope, in accordance with the manufacturer’s instructions and storaged in the −80。C freezer. The isolated plasma EVs were then eluted and stored. Usually, 400 μL of eluate is obtained per mL of plasma. The isolated plasma EVs were validated according to the International Society of Extracellular Vesicles guidelines, which include1.markers, including the presence of CD63 (ab59479, Abcam, Cambridge, UK), CD9(ab92726, Abcam, Cambridge, UK), tumor susceptibility gene 101 protein (GTX118736, GeneTex, CA, USA) and negative of cytochrome c (ab110325; Abcam, Cambridge, UK) 2. Physical characterization through the nanoparticle tracking analysis, which demonstrated the majority of the size of EV are mainly within 50-100nm 3. The morphology from the electron microscopy analysis. The validation had been described previously [29-31]. “

It's important to note that our primary focus was on exosomes, the smallest subtype of EVs. Through nanoparticle tracking analysis, we observed that the majority of isolated EVs fell within the diameter range of 50-150nm, exhibiting significant surface marker (i.e. CD63 and CD9) expression. Moreover, electron microscopy confirmed their vesicular morphology. These meticulously validated EVs were promptly analysed post-isolation.

However, we acknowledge that the plasma obtained from study participants might have undergone freezing prior to EV isolation. This freezing process has the potential to diminish the yield rate of EVs and result in some degree of fragmentation. We have duly included this issue as a limitation in our revised manuscript as following “The final technical issue in the present study was the relatively small size of the isolated EVs. Despite the primary focus on isolating exosomes, which are the smallest type of EVs, it's important to consider that the presence of small-sized EVs could potentially be attributed to EV fragmentation that occurs during the freezing and thawing processes.”

• Regarding synaptic protein quantification, the choice of western blotting may not be the best one. ELISA and other multiplex arrays are available. How the authors do justify their choice?

Response: Thanks for your comments. We appreciate your input regarding the semi-quantitative western blot analysis not being the most optimal approach. Owing to the limited quantity of isolated plasma EVs and the significant protein abundance of synaptic proteins within these EVs, we did explore the use of an ELISA assay. However, it's worth noting that for a specific subset of the samples, the readout obtained was lower than the lower limit of detection of the ELISA kit. In response, we have incorporated this point as limitation within the discussion section of the revised manuscript as following “Semiquantitative assessment of plasma EV synaptic protein (SNAP-25, GAP-43, and synaptotagmin-1) levels was performed using western blot analysis. The lack of absolute values, i.e. from the results of enzyme-linked immunosorbent assay, limits further clinical application.”

• Do the authors try to sort plasma EV by membrane-associated neuronal EV markers using either vesicle sorting or immunoprecipitation?

Response: Thanks for your comments. The current study did not specifically isolate neuron-derived extracellular vesicles (EVs), potentially introducing some bias to the results. However, it's important to note that synaptic proteins, such as SNAP-25, exhibit a high degree of neuron-specific expression, with a predominant presence in the brain (as indicated by https://www.proteinatlas.org/ENSG00000132639-SNAP25/tissue). Given this context, the limitation of not analyzing neuron-derived EVs could be mitigated to some extent. In response, we have incorporated this point as limitation within the discussion section of the revised manuscript as following “Furthermore, this study evaluated the overall plasma EVs rather than specifically focusing on neuron-derived exosomes, potentially introducing a bias towards somatic-origin EVs. Nonetheless, it is worth noting that synaptic proteins primarily originate from neurons. Even when considering neuron-derived exosomes, it's important to recognize that they are not exclusively derived from the brain, which can lead to contamination from the peripheral nervous system.”

• Many technical aspects may be improved. Such technical questions weakened the authors' conclusions.

Response: Thanks for your comments. We recognize that the aforementioned issues represent limitations of our current study. In response, we have incorporated these points as limitations, including the semi-quantitative assessments, the isolation of total but not neuron-derived exosomes in the plasma, and the short follow-up time within the discussion section of the revised manuscript.

• The discussion is pretty long to justify the data. It may be shortened by adding some information in the introduction.

Response: Thanks for your comments. We have repositioned a statement from the second paragraph of the discussion to the introduction. This adjustment serves to enrich the background understanding of the link between synaptic dysfunction and neurodegenerative diseases.

Author Response

Reviewer #1 (Public Review)

The manuscript by Singh et al proposes a new theoretical model for the phenomenon of planar cell polarity (PCP). The new model is simulating the emergence of the subcellular polarity of the Fat-Ds pathway, based on the interactions of the protocadherins Fat and Ds at the boundary between cells and in response to external gradients. Several mathematical models for PCP have been previously developed focusing on different aspects of PCP, including non-autonomy domineering (Amonlirdviman et al.), the effect of stochasticity on polarity (Burak et al.), gradient sensing (Mani et al), formation of molecular bridges (Fisher et al.) to name a few. The current modeling approach suggests a new model, based on a relatively simple set of equations for membrane Fat and Ds and their interactions, both in 1D (line of cells) and in 2D (hexagonal array). The equations are relatively simple on one hand, allowing performing tractable computational analysis as well as analytical approximations, while on the other hand allowing tracking membrane protein levels, which is what is measured experimentally. It has been previously shown that achieving polarity requires local feedback that amplify complexes in one orientation at the expense of complexes in the opposite orientation (e.g. Mani et al.). Interestingly, the current manuscript shows that a simple assumption, that Fat-DS complexes are stabilized when bound is sufficient to induce PCP when concentrations are high enough. The authors use the model to show how it captures several experimental observations, as well as to analyze the sensitivity to noise, the response to gradients, and the response to local perturbations (mutant clones). The manuscript is clear and the analysis is mostly coherent and sensible (although some parts need to be clarified, see below). The main issue I have with the manuscript is that it mostly describes how it captures different features that were mostly explained in previous models. I do think the authors should do more with their model to explain features that were not explained by other models, and/or generate non-trivial predictions that can be tested experimentally.

We thank the reviewer for the positive feedback and valuable comments We have comprehensively modified the manuscript by including new results and detailing the specific model prediction and their potential experimental tests to address the concerns.

Reviewer #2 (Public Review):

The setting of planar cell polarity in epithelial tissues involves a complex interplay of chemical interactions. While local interactions can spontaneously give rise to cell polarity, planar cell polarity also involves tissue scale gradients whose effects are not clear. To understand their role, the authors built a minimal mechanistic model in considering two atypical cadherins, Fat (Ft) and Dachsous (Ds) which can associate at cell-cell interfaces to form hetero-dimers in which monomers belong to adjacent cells. This association can be seen as a local interaction between cells and is also sensitive to overall concentration gradients. From their model which appears to capture diverse experimental observations, the authors conclude that tissue-scale gradients provide to planar cell polarity a directional cue and some robustness to cellular stochasticity. While this model comes after similar works reaching similar predictions, the quality of this model is in its simplicity, its convenience for experimental testing, and the diversity of experimental observations it recapitulates.

A strength of this work is to recapitulate many experimental observations made on planar cell polarity. It, for example, seems to capture the response of tissues to perturbations such as local downregulation of some important proteins, and the polarity patterns observed in the presence of noise in synthesis or cell-to-cell heterogeneity. It also gives a mechanistic description of planar cell polarity, making its experimental interpretation simple. Finally, the simplicity of the model facilitates its exploration and makes it easily testable because of the reduced amount of free model parameters.

A weakness of this work is that it comes after several models with similar hypotheses and similar predictions.

Another weakness is that some conclusions of this work rely on visual appreciation rather than quantification. This is particularly true for what concerns 2D patterns. An argument of the authors is for example that their model reproduces a variety of known spatial patterns, but the comparison with experiments is only visual and would be more convincing in being more quantitative.

We are grateful to the reviewer for a critical evaluation of the manuscript and for giving important suggestions. We have incorporated all the comments and revised the manuscript accordingly by including quantitative analysis of all the results presented.

Reviewer #3 (Public Review):

Using theory, the authors study mechanisms for establishing planar cell polarity (PCP) through local and global modules. These modules refer to the interaction between neighbouring cells and tissue-wide gradients, respectively. Whereas local interactions alone can lead to tissue-wide alignment PCP, a global gradient can set the direction of PCP and maintain the pattern in presence of noise. In contrast, the authors argue that a global gradient can only generate PCP to an extent that is proportional to the gradient magnitude.

The authors formulate a discrete model in one and two spatial dimensions that describe the assembly dynamics of PCP proteins on membranes. The number of proteins per cell remains constant. Additive noise is introduced to account for stochasticity in the attachment/detachment kinetics of proteins. Furthermore, ’quenched’ noise is introduced to account for variations of protein numbers between cells. The authors perform simulations of the stochastic discrete model in various situations. In addition, they derive a continuum description to perform some analytical computations.

The strength of this analysis relies clearly on showing that simple dynamics can lead to tissue-wide PCP even in absence of a gradient in protein expression. A number of phenomena observed in tissues are qualitatively reproduced. In two spatial dimensions, they find swirling patterns that resemble patterns found in tissues when a global gradient is absent. The model also captures qualitative effects due to the down-regulation of one of the PCP proteins in a certain region of the tissue.

The main weak point is that, from a physical point of view, the findings are not particularly surprising. Furthermore, some assumptions underlying the model, need some more justification. This holds notably for the question, of why additive noise is appropriate to account for the effect of stochasticity in the attachment-detachment dynamics of the proteins. Finally, the authors consider a situation that they consider to be one of the most interesting features of PCP, namely, the formation of PCP in the presence of a region with a down-regulated PCP protein and in presence of a gradient. Unfortunately, the effect is not very clear and the data provided remains limited.

We thank the reviewer for the valuable comments are critique of the work. We have considered all the concerns and revised the manuscript comprehensively. In particular, we have elaborated the sections on model assumptions and added new figures/figure-panels to quantitatively present the model predictions. We have also revised the details of the one-dimensional continuum theory for PCP which, we feel, presents a detailed quantitative picture of PCP and its dependence on model parameters.

Author Response

Reviewer #2 (Public Review):

In this study, Leiba et al. aim at establishing the developing zebrafish embryo as a suitable infection model to study Salmonella persistence in vivo. Under environmental stress (ex: macrophage phagosomes) a proportion of bacteria switch to a slow/arrested growth state conferring increased resistance to antibiotic treatments. Persisters are getting increasingly linked to infection relapses. Understanding how persistent infections emerge and bacteria survive in an organism for long time without replicating before switching back to a replicative state is essential. Zebrafish represents an alternative model to mice offering the possibility to image the whole organism and capture persistency with an amazing spatio-temporal resolution.

In this paper, the authors demonstrate that persistent infections of Salmonella can be reproduced in the developing zebrafish. The kinetics of infection have been well characterized and shows a very nice heterogeneity between animals demonstrating the complex host-pathogen interactions (Fig 1). From the perspective of persistence, the presence of Salmonella survivors to host clearing is reported until 14dpi demonstrating the possibility to induce persistent infection in this model. Through the manuscript, the authors have used a variety of state-of-the-art technics illustrating the flexibility of this model including microscopy and imaging of specific immune populations, various transgenic animals and selective depletion of macrophages or neutrophils to assess their relative contributions. Overall, the conclusions of the authors are well supported by the presented data. This said, the authors should strengthen the conclusions of the paper by providing a better characterization of the infection.

Major comments:

1) Figure 1: What is the general life-spam of the fish?

The general life-span of the zebrafish is approximately 3 years on average. Persistent infection is determined by the existence of a fraction of bacteria that endure over an extended period (after 96 hpi). Further, we observed Salmonella persistence for 14 days. In figure 1, we don’t think that the information of the general life-span of the zebrafish is critical.

2) Figure 2: It would be nice to clearly state what infection scenario we are looking at. Have the authors studied "high proliferation", "infected" or "cleared" zebrafish?

In Figure 2 we have studied the "infected" group. Both "high proliferation" and "cleared" larvae were excluded from the analysis. This is now clearly stated in the legend of Figure 2.

3) Figure 3 and 4: It would be very informative if the authors can tell us what proportion of Salmonella is associated with macrophages and neutrophils. From panel C and D (Figure 3) and Figure 4 C and D and Suppl Fig 1, it seems that a lot of bacteria are extracellular. Maybe an EM image of the tissue would help to understand if the bacteria is "all" intracellular or intracellular.

We apologize for any misunderstanding regarding the presence of intra- and extracellular bacteria depicted in Figure 3 C and D, Figure 4 C and D and Figure 3 -Suppl Fig 1. These figures illustrate infection experiments conducted in single-reporter larvae, limiting our analysis to bacteria associated with a single cell type. Figure 3G and Figure 4E-G, the panels depict infection experiments carried out in dual-reporter larvae, showing bacteria associated or not with macrophages and neutrophils. The present study aimed to establish the role of neutrophils and macrophages in the control of early and persistent Salmonella infection but further studies will focus on the exact localization of Salmonella during the course of the infection and, despite being a challenging technique for zebrafish, electron microscopy could be of great interest, allowing to visualize any type of cells (to determine if all bacteria are intracellular) at high resolution.

4) Figure 3 and 4: It would be very useful if the authors can tell us if the intracellular bacteria are mainly found individually (like in Figure 3C) or does host cells harbor many intracellular bacteria. Looking at figure 4G: it is not clear to me how many intracellular bacteria can be counted on this image.

This is an interesting suggestion. At present, an accurate quantification of the intracellular bacteria on microscopy 3D-datasets is challenging because bacteria aggregate inside the cells. At 4 hpi, single bacteria can occasionally be observed outside leukocytes, while most of infected macrophages harbored several intracellular bacteria (bacteria aggregates). To compare the levels of intracellular bacterial between acute and persistent stages, we measured the size of E2Crimson-positive (E2Crimson+) events. At 5 hpi, the median volume of E2Crimson+ events was lower than that at 4 dpi. The size distribution analysis of E2Crimson+ events indicated a higher representation of smaller volumes (0.5-1.5 m3 and 1.5-10 m3) at 5 hpi compared to 4 dpi, a stage during which very large E2Crimson+ events were observed (between 100-1000 m3, with some exceeding 1000 m3). This observation suggests an elevated presence of intracellular bacteria within the cells during persistent stages and that intracellular bacteria are predominantly observed as multiple rather than as solitary entities. This analysis has been incorporated in new Figure 5.

5) Figure 3 and 4: The authors should also perform an experiment with a Salmonella strain harboring a growth reporter to quantify the amount of replicating and non-replicating bacteria. This experiment is not absolutely necessary for the story, but if possible, it would provide a very nice add-up to the story and impact to the paper.

We welcome the reviewers’ suggestion, which we have indeed considered and planning to carry on in the future, along with experimented more oriented on the bacterial side.

6) Figure 6: The authors should provide in suppl. the flow cytometry scatter plots used to delineate the different subpopulations.

We agree with the reviewer that the flow cytometry scatter plots used to delineate the different subpopulations were missing and are now incorporated in new Fig 7 - figure supplement 2.

7) Figure 6: A specific characterization of macrophages harboring Salmonella persisters at 4dpi is missing. As shown by the authors in Figure 6, the tnfa- populations of macrophages at 4dpi are very similar for both infected and non-infected larvae. Persisters should indeed reside within tnfa- macrophages but they should also induce a specific signature through the actions of Salmonella effectors. Measuring this signature will allow a direct comparison with published data in mice and assess how accurately the zebrafish model recapitulates the manipulation of macrophages by Salmonella

We agree with the reviewer that a specific characterization of macrophages harboring persistent Salmonella at 4 dpi is missing. However due to the technical limitation inherent to the model (limited recovery of infected cells following FACS sorting), we were not able to specifically sort infected macrophages at 4 dpi.

Author Response

Reviewer #1 (Public Review):

This paper combines a number of cutting-edge approaches to explore the role of a specific mouse retinal ganglion cell type in visual function. The approaches used include calcium imaging to measure responses of RGC populations to a collection of visual stimuli and CNNs to predict the stimuli that maximally activate a given ganglion cell type. The predictions about feature selectivity are tested and used to generate a hypothesized role in visual function for the RGC type identified as interesting. The paper is impressive; my comments are all related to how the work is presented.

We thank the reviewer for appreciating our study and for the interesting comments.

Is the MEI approach needed to identify these cells?

To briefly summarize the approach, the paper fits a CNN to the measured responses to a range of stimuli, extracts the stimulus (over time, space, and color) that is predicted to produce a maximal response for each RGC type, and then uses these MEIs to investigate coding. This reveals that G28 shows strong selectivity for its own MEI over those of other RGC types. The feature of the G28 responses that differentiate it appears to be its spatially-coextensive chromatic opponency. This distinguishing feature, however, should be relatively easy to discover using more standard approaches.

The concern here is that the paper could be read as indicating that standard approaches to characterizing feature selectivity do not work and that the MEI/CNN approach is superior. There may be reasons why the latter is true that I missed or were not spelled out clearly. I do think the MEI/CNN approach as used in the paper provides a very nice way to compare feature selectivity across RGC types - and that it seems very well suited in this context. But it is less clear that it is needed for the initial identification of the distinguished response features of the different RGC types. What would be helpful for me, and I suspect for many readers, is a more nuanced and detailed description of where the challenges arise in standard feature identification approaches and where the MEI/CNN approaches help overcome those challenges.

Thank you for the opportunity for clarification. In fact, the MEI (or an alternative nonlinear approach) is strictly necessary to discover this selectivity: as we show above (response #1 to editorial summary), the traditional linear filter approach does not reveal the color opponency. We realize that this fact was not made sufficiently clear in the initial submission. In the revised manuscript, we now include this analysis. Moreover, throughout the manuscript, we added explanations on the differences between MEIs and standard approaches and more intuitions about how to interpret MEIs. We also added a section to the discussion dedicated to explaining the advantages and limitations of the MEI approach.

Interpretation of MEI temporal structure

Some aspects of the extracted MEIs look quite close to those that would be expected from more standard measurements of spatial and temporal filtering. Others - most notably some of the temporal filters - do not. In many of the cells, the temporal filters oscillate much more than linear filters estimated from the same cells. In some instances, this temporal structure appears to vary considerably across cells of the same type (Fig. S2). These issues - both the unusual temporal properties of the MEIs and the heterogeneity across RGCs of the same type - need to be discussed in more detail. Related to this point, it would be nice to understand how much of the difference in responses to MEIs in Figure 4d is from differences in space, time, or chromatic properties. Can you mix and match MEI components to get an estimate of that? This is particularly relevant since G28 responds quite well to the G24 MEI.

One advantage of the MEI approach is that it allows to distinguish between transient and sustained cells in a way that is not possible with the linear filter approach: Because we seek to maximize activity over an extended period of time, transient cells need to be repetitively stimulated whereas sustained cells will also respond in the absence of multiple contrast changes. In the revised manuscript, we add a section explaining this, together with Figure 3-supplement 2, illustrating this point by showing that oscillations disappear when we optimize the MEI for a short time window. The benefit of a longer time window lies in the increased discriminability between transient and sustained cells, which is also shown in the new supplementary figure.

Regarding the heterogeneity of MEIs, this is most likely due to heterogeneity within the RGC group: “The mixed non-direction-selective groups G17 and G31 probably contain more than one type, as supported by multiple distinct morphologies and genetic identities (for example, G31,32, Extended Data Fig. 5) or response properties (for example, G17, see below)” (Baden et al. Nature 2016). We added a paragraph in the Results section.

Concerning the reviewer’s last point: We agree that it is important to know whether the defining feature - i.e., the selectivity for chromatic contrast - is robust against variations in other stimulus properties. New electrophysiological data included in the manuscript (Fig. 6e,f) offers some insights here. We probed G28/tSbC cells with full-field flashed stimuli that varied in chromatic contrast. Despite not matching the cell’s preferred spatial and temporal properties, this stimulus still recovered the cell’s preference for chromatic contrast. While we think it is an interesting direction to systematically quantify the relative importance of temporal, spatial and chromatic MEI properties for an RGC type’s responses, we think this is beyond the scope of this manuscript.

Explanation of RDM analysis

I really struggled with the analysis in Figure 5b-c. After reading the text several times, this is what I think is happening. Starting with a given RGC type (#20 in Figure 5b), you take the response of each cell in that group to the MEI of each RGC type, and plot those responses in a space where the axes correspond to responses of each RGC of this type. Then you measure euclidean distance between the responses to a pair of MEIs and collect those distances in the RDM matrix. Whether correct or not, this took some time to arrive at and meant filling in some missing pieces in the text. That section should be expanded considerably.

We appreciate the reviewer’s efforts to understand this analysis and confirm that they interpreted it correctly. However, we decided to remove the analysis. The point we were trying to make with this analysis is that the transformation implemented by G28/tSbC cells “warps” stimulus space and increases the discriminability of stimuli with similar characteristics like the cell’s MEI. We now make this point in a - we think - more accessible manner by the new analysis about the nonlinearity of G28/tSbC cell’s color opponency (see above).

Centering of MEIs

How important is the lack of precise centering of the MEIs when you present them? It would be helpful to have some idea about that - either from direct experiments or using a model.

In the electrophysiological experiments, the MEIs were centered precisely (now Fig. 5 in revised manuscript) and these experiments yielded almost identical results to the 2P imaging experiments, where the MEIs were presented on a grid to approach the optimal position for the recorded cells. Additionally, all model simulations work with perfectly centered MEIs. We hence conclude that our grid-approach at presenting stimuli provided sufficient precision in stimulus positioning.

We added this information to the revised manuscript.

Reviewer #2 (Public Review):

This paper uses two-photon imaging of mouse ganglion cells responding to chromatic natural scenes along with convolutional neural network (CNN) models fit to the responses of a large set of ganglion cells. The authors analyze CNN models to find the most effective input (MEI) for each ganglion cell as a novel approach to identifying ethological function. From these MEIs they identify chromatic opponent ganglion cells, and then further perform experiments with natural stimuli to interpret the ethological function of those cells. They conclude that a type of chromatic opponent ganglion cell is useful for the detection of the transition from the ground to the sky across the horizon. The experimental techniques, data, and fitting of CNN models are all high quality. However, there are conceptual difficulties with both the use of MEIs to draw conclusions about neural function and the ethological interpretations of experiments and data analyses, as well as a lack of comparison with standard approaches. These bear directly both on the primary conclusions of the paper and on the utility of the new approaches.

We thank the reviewer for the detailed comments.

1) Claim of feature detection.

The color opponent cells are cast as a "feature detector" and the term 'detector' is in the title. However insufficient evidence is given for this, and it seems likely a mischaracterization. An example of a ganglion cell that might qualify as a feature detector is the W3 ganglion cell (Zhang et al., 2012). These cells are mostly silent and only fire if there is differential motion on a mostly featureless background. Although this previous work does not conduct a ROC analysis, the combination of strong nonlinearity and strong selectivity are important here, giving good qualitative support for these cells as participating in the function of detecting differential motion against the sky. In the present case, the color opponent cells respond to many stimuli, not just transitions across the horizon. In addition, for the receiver operator characteristic (ROC) analysis as to whether these cells can discriminate transitions across the horizon, the area under the curve (AUC) is on average 0.68. Although there is not a particular AUC threshold for a detector or diagnostic test to have good discrimination, a value of 0.5 is chance, and values between 0.5 and 0.7 are considered poor discrimination, 'not much better than a coin toss' (Applied Logistic Regression, Hosmer et al., 2013, p. 177). The data in Fig. 6F is also more consistent with a general chromatic opponent cell that is not highly selective. These cells may contribute information to the problem of discriminating sky from ground, but also to many other ethologically relevant visual determinations. Characterizing them as feature detectors seems inappropriate and may distract from other functional roles, although they may participate in feature detection performed at a higher level in the brain.

The reviewer apparently uses a rather narrow definition of a feature detector. We, however, argue for a broader definition, which, in our view, better captures the selectivities described for RGCs in the literature. For example, while W3 cells have been quite extensively studied, one can probably agree on that so far only a fraction of the possible stimulus space has been explored. Therefore, it cannot be excluded that W3 cells respond also to other features than small dark moving dots, but we (like the reviewer) still refer to it as a feature detector. Or, for instance, direction-selective (DS) RGCs are commonly considered feature detectors (i.e., responsive to a specific motion direction), although they also respond to flashes and spike when null-direction motion is paused (Barlow & Levick J Physiol 1965).

The G28/tSbC cells’ selectivity for full-field changes in chromatic contrast enables them to encode ground-sky horizon transitions reliably across stimulus parameters (e.g., see new Fig. 7i panel). This cell type is thus well-suited to contribute to detecting context changes, as elicited by ground-sky transitions.

Therefore, we think that the G28/tSbC RGC can be considered a feature detector and as such, could be used at a higher level in the brain to quickly detect changes in visual context (see also Kerschensteiner Annu Rev Vis Sci 2022). Still, their signals may also be useful for other computations (e.g., defocus, as discussed in our manuscript).

Regarding the ROC analysis, we acknowledge that an average AUC of .68 may seem comparatively low; however, this is based on the temporally downsampled information (i.e., by way of Ca2+ imaging) gathered from the activity of a single cell. A downstream area would have access to the activity of a local population of cells. This AUC value should therefore be considered a lower bound on the discrimination performance of a downstream area. We now comment on this in the manuscript.

2) Appropriateness of MEI analysis for interpretations of the neural code.

There is a fundamental incompatibility between the need to characterize a system with a complex nonlinear CNN and then characterizing cells with a single MEI. MEIs represent the peak in a complex landscape of a nonlinear function, and that peak may or may not occur under natural conditions. For example, MEIs do not account for On-Off cells, On-Off direction selectivity, nonlinear subunits, object motion sensitivity, and many other nonlinear cell properties where multiple visual features are combined. MEIs may be a useful tool for clustering and distinguishing cells, but there is not a compelling reason to think that they are representative of cell function. This is an open question, and thus it should not be assumed as a foundation for the study. This paper potentially speaks to this issue, but there is more work to support the usefulness of the approach. Neural networks enable a large set of analyses to understand complex nonlinear effects in a neural code, and it is well understood that the single-feature approach is inadequate for a full understanding of sensory coding. A great concern is that the message that the MEI is the most important representative statistic directs the field away from the primary promise of the analysis of neural networks and takes us back to the days when only a single sensory feature is appreciated, now the MEI instead of the linear receptive field. It is appropriate to use MEI analyses to create hypotheses for further experimental testing, and the paper does this (and states as much) but it further takes the point of view that the MEI is generally informative as the single best summary of the neural code. The representation similarity analysis (Fig. 5) acts on the unfounded assumption that MEIs are generally representative and conveys this point of view, but it is not clear whether anything useful can be drawn from this analysis, and therefore this analysis does not support the conclusions about changes in the representational space. Overall this figure detracts from the paper and can safely be removed. In addition, in going from MEI analysis to testing ethological function, it should be made much more clear that MEIs may not generally be representative of the neural code, especially when nonlinearities are present that require the use of more complex models such as CNNs, and thus testing with other stimuli are required.

The reviewer correctly characterizes MEIs as representing the peak in a nonlinear loss landscape that, in this case, describes the neurons’ tuning. As such, the MEI approach is indeed capable of characterizing nonlinear neuronal feature selectivities that are captured by a nonlinear model, such as the CNN we used here. We therefore disagree with the suggestion that MEIs should not be used “when nonlinearities are present that require the use of more complex models such as CNNs”. It is unclear what other “analysis of neural networks” the reviewer refers to. One approach to analyze the predictive neural network are MEIs.

We also want to clarify that, while the reviewer is correct in stating that the MEI approach as used here only identifies a single peak, this does not mean that it cannot capture neuronal selectivities for a combination of features, as long as this combination of features can be described as a point in high-dimensional stimulus space. In fact, this is demonstrated in our manuscript for the case of G28/tSbC cell’s selectivity for large or full-field, sustained changes in chromatic contrast (a combination of spatial, temporal, and chromatic features). While approaches similar to the one used here generate several diverse exciting inputs (Ding et al. bioRxiv 2023) and could therefore also fully capture On-Off selectivities, we pointed out the limitation of MEIs when describing On-Off cells in the manuscript (both original and revised).

Regarding the reviewer’s concern that “[...] the message that the MEI is the most important representative statistic [...] takes us back to the days when only a single sensory feature is appreciated”. It was certainly not our intention to proclaim MEIs as the ultimate representation of a cell’s response features and we have clarified this in the revised manuscript. However, we also think that (i) in applying a nonlinear method to extract chromatic, temporal, and spatial response properties from natural movie responses, we go beyond many characterizations that use linear methods to extract spatial or temporal only, achromatic response properties from static, white-noise stimuli. This said, we agree that (ii) expanding around the peak is desirable, and we do that in an additional analysis (new Fig. 6); but that reducing complexity to a manageable degree (at least, at first) is useful and even necessary when discovering novel response properties.

Concerning the representational similarity analysis (RSA): the point we were trying to make with this analysis is that the transformation implemented by G28 “warps” stimulus space and increases the discriminability of stimuli with similar characteristics like the cell’s MEI. We now made this point in a more accessible fashion through the above-mentioned analysis, where we extended the estimate around the peak. We therefore agree to remove the RSA from the paper.

In the revised manuscript, we (a) discuss the advantages and limitations of the MEI approach in more detail (in Results and Discussion; see also our reply #1) and (b) replaced the RSA analysis.

3) Usefulness of MEI approach over alternatives. It is claimed that analyzing the MEI is a useful approach to discovering novel neural coding properties, but to show the usefulness of a new tool, it is important to compare results to the traditional technique. The more standard approach would be to analyze the linear receptive field, which would usually come from the STA of white noise measurement, but here this could come from the linear (or linear-nonlinear) model fit to the natural scene response, or by computing an average linear filter from the natural scene model. It is important to assess whether the same conclusion about color opponency can come from this standard approach using the linear feature (average effective input), and whether the MEIs are qualitatively different from the linear feature. The linear feature should thus be compared to MEIs for Fig. 3 and 4, and the linear feature should be compared with the effects of natural stimuli in terms of chromatic contrast (Fig. 6b). With respect to the representation analysis (Fig. 5), although I don't believe this is meaningful for MEIs, if this analysis remains it should also be compared to a representation analysis using the linear feature. In fact, a representation analysis would be more meaningful when performed using the average linear feature as it summarizes a wider range of stimuli, although the most meaningful analysis would be directly on a broader range of responses, which is what is usually done.

We agree that the comparison with a linear model is an important validation. Therefore, we performed an additional analysis (see also reply #1, as well as Fig. 6 and corresponding section in the manuscript) which demonstrates that an LN model does not recover the chromatic feature selectivity. This finding supports our claims about the usefulness of the MEI approach over linear approaches.

Regarding the comment on the representation analysis, as mentioned above, we consider it replaced by the analysis comparing results from an LN model and a nonlinear CNN.

4) Definition of ethological problem. The ethological problem posed here is the detection of the horizon. The stimuli used do not appear to relate to this problem as they do not include the horizon and only include transitions across the horizon. It is not clear whether these stimuli would ever occur with reasonable frequency, as they would only occur with large vertical saccades, which are less common in mice. More common would be smooth transitions across the horizon, or smaller movements with the horizon present in the image. In this case, cells which have a spatial chromatic opponency (which the authors claim are distinct from the ones studied here) would likely be more important for use in chromatic edge detection or discrimination. Therefore the ethological relevance of any of these analyses remains in question.

It is further not clear if detection is even the correct problem to consider. The horizon is always present, but the problem is to determine its location, a conclusion that will likely come from a population of cells. This is a distinct problem from detecting a small object, such as a small object against the background of the sky, which may be a more relevant problem to consider.

Thank you for giving us the opportunity to clear these things up. First, we would like to clarify that we propose that G28/tSbC cells contribute to detecting context changes, such as transitions across the horizon from ground to sky, not to detecting the horizon itself. We acknowledge that we were not clear enough about this in the manuscript and corrected this. To back-up our hypothesis that G28 RGCs contribute to detecting context changes, we performed an additional simulation analysis, which is described in our reply #3 (see above).

5) Difference in cell type from those previously described. It is claimed that the chromatic opponent cells are different from those previously described based on the MEI analysis, but we cannot conclude this because previous work did not perform an MEI analysis. An analysis should be used that is comparable to previous work, the linear spatiotemporal receptive field should be sufficient. However, there is a concern that because linear features can change with stimulus statistics (Hosoya et al., 2005), a linear feature fit to natural scenes may be different than those from previous studies even for the same cell type. The best approach would likely be presenting a white noise stimulus to the natural scenes model to compute a linear feature, which still carries the assumption that this linear feature from the model fit to a natural stimulus would be comparable to previous studies. If the previous cells have spatial chromatic opponency and the current cells only have chromatic opponency in the center, there should be both types of cells in the current data set. One technical aspect relating to this is that MEIs were space-time separable. Because the center and surround have a different time course, enforcing this separability may suppress sensitivity in the surround. Therefore, it would likely be better if this separability were not enforced in determining whether the current cells are different than previously described cells. As to whether these cells are actually different than those previously described, the authors should consider the following uncited work; (Ekesten Gouras, 2005), which identified chromatic opponent cells in mice in approximate numbers to those here (~ 2%). In addition, (Yin et al., 2009) in guinea pigs and (Michael, 1968) in ground squirrels found color-opponent ganglion cells without effects of a spatial surround as described in the current study.

First of all, we did not intend to claim to have discovered a completely new type of color-opponent tuning in general; what we were trying to say is that tSbC cells display spatially co-extensive color opponency, a feature selectivity previously not described in this mouse RGC type, and which may be used to signal context changes as elicited by ground-sky transitions.

Concerning the reviewer’s first argument about a lack of comparability of our results to results previously obtained with a different approach: We think that this is now addressed by the new analysis (new Fig. 6), where we show why linear methods are limited in their capability to recover the type of color opponency that we discovered with the MEI approach.

Regarding the argument about center-surround opponency, we agree that “if the previous cells have spatial chromatic opponency and the current cells only have chromatic opponency in the center, there should be both types of cells in the current data set”. We did not focus on analyzing center-surround opponency in the present study, but from the MEIs, it is visible that many cells have a stronger antagonistic surround in the green channel compared to the UV channel (see Fig. 4a, example RGCs of G21, G23, G24; Figure 3-supplement 1 example RGCs of G21, G23, G24, G31, G32). Importantly, the MEIs shown in Fig. 4a were also shown in the verification experiment, and had G28 RGCs preferred this kind of stimulus, they would have responded preferentially to these MEIs, which was not the case (Fig. 4f).

It should also be noted here that, while the model’s filters were space-time separable, we did not impose a restriction on the MEIs to be space-time separable during optimization. However, we analyzed only the rank 1 components of the MEIs (see Methods section Validating MEIs experimentally). since our analysis focused on aspects of retinal processing not contingent on spatiotemporal interactions in the stimulus.

In summary, we are convinced that our finding of center-opponency in G28 is not an artifact of the methodology.

We discuss this in the manuscript and add the references mentioned by the reviewer to the respective part of the Discussion.

Reviewer #3 (Public Review):

This study aims to discover ethologically relevant feature selectivity of mouse retinal ganglion cells. The authors took an innovative approach that uses large-scale calcium imaging data from retinal ganglion cells stimulated with both artificial and natural visual stimuli to train a convolutional neural network (CNN) model. The resulting CNN model is able to predict stimuli that maximally excite individual ganglion cell types.

The authors discovered that modeling suggests that the "transient suppressed-by-contrast" ganglion cells are selectively responsive to Green-Off, UV-On contrasts, a feature that signals the transition from the ground to the sky when the animal explores the visual environment. They tested this hypothesis by measuring the responses of these suppressed-by-contrast cells to natural movies, and showed that these cells are preferentially activated by frames containing ground-to-sky transitions and exhibit the highest selectivity of this feature among all ganglion cell types. They further verified this novel feature selectivity by single-cell patch clamp recording.

This work is of high impact because it establishes a new paradigm for studying feature selectivity in visual neurons. The data and analysis are of high quality and rigor, and the results are convincing. Overall, this is a timely study that leverages rapidly developing AI tools to tackle the complexity of both natural stimuli and neuronal responses and provides new insights into sensory processing.

We thank the reviewer for appreciating our study.

Author Response

Reviewer #3 (Public Review):

This manuscript uses ASO to inhibit the self-cleaving ribozyme within CPEB intron 3 and test its effect on CPEB3 expression and memory consolidation. The authors conclude that the intronic ribozyme negatively affects CPEB3 mRNA splicing and expression, and suggests its implications for experience-induced gene expression underlying learning and memory.

The strength of the manuscript is in its exploration of a potentially novel mechanism of regulating CPEB3 expression in learning and memory, a combination of both biochemical and behavioral approaches to gain a wide perspective of this regulatory mechanism, and the application of ASO in this context. The introduction is sufficiently detailed. Statistics are thorough and appropriate. If the results could be more robust, the mechanism would provide a novel target and venue to modify learning and memory paradigm.

The weakness of the manuscript is that the magnitude of the activity-dependent regulation of ribozyme, the effects of ASOs on CPEB3 expression (mRNA and protein) and downstream target gene expression, in vitro and in vivo, are generally weak, raising concerns about the robustness of the result. This may have caused some of the inconsistencies between the data presentation (see below). Also unclear is whether the ribozyme activity is physiologically regulated by experience without ASO interference.

While the statistics tests support corresponding figure panels and their conclusions. The manuscript can be significantly strengthened by additional evidence, clarification of some methodologies, and reconciling some inconsistent results.

The premise of a comparable timescale between transcription and ribozyme activity as the foundation of the whole thesis was based on in vitro measurement of self-scission half-life and a broadly generalized transcription rate (which actually varies significantly between genes). This premise is weak and needs direct experimental support.

The physiological relevance of the proposed mechanism has yet to be demonstrated without ASO interference.

Fig2b: how were total and uncleaved Ribozymes measured by qRT-PCR? Where are the primers' locations? If the two products were amplified using different primers, their subtraction to derive % cleavage would not be appropriate.

We thank the reviewer for the thoughtful review. We measured the levels of the total ribozyme by measuring a 220-bp amplicon that starts 18 nts downstream from the ribozyme cleavage site. The uncleaved ribozyme levels were measured using oligos that amplify a region of the intron that starts 45 nts upstream and ends 238 nts downstream of the ribozyme cleavage site. We added this information to the Table of primers in the manuscript. For all PCR oligos we established independent standard curves and calculated RNA levels independently of other amplicons, as noted in the Methods section and now specified in the Results section as well (Page 15). The measurements were thus appropriate for the calculation of the cleaved ribozyme fractions in the various experiments. The fraction ribozyme cleaved was calculated from the uncleaved fraction as the difference between uncleaved fraction and unity (1 – fraction uncleaved), now specified on page 16 of the manuscript. Fraction uncleaved was calculated as [uncleaved ribozyme]/[total ribozyme], as was done previously (see Salehi-Ashtiani et al. Science 313:1788-1792 or Webb et al. Science 326:953).

Line 400-403: shouldn't ribozyme-blocking ASO prevent ribozyme self-cleavage, and as a result should further increase ribozyme levels? This would contradict the result in fig3a.

We showed that the ribozyme is inhibited in vitro (Fig. 1F and 1G) and all our data are consistent with ASO inhibition of the ribozyme in cellulo and in vivo. However, we do not have direct evidence for this ribozyme inhibition in vivo, because such an experiment would require a single-molecule FRET-type sensitivity in cells and this assay has not been developed for ribozyme cleavage in cellulo or in vivo. We measured the ribozyme levels by RT-qPCR and observed lower ribozyme levels in presence of ASO in cultured neurons (Fig. 3A) as well as in vivo (Fig. 5B), which is nominally in contrast to the observations in vitro. However, in these situations we do not measure the co-transcriptional fate of the intron or the ribozyme; rather, we measure the levels of the intron after splicing (evidenced by the increased levels of spliced exons 2–3) when the intron is likely already being degraded. We also do not know what effect the ribozyme ASO has on the intron stability once splicing occurs. Understandably, this is a weakness of the study—and we are fully open about this result— however, given the abundance of evidence that the ribozyme ASO leads to increase of CPEB3 mRNA under all conditions tested, we feel that there is strong, if indirect, evidence that our model for the ribozyme function is correct. Future studies will examine this issue closer, but a definitive experimental investigation for the mechanism and timing of ribozyme inhibition and intron degradation is out of scope of this study.

Author Response

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

Reviewer 1 (Public Review):

Weakness: Although the cross-links stimulate ATP hydrolysis, further controls are needed to convince me that the TM1 conformations observed in the structures are physiologically relevant, since they have been trapped by "large" substrates covalently-tethered by crosslinks.

Our response: Reviewer 1 raised concerns about the relatively large size of our covalently attached AAC substrate that would potentially distort TM1 in Pgp. We would like to clarify that AAC has a molecular weight of 462 Da, which, in comparison to many known Pgp substrates ranging from 250 to over 1,000 Da, is not a large compound. For instance, the few other Pgp substrates mentioned in our manuscript all have a comparable or larger size: verapamil, 455 Da; doxorubicin, 544 Da; FK506, 804 Da; valinomycin, 1,111 Da; cyclosporin A, 1,203 Da.

Furthermore, AAC was strategically attached to a site distant from TM1 in the inwardfacing Pgp conformation. After it was exported to the outward-facing state, several TM helices accommodate the compound. The observation that only TM1 exhibited significant conformational changes suggests its potential role in the transport mechanism. This hypothesis is supported by our findings, where a conservative substitution (G72A) in TM1 resulted in a dramatic loss of transport function for various drug substrates and impaired verapamil-stimulated ATPase activity.

Reviewer 1 (Recommendations for the Authors):

I understand the need for an unconventional approach to understanding the translocation pathway. What would help to support this model is to cross-link a much smaller substrate, as the one used is quite large and could potentially distort TM1 in the outward-state when cross-linked.

Our response: We thank the reviewer for this recommendation, and we have outlined plans for future experiments involving other substrates, including smaller ones, to further investigate our proposed model. However, it is important to acknowledge that conducting these studies will require a significant amount of effort and resources, which we believe extend beyond the scope of our current manuscript.

In unbiased MD simulations starting from the IF state are there any simulations where the substrate follows the same path as proposed here?

Our response: All our MD simulations were performed in the outward-facing state to focus on potential substrate release pathways. Starting MD simulations from the inwardfacing state would introduce complexities in capturing the necessary domain motions and nucleotide binding and hydrolysis required for substrate translocations. Therefore, we opted not to perform MD studies starting from the inward-facing state.

Reviewer 2 (Public Review):

Weakness: There is much to like about the experimental work here but I am less sanguine on the interpretation. The main idea is to covalently link via disulfide bonds a model tripeptide substrate under different conditions that mimic transport and then image the resulting conformations. The choice of the Pgp cysteine mutants here is critical but also poses questions regarding the interpretation. What seems to be missing, or not reported, is a series of control experiments for further cysteine mutations.

Our response: Reviewer 2 raised concerns about the interpretation of our results and suggested the need for additional mutant designs to validate our proposed TM1 mechanism. Firstly, we believe that the observed TM1 conformational changes are valid in our cryoEM structures, despite the use of different conditions and several mutants to capture Pgp in the outward-facing state.

Regarding the G72A mutant, we consider it conclusive that this single point mutation in the TM1 has a profound effect. Importantly, the G72A mutant was readily expressed and purifiable as a stable protein. We were able to resolve a high-resolution structure of the G72A mutant (without the substrate), confirming that the protein is not generally destabilized but properly folded.

Above all, we appreciate the Reviewer’s suggestion to explore additional mutations and intend to do so in future studies.

Reviewer 2 (Recommendations for the Authors):

I am sold on the results regarding TM1 conformational changes as they are evident in the cryoEM structures. However, the set of states compared between mutants are not biochemically equivalent: for 335 and 978 they used an ATP-impaired Pgp whereas for 971 they used what appears to be WT, and the conformation was imaged presumably subsequent to ATP hydrolysis and Vanadate trapping. This is significant if the authors were unable to trap the OF in the impaired mutant background and should be highlighted. I have to believe that they tried that condition but I could be wrong.

Our response: We acknowledge the point made by the Reviewer about the biochemical equivalence of mutant states and the potential significance of using an ATP-impaired mutant for trapping the outward-facing conformation of 971. We have not yet attempted to use the ATPase-deficient 971C mutant for crosslinking and intend to address this question in future studies.

In our current approach, we used the ATPase-active 971C for two specific reasons:

1) Our biochemistry data, as shown in Fig 1C, indicates that 971C only crosslinks in the presence of ATP hydrolysis conditions. Vanadate trapping was employed to stabilize the outward-facing conformation.

2) Based on our experience, we have observed that the conformations of ATP-bound (mutant) and vanadate-trapped states of an ABC transporter are structurally equivalent at this resolution level of our study (see ref. 21: Hoffmann et al. NATURE 2019).

The authors propose a new model for substrate translocation. It is based on three mutants and a number of structures. If the authors were not challenging the current dogma I would not have written the next comment. Considering the impact of the findings, I would have designed a couple more cysteine mutants based on their model. For instance, this pathway has a number of stabilizing interactions, can't they make a mutant that preserves conformational switching but eliminates substrate translocation? I like the G97A mutant result but I am worried that the effect could just be a general destabilization or misfolding as part of the cryoEM particles seem to suggest. The authors advance one interpretation of the disorder observed in this mutant but it could easily be my interpretation.

Our response: We thank the reviewer for the suggestion to design additional mutants to further validate our proposed model for substrate translocation. We agree that this would be highly valuable, considering the potential impact of our findings. However, given the time-intensive nature of our approach, we believe that presenting these additional designs in a future study is a reasonable course of action.

Regarding the G72A mutation, we believe that our current data fully supports our model and the role of TM1 in regulating the Pgp activity. Importantly, we would like to emphasize that the G72A mutant was readily expressed and purifiable as a stable protein. Additionally, our cryoEM structural determination of the G72A mutant at high resolution confirmed that the protein is not generally destabilized but properly folded.

There are a couple of troubling methodological questions that I want the authors to address or clarify:

1. In the methods they report that the final sample for cryoEM was prepared on a SEC devoid of detergent. It is obvious that the sample was folded but I was wondering why the detergent was removed? Was that critical for observing these structures with multiple ligands? Did they observe any lipids in their cryoEM?

Our response: We avoid detergent in the buffer on final SEC purification. This step is to remove free detergent from the background which helps during cryoEM imaging. Of course, this cannot be done with every detergent but due to the very low CMC of LMNG it is possible. By now, we have verified this method for several other transporters with the same success. While this procedure helps us to obtain better images it is not necessary to obtain specific conformations or ligand bound states, nor does it affect these states or conformations.

In our cryoEM structures , we did observe multiple cholesterol hemisuccinate (CHS) molecules on the outer transmembrane surface of Pgp.

1. Can the authors comment on why labeling was carried out in the presence of ATP? Does it matter if the substrate was added prior to ATP and incubated for a few minutes?

Our response: For every dataset, we first added the substrate to be cross-linked and afterwards added the ATP. In the cases of 335C and 978C, labeling was successful before ATP was added, as evidenced by the inward-facing structures with cross-linked substrate. However, for 971C, cross-linking only occurred after the addition of ATP. We interpret this data to suggest that the 971 site is inaccessible to the substrate in the inward-facing state, and cross-linking can only occur after the transporter transitions to outward-facing state. This is in line with our inward-facing structure which does not show a cross-linked substrate, and our biochemical data shown in Fig 1C, where 971C only crosslinked in the presence of ATP.

1. I am not an expert on MD simulations and I understand that carrying out simulations at higher temperatures used to be a trick to accelerate the process. Is this still necessary? Why didn't the author use approaches such as WESTPA?

Our response: Most so-called enhanced sampling methods, including WESTPA, explicitly define a reaction coordinate for the process of interest, usually based on intuition or prior studies. If this coordinate is chosen poorly, enhanced sampling usually fails, either because the sampling becomes inefficient or because the sampling biases the transition pathway (or both). Lacking reliable intuition or prior knowledge on which motions would result in substrate release, we chose temperature to speed up the process. High temperature largely avoids the introduction of an any bias through the definition of a progress coordinate. By contrast, the weighted ensemble method underlying WESTPA is a great method to simulate unbiased dynamics of a process with a known progress coordinate, but unfortunately requires to choose a progress coordinate prior to the simulation and will then mostly sample the process along this progress coordinate, because this is the only direction in which sampling is improved. High temperature MD on the other hand accelerates all processes in the system under study. Indeed, we have now confirmed that the pathway found at high temperature is also feasible at near-ambient conditions.

In new simulations, we have now observed a similar release pathway at T=330 K. As the only difference, the substrate has not fully dissociated from the protein after 2.5 us, with weak interactions persisting at the top part of TM1 from the extracellular side. Importantly, this is a configuration observed also in higher temperature simulations but with much shorter lifetime.

In response, we now included these new findings and a new Extended Data Fig. 15 in the revised manuscript.

1. One way to show that the two substrates binding mode is biochemically relevant is to measure Vmax at different substrate concentrations. One would expect a cooperative transition if that interaction is mechanistically important.<br /> Our response: We have measured Vmax as a function of QZ-Ala concentration in a previous report (ref. 24), supporting positive cooperativity for binding to two sites.

Reviewer 3 (Public Review):

We thank Reviewer 3 for recommending the acceptance of our manuscript as is.

Reviewer 3 (Recommendations for the Authors):

Page 4, last line: Pgp302 should be Pgp1302. In addition, I can only encourage the authors to add an additional table to the manuscript. Here, the mutation, the obtained structure(s), IF or OF, the resolution, and the main message should be summarized.

Our response: Following the reviewer’s suggestion, we have added Extended Data Table 2 summarizing the Pgp mutants and respective structural data in the revised manuscript.<br /> We verified that Pgp302 is the correct term on Page 4, last line.

Pg. 5, section 'Covalent ligand design for Pgp labeling', it is mentioned that even in the presence of Mg2+ATP, Pgp302 could not react with AAC-DNPT. Maybe it would be worthwhile to add the data either in Supplementary Information or state 'data not shown'.

Our response: We stated ‘data not shown’ in the text.

Pg. 47, last line : A space is missing between M68, and M74.

Our response: Space was added.

Pg. 7, line 2: The authors mention that a single dataset of ATP-bound Pgp335 revealed three different OF conformations: ligand-free, single-ligand-bound, and double-ligandbound. However, the percentage fraction of each dataset sums up to be more than 100%. Would request the authors to recalculate the fraction size of each conformation.

Our response: We have corrected the error in our calculation, based on the particle distribution in our dataset (OF335-nolig: 1,437,110 particles, 40.4%; OF335-1lig: 1,184,253 particles, 33.3%; and OF335-2lig: 939,924 particles, 26.4%).

Pg 53, Figure legend of Extended Data Fig. 11: Please include the color coding for the helix TM1 and also the residues colored plum.

Our response: We added the color coding for TM1 and other residues in the figure legend.

Pg. 8, line 3: While referring to the structure of OF971-1lig, the authors nicely point towards the conserved residues M74 and F78 which coordinate the ligand. However, in Fig. 3b, residues M74 and F78 should also be indicated.

Our response: We updated Fig. 3b by adding arrows pointing towards the residues M74 and F78.

Pg. 54, Extended data Fig. 12: The authors should adopt a single writing style. In some places, Pgp is referred to as P-gp while in others as Pgp.

Our response: We updated the protein labels in Extended Data Fig. 12.

Pg. 54, Extended data Fig. 12: The authors should clearly mention which OF335 structure (1st panel) was used for visualizing the interactions.

Our response: To clarify, we added the following sentences in the figure legend: “Pgp335 OF in the top panel refers to OF335-1lig. In the bottom panel describing OF335-2lig, the left and right diagrams refer to the binding positions of non-covalent and covalent ligand, respectively”.

Pg. 18, section 'synthesis of dipeptide 8': In the text it is mentioned that for the synthesis of thiazole acid 6, compound 3 was dissolved in a mixture of THF/MeOH/H2O (3:1:1), while in the corresponding figure (Extended Data Fig. 1), the ratio is stated as 5:1:2.

Our response: 3:1:1 ratio is correct. We made the correction in Extended Data Fig. 1.

Pg. 19, section 'synthesis of linear tripeptide 10': Same as above for compounds 10 and 4, respectively.

Our response: We corrected the conditions in the Extended Data Fig. 1 accordingly.

Pg. 20, section 'Synthesis of cyclic peptide 11': There seems to be a discrepancy in the synthesis protocol between the text and the extended figure 1, especially regarding the use of THF/MeOH/H20, followed by NaOH and TFA or only NaOH and TFA.

Our response: we further clarified the conditions of using NaOH in THF/MeOH/H2O (3:1:1) and TFA in DCM in the text for synthesis and Extend Data Fig. 1.

Pg. 40, Extended Data Fig. 1: In the bottom last panel showing the synthesis of peptide 11, the authors have missed showing peptide 10 as the starting material for the reaction.

Our response: Label for the peptide 10 was added following the suggestion.

Pg. 26, third last line: 'o' is missing from the last word cry'o'

Our response: We corrected the typo.

Pg. 63 and 64, Extended Data Table 1: The Cryo-EM data collection, refinement, and validation statistics for OF971-1lig, IF971-1lig, OF978-1lig, and IF978-2lig are mentioned twice in the table.

Our response: This was now corrected in the revision.

Author Response

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

Recommendations for the authors:

Reviewer #1 (Recommendations for the Authors):

The authors have addressed my recommendations in the previous review round in a satisfactory way. I only have one additional comment to the authors:

In the manuscript abstract lines 31-32, the author state that: "Using NIH data for the period 2006-2022, we report that ~230 K99 awards were made every year, representing ~$25 million annually."-- The "~$25 million" is under-stating the actual funds spent because this sum is just money spent on the first year of some k99s while the NIH is paying years 2,3,4 etc for others for k99 awards (~90% conversion rate to R00) awarded in previous years for a given year. The NIH is actually spending ~$230-$250 million a year on the k99 award mechanism in a given year. so the authors need to amend the stated amount in the manuscript.

Thank you for pointing this out. The reviewer is correct, that we had incorrectly only calculated the investment $ in new K99 awards made. We have corrected this in the revised manuscript. We appreciate your careful reading of our manuscript and the edits made based on your comments have improved the final version.

Reviewer #2 (Recommendations for the Authors):

Thank you for taking the time to revise this important work. I learned a lot reading this paper a second time, and appreciate the improvements you have made.

My only major thought while re-reading this is that I wish you all had written two papers! I see two themes in this work: one looking at faculty hiring networks from the Wapman et al. dataset, and another at K99/R00 conversions by institution, gender, and researcher mobility and its impact on subsequent funding success. After reading, I felt like I had many follow-up questions about both analyses, but it would be impractical for me to suggest all these follow-up analyses without making your paper unreasonably long.

Thank you for these comments. We agree that there are 2 general themes in this paper. While we feel that significantly expanding on both themes will be important in future research. Our hope is that this work continues to inspire others to critically examine funding practices and inequity in the same way that the work of Wapman, Pickett, etc. inspired the present work.

For example, regarding the results that more R00 are activated at different institutions, and that moving institutions improves subsequent funding success, I wonder: Do proportionally more women or men move institutions? Do proportionally more K99 awardees at less-funded places move for their R00, or less? The Cox proportional hazard models illustrate the impact of various characteristics on subsequent funding success, but they do not illustrate disparate impacts of mobility on different groups (if I am understanding them correctly). (You sort of dive into these questions in the very interesting subsection, "K99/R00 awardee self-hires are more common at institutions with top NIH funding." I wanted to read more!)

Thank you for these kind comments. These are fantastic follow-up questions. We do not feel that we can adequately address them within the present manuscript without potentially splitting it into 2 separate manuscripts. However, we may examine these in future analyses. We are particularly interested in examining additional aspects such as how the K99 MOSAIC funding mechanism may differ from the traditional K99 mechanism. Since the K99 MOSAIC mechanism is newer, there may not be enough K99 MOSAIC awards made for a thorough exploration.

As another example, for your analysis on faculty hiring networks, the prevalence of self-hiring amongst institutions and regions was one finding. However, this finding seems somewhat at odds with the previous takeaway about how researcher mobility improves subsequent funding success. Are institutions doing themselves a disfavor by hiring their own, then? I suspect there is more to say here about this pattern... maybe there are important differences between PhD institution and postdoc institution and its impact on hiring/subsequent funding success? Or is this a story about upward mobility into the top 25 well-funded NIH institutions?

Again, these are very insightful comments and follow-up questions. We hope to address these in potential future manuscripts. We also hope that others may become interested in finding answers to these questions by exploring our dataset as well as other publicly available datasets such as the Wapman et al. dataset.

I can completely understand how combining the faculty hiring network analysis with the K99/R00 conversions would seem like a natural fit, but I personally feel - emphasis on this being a personal opinion - that there would have been benefits to giving more space to the details of both analyses separately. Perhaps this is a "hindsight is 20/20" issue. Or an issue with the current times in which ones' brain can only hold so many main takeaways from a single body of work. (For example, I struggled to summarize your paper in my public review because I find so many takeaways important.)

I suppose this is all to say that I find your work important enough to warrant additional follow-up work! :)

Thank you for these very kind remarks. This work evolved over 8-10 months as evidenced by the updates to the biorXiv preprint. With unlimited time and foresight, it would probably be best to have separated the 2 themes into separate manuscripts and expanded both. Given current constraints, we plan to make some changes/updates to the present manuscript and hopefully include more in-depth analyses on each theme in future works. Thank you again for the thoughtful reading and critique of both our original manuscript and the revised version.

Minor comments/questions:

"K99 to R00 conversions are increasing in time"

• Assuming I am interpreting the figures correctly, in my opinion, the most important takeaway is that the number of R00 awards have increased, but only for awardees moving to another institution. This key result, best illustrated by panels A and C of Figure 1, is buried in the long paragraph in this section. The organization of content in this section could be improved and more focused. Consider renaming this subsection to be more declarative: "K99 tR00 conversions have increased, but only for awardees moving to another institution."

This is a very concise interpretation of this data. We have edited the paragraph referenced by the reviewer, split it into 2 paragraphs, and changed the title to “K99 awardees increasingly move to other institutions for R00 awards from 2008 to 2022” and the final sentence to “Thus, the number of K99 to R00 conversions is consistent over time, but increasingly more R00 awardees have moved to other institutions since 2013”

• Similarly, I personally found the current title of the subsection, "K99 to R00 conversions are increasing with time" is mildly confusing. An R00 award indicates a successful conversion, so why not simply call this an R00 award instead of saying K99-to-R00 conversion? Also, when I look at Figure 1B and exclude the conversion rates for 2007 and 2008 (because this is a 3 year rolling average), I see that conversion rates (or R00 awards) have remained stagnant. This comment is very much in-the-weeds and is mainly to do with clarity of language.

Thank you for these comments. We had “K99 to R00 conversion” to highlight the unique nature of this award mechanism that a person can only receive an R00 if they previously had a K99 award. Nevertheless, we have edited the text to “R00 awards” and “R00 awardees” to simplify things. We also want to note that we did not compute a 3-year rolling average. The function we used was: (X/(Y -1))x100 where X is the number of R00 awards made in a year and Y is the number of K99 awards made in a year. We did note an error in our calculation in the previous version of the manuscript. Previously, we included all R00 awards and K99 awards for each year from the NIH Reporter dataset; however, this is a flawed methodology. NIH reporter includes only extramural K99 award data and extramural R00 awards, but intramural K99 awardees can receive extramural R00 awards and thus are only included in the R00 dataset. There were 141 R00 awardees in our dataset from NIH Reporter that did not have K99 data, so we assume these are intramural K99 awards since it is required to have a K99 to be eligible for the R00 award. Since we do not know the awarding year for intramural K99 awardees or have data on intramural K99 awardees that fail to activate the R00 award (or stay internal at NIH), we have excluded these 141 R00 awardees. In the previous version, this mis-calculation exaggerated rolling conversion rate (we had correctly calculated the 78% total conversion rate). We re-analyzed our rolling conversion rate and found the average is 81.8% (excluding the first 2 years of the K99 program and the last 2 years).

This is a long explanation, but essentially, we overestimated the number of R00 awards which inadvertently increased the rolling conversion rate. We have corrected this and simplified the first 2 paragraphs of the Results section.

• I was also mildly confused looking at Figure 1c. The caption says that the percentages represent the K99 awardees that stayed at the same institution for the R00 activation, but the percentages are next to the solid circles which the legend labels as "different institution." Perhaps another or different way to show this is a stacked bar chart, where one bar represents the percentage of R00 awards activated at the same institution and another bar represents the percentage of R00 awards activated at a different institution. The bars always add to 100% but the change in proportions illustrates that proportionally fewer awards are being made to those remaining at the same institution.

Great idea. We have included a stacked bar chart here. Since the stacked bar chart is percentages, we felt it was important to also show the total numbers so we still included the previous chart also but removed the percentage numbers from it. We also changed the departmental analysis to stacked bar charts. This shows the stark difference between 2008-2012 and 2013 onward. These changes were made in the revised Fig. 1.

• Minor question: I would love to see Table 3 and Table 4 as a time-series. Has the proportion of recipients at various institution types changed with time?

This is a great suggestion and we felt it fit best in Figure 5, so we’ve added it there.

• Table 3 is useful but only indirectly addresses my first "Recommendation to the Authors" from my previous review. I did some number crunching myself from the data provided. Assuming I did this correctly: If you're a K99 awardee at a private institute, you had a 76.3% change of getting an R00 compared to 80.4% for a K99 awardee at a public institution. If you're a K99 awardee at a top-funded institution, you had a 76.8% chance of R00 compared to 78.6% for a lower-funded institution. I would have liked to see more figures and tables to illustrate conversion rates by institution type in this way. Interestingly, to me, these data suggest that there are not enormous conversion rate differences by institution type (though looking at these now, I am confused at the 89% statistic in line 174 and where that comes form, since it is much higher than what I've calculated).

Thank you for this suggestion and these comments. Please see above where we describe how we incorrectly overestimated the 89% statistic. This has been corrected. As the reviewer suggested, we now show yearly percent of grants to specific institution types in the revised Figure 5. We agree with the reviewer that showing the conversion rate by institution type is interesting; however, it is fairly obvious from the new panels in Figure 5 that there is not much difference in conversion rate. Thus, to avoid crowding too many panels into the figure, we opted to keep the stacked bar plot.

Reviewer #3 (Recommendations for the Authors):

-One minor change to Figure 1C would be to switch the color coding for the lines so that they match with 1D whereby "same institution" would be white circles, or whatever the authors decide would be best for consistency since they are similar comparisons.

Thank you for this suggestion. We have corrected this to be consistent.

-Minor note for lines 459-461: I would suggest changing the wording to "intersectional inequalities" as it is not that a scientist's identities impact their careers as much as how those identities are positioned within an unequal opportunity structure and differentially treated that produce varying career trajectories and experiences of marginalization and cumulative (dis)advantages.

Thank you and we agree with you. We have made this correction.

-To carry forward a suggestion for the authors in my previous review, future research that more fully explores the research infrastructure of institutions for how top NIH funded institutions continue to be top funded institutions year after year could help clarify some of the career mobility and same/similar institution hiring found in the data. Rather than hand coding institutions for some of the infrastructure, the National Center for Education Statistics' Integrated Postsecondary Education Data System (IPEDS) has data on colleges and universities including whether they operate a hospital, have a medical degree, and many other interesting data about student and faculty demographics, institutional expenditures (including research budgets), and degrees awarded in different fields of study (undergrad and grad) that may be helpful to the authors as they continue their research stream in this area.

Thank you very much. We will look into this data set as we continue our investigations in this area.

Author Response

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

Reviewer #1 (Recommendations For The Authors):

The discussion seems to imply that the ball-and-chain peptide is or is related to the common gate. (Although it isn't stated explicitly, it is implied based on the presentation of the gating model in Figure 8 immediately after the discussion of common gating, and the simultaneous opening of both pores in Figure 8). What does the asymmetric structure say about the relationship between the N-term peptide and common gating in ClC-2? It seems like this structure suggests that the CTDs can independently rotate, and independently bind N-terminal peptide, which might not be expected to impact both pores. Some additional clarification and/or discussion of these ideas could be helpful here.

We thank the reviewer for raising these very important points. We agree we should have been more explicit and have now expanded our discussion on this topic, highlighting the independent movement of the N-term peptide and CTDs and clarifying that it is currently unknown whether CLC-2 has a common gate (lines 431484).

Discussion of "Revised Framework for CLC-2 gating": I think this would be a little easier to follow if most of the legend from Figure 8 was in the main text at the end of that section. Also, additional labels in Figure 8 (of the glutamates, the N-terminal peptide, and what the CTD arrows represent).

We have revised this section of the text and added labels to the (revised) Figure as suggested.

Line 261: typo, misspelling of "hydrogen"

Fixed. (Now line 279.)

Figure 6 - supplement 2B: Looks like an error in numbering y-axis - should be 90/120/150, I think. Can you show the three data points for the WT initial current rectification? Can you clarify whether the 3 that you are analyzing are the ones where AK42 the AK42 "zero current" level is not more than the initial positive current?

We apologize for this error, which arose from the Y-axis label overlapping the tick labels, so 90/120/150 showed as 90/20/50. We have fixed this error and have added a new panel (C) to show three data points for the WT initial current rectification. In the Figure legend to panel C, we clarify that the 3 experiments we analyzed are the ones where the AK-42 current level is not more than the initial current at 80 mV.

Reviewer #2 (Recommendations For The Authors):

1. It appears from a close inspection of Figure 2 that the TM dimer is not quite symmetric, but I couldn't tell for sure from the figures as presented. No comment is made in the methods about symmetry imposed, and the authors explicitly comment on asymmetry in the cytoplasmic domain. It would be useful to have an explicit discussion of the TM dimer symmetry.

We have now explicitly stated that the TM dimer is symmetric, and we have clarified the wording in the Methods:

Main text, line 81: "The TM region of CLC-2 displays a typical CLC family symmetric homodimeric structure, with each subunit containing an independent Cl– pathway (Figure 2A, B)."

Methods (lines 557-558): "The following ab initio reconstruction and 3D refinement (for all structures presented in this paper) were performed with C1 symmetry (no symmetry imposed)."

1. For the simulations in Figure 5 Supplement 2, the N terminus flexibility is shown, but this of course can't be compared to a control. However, given the structural results, one might expect the JK helix to show changes in flexibility/mobility in the apo vs inactivated structures. Is this observed?

We agree that the structures strongly suggest the JK-helix is not as stable without the N-terminus bound. We did not perform comparative simulations on the JK helix in the apo vs inactivated structures. While we agree this could be of interest, we don’t think it is essential to our conclusions, and the simulations might need to be quite long to adequately capture dynamics of the JK helix. [In the simulation results shown in Figure 5 Supplement 2, our aim was to test the validity of the structure by determining whether the N-terminus remains bound to the channel in simulations. The plot shows that the N-terminus stays in the same binding pose with an average RMSD (to the initial structure) of less than 2 angstroms, which is generally considered to be relatively stable.]

1. I find the section "revised framework for ClC-2 gating" to be wanting. The ideas are illustrated in the cartoon, but should also be laid out in the text. In what ways are you revising the framework, and in what aspects are you carrying through ideas already proposed?

Thank you for raising this point, which was also raised by Reviewer 1. We have revised this section and the accompanying Figure (Figure 8 and Lines 431-484).

1. The authors mention in passing the idea that the hairpin could contribute to inward rectification (lines 227/8), but also suggest a role for the gating glutamate in this process. They also mention the idea of a common gate, but don't flesh out its function very much. These possibilities are very interesting and should be substantially fleshed out in the "framework" section, even if they cannot be fully answered yet.

We have expanded on these points in the “framework” section.

1. Figure 6E. points representing individual experiments should be shown.

We added points representing individual experiments for Delta N (normalized to WT) in the surface-expression experiments in Figure 6E. Individual data points for the electrophysiology experiments are in panel C; we did not replot these in panel E because some of the points would have been off scale.

1. The density in Figure 2A is hard to see, is there a better way to display it? Also, the orientation of the rightmost panel in Figure 2C is difficult to interpret.

We revised 2A to make the density easier to see. We revised Figure 2C so that the middle and rightmost panels have the same orientation.

1. P6. Line 87. This sentence is a little confusing, and perhaps could be a little clearer-the density is consistent with a Cl- ion, but no experiments have been done to support this, no?

We have clarified the wording as suggested (now line 89) and added references supporting Clˉ binding to the Sext site in CLCs (line 90).

1. P6 lines 89-98. Two lines of evidence, the conformation of the gate and the pinch point, both point to the structure representing a closed state. The wording as presented is a little hard to follow.

We have revised the wording in this paragraph (lines 92-111)

1. It's hard to distinguish water protons and oxygens in the lower right panel (QQQ).

We revised this panel (in Figure 3 – figure supplement 2) to better distinguish the water protons and oxygens.

Reviewer #3 (Recommendations For The Authors):

A few points to consider for improving the manuscript

1. It is intriguing that in the AK-42 structure, there is no density for the hairpin loop even though the CTD is in a symmetrical conformation as the apo. The authors could perhaps comment on whether there is any difference in the rectification properties of currents (or run-up) upon unblocking of AK-42 which may suggest that the hairpin binding is prevented by AK-42.

We have not yet performed the suggested experiment nor any experiments to examine state-dependence, though we agree such experiments would be informative. We have added a note on this point in the discussion, lines 334-337.

1. Although the conformation-dependent placement of the hairpin loop is convincing based on the density, the sequence assigned to this region is not conclusive.

To strengthen our conclusion concerning the hairpin assignment, we investigated fits of peptide segments from the disordered sections of the C-terminal cytoplasmic domain to the hairpin density. We found that these fits are not as good as that with the N-terminal peptide. This analysis is described in lines 179-181 and a new figure (Figure 5 – figure supplement 1). We appreciate the reviewer’s point that it is extremely difficult to conclusively assign residues that are not contiguous with the rest of the structure. Nevertheless, given the wide variety of evidence all pointing to the conclusion that the hairpin loop corresponds to residues 14-28, we think the assignment is on strong footing. We respectfully ask that you consider removing this criticism from the public review, as we think it will hinder the casual reader from recognizing the strength of the evidence: (1) of unresolved regions in CLC-2, residues 14-28 fit best; (2) residues 14-28 were previously identified as part of the ball blocking region (lines 158-161); (3) MD simulations support that the N-terminal residues stay stably bound (Figure 5 – figure supplement 4) (4) gain-of-function disease causing mutations map onto either the Nterminal residues or interacting residues on the TM domain (Figure 5 – figure supplement 6). Thank you for considering this request.

1. The authors should comment on the physiological relevance of the CBS domain rearrangements during gating.

We have added this sentence (lines 131-133): “The physiological relevance of C-terminal domain rearrangements is suggested by disease-causing mutations that alter channel gating (Estevez et al., 2004; Brenes et al., 2023).”

1. For the figures with cryo-EM maps, indicate the contour levels.

Contour levels are now indicated in the Figure legends.

1. It will be useful to the electrostatic map of the N-terminal peptide and the docking site.

This is now shown in Figure 5 – figure supplement 3 and Video 5.

1. Include a comment on the recent CLC-2 /AK-42 structure and if there are any differences in the structural features.

We added this text to lines 273-274: “The RMSD between our CLC2-TM-AK42 structure and that of Ma et al. is 0.655 Å, and the RMSD between the apo TM structures is 0.756 Å.”

Author Response

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

eLife assessment

The paper contains some useful analysis of existing data but there are concerns regarding the conclusion that there might be alternative mechanisms for determining the location of origins of DNA replication in human cells compared to the well known mechanism known from many eukaryotic systems, including yeast, Xenopus, C. elegans and Drosophila. The lack of overlap between binding sites for ORC1 and ORC2, which are known to form a complex in human cells, is a particular concern and points to the evidence for the accurate localization of their binding sites in the genome being incomplete.

Public Reviews:

Reviewer #1 (Public Review):

In the best genetically and biochemically understood model of eukaryotic DNA replication, the budding yeast, Saccharomyces cerevisiae, the genomic locations at which DNA replication initiates are determined by a specific sequence motif. These motifs, or ARS elements, are bound by the origin recognition complex (ORC). ORC is required for loading of the initially inactive MCM helicase during origin licensing in G1. In human cells, ORC does not have a specific sequence binding domain and origin specification is not specified by a defined motif. There have thus been great efforts over many years to try to understand the determinants of DNA replication initiation in human cells using a variety of approaches, which have gradually become more refined over time.

In this manuscript Tian et al. combine data from multiple previous studies using a range of techniques for identifying sites of replication initiation to identify conserved features of replication origins and to examine the relationship between origins and sites of ORC binding in the human genome. The authors identify a) conserved features of replication origins e.g. association with GC-rich sequences, open chromatin, promoters and CTCF binding sites. These associations have already been described in multiple earlier studies. They also examine the relationship of their determined origins and ORC binding sites and conclude that there is no relationship between sites of ORC binding and DNA replication initiation. While the conclusions concerning genomic features of origins are not novel, if true, a clear lack of colocalization of ORC and origins would be a striking finding. However, the majority of the datasets used do not report replication origins, but rather broad zones in which replication origins fire. Rather than refining the localisation of origins, the approach of combining diverse methods that monitor different objects related to DNA replication leads to a base dataset that is highly flawed and cannot support the conclusions that are drawn, as explained in more detail below.

Response: We are using the narrowly defined SNS-seq peaks as the gold standard origins and making sure to focus in on those that fall within the initiation zones defined by other methods. The objective is to make a list of the most reproducible origins. Unlike what the reviewer states, this actually refines the dataset to focus on the SNS origins that have also been reproduced by the other methods in multiple cell lines. We have changed the last box of Fig. 1A to make this clearer: Shared origins = reproducible SNS-seq origins that are contained in initiation zones defined by Repli-seq, OK-seq and Bubble-seq. This and the Fig. 2B (as it is) will make our strategy clearer.

Methods to determine sites at which DNA replication is initiated can be divided into two groups based on the genomic resolution at which they operate. Techniques such as bubble-seq, ok-seq can localise zones of replication initiation in the range ~50kb. Such zones may contain many replication origins. Conversely, techniques such as SNS-seq and ini-seq can localise replication origins down to less than 1kb. Indeed, the application of these different approaches has led to a degree of controversy in the field about whether human replication does indeed initiate at discrete sites (origins), or whether it initiates randomly in large zones with no recurrent sites being used. However, more recent work has shown that elements of both models are correct i.e. there are recurrent and efficient sites of replication initiation in the human genome, but these tend to be clustered and correspond to the demonstrated initiation zones (Guilbaud et al., 2022).

These different scales and methodologies are important when considering the approach of Tian et al. The premise that combining all available data from five techniques will increase accuracy and confidence in identifying the most important origins is flawed for two principal reasons. First, as noted above, of the different techniques combined in this manuscript, only SNS-seq can actually identify origins rather than initiation zones. It is the former that matters when comparing sites of ORC binding with replication origin sites, if a conclusion is to be drawn that the two do not co-localise.

Response: We agree. So the reviewer should agree that our method of finding SNS-seq peaks that fall within initiation zones actually refines the origins to find the most reproducible origins. We are not losing the spatial precision of the SNS-seq peaks.

Second, the authors give equal weight to all datasets. Certainly, in the case of SNS-seq, this is not appropriate. The technique has evolved over the years and some earlier versions have significantly different technical designs that may impact the reliability and/or resolution of the results e.g. in Foulk et al. (Foulk et al., 2015), lambda exonuclease was added to single stranded DNA from a total genomic preparation rather than purified nascent strands), which may lead to significantly different digestion patterns (ie underdigestion). Curiously, the authors do not make the best use of the largest SNS-seq dataset (Akerman et al., 2020) by ignoring these authors separation of core and stochastic origins. By blending all data together any separation of signal and noise is lost. Further, I am surprised that the authors have chosen not to use data and analysis from a recent study that provides subsets of the most highly used and efficient origins in the human genome, at high resolution (Guilbaud et al., 2022).

Response: 1) We are using the data from Akerman et al., 2020: Dataset GSE128477 in Supplemental Table 1. We have now separately examined the core origins defined by the authors to check its overlap with ORC binding (Supplementary Fig. S8b)

2) To take into account the refinement of the SNS-seq methods through the years, we actually included in our study only those SNS-seq studies after 2018, well after the lambda exonuclease method was introduced. Indeed, all 66 of SNS-seq datasets we used were obtained after the lambda exonuclease digestion step. To reiterate, we recognize that there may be many false positives in the individual origin mapping datasets. Our focus is on the True positives, the SNS-seq peaks that have some support from multiple SNS-seq studies AND fall within the initiation zones defined by the independent means of origin mapping (described in Fig. 1A and 2B). These True positives are most likely to be real and reproducible origins and should be expected to be near ORC binding sites.

We have changed the last box of Fig. 1A to make this clearer: Shared origins = reproducible SNS-seq origins that are contained in initiation zones defined by Repli-seq, OK-seq or Bubble-seq.

Ini-seq by Torsten Krude and co-workers (Guillbaud, 2022) does NOT use Lambda exonuclease digestion. So using Ini-seq defined origins is at odds with the suggestion above that we focus only on SNS-seq datasets that use Lambda exonuclease. However, Ini-seq identifies a much smaller subset of SNS-seq origins, so, as requested, we have also done the analysis with just that smaller set of origins, and it does show a better proximity to ORC binding sites, though even then the ORC proximate origins account for only 30% of the Ini-seq2 origins (Supplementary Fig. S8d). Note Ini-seq2 identifies DNA replication initiation sites seen in vitro on isolated nuclei.

Update in response to authors' comments on the original review:

While the authors have clarified their approach to some aspects of their analysis, I believe they and I are just going to have to disagree about the methodology and conclusions of this work. I do not find the authors responses sufficiently compelling to change my mind about the significance of the study or veracity of the conclusions. In my opinion, the method for identification of strong origins is not robust and of insufficient resolution. In addition, the resolution and the overlap of the MCM Chip-seq datasets is poor. While the conclusion of the paper would indeed be striking and surprising if true, I am not at all persuaded that it is based on the presented data.

Reviewer #2 (Public Review):

Tian et al. performed a meta-analysis of 113 genome-wide origin profile datasets in humans to assess the reproducibility of experimental techniques and shared genomics features of origins. Techniques to map DNA replication sites have quickly evolved over the last decade, yet little is known about how these methods fare against each other (pros and cons), nor how consistent their maps are. The authors show that high-confidence origins recapitulate several known features of origins (e.g., correspondence with open chromatin, overlap with transcriptional promoters, CTCF binding sites). However, surprisingly, they find little overlap between ORC/MCM binding sites and origin locations.

Overall, this meta-analysis provides the field with a good assessment of the current state of experimental techniques and their reproducibility, but I am worried about: (a) whether we've learned any new biology from this analysis; (b) how binding sites and origin locations can be so mismatched, in light of numerous studies that suggest otherwise; and (c) some methodological details described below.

• I understand better the inclusion/exclusion logic for the samples. But I'm still not sure about the fragments. As the authors wrote, there is both noise and stochasticity; the former is not important but the latter is essential to include. How can these two be differentiated, and what may be the expected overlap as a function of different stochasticity rates?

It is difficult to separate the effect of noise from the effect of stochastic firing of origins. We therefore took the simplest approach: focus only on the most reproducible origins (shared origins) and ignore the non-reproducible origins. At least the most reproducible origins can be used to test the hypotheses regarding origin firing.

• Many of the major genomic features analyzed have already been found to be associated with origin sites. For example, the correspondence with TSS has been reported before:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6320713/

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6547456/

• Line 250: The most surprising finding is that there is little overlap between ORC/MCM binding sites and origin locations. The authors speculate that the overlap between ORC1 and ORC2 could be low because they come from different cell types. Equally concerning is the lack of overlap with MCM. If true, these are potentially major discoveries that butts heads with numerous other studies that have suggested otherwise.

The key missing dataset is ORC1 and ORC2 CHiP-seq from the same cell type. This shouldn't be too expensive to perform, and I hope someone performs this test soon. Without this, I remain on the fence about how much existing datasets are "junk" vs how much the prevailing hypothesis about replication needs to be revisited. Nonetheless, the authors do perform a nice analysis showing that existing techniques should be carefully used and interpreted.

We agree that a thorough set of ChIP-seq data (with multiple antibodies or with equivalent techniques that do not use antibodies) for all six subunits of ORC in mammalian cells will be very useful for the field. Note, though, that just by simple cell lysis, it is very easy to divide human ORC into at least three different parts: ORC1, ORC2-5, and ORC6. The subunits do not form as robust a complex as seen in the yeasts and in flies.

Reviewer #3 (Public Review):

Summary: The authors present a thought-provoking and comprehensive re-analysis of previously published human cell genomics data that seeks to understand the relationship between the sites where the Origin Recognition Complex (ORC) binds chromatin, where the replicative helicase (Mcm2-7) is loaded, and where DNA replication actually beings (origins). The view that these should coincide is influenced by studies in yeast where ORC binds site-specifically to dedicated nucleosome-free origins where Mcm2-7 can be loaded and remains stably positioned for subsequent replication initiation. However, this is most certainly not the case in metazoans where it has already been reported that chromatin bindings sites of ORC and Mcm2-7 do not necessarily overlap, nor do they always overlap with origins. This is likely due to Mcm2-7 possessing linear mobility on DNA (i.e., it can slide) such that other chromatin-contextualized processes can displace it from the site in which it was originally loaded. Additionally, Mcm2-7 is loaded in excess and thus only a fraction of Mcm2-7 would be predicted to coincide with replication start sites. This study reaches a very similar conclusion of these previous studies: they find a high degree of discordance between ORC, Mcm2-7, and origin positions in human cells.

Strengths: The strength of this work is its comprehensive and unbiased analysis of all relevant genomics datasets. To my knowledge, this is the first attempt to integrate these observations. It also is an important cautionary tale to not confuse replication factor binding sites with the genomic loci where replication actually begins, although this point is already widely appreciated in the field. Response: Thank you for recognizing the comprehensive and unbiased nature of our analysis. Our findings will prevent the unwise adoption of ORC or MCM binding sites as surrogate markers of origins and will stimulate the field to try and improve methods of identifying ORC or MCM binding until the binding sites are found to be proximal to the most reproducible origins. The last possibility is that there are ORC- or MCM-independent modes of defining origins, but we have no evidence of that.

Weaknesses: The major weakness of this paper is the lack of novel biological insight and that the comprehensive approach taken failed to provide any additional mechanistic insight regarding how and why ORC, Mcm2-7, and origin sites are selected or why they may not coincide.

Response: we agree that we cannot provide a novel biological insight from this kind of meta-analysis. The importance of this study is in highlighting that there is either significant problems with the data collected till now (preventing the co-localization of ORC or MCM binding sites with the most reproducible origins) or ORC and MCM binding sites are often far away from where the most reproducible origins fire, which should make us consider ways in which origins could be activated kilobases away from ORC and MCM binding sites.

Recommendations for the authors:

Reviewer #2 (Recommendations For The Authors):

All suggestions and recommendations were described in a previous review.

Reviewer #3 (Recommendations For The Authors):

The most significant omission is a contextualization of the results in the discussion and an explanation of why these results matter for the biology of replication, disease, and/or our confidence in the genomic techniques reported on in this study. As written, the discussion simply restates the results without any interpretation towards novel insight. I suggest that the authors revise their discussion to fill this important gap.

A second important, unresolved point is whether replication origins identified by the various methods differ due to technical reasons or because different cell types were analyzed. Given the correlation between TSS and origins (reported in this study but many others too), it is somewhat expected that origins will differ between cell types as each will have a distinct transcriptional program. This critique is partly addressed in Figure S1C. However, given the conclusion that the techniques are only rarely in agreement (only 0.27% origins reproducibly detected by the four techniques), a more in-depth analysis of cell type specific data is warranted. Specifically, I would suggest that cell type-specific data be reported wherever origins have been defined by at least two methods in the same cell type, specifically reporting the percent of shared origins amongst the datasets. This type of analysis may also inform on whether one or more techniques produces the highest (or lowest) quality list of true origins.

We have done what has been suggested: used K562 cell type-specific data because here the origins have been defined by at least two methods in the same cell type and reported the percent of shared origins amongst the datasets (Supp. Fig. S4).

Other MINOR comments include:

• Line 215: the authors show that shared origins overlap with TF binding hotspots more often than union origins, which they claim suggests "that they are more likely to interact with transcription factors." As written, it sounds like the authors are proposing that ORC may have some direct physical interaction with transcription factors. Is this intended? If so, what support is there for this claim?

The reviewer is correct. We have rephrased because we have no experimental support for this claim.

• In the text, Figure 3G is discussed before Figure 3F. I suggest switching the order of these panels in Figure 3.

Done.

• It's not clear what Figure 5H to Figure 6 accomplishes. What specifically is added to the story by including these data? Is there something unique about the high confidence origins? If there is nothing noteworthy, I would suggest removing these data.

We want to keep them to highlight the small number of origins that meet the hypothesis that ORC and MCM must bind at or near reproducible origins. These would be the origins that the field can focus in on for testing the hypothesis rigorously. They also show the danger of evaluating proximity between ORC or MCM binding sites with origins based on a few browser shots. If we only showed this figure, we could conclude that ORC and MCM binding sites are very close to reproducible origins.

• Line 394: "Since ORC is an early factor for initiating DNA replication, we expected that shared human origins will be proximate to the reproducible ORC binding sites." This is only expected if one disbelieves the prior literature that shows that ORC and origins are not, in many cases, proximal. This statement should be revised, or the previous literature should be cited, and an explanation provided about why this prior work may have missed the mark.

We do not know of any genome-wide study in mammalian cell lines where ORC binding sites and MCM binding have been compared to highly reproducible origins, or that show that these binding sites and highly reproducible origins are mostly not proximal to each other. Most studies cherry pick a few origins and show by ChIP-PCR that ORC and/or MCM bind near those sites. Alternatively, studies sometimes show a selected browser shot, without a quantitative measure of the overlap genome wide and without doing a permutation test to determine if the observed overlap or proximity is higher than what would be expected at random with similar numbers of sites of similar lengths. In the revised manuscript we have discussed Dellino, 2013; Kirstein, 2021; Wang, 2017; Mas, 2023. None of them have addressed what we are addressing, is the small subset of the most reproducible origins proximal to ORC or MCM binding sites?

• Line 402-404: given the lack of agreement between ORC binding sites and origins the authors suggest as an explanation that "MCM2-7 loaded at the ORC binding sites move much further away to initiate origins far from the ORC binding sites, or that there are as yet unexplored mechanisms of origin specification in human cancer cells". The first part of this statement has been shown to be true (Mcm2-7 movement) and should be cited. But what do the authors mean by the second suggestion of "unexplored mechanisms"? Please expand.

We have addressed this point in the revised manuscript.

• The authors should better reference and discuss the previous literature that relates to their work, some of these include Gros et al., 2015 Mol Cell, Powell et al., 2015 EMBO J, Miotto et al., 2016 PNAS, but likely there are many others.

We have addressed this point in the revised manuscript.

Note for authors:

Line 107: The introduction discusses the mechanism for yeast ORC recognizes specific origins and discusses the Orc4 contribution, but it is known that Orc2 also binds DNA on a base-specific manner (see PMID 33056978). Thus Lee et al. did not "humanize ORC" as stated.

Done

Lines 117-119: Two of the cited papers are on endo-reduplication and not on initiation in a normal cell cycle and this should be pointed out. Second, there is contradictory evidence that ORC is essential in human cells and this should be cited (PMID 33522487)

Done

Author Response

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

Based on the reviewer comments (see below) and subsequent discussion between the reviewers and the Reviewing Editor, I would like to invite the authors to make major revisions, including new experiments. However, if major new experiments are not feasible, as may be the case, then at a minimum, I would urge the authors to:

1. Tone down the language regarding a causative role for changes in GH/IGF-I signaling in mediating the effects of Tmem63 on the skeleton, and also be very open in acknowledging the lack of mechanistic insight into how Tmem regulates GH signaling.

Response: We toned down the language as suggested and also acknowledged the lack of mechanistic insights into how Tmem263 regulates GH signaling.

1. Revise/redo or if not possible, then delete the problematic experiment in Fig. 5E.

Response: We have included additional Western blot data in Figure 5 from control WT and KO male mice without exogenous GH injection. In the absence of GH injection, we could not detect Jak2 and Stat5 phosphorylation in the liver of male WT and KO mice.

1. Address the comments about liver feminization.

Response: We have performed additional analysis as suggested by reviewer # 3. We have now included additional data to address the issue of liver feminization (new Fig. 6G-I and Figure 6-figure supplement 1). We plan to expand on this very topic in future studies as this is an interesting transcriptional phenomenon.

1. Revise the manuscript to address as many of the recommendations for the authors as possible, many of which can be addressed by textual edits. Response: We have addressed as many of the textual changes as suggested in the revised manuscript.

Reviewer #2 (Recommendations for The Authors):

TMEM263 has been suggested to be associated with bone mineral density and growth in humans and mice, but the functional role of this transmembrane protein in the regulation of bone metabolism is unknown. With the knockout mouse approach, this manuscript demonstrates that Tmem263 is essential for longitudinal bone growth in the mouse as deletion of Tmem263 in knockout (KO) mice developed severe postnatal growth impairment and proportional dwarfism. It is determined that the dwarfism was caused by a substantial reduction in liver expression of growth hormone receptor (GHR), a slight increase in serum GH, and a reduction in serum IGF-I, which resulted in disruptive of GH/IGF-I regulatory axis of endochondral bone formation.

The study was relatively well designed, and the results in general are supportive of the conclusions. While this study discloses new and intriguing functional information about a novel cytoplasmic membrane gene, there are a few minor issues that the authors may wish to address. These issues are listed in the following:

1. One of the intriguing findings of this manuscript is that deletion of a gene encoding a small cytoplasmic membrane protein could cause a substantial reduction in the expression and protein levels of GHR. Inasmuch as a couple of potential explanations were offered in the Discussion section (first complete paragraph of page 10), there has been no attempt to test any of the suggested causes, since many of these potential mechanisms can readily be tested experimentally. Accordingly, the lack of mechanistic investigation into this intriguing effect renders the manuscript largely descriptive in nature.

Response: The point made by the reviewer is well taken. We do plan to have follow up studies to establish which among the mechanisms we highlighted in the discussion is contributing to the reduction in GHR transcript and protein level. Our present study is the first functional characterization of this enigmatic novel membrane protein. We anticipate that multiple follow-up studies are needed to gain a deeper understanding of the biology of Tmem263. We believe that our present study represents an important first step.

1. Because a major conclusion is that the bone phenotype of Tmem263 KO mice was caused by deficient hepatic expression and/or action of GHR, it would be helpful to (or strengthen) the conclusion if a brief comparison of the bone phenotype between GHR KO mice and Tmem263 KO mice is included in the Discussion section.

Response: We have now included this information in the revised manuscript.

1. In Figure 3, the cortical bone parameters (i.e., Tt.Ar, Ct.Ar, and Ct.Th), but none of the trabecular bone parameters (i.e., BV/TV, Tb.N, Tb.Th), were normalized against femur length. The authors did not provide a rationale for this differential treatment with the cortical bone parameters from the trabecular bone parameters. If the reason to normalize the cortical bone parameters against bone length was to demonstrate that the reduced cortical bone mass in mutants was related to the impaired longitudinal bone growth, then why did the authors not also assess whether the observed reduction in these trabecular bone parameters in KO mutants was proportional to reduced longitudinal bone growth?

Response: We actually made the exact adjustments that the reviewer refers to, as stated in the methods section. Please see page 14. The regions of interest (ROIs) of both the trabecular bone analysis and the cortical analysis in the mutants was reduced proportional to the length of the bone (40% smaller). The normalization to Tt Ar to femur length in Figure 3I was only meant to show that the reduction in Tt Ar in the mutants was proportional. We have modified the text in our result section for clarity.

1. Elements described in Fig. 5A have been well documented. Therefore, Fig. 5A is unnecessary and can be deleted.

Response: We felt that Figure 5A should remain. It helps orient readers that are not familiar with the literature to be aware that both liver- and bone-derived IGF-1 contribute to longitudinal bone growth.

1. Figure 6 was performed with male KO mice. Were the altered gene expression profiles in female KO mice any different from male KO mice?

Response: We plan to perform RNA-seq in female mouse liver in our follow-up studies. We do not know, at present, whether and to what extent the liver transcriptomic profile would be different between male and female KO mice. As far as dwarfism and deficiency in skeletal acquisition, both male and female KO mice showed the same phenotypes.

1. The number of animals (or samples) per group in some of the Figures (i.e., Fig. 2G, 2I, 2J, 3A to J, the entire Fig. 4, 5D, 5F, and Suppl Fig. 1) is needed to be provided in the legends.

Response: We have included this information in the figure legends.

Reviewer #3 (Recommendations for The Authors):

1. Explain the discrepancy between the impact of KO on serum Igfbp3 (= decreased) vs. hepatic Igfbp3 (= unchanged).

Response: We do not have a plausible mechanism, at present, that can explain the reduction in circulating serum Igfbp3 level without an apparent reduction in Igfbp3 transcript level in the liver. In human studies, typically only serum IGFBP3 levels are measured but not the hepatic IGFBP3 transcript level. Therefore, it is unclear whether the circulating levels of IGFBP3 is being regulated at the posttranscriptional level, an issue that can be explored in future studies.

1. Line 215, 221, and elsewhere - Foxa1 does not show significant male-biased expression in mouse liver.

Response: We have removed Foxa1 from the text.

1. Line 225- According to the abstract of Ref. #45, Cux2 regulates a subset of sex-biased genes in the liver. The authors should compare the genes dysregulated by TMEM263-KO (Fig. 6) to those altered by Cux2 loss (Ref. #45) to ascertain whether the results of Fig. 6 are partially or entirely explained by Cux2 overexpression.

Response: We agree that this is a great area of future study. We do feel this, however, would be better explored in a more in-depth follow-up article. We felt, given the current direction of the paper it made more sense to include differential expression comparisons of male vs female, hypophysectomized vs sham control, and Stat5b-KO vs WT mouse liver gene expression data. Our future work will explore the transcriptomes of male and female WT and Tmem263-KO liver gene expression in the context of the observed physiology.

1. Line 262- "lower transcription of Ghr gene". A decrease in mRNA levels does NOT equate with a decrease in transcription per se. Altered mRNA splicing, poly A, export, cytoplasmic stability, etc. are all potential contributors.

Response: We have included these possibilities highlighted by the reviewer in our revised Discussion section.

1. Line 273, "TMEM263... most highly expressed in liver" Not correct - see Fig. 1C for TMEM263 RNA levels in mouse tissues.

Response: We have corrected the text on page 11.

1. Line 425 - Include GEO accession number.

Response: We have already uploaded our RNA-seq data to the NCBI Sequence Read Archive (SRA), and the data can be accessed under accession number # PRJNA938158.

1. Fig. 6 - Line 796 - Specify the age and sex of mice analyzed.

Response: We have included the information in the revised figure 6 legend.

1. Fig.2 - Suppl 1- Specify age of mice.

Response: We have included the information in the revised Figure 2-figure supplement 2.

1. Fig.2G -Specify the sex of the mice.

Response: For the P1 to P21 pups’ data, we did not separate by sex, as gender determination of pups at P1 and P7 can be challenging. We now indicated this in the figure legend.

1. Fig. 6A and 6C-6F: Which of these genes shows sex-dependent expression in wild-type liver? Use color to highlight gene names for genes that show male-biased or female-biased expression.

Response: We agree with the reviewer that additional labels on Figure 6A and 6C-F would be helpful to show genes of sex-bias. However, this is not the primary point of the paper. This topic deserves a much more in-depth analysis in follow up studies focused on defining the exact type and degree of transcript feminization in the liver of Tmem263-KO mice, as well as, its physiologic consequences. For readers interested in this topic, we have included the subfigures G-I in Figure 6 and for greater transcript level detail, figure 6 supplement 1.

Author Response

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

Reviewer #1

Recommendation 1: The authors reasoned upon the presence of a differential basal hydraulic stress in waves' valleys vs hills at first from the observation of "domes" formation upon 48h cultivation. I suggest performing a quantification to support the statement as a good scientific practice. Furthermore, it would strengthen the concept when the formation of domes was compared between the waves' dimensions as a different grade of cell extrusion was quantified. i.e., 50, 100, and 200 µm.

Response 1: Upon seeing the phenomenon (Author response image 1 A), we performed a count for domes on the 100 µm and saw a significant effect. We refrained from including the results as it is the subject of ongoing research in our lab. In response to the reviewer’s suggestion, we have included a graph (Author response image 1 B) showing the increasing number of domes over 48 hours from three 100 µm wave samples.

We have updated Figure 2A and B in the manuscript to include the new graph.

Author response image 1.

(A) shows dome (white arrows) over a 100 µm wave substrate. (B) is the number of accumulated domes in valley and hill regions, for 3 independent samples, over 48 hours.

Recommendation 2: Using RICM microscopy to quantify the cell basal separation with the substrate and hydraulic stress is very clever. Nevertheless, I am in doubt if the different intensity reported for the hills vs valley (Fig. 2G and H) is a result of the signal reduction at deeper Z levels. Since there is no difference in extrusion and forces between valleys and hills in the 200 µm waves but only in 50µm and 100µm, I would add this to the quantification. I would expect no intensity difference from RICM for the 200 µm sample if this is not an artefact of imaging.

Response 2: We performed additional experiments on blank wave substrates (both 100 and 200 µm) to ascertain the extent of reflection intensity drop (Author response image 2A). And, as correctly pointed out by Reviewer #1, there was a drop in intensity even without cells. On the 100 µm waves, hill reflections are on average ~27 % dimmer than valley reflections. Whereas, on the 200 µm waves, hill reflections are on average ~39 % dimmer.

Using this information, we performed a calibration on the RICM results obtained from both the 100 and 200 µm waves (Author response image 3B). The calibrated 100 µm data showed residual signatures of difference, whereas the calibrated 200 µm distributions appeared very similar. We noticed large cross- sample variations in the registered intensities, which will negatively impact effect size if not accounted for. To do this, we subsequently normalized both hill and valley intensities against planar region intensities for each sample. As shown by the final output (Author response image 3C), we were able to remove the skewness in the distributions. Moreover, 1-way ANOVA followed by a post hoc analysis with BH correction revealed a significant reduction in 100 µm hill/flat intensity ratio compared to 100 µm valley/flat intensity ratios (Δ~-23 %). Conversely, no significance was observed for the same comparison on the 200 µm waves.

Author response image 2.

(A). RICM from blank wave samples reveal a reduction in reflection intensity in hill regions compared to flat and valley regions.

Author response image 3.

(B) shows the RICM intensities after adjusting for the inherent reflection intensity drop shown in (A). (C) show the RICM intensities after normalization against planar region signals; this removes cross-sample variations and improve effect size of differences.

We have updated the manuscript Figure 2I and text accordingly. The blank wave results are included in Figure 2-figure supplement 1 along with updated text and summary data table in Supplementary File 4.

Recommendation 3: To measure 3D forces on top of the hills and valleys, the use of PAA gels is necessary. Since in Fig 3B, the authors show a difference in cell extrusion number between substrates and stiffnesses, I think it is necessary to confirm the presence of more extrusion in valleys vs hills on PAA gels. This would ensure the conclusion between normal forces and extrusion.

Response 3: We do have time-lapse data with monolayers on the PAA waves. However, we felt results from the flat regions were sufficient in supporting the point being made in the text. Specifically, our original intention with PAA gels was to show that the extrusion reductions seen in osmotic perturbations were by virtue of removing basal stress and not some cryptic osmotic response. Hydrogels were chosen because they can effectively dilute basal solute concentration and thereby reduce the osmotically induced water transport. Moreover, as fluid could freely move within the gel, the fluid stress can quickly equilibrate across the basal surface. In contrast, poorly water/solute permeable substrates could lead to localized spikes in solute concentration and transient basal regions with high fluid stress.

To get a sense of the potential difference in basal solute concentration between the two materials, we can do a quick hand-waving estimation. For monolayers on non-water/solute permeable PDMS of 20x20 mm and using the laser wavelength (640 nm) for RICM as an extreme estimate of basal separation, we should expect ~0.25 µl of total basal water content. On the other hand, we typically produce our PAM gel slabs using ~150 µl of precursor solutions. This means that, given similar amounts of solute, PAM gels will lead to monolayer basal osmolarity that is around 3 orders of magnitude lower than monolayers on PDMS, producing significantly lower osmotic potential. This implies from the outset that we should expect high survivability of cells on these substrates irrespective of curvature domains. Indeed, later immunoblotting experiments showed MDCKs exhibiting hyper activated FAK and Akt on PAM gels.

In response to Reviewer #1’s suggestion then, we have added another supporting time-lapse (Video 19) showing typical response of MDCK monolayers on 100 µm PAA waves (Author response image 4). Evident from the time-lapses, like the planar regions, cell extrusions were very rare. This supports the idea that on PAM gels the effects of basal hydraulic stress and asymmetric forces are marginal against the strong survival signals. And the response is similar to hyper-osmotic perturbations; there, we did not see a significant difference between valley and hill extrusions.

Author response image 4.

Time-lapse snapshot showing negligible MDCK extrusions 24 hours after confluency over PAM gel wave substrates.

Recommendation 4: Before proceeding with the FAK inhibitor experiment, the authors should better justify why the 4.1 wt % sucrose vs DMSO or NaCl is the most inert treatment. This can be done by citing relevant papers or showing time-lapses (as it is done for the higher FAKI14 dose).

Response 4: Although some cells have recently been shown to be able to transport and utilize sucrose, mammalian cells generally cannot directly take up polysaccharides for metabolism and this is frequently mentioned in literature: see (Ref. R1) for example. Without special enzymes to break sucrose down into monosaccharides, such as sucrase found in the gut, the sugars should remain spectators in the culture medium, contributing only to osmotic effects.

DMSO on the other hand, besides changing osmolarity, can also be integrated into cell membrane and pass through cells over time. It has been reported to chronically affect cell membrane properties and gene expressions (Ref. R2).

Finally, it is well known that both sodium and chloride ions are readily taken up and transported by cells (Ref R3). They help to regulate the transmembrane potential, which in turn can affect membrane bound proteins and biochemical reactions within a cell.

Hence, comparing the 3 hyper-osmotic perturbations, adding sucrose should have the least off- target effects on both the inhibitor study and the subsequent immunoblotting. And, in response to the reviewer’s recommendation, we have updated the text accordingly and included new references to support our statement.

Ref R1. H. Meyer, O. Vitavska, H. Wieczorek; Identification of an animal sucrose transporter. Journal of Cell Science 124, 1984–1991 (2011). Doi: 10.1242/jcs.082024

Ref R2. B. Gironi, Z. Kahveci, B. McGill, B.-D. Lechner, S. Pagliara, J. Metz, A. Morresi, F. Palombo, P. Sassi, P. G. Petrov; Effect of DMSO on the Mechanical and Structural Properties of Model and Biological Membranes. Biophysical Journal 119, 274-286 (2020). Doi: doi.org/10.1016/j.bpj.2020.05.037

Ref R3. X. Zhang, H. Li; Interplay between the electrostatic membrane potential and conformational changes in membrane proteins. Protein Science 28, 502-512 (2019). Doi: 10.1002/pro.3563

Recommendation 5: The data showing a FAK-dependent phosphorylation of AKT responsible for a higher cell survival rate in the hills is not yet completely convincing. Please show a reduced AKT phosphorylation level after FAK inhibition in high osmolarity levels. Furthermore, the levels of AKT activation seem to increase slightly upon substrate softening independently of FAK activation or osmotic pressure (i.e., Fig. 4E, Soft PDMS). The authors should comment on this in connection with the results shown for PAA gels.

Response 5: For the additional immunoblotting experiments, work is currently underway. We could not, however, complete these experiments in time for this revision, as both Cheng-Kuang and Xianbin will shortly be taking on new jobs elsewhere. David will continue with the immunoblotting studies and should be able to include the results in an update in the coming months. As for the apparent elevated levels of AKT seen on soft silicones, we speculate that it is because we cannot immunoblot cells that have died and were inevitably washed out at the start of the procedure. Inferring from the higher extrusion rates on these soft substrates, we could be missing a significant portion of stats. Specifically, we are missing all the cells that would have lowered AKT activation but died, and had we been able to collect those statistics, perhaps both the FAK and AKT should have shown lower levels. We risk committing survival bias on the results if we read too much into the data as is.

Alternatively, another explanation could be that, by virtue of survival of the fittest, we might have effectively selected a subpopulation of cells that were able to survive on lower FAK signals, or completely irrespectively of it.

At any rate, to prove our foregoing hypothesis would require us to perform comprehensive immunoblotting and total transcriptome analysis over different duration conditions. Unfortunately, we do not have the time to do that for the current article, but it could be developed into a stand-alone molecular biology investigation in future. We have included similar discussion in the main text.

Recommendation 6: In the discussion, the authors suggest the reported findings be especially relevant for epithelia that significantly separate compartments and regulate water and soluble transport. These are for example kidney epithelia (i.e., MDCK is the best experimental choice), retinal epithelium or intestinal epithelium. I would suggest that some proof-of-concept experiments could be done to support this concept. For example, I would expect keratinocytes (i.e., HaCaT) not to show a strong difference in extrusion rate between valleys and hills since the monolayer is not so sealed as kidney epithelium. In general, this kind of experiment would significantly strengthen the finding of this work.

Response 6: As recommended, we tracked the behavior of retina pigment epithelial cells (hTERT RPE-1 from ATCC) which do not form tight monolayers like MDCKs (Ref. R4). We did not detect extrusion events occurring from monolayers of these cells (Author response image 5). This is true even for portions of monolayers over waved regions.

Author response image 5.

Time-lapse snapshot showing non-existent o cell extrusions from RPE monolayers confluent for over 21 hours.

We have updated these findings in the main text discussions and included a new supporting time- lapse (Video 15) in our article.

Ref R4 F. Liu, T. Xu, S. Peng, R. A. Adelman, L. I. Rizzolo; Claudins regulate gene and protein expression of the retinal pigment epithelium independent of their association with tight junctions. Experimental Eye Research 198, 108157 (2020). Doi: 10.1016/j.exer.2020.108157

Recommendation 7 (minor point): Figure S1 needs to have clear notes indicating in each step what is what. i.e., where is glass, PDMS, NOA73, etc? A more detailed caption will help the figure's comprehension. Also "Cy52" should be changed to "soft silicone" to be consistent with the text (or Cy52 should be mentioned in the text).

Response 7 (minor point): Changes were made to Figure 1-figure supplement 1 to improve comprehension accordingly. CY52 was added to the main-text, next to the first appearance of the word soft silicone, to be consistent with the figures.

Recommendation 8 (minor point): The authors often mentioned that epithelial monolayers are denser on PAA gels. Please add a reference(s) to this statement.

Response 8 (minor point): The statement is an inference from visually comparing monolayers on PAM gels and PDMS. The difference is quite evident (Author response image 6). The density difference is in spite of the fact that the substrates share similar starting cell numbers.

To address the reviewer’s comment, we have combined time-lapses of monolayers on silicones and PAM gels side-by-side in Video 17 to facilitate convenient comparisons.

Author response image 6.

Time-lapse snapshot at 24 hours after confluence, showing conspicuously higher density of MDCK monolayers on PAM gel compared to those on silicon elastomer.

Reviewer #2

Recommendation 1: The sinusoidal wavy substrate that the authors use in their investigation is interesting and relevant, but it is important to realize that this is a single-curved surface (also known as a developable surface). This means that the Gaussian curvature is zero and that monolayers need to undergo (almost) no stretching to conform to the curvature. The authors should at least discuss other curved surfaces as an option for future research, and highlight how the observations might change. Convex and concave hemispherical surfaces, for example, might induce stronger differences than observed on the sinusoidal substrates, due to potentially higher vertical resultant forces that the monolayer would experience. The authors could discuss this geometry aspect more in their manuscript and potentially link it to some other papers exploring cell-curvature interactions in more complex environments (e.g. non-zero Gaussian curvature).

Response 1: In response to reviewer #2’s recommendation we have highlighted in the discussion of our text that our waves constitute a developable surface and that cells will experience little stretching for the most part. Based on our knowledge of how curvature can modulate forces and thus osmotic effects, we included some rudimentary analysis of what one would expect on hemispherical surfaces of two types: one that is periodic and contiguous (Ref. R5), and another with delineating flat regions (Ref. R6).

For epithelial monolayers in the first scenario, and on poorly solute/water permeable substrates, we should also expect to see a relatively higher likelihood of extrusions from concave regions compared to convex ones. Moreover, as the surfaces are now curved in both principal directions (producing larger out-of-plane forces), we should see the onset of differential extrusions seen in this study, but at larger length scales. For example, the effects seen on 100 µm hemicylindrical waves might now happen at larger feature size for hemispherical waves. Furthermore, as this kind of surface would invariably contain hyperbolic regions (saddle points), we might expect an intermediate response from these locations. If the forces in both principal directions offset each other, the extrusion response may parallel planar regions. On the other hand, if one dominates over the other, we may see extrusion responses tending to the dominating curvature (concave of convex).

On the other hand, on curved landscapes with discrete convex or concave regions, we should expect, within the curved surface, extrusion behaviors paralleling findings in this study. What would be interesting would be to see what happens at the rims (or skirt regions) of the features. At these locations we effectively have hyperbolically curved surfaces, and like before, we should expect some sort of competing effect between the forces generated from the principal directions. So, for dome skirts, we should see fewer extrusions when the domes are small, and vice versa, when they are larger. Meanwhile, for pit rims, we should see a reversed behavior. It should also be noted that the transitioning curvature between convex/concave and planar regions would also modulate the effect.

These effects might have interesting developmental implications. For instance, in developing pillar like tissues (e.g., villi) structures, the strong curvatures of nascent lumps would favor accumulation of cell numbers. However, once the size of the lumps reaches some critical value, epithelial cell extrusions might begin to appear at the roots of the developing structures, offsetting cell division, and eventually halting growth.

Ref R5. L. Pieuchot, J. Marteau, A. Guignandon, T. Dos Santos, I. Brigaud, P. Chauvy, T. Cloatre, A. Ponche, T. Petithory, P. Rougerie, M. Vassaux, J. Milan, N. T. Wakhloo, A. Spangenberg, M. Bigerelle, K. Anselme, Curvotaxis directs cell migration through cell-scale curvature landscapes. Nature Communications 9, 3995 (2018). Doi: 10.1038/s41467-018-06494-6

Ref R6. M. Werner, S. B.G. Blanquer, S. P. Haimi, G. Korus, J. W. C. Dunlop, G. N. Duda, D. W. Grijpma, A. Petersen, Surface curvature differentially regulates stem cell migration and differentiation via altered attachment morphology and nuclear deformation. Advanced Science 4, 1–11 (2017). Doi: 10.1002/advs.201600347

Recommendation 2: The discussion of the experiments on PAM gels is rather limited. The authors describe that cells on the PAM gels experience fewer extrusions than on the PDMS substrates, but this is not discussed in sufficient detail (e.g. why is this the case). Additionally, the description of the 3D traction force microscopy and its validation is quite limited and should be extended to provide more convincing evidence that the measured force differences are not an artefact of the undulations of the surface.

Response 2: We first saw a significant reduction in cell extrusions when we performed hyper-osmotic perturbations, and to eliminate possible off-target effects of the compounds used to increase osmolarity, we used three different compounds to be sure. In spite of this, we felt it would further support our argument, that basal accumulation of fluid stress was responsible for the extrusions, if we had some other independent means of removing fluid stress without directly tuning osmolarity through addition of extraneous solutes. We hence thought of culturing MDCK monolayers on hydrogels.

Hydrogels were chosen because they can effectively dilute basal solute concentration (for reference ions (Na+) are continuously pumped out basally by the monolayer) and thereby reduce the associated osmotically induced water transport. Moreover, as fluid could freely move within the gel, the fluid stress can quickly equilibrate across the basal surface. In contrast, poorly water/solute permeable substrates will lead to localized spikes in solute concentration and transient basal regions with high fluid stress.

To get a sense of the extent of difference in basal solute concentration between the two materials, we can do a quick hand-waving estimation. For monolayers on non-water-permeable PDMS of 20x20 mm, and using the laser wavelength (640 nm) for RICM as an extreme estimate of basal separation, we should expect ~0.25 µl of total basal water content. On the other hand, we typically produce our PAM gel slabs using ~150 µl of precursor solutions. This means that, given similar amounts of solute, PAM gels will lead to monolayer basal osmolarity that is around 3 orders of magnitude lower than monolayers on PDMS, producing significantly lower osmotic potential. This implies from the outset that we should expect high survivability of cells on these substrates. Indeed, later immunoblotting experiments showed MDCKs exhibiting hyper activated FAK and Akt on PAM gels.

As for the 3D TFM used in this study, it is actually implemented from a well-established finite element method to solve inverse problems in engineering and has been repeatedly validated in larger scale engineering contexts (Ref. R7). The novelty and contribution of our article is in its adaptation to reconstruct cellular forces at microscopic scales.

In brief, soft materials, such as hydrogels used in our case, are doped with fluorescent particles, coated with ECM, and then seeded with cells. The cells would exert forces that deform the soft substrate, thereby displacing the fluorescent particles from their equilibrium positions. This particle displacement can be extracted by producing an image pair with microscopy; first one with the cells, and subsequent one of relaxed gel after removal of cells with acutely cytotoxic reagents, such as SDS. There are several ways in which the displacement field can be extracted from the image pair. These include particle tracking velocimetry, particle image velocimetry, digital volume correlation, and optical flow.

We employed 3D Farneback optical flow in our study for its superior computational performance. The method was validated using synthetically generated images from Sample 14 of the Society for Experimental Mechanics DIC challenge. The accuracy of the calculated displacements using the 3D Farneback optical flow was then compared to the provided ground truth displacements. For the highest frequency displacement image pairs, an x-component root-mean-square-error (RMSE) value of 0.0113 was observed. This was lower than the 0.0141 RMSE value for the Augmented Lagrangian Digital Volume Correlation method. This suggested that the 3D Farneback optical flow is capable of accurately calculating the displacement between two bead images.

The displacement fields are then fed into a finite element suite (ANSYS in our case) along with the model and mesh of the underlying substrate structure to obtain node specific displacements. This is required because mech nodes do not typically align with voxel positions of displacements. With these node specific displacements, we subsequently solve the inverse problem for the forces using Tikhonov regularization (Ref. R8). The outcome is a vector of node specific forces.

In light of the above, to physically validate the method in our context would require the generation of a known ground truth force on the scale of pico- to nano-newtons and subsequently image the particle displacements from this force using confocal microscopy. The force must then be released in situ in order for the relaxed gel to be imaged again. This is not a straightforward feat at this scale, and a method that immediately springs to mind is magnetic tweezers. Unfortunately, this is a tool that we cannot develop within reasonable timeframes, as the method will have to be seamlessly integrated with our spinning-disk confocal. However, as a compromise, we have included an in-silico validation with our revised manuscript.

Specifically, given a finite element model with a predefined curvature, a known force was applied to the surface of the model (Author response image 7A). The resulting displacements were then calculated from the finite element solution. A 10% random noise is then added to the resulting displacement. The traction force recovery (Fig. R2-1 B) was then performed using the in-silico noisy displacements. To evaluate the accuracy of the recovery, the cosine similarity along with the mean norm of the force vectors were calculated. A value closer to 1 for both evaluation metrics indicates a more accurate reconstruction of the simulated traction force. The cosine similarity of the recovered traction forces to the original applied force was 0.977±0.056 while the norm of the recovered traction forces as a proportion of the original applied force was 1.016±0.165. As both values are close to 1 (i.e., identical), this suggested that the traction forces could be satisfactorily recovered using the finite-element based method.

In response to the reviewer’s recommendations then, additional content has been included in the main text to explain the use of PAM gels and the workings of our 3D TFM pipeline.

Ref R7. James F. Doyle, Modern Experimental Stress Analysis: Completing the Solution of Partially Specified Problems (John Wiley & Sons, Chichester, 2004).

Ref R8. Per Christian Hansen, Discrete Inverse Problems: Insight and Algorithms (siam, Philadelphia, 2010).

Author response image 7.

(A) shows simulated force field to generate simulated displacements. (B) shows force field reconstructed from simulated displacements with noise.

Recommendation 3: The authors show nuclear deformation on the hills and use this as evidence for a resultant downward-pointing force vector. This has, indeed, also been observed in other works referenced by the authors (e.g. Werner et al.), and could be interesting evidence to support the current observations, provided the authors also show a nuclear shape on the concave and flat regions. The authors could potentially also characterize this shape change better using higher-resolution data.

Response 3: We characterized nucleus deformation using Hoechst-stained samples as per recommendation. The deformation is estimated by dividing segmented nuclei volumes by best-fit ellipsoid volumes of same objects. In this way, objects exhibiting minimal bending will lead to values close to 1.0. The obtained graph is shown in figure Author response image 8B (and manuscript Figure 3D).

Author response image 8.

(A) an example of deformed nuclei on 50 µm wave hill region. (B) a Violin plot of calculated nuclear deformations across dimensions and features using segmented volume normalized against best-fit ellipsoid volume.

Our quantifications show a statistically significant difference in nuclei deformation measure medians between hill and valley cells on the 50 µm (0.973 vs 0.982) and 100 µm (0.971 vs 0.979) waves; this indicates that cells on the hills tend to have more deformed nuclei compared to cells in the valleys. Meanwhile, no significant difference was found for a similar comparison on 200 µm (0.978 vs 0.978) samples. For reference, the median found for cells pooled from planar regions was 0.975.

In response to the reviewer’s suggestions Figure 3 of our manuscript has been updated to include the new results on nuclei deformation. The text has also been updated to account for the new information to support our claims. The statistics are included in a new summary data table in Supplementary File 6.

Recommendation 4: The U-net for extrusion detection is a central tool used within this study, though the explanation and particularly validation of the tool are somewhat lacking. More clarity in the explanation and more examples of good (or bad) detections would help establish this tool as a more robust component of the data collection (on all geometries).

Response 4: The architecture of the neural network used in this study is outlined in supplementary figure S5a. To validate the performance of the model, a test dataset consisting of 200 positive examples and 100 negative examples were fed into the network and the resulting prediction was obtained from model. The confusion matrix of the model is shown in supplementary figure S5c. The weighted precision and recall of the model are 0.958 and 0.953 respectively.

Additionally, we have included examples of false positive and false negative detections in Figure 1-figure supplement 5 (Author response image 8). For false positive detections, these were typically observed to be extrusions that were labelled to have occurred the frame prior to the frame of interest (Author response image 9 bottom sequence). However, as the extrusion process is incomplete in the prior frame, there are still changes in the extruded cell body and the network falsely predicts this as a detection.

Author response image 9.

Examples of false negative and false positive extrusions registration.

Recommendation 5: The authors study the involvement of FAK in the observed curvature-dependent and hydraulic stress-dependent spatial regulation of cell extrusion. In one of the experiments, the authors supplement the cell medium with FAK inhibitors, though only in a hyper-osmotic medium. They show that FAK inhibition counteracts the extrusion-suppressing effect of a hyper-osmotic medium. However, no data is shown on the effect of FAK inhibitors within the control medium. Would the extrusion rates be even higher then?

Response 4: We proceeded, as suggested by the reviewer, to explore the effects of the FAK inhibitor on MDCK monolayers in our control medium. The results revealed that, at the 3 µM FAK concentration, where cells in sucrose media showed an elevated extrusion rate, monolayers in control medium quickly suffered massive cell death (Author response image 10) similar to what was seen when 6 µM FAK was introduced to sucrose medium.

This finding suggests that osmolarity protects against FAK inhibitors in a dose dependent manner. Moreover, as cell extrusions require an intact monolayer, its rates cannot increase indefinitely: a point will be reached where an intact monolayer can no longer be maintained.

We have updated the main text of our article to mention this observation, and also included a new time-lapse (Video 22) to demonstrate the effect.

Author response image 10.

Timelapse snapshot of MDCK monolayers over waves 4 hours after inclusion of focal adhesion kinase inhibitor.

Recommendation 6: The supplementary videos show two fields of view next to each other, which is not immediately clear to the viewer. I strongly advise the authors to add a clear border between the two panels, so that it is clear that the cells from one panel are not migrating into the next panel.

Response 6: A distinctive border has been added to the movies to separate panels showing different focal planes of the same stack.

Recommendation 7: The general quality and layout of the figures could be improved. Some figures would benefit from higher-resolution or larger cell images (e.g. Figure 2A, C, D), and the organisation of subpanels could be improved (e.g. especially in Figure 2). The box plots and bar graphs are also not consistent throughout the manuscript in terms of colouring and style, which should be improved.

Response 7: We have enlarged the figures in question accordingly, at the cost of reducing some information. However, the full scope of the sub-figures remains accessible in the supplementary movies. We have also tried to change the placement of the panels to improve readability. We have also adjusted the valley, hill, and flat coloring scheme for the extrusion boxplots in Figures 1 and 2 to make them consistent.

Recommendation 8: The graphs in Figures 3E and F are confusing and difficult to interpret. The x-axis states "Position along curve in radians" but it is unclear how to relate this to the position on the wavy substrate. The graphs also have a second vertical axis on the right ("valley-interface-hill"), which adds to the confusion. I would recommend the authors provide more explanation and consider a different approach of plotting this.

Response 8: We have removed the confusing plot of cross-sectional profile from the force graphs. To indicate positions on the waves, we have augmented radian values with Hill, Interface, and Valley accordingly.

Recommendation 9: Specify which silicone was used for the low-stiffness silicone substrates in the methods and in the main text.

Response 9: CY52 has been added to the main-text, next to the first appearance of the word soft silicone, to be consistent with the figures.

Recommendation 10: The flow lines that are plotted over the RICM data make it difficult to see the underlying RICM images. I would advise to also show the RICM images without the flow lines.

Response 10: The original movie S15 (now Video 16) showing the RICM overlapped with optical flow paths has now been replaced by a movie showing the same, but with the flow paths and RICM in separate panels.

Recommendation 11: In the first paragraph of the discussion, the authors write: "And this difference was both dependent on the sense (positive or negative)...". This is superfluous since the authors already mentioned earlier in the paragraph that the convex and concave regions (i.e. different signs of curvature) show differences in extrusion rates.

Response 11: The sentence has been changed to “And this difference was also dependent on the degree of curvature.”

Recommendation 12: In the second paragraph of the discussion, the authors mention that "basal fluid spaces under monolayers in hill regions were found consistently smaller than those in valley regions". Is this data shown in the figures of the manuscript? If so, a reference should be made because it was unclear to me.

Response 12: This statement is an inference from the comparison of the hill and valley RICM grey values. Specifically, RICM intensities are direct surrogates for basal separations (i.e., fluid space (as there cannot be a vacuum)) by virtue of the physics underlying the effect. To be more precise then, “inferred from RICM intensity differences (Figure 2I)” has been added to support the statement.

Recommendation 13: On page 7 of the discussion, the authors talk about positively and negatively curved surfaces. This type of description should be avoided, as this depends on the definition of the surface normal (i.e. is positive convex or concave?). Rather use convex and concave in this context.

Response 13: The wording has been changed accordingly.

Recommendation 14: The label of Table 8 reads "Table 2".

Response 14: The error has been corrected.

Reviewer #3

Recommendation 1: The central finding seems to be opposite to an earlier report (J Cell Sci (2019) 132, jcs222372), where MDCK cells in curved alginate tubes exhibit increased extrusion on a convex surface. I suggest that you comment on possible explanations for the different behaviors.

Response 1: The article in question primarily reported the phenomenon of MDCK and J3B1A monolayers detaching from the concave alginate tube walls coated with Matrigel. The authors attributed this to the curvature induced out-of-plane forces towards the center of the tubes. Up to this point, the findings and interpretation are consistent with our current study where we also find a similar force trend in concave regions.

To further lend support to the importance of curvature in inducing detachment, the authors cleverly bent the tubes to introduce asymmetry in curvature between outer and inner surfaces. Specifically, the outside bend is concave in both principal directions, whereas the inside bend is convex in one of its principal directions. As expected, the authors found that detachment rates from the outer surface were much larger compared to the inner one. Again, the observations and interpretations are consistent with our own findings; the convex direction will generate out-of-plane forces pointing into the surface, serving to stabilize the monolayer against the substrate. It should be noted however, since the inner-side tube is characterized by both convex and concave curvatures in its two principal directions, the resulting behavior of overlaying monolayers will depend on which of the two resulting forces become dominant. So, for gradual bends, one should expect the monolayers to still be able to detach from the inner tube surface. This is what was reported in their findings.

For their extrusion observations, I am surprised. Because their whole material (hydrogels) is presumably both solute and water permeable, I would be more inclined to expect very few extrusions irrespective of curvature. This is indeed the case with our study of MDCKs on PAM hydrogels, where the hydrogel substrate effectively buffers against the quick build-up of solute concentration and basal hydraulic stress. Without the latter, concave monolayer forces alone are unlikely to be able to disrupt cell focal adhesions. Indeed, the detachments seen in their study are more likely by exfoliation of Matrigel rather than pulling cells off Matrigel matrix entirely.

My guess is that the extrusions seen in their study are solely of the canonical crowding effect. If this was the case, then the detached monolayer on the outside bend could buffer against crowding pressure by buckling. Meanwhile, the monolayer on the inside bend, being attached to the surface, can only regulate crowding pressure by removing cells through extrusions. This phenomenon should be particular to soft matrices such as Matrigel. Using stiffer and covalently bonded ECM should be sufficient to prevent monolayers from detaching, leading to similar extrusion behaviors. In response to the reviewer’s recommendation then, we have included a short paragraph to state the points discussed in this response.

Recommendation 2: Fig 3E, F: The quantities displayed on the panels are not forces, but have units of pressure (or stress).

Response 2: we have changed “force” to “stress” according to the reviewer’s suggestion. The reason we kept the use of force in the original text was due to the fact that we were reconstructing forces. Due to discretization, the resulting forces will inevitably be assigned to element nodes. In between the nodes, in the faces, there will be no information. So, in order to have some form of continuity to plot, the face forces are obtained by averaging the 4 nodes around the element face. Unfortunately, element face areas are not typically of the same size, therefore the average forces obtained needs to be further normalized against the face area, leading to a quantity that has units of stress.

Recommendation 3: Fig 2D: Asterisks are hard to see.

Response 3: the color of the asterisks has been changed to green for better clarity against a B&W background.

Recommendation 4: p 19, l 7: Word missing in "the of molding"

Response 4: the typo has been amended to “the molding of”.

Author response

Reviewer #1 (Public Review):

Loss of skeletal muscle tissue from traumatic injury is debilitating. Restoring muscle mass and function remains a challenge. Using a mouse model, the authors performed punch biopsy injuries of the tibialis anterior in which the volume of muscle loss was varied to result in either successful muscle regeneration with a smaller injury or the unsuccessful outcome of fibrosis with a larger injury. For both conditions, a novel lipidomic profiling approach was used to evaluate pro-inflammatory and anti-inflammatory lipids at key time points post-injury with respect to collagen deposition, macrophage infiltration, muscle fiber regeneration, and force produced during isometric contractions. A key finding was that while all lipids increased at 3 days post-injury (dpi) and then declined through 14 dpi, pro-inflammatory lipids remained elevated during recovery from greater muscle loss which led to fibrosis. Maresin 1 was identified as an anti-inflammatory lipid that, when injected into injured muscle, reduced fibrosis, improved muscle regeneration, and partially restored the strength of contraction.

Strengths: The metabolipidomic profiling demonstrated here represents a novel approach to identifying pro-inflammatory and anti-inflammatory mediators of successful vs unsuccessful skeletal muscle regeneration. These findings may translate into a new therapeutic approach for promoting successful regeneration following volumetric muscle loss.

Weaknesses: Certain aspects of the data are overinterpreted; while some measures appear to have an adequate sample size to make sound conclusions, other measures are likely to lack sufficient statistical power given their variability. Presentation of the results would be strengthened by adhering to consistent terminology and labeling of figures throughout; specific examples are identified in recommendations to the authors. Several of the images used to illustrate differences between treatments are unconvincing because differences are not readily.

We agree with the reviewer and have scaled back some of the interpretation as well as clarified the sample sizes. We have also amended the text to maintain a consistent terminology.

Reviewer #2 (Public Review):

The study is novel and valuable to the field and provides new and important insights into the role of lipid mediators in VML injuries. By expanding our understanding of the mechanisms that regulate muscle regeneration following VML injuries, the study has the potential to guide the development of novel therapeutic interventions that promote tissue repair and recovery. The data presented in the manuscript is of good quality. The findings and conclusions are supported by a variety of different analyses (e.g., gene expression, histology, flow cytometry).

Despite the strengths of the study, some limitations are identified. Specifically, the impact of maresin 1 on macrophage phenotypes (M1/M2) could have been explored in more detail using histological or protein expression analysis. Moreover, additional data are needed to substantiate the claims about increased muscle regeneration. Lastly, the study does not address myofiber innervation, myofiber-type transitions, or motor unit remodeling.

We thank the reviewer for the suggestions and have performed a more in-depth exploration of macrophage phenotypes through additional scRNA-sequencing analysis. We have also included additional data describing how Maresin 1 impacts muscle stem cells through cyclic AMP. Respectfully, profiling myofiber innervation, motor unit remodeling and myofiber-type transitions are beyond the scope of this manuscript.