Author Response

We are grateful for the constructive comments of the reviewers. Here is a provisional response to major questions.

To Question 1, we appreciate that you point out that the phenotypes of pan-neuronal knockout of PDFR by unmodified Cas9 (Fig 2H-2I, in previous manuscript) whose morning anticipation still exist at some level (Fig a) though the decreases of morning anticipation index (Fig b) and advanced evening activity were not as pronounced as observed in han5304 (Fig 3C Hyun et al., 2005), our response is that the difference between pan-neuronal knockout of PDFR by unmodified Cas9 might be caused by the limited efficiency of unmodified Cas9 in our conditional system. We will adjust the relevant conclusions in the revised version, and these findings underscore the necessity to enhance the efficiency of the original Cas9

Author response image 1.

To Question 2, that some expression profiles of clock neurons are not consistent with previous reports, such as Dh31 and ChAT in s-LNvs, our response is that the differences can be attributed to the variation in expression patterns between 3’ terminal KI-LexA (used in this gene expression dissection) and KO-GAL4, KI-GAL4, or transgenic GAL4. We have indeed observed differences when identical sites were inserted in frame with Gal4 or LexA.

To Question 3, that our description of advanced morning anticipation versus no morning anticipation with the term "opposite" is not accurate enough, our response is that we will modify that. Mutants of CNMa or CNMaR exhibit advanced morning activity, suggesting an inhibitory role of CNMa/CNMaR. Mutants of Pdf/Pdfr, on the other hand, showed no morning anticipation, indicating a promoting role in morning anticipation.

To Question 4, whether we have generated transgenic UAS-sgRNA flies for all CCT genes or only a subset, our response is that we have indeed generated UAS-sgRNA flies for all CCT genes.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

In their previous publication (Dong et al. Cell Reports 2024), the authors showed that citalopram treatment resulted in reduced tumor size by binding to the E380 site of GLUT1 and inhibiting the glycolytic metabolism of HCC cells, instead of the classical citalopram receptor. Given that C5aR1 was also identified as the potential receptor of citalopram in the previous report, the authors focused on exploring the potential of the immune-dependent anti-tumor effect of citalopram via C5aR1. C5aR1 was found to be expressed on tumor-associated macrophages (TAMs) and citalopram administration showed potential to improve the stability of C5aR1 in vitro. Through macrophage depletion and adoptive transfer approaches in HCC mouse models, the data demonstrated the potential importance of C5aR1-expressing macrophage in the anti-tumor effect of citalopram in vivo. Mechanistically, their in vitro data suggested that citalopram may regulate the phagocytosis potential and polarization of macrophages through C5aR1. Next, they tried to investigate the direct link between citalopram and CD8+T cells by including an additional MASH-associated HCC mouse model. Their data suggest that citalopram may upregulate the glycolytic metabolism of CD8+T cells, probability via GLUT3 but not GLUT1-mediated glucose uptake. Lastly, as the systemic 5-HT level is down-regulated by citalopram, the authors analyzed the association between a low 5-HT and a superior CD8+T cell function against a tumor. Although the data is informative, the rationale for working on additional mechanisms and logical links among different parts is not clear. In addition, some of the conclusion is also not fully supported by the current data.

Thanks very much for your insightful evaluation and the constructive suggestions. We have thoroughly studied the comments and a provisional point-to-point response is shown as follows.

Strengths:

The idea of repurposing clinical-in-used drugs showed great potential for immediate clinical translation. The data here suggested that the anti-depression drug, citalopram displayed an immune regulatory role on TAM via a new target C5aR1 in HCC.

Thank you for your constructive comments. We believe that further investigation into the mechanisms by which citalopram modulates TAM function could provide valuable insights into its potential role in HCC therapy.

Weaknesses:

(1) The authors concluded that citalopram had a 'potential immune-dependent effect' based on the tumor weight difference between Rag-/- and C57 mice in Figure 1. However, tumor weight differences may also be attributed to a non-immune regulatory pathway. In addition, how do the authors calculate relative tumor weight? What is the rationale for using relative one but not absolute tumor weight to reflect the anti-tumor effect?

We appreciate your insights into the potential contributions of non-immune regulatory pathways to the observed tumor weight differences between Rag-/- and C57 mice, and we will further address this issue in our discussion. The relative tumor weight was calculated by assigning an arbitrary value of 1 to the Rag1^-/- mice in the DMSO treatment group, with all other tumor weights expressed relative to this baseline. As suggested, we will include absolute tumor weight data in our revised manuscript.

(2) The authors used shSlc6a4 tumor cell lines to demonstrate that citalopram's effects are independent of the conventional SERT receptor (Figure 1C-F). However, this does not entirely exclude the possibility that SERT may still play a role in this context, as it can be expressed in other cells within the tumor microenvironment. What is the expression profiling of Slc6a4 in the HCC tumor microenvironment? In addition, in Figure 1F, the tumor growth of shSlc6a4 in C57 mice displayed a decreased trend, suggesting a possible role of Slc6a4.

To identify the expression patterns of Slc6a4 in different cellular contexts within the HCC tumor microenvironment, we will conduct a thorough screening of HCC datasets that include single-cell sequencing analysis. The possible role of Slc6a4 on tumor growth will be verified with in vitro loss-of-function experiments.

(3) Why did the authors choose to study phagocytosis in Figures 3G-H? As an important player, TAM regulates tumor growth via various mechanisms.

Thank you for your question. We focused on this aspect because citalopram targets C5aR1-expressing TAM. C5aR1 is a receptor for complement component C5a, and complement components play a significant role in mediating the phagocytosis process in macrophages. In the revised manuscript, we will emphasize this rationale clearly.

(4) The information on unchanged deposition of C5a has been mentioned in this manuscript (Figures 3D and 3F), the authors should explain further in the manuscript, for example, C5a could bind to receptors other than C5aR1 and/or C5a bind to C5aR1 by different docking anchors compared with citalopram.

Thank you for your insightful comment. First, we will investigate the docking anchors involved in the binding of C5a to C5aR1 and compare these interactions with those of C5aR1 and citalopram. Additionally, we will discuss the potential binding of C5a to other receptors, providing a broader perspective on the signaling mechanisms.

(5) Figure 3I-M - the flow cytometry data suggested that citalopram treatment altered the proportions of total TAM, M1 and M2 subsets, CD4+ and CD8+T cells, DCs, and B cells. Why does the author conclude that the enhanced phagocytosis of TAM was one of the major mechanisms of citalopram? As the overall TAM number was regulated, the contribution of phagocytosis to tumor growth may be limited.

As suggested, we will restate the conclusion to enhance clarity and better articulate the relationship between citalopram treatment, TAM populations, and their phagocytic activity. Thank you for your valuable input.

(6) Figure 4 - what is the rationale for using the MASH-associated HCC mouse model to study metabolic regulation in CD8+T cells? The tumor microenvironment and tumor growth would be quite different. In addition, how does this part link up with the mechanisms related to C5aR1 and TAM? The authors also brought GLUT1 back in the last part and focused on CD8+T cell metabolism, which was totally separated from previous data.

We chose the MASH-associated HCC mouse model because it closely mimics the etiology of metabolic-associated fatty liver disease (MAFLD), which is a significant contributor to the development of cirrhosis and HCC. The inclusion of CD8⁺ T cells in our study is based on the understanding that citalopram targets GLUT1, which plays a crucial role in glucose uptake. CD8⁺ T cell function is heavily reliant on glycolytic metabolism, making it essential to investigate how citalopram’s effects on GLUT1 influence the metabolic pathways and functionality of these immune cells. The data presented in this section primarily aim to demonstrate how citalopram influences peripheral 5-HT levels, which subsequently affects CD8⁺ T cell functionality. By linking these findings, we will clarify how citalopram impacts both TAM and CD8⁺ T cells. In the revised manuscript, we will enhance the background information and provide relevant data support to avoid any gaps.

(7) Figure 5, the authors illustrated their mechanism that citalopram regulates CD8+T cell anti-tumor immunity through proinflammatory TAM with no experimental evidence. Using only CD206 and MHCII to represent TAM subsets obviously is not sufficient.

As suggested, more relevant experimental data will be included in the revised manuscript to better characterize the TAM populations and their roles in mediating the effects of citalopram on CD8⁺ T cells.

Reviewer #2 (Public review):

Summary:

Dong et al. present a thorough investigation into the potential of repurposing citalopram, an SSRI, for hepatocellular carcinoma (HCC) therapy. The study highlights the dual mechanisms by which citalopram exerts anti-tumor effects: reprogramming tumor-associated macrophages (TAMs) toward an anti-tumor phenotype via C5aR1 modulation and suppressing cancer cell metabolism through GLUT1 inhibition while enhancing CD8+ T cell activation. The findings emphasize the potential of drug repurposing strategies and position C5aR1 as a promising immunotherapeutic target. However, certain aspects of experimental design and clinical relevance could be further developed to strengthen the study's impact.

Thank you for your thoughtful review and constructive feedback, and we look forward to improving our manuscript accordingly.

Strength:

It provides detailed evidence of citalopram's non-canonical action on C5aR1, demonstrating its ability to modulate macrophage behavior and enhance CD8+ T cell cytotoxicity. The use of DARTS assays, in silico docking, and gene signature network analyses offers robust validation of drug-target interactions. Additionally, the dual focus on immune cell reprogramming and metabolic suppression presents a thorough strategy for HCC therapy. By emphasizing the potential for existing drugs like citalopram to be repurposed, the study also underscores the feasibility of translational applications.

Your insights reinforce the significance of our findings, and we will ensure that these points are clearly articulated in the revised manuscript to enhance its impact.

Major weaknesses/suggestions:

The dataset and signature database used for GSEA analyses are not clearly specified, limiting reproducibility. The manuscript does not fully explore the potential promiscuity of citalopram's interactions across GLUT1, C5aR1, and SERT1, which could provide a deeper understanding of binding selectivity. The absence of GLUT1 knockdown or knockout experiments in macrophages prevents a complete assessment of GLUT1's role in macrophage versus tumor cell metabolism. Furthermore, there is minimal discussion of clinical data on SSRI use in HCC patients. Incorporating survival outcomes based on SSRI treatment could strengthen the study's translational relevance.

By addressing these limitations, the manuscript could make an even stronger contribution to the fields of cancer immunotherapy and drug repurposing.

We appreciate your valuable suggestions. As suggested, we will take the following actions:

(1) GSEA analysis: we will clearly specify the datasets and signature databases used for the GSEA in the revised manuscript.

(2) Exploration of binding selectivity: we recognize the importance of exploring the potential promiscuity of citalopram’s interactions across GLUT1, C5aR1, and SERT1. As suggested, we will include a more detailed analysis of these interactions, which will help elucidate binding selectivity and its implications for therapeutic outcomes.

(3) GLUT1 knockdown in macrophages: to address the gap in our assessment of GLUT1’s role in macrophages, we will incorporate GLUT1 knockdown or knockout experiments in macrophages upon citalopram treatment. Moreover, a DARTS assay for GLUT1 in THP-1 cells will be conducted.

(4) Clinical data on SSRI use in HCC patients: Related data have been reported previously in PMID: 39388353 (Cell Rep. 2024 Oct 22;43(10):114818.). As detailed below:

“SSRIs use is associated with reduced disease progression in HCC patients

We determined whether SSRIs for alleviating HCC are supported by real-world data. A total of 3061 patients with liver cancer were extracted from the Swedish Cancer Register. Among them, 695 patients had been administrated with post-diagnostic SSRIs. The Kaplan-Meier survival analysis suggested that patients who utilized SSRIs exhibited a significantly improved metastasis-free survival compared to those who did not use SSRIs, with a P value of log-rank test at 0.0002. Cox regression analysis showed that SSRI use was associated with a lower risk of metastasis (HR = 0.78; 95% CI, 0.62-0.99).”

Author response image 1.

Author Response

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.

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 sgRNAs used to generate initial PARG KO cells in this manuscript target all three catalytically active isoforms (isoforms 1, 2 and 3), while isoforms 4 and 5 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 in Author response image 1. 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.

2) 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.

3) 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 sgRNAs used to generate PARG KO cells in this manuscript target all three catalytically active isoforms (isoforms 1, 2 and 3), while isoforms 4 and 5 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 in Author response image 2.

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 in Author response image 3. 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.

4) 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.

5) 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.

6) 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.

Author Response:

Reviewer #1 (Public review):

The authors of this study use electron microscopy and 3D reconstruction techniques to study the morphology of distinct classes of Drosophila sensory neurons *across many neurons of the same class.* This is a comprehensive study attempting to look at nearly all the sensory neurons across multiple sensilla to determine a) how much morphological variability exists between and within neurons of different and similar sensory classes, and 2) identify dendritic features that may have evolved to support particular sensory functions. This study builds upon the authors' previous work, which allowed them to identify and distinguish sensory neuron subtypes in the EM volumes without additional staining so that reconstructed neurons could reliably be placed in the appropriate class. This work is unique in looking at a large number of individual neurons of the same class to determine what is consistent and what is variable about their class-specific morphologies.

This means that in addition to providing specific structural information about these particular cells, the authors explore broader questions of how much morphological diversity exists between sensory neurons of the same class and how different dendritic morphologies might affect sensory and physiological properties of neurons.

The authors found that CO2-sensing neurons have an unusual, sheet-like morphology in contrast to the thin branches of odor-sensing neurons. They show that this morphology greatly increases the surface area to volume ratio above what could be achieved by modest branching of thin dendrites, and posit that this might be important for their sensory function, though this was not directly tested in their study. The study is mainly descriptive in nature, but thorough, and provides a nice jumping-off point for future functional studies. One interesting future analysis could be to examine all four cell types within a single sensilla together to see if there are any general correlations that could reveal insights about how morphology is determined and the relative contributions of intrinsic mechanisms vs interactions with neighboring cells. For example, if higher than average branching in one cell type correlated with higher than average branching in another type, if in the same sensilla. This might suggest higher extracellular growth or branching cues within a sensilla. Conversely, if higher branching in one cell type consistently leads to reduced length or branching in another, this might point to dendrite-dendrite interactions between cells undergoing competitive or repulsive interactions to define territories within each sensilla as a major determinant of the variability.

We thank the reviewer for the insightful comments and appreciation for our study.

Reviewer #2 (Public review):

Summary:

The manuscript employs serial block‐face electron microscopy (SBEM) and cryofixation to obtain high‐resolution, three‐dimensional reconstructions of Drosophila antennal sensilla containing olfactory receptor neurons (ORNs) that detect CO2. This method has been used previously by the same lab in Gonzales et. al, 2021. (https://elifesciences.org/articles/69896), which had provided an exemplary model by integrating high-resolution EM with electrophysiology and cell-type-specific labeling.

We thank the reviewer for expressing appreciation for our published study.

The previous study ended up correlating morphology with activity for multiple olfactory sensillar types. Compared to the 2021 study, this current manuscript appears somewhat incomplete and lacks integration with activity.

We thank the reviewer for their feedback. However, we would like to clarify that our previous study did not correlate morphology with activity to a greater extent than the current study. Both employed the same cryofixation, SBEM-based approach without recording odor-induced activity, but the focus of the current work is fundamentally different. While the previous study examined multiple sensillum types, the current study concentrates on a single sensillum type to address a distinct biological question regarding morphological heterogeneity. We appreciate the opportunity to clarify this distinction, and we hope that the revised manuscript more clearly conveys the unique scope and contributions of this study.

In fact older studies have also reported two-dimensional TEM images of the putative CO2 neuron in Drosophila (Shanbhag et al., 1999) and in mosquitoes (McIver and Siemicki, 1975; Lu et al, 2007), and in these instances reported that the dendritic architecture of the CO2 neuron was somewhat different (circular and flattened, lamellated) from other olfactory neurons.

We thank the reviewer for pointing this out. As noted in both the Introduction and Discussion sections, previous studies—including those cited by the reviewer—suggested that CO2-sensing neurons may have a distinct dendritic morphology. However, those earlier studies lacked the means to definitively link the observed morphology to CO2 neuron identity.

In contrast, our study assigns neuronal identity based on quantitative morphometric measurements, allowing us to confidently associate the unique dendritic architecture with CO2 neurons. Furthermore, we extend previous observations by providing full 3D reconstructions and nanoscale morphometric analyses, offering a much more comprehensive and definitive characterization of these neurons. We believe this represents a significant advancement over earlier work.

The authors claim that this approach offers an artifact‐minimized ultrastructural dataset compared to earlier. In this study, not only do they confirm this different morphology but also classify it into distinct subtypes (loosely curled, fully curled, split, and mixed). This detailed morphological categorization was not provided in prior studies (e.g., Shanbhag et al., 1999 ).

We thank the reviewer for acknowledging the significance of our study.

The authors would benefit from providing quantitative thresholds or objective metrics to improve reproducibility and to clarify whether these structural distinctions correlate with distinct functional roles.

We thank the reviewer for raising this point. However, we would like to clarify that assigning neurons to strict morphological subtypes was not the primary aim of our study. In practice, dendritic architectures can be highly complex, with individual neurons often displaying features characteristic of multiple subtypes. This is precisely why we included a “mixed” subtype category—to acknowledge and capture this morphological heterogeneity rather than impose rigid classification boundaries.

Our intent in defining subtypes was not to imply discrete functional classes, but rather to highlight the range of morphological variation observed across ab1C neurons. While we agree that exploring potential correlations between structure and function is an important future direction, the current study focuses on characterizing this diversity using 3D reconstruction and morphometric analysis. We hope this clarifies the purpose and scope of our morphological categorization.

Strengths:

The study makes a convincing case that ab1C neurons exhibit a unique, flattened dendritic morphology unlike the cylindrical dendrites found in ab1D neurons. This observation extends previous qualitative TEM findings by not only confirming the presence of flattened lamellae in CO₂ neurons but also quantifying key morphometrics such as dendritic length, surface area, and volume, and calculating surface area-to-volume ratios. The enhanced ratios observed in the flattened segments are speculated to be linked to potential advantages in receptor distribution (e.g., Gr21a/Gr63a) and efficient signal propagation.

We thank the reviewer for appreciating the significance our current study.

Weaknesses:

While the manuscript offers valuable ultrastructural insights and reveals previously unappreciated heterogeneity among CO₂-sensing neurons, several issues warrant further investigation in addition to the points made above.

(1) Although this quantitative approach is robust compared to earlier descriptive reports, its impact is somewhat limited by the absence of direct electrophysiological data to confirm that ultrastructural differences translate into altered neuronal function. A direct comparison or discussion of how the present findings align with the functional data obtained from electrophysiology would strengthen the overall argument.

We thank the reviewer for this comment. We would like to clarify, however, that our study does not claim that the observed morphological heterogeneity necessarily leads to functional diversity. Rather, we consider this as a possible implication and discuss it as a potential question for future research. This idea is raised only in the Discussion section, and we are carefully not to present functional diversity as a conclusion of our study. Nonetheless, we have reviewed the relevant paragraph to ensure the language remains cautious and does not overstate our interpretation.

We also acknowledge the significance of directly linking ultrastructural features to neuronal function through electrophysiological recordings. However, at present, it is technically challenging to correlate the nanoscale morphology of individual ORNs with their functional activity, as this would require volume EM imaging of the very same neurons that were recorded via electrophysiology. Currently, there is no dye-labeling method compatible with single-sensillum recording and SBEM sample preparation that allows for unambiguous identification and segmentation of recorded ORNs at the necessary ultrastructural resolution.

To acknowledge this important limitation, we have added a paragraph in the Discussion section, as suggested, to clarify the current technical barriers and to highlight this as a promising direction for future methodological advances.

(2) Clarifying the criteria for dendritic subtype classification with quantitative parameters would enhance reproducibility and interpretability. Moreover, incorporating electrophysiological recordings from ab1C neurons would provide compelling evidence linking structure and function, and mapping key receptor proteins through immunolabeling could directly correlate receptor distribution with the observed morphological diversity.

Please see our response to the comment regarding the technical limitations of directly correlating ultrastructure with electrophysiological data.

In addition, we would like to address the suggestion of using immunolabeling to map receptor distribution in relation to the 3D EM models. Currently, antibodies against Gr21a or Gr63a (the receptors expressed in ab1C neurons) are not available. Even if such antibodies were available, immunogold labeling for electron microscopy requires harsh detergent treatment to increase antibody permeability, damaging morphological integrity. These treatments would compromise the very morphological detail that our study aims to capture and quantify.

(3) Even though Cryofixation is claimed to be superior to chemical fixation for generating fewer artifacts, authors need to confirm independently the variation observed in the CO2 neuron morphologies across populations. All types of fixation in TEMs cause some artifacts, as does serial sectioning. Without understanding the error rates or without independent validation with another method, it is hard to have confidence in the conclusions drawn by the authors of the paper.

We thank the reviewer for raising concerns regarding potential artifacts in morphological analyses. However, we would like to clarify that cryofixation is widely regarded as a gold standard for ultrastructural preservation and minimizing fixation-induced artifacts, as supported by extensive literature. This is why we adopted high-pressure freezing and freeze substitution in our study.

We have also published a separate methods paper (Tsang et al., eLife, 2018) directly comparing our cryofixation-based protocol with conventional chemical fixation, demonstrating substantial improvements in morphological preservation (see the image below, adapted from Figure 2 of our 2018 eLife paper). This provides strong empirical support for the reliability of our approach.

Author response image 1.

Regarding the suggestion to validate observed morphological variation across populations: we note that determining the presence of artifacts requires a known ground truth, which is inherently unavailable as we could not measure the morphometrics of fly olfactory receptor neurons in their native state. In the absence of such a benchmark, we have instead prioritized using the best-available preparation methods and high-resolution imaging to ensure structural integrity.

Addressing these concerns and integrating additional experiments would significantly bolster the manuscript's completeness and advancement.

We appreciate the reviewer’s feedback. As discussed in our responses to the specific comments above, certain suggested experiments are currently limited by technical constraints, particularly in the context of high-resolution volume EM for insect tissues enclosed in cuticles.

Nevertheless, we have carefully addressed the reviewer’s concerns to the fullest extent possible within the scope of this study. We have revised the manuscript to clarify methodological limitations, added new explanatory content where appropriate, and ensured that our interpretations remain well grounded in the data. We hope these revisions strengthen the clarity and completeness of the manuscript.

Reviewer #3 (Public review):

Summary:

In the current manuscript entitled "Population-level morphological analysis of paired CO2- and odor-sensing olfactory neurons in D. melanogaster via volume electron microscopy", Choy, Charara et al. use volume electron microscopy and neuron reconstruction to compare the dendritic morphology of ab1C and ab1D neurons of the Drosophila basiconic ab1 sensillum. They aim to investigate the degree of dendritic heterogeneity within a functional class of neurons using ab1C and ab1D, which they can identify due to the unique feature of ab1 sensilla to house four neurons and the stereotypic location on the third antennal segment. This is a great use of volumetric electron imaging and neuron reconstruction to sample a population of neurons of the same type. Their data convincingly shows that there is dendritic heterogeneity in both investigated populations, and their sample size is sufficient to strongly support this observation. This data proposes that the phenomenon of dendritic heterogenity is common in the Drosophila olfactory system and will stimulate future investigations into the developmental origin, functional implications, and potential adaptive advantage of this feature.

Moreover, the authors discovered that there is a difference between CO2- and odour-sensing neurons of which the first show a characteristic flattened and sheet-like structure not observed in other sensory neurons sampled in this and previous studies. They hypothesize that this unique dendritic organization, which increases the surface area to volume ratio, might allow more efficient Co2 sensing by housing higher numbers of Co2 receptors. This is supported by previous attempts to express Co2 sensors in olfactory sensory neurons, which lack this dendritic morphology, resulting in lower Co2 sensitivity compared to endogenous neurons.

Overall, this detailed morphological description of olfactory sensory neurons' dendrites convincingly shows heterogeneity in two neuron classes with potential functional impacts for odour sensing.

Strength:

The volumetric EM imaging and reconstruction approach offers unprecedented details in single cell morphology and compares dendrite heterogeneity across a great fraction of ab1 sensilla.<br /> The authors identify specific shapes for ab1C sensilla potentially linked to their unique function in CO2 sensing.

We thank the reviewer for the insightful comments and appreciation for our study.

Weaknesses:

While the morphological description is highly detailed, no attempts are made to link this to odour sensitivity or other properties of the neurons. It would have been exciting to see how altered morphology impacts physiology in these olfactory sensory cells.

We agree that linking morphological variation to physiological properties, such as odor sensitivity, would be a highly valuable direction for future research. However, the aim of the current study is to provide an in-depth nanoscale characterization based on a substantial proportion of ab1 sensilla, highlighting morphological heterogeneity among homotypic ORNs.

At present, it is technically challenging to correlate the nanoscale morphology of individual ORNs with their physiological responses, as this would require volume EM imaging of the exact neurons recorded via single-sensillum electrophysiology. Currently, no dye-labeling method exists that is compatible with both single-sensillum recording and the stringent requirements of SBEM sample preparation to allow for unambiguous identification and segmentation of recorded ORNs.

To acknowledge this important limitation, we have added a paragraph in the Discussion section clarifying the current technical barriers and highlighting this as a promising area for future methodological development. Please also see our responses to the reviewer’s 4th comment below, where we present preliminary experiments examining whether odor sensitivity varies among homotypic ORNs.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this study, the authors showed that enalapril was able to reduce cellular senescence and improve health status in aged mice. The authors further showed that phosphorylated Smad1/5/9 was significantly elevated and blocking this pathway attenuated the protection of cells from senescence. When middle-aged mice were treated with enalapril, the physiological performance in several tissues, including memory capacity, renal function, and muscle strength, exhibited significant improvement.

Strengths:

The strength of the study lies in the identification of the pSMAD1/5/9 pathway as the underlying mechanism mediating the anti-senescence effects of enalapril with comprehensive evaluation both in vitro and in vivo.

Thanks very much for your insightful evaluation and the constructive suggestions. We have thoroughly studied the comments and a provisional point-to-point response is shown as follows.

Weaknesses:

The major weakness of the study is the in vivo data. Despite the evidence shown in the in vitro study, there is no data to show that blocking the pSmad1/5/9 pathway is able to attenuate the anti-aging effects of enalapril in the mice. In addition, the aging phenotypes mitigation by enalapril is not evidenced by the extension of lifespan.

Thanks for your comment. As suggested, we will feed LDN193189 to mice while using LDN193189 to block pSmad1/5/9, and will assess age-related phenotypes in the mice to demonstrate that the anti-aging effect of enalapril in mice is mediated through pSmad1/5/9.

We only assess the improvement in the health status of the aging mice, which indicate that enalapril can extend the healthy lifespan of aging mice. This is because we believe that lifespan is controlled by genetics. Therefore, this study focuses solely on the improvement of health phenotypes in aging mice by enalapril.

If it is necessary to show that NAC is able to attenuate enalapril effects in the aging mice. In addition, it would be beneficial to test if enalapril is able to achieve similar rescue in a premature aging mouse model.

Thanks for your suggestion. To our knowledge, NAC is an inhibitor of ROS, which is consistent with the antioxidant effect of enalapril. Therefore, we believe that NAC will not diminish the effect of enalapril.

For the premature aging mouse models, we examined the effect of enalapril on Lmna^G609G mice and other premature aging models and found that the effect was relatively modest. This may be due to differences in the genetic background of premature aging mice, leading to a less pronounced effect of enalapril compared to its impact on naturally aged mice.

Reviewer #2 (Public review):

This manuscript presents an interesting study of enalapril for its potential impact on senescence through the activation of Smad1/5/9 signaling with a focus on antioxidative gene expression. Repurposing enalapril in this context provides a fresh perspective on its effects beyond blood pressure regulation. The authors make a strong case for the importance of Smad1/5/9 in this process, and the inclusion of both in vitro and in vivo models adds value to the findings. Below, I have a few comments and suggestions which may help improve the manuscript.

Thanks very much for your insightful evaluation and the constructive suggestions. We have thoroughly studied the comments and a provisional point-to-point response is shown as follows.

A major finding in the study is that phosphorylated Smad1/5/9 mediates the effects of enalapril. However, the manuscript focused on the Smad pathway relatively abruptly, and the rationale behind targeting this specific pathway is not fully explained. What makes Smad1/5/9 particularly relevant to the context of this study?

Thanks for your comment. As stated in the manuscript, after we found that enalapril could improve the cellular senescence phenotype, we screened and examined key targets in important aging-related signaling pathways, such as AKT, mTOR, ERK (Fig. S2A), Smad2/3 and Smad1/5/9 (Fig. 2A). We found that only the phosphorylation levels of Smad1/5/9 significantly increased after enalapril treatment. Therefore, the subsequent focus of this study is on pSmad1/5/9.

Furthermore, their finding that activation of Smad1/5/9 leads to a reduction of senescence appears somewhat contradictory to the established literature on Smad1/5/9 in senescence. For instance, studies have shown that BMP4-induced senescence involves the activation of Smad1/5/8 (Smad1/5/9), leading to the upregulation of senescence markers like p16 and p21 (JBC, 2009, 284, 12153). Similarly, phosphorylated Smad1/5/8 has been shown to promote and maintain senescence in Ras-activated cells (PLOS Genetics, 2011, 7, e1002359). Could the authors provide more detailed mechanistic insights into why enalapril seems to reverse the typical pro-senescent role of Smad1/5/9 in their study?

Thanks for your comment. The downstream regulatory network of BMP-pSmad1/5/9 is highly complex. The BMP-SMAD-ID axis has been mentioned in many studies, and its downstream signaling inhibits the expression of p16 and p21 (PNAS, 2016, 113(46), 13057-13062; Cell, 2003, 115(3), 281-292). Additionally, studies have also found that the Smad1-Stat1-P21 axis inhibits osteoblast senescence (Cell Death Discovery, 2022, 8:254). In our study, enalapril was found to increase the expression of ID1, which is a classic downstream target of pSmad1/5/9 (Cell Stem Cell, 2014, 15(5), 619-633). Therefore, pSmad1/5/9 inhibits cellular senescence markers such as p16, p21 and SASP through ID1, thereby promoting cell proliferation (Fig. 3). Furthermore, we also found that pSmad1/5/9 increases the expression of antioxidant genes and reduces ROS levels, exerting antioxidant effects (Fig. 4). Together, ID1 and antioxidant genes enable pSmad1/5/9 to exert its anti-aging effects.

While the authors showed that enalapril increases pSmad1/5/9 phosphorylation, what are the expression levels of other key and related factors like Smad4, pSmad2, pSmad3, BMP2, and BMP4 in both senescent and non-senescent cells? These data will help clarify the broader signaling effects.

Thanks for your suggestion. We observed an increase in Smad4 expression, while the levels of pSmad2 and pSmad3 remained unchanged after enalapril treatment (Fig. 2A). We will supplement data on the expression changes of these key factors in both senescent and non-senescent cells.

They used BMP receptor inhibitor LDN193189 to pharmacologically inhibit BMP signaling, but it would be more convincing to also include genetic validation (e.g., knockdown or knockout of BMP2 or BMP4). This will help confirm that the observed effects are truly due to BMP-Smad signaling and not off-target effects of the pharmacological inhibitor LDN.

Thanks for your suggestion. We will use shRNA or siRNA to knockdown BMP and examine the related changes to clarify the role of BMP-Smad signaling.

I don't see the results on the changes in senescence markers p16 and p21 in the mouse models treated with enalapril. Similarly, the effects of enalapril treatment on some key SASP factors, such as TNF-α, MCP-1, IL-1β, and IL-1α, are missing, particularly in serum and tissues. These are important data to evaluate the effect of enalapril on senescence.

Thanks for your comment. As for the markers p16 and p21, we observed no change in p16, while the changes in p21 varied across different organs and tissues. (Author response image 1). Nevertheless, behavioral experiments and physiological and biochemical indicators at the individual level consistently demonstrated the significant anti-aging effects of enalapril (Fig. 6).

Author response image 1.

p21(Cdkn1a) expression levels in organs of mice after enalapril feeding.

We also examined the changes in SASP factors in the serum of mice after enalapril treatment. Notably, SASP factors such as CCL (MCP), CXCL and TNFRS11B showed significant decreases (Fig. 5C). The expression changes of SASP factors varied across different organs. In the liver, kidneys and spleen, the expression of IL1a and IL1b decreased, while TNFRS11B expression decreased in both the liver and muscles (Fig. 5B). Additionally, CCL (MCP) levels decreased in all organs (Fig. 5B).

Given that enalapril is primarily known as an antihypertensive, it would be helpful to include data on how it affects blood pressure in the aged mouse models, such as systolic and diastolic blood pressure. This will clarify whether the observed effects are independent of or influenced by changes in blood pressure.

Thanks for your comment. We measured the blood pressure in mice, and found no significant change in blood pressure after enalapril treatment, which has also been validated in other studies (J Gerontol A Biol Sci Med Sci, 2019, 74(8), 1149–1157). Therefore, our results are independent of changes in blood pressure.

Author response:

Reviewer #1 (Public review):

Summary:

Through a series of CRISPR-Cas9 screens, the GPX4 antioxidant pathway was identified as a critical suppressor of cold-induced cell death in hibernator-derived cells. Hamster BHK-21 cells exposed to repeated cold and rewarming cycles revealed five genes (Gpx4, Eefsec, Pstk, Secisbp2, and Sepsecs) as critical components of the GPX4 pathway, which protects against cold-induced ferroptosis. A second screen with continuous cold exposure confirmed the essential role of GPX4 in prolonged cold tolerance. GPX4 knockout lines exhibited complete cell death within four days of cold exposure, and pharmacological inhibition of GPX4 further increased cell death, underscoring the necessity of GPX4's catalytic activity in cold conditions.

An additional CRISPR screen in human cold-sensitive K562 cells identified 176 genes for cold survival. The GPX4 pathway was found to confer significant resistance to cold in hibernators and human cells, with GPX4 loss significantly increasing cold-induced cell death.

Comparing hamster and human GPX4, overexpression of GPX4 in human K562 cells, whether hamster or human GPX4, dramatically improved cold tolerance, while catalytically dead mutants showed no such effect. These findings suggest that GPX4 abundance is a key limiting factor for cold tolerance in human cells, and primary cell types show strong sensitivity to GPX4 loss, highlighting that differences in cold tolerance across species may be due to varying GPX4-mediated protection.

Strengths:

(1) Innovative Approach: The study employs a series of unbiased genome-wide CRISPR-Cas9 screens in both hibernator- and non-hibernator-derived cells to investigate the mechanisms controlling cellular cold tolerance. Notably, this is the first genome-scale CRISPR-Cas9 screen conducted in cells derived from a hibernator, the Syrian hamster.

(2) Identification of the GPX4 Pathway: Identifying glutathione peroxidase 4 (GPX4) as a critical suppressor of cold-induced cell death significantly contributes to the field. Recently, GPX4 was also reported as a potent regulator of cold tolerance through overexpression screening (Sone et al.) in hamsters, which further supports this finding.

(3) Improved Cold Viability Assessment: The study identifies an important technical artifact in using trypan blue to assess cell viability following cold exposure. It reveals that cells stained immediately after cold exposure retain the dye, inaccurately indicating cell death. By introducing a brief rewarming period before viability assessment, the authors significantly improve the accuracy of detecting cold-induced cell death. This refinement in methodology ensures more reliable results and sets a new standard for future research on cold stress in cells.

Weaknesses:

(1) Mechanisms Regulating GPX4 Levels: While the study highlights GPX4 levels as a major determinant of cellular cold tolerance, it does not discuss how these levels are regulated or why they differ between hibernators and non-hibernators. This omission leaves an important aspect of GPX4's role in cold tolerance unexplored.

(2) Generalizability Across Species: Although the study demonstrates the role of GPX4 in several mammalian species, it does not investigate whether this mechanism extends to other vertebrates (e.g., fish and amphibians) that also face cold challenges. This limitation could restrict the broader evolutionary claims made by the study.

(3) Variability in Cold Sensitivity Across Human Cell Lines: The study observes significant variability in cold tolerance among different human cell lines but does not explain these differences clearly. This leaves a key aspect of human cell cold sensitivity insufficiently addressed.

We thank the reviewer for the positive evaluation and thoughtful comments on the manuscript. We acknowledge that our study does not delve into the mechanisms regulating GPX4 levels, including differences between hibernators and non-hibernators, differences between cell types, or the possibility that GPX4 levels are dynamically regulated by environmental conditions. We consider these as interesting open questions that could be addressed in future studies.

While our study focused entirely on mammalian species, we agree that examining cold tolerance mechanisms across a broader range of vertebrates, including fish and amphibians, could enhance our evolutionary perspective. Interestingly, previous work has indicated that C.elegans adapt to cold temperatures through ferritin mediated Fe2+ detoxification. This suggests that cold induces Fe2+-mediated toxicity in C.elegans as well as mammalian cells, but that the mechanisms through which distantly related species counteract cold-mediated cell death may vary. 

Finally, we agree that the variability in cold sensitivity across human cell lines could be further explored, and we will strongly consider conducting follow up experiments to examine the extent to which this variability is driven by levels of GPX4.

We are grateful for these insightful comments, as they highlight important avenues for future research. Addressing these questions will enable a more comprehensive understanding of GPX4's role in cold tolerance and its evolutionary significance across diverse organisms.

Reviewer #2 (Public review):

Summary:

Lam et al., present a very intriguing whole genome CRISPR screen in Syrian Hamster cells as well as K562 cells to identify key genes involved in hypothermia-rewarming tolerance. Survival screens were performed by exposing cells to 4C in a cooled CO2 incubator followed by a rewarming period of 30 minutes prior to survival analysis. In this paradigm, Syrian hamster-derived cell lines exhibit more robust survival than human cell lines (BHK-21 and HaK vs HT1080, HeLa, RPE1, and K562). A genome-wide Syrian hamster CRISPR library was created targeting all annotated genes with 10 guides/gene. LV transduction of the library was performed in BHK-21 cells and the survival screen procedures involved 3 cycles of 4C cold exposure x4 days followed by 2 days of re-warming.

When compared to controls maintained at 37C, 9 genes were required for BHK-21 survival of cold cycling conditions and 5 of these 9 are known components of the GPX4 antioxidant pathway. GPX4 KO BHK-21 cells had reduced cell growth at 37C and profoundly worse cold tolerance which could be reduced by GPX4 expression. GPX4 inhibitors also reduced survival in cold. CRISPR KO screens and GPX4 KO in K562 cells revealed comparable results (though intriguingly glutathione biosynthesis genes were more critical to K562 cells than BHK-21 cells). Human or Syrian hamster GPX4 overexpression improved cold tolerance.

Strengths:

This is a very nicely written paper that clearly communicates in figures and text complicated experimental manipulations and in vitro genetic screening and cell survival data. The focus on GPX4 is interesting and relatively novel. The converging pharmacologic, loss-of-function, and gain-of-function experiments are also a strength.

Weaknesses:

A recently published article (Reference 43, Sone et al.) also independently explored the role of GPX4 in Syrian hamster cold tolerance through gain-of-function screening. Further exploration of the GPX4 species-specific mechanisms would be of great interest, but this is considered a minor weakness given the already very comprehensive and compelling data presented.

We greatly appreciate the reviewer’s compliments and thoughtful comments on our manuscript. We agree with the reviewer that our approach (dual unbiased genome-scale screens in human and hamster cells) and the recent investigation by Sone et al (gain-of-function screening involving the insertion of hamster cDNA into human cells) mutually strengthen the importance of GPX4 in cold tolerance across cell types and species.

Reviewer #3 (Public review):

Summary:

This work aims to address a fundamental biological question: how do mammalian cells achieve/lose tolerance to cold exposure? The authors first tried to establish an experimental system for cell cold exposure and evaluation of cell death and then performed genome-scale CRISPR-Cas9 screening on immortalized cell lines from Syrian Hamster (BHK-21) and human (K562) for key genes that are associated with cell survival during prolonged cold exposure. From these screenings, they focused on glutathione peroxidase 4 (GPX4). Using genetic modifications or pharmacological interventions, and multiple cell models including primary cells from various mammalian species, they showed that GPX4 proteins are likely to retain their activities at 4 {degree sign}C, functioning to prevent cold-induced cell ferroptosis.

Strengths:

(1) This paper is neatly written and hence easy to follow.

(2) Experiments are well designed.

(3) The data showing the overall good cell survival after a prolonged cold exposure or repeated cold-warm cycles are helpful to show the advantages of the experimental instruments and methods the authors used, and hence the validity of their results.

(4) The CRISPR-Cas9 screening is a great attempt.

(5) Multiple cell types from hibernating mammals (cold tolerant) and cold-intolerant species are used to test their findings.

(6) Although some may argue that other labs have published works with different approaches that have pointed out the importance of GPX4 and ferroptosis in hamster cell survival from anoxia-reoxygenation or cold exposure models, hence hurting the novelty of this work, this reviewer thinks that it is highly valuable to have independent research groups and different methods/systems to validate an important concept.

Weaknesses:

(1) Only cell death was robustly surveyed; though cell proliferation was evaluated too in some experiments, other cellular functions, such as mitochondrial ATP production vs. glycolysis, and the extent of lipid peroxidation, could have been measured to reflect cellular physiology.

Validations on complex tissues or in vivo systems would have further strengthened the work and its impact.

CRISPR-Cas9 screening may have technical limitations as knock-out of some essential genes/pathways may lead to cell lethality during screening, and hence the relevance of these genes/pathways to cell cold tolerance may not be noted. From the data presented in this study, this reviewer thinks that the GPX4 pathway is likely a conserved mechanism for long-term cold survival, but not for cold sensitivity or acute cell death from cold exposure. In line with my such speculation, their CRISPR-Cas9 screening revealed genes in the GPX4 pathway from a relatively cold-sensitive human cell line, but the endogenous GPX4 pathway is seemingly operational in this cold-sensitive cell line. Also, these cells are viable after GPX4 knock-out. Dead cells from the acute cold exposure phase may detached, or their genomic DNAs have been severely damaged by the time of sample collection, hence not giving any meaningful sequencing reads. Crippling other factors/pathways such as FOXO1 (PMID: 38570500) or 5-aminolevulinic acid (ALA) metabolism (PMID: 35401816) have been shown to severely aggravate cold-induced cell death, including TUNEL-revealed DNA damage, within a much shorter time scale, whilst loss-function knockouts of FOXO1 or ALA Synthase 1 (ALAS1) are usually cell lethal. Thus, they and other possible essential genes may not be screenable from the current experimental protocol. These important points need to be taken into consideration by the authors.

We thank the reviewer for highlighting the novelty of using genome-scale CRISPR-Cas9 screens and the validation of GPX4 function across cell types and mammalian species. 

We acknowledge that our study primarily focused on measuring cell death using Trypan Blue dye exclusion. To validate the Trypan Blue assay, cell survival data was orthogonally measured using the LDH release assays (Fig. 1g). The proliferation potential of putatively live cells was assessed by counting the increase in live cells following 24 h at 37°C (Fig. 1b). Prompted by your question, we will add additional data to the final version of the manuscript in which we show that following 1 day at 4°C, K562 cells rapidly restarted their cell cycle and double in numbers every 21 hours (Author response image 1). This rate is indistinguishable from the replication rate of cells that were not previously exposed to 4°C, suggesting that the cells following cold exposure are both alive and functionally capable of replicating.

Author response image 1.

Population doubling time of K562 cells cultured at 37°C (pink) and cells that are rewarmed to 37°C following 1 day of 4°C exposure

We agree that assessing additional cellular functions, such as mitochondrial ATP production, glycolysis, lipid metabolism and peroxidation could provide a more comprehensive understanding of cellular physiology under cold stress and would be valuable future studies. Similarly, we appreciate the suggestion to validate our findings in complex tissues or in vivo models. We recognize that such validation could strengthen the implications of our study and enhance its translational potential; however, due to their complexity, we believe that these additional studies are beyond the scope of our current study.

We agree with the reviewer that CRISPR-Cas9 screens have limitations. For example our screen was designed to identify genes that are preferentially required for cellular fitness at 4°C versus 37°C. There are many genes that are required for cellular survival at 4°C as well as 37°C that are not discussed (Table S2, S5). Also, given that the screen is designed to disrupt a single gene per cell, genes that have redundant functions in cold-tolerance will likely be missed. Given the reviewer’s questions, we will expand the discussion of the paper to highlight limitations of the screen.

We apologize for any lack of clarity about the methods we employed during the screen and will expand the methods section to provide further details. For example, for the BHK-21 screen we eliminated dead cells by sequencing cells that reattached after rewarming to 37°C for either 30 minutes (15 day cold exposure screen) or 24 hours (4°C cycling screen). Indeed, at the point of cell collection for both BHK-21 and K562 screens, the fraction of live cells was greater than 92% and 95%, respectively.  We respectfully disagree with the reviewer that our screens would miss genes that affect acute cold tolerance. Any cells that would have died either early or late during cold exposure would have not been sequenced, and thus the sgRNAs targeting a specific gene in those cells would appear depleted, regardless of whether these cells died early/acutely or later during cold exposure. 

We thank the reviewer for pointing out two additionally highly relevant studies. Interestingly, the genes implicated in cold tolerance in these studies, FOXO1 and ALAS1, did not appear essential for survival at 37°C or 4°C  in BHK-21 or K562 cells. There are several possibilities that could explain this finding: 1) our screen may not have successfully knocked out these genes, 2) other proteins may have compensated for their loss, or 3) these pathways may regulate cold tolerance in some but not all cell types. We apologize that in the current version of the manuscript we did not reflect on these recent studies. We will expand our discussion to include their findings. 

Once again, we are grateful for the reviewer’s insights, which have highlighted key areas for further exploration as well as pointed to specific ways to improve our manuscript.

Author Response

Joint Public Review

The molecular composition of synaptic vesicles (SVs) has been defined in substantial detail, but the function of many SV-resident proteins are still unknown. The present study focused on one such protein, the 'orphan' SV-resident transporter SLC6A17. By utilizing sophisticated and extensive mouse genetics and behavioral experiments, the authors provide convincing support for the notion that certain SLC6A17 variants cause intellectual disability (ID) in humans carrying such genetic variations. This is an important and novel finding. Furthermore, the authors propose, based on LCMS analyses of isolated SVs, that SLC6A17 is responsible for glutamine (Gln) transport into SVs, leading to the provocative idea that Gln functions as a neurotransmitter and that deficits in Gln transport into SVs by SLC6A17 represents a key pathogenetic mechanism in human ID patients carrying variants of the SLC6A17 gene.

This latter aspect of the present paper is not adequately supported by the experimental evidence so that the main conceptual claims of the study appear insufficiently justified at this juncture. Key weaknesses are as follows:

A) Detection of Gln, along with classical neurotransmitters such as glutamate, GABA, or ACh, in isolated SV fractions does not prove that Gln is transported into SVs by active transport. Gln is quite abundant in extracellular compartments. Its appearance in SV samples can therefore also be explained by trapping in SVs during endocytosis, presence in other - contaminating - organelles, binding to membrane surfaces, and other processes. Direct assays of Gln uptake into SVs, which have the potential to stringently test key postulates of the authors, are lacking.

We have conducted multiple control experiments to exclude the possibility of contamination.

1). Western blot analysis of SLC6A17-HA immunoisolation (Figure 4D and Figure 4—figure supplement 1) has shown that this faction contained little other organelles and membranes. These results are strong argument that contaminations in our isolated fraction were in very low level.

2). We then examined the proportion of SLC6A17 localized SVs through quantifying the co-localization of Syp and SLC6A17 by anti-Syp immunoisolation in Slc6a17-2A-HA-iCre mice. We found that SLC6A17 is predominately localized on SVs (with 98.7% compared with classical SV marker, Author response image 1A). This further showed that immunoisolated SLC6A17 fraction was mainly composed of SVs.

3). We also analyzed other SV marker proteins such as Syt1 and Syb2 for IP-LC-MS, all results supported Gln enrichment (Author response image 1B).

4). Importantly, immunoisolation of the SLC6A17P633R-HA protein, which caused SLC6A17 mislocalization away from the SVs (Figure 3B and Figure 3—figure supplement 1C, D), showed no Gln enrichment (Author response image 1C).

5). Moreover, immunoisolation of AAV-PHP.eb overexpressed cytoplasmic membrane Gln transporter SLC38A1-HA did not show Gln enrichment (Author response image 1D).

6). We also tested whether trafficking organelles such as the lysosome could enrich Gln. As is shown in Author response image 1E, immunoisolation of AAV-PHP.eb overexpressed TMEM192-HA did not show Gln enrichment. For active transport, we tested the effects of proton dissipator FCCP, v-ATPase inhibitor NEM and ΔpH dissipator nigercin. As is shown in Author response image 1F, 1G, Gln level was reduced by these inhibitors, supporting active transport of Gln.

Author response image 1.

Control experiments to test for contamination. A. Anti-Syp immunoisolation in Slc6a17-2A-HA-iCre mice. B. Quantification of Gln level in anti-Syt1 and anti-Syb2 immunoisolated fraction. C. Anti-HA immunoisolation in SLC6A7-2A-HA and anti-Slc6a17P633R mice. D. Anti-HA immunoisolation in AAV-PHP.eb-hSyn-SLC38A1-HA overexperssion mice. E. Anti-HA immunoisolation in AAV-PHP.eb-hSyn-TMEM192-HA overexperssion mice. F. Anti-HA immunoisolation in SLC6A7-2A-HA mice under FCCP (50 μM) and NEM (200 μM). G. Anti-Syp immunoisolation in wild type mice under FCCP (50 μM) and Nigercin (20 μM).

B) The authors generated multiple potentially very useful genetic tools and models. However, the validation of these models is incomplete. Most importantly, it remains unclear whether the different mutations affect SLC6A17 expression levels, subcellular localization, or the expression and trafficking of other SV and synapse components.

The verification of transgenic mouse line is described in the Material and Methods section of our manuscript. There are numerous literatures published for CRISPR mediated gene editing in animals and the off-target effect of CRISPR-Cas9 system is widely studied with optimized design tools developed by many groups (Platt et al., 2014; Chu et al., 2015, 2016; Liu et al., 2017; Gemberling et al., 2021; Singh et al., 2022). The gRNAs used for animal generation were chosen carefully based on publically available tools. Apart from basic genomic PCR sequencing of target regions of all gene edited mouse models, Southern blots were performed by Biocytogen company for Slc6a17-HA-2A-iCre and Slc6a17P633R mice to rule out random insertions. Expression levels in Slc6a17-KO and Slc6a17P633R mice were not affected, as shown in Figure R2. HA-tagged protein in Slc6a17-HA-2A-iCre and Slc6a17P633R mice were detected by immunoisolation, immunofluorescence, and fractionation (Figure 3, 4, Figure 3—figure supplement 1, Figure 4—figure supplement 1). Both showed localizations expected from previous reports ().

C) Apart from the caveats mentioned above regarding Gln uptake into SVs, the data interpretation provided by the authors lacks stringency with respect to the biophysics of plasma membrane and SV transporters.

The biophysics of SLC6A17 was carefully studied (Para et al 2008; Zaia and Reimer, 2009). Our work focused on in vivo biochemical results, not biophysics.

Author response image 2.

Verification of genetic mouse models. A. q-PCR verification of Slc6a17-KO mice; B. q-PCR verification of Slc6a17P633R mice; C. Example of genomic primer design for Slc6a17-HA-2A-iCre mice founder mice screen; D. Example of genomic PCR for Slc6a17-HA-2A-iCre mice founder mice screen; E. Southern blot performed for Slc6a17-HA-2A-iCre mice.

Reference

Chu, Van Trung et al. “Increasing the efficiency of homology-directed repair for CRISPR-Cas9-induced precise gene editing in mammalian cells.” Nature biotechnology vol. 33,5 (2015): 543-8. doi:10.1038/nbt.3198

Chu, Van Trung, et al. "Efficient generation of Rosa26 knock-in mice using CRISPR/Cas9 in C57BL/6 zygotes." BMC biotechnology 16.1 (2016): 1-15.

Gemberling, Matthew P et al. “Transgenic mice for in vivo epigenome editing with CRISPR-based systems.” Nature methods vol. 18,8 (2021): 965-974. doi:10.1038/s41592-021-01207-2

Liu, Edison T., et al. "Of mice and CRISPR: The post‐CRISPR future of the mouse as a model system for the human condition." EMBO reports 18.2 (2017): 187-193.

Madisen, Linda, et al. "A robust and high-throughput Cre reporting and characterization system for the whole mouse brain." Nature neuroscience 13.1 (2010): 133-140.

Parra, Leonardo A., et al. "The orphan transporter Rxt1/NTT4 (SLC6A17) functions as a synaptic vesicle amino acid transporter selective for proline, glycine, leucine, and alanine." Molecular pharmacology 74.6 (2008): 15211532.

Platt, R.J., Chen, S., Zhou, Y., Yim, M.J., Swiech, L., Kempton, H.R., Dahlman, J.E., Parnas, O., Eisenhaure, T.M., Jovanovic, M., et al. (2014). CRISPR-Cas9 knockin mice for genome editing and cancer mode Yang, Hui, Haoyi Wang, and Rudolf Jaenisch. "Generating genetically modified mice using CRISPR/Cas-mediated genome engineering." Nature protocols 9.8 (2014): 1956-1968.ling. Cell 159, 440-455.

Singh, Surender et al. “Opportunities and challenges with CRISPR-Cas mediated homologous recombination based precise editing in plants and animals.” Plant molecular biology, 10.1007/s11103-022-01321-5. 31 Oct. 2022, doi:10.1007/s11103-022-01321-5

Zaia, K.A., and Reimer, R.J. (2009). Synaptic vesicle protein NTT4/XT1 (SLC6A17) catalyzes Na+-coupled neutral amino acid transport. J Biol Chem 284, 8439-8448.

Author response:

We would like to thank the editors and the reviewers for constructive feedback on our first version of the manuscript. Before submitting a fully revised version with detailed response to each point, we would like to provide a brief clarification on some of the key issues.

Reviewer 2 raised a concern about the precision and specificity of holographic stimulation, regarding its potential effect on out-of-focus stimulation points and planes. We further verified whether the laser power at the targeted z-plane influences cells’ activity at nearby z-planes. As the Reviewer pointed out, the previous x- and y-axis shifts were tested by single-cell stimulation. This time, we stimulated five cells simultaneously, to match the actual experiment setup and assess potential artifacts in other planes. We observed no stimulation-driven activity increase in cells at a z-planed shifted by 20 µm (Author response image 1). This confirms the holographic stimulation accurately manipulates the pre-selected target cells and the effects we observe is not likely due to out-of-focus stimulation artifacts. It is true that not all of pre-selected cells showing significant response changes prior to the main experiment are effectively activated t every trial during the experiments. While further analyses will be included in the revised manuscript, we varied the target cell distances across FOVs, from nearby cells to those farther apart within the FOV. We have not observed a significant relationship between the target cell distances and stimulation effect. Lastly, cells within < 15 µm of the target were excluded to prevent potential excitation due to the holographic stimulation power. Given the spontaneous movements of the FOV during imaging sessions due to animal’s movement, despite our efforts to minimize them, we believe that any excitation from these neighboring neurons would be directly from the stimulation rather than the light pattern artifact itself.

Author response image 1.

Stimulation effect on five pre-selected cells at the target z-plane (left) and 20 µm off-target z-plane (right). No stimulation-driven effect was observed on the off-target cells.

Reviewers also raised concerns regarding the interpretation of homeostatic balance. While we are working on further analyses to strengthen our findings based on the reviewers’ suggestions, the observed response changes in co-tuned neuronal ensembles, specifically during the processing of their preferred frequency information, highlights an interaction between sensory processing and network dynamics. We believe this specificity indicates a functional mechanism beyond broad suppression or simple inhibitory effects, possibly aligning with homeostatic principles in cortical circuits. Regarding the post-stimulation effect, it is true neither the stimulation nor the control condition showed further response changes during the post-stimulation session. For the control condition, this is likely due to the repetitive tone presentation that could already triggered neural adaptation to a plateau by first two imaging sessions (baseline and stimulation sessions), preventing further changes in the last session. However, as the stimulation condition induced a greater amplitude decrease during the stimulation session compared to the control condition, if this extra suppression had not persisted during the post-stimulation session, we would have expected response amplitudes to rebound, increasing between the stimulation and post-stimulation sessions, which was not the case. Therefore, we propose that the persistence of this rebalanced network state is more indicative of a potential homeostatic mechanism in response to the activity manipulation within the network.

Author response:

Point-by-point description of the revisions

Reviewer #1 (Evidence, reproducibility, and clarity):

The work by Pinon et al describes the generation of a microvascular model to study Neisseria meningitidis interactions with blood vessels. The model uses a novel and relatively high throughput fabrication method that allows full control over the geometry of the vessels. The model is well characterized. The authors then study different aspects of Neisseriaendothelial interactions and benchmark the bacterial infection model against the best disease model available, a human skin xenograft mouse model, which is one of the great strengths of the paper. The authors show that Neisseria binds to the 3D model in a similar geometry that in the animal xenograft model, induces an increase in permeability short after bacterial perfusion, and induces endothelial cytoskeleton rearrangements. Finally, the authors show neutrophil recruitment to bacterial microcolonies and phagocytosis of Neisseria. The article is overall well written, and it is a great advancement in the bioengineering and sepsis infection field, and I only have a few major comments and some minor.

Major comments:

Infection-on-chip. I would recommend the authors to change the terminology of "infection on chip" to better reflect their work. The term is vague and it decreases novelty, as there are multiple infection on chips models that recapitulate other infections (recently reviewed in https://doi.org/10.1038/s41564-024-01645-6) including Ebola, SARS-CoV-2, Plasmodium and Candida. Maybe the term "sepsis on chip" would be more specific and exemplify better the work and novelty. Also, I would suggest that the authors carefully take a look at the text and consider when they use VoC or to current term IoC, as of now sometimes they are used interchangeably, with VoC being used occasionally in bacteria perfused experiments.

We thank Reviewer #1 for this suggestion. Indeed, we have chosen to replace the term "Infection-on-Chip" by "infected Vessel-on-chip" to avoid any confusion in the title and the text. Also, we have removed all the terms "IoC" which referred to "Infection-on-Chip" and replaced with "VoC" for "Vessel-on-Chip". We think these terms will improve the clarity of the main text.

Author response image 1.

F-actin (red) and ezrin (yellow) staining after 3h of infection with N. meningitidis (green) in 2D (top) and 3D (bottom) vessel-on-chip models.

Fig 3 and Supplementary 3: Permeability. The authors suggest that early 3h infection with Neisseria do not show increase in vascular permeability in the animal model, contrary to their findings in the 3D in vitro model. However, they show a non-significant increase in permeability of 70 KDa Dextran in the animal xenograft early infection. This seems to point that if the experiment would have been done with a lower molecular weight tracer, significant increases in permeability could have been detected. I would suggest to do this experiment that could capture early events in vascular disruption.

Comparing permeability under healthy and infected conditions using Dextran smaller than 70 kDa is challenging. Previous research (1) has shown that molecules below 70 kDa already diffuse freely in healthy tissue. Given this high baseline diffusion, we believe that no significant difference would be observed before and after N. meningitidis infection and these experiments were not carried out. As discussed in the manuscript, bacteria induced permeability in mouse occurs at later time points, 16h post infection as shown previoulsy (2). As discussed in the manuscript, this difference between the xenograft model and the chip likely reflect the absence in the chip of various cell types present in the tissue parenchyma.

The authors show the formation of actin of a honeycomb structure beneath the bacterial microcolonies. This only occurred in 65% of the microcolonies. Is this result similar to in vitro 2D endothelial cultures in static and under flow? Also, the group has shown in the past positive staining of other cytoskeletal proteins, such as ezrin in the ERM complex. Does this also occur in the 3D system?

We thank the Reviewer #1 for this suggestion.

• According to this recommendation, we imaged monolayers of endothelial cells in the flat regions of the chip (the two lateral channels) using the same microscopy conditions (i.e., Obj. 40X N.A. 1.05) that have been used to detect honeycomb structures in the 3D vessels in vitro. We showed that more than 56% of infected cells present these honeycomb structures in 2D, which is 13% less than in 3D, and is not significant due to the distributions of both populations. Thus, we conclude that under both in vitro conditions, 2D and 3D, the amount of infected cells exhibiting cortical plaques is similar. We have added the graph and the confocal images in Figure S4B and lines 418-419 of the revised manuscript.

• We recently performed staining of ezrin in the chip and imaged both the 3D and 2D regions. Although ezrin staining was visible in 3D (Fig. 1 of this response), it was not as obvious as other markers under these infected conditions and we did not include it in the main text. Interpretation of this result is not straight forward as for instance the substrate of the cells is different and it would require further studies on the behaviour of ERM proteins in these different contexts.

One of the most novel things of the manuscript is the use of a relatively quick photoablation system. I would suggest that the authors add a more extensive description of the protocol in methods. Could this technique be applied in other laboratories? If this is a major limitation, it should be listed in the discussion.

Following the Reviewer’s comment, we introduced more detailed explanations regarding the photoablation:

• L157-163 (Results): "Briefly, the chosen design is digitalized into a list of positions to ablate. A pulsed UV-LASER beam is injected into the microscope and shaped to cover the back aperture of the objective. The laser is then focused on each position that needs ablation. After introducing endothelial cells (HUVEC) in the carved regions,…"

• L512-516 (Discussion): "The speed capabilities drastically improve with the pulsing repetition rate. Given that our laser source emits pulses at 10kHz, as compared to other photoablation lasers with repetitions around 100 Hz, our solution could potentially gain a factor of 100."

• L1082-1087 (Materials and Methods): "…, and imported in a python code. The control of the various elements is embedded and checked for this specific set of hardware. The code is available upon request." Adding these three paragraphs gives more details on how photoablation works thus improving the manuscript.

Minor comments:

Supplementary Fig 2. The reference to subpanels H and I is swapped.

The references to subpanels H and I have been correctly swapped back in the reviewed version.

Line 203: I would suggest to delete this sentence. Although a strength of the submitted paper is the direct comparison of the VoC model with the animal model to better replicate Neisseria infection, a direct comparison with animal permeability is not needed in all vascular engineering papers, as vascular permeability measurements in animals have been well established in the past.

The sentence "While previously developed VoC platforms aimed at replicating physiological permeability properties, they often lack direct comparisons with in vivo values." has been removed from the revised text.

Fig 3: Bacteria binding experiments. I would suggest the addition of more methodological information in the main results text to guarantee a good interpretation of the experiment. First, it would be better that wall shear stress rather than flow rate is described in the main text, as flow rate is dependent on the geometry of the vessel being used. Second, how long was the perfusion of Neisseria in the binding experiment performed to quantify colony doubling or elongation? As per figure 1C, I would guess than 100 min, but it would be better if this information is directly given to the readers.

We thank Reviewer #1 for these two suggestions that will improve the text clarity (e.g., L316). (i) Indeed, we have changed the flow rate in terms of shear stress. (ii) Also, we have normalized the quantification of the colony doubling time according to the first time-point where a single bacteria is attached to the vessel wall. Thus, early adhesion bacteria will be defined by a longer curve while late adhesion bacteria by a shorter curve. In total, the experiment lasted for 3 hours (modifications appear in L318 and L321-326).

Fig 4: The honeycomb structure is not visible in the 3D rendering of panel D. I would recommend to show the actin staining in the absence of Neisseria staining as well.

According to this suggestion, a zoom of the 3D rendering of the cortical plaque without colony had been added to the figure 4 of the revised manuscript.

Line 421: E-selectin is referred as CD62E in this sentence. I would suggest to use the same terminology everywhere.

We have replaced the "CD62E" term with "E-selectin" to improve clarity.

Line 508: "This difference is most likely associated with the presence of other cell types in the in vivo tissues and the onset of intravascular coagulation". Do the authors refer to the presence of perivascular cells, pericytes or fibroblasts? If so, it could be good to mention them, as well as those future iterations of the model could include the presence of these cell types.

By "other cell types", we refer to pericytes (3), fibroblasts (4), and perivascular macrophages (5), which surround endothelial cells and contribute to vessel stability. The main text was modified to include this information (Lines 548 and 555-570) and their potential roles during infection disussed.

Discussion: The discussion covers very well the advantages of the model over in vitro 2D endothelial models and the animal xenograft but fails to include limitations. This would include the choice of HUVEC cells, an umbilical vein cell line to study microcirculation, the lack of perivascular cells or limitations on the fabrication technique regarding application in other labs (if any).

We thank Reviewer #1 for this suggestion. Indeed, our manuscript may lack explaining limitations, and adding them to the text will help improve it:

• The perspectives of our model include introducing perivascular cells surrounding the vessel and fibroblasts into the collagen gel as discussed previously and added in the discussion part (L555-570).

• Our choice for HUVEC cells focused on recapitulating the characteristics of venules that respect key features such as the overexpression of CD62E and adhesion of neutrophils during inflammation. Using microvascular endothelial cells originating from different tissues would be very interesting. This possibility is now mentioned in the discussion lines 567-568.

• Photoablation is a homemade fabrication technique that can be implemented in any lab harboring an epifluorescence microscope. This method has been more detailed in the revised manuscript (L1085-1087).

Line 576: The authors state that the model could be applied to other systemic infections but failed to mention that some infections have already been modelled in 3D bioengineered vascular models (examples found in https://doi.org/10.1038/s41564-024-01645-6). This includes a capillary photoablated vascular model to study malaria (DOI: 10.1126/sciadv.aay724).

Thes two important references have been introduced in the main text (L84, 647, 648).

Line 1213: Are the 6M neutrophil solution in 10ul under flow. Also, I would suggest to rewrite this sentence in the following line "After, the flow has been then added to the system at 0.7-1 µl/min."

We now specified that neutrophils are circulated in the chip under flow conditions, lines 1321-1322.

Significance

The manuscript is comprehensive, complete and represents the first bioengineered model of sepsis. One of the major strengths is the carful characterization and benchmarking against the animal xenograft model. Its main limitations is the brief description of the photoablation methodology and more clarity is needed in the description of bacteria perfusion experiments, given their complexity. The manuscript will be of interest for the general infection community and to the tissue engineering community if more details on fabrication methods are included. My expertise is on infection bioengineered models.

Reviewer #2 (Evidence, reproducibility, and clarity):

Summary:

The authors develop a Vessel-on-Chip model, which has geometrical and physical properties similar to the murine vessels used in the study of systemic infections. The vessel was created via highly controllable laser photoablation in a collagen matrix, subsequent seeding of human endothelial cells and flow perfusion to induce mechanical cues. This vessel could be infected with Neisseria meningitidis, as a model of systemic infection. In this model, microcolony formation and dynamics, and effects on the host were very similar to those described for the human skin xenograft mouse, which is the current gold standard for these studies, and were consistent with observations made in patients. The model could also recapitulate the neutrophil response upon N. meningitidis systemic infection.

Major comments:

I have no major comments. The claims and the conclusions are supported by the data, the methods are properly presented and the data is analyzed adequately. Furthermore, I would like to propose an optional experiment could improve the manuscript. In the discussion it is stated that the vascular geometry might contribute to bacterial colonization in areas of lower velocity. It would be interesting to recapitulate this experimentally. It is of course optional but it would be of great interest, since this is something that can only be proven in the organ-on-chip (where flow speed can be tuned) and not as much in animal models. Besides, it would increase impact, demonstrating the superiority of the chip in this area rather than proving to be equal to current models.

We have conducted additional experiments on infection in different vascular geometries now added these results figure 3/S3 and lines 288-305. We compared sheared stress levels as determined by Comsol simulation and experimentally determined bacterial adhesion sites. In the conditions used, the range of shear generated by the tested geometries do not appear to change the efficiency of bacterial adhesion. These results are consistent with a previous study from our group which show that in this range of shear stresses the effect on adhesion is limited (6) . Furthermore, qualitative observations in the animal model indicate that bacteria do not have an obvious preference in terms of binding site.

Minor comments:

I have a series of suggestions which, in my opinion, would improve the discussion. They are further elaborated in the following section, in the context of the limitations.

• How to recapitulate the vessels in the context of a specific organ or tissue? If the pathogen is often found in the luminal space of other organs after disseminating from the blood, how can this process be recapitulated with this mode, if at all?

For reasons that are not fully understood, postmortem histological studies reveal bacteria only inside blood vessels but rarely if ever in the organ parenchyma. The presence of intravascular bacteria could nevertheless alter cells in the tissue parenchyma. The notable exception is the brain where bacteria exit the bacterial lumen to access the cerebrospinal fluid. The chip we describe is fully adapted to develop a blood brain barrier model and more specific organ environments. This implies the addition of more cell types in the hydrogel. A paragraph on this topic has been added (Lines 548 and 552-570).

• Similarly, could other immune responses related to systemic infection be recapitulated? The authors could discuss the potential of including other immune cells that might be found in the interstitial space, for example.

This important discussion point has been added to the manuscript (L623-636). As suggested by Reviewer #2, other immune cells respond to N. meningitis and can be explored using our model. For instance, macrophages and dendritic cells are activated upon N. meningitis infection, eliminate the bacteria through phagocytosis, produce pro-inflammatory cytokines and chemokines potentially activating lymphocytes (7). Such an immune response, yet complex, would be interesting to study in our model as skin-xenograft mice are deprived of B and T lymphocytes to ensure acceptance of human skin grafts.

• A minor correction: in line 467 it should probably be "aspects" instead of "aspect", and the authors could consider rephrasing that sentence slightly for increased clarity.

We have corrected the sentence with "we demonstrated that our VoC strongly replicates key aspects of the in vivo human skin xenograft mouse model, the gold standard for studying meningococcal disease under physiological conditions." in lines 499-503.

Strengths and limitations

The most important strength of this manuscript is the technology they developed to build this model, which is impressive and very innovative. The Vessel-on-Chip can be tuned to acquire complex shapes and, according to the authors, the process has been optimized to produce models very quickly. This is a great advancement compared with the technologies used to produce other equivalent models. This model proves to be equivalent to the most advanced model used to date, but allows to perform microscopy with higher resolution and ease, which can in turn allow more complex and precise image-based analysis. However, the authors do not seem to present any new mechanistic insights obtained using this model. All the findings obtained in the infection-on-chip demonstrate that the model is equivalent to the human skin xenograft mouse model, and can offer superior resolution for microscopy. However, the advantages of the model do not seem to be exploited to obtain more insights on the pathogenicity mechanisms of N. meningitidis, host-pathogen interactions or potential applications in the discovery of potential treatments. For example, experiments to elucidate the role of certain N. meningiditis genes on infection could enrich the manuscript and prove the superiority of the model. However, I understand these experiments are time-consuming and out of the scope of the current manuscript. In addition, the model lacks the multicellularity that characterizes other similar models. The authors mention that the pathogen can be found in the luminal space of several organs, however, this luminal space has not been recapitulated in the model. Even though this would be a new project, it would be interesting that the authors hypothesize about the possibilities of combining this model with other organ models. The inclusion of circulating neutrophils is a great asset; however it would also be interesting to hypothesize about how to recapitulate other immune responses related to systemic infection.

We thank Reviewer #2 for his/her comment on the strengths and limitations of our work. The difficulty is that our study opens many futur research directions and applications and we hope that the work serves as the basis for many future studies but one can only address a limited set of experiments in a single manuscript.

• Experiments investigating the role of N. meningitidis genes require significant optimization of the system. Multiplexing is a potential avenue for future development, which would allow the testing of many mutants. The fast photoablation approach is particularly amenable to such adaptation.

• Cells and bacteria inside the chambers could be isolated and analyzed at the transcriptomic level or by flow cytometry. This would imply optimizing a protocol for collecting cells from the device via collagenase digestion, for instance. This type of approach would also benefit from multiplexing to enhance the number of cells.

• As mentioned above, the revised manuscript discusses the multicellular capabilities of our model, including the integration of additional immune cells and potential connections to other organ systems. We believe that these approaches are feasible and valuable for studying various aspects of N. meningitidis infection.

Advance

The most important advance of this manuscript is technical: the development of a model that proves to be equivalent to the most complex model used to date to study meningococcal systemic infections. The human skin xenograft mouse model requires complex surgical techniques and has the practical and ethical limitations associated with the use of animals. However, the Infection-on-chip model is completely in vitro, can be produced quickly, and allows to precisely tune the vessel’s geometry and to perform higher resolution microscopy. Both models were comparable in terms of the hallmarks defining the disease, suggesting that the presented model can be an effective replacement of the animal use in this area.

Other vessel-on-chip models can recapitulate an endothelial barrier in a tube-like morphology, but do not recapitulate other complex geometries, that are more physiologically relevant and could impact infection (in addition to other non-infectious diseases). However, in the manuscript it is not clear whether the different morphologies are necessary to study or recapitulate N. meningitidis infection, or if the tubular morphologies achieved in other similar models would suffice.

Audience

This manuscript might be of interest for a specialized audience focusing on the development of microphysiological models. The technology presented here can be of great interest to researchers whose main area of interest is the endothelium and the blood vessels, for example, researchers on the study of systemic infections, atherosclerosis, angiogenesis, etc. Thus, the tool presented (vessel-on-chip) can have great applications for a broad audience. However, even when the method might be faster and easier to use than other equivalent methods, it could still be difficult to implement in another laboratory, especially if it lacks expertise in bioengineering. Therefore, the method could be more of interest for laboratories with expertise in bioengineering looking to expand or optimize their toolbox. Alternatively, this paper present itself as an opportunity to begin collaborations, since the model could be used to test other pathogen or conditions.

Field of expertise:

Infection biology, organ-on-chip, fungal pathogens.

I lack the expertise to evaluate the image-based analysis.

References

(1) Gyohei Egawa, Satoshi Nakamizo, Yohei Natsuaki, Hiromi Doi, Yoshiki Miyachi, and Kenji Kabashima. Intravital analysis of vascular permeability in mice using two-photon microscopy. Scientific Reports, 3(1):1932, Jun 2013. ISSN 2045-2322. doi: 10.1038/srep01932.

(2) Valeria Manriquez, Pierre Nivoit, Tomas Urbina, Hebert Echenique-Rivera, Keira Melican, Marie-Paule Fernandez-Gerlinger, Patricia Flamant, Taliah Schmitt, Patrick Bruneval, Dorian Obino, and Guillaume Duménil. Colonization of dermal arterioles by neisseria meningitidis provides a safe haven from neutrophils. Nature Communications, 12(1):4547, Jul 2021. ISSN 2041-1723. doi: 10.1038/s41467-021-24797-z.

(3) Mats Hellström, Holger Gerhardt, Mattias Kalén, Xuri Li, Ulf Eriksson, Hartwig Wolburg, and Christer Betsholtz. Lack of pericytes leads to endothelial hyperplasia and abnormal vascular morphogenesis. Journal of Cell Biology, 153(3):543–554, Apr 2001. ISSN 0021-9525. doi: 10.1083/jcb.153.3.543.

(4) Arsheen M. Rajan, Roger C. Ma, Katrinka M. Kocha, Dan J. Zhang, and Peng Huang. Dual function of perivascular fibroblasts in vascular stabilization in zebrafish. PLOS Genetics, 16(10):1–31, 10 2020. doi: 10.1371/journal.pgen.1008800.

(5) Huanhuan He, Julia J. Mack, Esra Güç, Carmen M. Warren, Mario Leonardo Squadrito, Witold W. Kilarski, Caroline Baer, Ryan D. Freshman, Austin I. McDonald, Safiyyah Ziyad, Melody A. Swartz, Michele De Palma, and M. Luisa Iruela-Arispe. Perivascular macrophages limit permeability. Arteriosclerosis, Thrombosis, and Vascular Biology, 36(11):2203–2212, 2016. doi: 10.1161/ATVBAHA. 116.307592.

(6) Emilie Mairey, Auguste Genovesio, Emmanuel Donnadieu, Christine Bernard, Francis Jaubert, Elisabeth Pinard, Jacques Seylaz, Jean-Christophe Olivo-Marin, Xavier Nassif, and Guillaume Dumenil. Cerebral microcirculation shear stress levels determine Neisseria meningitidis attachment sites along the blood–brain barrier . Journal of Experimental Medicine, 203(8):1939–1950, 07 2006. ISSN 0022-1007. doi: 10.1084/jem.20060482.

(7) Riya Joshi and Sunil D. Saroj. Survival and evasion of neisseria meningitidis from macrophages. Medicine in Microecology, 17:100087, 2023. ISSN 2590-0978. doi: https://doi.org/10.1016/j.medmic. 2023.100087.

Author Response:

Assessment note: “Whereas the results and interpretations are generally solid, the mechanistic aspect of the work and conclusions put forth rely heavily on in vitro studies performed in cultured L6 myocytes, which are highly glycolytic and generally not viewed as a good model for studying muscle metabolism and insulin action.”

While we acknowledge that in vitro models may not fully recapitulate the complexity of in vivo systems, we believe that our use of L6 myotubes is appropriate for studying the mechanisms underlying muscle metabolism and insulin action. As mentioned below (reviewer 2, point 1), L6 myotubes possess many important characteristics relevant to our research, including high insulin sensitivity and a similar mitochondrial respiration sensitivity to primary muscle fibres. Furthermore, several studies have demonstrated the utility of L6 myotubes as a model for studying insulin sensitivity and metabolism, including our own previous work (PMID: 19805130, 31693893, 19915010).

In addition, we have provided evidence of the similarities between L6 cells overexpressing SMPD5 and human muscle biopsies at protein levels and the reproducibility of the negative correlation between ceramide and Coenzyme Q observed in L6 cells in vivo, specifically in the skeletal muscle of mice in chow diet. These findings support the relevance of our in vitro results to in vivo muscle metabolism.

Finally, we will supplement our findings by demonstrating a comparable relationship between ceramide and Coenzyme Q in mice exposed to a high-fat diet, to be shown in Supplementary Figure 4 H-I. Further animal experiments will be performed to validate our cell-line based conclusions. We hope that these additional results address the concerns raised by the reviewer and further support the relevance of our in vitro findings to in vivo muscle metabolism and insulin action.

Points from reviewer 1:

1. Although the authors' results suggest that higher mitochondrial ceramide levels suppress cellular insulin sensitivity, they rely solely on a partial inhibition (i.e., 30%) of insulin-stimulated GLUT4-HA translocation in L6 myocytes. It would be critical to examine how much the increased mitochondrial ceramide would inhibit insulin-induced glucose uptake in myocytes using radiolabel deoxy-glucose.

Response: The primary impact of insulin is to facilitate the translocation of glucose transporter type 4 (GLUT4) to the cell surface, which effectively enhances the maximum rate of glucose uptake into cells. Therefore, assessing the quantity of GLUT4 present at the cell surface in non-permeabilized cells is widely regarded as the most reliable measure of insulin sensitivity (PMID: 36283703, 35594055, 34285405). Additionally, plasma membrane GLUT4 and glucose uptake are highly correlated. Whilst we have routinely measured glucose uptake with radiolabelled glucose in the past, we do not believe that evaluating glucose uptake provides a better assessment of insulin sensitivity than GLUT4.

We will clarify the use of GLUT4 translocation in the Results section:

“...For this reason, several in vitro models have been employed involving incubation of insulin sensitive cell types with lipids such as palmitate to mimic lipotoxicity in vivo. In this study we will use cell surface GLUT4-HA abundance as the main readout of insulin response...”

1. Another important question to be addressed is whether glycogen synthesis is affected in myocytes under these experimental conditions. Results demonstrating reductions in insulin-stimulated glucose transport and glycogen synthesis in myocytes with dysfunctional mitochondria due to ceramide accumulation would further support the authors' claim.

Response: We have carried out supplementary experiments to investigate glycogen synthesis in our insulin-resistant models. Our approach involved L6-myotubes overexpressing the mitochondrial-targeted construct ASAH1 (as described in Fig. 3). We then challenged them with palmitate and measured glycogen synthesis using 14C radiolabeled glucose. Our observations indicated that palmitate suppressed insulin-induced glycogen synthesis, which was effectively prevented by the overexpression of ASAH1 (N = 5, * p<0.05). These results provide additional evidence highlighting the role of dysfunctional mitochondria in muscle cell glucose metabolism.

These data will be added to Supplementary Figure 4K and the results modified as follows:

“Notably, mtASAH1 overexpression protected cells from palmitate-induced insulin resistance without affecting basal insulin sensitivity (Fig. 3E). Similar results were observed using insulin-induced glycogen synthesis as an ortholog technique for Glut4 translocation. These results provide additional evidence highlighting the role of dysfunctional mitochondria in muscle cell glucose metabolism (Sup. Fig. 5K). Importantly, mtASAH1 overexpression did not rescue insulin sensitivity in cells depleted…”

We will add to the method section:

“L6 myotubes overexpressing ASAH were grown and differentiated in 12-well plates, as described in the Cell lines section, and stimulated for 16 h with palmitate-BSA or EtOH-BSA, as detailed in the Induction of insulin resistance section.

On day seven of differentiation, myotubes were serum starved in plain DMEM for 3 and a half hours. After incubation for 1 hour at 37C with 2 µCi/ml D-[U-14C]-glucose in the presence or absence of 100 nM insulin, glycogen synthesis assay was performed, as previously described (Zarini S. et al., J Lipid Res, 63(10): 100270, 2022).”

1. In addition, it would be critical to assess whether the increased mitochondrial ceramide and consequent lowering of energy levels affect all exocytic pathways in L6 myoblasts or just the GLUT4 trafficking. Is the secretory pathway also disrupted under these conditions?

Response: As the secretory pathway primarily involves the synthesis and transportation of soluble proteins that are secreted into the extracellular space, and given that the majority of cellular transmembrane proteins (excluding those of the mitochondria) use this pathway to arrive at their ultimate destination, we believe that the question posed by the reviewer is highly challenging and beyond the scope of our research. We will add this to the discussion:

“...the abundance of mPTP associated proteins suggesting a role of this pore in ceramide induced insulin resistance (Sup. Fig. 6E). In addition, it is yet to be determined whether the trafficking defect is specific to Glut4 or if it affects the exocytic-secretory pathway more broadly…”

Points from reviewer 2:

1. The mechanistic aspect of the work and conclusions put forth rely heavily on studies performed in cultured myocytes, which are highly glycolytic and generally viewed as a poor model for studying muscle metabolism and insulin action. Nonetheless, the findings provide a strong rationale for moving this line of investigation into mouse gain/loss of function models.

Response: The relative contribution of the anaerobic (glycolysis) and aerobic (mitochondria) contribution to the muscle metabolism can change in L6 depending on differentiation stage. For instance, Serrage et al (PMID30701682) demonstrated that L6-myotubes have a higher mitochondrial abundance and aerobic metabolism than L6-myoblasts. Others have used elegant transcriptomic analysis and metabolic characterisation comparing different skeletal muscle models for studying insulin sensitivity. For instance, Abdelmoez et al in 2020 (PMID31825657) reported that L6 myotubes exhibit greater insulin-stimulated glucose uptake and oxidative capacity compared with C2C12 and Human Mesenchymal Stem Cells (HMSC). Overall, L6 cells exhibit higher metabolic rates and primarily rely on aerobic metabolism, while C2C12 and HSMC cells rely on anaerobic glycolysis. It is worth noting that L6 myotubes are the cell line most closely related to adult human muscle when compared with other muscle cell lines (PMID31825657). Our presented results in Figure 6 H and I provide evidence for the similarities between L6 cells overexpressing SMPD5 and human muscle biopsies. Additionally, in Figure 3J-K, we demonstrate the reproducibility of the negative correlation between ceramide and Coenzyme Q observed in L6 cells in vivo, specifically in the skeletal muscle of mice in chow diet. Furthermore, we have supplemented these findings by demonstrating a comparable relationship in mice exposed to a high-fat diet, as shown in Supplementary Figure 4 H-I (refer to point 4). We will clarify these points in the Discussion:

“In this study, we mainly utilised L6-myotubes, which share many important characteristics with primary muscle fibres relevant to our research. Both types of cells exhibit high sensitivity to insulin and respond similarly to maximal doses of insulin, with Glut4 translocation stimulated between 2 to 4 times over basal levels in response to 100 nM insulin (as shown in Fig. 1-4 and (46,47)). Additionally, mitochondrial respiration in L6-myotubes have a similar sensitivity to mitochondrial poisons, as observed in primary muscle fibres (as shown in Fig. 5 (48)). Finally, inhibiting ceramide production increases CoQ levels in both L6-myotubes and adult muscle tissue (as shown in Fig. 2-3). Therefore, L6-myotubes possess the necessary metabolic features to investigate the role of mitochondria in insulin resistance, and this relationship is likely applicable to primary muscle fibres”.

We will also add additional data - in point 2 - from differentiated human myocytes that are consistent with our observations from the L6 models. Additional experiments are in progress to further extend these findings.

1. One caveat of the approach taken is that exposure of cells to palmitate alone is not reflective of in vivo physiology. It would be interesting to know if similar effects on CoQ are observed when cells are exposed to a more physiological mixture of fatty acids that includes a high ratio of palmitate, but better mimics in vivo nutrition.

Response: Palmitate is widely recognized as a trigger for insulin resistance and ceramide accumulation, which mimics the insulin resistance induced by a diet in rodents and humans. Previous studies have compared the effects of a lipid mixture versus palmitate on inducing insulin resistance in skeletal muscle, and have found that the strong disruption in insulin sensitivity caused by palmitate exposure was lessened with physiologic mixtures of fatty acids, even with a high proportion of saturated fatty acids. This was associated, in part, to the selective partitioning of fatty acids into neutral lipids (such as TAG) when muscle cells are exposed to physiologic lipid mixtures (Newsom et al PMID25793412). Hence, we think that using palmitate is a better strategy to study lipid-induced insulin resistance in vitro. We will add to results:

“In vitro, palmitate conjugated with BSA is the preferred strategy for inducing insulin resistance, as lipid mixtures tend to partition into triacylglycerides (33)”.

We are also performing additional in vivo experiments to add to the physiological relevance of the findings.

1. While the utility of targeting SMPD5 to the mitochondria is appreciated, the results in Figure 5 suggest that this manoeuvre caused a rather severe form of mitochondrial dysfunction. This could be more representative of toxicity rather than pathophysiology. It would be helpful to know if these same effects are observed with other manipulations that lower CoQ to a similar degree. If not, the discrepancies should be discussed.

Response: We conducted a staining procedure using the mitochondrial marker mitoDsRED to observe the effect of SMPD5 overexpression on cell toxicity. The resulting images, displayed in the figure below (Author response image 1), demonstrate that the overexpression of SMPD5 did not result in any significant changes in cell morphology or impact the differentiation potential of our myoblasts into myotubes.

Author response image 1.

In addition, we evaluated cell viability in HeLa cells following exposure to SACLAC (2 uM) to induce CoQ depletion (left panel). Specifically, we measured cell death by monitoring the uptake of Propidium iodide (PI) as shown in the right panel. Our results demonstrated that Saclac-induced CoQ depletion did not lead to cell death at the doses used for CoQ depletion (Author response image 2).

Author response image 2.

Therefore, we deemed it improbable that the observed effect is caused by cellular toxicity, but rather represents a pathological condition induced by elevated levels of ceramides. We will add to discussion:

“...downregulation of the respirasome induced by ceramides may lead to CoQ depletion. Despite the significant impact of ceramide on mitochondrial respiration, we did not observe any indications of cell damage in any of the treatments, suggesting that our models are not explained by toxic/cell death events.”

1. The conclusions could be strengthened by more extensive studies in mice to assess the interplay between mitochondrial ceramides, CoQ depletion and ETC/mitochondrial dysfunction in the context of a standard diet versus HF diet-induced insulin resistance. Does P053 affect mitochondrial ceramide, ETC protein abundance, mitochondrial function, and muscle insulin sensitivity in the predicted directions?

Response: We would like to note that the metabolic characterization and assessment of ETC/mitochondrial function in these mice (both fed a high-fat (HF) and chow diet, with or without P053) were previously published (Turner N, PMID30131496). In addition to this, we have conducted targeted metabolomic and lipidomic analyses to investigate the impact of P053 on ceramide and CoQ levels in HF-fed mice. As illustrated in the figures below (Author response image 3), the administration of P053 led to a reduction in ceramide levels (left panel) and an increase in CoQ levels (right panel) in HF-fed mice, which is consistent with our in vitro findings.

Author response image 3.

We will add to results:

“…similar effect was observed in mice exposed to a high fat diet for 5 wks (Supp. Fig. 4H-I further phenotypic and metabolic characterization of these animals can be found in (41))”

We will further perform more in-vivo studies to corroborate these findings.

Author Response

Reviewer #1 (Public Review):

Summary:

Alonso-Calleja and colleagues explore the role of TGR5 in adult hematopoiesis at both steady state and post-transplantation. The authors utilize two different mouse models including a TGR5-GFP reporter mouse to analyze the expression of TGR5 in various hematopoietic cell subsets. Using germline Tgr5-/- mice it's reported that loss of Tgr5 has no significant impact on steady-state hematopoiesis, with a small decrease in trabecular bone fraction, associated with a reduction in proximal tibia adipose tissue, and an increase in marrow phenotypic adipocytic precursors. The authors further explored the role of stroma TGR5 expression in the hematopoietic recovery upon bone marrow transplantation of wild-type cells, although the studies supporting this claim are weak. Overall, while most of the hematopoietic phenotypes have negative results or small effects, the role of TGR5 in adipose tissue regulation is interesting to the field.

We thank Reviewer 1 for having identified some strengths and weaknesses of our study. As summarized below, we will work to consolidate the weaknesses of our study.

Strengths:

• This is the first time the role of TGR5 has been examined in the bone marrow.

• This paper supports further exploration of the role of bile acids in bone marrow transplantation and possible therapeutic strategies.

Weaknesses:

• The authors fail to describe whether niche stroma cells or adipocyte progenitor cells (APCs) express TGR5.

We are currently working to address this question using our reporter model and expect to be able to provide the data in the next version of the reviewed preprint.

• Although the authors note a significant reduction in bone marrow adipose tissue in Tgr5-/- mice, they do not address whether this is white or brown adipose tissue especially since BA-TGR5 signaling has been shown to play a role in beiging.

The nature of BMAT and how it relates to brown, white or brown/beige adipose tissue has been a persistent question in the field. Our understanding is that BMAT is currently considered a distinct adipose depot that is neither white nor brown/beige. BMAT does not express UCP1 to an appreciable extent, with reports showing its expressing possibly detecting contamination by tissues surrounding bone (Craft et al., 2019). Beyond this consideration, as the regulated BMAT in TGR5-/- mice is almost absent, determination of the brown/beige vs white nature of the regulated BMAT remains technically challenging.

In Figure 1, the authors explore different progenitor subsets but stop short of describing whether TGR5 is expressed in hematopoietic stem cells (HSCs).

Figure 1 of the originally submitted manuscript described TGR5 expression in committed myeloid progenitors (CMP, GMP and MEP). Below we provide the requested data (expression in MPPs and HSCs in Author response image 1) and we have further expanded our data with the expression in megakaryocyte progenitors (MkProg - Lin-cKit+Sca1-CD41+CD150+) as shown in Author response image 2.

Author response image 1.

Frequencies of GFP+ cells in MPPs and HSCs in the BM of 8-12-week-old male TGR5:GFP mice and their controls (n=9 for Wild-type control mice, n=11 for TGR5:GFP mice). Results represent the mean ± s.e.m., n represents biologically independent replicates. Two-tailed Student’s t-test was used for statistical analysis. p-values (exact value) are indicated.

Author response image 2.

A, representative flow cytometry gating strategy used to identify megakaryocyte progenitors (MkProg) and GFP positivity in TGR5:GFP mice and their wild-type controls. B, frequencies of GFP+ cells in MkProg population in the BM of 8-12-week-old male TGR5:GFP mice and their controls (n=3 for Wild-type control mice, n=4 for TGR5:GFP mice). Results represent the mean ± s.e.m., n represents biologically independent replicates. Two-tailed Student’s t-test (B) was used for statistical analysis. p-values (exact value) are indicated.

• Are there more CD45+ cells in the BM because hematopoietic cells are proliferating more due to a direct effect of the loss of Tgr5 or is it because there is just more space due to less trabecular bone?

While we do not have direct evidence to address this question, we see approximately an average 20% increase in CD45+ cell counts in the baseline Tgr5-/- mice. The absolute volume of bone and BMAT lost in these animals does not account for 20% of the total volume of the medullary cavity, so we speculate that the increase in CD45+ counts is not due exclusively to an increase in available volume.

• In Figure 4 no absolute cell counts are provided to support the increase in immunophenotypic APCs (CD45-Ter119-CD31-Sca1+CD24-) in the stroma of Tgr5-/- mice. Accordingly, the absolute number of total stromal cells and other stroma niche cells such as MSCs, ECs are missing.

We initially chose not to report the total number of cells per leg, as the processing of the bones for stroma isolation is less homogenous than that of the HSPC populations (which we do by crushing whole bones with a mortar and pestle). Regardless of these considerations, the data for absolute counts of APCs (left panel), the stroma-enriched fraction (CD45-Ter119-CD31- - middle panel) and endothelial cells (CD45-Ter119-CD31+ - right panel) is provided in Author response image 3. Note that the number of cells plated for CFU-F and BMSC in vitro differentiation is constant between the genotypes, thus confirming the importance of ther elative abundance data shown in the submitted version of the manuscript. In conclusion, we have prioritized the data showing the relative overrepresentation of APC progenitors in the BM stroma as measured by flow cytometry in a per cell basis, which is in line with the functional in vitro data. Further studies could address the specific question through 3D wholemount studies once APC in situ markers are firmly characterized.

Author response image 3.

Left panel: absolute number of adipocyte progenitor cells (APCs) in the CD45-Ter119-CD31- BM stromal gate for bothTgr5+/+ and Tgr5−/− (n=5). Middle panel: absolute number of cells isolated from the stroma-enriched BM fraction (CD45-Ter119-CD31-) in the same mice. Right panel: absolute number of endothelial cells, defined as CD45-Ter119-CD31+, in the same BM isolates.

• There are issues with the reciprocal transplantation design in Fig 4. Why did the authors choose such a low dose (250 000) of BM cells to transplant? If the effect is true and relevant, the early recovery would be observed independently of the setup and a more robust engraftment dataset would be observed without having lethality post-transplant. On the same note, it's surprising that the authors report ~70% lethality post-transplant from wild-type control mice (Fig 4E), according to the literature 200 000 BM cells should ensure the survival of the recipient post-TBI. Overall, the results even in such a stringent setup still show minimal differences and the study lacks further in-depth analyses to support the main claim.

We thank the reviewer for this comment. On the one hand, we disagree on the relevance of the effect size, as Tgr5-/- mice recover from low levels of platelets significantly faster than the Tgr5+/+ controls. Underlining the relevance, in a clinical setting, G-CSF is administered to patients routinely even if the acceleration of recovery is of 1-2 days (Trivedi et al., 2009).

From the point of view of the mortality, we agree that it is higher than expected. We have suffered from cases of swollen muzzles syndrome in our facilities that have greatly hampered our ability to perform myeloablation experiments (Garrett et al., 2019), as even sublethal doses have resulted in the appearance of severe side effects that are reasons for euthanasia under Swiss legislation. For example, a strong reduction in mobility requires immediate euthanasia. All experiments were performed blinded to genotype allocation, so we can reasonably exclude experimenter bias. Finally, it could be argued that mice with more marked symptomatology leading to euthanasia are more likely to have hematopoietic deficits, which in our case was mostly seen for Tgr5+/+animals. We have therefore chosen to report mortality together with the longitudinal assessment of peripheral blood counts.

• Mechanistically, how does the loss of Tgr5 impact hematopoietic regeneration following sublethal irradiation?

The question of a non-lethal hematopoietic stress is a very relevant one. Unfortunately, and as delineated in the previous point, we have been seriously conditioned by cases of swollen muzzles syndrome (Garrett et al., 2019) that have stopped us from proceeding with more irradiation studies. We will profit from the change of animal facility that will consolidate during the upcoming year Labora(tory of Regenerative Hematopoiesis) to address this point in follow-up studies.

• Only male mice were used throughout this study. It would be beneficial to know whether female mice show similar results.

We agree with this comment, and we expect to include the characterization of BM microenvironment (Figure 3 of the current manuscript) in females in the reviewed version of the manuscript when a suitable cohort becomes available.

Reviewer #2 (Public Review):

Summary: In this manuscript, the authors examined the role of the bile acid receptor TGR5 in the bone marrow under steady-state and stress hematopoiesis. They initially showed the expression of TGR5 in hematopoietic compartments and that loss of TGR5 doesn't impair steady-state hematopoiesis. They further demonstrated that TGR5 knockout significantly decreases BMAT, increases the APC population, and accelerates the recovery upon bone marrow transplantation.

Strengths: The manuscript is well-structured and well-written.

We thank Reviewer #2 for this comment.

Weaknesses: The mechanism is not clear, and additional studies need to be performed to support the authors' conclusion.

We agree with Reviewer #2 that more studies are needed to understand what the role of TGR5 in the hematopoietic system is. We have been hampered in our studies of stress hematopoiesis because of frequent cases of swollen muzzles syndrome (Garrett et al., 2019), which has made difficult to continue with experiments involving myelosuppression (see response to Reviewer #1 as well). Further studies are planned or ongoing, including determining the role of the microbiome on the observed TGR5 bone and hematopoiesis stress phenotypes, but will be the focus of a separate study.

References

Craft, C.S., Robles, H., Lorenz, M.R., Hilker, E.D., Magee, K.L., Andersen, T.L., Cawthorn, W.P., MacDougald, O.A., Harris, C.A., Scheller, E.L., 2019. Bone marrow adipose tissue does not express UCP1 during development or adrenergic-induced remodeling. Sci Rep 9, 17427. https://doi.org/10.1038/s41598-019-54036-x

Garrett, J., Sampson, C.H., Plett, P.A., Crisler, R., Parker, J., Venezia, R., Chua, H.L., Hickman, D.L., Booth, C., MacVittie, T., Orschell, C.M., Dynlacht, J.R., 2019. Characterization and Etiology of Swollen Muzzles in Irradiated Mice. Radiat Res 191, 31–42. https://doi.org/10.1667/RR14724.1

Trivedi, M., Martinez, S., Corringham, S., Medley, K., Ball, E.D., 2009. Optimal use of G-CSF administration after hematopoietic SCT. Bone Marrow Transplant 43, 895–908. https://doi.org/10.1038/bmt.2009.75

Author Response

eLife assessment

In this valuable study, the authors investigate the mechanism of amyloid nucleation in a cellular system using their novel ratiometric measurements and uncover interesting insights regarding the role of polyglutamine length and the sequence features of glutamine-rich regions on amyloid formation. Overall, the problem is significant and being able to assess nucleation in cells is of considerable relevance. The data, as presented and analyzed, are currently still incomplete. The specific claims would be stronger if based on in vitro measurements that avoid the intricacies of specific cellular systems and that are more suitable for assessing sequence-intrinsic properties.

We are pleased that the editors find our study valuable. We find that the reviewers’ criticisms largely arise from misunderstandings inherent to the conceptually challenging nature of the topic, rather than fundamental flaws, as we will elaborate here. We are grateful for the opportunity afforded by eLife to engage reviewers in a constructive public dialogue.

Reviewer #1 (Public Review):

The authors take on the challenge of defining the core nucleus for amyloid formation by polyglutamine tracts. This rests on the assertion that polyQ forms amyloid structures to the exclusion of all other forms of solids. Using their unique assay, deployed in yeast, the authors attempt to infer the size of the nucleus that templates amyloid formation by polyQ. Further, through a series of sequence titrations, all studied using a single type of assay, the authors converge on an assertion stating that a single polyQ molecule is the nucleus for amyloid formation, that 12-residues make up the core of the nucleus, that it takes ca. 60 Qs in a row to unmask this nucleation potential, and that polyQ amyloid formation belongs to the same universality class as self-poisoned crystallization, which is the hallmark of crystallization from polymer melts formed by large, high molecular weight synthetic polymers. Unfortunately, the authors have decided to lean in hard on their assertions without a critical assessment of whether their findings stand up to scrutiny. If their findings are truly an intrinsic property of polyQ molecules, then their findings should be reconstituted in vitro. Unfortunately, careful and rigorous experiments in vitro show that there is a threshold concentration for forming fibrillar solids. This threshold concentration depends on the flanking sequence context on temperature and on solution conditions. The existence of a threshold concentration defies the expectation of a monomer nucleus. The findings disagree with in vitro data presented by Crick et al., and ignored by the authors. Please see: https://doi.org/10.1073/pnas.1320626110. These reports present data from very different assays, the importance of which was underscored first by Regina Murphy and colleagues. The work of Crick et al., provides a detailed thermodynamic framework - see the SI Appendix. This framework dove tails with theory and simulations of Zhang and Muthukumar, which explains exactly how a system like polyQ might work (https://doi.org/10.1063/1.3050295). The picture one paints is radically different from what the authors converge upon. One is inclined to lean toward data that are gleaned using multiple methods in vitro because the test tube does not have all the confounding effects of a cellular milieu, especially when it comes to focusing on sequence-intrinsic conformational transitions of a protein. In addition to concerns about the limitations of the DAmFRET method, which based on the work of the authors in their collaborative paper by Posey et al., are being stretched to the limit, there is the real possibility that the cellular milieu, unique to the system being studied, is enabling transitions that are not necessarily intrinsic to the sequence alone. A nod in this direction is the work of Marc Diamond, which showed that having stabilized the amyloid form of Tau through coacervation, there is a large barrier that limits the loss of amyloid-like structure for Tau. There may well be something similar going on with the polyQ system. If the authors could show that their data are achievable in vitro without anything but physiological buffers one would have more confidence in a model that appears to contradict basic physical principles of how homopolymers self-assemble. Absent such additional evidence, numerous statements seem to be too strong. There are also several claims that are difficult to understand or appreciate.

Rebuttal to the perceived necessity of in vitro experiments

The overarching concern of this reviewer and reviewing editor is whether in-cell assays can inform on sequence-intrinsic properties. We understand this concern. We believe however that the relative merit of in-cell assays is largely a matter of perspective. The truly sequence-intrinsic behavior of polyQ, i.e. in a vacuum, is less informative than the “sequence-intrinsic” behaviors of interest that emerge in the presence of extraneous molecules from the appropriate biological context. In vitro experiments typically include a tiny number of these -- water, ions, and sometimes a crowding agent meant to approximate everything else. Obviously missing are the myriad quinary interactions with other proteins that collectively round out the physiological solvent. The question is what experimental context best approximates that of a living human neuron under which the pathological sequence-dependent properties of polyQ manifest. We submit that a living yeast cell comes closer to that ideal than does buffer in a test tube.

The reviewer’s statements that our findings must be validated in vitro ignores the fact -- stressed in our introduction -- that decades of in vitro work have not yet generated definitive evidence for or against any specific nucleus model. In addition to the above, one major problem concerns the large sizes of in vitro systems that obscure the effects of primary nucleation. For example, a typical in vitro experimental volume of e.g. 1.5 ml is over one billion-fold larger than the femtoliter volume of a cell. This means that any nucleation-limited kinetics of relevant amyloid formation are lost, and any alternative amyloid polymorphs that have a kinetic growth advantage -- even if they nucleate at only a fraction the rate of relevant amyloid -- will tend to dominate the system (Buell, 2017). Novel approaches are clearly needed to address these problems. We present such an approach, stretch it to the limit (as the reviewer notes) across multiple complementary experiments, and arrive at a novel finding that is fully and uniquely consistent with all of our own data as well as the collective prior literature.

That the preceding considerations are collectively essential to understand relevant amyloid behavior is evident from recent cryoEM studies showing that in vitro-generated amyloid structures generally differ from those in patients (Arseni et al., 2022; Bansal et al., 2021; Radamaker et al., 2021; Schmidt et al., 2019; Schweighauser et al., 2020; Yang et al., 2022). This is highly relevant to the present discourse because each amyloid structure is thought to emanate from a different nucleating structure. This means that in vitro experiments have broadly missed the mark in terms of the relevant thermodynamic parameters that govern disease onset and progression. Note that the rules laid out via our studies are not only consistent with structural features of polyQ amyloid in cells, but also (as described in the discussion) explain why the endogenous structure of a physiologically relevant Q zipper amyloid differs from that of polyQ.

A recent collaboration between the Morimoto and Knowles groups (Sinnige et al.) investigated the kinetics of aggregation by Q40-YFP expressed in C. elegans body wall muscle cells, using quantitative approaches that have been well established for in vitro amyloid-forming systems of the type favored by the reviewer. They calculate a reaction order of just 1.6, slightly higher than what would be expected for a monomeric nucleus but nevertheless fully consistent with our own conclusions when one accounts for the following two aspects of their approach. First, the polyQ tract in their construct is flanked by short poly-Histidine tracts on both sides. These charges very likely disfavor monomeric nucleation because all possible configurations of a four-stranded bundle position the beginning and end of the Q tract in close proximity, and Q40 is only just long enough to achieve monomeric nucleation in the absence of such destabilization. Second, the protein is fused to YFP, a weak homodimer (Landgraf et al., 2012; Snapp et al., 2003). With these two considerations, our model -- which was generated from polyQ tracts lacking flanking charges or an oligomeric fusion -- predicts that amyloid nucleation by their construct will occur more frequently as a dimer than a monomer. Indeed, their observed reaction order of 1.6 supports a predominantly dimeric nucleus. Like us and others, Sinnige et al. did not observe phase separation prior to amyloid formation. This is important because it not only argues against nucleation occurring in a condensate, it also suggests that the reaction order they calculated has not been limited by the concentration-buffering effect of phase separation.

While we agree that our conclusions rest heavily on DAmFRET data (for good reason), we do provide supporting evidence from molecular dynamics simulations, SDD-AGE, and microscopy.

To summarize, given the extreme limitations of in vitro experiments in this field, the breadth of our current study, and supporting findings from another lab using rigorous quantitative approaches, we feel that our claims are justified without in vitro data.

Rebuttal to the perceived incompatibility of monomeric nucleation with the existence of a critical concentration for amyloid

We appreciate that the concept of a monomeric nucleus can superficially appear inconsistent with the fact that crystalline solids such as polyQ amyloid have a saturating concentration, but this is only true if one neglects that polyQ amyloids are polymer crystals with intramolecular ordering. The perceived discrepancy is perhaps most easily dispelled by protein crystallography. Folded proteins form crystals. These crystals have critical concentrations, and the protein subunits within them each have intramolecular crystalline order (in the form of secondary structure). To extrapolate these familiar examples to our present finding with polyQ, one need only appreciate the now well-established phenomenon of secondary nucleation, whereby transient interactions of soluble species with the ordered species leads to their own ordering (Törnquist et al., 2018). Transience is important here because it implies that intramolecular ordering can in principle propagate even in solutions that are subsaturated with respect to bulk crystallization. This is possible in the present case because the pairing of sufficiently short beta strands (equivalent to “stems” in the polymer crystal literature) will be more stable intramolecularly than intermolecularly, due to the reduced entropic penalty of the former. Our elucidation that Q zipper ordering can occur with shorter strands intramolecularly than intermolecularly (Fig. S4C-D) demonstrates this fact. It is also evident from published descriptions of single molecule “crystals” formed in sufficiently dilute solutions of sufficiently long polymers (Hong et al., 2015; Keller, 1957; Lauritzen and Hoffman, 1960).

In suggesting that a saturating concentration for amyloid rules out monomeric nucleation, the reviewer assumes that the Q zipper-containing monomer must be stable relative to the disordered ensemble. This is not inherent to our claim and in fact opposes the definition of a nucleus. The monomeric nucleating structure need not be more stable than the disordered state, and monomers may very well be disordered at equilibrium at low concentrations. To be clear, our claim requires that the Q zipper-containing monomer is both on pathway to amyloid and less stable than all subsequent species that are on pathway to amyloid. The former requirement is supported by our extensive mutational analysis. The latter requirement is supported by our atomistic simulations showing the Q zipper-containing monomer is stabilized by dimerization (see our 2021 preprint). Hence, requisite ordering in the nucleating monomer is stabilized by intermolecular interactions. We provide in Author response image 1 an illustration to clarify what we believe to be the discrepancy between our claim and the reviewer’s interpretation.

Author response image 1.

That the rate-limiting fluctuation for a crystalline phase can occur in a monomer can also be understood as a consequence of Ostwald’s rule of stages, which describes the general tendency of supersaturated solutes, including amyloid forming proteins (Chakraborty et al., 2023), to populate metastable phases en route to more stable phases (De Yoreo, 2022; Schmelzer and Abyzov, 2017). Our findings with polyQ are consistent with a general mechanism for Ostwald’s rule wherein the relative stabilities of competing polymorphs differ with the number of subunits (De Yoreo, 2022; Navrotsky, 2004). As illustrated in Fig. 6 of Navrotsky, a polymorph that is relatively stable at small particle sizes tends to give way to a polymorph that -- while initially unstable -- becomes more stable as the particles grow. The former is analogous to our early stage Q zipper composed of two short sheets with an intramolecular interface, while the latter is analogous to the later stage Q zipper composed of longer sheets with an intermolecular interface. Subunit addition stabilizes the latter more than the former, hence the initial Q zipper that is stabilized more by intra- than intermolecular interactions will mature with growth to one that is stabilized more by intermolecular interactions.

We apologize to the Pappu group for neglecting to cite Crick et al. 2013 in the current preprint. Contrary to the reviewer’s assessment, however, we find that the conclusions of this valuable study do more to support than to refute our findings. Briefly, Crick et al. investigated the aggregation of synthetic Q30 and Q40 peptides in vitro, wherein fibrils assembled from high concentrations of peptide were demonstrated to have saturating concentrations in the low micromolar range. As explained above, this finding of a saturating concentration does not refute our results. More relevant to the present work are their findings that “oligomers” accumulated over an hours-long timespan in solutions that are subsaturated with respect to fibrils, and these oligomers themselves have (nanomolar) critical concentrations. The authors postulated that the oligomers result from liquid–liquid demixing of intrinsically disordered polyglutamine. However, phase separation by a peptide is expected to fix its concentration in both the solute and condensed phases, and, because disordered phase separation is inherently faster than amyloid formation, the postulated explanation removes the driving force for any amyloid phase with a critical solubility greater than that of the oligomers. In place of this interpretation that truly does appear to -- in the reviewer’s words -- “contradict basic physical principles of how homopolymers self-assemble”, we interpret these oligomers as evidence of our Q zipper-containing self-poisoned multimers, rounded as an inherent consequence of self-poisoning (Ungar et al., 2005), and likely akin to semicrystalline spherulites that have been observed in other polymer crystal and amyloid-forming systems (Crist and Schultz, 2016; Vetri and Foderà, 2015). That Crick et al. also observed the formation of a relatively labile amyloid phase when the reactions were started with 50 uM peptide is unsurprising in light of the aforementioned kinetic advantage that large reaction volumes can confer to labile polymorphs, and that high concentrations (in this case, orders of magnitude higher than the likely physiological concentration of polyQ (Wild et al., 2015)) can favor the formation of labile amyloid polymorphs (Ohhashi et al., 2010). Indeed, a contemporaneous study by the Wetzel group using very similar peptide constructs and polyQ lengths -- but beginning with lower concentrations -- found that the relevant saturating concentrations for amyloid lie below their limit of detection of 100 nM (Sahoo et al., 2014).

Rebuttals to other critiques

The reviewer states that we found nucleation potential to require 60 Qs in a row. Our data are collectively consistent with nucleation occurring at and above approximately 36 Qs, a point repeated in the paper. The reviewer may be referring to our statement, ”Sixty residues proved to be the optimum length to observe both the pre- and post-nucleated states of polyQ in single experiments”. The purpose of this statement is simply to describe the practical consideration that led us to use 60 Qs for the bulk of our assays. We do appreciate that the fraction of AmFRET-positive cells is very low for lengths just above the threshold, especially Q40. They are nevertheless highly significant (p = 0.004 in [PIN+] cells, one-tailed T-test), and we will modify the figure and text to clarify this.

The reviewer characterizes self-poisoning as the hallmark of crystallization from polymer melts, which would be problematic for our conclusions if self-poisoning were limited to this non-physiological context. In fact the term was first used to describe crystallization from solution (Organ et al., 1989), wherein the phenomenon is more pronounced (Ungar et al., 2005).

Reviewer #2 (Public Review):

Numerous neurodegenerative diseases are thought to be driven by the aggregation of proteins into insoluble filaments known as "amyloids". Despite decades of research, the mechanism by which proteins convert from the soluble to insoluble state is poorly understood. In particular, the initial nucleation step is has proven especially elusive to both experiments and simulation. This is because the critical nucleus is thermodynamically unstable, and therefore, occurs too infrequently to directly observe. Furthermore, after nucleation much faster processes like growth and secondary nucleation dominate the kinetics, which makes it difficult to isolate the effects of the initial nucleation event. In this work Kandola et al. attempt to surmount these obstacles using individual yeast cells as microscopic reaction vessels. The large number of cells, and their small size, provides the statistics to separate the cells into pre- and post-nucleation populations, allowing them to obtain nucleation rates under physiological conditions. By systematically introducing mutations into the amyloid-forming polyglutamine core of huntingtin protein, they deduce the probable structure of the amyloid nucleus. This work shows that, despite the complexity of the cellular environment, the seemingly random effects of mutations can be understood with a relatively simple physical model. Furthermore, their model shows how amyloid nucleation and growth differ in significant ways, which provides testable hypotheses for probing how different steps in the aggregation pathway may lead to neurotoxicity.

In this study Kandola et al. probe the nucleation barrier by observing a bimodal distribution of cells that contain aggregates; the cells containing aggregates have had a stochastic fluctuation allowing the proteins to surmount the barrier, while those without aggregates have yet to have a fluctuation of suitable size. The authors confirm this interpretation with the selective manipulation of the PIN gene, which provides an amyloid template that allows the system to skip the nucleation event.

In simple systems lacking internal degrees of freedom (i.e., colloids or rigid molecules) the nucleation barrier comes from a significant entropic cost that comes from bringing molecules together. In large aggregates this entropic cost is balanced by attractive interactions between the particles, but small clusters are unable to form the extensive network of stabilizing contacts present in the larger aggregates. Therefore, the initial steps in nucleation incur an entropic cost without compensating attractive interactions (this imbalance can be described as a surface tension). When internal degrees of freedom are present, such as the conformational states of a polypeptide chain, there is an additional contribution to the barrier coming from the loss of conformational entropy required to the adopt aggregation-prone state(s). In such systems the clustering and conformational processes do not necessarily coincide, and a major challenge studying nucleation is to separate out these two contributions to the free energy barrier. Surprisingly, Kandola et al. find that the critical nucleus occurs within a single molecule. This means that the largest contribution to the barrier comes from the conformational entropy cost of adopting the beta-sheet state. Once this state is attained, additional molecules can be recruited with a much lower free energy barrier.

There are several caveats that come with this result. First, the height of the nucleation barrier(s) comes from the relative strength of the entropic costs compared to the binding affinities. This balance determines how large a nascent nucleus must grow before it can form interactions comparable to a mature aggregate. In amyloid nuclei the first three beta strands form immature contacts consisting of either side chain or backbone contacts, whereas the fourth strand is the first that is able to form both kinds of contacts (as in a mature fibril). This study used relatively long polypeptides of 60 amino acids. This is greater than the 20-40 amino acids found in amyloid-forming molecules like ABeta or IAPP. As a result, Kandola et al.'s molecules are able to fold enough times to create four beta strands and generate mature contacts intramolecularly. The authors make the plausible claim that these intramolecular folds explain the well-known length threshold (L~35) observed in polyQ diseases. The intramolecular folds reduce the importance of clustering multiple molecules together and increase the importance of the conformational states. Similarly, manipulating the sequence or molecular concentrations will be expected to manipulate the relative magnitude of the binding affinities and the clustering entropy, which will shift the relative heights of the entropic barriers.

The reviewer correctly notes that the majority of our manipulations were conducted with 60-residue long tracts (which corresponds to disease onset in early adulthood), and this length facilitates intramolecular nucleation. However, we also analyzed a length series of polyQ spanning the pathological threshold, as well as a synthetic sequence designed explicitly to test the model nucleus structure with a tract shorter than the pathological threshold, and both experiments corroborate our findings.

The authors make an important point that the structure of the nucleus does not necessarily resemble that of the mature fibril. They find that the critical nucleus has a serpentine structure that is required by the need to form four beta strands to get the first mature contacts. However, this structure comes at a cost because residues in the hairpins cannot form strong backbone or zipper interactions. Mature fibrils offer a beta sheet template that allows incoming molecules to form mature contacts immediately. Thus, it is expected that the role of the serpentine nucleus is to template a more extended beta sheet structure that is found in mature fibrils.

A second caveat of this work is the striking homogeneity of the nucleus structure they describe. This homogeneity is likely to be somewhat illusory. Homopolymers, like polyglutamine, have a discrete translational symmetry, which implies that the hairpins needed to form multiple beta sheets can occur at many places along the sequence. The asparagine residues introduced by the authors place limitations on where the hairpins can occur, and should be expected to increase structural homogeneity. Furthermore, the authors demonstrate that polyglutamine chains close to the minimum length of ~35 will have strict limitations on where the folds must occur in order to attain the required four beta strands.

We are unsure how to interpret the above statements as a caveat. We agree that increasing sequence complexity will tend to increase homogeneity, but this is exactly the motivation of our approach. We explicitly set out to determine the minimal complexity sequence sufficient to specify the nucleating conformation, which we ultimately identified in terms of secondary and tertiary structure. We do not specify which parts of a long polyQ tract correspond to which parts of the structure, because, as the reviewer points out, they can occur at many places. Hence, depending on the length of the polyQ tract, the nucleus we describe may have any length of sequence connecting the strand elements. We do not think that the effects of N-residue placement can be interpreted as a confounding influence on hairpin position because the striking even-odd pattern we observe implicates the sides of beta strands rather than the lengths. Moreover, we observe this pattern regardless of the residue used (Gly, Ser, Ala, and His in addition to Asn).

A novel result of this work is the observation of multiple concentration regimes in the nucleation rate. Specifically, they report a plateau-like regime at intermediate regimes in which the nucleation rate is insensitive to protein concentration. The authors attribute this effect to the "self-poisoning" phenomenon observed in growth of some crystals. This is a valid comparison because the homogeneity observed in NMR and crystallography structures of mature fibrils resemble a one-dimensional crystal. Furthermore, the typical elongation rate of amyloid fibrils (on the order of one molecule per second) is many orders of magnitude slower than the molecular collision rate (by factors of 10^6 or more), implying that the search for the beta-sheet state is very slow. This slow conformational search implies the presence of deep kinetic traps that would be prone to poisoning phenomena. However, the observation of poisoning in nucleation during nucleation is striking, particularly in consideration of the expected disorder and concentration sensitivity of the nucleus. Kandola et al.'s structural model of an ordered, intramolecular nucleus explains why the internal states responsible for poisoning are relevant in nucleation.

We thank the reviewer for noting the novelty and plausibility of the self-poisoning connection. We would like to elaborate on our finding that self-poisoning inhibits nucleation (in addition to elongation), as this could prove confusing to some readers. While self-poisoning is claimed to inhibit primary nucleation in the polymer crystal literature (Ungar et al., 2005; Zhang et al., 2018), the semantics of “nucleation” in this context warrants clarification. Technically, the same structure can be considered a nucleus in one context but not in another. The Q zipper monomer, even if it is rate-limiting for amyloid formation at low concentrations (and is therefore the “nucleus”), is not necessarily rate-limiting when self-poisoned at high concentrations. Whether it comprises the nucleus in this case depends on the rates of Q zipper formation relative to subunit addition to the poisoned state. If the latter happens slower than Q zipper formation de novo, it can be said that self-poisoning inhibits nucleation, regardless of whether the Q zipper formed. We suspect this to be the mechanism by which preemptive oligomerization blocks nucleation in the case of polyQ, though other mechanisms may be possible.

To achieve these results the authors used a novel approach involving a systematic series of simple sequences. This is significant because, while individual experiments showed seemingly random behavior, the randomness resolved into clear trends with the systematic approach. These trends provided clues to build a model and guide further experiments.

Reviewer #3 (Public Review):

Kandola et al. explore the important and difficult question regarding the initiating event that triggers (nucleates) amyloid fibril growth in glutamine-rich domains. The researchers use a fluorescence technique that they developed, dAMFRET, in a yeast system where they can manipulate the expression level over several orders of magnitude, and they can control the length of the polyglutamine domain as well as the insertion of interfering non-glutamine residues. Using flow cytometry, they can interrogate each of these yeast 'reactors' to test for self-assembly, as detected by FRET.

In the introduction, the authors provide a fairly thorough yet succinct review of the relevant literature into the mechanisms of polyglutamine-mediated aggregation over the last two decades. The presentation as well as the illustrations in Figure 1A and 1B are difficult to understand, and unfortunately, there is no clear description of the experimental technique that would allow the reader to connect the hypothetical illustrations to the measurement outcomes. The authors do not explain what the FRET signal specifically indicates or what its intensity is correlated to. FRET measures distance between donor and acceptor, but can it be reliably taken as an indicator of a specific beta-sheet conformation and of amyloid? Does the signal increase with both nucleation and with elongation, and is the signal intensity the same if, e.g., there were 5 aggregates of 10 monomers each versus 50 monomeric nuclei? Is there a reason why the AmFRET signal intensity decreases at longer Q even though the number of cells with positive signal increases? Does the number of positive cells increase with time? The authors state later that 'non-amyloid containing cells lacked AmFRET altogether', but this seems to be a tautology - isn't the lack of AmFRET taken as a proof of lack of amyloid? Overall, a clearer description of the experimental method and what is actually measured (and validation of the quantitative interpretation of the FRET signal) would greatly assist the reader in understanding and interpreting the data.

We believe the difficulty in understanding the illustrations in Figure 1A and 1B is inherent to the subject. We agree that elaborating how DAmFRET works would help the reader, and will add a few sentences to this end. Beyond this, we refer the reviewer and readers to our cited prior work describing the theory and interpretation of DAmFRET. Note that the y-axes of DAmFRET plots are not raw FRET but rather “AmFRET”, a ratio of FRET to total expression level. As explained thoroughly in our cited prior work, the discontinuity of AmFRET with expression level indicates that the high AmFRET-population formed via a disorder-to-order transition. When the query protein is predicted to be intrinsically disordered, the discontinuous transition to high AmFRET invariably (among hundreds of proteins tested in prior published and unpublished work) signifies amyloid formation as corroborated by SDD-AGE and tinctorial assays.

When performed using standard flow cytometry as in the present study, every AmFRET measurement corresponds to a cell-wide average, and hence does not directly inform on the distribution of the protein between different stoichiometric species. As there is only one fluorophore per protein molecule, monomeric nuclei have no signal. DAmFRET can distinguish cells expressing monomers from stable dimers from higher order oligomers (see e.g. Venkatesan et al. 2019), and we are therefore quite confident that AmFRET values of zero correspond to cells in which a vast majority of the respective protein is not in homo-oligomeric species (i.e. is monomeric or in hetero-complexes with endogenous proteins). The exact value of AmFRET, even for species with the same stoichiometry, will depend both on the effect of their respective geometries on the proximity of mEos3.1 fluorophores, and on the fraction of protein molecules in the species. Hence, we only attempt to interpret the plateau values of AmFRET (where the fraction of protein in an assembled state approaches unity) as directly informing on structure, as we did in Fig. S3A.

We believe that AmFRET decreases with longer polyQ because the mass fraction of fluorophore decreases in the aggregate, simply because the extra polypeptide takes up volume in the aggregate.

Yes, the fraction of positive cells in a discontinuous DAmFRET plot does increase with time. However, given the more laborious data collection and derivation of nucleation kinetics in a system with ongoing translation, especially across hundreds of experiments with other variables, ours is a snapshot measurement to approximately derive the relative contributions of intra- and intermolecular fluctuations to the nucleation barrier, rather than the barrier’s magnitude.

We will revise the tautological statement by removing “non-amyloid containing”.

The authors demonstrate that their assay shows that the fraction of cells with AmFRET signal increases strongly with an increase in polyQ length, with a 'threshold around 50-60 glutamines. This roughly correlates with the Q-length dependence of disease. The experiments in which asparagine or other amino acids are inserted at variable positions in the glutamine repeat are creative and thorough, and the data along with the simulations provide compelling support for the proposed Q zipper model. The experiments shown in Figure 5 are strongly supportive of a model where formation of the beta-sheet nucleus is within a monomer. This is a potentially important result, as there are conflicting data in the literature as to whether the nucleus in polyQ is monomer.

We thank the reviewer for these comments. We wish to clarify one important point, however, concerning the correlation of our data with the pathological length threshold. As we state in the first results section, “Our data recapitulated the pathologic threshold -- Q lengths 35 and shorter lacked AmFRET, indicating a failure to aggregate or even appreciably oligomerize, while Q lengths 40 and longer did acquire AmFRET in a length and concentration-dependent manner”. Hence, most of our experiments were conducted with 60Q not because it resembles the pathological threshold, but rather because it was most convenient for DAmFRET experiments.

I did not find the argument, that their data shows the Q zipper grows in two dimensions, compelling; there are more direct experimental methods to answer this question. I was also confused by the section that Q zippers poison themselves. It would be easier for the reader to follow if the authors first presented their results without interpretation. The data seem more consistent with an argument that, at high concentrations, non-structured polyQ oligomers form which interfere with elongation into structured amyloid assemblies - but such oligomers would not be zippers.

Self-poisoning is a widely observed and heavily studied phenomenon in polymer crystal physics, though it seems not yet to have entered the lexicon of amyloid biologists. We were new to this concept before it emerged as an extremely parsimonious explanation for our results. As described in the text, two pieces of evidence exclude the alternative mechanism suggested by the reviewer -- that non-structured oligomers form and subsequently engage and inhibit the template. Specifically, 1) inhibition occurs without any detectable FRET, even at high total protein concentration, indicating the species do not form in a concentration-dependent manner that would be expected of disordered oligomers; and 2) inhibition itself has strict sequence requirements that match those of Q zippers. Hence our data collectively suggest that inhibition is a consequence of the deposition of partially ordered molecules onto the templating surface.

Although some speculation or hypothesizing is perfectly appropriate in the discussion, overall the authors stretch this beyond what can be supported by the results. A couple of examples: The conclusion that toxicity arises from 'self-poisoned polymer crystals' is not warranted, as there is no relevant data presented in this manuscript. The authors refer to findings 'that kinetically arrested aggregates emerge from the same nucleating event responsible for amyloid formation', but I cannot recall any evidence for this statement in the results section.

We restricted any mention of toxicity to the introduction and a section in the discussion that is not worded as conclusive. Nevertheless, we will soften the subheading and text of the relevant section in the discussion to more clearly indicate the speculative nature of the statements.

We stand by our statement 'that kinetically arrested aggregates emerge from the same nucleating event responsible for amyloid formation', as this follows directly from self-poisoning.

Bibliography

Arseni D, Hasegawa M, Murzin AG, Kametani F, Arai M, Yoshida M, Ryskeldi-Falcon B. 2022. Structure of pathological TDP-43 filaments from ALS with FTLD. Nature 601:139–143. doi:10.1038/s41586-021-04199-3

Bansal A, Schmidt M, Rennegarbe M, Haupt C, Liberta F, Stecher S, Puscalau-Girtu I, Biedermann A, Fändrich M. 2021. AA amyloid fibrils from diseased tissue are structurally different from in vitro formed SAA fibrils. Nat Commun 12:1013. doi:10.1038/s41467-021-21129-z

Buell AK. 2017. The Nucleation of Protein Aggregates - From Crystals to Amyloid Fibrils. Int Rev Cell Mol Biol 329:187–226. doi:10.1016/bs.ircmb.2016.08.014

Chakraborty D, Straub JE, Thirumalai D. 2023. Energy landscapes of Aβ monomers are sculpted in accordance with Ostwald’s rule of stages. Sci Adv 9:eadd6921. doi:10.1126/sciadv.add6921 Crist B, Schultz JM. 2016. Polymer spherulites: A critical review. Prog Polym Sci 56:1–63. doi:10.1016/j.progpolymsci.2015.11.006

De Yoreo JJ. 2022. Casting a bright light on Ostwald’s rule of stages. Proc Natl Acad Sci USA 119. doi:10.1073/pnas.2121661119

Hong Y, Yuan S, Li Z, Ke Y, Nozaki K, Miyoshi T. 2015. Three-Dimensional Conformation of Folded Polymers in Single Crystals. Phys Rev Lett 115:168301. doi:10.1103/PhysRevLett.115.168301

Keller A. 1957. A note on single crystals in polymers: Evidence for a folded chain configuration. Philosophical Magazine 2:1171–1175. doi:10.1080/14786435708242746

Landgraf D, Okumus B, Chien P, Baker TA, Paulsson J. 2012. Segregation of molecules at cell division reveals native protein localization. Nat Methods 9:480–482. doi:10.1038/nmeth.1955

Lauritzen JI, Hoffman JD. 1960. Theory of Formation of Polymer Crystals with Folded Chains in Dilute Solution. J Res Natl Bur Stand A Phys Chem 64A:73–102. doi:10.6028/jres.064A.007

Navrotsky A. 2004. Energetic clues to pathways to biomineralization: precursors, clusters, and nanoparticles. Proc Natl Acad Sci USA 101:12096–12101. doi:10.1073/pnas.0404778101

Ohhashi Y, Ito K, Toyama BH, Weissman JS, Tanaka M. 2010. Differences in prion strain conformations result from non-native interactions in a nucleus. Nat Chem Biol 6:225–230. doi:10.1038/nchembio.306

Organ SJ, Ungar G, Keller A. 1989. Rate minimum in solution crystallization of long paraffins. Macromolecules 22:1995–2000. doi:10.1021/ma00194a078

Radamaker L, Baur J, Huhn S, Haupt C, Hegenbart U, Schönland S, Bansal A, Schmidt M, Fändrich M. 2021. Cryo-EM reveals structural breaks in a patient-derived amyloid fibril from systemic AL amyloidosis. Nat Commun 12:875. doi:10.1038/s41467-021-21126-2

Sahoo B, Singer D, Kodali R, Zuchner T, Wetzel R. 2014. Aggregation behavior of chemically synthesized, full-length huntingtin exon1. Biochemistry 53:3897–3907. doi:10.1021/bi500300c

Schmelzer JWP, Abyzov AS. 2017. How do crystals nucleate and grow: ostwald’s rule of stages and beyond In: Šesták J, Hubík P, Mareš JJ, editors. Thermal Physics and Thermal Analysis, Hot Topics in Thermal Analysis and Calorimetry. Cham: Springer International Publishing. pp. 195–211. doi:10.1007/978-3-319-45899-1_9

Schmidt M, Wiese S, Adak V, Engler J, Agarwal S, Fritz G, Westermark P, Zacharias M, Fändrich M. 2019. Cryo-EM structure of a transthyretin-derived amyloid fibril from a patient with hereditary ATTR amyloidosis. Nat Commun 10:5008. doi:10.1038/s41467-019-13038-z

Schweighauser M, Shi Y, Tarutani A, Kametani F, Murzin AG, Ghetti B, Matsubara T, Tomita T, Ando T, Hasegawa K, Murayama S, Yoshida M, Hasegawa M, Scheres SHW, Goedert M. 2020. Structures of α-synuclein filaments from multiple system atrophy. Nature 585:464–469. doi:10.1038/s41586-020-2317-6

Snapp EL, Hegde RS, Francolini M, Lombardo F, Colombo S, Pedrazzini E, Borgese N, Lippincott-Schwartz J. 2003. Formation of stacked ER cisternae by low affinity protein interactions. J Cell Biol 163:257–269. doi:10.1083/jcb.200306020

Törnquist M, Michaels TCT, Sanagavarapu K, Yang X, Meisl G, Cohen SIA, Knowles TPJ, Linse S. 2018. Secondary nucleation in amyloid formation. Chem Commun 54:8667–8684. doi:10.1039/c8cc02204f

Ungar G, Putra EGR, de Silva DSM, Shcherbina MA, Waddon AJ. 2005. The Effect of Self-Poisoning on Crystal Morphology and Growth Rates In: Allegra G, editor. Interphases and Mesophases in Polymer Crystallization I, Advances in Polymer Science. Berlin, Heidelberg: Springer Berlin Heidelberg. pp. 45–87. doi:10.1007/b107232

Vetri V, Foderà V. 2015. The route to protein aggregate superstructures: Particulates and amyloid-like spherulites. FEBS Lett 589:2448–2463. doi:10.1016/j.febslet.2015.07.006

Wild EJ, Boggio R, Langbehn D, Robertson N, Haider S, Miller JRC, Zetterberg H, Leavitt BR, Kuhn R, Tabrizi SJ, Macdonald D, Weiss A. 2015. Quantification of mutant huntingtin protein in cerebrospinal fluid from Huntington’s disease patients. The Journal of Clinical Investigation.

Yang Y, Arseni D, Zhang W, Huang M, Lövestam S, Schweighauser M, Kotecha A, Murzin AG, Peak-Chew SY, Macdonald J, Lavenir I, Garringer HJ, Gelpi E, Newell KL, Kovacs GG, Vidal R, Ghetti B, Ryskeldi-Falcon B, Scheres SHW, Goedert M. 2022. Cryo-EM structures of amyloid-β 42 filaments from human brains. Science 375:167–172. doi:10.1126/science.abm7285

Zhang X, Zhang W, Wagener KB, Boz E, Alamo RG. 2018. Effect of Self-Poisoning on Crystallization Kinetics of Dimorphic Precision Polyethylenes with Bromine. Macromolecules 51:1386–1397. doi:10.1021/acs.macromol.7b02745

Author response:

Reviewer #1 (Public Review):

Summary:

In this manuscript, the authors investigated the effect of chronic activation of dopamine neurons using chemogenetics. Using Gq-DREADDs, the authors chronically activated midbrain dopamine neurons and observed that these neurons, particularly their axons, exhibit increased vulnerability and degeneration, resembling the pathological symptoms of Parkinson's disease. Baseline calcium levels in midbrain dopamine neurons were also significantly elevated following the chronic activation. Lastly, to identify cellular and circuit-level changes in response to dopaminergic neuronal degeneration caused by chronic activation, the authors employed spatial genomics (Visium) and revealed comprehensive changes in gene expression in the mouse model subjected to chronic activation. In conclusion, this study presents novel data on the consequences of chronic hyperactivation of midbrain dopamine neurons.

Strengths:

This study provides direct evidence that the chronic activation of dopamine neurons is toxic and gives rise to neurodegeneration. In addition, the authors achieved the chronic activation of dopamine neurons using water application of clozapine-N-oxide (CNO), a method not commonly employed by researchers. This approach may offer new insights into pathophysiological alterations of dopamine neurons in Parkinson's disease. The authors also utilized state-of-the-art spatial gene expression analysis, which can provide valuable information for other researchers studying dopamine neurons. Although the authors did not elucidate the mechanisms underlying dopaminergic neuronal and axonal death, they presented a substantial number of intriguing ideas in their discussion, which are worth further investigation.

We thank the reviewer for these positive comments.

Weaknesses:

Many claims raised in this paper are only partially supported by the experimental results. So, additional data are necessary to strengthen the claims. The effects of chronic activation of dopamine neurons are intriguing; however, this paper does not go beyond reporting phenomena. It lacks a comprehensive explanation for the degeneration of dopamine neurons and their axons. While the authors proposed possible mechanisms for the degeneration in their discussion, such as differentially expressed genes, these remain experimentally unexplored.

We thank the reviewer for this review. We do believe that the manuscript has a mechanistic component, as the central experiments involve direct manipulation of neuronal activity, and we show an increase in calcium levels and gene expression changes in dopamine neurons that coincide with the degeneration. However, we agree that deeper mechanistic investigation would strengthen the conclusions of the paper. We have planned several important revisions, including the addition of CNO behavioral controls, manipulation of intracellular calcium using isradipine, additional transcriptomics experiments and further validation of findings. We anticipate that these additions will significantly bolster the conclusions of the paper.

Reviewer #2 (Public Review):

Summary:

Rademacher et al. present a paper showing that chronic chemogenetic excitation of dopaminergic neurons in the mouse midbrain results in differential degeneration of axons and somas across distinct regions (SNc vs VTA). These findings are important. This mouse model also has the advantage of showing a axon-first degeneration over an experimentally-useful time course (2-4 weeks). 2. The findings that direct excitation of dopaminergic neurons causes differential degeneration sheds light on the mechanisms of dopaminergic neuron selective vulnerability. The evidence that activation of dopaminergic neurons causes degeneration and alters mRNA expression is convincing, as the authors use both vehicle and CNO control groups, but the evidence that chronic dopaminergic activation alters circadian rhythm and motor behavior is incomplete as the authors did not run a CNO-control condition in these experiments.

Strengths:

This is an exciting and important paper.

The paper compares mouse transcriptomics with human patient data.

It shows that selective degeneration can occur across the midbrain dopaminergic neurons even in the absence of a genetic, prion, or toxin neurodegeneration mechanism.

We thank the reviewer for these insightful comments.

Weaknesses:

Major concerns:

(1) The lack of a CNO-positive, DREADD-negative control group in the behavioral experiments is the main limitation in interpreting the behavioral data. Without knowing whether CNO on its own has an impact on circadian rhythm or motor activity, the certainty that dopaminergic hyperactivity is causing these effects is lacking.

This is an important point. Although we show that CNO does not produce degeneration of DA neuron terminals, we do not exclude a contribution to the behavioral changes. We agree that this behavioral control is necessary, and will address it in revision with a CNO-only running wheel cohort.

(2) One of the most exciting things about this paper is that the SNc degenerates more strongly than the VTA when both regions are, in theory, excited to the same extent. However, it is not perfectly clear that both regions respond to CNO to the same extent. The electrophysiological data showing CNO responsiveness is only conducted in the SNc. If the VTA response is significantly reduced vs the SNc response, then the selectivity of the SNc degeneration could just be because the SNc was more hyperactive than the VTA. Electrophysiology experiments comparing the VTA and SNc response to CNO could support the idea that the SNc has substantial intrinsic vulnerability factors compared to the VTA.

We agree that additional electrophysiology conducted in the VTA dopamine neurons would meaningfully add to our understanding of the selective vulnerability in this model, and will complete these experiments in revision.

(3) The mice have access to a running wheel for the circadian rhythm experiments. Running has been shown to alter the dopaminergic system (Bastioli et al., 2022) and so the authors should clarify whether the histology, electrophysiology, fiber photometry, and transcriptomics data are conducted on mice that have been running or sedentary.

We will explicitly clarify which mice had access to a running wheel in our revision. Briefly, mice for histology, electrophysiology, and transcriptomics all had access to a running wheel during their treatment. The mice used for photometry underwent about 7 days of running wheel access approximately 3 weeks prior to the beginning of the experiment. The photometry headcaps sterically prevented mice from having access to a running wheel in their home cage.

Reviewer #3 (Public Review):

Summary:

In this manuscript, Rademacher and colleagues examined the effect on the integrity of the dopamine system in mice of chronically stimulating dopamine neurons using a chemogenetic approach. They find that one to two weeks of constant exposure to the chemogenetic activator CNO leads to a decrease in the density of tyrosine hydroxylase staining in striatal brain sections and to a small reduction of the global population of tyrosine hydroxylase positive neurons in the ventral midbrain. They also report alterations in gene expression in both regions using a spatial transcriptomics approach. Globally, the work is well done and valuable and some of the conclusions are interesting. However, the conceptual advance is perhaps a bit limited in the sense that there is extensive previous work in the literature showing that excessive depolarization of multiple types of neurons associated with intracellular calcium elevations promotes neuronal degeneration. The present work adds to this by showing evidence of a similar phenomenon in dopamine neurons.

We thank the reviewer for the careful and thoughtful review of our manuscript.

While extensive depolarization and associated intracellular calcium elevations promotes degeneration generally, we emphasize that the process we describe is novel. Indeed, prior studies delivering chronic DREADDs to vulnerable neurons in models of Alzheimer’s disease did not report an increase in neurodegeneration, despite seeing changes in protein aggregation (e.g. Yuan and Grutzendler, J Neurosci 2016, PMID: 26758850; Hussaini et al., PLOS Bio 2020, PMID: 32822389). Further, a critical finding from our study is that in our paradigm, this stressor does not impact all dopamine neurons equally, as the SNc DA neurons are more vulnerable than the VTA, mirroring selective vulnerability characteristic of Parkinson’s disease. This is consistent with a large body of literature that SNc dopamine neurons are less capable of handling large energetic and calcium loads compared to neighboring VTA neurons, and the finding that chronically altered activity is sufficient to drive this preferential loss is novel.

In addition, we are not aware of prior studies that have chronically activated DREADDs to produce neurodegeneration. Other studies have shown that acute excitotoxic stressors can produce neuronal degeneration, but the chronic increase in activity is central to our approach.

In terms of the mechanisms explaining the neuronal loss observed after 2 to 4 weeks of chemogenetic activation, it would be important to consider that dopamine neurons are known from a lot of previous literature to undergo a decrease in firing through a depolarization-block mechanism when chronically depolarized. Is it possible that such a phenomenon explains much of the results observed in the present study? It would be important to consider this in the manuscript.

As discussed in greater detail in the results section below, our data suggests this may not be a prominent feature in our model. However, we cannot rule out a contribution of depolarization block, and will expand on the discussion of this possibility in the revised manuscript.

The relevance to Parkinson's disease (PD) is also not totally clear because there is not a lot of previous solid evidence showing that the firing of dopamine neurons is increased in PD, either in human subjects or in mouse models of the disease. As such, it is not clear if the present work is really modelling something that could happen in PD in humans.

We completely agree that evidence of increased dopamine neuron activity from human PD patients is lacking and the existing data are difficult to interpret without human controls. However, as we outline in the manuscript, multiple lines of evidence suggest that the activity level of dopamine neurons almost certainly does change in PD. Therefore, it is very important that we understand how changes in the level of neural activity influence the degeneration of DA neurons. In this paper we examine the impact of increased activity. Increased activity may be compensatory after initial dopamine neuron loss, or may be an initial driver of death (Rademacher & Nakamura, Exp Neurol 2024, PMID: 38092187). Beyond what is already discussed in the manuscript, additional support for increased activity in PD models include:

- Elevated firing rates in asymptomatic MitoPark mice (Good et al., FASEB J 2011, PMID: 21233488)

- Increased frequency of spontaneous firing in patient-derived iPSC dopamine neurons and primary mouse dopamine neurons that overexpress synuclein (Lin et al., Acta Neuropath Comm 2021, PMID: 34099060)

- Increased spontaneous firing in dopamine neurons of rats injected with synuclein preformed fibrils compared to sham (Tozzi et al., Brain 2021, PMID: 34297092)

We will include and further discuss these important examples in our revision.

Similarly, in future studies, it will also be important to study the impact of decreasing DA neuron activity. There will be additional levels of complexity to accurately model changes in PD, which may differ between subtypes of the disease, the disease stage, and the subtype of dopamine neuron. Our study models the possibility of chronically increased pacemaking, and interpretation of our results will be informed as we learn more about how the activity of DA neurons changes in humans in PD. We will discuss and elaborate on these important points in the revision.

Comments on the introduction:

The introduction cites a 1990 paper from the lab of Anthony Grace as support of the fact that DA neurons increase their firing rate in PD models. However, in this 1990 paper, the authors stated that: "With respect to DA cell activity, depletions of up to 96% of striatal DA did not result in substantial alterations in the proportion of DA neurons active, their mean firing rate, or their firing pattern. Increases in these parameters only occurred when striatal DA depletions exceeded 96%." Such results argue that an increase in firing rate is most likely to be a consequence of the almost complete loss of dopamine neurons rather than an initial driver of neuronal loss. The present introduction would thus benefit from being revised to clarify the overriding hypothesis and rationale in relation to PD and better represent the findings of the paper by Hollerman and Grace.

We agree that the findings of Hollerman and Grace support compensatory changes in dopamine neuron activity in response to loss of dopamine neurons, rather than informing whether dopamine neuron loss can also be an initial driver of activity. We will clarify this point in our revision. In addition, the results of other studies on this point are mixed: a 50% reduction in dopamine neurons didn’t alter firing rate or bursting (Harden and Grace, J Neurosci 1995, PMID: 7666198; Bilbao et al, Brain Res 2006, PMID: 16574080), while a 40% loss was found to increase firing rate and bursting (Chen et al, Brain Res 2009. PMID: 19545547) and larger reductions alter burst firing (Hollerman & Grace, Brain Res 1990, PMID: 2126975; Stachowiak et al, J Neurosci 1987, PMID: 3110381). Importantly, even if compensatory, such late-stage increases in dopamine neuron activity may contribute to disease progression and drive a vicious cycle of degeneration in surviving neurons. In addition, we also don’t know how the threshold of dopamine neuron loss and altered activity may differ between mice and humans, and PD patients do not present with clinical symptoms until ~30-60% of nigral neurons are lost (Burke & O’Malley, Exp Neurol 2013, PMID: 22285449; Shulman et al, Annu Rev Pathol 2011, PMID: 21034221).

Other lines of evidence support the potential role of hyperactivity in disease initiation, including increased activity before dopamine neuron loss in MitoPark mice (Good et al., FASEB J 2011, PMID: 21233488), increased spontaneous firing in patient-derived iPSC dopamine neurons (Lin et al., Acta Neuropath Comm 2021, PMID: 34099060), and increased activity observed in genetic models of PD (Bishop et al., J Neurophysiol 2010, PMID: 20926611; Regoni et al., Cell Death Dis 2020,  PMID: 33173027).

It would be good that the introduction refers to some of the literature on the links between excessive neuronal activity, calcium, and neurodegeneration. There is a large literature on this and referring to it would help frame the work and its novelty in a broader context.

We agree that a discussion of hyperactivity, calcium, and neurodegeneration would benefit the introduction. While we briefly discuss calcium and neurodegeneration in the discussion, we will expand on this literature in both the introduction and discussion sections. We will carefully review and contextualize our work within existing frameworks of calcium and neurodegeneration (e.g. Surmeier & Schumacker, J Biol Chem 2013, PMID: 23086948; Verma et al., Transl Neurodegener 2022, PMID: 35078537). We believe that the novelty of our study lies in 1) a chronic chemogenetic activation paradigm via drinking water, 2) demonstrating selective vulnerability of dopamine neurons as a result of altering their activity/excitability alone, and 3) comparing mouse and human spatial transcriptomics.

Comments on the results section:

The running wheel results of Figure 1 suggest that the CNO treatment caused a brief increase in running on the first day after which there was a strong decrease during the subsequent days in the active phase. This observation is also in line with the appearance of a depolarization block.

The authors examined many basic electrophysiological parameters of recorded dopamine neurons in acute brain slices. However, it is surprising that they did not report the resting membrane potential, or the input resistance. It would be important that this be added because these two parameters provide key information on the basal excitability of the recorded neurons. They would also allow us to obtain insight into the possibility that the neurons are chronically depolarized and thus in depolarization block.

We do report the input resistance in Supplemental Figure 1C, which was unchanged in CNO-treated animals compared to controls. We did not report the resting membrane potential because many of the DA neurons were spontaneously firing. However, we will report the initial membrane potential on first breaking into the cell for the whole cell recordings in the revision, which did not vary between groups. This is still influenced by action potential activity, but is the timepoint in the recording least impacted by dialyzing of the neuron by the internal solution. We observed increased spontaneous action potential activity ex vivo in slices from CNO-treated mice (Figure 1D), thus at least under these conditions these dopamine neurons are not in depolarization block. We also did not see strong evidence of changes in other intrinsic properties of the neurons with whole cell recordings (e.g. Figure S1C). Overall, our electrophysiology experiments are not consistent with the depolarization block model, at least not due to changes in the intrinsic properties of the neurons. Although our ex vivo findings cannot exclude a contribution of depolarization block in vivo, we do show that CNO-treated mice removed from their cages for open field testing continue to have a strong trend for increased activity for approximately 10 days (S1E).  This finding is also consistent with increased activity of the DA neurons. We will add discussion of these important considerations in the revision.

It is great that the authors quantified not only TH levels but also the levels of mCherry, co-expressed with the chemogenetic receptor. This could in principle help to distinguish between TH downregulation and true loss of dopamine neuron cell bodies. However, the approach used here has a major caveat in that the number of mCherry-positive dopamine neurons depends on the proportion of dopamine neurons that were infected and expressed the DREADD and this could very well vary between different mice. It is very unlikely that the virus injection allowed to infect 100% of the neurons in the VTA and SNc. This could for example explain in part the mismatch between the number of VTA dopamine neurons counted in panel 2G when comparing TH and mCherry counts. Also, I see that the mCherry counts were not provided at the 2-week time point. If the mCherry had been expressed genetically by crossing the DAT-Cre mice with a floxed fluorescent reported mice, the interpretation would have been simpler. In this context, I am not convinced of the benefit of the mCherry quantifications. The authors should consider either removing these results from the final manuscript or discussing this important limitation.

We thank the reviewer for this insightful comment, and we agree that this is a caveat of our mCherry quantification. Quantitation of the number of mCherry+ DA neurons specifically informs the impact on transduced DA neurons, and mCherry appears to be less susceptible to downregulation versus TH. As the reviewer points out, it carries the caveat that there is some variability between injections. Nonetheless, we believe that it conveys useful complementary data. As suggested, we will discuss this caveat in our revision. Note that mCherry was not quantified at the two-week timepoint because there is no loss of TH+ cells at that time.

Although the authors conclude that there is a global decrease in the number of dopamine neurons after 4 weeks of CNO treatment, the post-hoc tests failed to confirm that the decrease in dopamine number was significant in the SNc, the region most relevant to Parkinson's. This could be due to the fact that only a small number of mice were tested. A "n" of just 4 or 5 mice is very small for a stereological counting experiment. As such, this experiment was clearly underpowered at the statistical level. Also, the choice of the image used to illustrate this in panel 2G should be reconsidered: the image suggests that a very large loss of dopamine neurons occurred in the SNc and this is not what the numbers show. A more representative image should be used.

We agree that the stereology experiments were performed on relatively small numbers of animals. Combined with the small effect size, this may have contributed to the post-hoc tests showing a trend of p=0.1 for both the TH and mCherry dopamine cell counts in the SN at 4 weeks. As part of the planned experiments for our revision, we will perform an additional stereologic analysis to further assess the loss of SNc dopamine neurons. We will also review and ensure the images are representative.

In Figure 3, the authors attempt to compare intracellular calcium levels in dopamine neurons using GCaMP6 fluorescence. Because this calcium indicator is not quantitative (unlike ratiometric sensors such as Fura2), it is usually used to quantify relative changes in intracellular calcium. The present use of this probe to compare absolute values is unusual and the validity of this approach is unclear. This limitation needs to be discussed. The authors also need to refer in the text to the difference between panels D and E of this figure. It is surprising that the fluctuations in calcium levels were not quantified. I guess the hypothesis was that there should be more or larger fluctuations in the mice treated with CNO if the CNO treatment led to increased firing. This needs to be clarified.

We thank the reviewer for this comment. We understand that this method of comparing absolute values is unconventional. However, these animals were tested concurrently on the same system, and a clear effect on the absolute baseline was observed. We will include a caveat of this in our discussion. Panel D of this figure shows the raw, uncorrected photometry traces, whereas panel E shows the isosbestic corrected traces for the same recording. In panel E, the traces follow time in ascending order. We will also include frequency and amplitude data for these recordings.   

Although the spatial transcriptomic results are intriguing and certainly a great way to start thinking about how the CNO treatment could lead to the loss of dopamine neurons, the presented results, the focusing of some broad classes of differentially expressed genes and on some specific examples, do not really suggest any clear mechanism of neurodegeneration. It would perhaps be useful for the authors to use the obtained data to validate that a state of chronic depolarization was indeed induced by the chronic CNO treatment. Were genes classically linked to increased activity like cfos or bdnf elevated in the SNc or VTA dopamine neurons? In the striatum, the authors report that the levels of DARP32, a gene whose levels are linked to dopamine levels, are unchanged. Does this mean that there were no major changes in dopamine levels in the striatum of these mice?

We will review the expression of activity-related genes in our dataset, although we must keep in mind that these genes may behave differently in the context of chronic activation as opposed to acutely increased activity. We will also include experiments assessing striatal dopamine levels by HPLC in the revision.

The usefulness of comparing the transcriptome of human PD SNc or VTA sections to that of the present mouse model should be better explained. In the human tissues, the transcriptome reflects the state of the tissue many years after extensive loss of dopamine neurons. It is expected that there will be few if any SNc neurons left in such sections. In comparison, the mice after 7 days of CNO treatment do not appear to have lost any dopamine neurons. As such, how can the two extremely different conditions be reasonably compared?

Our mouse model and human PD progress over distinct timescales, as is the case with essentially all mouse models of neurodegenerative diseases. Nonetheless, in our view there is still great value in comparing gene expression changes in mouse models with those in human disease. It seems very likely that the same pathologic processes that drive degeneration early in the disease continue to drive degeneration later in the disease. Note that we have tried to address the discrepancy in time scales in part by comparing to early PD samples when there is more limited SNc DA neuron loss. Please note the numbers of DA neurons within the areas we have selected for sampling (Figure at right). Therefore, we can indeed use spatial transcriptomics to compare dopamine neurons from mice with initial degeneration and patients where degeneration is ongoing during their disease.

Author response image 1.

Violin plot of DA neuron proportions sampled within the vulnerable SNV (deconvoluted RCTD method used in unmasked tissue sections of the SNV). Control and early PD subjects.

Comments on the discussion:

In the discussion, the authors state that their calcium photometry results support a central role of calcium in activity-induced neurodegeneration. This conclusion, although plausible because of the very broad pre-existing literature linking calcium elevation (such as in excitotoxicity) to neuronal loss, should be toned down a bit as no causal relationship was established in the experiments that were carried out in the present study.

Our model utilizes hM3Dq-DREADDs that function by increasing intracellular calcium to increase neuronal excitability, and our results show increased Ca2+ by fiber photometry and changes to Ca2+-related genes, strongly suggesting a causal relation and crucial role of calcium in the mechanism of degeneration. However, we agree that we have not experimentally proven this point, as we acknowledged in the text. Additionally, we have planned revision experiments involving chronic isradipine treatment to further test the role of calcium in the mechanism of degeneration in this model.

In the discussion, the authors discuss some of the parallel changes in gene expression detected in the mouse model and in the human tissues. Because few if any dopamine neurons are expected to remain in the SNc of the human tissues used, this sort of comparison has important conceptual limitations and these need to be clearly addressed.

As discussed, we can sample SN DA neurons in early PD (see figure above), and in our view there is great value for such comparisons. We agree that discussion of appropriate caveats is warranted and this will be clearly addressed in the revision.

A major limitation of the present discussion is that it does not discuss the possibility that the observed phenotypes are caused by the induction of a chronic state of depolarization block by the chronic CNO treatment. I encourage the authors to consider and discuss this hypothesis.

As discussed above, our analyses of DA neuron firing in slices and open field testing to date do not support a prominent contribution of depolarization block with chronic CNO treatment. However, we cannot rule out this hypothesis, therefore we will include additional electrophysiology experiments and add discussion of this important consideration.  

Also, the authors need to discuss the fact that previous work was only able to detect an increase in the firing rate of dopamine neurons after more than 95% loss of dopamine neurons. As such, the authors need to clearly discuss the relevance of the present model to PD. Are changes in firing rate a driver of neuronal loss in PD, as the authors try to make the case here, or are such changes only a secondary consequence of extensive neuronal loss (for example because a major loss of dopamine would lead to reduced D2 autoreceptor activation in the remaining neurons, and to reduced autoreceptor-mediated negative feedback on firing). This needs to be discussed.

As discussed above, while increases in dopamine neuron activity may be compensatory after loss of neurons, the precise percentage required to induce such compensatory changes is not defined in mice and varies between paradigms, and the threshold level is not known in humans. We also reiterate that a compensatory increase in activity could still promote the degeneration of critical surviving DA neurons, whose loss underlies the substantial decline in motor function that typically occurs over the course of PD. Moreover, there are also multiple lines of evidence to suggest that changes in activity can initiate and drive dopamine neuron degeneration (Rademacher & Nakamura, Exp Neurol 2024). For example, overexpression of synuclein can increase firing in cultured dopamine neurons (Dagra et al., NPJ Parkinsons Dis 2021, PMID: 34408150) while mice expressing mutant Parkin have higher mean firing rates (Regoni et al., Cell Death Dis 2020,  PMID: 33173027). Similarly, an increased firing rate has been reported in the MitoPark mouse model of PD at a time preceding DA neuron degeneration (Good et al., FASEB J 2011, PMID: 21233488). We also acknowledge that alterations to dopamine neuron activity are likely complex in PD, and that dopamine neuron health and function can be impacted not just by simple increases in activity, but also by changes in activity patterns and regularity. We will amend our discussion to include the important caveat of changes in activity occurring as compensation, as well as further evidence of changes in activity preceding dopamine neuron death.

There is a very large, multi-decade literature on calcium elevation and its effects on neuronal loss in many different types of neurons. The authors should discuss their findings in this context and refer to some of this previous work. In a nutshell, the observations of the present manuscript could be summarized by stating that the chronic membrane depolarization induced by the CNO treatment is likely to induce a chronic elevation of intracellular calcium and this is then likely to activate some of the well-known calcium-dependent cell death mechanisms. Whether such cell death is linked in any way to PD is not really demonstrated by the present results. The authors are encouraged to perform a thorough revision of the discussion to address all of these issues, discuss the major limitations of the present model, and refer to the broad pre-existing literature linking membrane depolarization, calcium, and neuronal loss in many neuronal cell types.

While our model demonstrates classic excitotoxic cell death pathways, we would like to emphasize both the chronic nature of our manipulation and the progressive changes observed, with increasing degeneration seen at 1, 2, and 4 weeks of hyperactivity in an axon-first manner. This is a unique aspect of our study, in contrast to much of the previous literature which has focused on shorter timescales. Thus, while we will revise the discussion to more comprehensively acknowledge previous studies of calcium-dependent neuron cell death, we believe we have made several new contributions that are not predicted by existing literature. We have shown that this chronic manipulation is specifically toxic to nigral dopamine neurons, and the data that VTA dopamine neurons continue to be resilient even at 4 weeks is interesting and disease-relevant. We therefore do not want to use findings from other neuron types to draw assumptions about DA neurons, which are a unique and very diverse population. We acknowledge that as with all preclinical models of PD, we cannot draw definitive conclusions about PD with this data. However, we reiterate that we strongly believe that drawing connections to human disease is important, as dopamine neuron activity is very likely altered in PD and a clearer understanding of how dopamine neuron survival is impacted by activity will provide insight into the mechanisms of PD.

Author response:

Reviewer #1 (Public review):

From the Reviewing Editor:

Four reviewers have assessed your manuscript on valence and salience signaling in the central amygdala. There was universal agreement that the question being asked by the experiment is important. There was consensus that the neural population being examined (GABA neurons) was important and the circular shift method for identifying task-responsive neurons was rigorous. Indeed, observing valenced outcome signaling in GABA neurons would considerably increase the role the central amygdala in valence. However, each reviewer brought up significant concerns about the design, analysis and interpretation of the results. Overall, these concerns limit the conclusions that can be drawn from the results. Addressing the concerns (described below) would work towards better answering the question at the outset of the experiment: how does the central amygdala represent salience vs valence.

A weakness noted by all reviewers was the use of the terms 'valence' and 'salience' as well as the experimental design used to reveal these signals. The two outcomes used emphasized non-overlapping sensory modalities and produced unrelated behavioral responses. Within each modality there are no manipulations that would scale either the value of the valenced outcomes or the intensity of the salient outcomes. While the food outcomes were presented many times (20 times per session over 10 sessions of appetitive conditioning) the shock outcomes were presented many fewer times (10 times in a single session). The large difference in presentations is likely to further distinguish the two outcomes. Collectively, these experimental design decisions meant that any observed differences in central amygdala GABA neuron responding are unlikely to reflect valence, but likely to reflect one or more of the above features.

We appreciate the reviewers’ comments regarding the experimental design. When assessing fear versus reward, we chose stimuli that elicit known behavioral responses, freezing versus consumption. The use of stimuli of the same modality is unlikely to elicit easily definable fear or reward responses or to be precisely matched for sensory intensity. For example, sweet or bitter tastes can be used, but even these activate different taste receptors and vary in the duration of the activation of taste-specific signaling (e.g. how long the taste lingers in the mouth). The approach we employed is similar to that of Yang et al., 2023 (doi: 10.1038/s41586-023-05910-2) that used water reward and shock to characterize the response profiles of somatostatin neurons of the central amygdala. Similar to what was reported by Yang and colleagues we observed that the majority of CeA GABA neurons responded selectively to one unconditioned stimulus (~52%). We observed that 15% of neurons responded in the same direction, either activated or inhibited, by the food or shock US. These were defined as salience based on the definitions of Lin and Nicolelis, 2008 (doi: 10.1016/j.neuron.2008.04.031) in which basal forebrain neurons responded similarly to reward or punishment irrespective of valence. The designation of valence encoding based opposite responses to the food or shock is straightforward (~10% of cells); however, we agree that the designation of modality-specific encoding neurons as valence encoding is less straightforward.

A second weakness noted by a majority of reviewers was a lack of cue-responsive unit and a lack of exploration of the diversity of response types, and the relationship cue and outcome firing. The lack of large numbers of neurons increasing firing to one or both cues is particularly surprising given the critical contribution of central amygdala GABA neurons to the acquisition of conditioned fear (which the authors measured) as well as to conditioned orienting (which the authors did not measure). Regression-like analyses would be a straightforward means of identifying neurons varying their firing in accordance with these or other behaviors. It was also noted that appetitive behavior was not measured in a rigorous way. Instead of measuring time near hopper, measures of licking would have been better. Further, measures of orienting behaviors such as startle were missing.

The authors also missed an opportunity for clustering-like analyses which could have been used to reveal neurons uniquely signaling cues, outcomes or combinations of cues and outcomes. If the authors calcium imaging approach is not able to detect expected central amygdala cue responding, might it be missing other critical aspects of responding?

As stated in the manuscript, we were surprised by the relatively low number of cue responsive cells; however, when using a less stringent statistical method (Figure 5 - Supplement 2), we observed 13% of neurons responded to the food associated cue and 23% responded to the shock associated cue. The differences are therefore likely a reflection of the rigor of the statistical measure to define the responsive units. The number of CS responsive units is less than reported in the CeAl by Ciocchi et al., 2010 (doi: 10.1038/nature09559 ) who observed 30% activated by the CS and 25% inhibited, but is not that dissimilar from the results of Duvarci et al., 2011 (doi: 10.1523/JNEUROSCI.4985-10.2011 ) who observed 11% activated in the CeAl and 25% inhibited by the CS. These numbers are also consistent with previous single cell calcium imaging of cell types in the CeA. For example, Yang et al., 2023 (doi: 10.1038/s41586-023-05910-2) observed that 13% of somatostatin neurons responded to a reward CS and 8% responded to a shock CS. Yu et al., 2017 (doi: 10.1038/s41593-017-0009-9) observed 26.5% of PKCdelta neurons responded to the shock CS. It should also be noted that our analysis was not restricted to the CeAl. Finally, Food learning was assessed in an operant chamber in freely moving mice with reward pellet delivery. Because liquids were not used for the reward US, licking is not a metric that can be used.

All reviewers point out that the evidence for salience encoding is even more limited than the evidence for valence. Although the specific concern for each reviewer varied, they all centered on an oversimplistic definition of salience. Salience ought to scale with the absolute value and intensity of the stimulus. Salience cannot simply be responding in the same direction. Further, even though the authors observed subsets of central amygdala neurons increasing or decreasing activity to both outcomes - the outcomes can readily be distinguished based on the temporal profile of responding.

We thank the reviewers for their comments relating to the definition of salience and valence encoding by central amygdala neurons. We have addressed each of the concerns below.

Additional concerns are raised by each reviewer. Our consensus is that this study sought to answer an important question - whether central amygdala signal salience or valence in cue-outcome learning. However, the experimental design, analyses, and interpretations do not permit a rigorous and definitive answer to that question. Such an answer would require additional experiments whose designs would address the significant concerns described here. Fully addressing the concerns of each reviewer would result in a re-evaluation of the findings. For example, experimental design better revealing valence and salience, and analyses describing diversity of neuronal responding and relationship to behavior would likely make the results Important or even Fundamental.

We appreciate the reviewers’ comments and have addressed each concern below.

Reviewer #2 (Public review):

In this article, Kong and authors sought to determine the encoding properties of central amygdala (CeA) neurons in response to oppositely valenced stimuli and cues predicting those stimuli. The amygdala and its subregional components have historically been understood to be regions that encode associative information, including valence stimuli. The authors performed calcium imaging of GABA-ergic CeA neurons in freely-moving mice conditioned in Pavlovian appetitive and fear paradigms, and showed that CeA neurons are responsive to both appetitive and aversive unconditioned and conditioned stimuli. They used a variant of a previously published 'circular shifting' technique (Harris, 2021), which allowed them to delineate between excited/non-responsive/inhibited neurons. While there is considerable overlap of CeA neurons responding to both unconditioned stimuli (in this case, food and shock, deemed "salience-encoding" neurons), there are considerably fewer CeA neurons that respond to both conditioned stimuli that predict the food and shock. The authors finally demonstrated that there are no differences in the order of Pavlovian paradigms (fear - shock vs. shock - fear), which is an interesting result, and convincingly presented given their counterbalanced experimental design.

In total, I find the presented study useful in understanding the dynamics of CeA neurons during a Pavlovian learning paradigm. There are many strengths of this study, including the important question and clear presentation, the circular shifting analysis was convincing to me, and the manuscript was well written. We hope the authors will find our comments constructive if they choose to revise their manuscript.

While the experiments and data are of value, I do not agree with the authors interpretation of their data, and take issue with the way they used the terms "salience" and "valence" (and would encourage them to check out Namburi et al., NPP, 2016) regarding the operational definitions of salience and valence which differ from my reading of the literature. To be fair, a recent study from another group that reports experiments/findings which are very similar to the ones in the present study (Yang et al., 2023, describing valence coding in the CeA using a similar approach) also uses the terms valence and salience in a rather liberal way that I would also have issues with (see below). Either new experiments or revised claims would be needed here, and more balanced discussion on this topic would be nice to see, and I felt that there were some aspects of novelty in this study that could be better highlighted (see below).

One noteworthy point of alarm is that it seems as if two data panels including heatmaps are duplicated (perhaps that panel G of Figure 5-figure supplement 2 is a cut and paste error? It is duplicated from panel E and does not match the associated histogram).

We thank the reviewer for their insightful comments and assessment of the manuscript.

Major concerns:

(1) The authors wish to make claims about salience and valence. This is my biggest gripe, so I will start here.

(1a) Valence scales for positive and negative stimuli and as stated in Namburi et al., NPP, 2016 where we operationalize "valence" as having different responses for positive and negative values and no response for stimuli that are not motivational significant (neutral cues that do not predict an outcome). The threshold for claiming salience, which we define as scaling with the absolute value of the stimulus, and not responding to a neutral stimulus (Namburi et al., NPP, 2016; Tye, Neuron, 2018; Li et al., Nature, 2022) would require the lack of response to a neutral cue.

We appreciate the reviewer’s comment on the definitions of salience and valence and agree that there is not a consistent classification of these response types in the field. As stated above, we used the designation of salience encoding if the cells respond in the same direction to different stimuli regardless of the valence of the stimulus similar to what was described previously (Lin and Nicolelis, 2008, doi: 10.1016/j.neuron.2008.04.031). Similar definitions of salience have also been reported elsewhere (for examples see: Stephenson-Jones et al., 2020, doi: 10.1016/j.neuron.2019.12.006,  Zhu et al., 2018 doi: 10.1126/science.aat0481, and  Comoli et al., 2003, doi: 10.1038/nn1113P). Per the suggestion of the reviewer, we longitudinally tracked cells on the first day of Pavlovian reward conditioning the fear conditioning day. Although there were considerably fewer head entries on the first day of reward conditioning, we were able to identify 10 cells that were activated by both the food US and shock US. We compared the responses to the first five head entries and last head entries and the first 5 shocks and last five shocks. Consistent with what has been reported for salience encoding neurons in the basal forebrain (Lin and Nicolelis, 2008, doi: 10.1016/j.neuron.2008.04.031), we observed that the responses were highest when the US was most unexpected and decreased in later trials.

Author response image 1.

(1b) The other major issue is that the authors choose to make claims about the neural responses to the USs rather than the CSs. However, being shocked and receiving sucrose also would have very different sensorimotor representations, and any differences in responses could be attributed to those confounds rather than valence or salience. They could make claims regarding salience or valence with respect to the differences in the CSs but they should restrict analysis to the period prior to the US delivery.

Perhaps the reviewer missed this, but analysis of valence and salience encoding to the different CSs are presented in Figure 5G, Figure 5 -Supplement 1 C-D, and Figure 5 -Supplement 2 N-O. Analysis of CS responsiveness to CSFood and CSShock were analyzed during the conditioning sessions Figure 3E-F, Figure 4B-C, Figure 5 – Supplement 2J-O and Figure 5 – Supplement 3K-L, and during recall probe tests for both CSFood and CSShock, Figure 5 – Supplement 1C-J.

(1c) The third obstacle to using the terms "salience" or "valence" is the lack of scaling, which is perhaps a bigger ask. At minimum either the scaling or the neutral cue would be needed to make claims about valence or salience encoding. Perhaps the authors disagree - that is fine. But they should at least acknowledge that there is literature that would say otherwise.

(1d) In order to make claims about valence, the authors must take into account the sensory confound of the modality of the US (also mentioned in Namburi et al., 2016). The claim that these CeA neurons are indeed valence-encoding (based on their responses to the unconditioned stimuli) is confounded by the fact that the appetitive US (food) is a gustatory stimulus while the aversive US (shock) is a tactile stimulus.

We provided the same analysis for the US and CS. The US responses were larger and more prevalent, but similar types of encoding were observed for the CS. We agree that the food reward and the shock are very different sensory modalities. As stated above, the use of stimuli of the same modality is unlikely to elicit easily definable fear or reward responses or to be precisely matched for sensory intensity. We agree that the definition of cells that respond to only one stimulus is difficult to define in terms of valence encoding, as opposed to being specific for the sensory modality and without scaling of the stimulus it is difficult to fully address this issue. It should be noted however, that if the cells in the CeA were exclusively tuned to stimuli of different sensory modalities, we would expect to see a similar number of cells responding to the CS tones (auditory) as respond to the food (taste) and shock (somatosensory) but we do not. Of the cells tracked longitudinally 80% responded to the USs, with 65% of cells responding to food (activated or inhibited) and 44% responding to shock (activated or inhibited).

(2) Much of the central findings in this manuscript have been previously described in the literature. Yang et al., 2023 for instance shows that the CeA encodes salience (as demonstrated by the scaled responses to the increased value of unconditioned stimuli, Figure 1 j-m), and that learning amplifies responsiveness to unconditioned stimuli (Figure 2). It is nice to see a reproduction of the finding that learning amplifies CeA responses, though one study is in SST::Cre and this one in VGAT::cre - perhaps highlighting this difference could maximize the collective utility for the scientific community?

We agree that the analysis performed here is similar to what was conducted by Yang et al., 2023. With the major difference being the types of neurons sampled. Yang et al., imaged only somatostatin neurons were as we recorded all GABAergic cell types within the CeA. Moreover, because we imaged from 10 mice, we sampled neurons that ostensibly covered the entire dorsal to ventral extent of the CeA (Figure 1 – Supplement 1). Remarkably, we found that the vast majority of CeA neurons (80%) are responsive to food or shock. Within this 80% there are 8 distinct response profiles consistent with the heterogeneity of cell types within the CeA based on connectivity, electrophysiological properties, and gene expression. Moreover, we did not find any spatial distinction between food or shock responsive cells, with the responsive cell types being intermingled throughout the dorsal to ventral axis (Figure 5 – Supplement 3).

(3) There is at least one instance of copy-paste error in the figures that raised alarm. In the supplementary information (Figure 5- figure supplement 2 E;G), the heat maps for food-responsive neurons and shock-responsive neurons are identical. While this almost certainly is a clerical error, the authors would benefit from carefully reviewing each figure to ensure that no data is incorrectly duplicated.

We thank the reviewer for catching this error. It has been corrected.

(4) The authors describe experiments to compare shock and reward learning; however, there are temporal differences in what they compare in Figure 5. The authors compare the 10th day of reward learning with the 1st day of fear conditioning, which effectively represent different points of learning and retrieval. At the end of reward conditioning, animals are utilizing a learned association to the cue, which demonstrates retrieval. On the day of fear conditioning, animals are still learning the cue at the beginning of the session, but they are not necessarily retrieving an association to a learned cue. The authors would benefit from recording at a later timepoint (to be consistent with reward learning- 10 days after fear conditioning), to more accurately compare these two timepoints. Or perhaps, it might be easier to just make the comparison between Day 1 of reward learning and Day 1 of fear learning, since they must already have these data.

We agree that there are temporal differences between the food and shock US deliveries. This is likely a reflection of the fact that the shock delivery is passive and easily resolved based on the time of the US delivery, whereas the food responses are variable because they are dependent upon the consumption of the sucrose pellet. Because of these differences the kinetics of the responses cannot be accurately compared. This is why we restricted our analysis to whether the cells were food or shock responsive. Aside from reporting the temporal differences in the signals did not draw major conclusions about the differences in kinetics. In our experimental design we counterbalanced the animals that received fear conditioning firs then food conditioning, or food conditioning then fear conditioning to ensure that order effects did not influence the outcome of the study. It is widely known that Pavlovian fear conditioning can facilitate the acquisition of conditioned stimulus responses with just a single day of conditioning. In contrast, Pavlovian reward conditioning generally progresses more slowly. Because of this we restricted our analysis to the last day of reward conditioning to the first and only day of fear conditioning. However, as stated above, we compared the responses of neurons defined as salience during day 1 of reward conditioning and fear conditioning. As would be predicted based on previous definitions of salience encoding (Lin and Nicolelis, 2008, doi: 10.1016/j.neuron.2008.04.031), we observed that the responses were highest when the US was most unexpected

(5) The authors make a claim of valence encoding in their title and throughout the paper, which is not possible to make given their experimental design. However, they would greatly benefit from actually using a decoder to demonstrate their encoding claim (decoding performance for shock-food versus shuffled labels) and simply make claims about decoding food-predictive cues and shock-predictive cues. Interestingly, it seems like relatively few CeA neurons actually show differential responses to the food and shock CSs, and that is interesting in itself.

As stated above, valence and salience encoding were defined similar to what has been previously reported (Li et al., 2019, doi: 10.7554/eLife.41223; Yang et al., 2023, doi: 10.1038/s41586-023-05910-2; Huang et al., 2024, doi: 10.1038/s41586-024-07819; Lin and Nicolelis, 2008, doi: 10.1016/j.neuron.2008.04.031; Stephenson-Jones et al., 2020, doi: 10.1016/j.neuron.2019.12.006; Zhu et al., 2018, doi: 10.1126/science.aat0481; and Comoli et al., 2003, doi: 10.1038/nn1113P). Interestingly, many of these studies did not vary the US intensity.

Reviewer #3 (Public review):

Summary:

In their manuscript entitled Kong and colleagues investigate the role of distinct populations of neurons in the central amygdala (CeA) in encoding valence and salience during both appetitive and aversive conditioning. The study expands on the work of Yang et al. (2023), which specifically focused on somatostatin (SST) neurons of the CeA. Thus, this study broadens the scope to other neuronal subtypes, demonstrating that CeA neurons in general are predominantly tuned to valence representations rather than salience.

We thank the reviewer for their insightful comments and assessment of the manuscript.

Strengths:

One of the key strengths of the study is its rigorous quantitative approach based on the "circular-shift method", which carefully assesses correlations between neural activity and behavior-related variables. The authors' findings that neuronal responses to the unconditioned stimulus (US) change with learning are consistent with previous studies (Yang et al., 2023). They also show that the encoding of positive and negative valence is not influenced by prior training order, indicating that prior experience does not affect how these neurons process valence.

Weaknesses:

However, there are limitations to the analysis, including the lack of population-based analyses, such as clustering approaches. The authors do not employ hierarchical clustering or other methods to extract meaning from the diversity of neuronal responses they recorded. Clustering-based approaches could provide deeper insights into how different subpopulations of neurons contribute to emotional processing. Without these methods, the study may miss patterns of functional specialization within the neuronal populations that could be crucial for understanding how valence and salience are encoded at the population level.

We appreciate the reviewer’s comments regarding clustering-based approaches. In order to classify cells as responsive to the US or CS we chose to develop a statistically rigorous method for classifying cell response types. Using this approach, we were able to define cell responses to the US and CS. Importantly, we identified 8 distinct response types to the USs. It is not clear how additional clustering analysis would improve cell classifications.

Furthermore, while salience encoding is inferred based on responses to stimuli of opposite valence, the study does not test whether these neuronal responses scale with stimulus intensity-a hallmark of classical salience encoding. This limits the conclusions that can be drawn about salience encoding specifically.

As stated above, we used salience classifications similar to those previously described (Lin and Nicolelis, 2008, doi: 10.1016/j.neuron.2008.04.031; Stephenson-Jones et al., 2020, doi: 10.1016/j.neuron.2019.12.006; Zhu et al., 2018, doi: 10.1126/science.aat0481; and Comoli et al., 2003, doi: 10.1038/nn1113P). We agree that varying the stimulus intensity would provide a more rigorous assessment of salience encoding; however, several of the studies mentioned above classify cells as salience encoding without varying stimulus intensity. Additionally, the inclusion of recordings with varying US intensities on top of the Pavlovian reward and fear conditioning would further decrease the number of cells that can be longitudinally tracked and would likely decrease the number of cells that could be classified.

In sum, while the study makes valuable contributions to our understanding of CeA function, the lack of clustering-based population analyses and the absence of intensity scaling in the assessment of salience encoding are notable limitations.

Reviewer #4 (Public review):

Summary:

The authors have performed endoscopic calcium recordings of individual CeA neuron responses to food and shock, as well as to cues predicting food and shock. They claim that a majority of neurons encode valence, with a substantial minority encoding salience.

Strengths:

The use of endoscopic imaging is valuable, as it provides the ability to resolve signals from single cells, while also being able to track these cells across time. The recordings appear well-executed, and employ a sophisticated circular shifting analysis to avoid statistical errors caused by correlations between neighboring image pixels.

Weaknesses:

My main critique is that the authors didn't fully test whether neurons encode valence. While it is true that they found CeA neurons responding to stimuli that have positive or negative value, this by itself doesn't indicate that valence is the primary driver of neural activity. For example, they report that a majority of CeA neurons respond selectively to either the positive or negative US, and that this is evidence for "type I" valence encoding. However, it could also be the case that these neurons simply discriminate between motivationally relevant stimuli in a manner unrelated to valence per se. A simple test of this would be to check if neural responses generalize across more than one type of appetitive or aversive stimulus, but this was not done. The closest the authors came was to note that a small number of neurons respond to CS cues, of which some respond to the corresponding US in the same direction. This is relegated to the supplemental figures (3 and 4), and it is not noted whether the the same-direction CS-US neurons are also valence-encoding with respect to different USs. For example, are the neurons excited by CS-food and US-food also inhibited by shock? If so, that would go a long way toward classifying at least a few neurons as truly encoding valence in a generalizable way.

As stated above, valence and salience encoding were defined similar to what has been previously reported (Li et al., 2019, doi: 10.7554/eLife.41223; Yang et al., 2023, doi: 10.1038/s41586-023-05910-2; Huang et al., 2024, doi: 10.1038/s41586-024-07819; Lin and Nicolelis, 2008, doi: 10.1016/j.neuron.2008.04.031; Stephenson-Jones et al., 2020, doi: 10.1016/j.neuron.2019.12.006; Zhu et al., 2018, doi: 10.1126/science.aat0481; and Comoli et al., 2003, doi: 10.1038/nn1113P). As reported in Figure 5 and Figure 5 – Supplement 3, ~29% of CeA neurons responded to both food and shock USs (15% in the same direction and 13.5% in the opposite direction). In contrast, only 6 of 303 cells responded to both the CSfood and CSshock, all in the same direction.

A second and related critique is that, although the authors correctly point out that definitions of salience and valence are sometimes confused in the existing literature, they then go on themselves to use the terms very loosely. For example, the authors define these terms in such a way that every neuron that responds to at least one stimulus is either salience or valence-encoding. This seems far too broad, as it makes essentially unfalsifiable their assertion that the CeA encodes some mixture of salience and valence. I already noted above that simply having different responses to food and shock does not qualify as valence-encoding. It also seems to me that having same-direction responses to these two stimuli similarly does not quality a neuron as encoding salience. Many authors define salience as being related to the ability of a stimulus to attract attention (which is itself a complex topic). However, the current paper does not acknowledge whether they are using this, or any other definition of salience, nor is this explicitly tested, e.g. by comparing neural response magnitudes to any measure of attention.

As stated in response to reviewer 2, we longitudinally tracked cells on the first day of Pavlovian reward conditioning the fear conditioning day. Although there were considerably fewer head entries on the first day of reward conditioning, we were able to identify 10 cells that were activated by both the food US and shock US. We compared the responses to the first five head entries and last head entries and the first 5 shocks and last five shocks. Consistent with what has been reported for salience encoding neurons in the basal forebrain (Lin and Nicolelis, 2008, doi: 10.1016/j.neuron.2008.04.031), we observed that the responses were highest when the US was most unexpected and decreased in later trials.

The impression I get from the authors' data is that CeA neurons respond to motivationally relevant stimuli, but in a way that is possibly more complex than what the authors currently imply. At the same time, they appear to have collected a large and high-quality dataset that could profitably be made available for additional analyses by themselves and/or others.

Lastly, the use of 10 daily sessions of training with 20 trials each seems rather low to me. In our hands, Pavlovian training in mice requires considerably more trials in order to effectively elicit responses to the CS. I wonder if the relatively sparse training might explain the relative lack of CS responses?

It is possible that learning would have occurred more quickly if we had used greater than 20 trials per session. However, we routinely used 20-25 trials for Pavlovian reward conditioning (doi: 10.1073/pnas.1007827107; doi: 10.1523/JNEUROSCI.5532-12.2013; doi: 10.1016/j.neuron.2013.07.044; and doi: 10.1016/j.neuron.2019.11.024).

Author Response

Response to Reviewer 1:

Summary of what the author was trying to achieve: In this study, the author aimed to develop a method for estimating neuronal-type connectivity from transcriptomic gene expression data, specifically from mouse retinal neurons. They sought to develop an interpretable model that could be used to characterize the underlying genetic mechanisms of circuit assembly and connectivity.

Strengths: The proposed bilinear model draws inspiration from commonly implemented recommendation systems in the field of machine learning. The author presents the model clearly and addresses critical statistical limitations that may weaken the validity of the model such as multicollinearity and outliers. The author presents two formulations of the model for separate scenarios in which varying levels of data resolution are available. The author effectively references key work in the field when establishing assumptions that affect the underlying model and subsequent results. For example, correspondence between gene expression cell types and connectivity cell types from different references are clearly outlined in Tables 1-3. The model training and validation are sufficient and yield a relatively high correlation with the ground truth connectivity matrix. Seemingly valid biological assumptions are made throughout, however, some assumptions may reduce resolution (such as averaging over cell types), thus missing potentially important single-cell gene expression interactions.

Thank you for acknowledging the strengths of this work. The assumption to average gene expression data across individual cells within a given cell type was made in response to the inherent limitations of, for example, the mouse retina dataset, where individual cell-level connectivity and gene expression data are not profiled jointly (the second scenario in our paper). This approach was a necessary compromise to facilitate the analysis at the cell type level. However, in datasets where individual cell-level connectivity and gene expression data are matched, such as the C.elegans dataset referenced below, our model can be applied to achieve single-cell resolution (the first scenario in our paper), offering a more detailed understanding of genetic underpinnings in neuronal connectivity.

Weaknesses: The main results of the study could benefit from replication in another dataset beyond mouse retinal neurons, to validate the proposed method. Dimensionality reduction significantly reduces the resolution of the model and the PCA methodology employed is largely non-deterministic. This may reduce the resolution and reproducibility of the model. It may be worth exploring how the PCA methodology of the model may affect results when replicating. Figure 5, ’Gene signatures associated with the two latent dimensions’, lacks some readability and related results could be outlined more clearly in the results section. There should be more discussion on weaknesses of the results e.g. quantification of what connectivity motifs were not captured and what gene signatures might have been missed.

I value the suggestion of validating the propose method in another dataset. In response, I found the C.elegans dataset in the references the reviewer suggested below a good candidate for this purpose, and I plan to explore this dataset and incorporate findings in the revised manuscript. I understand the concerns regarding the PCA methodology and its potential impact on the model’s resolution and reproducibility. In response, alternative methods, such as regularization techniques, will be explored to address these issues. Additionally, I agree that enhancing the clarity and readability of Figure 5, as well as including a more comprehensive discussion of the model’s limitations, would significantly strengthen the manuscript.

The main weakness is the lack of comparison against other similar methods, e.g. methods presented in Barabási, Dániel L., and Albert-László Barabási. "A genetic model of the connectome." Neuron 105.3 (2020): 435-445. Kovács, István A., Dániel L. Barabási, and Albert-László Barabási. "Uncovering the genetic blueprint of the C. elegans nervous system." Proceedings of the National Academy of Sciences 117.52 (2020): 33570-33577. Taylor, Seth R., et al. "Molecular topography of an entire nervous system." Cell 184.16 (2021): 4329-4347.

Thank you for highlighting the importance of comparing our model with others, particularly those mentioned in your comments. After reviewing these papers, I find that our bilinear model aligns closely with the methods described, especially in [1, 2]. To see this, let’s start with Equation 1 in Kovács et al. [2]:

In this equation, B represents the connectivity matrix, while X denotes the gene expression patterns of individual neurons in C.elegans. The operator O is the genetic rule operator governing synapse formation, linking connectivity with individual neuronal expression patterns. It’s noteworthy that the work of Barabási and Barabási [1] explores a specific application of this framework, focusing on O for B that represents biclique motifs in the C.elegans neural network.

To identify the the operator O, the authors sought to minimize the squared residual error:

with regularization on O.

Adopting the notation from our bilinear model paper and using Z to represent the connectivity matrix, the above becomes

Coming back to the bilinear model formulation, the optimization problem, as formulated for the C.elegans dataset where individual neuron connectivity and gene expression are accessible, takes the form:

where we consider each neuron as a distinct neuronal type. In addition, we extend the dimensions of X and Y to encompass the entire set of neurons in C.elegans, with X = Y ∈ Rn×p, where n signifies the total number of neurons and p the number of genes. Accordingly, our optimization challenge evolves into:

Upon comparison with the earlier stated equation, it becomes clear that our approach aligns consistently with the notion of O = ABT. This effectively results in a decomposition of the genetic rule operator O. This decomposition extends beyond mere mathematical convenience, offering several substantial benefits reminiscent of those seen in the collaborative filtering of recommendation systems:

• Computational Efficiency: The primary advantage of this approach is its improvement in computational efficiency. For instance, solving for O ∈ Rp×p necessitates determining p2 entries. In contrast, solving for A ∈ Rp×d and B ∈ Rp×d involves determining only 2pd entries, where p is the number of genes, and d is the number of latent dimensions. Assuming the existence of a lower-dimensional latent space (d << p) that captures the essential variability in connectivity, resolving A and B becomes markedly more efficient than resolving O. Additionally, from a computational system design perspective, inferring the connectivity of a neuron allows for caching the latent embeddings of presynaptic neurons XA or postsynaptic neurons XB with a space complexity of O(nd). This is significantly more space-efficient than caching XO or OXT, which has a space complexity of O(np). This difference is particularly notable when dealing with large numbers of neurons, such as those in the entire mouse brain. The bilinear modeling approach thus enables effective handling of large datasets, simplifying the optimization problem and reducing computational load, thereby making the model more scalable and faster to execute.

• Interpretability: The separation into A for presynaptic features and B for postsynaptic features provides a clearer understanding of the distinct roles of pre- and post- synaptic neurons in forming the connection. By projecting the pre- and post- synaptic neurons into a shared latent space through XA and YB, one can identify meaningful representations within each axis, as exemplified in different motifs from the mouse retina dataset. The linear characteristics of A and B facilitate direct evaluation of each gene’s contribution to a latent dimension. This interpretability, offering insights into the genetic factors influencing synaptic connections, is beyond what O could provide itself.

• Flexibility and Adaptability: The bilinear model’s adaptability is another strength. Much like collaborative filtering, which can manage very different user and item features, our bilinear model can be tailored to synaptic partners with genetic data from varied sources. A potential application of this model is in deciphering the genetic correlates of long-range projectomic rules, where pre- and post-synaptic neurons are processed and sequenced separately, or even involving post-synaptic targets being brain regions with genetic information acquired through bulk sequencing. This level of flexibility also allows for model adjustments or extensions to incorporate other biological factors, such as proteomics, thereby broadening its utility across various research inquiries into the determinants of neuronal connectivity.

In the study by Taylor et al. [3], the authors introduced a generalization of differential gene expressions (DGE) analysis called network DGE (nDGE) to identify genetic determinants of synaptic connections. It focuses on genes co-expressed across pairs of neurons connected, compared with pairs without connection.

As the authors acknowledged in the method part of the paper, nDGE can only examine single genes co-expressed at synaptic terminals: "While the nDGE technique introduced here is a generalization of standard DGE, interrogating the contribution of pairs of genes in the formation and maintenance of synapses between pairs of neurons, nDGE can only account for a single co-expressed gene in either of the two synaptic terminals (pre/post)."

In contrast, the bilinear model offers a more comprehensive analysis by seeking a linear combination of gene expressions in both pre- and post-synaptic neurons. This model goes beyond the scope of examining individual co-expressed genes, as it incorporates different weights for the gene expressions of pre- and post-synaptic neurons. This feature of the bilinear model enables it to capture not only homogeneous but also complex and heterogeneous genetic interactions that are pivotal in synaptic connectivity. This highlights the bilinear model’s capability to delve into the intricate interactions of synaptic gene expression.

Appraisal of whether the author achieved their aims, and whether results support their conclusions: The author achieved their aims by recapitulating key connectivity motifs from single-cell gene expression data in the mouse retina. Furthermore, the model setup allowed for insight into gene signatures and interactions, however could have benefited from a deeper evaluation of the accuracy of these signatures. The author claims the method sets a new benchmark for single-cell transcriptomic analysis of synaptic connections. This should be more rigorously proven. (I’m not sure I can speak on the novelty of the method)

I value your appraisal. In response, additional validation of the bilinear model on a second dataset will be undertaken.

Discussion of the likely impact of the work on the field, and the utility of methods and data to the community : This study provides an understandable bilinear model for decoding the genetic programming of neuronal type connectivity. The proposed model leaves the door open for further testing and comparison with alternative linear and/or non-linear models, such as neural networkbased models. In addition to more complex models, this model can be built on to include higher resolution data such as more gene expression dimensions, different types of connectivity measures, and additional omics data.

Thank you for your positive assessment of the potential impact of the study.

Response to Reviewer 2:

Summary: In this study, Mu Qiao employs a bilinear modeling approach, commonly utilized in recommendation systems, to explore the intricate neural connections between different pre- and post-synaptic neuronal types. This approach involves projecting single-cell transcriptomic datasets of pre- and post-synaptic neuronal types into a latent space through transformation matrices. Subsequently, the cross-correlation between these projected latent spaces is employed to estimate neuronal connectivity. To facilitate the model training, connectomic data is used to estimate the ground-truth connectivity map. This work introduces a promising model for the exploration of neuronal connectivity and its associated molecular determinants. However, it is important to note that the current model has only been tested with Bipolar Cell and Retinal Ganglion Cell data, and its applicability in more general neuronal connectivity scenarios remains to be demonstrated.

Strengths: This study introduces a succinct yet promising computational model for investigating connections between neuronal types. The model, while straightforward, effectively integrates singlecell transcriptomic and connectomic data to produce a reasonably accurate connectivity map, particularly within the context of retinal connectivity. Furthermore, it successfully recapitulates connectivity patterns and helps uncover the genetic factors that underlie these connections.

Thank you for your positive assessment of the paper.

Weaknesses:

1. The study lacks experimental validation of the model’s prediction results.

Thank you for pointing out the importance of experimental validation. I acknowledge that the current version of the study is focused on the development and validation of the computational model, using the datasets presently available to us. Moving forward, I plan to collaborate with experimental neurobiologists. These collaborations are aimed at validating our model’s predictions, including the delta-protocadherins mentioned in the paper. However, considering the extensive time and resources required for conducting and interpreting experimental results, I believe it is more pragmatic to present a comprehensive experimental study, including the design and execution of experiments informed by the model’s predictions, in a separate follow-up paper. I intend to include a paragraph in the discussion of this paper outlining the future direction for experimental validation.

1. The model’s applicability in other neuronal connectivity settings has not been thoroughly explored.

I recognize the importance of assessing the model across different neuronal systems. In response to similar feedback from Reviewer 1, I am keen to extend the study to include the C.elegans dataset mentioned earlier. The results from applying our bilinear model to the second dataset will be incorporated into the revised manuscript.

1. The proposed method relies on the availability of neuronal connectomic data for model training, which may be limited or absent in certain brain connectivity settings.

The concern regarding the dependency of our model on the availability of connectomic data is valid. While complete connectomes are available for organisms like C.elegans and Drosophila, and efforts are underway to map the connectome of the entire mouse brain, such data may not always be accessible for all research contexts. Recognizing this limitation, part of the ongoing research is to explore ways to adapt our model to the available data, such as projectomic data. Furthermore, our bilinear model is compatible with trans-synaptic virus-based sequencing techniques [4, 5], allowing us to leverage data from these experimental approaches to uncover the genetic underpinnings of neuronal connectivity. These initiatives are crucial steps towards broadening the applicability of our model, ensuring its relevance and usefulness in diverse brain connectivity studies where detailed connectomic data may not be readily available.

References

[1] Dániel L. Barabási and Albert-László Barabási. A genetic model of the connectome. Neuron, 105(3):435–445, 2020.

[2] István A. Kovács, Dániel L. Barabási, and Albert-László Barabási. Uncovering the genetic blueprint of the c. elegans nervous system. Proceedings of the National Academy of Sciences, 117(52):33570–33577, 2020.

[3] Seth R. Taylor, Gabriel Santpere, Alexis Weinreb, Alec Barrett, Molly B. Reilly, Chuan Xu, Erdem Varol, Panos Oikonomou, Lori Glenwinkel, Rebecca McWhirter, Abigail Poff, Manasa Basavaraju, Ibnul Rafi, Eviatar Yemini, Steven J. Cook, Alexander Abrams, Berta Vidal, Cyril Cros, Saeed Tavazoie, Nenad Sestan, Marc Hammarlund, Oliver Hobert, and David M. 3rd Miller. Molecular topography of an entire nervous system. Cell, 184(16):4329–4347, 2021.

[4] Nicole Y. Tsai, Fei Wang, Kenichi Toma, Chen Yin, Jun Takatoh, Emily L. Pai, Kongyan Wu, Angela C. Matcham, Luping Yin, Eric J. Dang, Denise K. Marciano, John L. Rubenstein, Fan Wang, Erik M. Ullian, and Xin Duan. Trans-seq maps a selective mammalian retinotectal synapse instructed by nephronectin. Nat Neurosci, 25(5):659–674, May 2022.

[5] Aixin Zhang, Lei Jin, Shenqin Yao, Makoto Matsuyama, Cindy van Velthoven, Heather Sullivan, Na Sun, Manolis Kellis, Bosiljka Tasic, Ian R. Wickersham, and Xiaoyin Chen. Rabies virusbased barcoded neuroanatomy resolved by single-cell rna and in situ sequencing. bioRxiv, 2023.

Author response:

Reviewer #1:

The only minor weakness that I found is the assumption of independence of bacterial species, which is expressed as the well-stirred approximation. One could imagine that bacterial species might cooperate, leading to non-uniform distributions that are real. How to distinguish such situations? I believe that this method can be extended to determine if this is the case or not before the application. For example, if the bacteria species are independent of each other and one can use the binomial distributions, then the Fano factor would be proportional to the overall relative fraction of bacterial species. Maybe a simple test can be added to test it before the application of REPOP. However, I believe that this is a minor issue.

This is an interesting point raised by the reviewer.

First, we need to clarify an important point–we do not make a well-stirred assumption. Samples can be drawn and plated from any region of space however small and that region’s population can be quantified using our method. The stirring only occurs after we collect a sample in order to dilute the contents and pour the solution homogeneously over the plate.

As such, learning multiple independent species is possible and not impacted by the dilution (“wellstirred” assumption). In the revised manuscript we will make it clear that this assumption concerns the dilution process. Any correlation between species arises in the initial sample and should be retained in the plating. Once given the sample, the dilution itself produces independent binomial draws from that point in space from which cultures were harvested. REPOP is designed to recover the true underlying heterogeneity in species abundance (even from limited data) by leveraging a Bayesian framework that remains valid regardless of whether species are independent or correlated.

If one applies the method for multiple species as is, REPOP can recover the marginal distribution of each species in each plate if they are selectively cultured or many species at once if the colonies are sufficiently distinct. To demonstrate this, we will add a synthetic example with two species whose populations in a sample are correlated to the manuscript.

However, in order to learn the joint distribution and capture correlations between species within samples, the method would need to be extended. At present, in Eq. 5 we sum the likelihood over all values of n, using a data-driven cutoff (twice the na¨ıvely estimated count times the dilution factor). Extending this to multiple species adding up to (n1,n2), while retain the generality of the method, would require quadratically scaling memory with this cutoff in the population number. For this reason while we will comment on this in the next version of the manuscript, it will not be implemented as part of REPOP.

Reviewer #2:

A more thorough discussion of when and by how much estimated microbial population abundance distributions differ from the ground truth would be helpful in determining the best practices for applying this method. Not only would this allow researchers to understand the sampling effort necessary to achieve the results presented here, but it would also contextualize the experimental results presented in the paper. Particularly, there is a disconnect between the discussion of the large sample sizes necessary to achieve accurate multimodal distribution estimates and the small sample sizes used in both experiments.

That is a great suggestion from the reviewer. To address it, we will expand Appendix B, which currently presents the relative error between the means for the experimental results in Fig. 3, to also include a comparable evaluation for the synthetic data example in Fig. 2.

Specifically, for each example, we will report (1) the relative error in the estimated means (as already done for Fig. 3), and (2) the Kullback-Leibler (KL) divergence between the reconstructed and ground truth distributions. These metrics will be shown as a function of the size of the dataset, enabling a direct assessment of how the sampling effort affects the precision of the inference.

That said, we highlight that by explicitly modeling the dilution process within a Bayesian framework, REPOP extracts the mathematically optimal amount of information from each individual sample no matter the sample size. Our strategy therefore leads to better inference with fewer measurements, which is particularly important in applications such as plate counting, where data acquisition is laborintensive.

Reviewer #3:

While the study is promising, there are a few areas where the paper could be strengthened to increase its impact and usability. First, the extent to which dilution and plating introduce noise is not fully explored. Could this noise significantly affect experimental conclusions? And under what conditions does it matter most? Does it depend on experimental design or specific parameter values? Clarifying this would help readers appreciate when and why REPOP should be used.

We agree with the reviewer that this is an important point, and we will expand Appendix B to include a quantitative analysis using simulated data (Fig. 2), reporting both relative error and KL divergence as a function of dataset size. This complements our response to Reviewer #2 clarifying when REPOP offers the greatest benefit.

In addition, we will expand the discussion on how modeling dilution noise becomes essential when learning population dynamics. In particular, we will emphasize the role of Model 3, especially relevant when working with multiple plates and approaching the asymptotic regime—an aspect that was alluded to in Fig. 3 but not fully explored.

Second, more practical details about the tool itself would be very helpful. Simply stating that it is available on GitHub may not be enough. Readers will want to know what programming language it uses, what the input data should look like, and ideally, see a step-by-step diagram of the workflow. Packaging the tool as an easy-to-use resource, perhaps even submitting it to CRAN or including example scripts, would go a long way, especially since microbiologists tend to favor user-friendly, recipe-like solutions.

We will update the introduction to reinforce that REPOP is written in Python(PyTorch), installable via pip, and designed for ease of use. We are also expanding the tutorials to include clearer guidance on data formatting and common workflows. Author response image 1 will be added in the revised manuscript to better illustrate the full application process.

Author response image 1.

Third, it would be great to see the method tested on existing datasets, such as those from Nic Vega and Jeff Gore (2017), which explore how colonization frequency impacts abundance fluctuation distributions. Even if the general conclusions remain unchanged, showing that REPOP can better match observed patterns would strengthen the paper’s real-world relevance.

That is a great suggestion from the reviewer. We will demonstrate the application of REPOP to datasets such as that of Vega and Gore (Ref. 27 in the manuscript), as well as other publicly available datasets, in the revised version.

Lastly, it would be helpful for the authors to briefly discuss the limitations of their method, as no approach is without its constraints. Acknowledging these would provide a more balanced and transparent perspective.

We agree with the reviewer on that. A new subsection will explicitly address the assumptions of our method, and therefore its limitations, including assumptions about species classification, computational cost of joint inference, and dependence on accurate dilution modeling. This discussion will synthesize points raised throughout our response to all reviewers.

Author response:

Reviewer #1 (Public review):

Strengths:

The genetic approaches here for visualizing the recombination status of an endogenous allele are very clever, and by comparing the turnover of wildtype and mutant cells in the same animal the authors can make very convincing arguments about the effect of chronic loss of pu.1. Likely this phenotype would be either very subtle or nonexistent without the point of comparison and competition with the wildtype cells.

Using multiple species allows for more generalizable results, and shows conservation of the phenomena at play.

The demonstration of changes to proliferation and cell death in concert with higher expression of tp53 is compelling evidence for the authors' argument.

Weaknesses:

This paper is very strong. It would benefit from further investigating the specific relationship between pu.1 and tp53 specifically. Does pu.1 interact with the tp53 locus? Specific molecular analysis of this interaction would strengthen the mechanistic findings.

We agree with the reviewer’s assessment regarding the significance of the relationship between PU.1 and TP53. A previous study by Tschan et al(1) has shown that PU.1 attenuates the transcriptional activity of the p53 tumor suppressor family through direct binding to the DNA-binding and/or the oligomerization domains of p53/p73 proteins. We will discuss this point in the revised manuscript and cite this paper accordingly. Moreover, to further investigate the interaction between Pu.1 and Tp53 in zebrafish, we intend to perform a comprehensive analysis of the tp53 promoter region utilizing bioinformatic prediction tools. This approach aims to identify potential Pu.1 binding sites, thereby providing insights into the direct regulatory interactions between Pu.1 and the tp53 promoter in zebrafish. 

Reviewer #2 (Public review):

Strengths:

Generation of an elegantly designed conditional pu.1 allele in zebrafish that allows for the visual detection of expression of the knockout allele.

The combination of analysis of pu.1 function in two model systems, zebrafish and mouse, strengthens the conclusions of the paper.

Confirmation of the functional significance of the observed upregulation of tp53 in mutant microglia through double mutant analysis provides some mechanistic insight.

Weaknesses:

(1) The presented RNA-Seq analysis of mutant microglia is underpowered and details on how the data was analyzed are missing. Only 9-15 cells were analyzed in total (3 pools of 3-5 cells each). Further, the variability in relative gene expression of ccl35b.1, which was used as a quality control and inclusion criterion to define pools consisting of microglia, is extremely high (between ~4 and ~1600, Figure S7A).

In the revised manuscript, we will elaborate on the methodological details of the RNA analysis. Owing to the technical challenge of unambiguously distinguishing microglia from dendritic cells (DCs) in brain cell suspensions, we employed a strategy of isolating 3-5 cells per pool and quantifying the relative expression of the microglia-specific marker ccl34b.1 normalized to the DC-specific marker ccl19a.1. This approach aimed to reduce DC contamination in downstream analyses. Across all experimental groups subjected to RNA-seq analysis, the ccl34b.1/ccl19a.1 expression ratios exceeded 5, confirming microglia as the dominant cell population. Nonetheless, residual DC contamination in the RNA-seq data cannot be entirely ruled out. We will explicitly acknowledge this technical constraint in the revised manuscript to ensure methodological transparency.

(2) The authors conclude that the reduction of microglia observed in the adult brain after cKO of pu.1 in the spi-b mutant background is due to apoptosis (Lines 213-215). However, they only provide evidence of apoptosis in 3-5 dpf embryos, a stage at which loss of pu.1 alone does lead to a complete loss of microglia (Figure 2E). A control of pu.1 KI/d839 mutants treated with 4OHT should be added to show that this effect is indeed dependent on the loss of spi-b. In addition, experiments should be performed to show apoptosis in the adult brain after cKO of pu.1 in spi-b mutants as there seems to be a difference in the requirement of pu.1 in embryonic and adult stages.

We apologize for the omission of data regarding conditional pu.1 knockout alone in the embryos in our manuscript which may have led to ambiguity. We would like to clarify that conditional pu.1 knockout alone at the embryonic stage does not induce microglial death (Author response image 1). Microglial death occurs only when Pu.1 is disrupted in the spi-b mutant background, in both embryonic and adult brains. The blebbing morphology of some microglia after pu.1 conditional knock out in adult spi-b mutant indicated microglia undergo apoptosis at both embryonic (Figure S4) and adult stages Author response image 2). The reviewer’s concern likely arises from the distinct outcomes of global pu.1 knockout (Figure 2) versus conditional pu.1 ablation. Global knockout eliminates microglia during early development due to Pu.1’s essential role in myeloid lineage specification. We plan to include this clarification in the revised manuscript.

Author response image 1.

Conditional depletion of Pu.1 in embryonic microglia had no effect for their short-term survival. (A) Schematics of 4-OHT treatment for pu.1^KI/WT Tg(coro1a:CreER) and pu.1^KI/Δ839 Tg(coro1a:CreER) at embryonic stage. (B) Representative images of DsRed⁺ microglia in pu.1^KI/WT and pu.1^KI/Δ839 at 5 dpf. (C) Quantification of DsRed⁺ microglia in pu.1^KI/WT and pu.1^KI/Δ839 at 3 dpf and 5 dpf. Values represent means ± SD, n.s., P >0.05.

Author response image 2. Simultaneous inactivation of Pu.1 and Spi-b lead to microglia death in adult zebrafish. (A) The experimental setup for pu.1 conditional knockout in adult spi-b^Δ232/Δ232 mutants (B) the representative images of the midbrain cross section of adult pu.1^KI/+;spi-b^Δ232/Δ232;Tg(coro1a:CreER) and pu.1^KI/WTspi-b^Δ232/Δ232;Tg(coro1a:CreER) fish at 2 dpi. The white arrow indicates microglia with blebbing morphology.

(3) The number of microglia after pu.1 knockout in zebrafish did only show a significant decrease 3 months after 4-OHT injection, whereas microglia were almost completely depleted already 7 days after injection in mice. This major difference is not discussed in the paper.

We propose that zebrafish Pu.1 and Spi-b function cooperatively to regulate microglial maintenance, analogous to the role of PU.1 alone in mice. This cooperative mechanism likely explains the observed difference in microglial depletion kinetics between zebrafish and mice following pu.1 conditional knockout. Specifically, the compensatory activity of Spi-b in zebrafish may buffer the immediate loss of Pu.1, whereas in mice, the absence of SPI-B expression in microglia eliminates this redundancy, resulting in rapid microglial depletion. Furthermore, during evolution, SPI-B appears to have acquired lineagespecific roles, becoming absent in microglia. We will expand on this evolutionary divergence and its implications for microglial regulation in the revised manuscript.

(4) Data is represented as mean +/-.SEM. Instead of SEM, standard deviation should be shown in all graphs to show the variability of the data. This is especially important for all graphs where individual data points are not shown. It should also be stated in the figure legend if SEM or SD is shown

We plan to represent our data as mean ± SD in the revised manuscript.

Reference:

(1) Tschan MP, Reddy VA, Ress A, Arvidsson G, Fey MF, Torbett BE. PU.1 binding to the p53 family of tumor suppressors impairs their transcriptional activity. Oncogene. 2008 May 29;27(24):3489-93.

Author response:

eLife assessment

This useful study reports how neuronal activity in the prefrontal cortex maps time intervals during which animals have to wait until reaching a reward and how this mapping is preserved across days. However, the evidence supporting the claims is incomplete as these sequential neuronal patterns do not necessarily represent time but instead may be correlated with stereotypical behavior and restraint from impulsive decision, which would require further controls (e.g. behavioral analysis) to clarify the main message. The study will be of interest to neuroscientists interested in decision making and motor control. 

We thank the editors and reviewers for the constructive comments. In light of the questions mentioned by the reviewers, we plan to perform additional analyses in our revision, particularly aiming to address issues related to single-cell scalability, and effects of motivation and movement. We believe these additional data will greatly improve the rigor and clarity of our study. We are grateful for the review process of eLife.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This paper investigates the neural population activity patterns of the medial frontal cortex in rats performing a nose poking timing task using in vivo calcium imaging. The results showed neurons that were active at the beginning and end of the nose poking and neurons that formed sequential patterns of activation that covaried with the timed interval during nose poking on a trial-by-trial basis. The former were not stable across sessions, while the latter tended to remain stable over weeks. The analysis on incorrect trials suggests the shorter non-rewarded intervals were due to errors in the scaling of the sequential pattern of activity. 

Strengths:

This study measured stable signals using in vivo calcium imaging during experimental sessions that were separated by many days in animals performing a nose poking timing task. The correlation analysis on the activation profile to separate the cells in the three groups was effective and the functional dissociation between beginning and end, and duration cells was revealing. The analysis on the stability of decoding of both the nose poking state and poking time was very informative. Hence, this study dissected a neural population that formed sequential patterns of activation that encoded timed intervals. 

We thank the reviewer for the positive comments.

Weaknesses: 

It is not clear whether animals had enough simultaneously recorded cells to perform the analyzes of Figures 2-4. In fact, rat 3 had 18 responsive neurons which probably is not enough to get robust neural sequences for the trial-by-trial analysis and the correct and incorrect trial analysis. 

We thank the reviewer for the comment. We would like to mention that the 18 cells plotted in Supplementary figure 1 were only from the duration cell category. To improve the clarity of our results, we are going to provide information regarding the number of cells from each rat in our revision. In general, we imaged more than 50 cells from each rat. We would also like to point to the data from individual trials in Supplementary figure 1B showing robust sequentiality.

In addition, the analysis of behavioral errors could be improved. The analysis in Figure 4A could be replaced by a detailed analysis on the speed, and the geometry of neural population trajectories for correct and incorrect trials.

We thank the reviewer for the suggestions. We are going to conduct the analysis as the reviewer recommended. We agree with the reviewer that better presentation of the neural activity will be helpful for the readers.

In the case of Figure 4G is not clear why the density of errors formed two clusters instead of having a linear relation with the produce duration. I would be recommendable to compute the scaling factor on neuronal population trajectories and single cell activity or the computation of the center of mass to test the type III errors. 

We would like to mention that the prediction errors plotted in this graph were calculated from two types of trials. The correct trials tended to show positive time estimation errors while the incorrect trials showed negative time estimation errors. We believe that the polarity switch between these two types suggested a possible use of this neural mechanism to time the action of the rats.

In addition, we are going to perform the analysis suggested by the reviewer in our revision. We agree that different ways of analyzing the data would provide better characterization of the scaling effect.

Due to the slow time resolution of calcium imaging, it is difficult to perform robust analysis on ramping activity. Therefore, I recommend downplaying the conclusion that: "Together, our data suggest that sequential activity might be a more relevant coding regime than the ramping activity in representing time under physiological conditions." 

We agree with the reviewer and we have mentioned this caveat in our original manuscript. We are going to rephrase the sentence as the reviewer suggested during our revision.

Reviewer #2 (Public Review):

In this manuscript, Li and collaborators set out to investigate the neuronal mechanisms underlying "subjective time estimation" in rats. For this purpose, they conducted calcium imaging in the prefrontal cortex of water-restricted rats that were required to perform an action (nosepoking) for a short duration to obtain drops of water. The authors provided evidence that animals progressively improved in performing their task. They subsequently analyzed the calcium imaging activity of neurons and identify start, duration, and stop cells associated with the nose poke. Specifically, they focused on duration cells and demonstrated that these cells served as a good proxy for timing on a trial-by-trial basis, scaling their pattern of actvity in accordance with changes in behavioral performance. In summary, as stated in the title, the authors claim to provide mechanistic insights into subjective time estimation in rats, a function they deem important for various cognitive conditions. 

This study aligns with a wide range of studies in system neuroscience that presume that rodents solve timing tasks through an explicit internal estimation of duration, underpinned by neuronal representations of time. Within this framework, the authors performed complex and challenging experiments, along with advanced data analysis, which undoubtedly merits acknowledgement. However, the question of time perception is a challenging one, and caution should be exercised when applying abstract ideas derived from human cognition to animals. Studying so-called time perception in rats has significant shortcomings because, whether acknowledged or not, rats do not passively estimate time in their heads. They are constantly in motion. Moreover, rats do not perform the task for the sake of estimating time but to obtain their rewards are they water restricted. Their behavior will therefore reflects their motivation and urgency to obtain rewards. Unfortunately, it appears that the authors are not aware of these shortcomings. These alternative processes (motivation, sensorimotor dynamics) that occur during task performance are likely to influence neuronal activity. Consequently, my review will be rather critical. It is not however intended to be dismissive. I acknowledge that the authors may have been influenced by numerous published studies that already draw similar conclusions. Unfortunately, all the data presented in this study can be explained without invoking the concept of time estimation. Therefore, I hope the authors will find my comments constructive and understand that as scientists, we cannot ignore alternative interpretations, even if they conflict with our a priori philosophical stance (e.g., duration can be explicitly estimated by reading neuronal representation of time) and anthropomorphic assumptions (e.g., rats estimate time as humans do). While space is limited in a review, if the authors are interested, they can refer to a lengthy review I recently published on this topic, which demonstrates that my criticism is supported by a wide range of timing experiments across species (Robbe, 2023). In addition to this major conceptual issue that cast doubt on most of the conclusions of the study, there are also several major statistical issues. 

Main Concerns 

(1) The authors used a task in which rats must poke for a minimal amount of time (300 ms and then 1500 ms) to be able to obtain a drop of water delivered a few centimeters right below the nosepoke. They claim that their task is a time estimation task. However, they forget that they work with thirsty rats that are eager to get water sooner than later (there is a reason why they start by a short duration!). This task is mainly probing the animals ability to wait (that is impulse control) rather than time estimation per se. Second, the task does not require to estimate precisely time because there appear to be no penalties when the nosepokes are too short or when they exceed. So it will be unclear if the variation in nosepoke reflects motivational changes rather than time estimation changes. The fact that this behavioral task is a poor assay for time estimation and rather reflects impulse control is shown by the tendency of animals to perform nose-pokes that are too short, the very slow improvement in their performance (Figure 1, with most of the mice making short responses), and the huge variability. Not only do the behavioral data not support the claim of the authors in terms of what the animals are actually doing (estimating time), but this also completely annhilates the interpretation of the Ca++ imaging data, which can be explained by motivational factors (changes in neuronal activity occurring while the animals nose poke may reflect a growing sens of urgency to check if water is available). 

We would like to respond to the reviewer’s comments 1, 2 and 4 together since they all focus on the same issue. We thank the reviewer for the very thoughtful comments and for sharing his detailed reasoning from a recently published review (Robbe, 2023). A lot of the discussion goes beyond the scope of this study and we agree that whether there is an explicit representation of time (an internal clock) in the brain is a difficult question to answer, particularly by using animal behaviors. In fact, even with fully conscious humans and elaborated task design, we think it is still questionable to clearly dissociate the neural substrate of “timing” from “motor”. In the end, it may as well be that as the reviewer cited from Bergson’s article, the experience of time cannot be measured.

Studying the neural representation of any internal state may suffer from the same ambiguity. With all due respect, however, we would like to limit our response in the scope of our results. According to the reviewer, two alternative interpretations of the task-related sequential activity exist: 1, duration cells may represent fidgeting or orofacial movements and 2, duration cells may represent motivation or motion plan of the rats. To test the first alternative interpretation, we will perform a more comprehensive analysis of the behavior data at all the limbs and visible body parts of the rat during nose poke and analyze its periodicity among different trials, although the orofacial movements may not be visible to us.

Regarding the second alternative interpretation, we think our data in the original Figure 4G argues against it. In this graph, we plotted the decoding error of time using the duration cells’ activity against the actual duration of the trials. If the sequential activity of durations cells only represents motivation, then the errors should distribute evenly across different trial times, or linearly modulated by trial durations. The unimodal distribution we observed (Figure 4G and see Author response image 1 below for a re-plot without signs) suggests that the scaling factor of the sequential activity represents information related to time. And the fact that this unimodal distribution centered at the time threshold of the task provides strong evidence for the active use of scaling factor for time estimation. In order to further test the relationship to motivation, we will measure the time interval between exiting nose poke to the start of licking water reward as an independent measurement of motivation for each trial. We will analyze and report whether this measurement correlates with the nose poking durations in our data in the revision.

Author response image 1.

Furthermore, whether the scaling sequential activity we report represents behavioral timing or true time estimation, the reviewer would agree that these activities correlate with the animal’s nose poking durations, and a previous study has showed that PFC silencing led to disruption of the mouse’s timing behavior (PMID: 24367075). The main surprising finding of the paper is that these duration cells are different from the start and end cells in terms of their coding stability. Thus, future studies dissecting the anatomical microcircuit of these duration cells may provide further clue regarding whether they receive inputs from thirst or reward-related brain regions. This may help partially resolve the “time” vs. “motor” debate the reviewer mentioned.

(2) A second issue is that the authors seem to assume that rats are perfectly immobile and perform like some kind of robots that would initiate nose pokes, maintain them, and remove them in a very discretized manner. However, in this kind of task, rats are constantly moving from the reward magazine to the nose poke. They also move while nose-poking (either their body or their mouth), and when they come out of the nose poke, they immediately move toward the reward spout. Thus, there is a continuous stream of movements, including fidgeting, that will covary with timing. Numerous studies have shown that sensorimotor dynamics influence neural activity, even in the prefrontal cortex. Therefore, the authors cannot rule out that what the records reflect are movements (and the scaling of movement) rather than underlying processes of time estimation (some kind of timer). Concretely, start cells could represent the ending of the movement going from the water spout to the nosepoke, and end cells could be neurons that initiate (if one can really isolate any initiation, which I doubt) the movement from the nosepoke to the water spout. Duration cells could reflect fidgeting or orofacial movements combined with an increasing urgency to leave the nose pokes.

(3)The statistics should be rethought for both the behavioral and neuronal data. They should be conducted separately for all the rats, as there is likely interindividual variability in the impulsivity of the animals.

We thank the reviewer for the comment, yet we are not quite sure what specifically was asked by the reviewer. There is undoubtedly variance among individual animals. One of the core reasons for statistical comparison is to compare the group difference with the variance due to sampling. It appears that the reviewer would like to require we conduct our analysis using each rat individually. We will conduct and report analysis with individual rat in Figure 1C, Figure 2C, G, K, Figure 4F in our revised manuscript.

(4) The fact that neuronal activity reflects an integration of movement and motivational factors rather than some abstract timing appears to be well compatible with the analysis conducted on the error trials (Figure 4), considering that the sensorimotor and motivational dynamics will rescale with the durations of the nose poke. 

(5) The authors should mention upfront in the main text (result section) the temporal resolution allowed by their Ca+ probe and discuss whether it is fast enough in regard of behavioral dynamics occurring in the task. 

We thank the reviewer for the suggestion. We have originally mentioned the caveat of calcium imaging in the interpretation of our results. We will incorporate more texts for this purpose during our revision. In terms of behavioral dynamics (start and end of nose poke in this case), we think calcium imaging could provide sufficient kinetics. However, the more refined dynamics related to the reproducibility of the sequential activity or the precise representation of individual cells on the scaled duration may be benefited from improved time resolution.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) Please refer explicitly to the three types of cells in the abstract. 

We will modify the abstract as suggested during revision.

(2) Please refer to the work of Betancourt et al., 2023 Cell Reports, where a trial-by-trail analysis on the correlation between neural trajectory dynamics in MPC and timing behavior is reported. In that same paper the stability of neural sequences across task parameters is reported. 

We will cite and discuss this study in our revised paper.

(3) Please state the number of studied animals at the beginning of the results section. 

We will provide this information as requested. The number of animals were also plotted in Figure 1D for each analysis.

(4) Why do the middle and right panels of Figure 2E show duration cells. 

Figure 2E was intended to show examples of duration cells’ activity. We included different examples of cells that peak at different points in the scaled duration. We believe these multiple examples would give the readers a straight forward impression of these cells’ activity patterns.

(5) Which behavioral sessions of Figure 1B were analyzed further. 

We will label the analyzed sessions in Figure 1B during our revision.

(6) In Figure 3A-C please increase the time before the beginning of the trial in order to visualize properly the activation patterns of the start cells. 

We thank the reviewer for the suggestion and will modify the figure accordingly during revision.

(7) Please state what could be the behavioral and functional effect of the ablation of the cortical tissue on top of mPFC. 

We thank the reviewer for the question. In our experience, mice with lens implanted in mPFC did not show observable different to mice without surgery regarding the acquisition of the task and the distribution of the nose-poke durations. Although we could not rule out the effect on other cognitive process, the mice appeared to be intact in the scope of our task. We will provide these behavior data during our revision.

Author response:

Public Reviews:

Reviewer #1 (Public Review):

Summary:

SUFU modulates Sonic hedgehog (SHH) signaling and is frequently mutated in the B-subtype of SHH-driven medulloblastoma. The B-subtype occurs mostly in infants, is often metastatic, and lacks specific treatment. Yabut et al. found that Fgf5 was highly expressed in the B-subtype of SHH-driven medulloblastoma by examining a published microarray expression dataset. They then investigated how Fgf5 functions in the cerebellum of mice that have embryonic Sufu loss of function. This loss was induced using the hGFAP-cre transgene, which is expressed in multiple cell types in the developing cerebellum, including granule neuron precursors (GNPs) derived from the rhombic lip. By measuring the area of Pax6+ cells in the external granule cell layer (EGL) of Sufu-cKO mice at postnatal day 0, they find Pax6+ cells occupy a larger area in the posterior lobe adjacent to the secondary fissure, which is poorly defined. They show that Fgf5 RNA and phosphoErk1/2 immunostaining are also higher in the same disrupted region. Some of the phosphoErk1/2+ cells are proliferative in the Sufu-cKO. Western blot analysis of Gli proteins that modulate SHH signaling found reduced expression and absence of Gli1 activity in the region of cerebellar dysgenesis in Sufu-cKO mice. This suggests the GNP expansion in this region is independent of SHH signaling. Amazingly, intraventricular injection of the FGFR1-2 antagonist AZD4547 from P0-4 and examined histologically at P7 found the treatment restored cytoarchitecture in the cerebella of Sufu-cKO mice. This is further supported by NeuN immunostaining in the internal granule cell layer, which labels mature, non-diving neurons, and KI67 immunostaining, indicating dividing cells, and primarily found in the EGL. The mice were treated beginning at a timepoint when cerebellar cytoarchitecture was shown to be disrupted and it is indistinguishable from control following treatment. Figure 3 presents the most convincing and exciting data in this manuscript.

Sufu-cKO do not readily develop cerebellar tumors. The authors detected phosphorylated H2AX immunostaining, which labels double-strand breaks, in some cells in the EGL in regions of cerebellar dysgenesis in the Sufu-cKO, as was cleaved Caspase 3, a marker of apoptosis. P53, downstream of the double-strand break pathway, the protein was reduced in Sufu-cKO cerebellum. Genetically removing p53 from the Sufu-cKO cerebellum resulted in cerebellar tumors in 2-month old mice. The Sufu;p53-dKO cerebella at P0 lacked clear foliation, and the secondary fissure, even more so than the Sufu-cKO. Fgf5 RNA and signaling (pERK1/2) were also expressed ectopically.

The conclusions of the paper are largely supported by the data, but some data analysis need to be clarified and extended.

(1) The rationale for examining Fgf5 in medulloblastoma is not sufficiently convincing. The authors previously reported that Fgf15 was upregulated in neocortical progenitors of mice with conditional loss of Sufu (PMID: 32737167). In Figure 1, the authors report FGF5 expression is higher in SHH-type medulloblastoma, especially the beta and gamma subtypes mostly found in infants. These data were derived from a genome-wide dataset and are shown without correction for multiple testing, including other Fgfs. Showing the expression of other Fgfs with FDR correction would better substantiate their choice or moving this figure to later in the manuscript as support for their mouse investigations would be more convincing.

To assess FGF5 (ENSG00000138675) expression in MB tissues, we used Geo2R (Barrett et al., 2013) to analyze published human MB subtype expression arrays from accession no. GSE85217 (Cavalli et al., 2017). GEO2R is an interactive web tool that compares expression levels of genes of interest (GOI) between sample groups in the GEO series using original submitter-supplied processed data tables. We entered the GOI Ensembl ID and organized data sets according to age and MB subgroup or MBSHH subtype classifications. GEO2R results presented gene expression levels as a table ordered by FDR-adjusted (Benjamini & Hochberg) p-values, with significance level cut-off at 0.05, processed by GEO2R’s built-in limma statistical test. Resulting data were subsequently exported into Prism (GraphPad). We generated scatter plots presenting FGF5 expression levels across all MB subgroups (Figure 1A) and MBSHH subtypes (Figure 1D). We performed additional statistical analyses to compare FGF5 expression levels between MB subgroups and MBSHH subtypes and graphed these data as violin plots (Figure 1B, 1C, and 1E). For these analyses, we used one-way ANOVA with Holm-Sidak’s multiple comparisons test, single pooled variance. P value ≤0.05 was considered statistically significant. Graphs display the mean ± standard error of the mean (SEM).

Author response image 1.

Comparative expression of FGF ligands, FGF5, FGF10, FGF12, and FGF19, across all MB subgroups. FGF12 expression is not significantly different, while FGF5, FGF10, and FGF19, show distinct upregulation in MBSHH subgroup (MBWNT n=70, MBSHH n=224, MBGR3 n=143, MBGR4 n=326).

Expression of the 21 known FGF ligands were also analyzed. Many FGFs did not exhibit differential expression levels in MBSHH compared to other MB subgroups, such as with FGF12 in Figure 1. FGF5, FGF10, and FGF19 (the human orthologue of mouse FGF15) all showed specific upregulation in MBSHH compared to other MB subgroups (Author response image 1), supporting our previous observations that FGF15 is a downstream target of SHH signaling (Yabut et al., 2020), as the reviewer pointed out. However, further stratification of MBSHH patient data revealed that only FGF5 specifically showed upregulation in infants with MBSHH (MBSHHb and MBSHHg Author response image 2) indicating a more prominent role for FGF5 in the developing cerebellum and driver of MBSHH tumorigenesis in this dynamic environment.

Author response image 2.

Comparative expression of FGF5, FGF10, and FGF19 in different MBSHH subtypes. FGF5 specifically show mRNA relative levels above 6 in 81% of MBSHH infant patient tumors (n=80 MBSHHb and MBSHHg tumors) unlike 35% of MBSHHa  (n=65) or 0% of MBSHHd  (n=75) tumors.

(2) The Sufu-cKO cerebellum lacks a clear anchor point at the secondary fissure and foliation is disrupted in the central and posterior lobes. It would be helpful for the authors to review Sudarov & Joyner (PMID: 18053187) for nomenclature specific to the developing cerebellum.

The reviewers are correct that the cerebellar foliation is severely disrupted in central and posterior lobes, as per Sudarov and Joyner (Neural Development 2007). This nomenclature may be referred to describe the regions referred in this manuscript.

(3) The metrics used to quantify cerebellar perimeter and immunostaining are not sufficiently described. It is unclear whether the individual points in the bar graph represent a single section from independent mice, or multiple sections from the same mice. For example, in Figures 2B-D. This also applies to Figure 3C-D.

All quantification were performed from 2-3 20 um cerebellar sections of 3-6 independent mice per genotype analyzed. Individual points in the bar graphs represent the average cell number (quantified from 2-3 sections) from each mice. Figure 2B show data points from n=4 mice per genotype. Figure 2C show data from n=3 mice per genotype. Figure 2D show data from n=6 mice per genotype.  Figure 3C-D show data from n=3 mice per genotype.

(4) The data on Fgf5 RNA expression presented in Figure 2E are not sufficiently convincing. The perimeter and cytoarchitecture of the cerebellum are difficult to see and the higher magnification shown in 2F should be indicated in 2E.

The lack of foliation in Sufu-cKO cerebellum is clear particularly when visualizing the perimeter via DAPI labeling (Figure 2E). The expression area of FGF5 is also visibly larger, given that all images in Figure 2E are presented in the same scale (scale bars = 500 um). 

(5) The data presented in Figure 3 are not sufficiently convincing. The number of cells double positive for pErk and KI67 (Figure 3B) are difficult to see and appear to be few, suggesting the quantification may be unreliable.

We used KI67+ expression to provide a molecular marker of regions to be quantified in both WT and Sufu-cKO sections. Quantification of labeled cells were performed in images obtained by confocal microscopy, enabling imaging of 1-2 um optical slices since Ki67 or pERK expression might not localize within the same cellular compartments. We relied on continuous DAPI nuclear staining to distinguish individual cells in each optical slice and the colocalization of of Ki67 and pERK. All quantification were performed from 2-3 20 um cerebellar sections of 3-6 independent mice per genotype analyzed. Individual points in the bar graphs represent the average cell number (quantified from 2-3 sections) from each mice.

(6) The data presented in Figure 4F-J would be more convincing with quantification. The Sufu;p53-dKO appears to have a thickened EGL across the entire vermis perimeter, and very little foliation, relative to control and single cKO cerebella. This is a more widespread effect than the more localized foliation disruption in the Sufu-cKO. 

We agree with the reviewers that quantification of these phenotypes provide a solid measure of the defects. The phenotypes of Sufu:p53-dKO cerebellum are so profound requiring  in-depth characterization that will be the focus of future studies.

(7) Figure 5 does not convincingly summarize the results. Blue and purple cells in sagittal cartoon are not defined. Which cells express Fgf5 (or other Fgfs) has not been determined. The yellow cells are not defined in relation to the initial cartoon on the left.

The revised manuscript will address this confusion by clearly labeling the cells and their roles in the schematic diagram.

Reviewer #2 (Public Review):

Summary:

Mutations in SUFU are implicated in SHH medulloblastoma (MB). SUFU modulates Shh signaling in a context-dependent manner, making its role in MB pathology complex and not fully understood. This study reports that elevated FGF5 levels are associated with a specific subtype of SHH MB, particularly in pediatric cases. The authors demonstrate that Sufu deletion in a mouse model leads to abnormal proliferation of granule cell precursors (GCPs) at the secondary fissure (region B), correlating with increased Fgf5 expression. Notably, pharmacological inhibition of FGFR restores normal cerebellar development in Sufu mutant mice.

Strengths:

The identification of increased FGF5 in subsets of MB is novel and a key strength of the paper.

Weaknesses:

The study appears incomplete despite the potential significance of these findings. The current paper does not fully establish the causal relationship between Fgf5 and abnormal cerebellar development, nor does it clarify its connection to SUFU-related MB. Some conclusions seem overstated, and the central question of whether FGFR inhibition can prevent tumor formation remains untested.

Reviewer #3 (Public Review):

Summary:

The interaction between FGF signaling and SHH-mediated GNP expansion in MB, particularly in the context of Sufu LoF, has just begun to be understood. The manuscript by Yabut et al. establishes a connection between ectopic FGF5 expression and GNP over-expansion in a late-stage embryonic Sufu LoF model. The data provided links region-specific interaction between aberrant FGF5 signaling with the SHH subtype of medulloblastoma. New data from Yabut et al. suggest that ectopic FGF5 expression correlates with GNP expansion near the secondary fissure in Sufu LoF cerebella. Furthermore, pharmacological blockade of FGF signaling inhibits GNP proliferation. Interestingly, the data indicate that the timing of conditional Sufu deletion (E13.5 using the hGFAP-Cre line) results in different outcomes compared to later deletion (using Math1-cre line, Jiwani et al., 2020). This study provides significant insights into the molecular mechanisms driving GNP expansion in SHH subgroup MB, particularly in the context of Sufu LoF. It highlights the potential of targeting FGF5 signaling as a therapeutic strategy. Additionally, the research offers a model for better understanding MB subtypes and developing targeted treatments.

Strengths:

One notable strength of this study is the extraction and analysis of ectopic FGF5 expression from a subset of MB patient tumor samples. This translational aspect of the study enhances its relevance to human disease. By correlating findings from mouse models with patient data, the authors strengthen the validity of their conclusions and highlight the potential clinical implications of targeting FGF5 in MB therapy.

The data convincingly show that FGFR signaling activation drives GNP proliferation in Sufu, conditional knockout models. This finding is supported by robust experimental evidence, including pharmacological blockade of FGF signaling, which effectively inhibits GNP proliferation. The clear demonstration of a functional link between FGFR signaling and GNP expansion underscores the potential of FGFR as a therapeutic target in SHH subgroup medulloblastoma.

Previous studies have demonstrated the inhibitory effect of FGF2 on tumor cell proliferation in certain MB types, such as the ptc mutant (Fogarty et al., 2006)(Yaguchi et al., 2009). Findings in this manuscript provide additional support suggesting multiple roles for FGF signaling in cerebellar patterning and development.

Weaknesses:

In the GEO dataset analysis, where FGF5 expression is extracted, the reporting of the P-value lacks detail on the statistical methods used, such as whether an ANOVA or t-test was employed. Providing comprehensive statistical methodologies is crucial for assessing the rigor and reproducibility of the results. The absence of this information raises concerns about the robustness of the statistical analysis.

The revised manuscript will include the following detailed explanation of the statistical analyses of the GEO dataset:

For the analysis of expression values of FGF5 (ENSG00000138675), we obtained these values using Geo2R (Barrett et al., 2013), which directly analyze published human MB subtype expression arrays from accession no. GSE85217 (Cavalli et al., 2017). GEO2R is an interactive web tool that compares expression levels of genes of interest (GOI) between sample groups in the GEO series using original submitter-supplied processed data tables. We simply entered the GOI Ensembl ID and organized data sets according to age and MB subgroup or MBSHH subtype classifications. GEO2R results presented gene expression levels as a table ordered by FDR-adjusted (Benjamini & Hochberg) p-values, with significance level cut-off at 0.05, processed by GEO2R’s built-in limma statistical test. Resulting data were subsequently exported into Prism (GraphPad). We generated scatter plots presenting FGF5 expression levels across all MB subgroups (Figure 1A) and MBSHH subtypes (Figure 1D). We performed additional statistical analyses to compare FGF5 expression levels between MB subgroups and MBSHH subtypes and graphed these data as violin plots (Figure 1B, 1C, and 1E). For these analyses, we used one-way ANOVA with Holm-Sidak’s multiple comparisons test, single pooled variance. P value ≤0.05 was considered statistically significant. Graphs display the mean ± standard error of the mean (SEM). Sample sizes were:

Author response table 1.

Another concern is related to the controls used in the study. Cre recombinase induces double-strand DNA breaks within the loxP sites, and the control mice did not carry the Cre transgene (as stated in the Method section), while Sufu-cKO mice did. This discrepancy necessitates an additional control group to evaluate the effects of Cre-induced double-strand breaks on phosphorylated H2AX-DSB signaling. Including this control would strengthen the validity of the findings by ensuring that observed effects are not artifacts of Cre recombinase activity.

The breeding scheme we used to generate homozygous SUFU conditional mutants will not generate pups carrying only hGFAP-Cre. Thus, we are unable to compare expression of gH2AX expression in littermates that do not carry loxP sites. The reviewer is correct in pointing out the possibility of Cre recombinase activity inducing double-strand breaks on its own. However, it is likely that any hGFAP-Cre induced double-strand breaks does not sufficiently cause the phenotypes we observed in homozygous mutants (Sufu-cKO) mice because the cerebellum of mice carry heterozygous SUFU mutations (hGFAP-Cre;Sufu-fl/+) do not display the profound cerebellar phenotypes observed in Sufu-cKO mice. We cannot rule out, however, any undetectable abnormalities that could be present which may require further analyses.

Although the use of the hGFAP-Cre line allows genetic access to the late embryonic stage, this also targets multiple celltypes, including both GNPs and cerebellar glial cells. However, the authors focus primarily on GNPs without fully addressing the potential contributions of neuron-glial interaction. This oversight could limit the understanding of the broader cellular context in which FGF signaling influences tumor development. 

The reviewer is correct in that hGFAP-Cre also targets other cell types, such as cerebellar glial cells, which are generated when Cre-expression has begun. It is possible that cerebellar glial cell development is also compromised in Sufu-cKO mice and may disrupt neuron-glial interaction, due to or independently of FGF signaling. In-depth studies are required to interrogate how loss of SUFU specifically affect development of cerebellar glial cells and influence their cellular interactions in the developing cerebellum.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The study by McKim et al seeks to provide a comprehensive description of the connectivity of neurosecretory cells (NSCs) using a high-resolution electron microscopy dataset of the fly brain and several single-cell RNA seq transcriptomic datasets from the brain and peripheral tissues of the fly. They use connectomic analyses to identify discrete functional subgroups of NSCs and describe both the broad architecture of the synaptic inputs to these subgroups as well as some of the specific inputs including from chemosensory pathways. They then demonstrate that NSCs have very few traditional presynapses consistent with their known function as providing paracrine release of neuropeptides. Acknowledging that EM datasets can't account for paracrine release, the authors use several scRNAseq datasets to explore signaling between NSCs and characterize widespread patterns of neuropeptide receptor expression across the brain and several body tissues. The thoroughness of this study allows it to largely achieve it's goal and provides a useful resource for anyone studying neurohormonal signaling.

Strengths:

The strengths of this study are the thorough nature of the approach and the integration of several large-scale datasets to address short-comings of individual datasets. The study also acknowledges the limitations that are inherent to studying hormonal signaling and provides interpretations within the the context of these limitations.

Weaknesses:

Overall, the framing of this paper needs to be shifted from statements of what was done to what was found. Each subsection, and the narrative within each, is framed on topics such as "synaptic output pathways from NSC" when there are clear and impactful findings such as "NSCs have sparse synaptic output". Framing the manuscript in this way allows the reader to identify broad takeaways that are applicable to other model system. Otherwise, the manuscript risks being encyclopedic in nature. An overall synthesis of the results would help provide the larger context within which this study falls.

We agree with the reviewer and will replace all the subsection titles as suggested.

The cartoon schematic in Figure 5A (which is adapted from a 2020 review) has an error. This schematic depicts uniglomerular projection neurons of the antennal lobe projecting directly to the lateral horn (without synapsing in the mushroom bodies) and multiglomerular projection neurons projecting to the mushroom bodies and then lateral horn. This should be reversed (uniglomerular PNs synapse in the calyx and then further project to the LH and multiglomerular PNs project along the mlACT directly to the LH) and is nicely depicted in a Strutz et al 2014 publication in eLife.

We thank the reviewer for spotting this error. We will modify the schematic as suggested.

Reviewer #2 (Public review):

Summary:

The authors aim to provide a comprehensive description of the neurosecretory network in the adult Drosophila brain. They sought to assign and verify the types of 80 neurosecretory cells (NSCs) found in the publicly available FlyWire female brain connectome. They then describe the organization of synaptic inputs and outputs across NSC types and outline circuits by which olfaction may regulate NSCs, and by which Corazon-producing NSCs may regulate flight behavior. Leveraging existing transcriptomic data, they also describe the hormone and receptor expressions in the NSCs and suggest putative paracrine signaling between NSCs. Taken together, these analyses provide a framework for future experiments, which may demonstrate whether and how NSCs, and the circuits to which they belong, may shape physiological function or animal behavior.

Strengths:

This study uses the FlyWire female brain connectome (Dorkenwald et al. 2023) to assign putative cell types to the 80 neurosecretory cells (NSCs) based on clustering of synaptic connectivity and morphological features. The authors then verify type assignments for selected populations by matching cluster sizes to anatomical localization and cell counts using immunohistochemistry of neuropeptide expression and markers with known co-expression.

The authors compare their findings to previous work describing the synaptic connectivity of the neurosecretory network in larval Drosophila (Huckesfeld et al., 2021), finding that there are some differences between these developmental stages. Direct comparisons between adults and larvae are made possible through direct comparison in Table 1, as well as the authors' choice to adopt similar (or equivalent) analyses and data visualizations in the present paper's figures.

The authors extract core themes in NSC synaptic connectivity that speak to their function: different NSC types are downstream of shared presynaptic outputs, suggesting the possibility of joint or coordinated activation, depending on upstream activity. NSCs receive some but not all modalities of sensory input. NSCs have more synaptic inputs than outputs, suggesting they predominantly influence neuronal and whole-body physiology through paracrine and endocrine signaling.

The authors outline synaptic pathways by which olfactory inputs may influence NSC activity and by which Corazon-releasing NSCs may regulate flight. These analyses provide a basis for future experiments, which may demonstrate whether and how such circuits shape physiological function or animal behavior.

The authors extract expression patterns of neuropeptides and receptors across NSC cell types from existing transcriptomic data (Davie et al., 2018) and present the hypothesis that NSCs could be interconnected via paracrine signaling. The authors also catalog hormone receptor expression across tissues, drawing from the Fly Cell Atlas (Li et al., 2022).

Weaknesses:

The clustering of NSCs by their presynaptic inputs and morphological features, along with corroboration with their anatomical locations, distinguished some, but not all cell types. The authors attempt to distinguish cell types using additional methodologies: immunohistochemistry (Figure 2), retrograde trans-synaptic labeling, and characterization of dense core vesicle characteristics in the FlyWire dataset (Figure 1, Supplement 1). However, these corroborating experiments often lacked experimental replicates, were not rigorously quantified, and/or were presented as singular images from individual animals or even individual cells of interest. The assignments of DH44 and DMS types remain particularly unconvincing.

We thank the reviewer for this comment. We would like to clarify that the images presented in Figure 2 and Figure 1 Supplement 1 are representative images based on at least 5 independent samples. We will clarify this in the figure caption and methods. The electron micrographs showing dense core vesicle (DCV) characteristics (Figure 1 Supplement E-G) are also representative images based on examination of multiple neurons. However, we agree with the reviewer that a rigorous quantification would be useful to showcase the differences between DCVs from NSC subtypes. Therefore, we have now performed a quantitative analysis of the DCVs in putative m-NSC^DH44 (n=6), putative m-NSC^DMS (n=6) and descending neurons (n=4) known to express DMS. For consistency, we examined the cross section of each cell where the diameter of nuclei was the largest. We quantified the mean gray value of at least 50 DCV per cell. Our analysis shows that mean gray values of putative m-NSC^DMS and DMS descending neurons are not significantly different, whereas the mean gray values of m-NSC^DH44 are significantly larger. This analysis is in agreement with our initial conclusion.

Author response image 1.

The authors present connectivity diagrams for visualization of putative paracrine signaling between NSCs based on their peptide and receptor expression patterns. These transcriptomic data alone are inadequate for drawing these conclusions, and these connectivity diagrams are untested hypotheses rather than results. The authors do discuss this in the Discussion section.

We fully agree with the reviewer and will further elaborate on the limitations of our approach in the revised manuscript. However, there is a very high-likelihood that a given NSC subtype can signal to another NSC subtype using a neuropeptide if its receptor is expressed in the target NSC. This is due to the fact that all NSC axons are part of the same nerve bundle (nervi corpora cardiaca) which exits the brain. The axons of different NSCs form release sites that are extremely close to each other. Neuropeptides from these release sites can easily diffuse via the hemolymph to peripheral tissues that (e.g. fat body and ovaries) that are much further away from the release sites on neighboring NSCs. We believe that neuropeptide receptors are expressed in NSCs near these release sites where they can receive inputs not just from the adjacent NSCs but also from other sources such as the gut enteroendocrine cells. Hence, neuropeptide diffusion is not a limiting factor preventing paracrine signaling between NSCs and receptor expression is a good indicator for putative paracrine signaling.

Reviewer #3 (Public review):

Summary:

The manuscript presents an ambitious and comprehensive synaptic connectome of neurosecretory cells (NSC) in the Drosophila brain, which highlights the neural circuits underlying hormonal regulation of physiology and behaviour. The authors use EM-based connectomics, retrograde tracing, and previously characterised single-cell transcriptomic data. The goal was to map the inputs to and outputs from NSCs, revealing novel interactions between sensory, motor, and neurosecretory systems. The results are of great value for the field of neuroendocrinology, with implications for understanding how hormonal signals integrate with brain function to coordinate physiology.

The manuscript is well-written and provides novel insights into the neurosecretory connectome in the adult Drosophila brain. Some, additional behavioural experiments will significantly strengthen the conclusions.

Strengths:

(1) Rigorous anatomical analysis

(2) Novel insights on the wiring logic of the neurosecretory cells.

Weaknesses:

(1) Functional validation of findings would greatly improve the manuscript.

We agree with this reviewer that assessing the functional output from NSCs would improve the manuscript. Given that we currently lack genetic tools to measure hormone levels and that behaviors and physiology are modulated by NSCs on slow timescales, it is difficult to assess the immediate functional impact of the sensory inputs to NSC using approaches such as optogenetics. However, since l-NSC^CRZ are the only known cell type that provide output to descending neurons, we will functionally test this output pathway using different behavioral assays recommended by this reviewer.

Author response:

Public Reviews:

Reviewer #1 (Public review):

The authors present exciting new experimental data on the antigenic recognition of 78 H3N2 strains (from the beginning of the 2023 Northern Hemisphere season) against a set of 150 serum samples. The authors compare protection profiles of individual sera and find that the antigenic effect of amino acid substitutions at specific sites depends on the immune class of the sera, differentiating between children and adults. Person-to-person heterogeneity in the measured titers is strong, specifically in the group of children's sera. The authors find that the fraction of sera with low titers correlates with the inferred growth rate using maximum likelihood regression (MLR), a correlation that does not hold for pooled sera. The authors then measure the protection profile of the sera against historical vaccine strains and find that it can be explained by birth cohort for children. Finally, the authors present data comparing pre- and post- vaccination protection profiles for 39 (USA) and 8 (Australia) adults. The data shows a cohort-specific vaccination effect as measured by the average titer increase, and also a virus-specific vaccination effect for the historical vaccine strains. The generated data is shared by the authors and they also note that these methods can be applied to inform the bi-annual vaccine composition meetings, which could be highly valuable.

Thanks for this nice summary of our paper.

The following points could be addressed in a revision:

(1) The authors conclude that much of the person-to-person and strain-to-strain variation seems idiosyncratic to individual sera rather than age groups. This point is not yet fully convincing. While the mean titer of an individual may be idiosyncratic to the individual sera, the strain-to-strain variation still reveals some patterns that are consistent across individuals (the authors note the effects of substitutions at sites 145 and 275/276). A more detailed analysis, removing the individual-specific mean titer, could still show shared patterns in groups of individuals that are not necessarily defined by the birth cohort.

As the reviewer suggests, we normalized the titers for all sera to the geometric mean titer for each individual in the US-based pre-vaccination adults and children. This is only for the 2023-circulating viral strains. We then faceted these normalized titers by the same age groups we used in Figure 6, and the resulting plot is shown below. Although there are differences among virus strains (some are better neutralized than others), there are not obvious age group-specific patterns (eg, the trends in the two facets are similar). To us this suggests that at least for these relatively closely related recent H3N2 strains, the strain-to-strain variation does not obviously segregate by age group. Obviously, it is possible (we think likely) that there would be more obvious age-group specific trends if we looked at a larger swath of viral strains covering a longer time range (eg, over decades of influenza evolution). We plan to add the new plots shown below to a supplemental figure in the revised manuscript.

Author response image 1.

Author response image 2.

(2) The authors show that the fraction of sera with a titer below 138 correlates strongly with the inferred growth rate using MLR. However, the authors also note that there exists a strong correlation between the MLR growth rate and the number of HA1 mutations. This analysis does not yet show that the titers provide substantially more information about the evolutionary success. The actual relation between the measured titers and fitness is certainly more subtle than suggested by the correlation plot in Figure 5. For example, the clades A/Massachusetts and A/Sydney both have a positive fitness at the beginning of 2023, but A/Massachusetts has substantially higher relative fitness than A/Sydney. The growth inference in Figure 5b does not appear to map that difference, and the antigenic data would give the opposite ranking. Similarly, the clades A/Massachusetts and A/Ontario have both positive relative fitness, as correctly identified by the antigenic ranking, but at quite different times (i.e., in different contexts of competing clades). Other clades, like A/St. Petersburg are assigned high growth and high escape but remain at low frequency throughout. Some mention of these effects not mapped by the analysis may be appropriate.

Thanks for the nice summary of our findings in Figure 5. However, the reviewer is misreading the growth charts when they say that A/Massachusetts/18/2022 has a substantially higher fitness than A/Sydney/332/2023. Figure 5a shows the frequency trajectory of different variants over time. While A/Massachusetts/18/2022 reaches a higher frequency than A/Sydney/332/2023, the trajectory is similar and the reason that A/Massachusetts/18/2022 reached a higher max frequency is that it started at a higher frequency at the beginning of 2023. The MLR growth rate estimates differ from the maximum absolute frequency reached: instead, they reflect how rapidly each strain grows relative to others. In fact, A/Massachusetts/18/2022 and A/Sydney/332/2023 have similar growth rates, as shown in Supplementary Figure 6b. Similarly, A/Saint-Petersburg/RII-166/2023 starts at a low initial frequency but then grows even as A/Massachusetts/18/2022 and A/Sydney/332/2023 are declining, and so has a higher growth rate than both of those. In the revised manuscript, we will clarify how viral growth rates are estimated from frequency trajectories, and how growth rate differs from max frequency.

(3) For the protection profile against the vaccine strains, the authors find for the adult cohort that the highest titer is always against the oldest vaccine strain tested, which is A/Texas/50/2012. However, the adult sera do not show an increase in titer towards older strains, but only a peak at A/Texas. Therefore, it could be that this is a virus-specific effect, rather than a property of the protection profile. Could the authors test with one older vaccine virus (A/Perth/16/2009?) whether this really can be a general property?

We are interested in studying immune imprinting more thoroughly using sequencing-based neutralization assays, but we note that the adults in the cohorts we studied would have been imprinted with much older strains than included in this library. As this paper focuses on the relative fitness of contemporary strains with minor secondary points regarding imprinting, these experiments are beyond the scope of this study. We’re excited for future work (from our group or others) to explore these points by making a new virus library with strains from multiple decades of influenza evolution.

Reviewer #2 (Public review):

This is an excellent paper. The ability to measure the immune response to multiple viruses in parallel is a major advancement for the field, which will be relevant across pathogens (assuming the assay can be appropriately adapted). I only have a few comments, focused on maximising the information provided by the sera.

Thanks very much!

Firstly, one of the major findings is that there is wide heterogeneity in responses across individuals. However, we could expect that individuals' responses should be at least correlated across the viruses considered, especially when individuals are of a similar age. It would be interesting to quantify the correlation in responses as a function of the difference in ages between pairs of individuals. I am also left wondering what the potential drivers of the differences in responses are, with age being presumably key. It would be interesting to explore individual factors associated with responses to specific viruses (beyond simply comparing adults versus children).

We’re excited by this idea! We plan to include these analyses in our revised pre-print.

Relatedly, is the phylogenetic distance between pairs of viruses associated with similarity in responses?

As above, we like this idea and our revised pre-print will include this analysis.

Figure 5C is also a really interesting result. To be able to predict growth rates based on titers in the sera is fascinating. As touched upon in the discussion, I suspect it is really dependent on the representativeness of the sera of the population (so, e.g., if only elderly individuals provided sera, it would be a different result than if only children provided samples). It may be interesting to compare different hypotheses - so e.g., see if a population-weighted titer is even better correlated with fitness - so the contribution from each individual's titer is linked to a number of individuals of that age in the population. Alternatively, maybe only the titers in younger individuals are most relevant to fitness, etc.

We’re very interested in these analyses, but suggest they may be better explored in subsequent works that could sample more children, teenagers and adults across age groups. Our sera set, as the reviewer suggests, may be under-powered to perform the proposed analysis on subsetted age groups of our larger age cohorts.

In Figure 6, the authors lump together individuals within 10-year age categories - however, this is potentially throwing away the nuances of what is happening at individual ages, especially for the children, where the measured viruses cross different groups. I realise the numbers are small and the viruses only come from a small numbers of years, however, it may be preferable to order all the individuals by age (y-axis) and the viral responses in ascending order (x-axis) and plot the response as a heatmap. As currently plotted, it is difficult to compare across panels

This is a good suggestion, and a revised pre-print will include heatmaps of the different cohorts, ordered by ages of individuals.

Reviewer #3 (Public review):

The authors use high-throughput neutralisation data to explore how different summary statistics for population immune responses relate to strain success, as measured by growth rate during the 2023 season. The question of how serological measurements relate to epidemic growth is an important one, and I thought the authors present a thoughtful analysis tackling this question, with some clear figures. In particular, they found that stratifying the population based on the magnitude of their antibody titres correlates more with strain growth than using measurements derived from pooled serum data. However, there are some areas where I thought the work could be more strongly motivated and linked together. In particular, how the vaccine responses in US and Australia in Figures 6-7 relate to the earlier analysis around growth rates, and what we would expect the relationship between growth rate and population immunity to be based on epidemic theory.

Thank you for this nice summary. This reviewer also notes that the text related to figures 6 and 7 are more secondary to the main story presented in figures 3-5. The main motivation for including figures 6 and 7 were to demonstrate the wide-ranging applications of sequencing-based neutralization data, and this can certainly be clarified in minor text revisions.

Author Response

Public Reviews

We thank both reviewers for taking the time and effort to think critically about our paper and point out areas where it can be improved. In this document, we do our best to clarify any misunderstandings with the hope that further consideration about the strengths and weaknesses of our approach will be possible. Our responses are in bold.

Reviewer #1 (Public Review):

Summary:

In their manuscript, Schmidlin, Apodaca, et al try to answer fundamental questions about the evolution of new phenotypes and the trade-offs associated with this process. As a model, they use yeast resistance to two drugs, fluconazole and radicicol. They use barcoded libraries of isogenic yeasts to evolve thousands of strains in 12 different environments. They then measure the fitness of evolved strains in all environments and use these measurements to examine patterns in fitness trade-offs. They identify only six major clusters corresponding to different trade-off profiles, suggesting the vast genotypic landscape of evolved mutants translates to a highly constrained phenotypic space. They sequence over a hundred evolved strains and find that mutations in the same gene can result in different phenotypic profiles.

Overall, the authors deploy innovative methods to scale up experimental evolution experiments, and in many aspects of their approach tried to minimize experimental variation.

We thank the reviewer for this positive assessment of our work. We are happy that the reviewer noted what we feel is a unique strength of our approach: we scaled up experimental evolution by using DNA barcodes and by exploring 12 related selection pressures. Despite this scaling up, we still see phenotypic convergence among the 744 adaptive mutants we study.

The environments we study represent 12 different concentrations or combinations of two drugs, radicicol and fluconazole. Our hope is that this large dataset (774 mutants x 12 environments) will be useful, both to scientists who are generally interested in the genetic and phenotypic underpinnings of adaptation, and to scientists specifically interested in the evolution of drug resistance.

Weaknesses:

(1) One of the objectives of the authors is to characterize the extent of phenotypic diversity in terms of resistance trade-offs between fluconazole and radicicol. To minimize noise in the measurement of relative fitness, the authors only included strains with at least 500 barcode counts across all time points in all 12 experimental conditions, resulting in a set of 774 lineages passing this threshold. This corresponds to a very small fraction of the starting set of ~21 000 lineages that were combined after experimental evolution for fitness measurements.

This is a misunderstanding that we will work to clarify in the revision. Our starting set did not include 21,000 adaptive lineages. The total number of unique adaptive lineages in this starting set is much lower than 21,000 for two reasons.

First, ~21,000 represents the number of single colonies we isolated in total from our evolution experiments. Many of these isolates possess the same barcode, meaning they are duplicates. Second, and more importantly, most evolved lineages do not acquire adaptive mutations, meaning that many of the 21,000 isolates are genetically identical to their ancestor. In our revised manuscript, we will explicitly state that these 21,000 isolated lineages do not all represent unique, adaptive lineages. In figure 2 and all associated text, we will change the word “lineages” to “isolates,” where relevant.

More broadly speaking, several previous studies have demonstrated that diverse genetic mutations converge at the level of phenotype, and have suggested that this convergence makes adaptation more predictable (PMID33263280, PMID37437111, PMID22282810, PMID25806684). Our study captures mutants that are overlooked in previous studies, such as those that emerge across subtly different selection pressures (e.g., 4 𝜇g/ml vs. 8 𝜇g/ml flu) and those that are undetectable in evolutions lacking DNA barcodes. Thus, while our experimental design misses some mutants (see next comment), it captures many others. Note that 774 adaptive lineages is more than most previous studies. Thus, we feel that “our work – showing that 774 mutants fall into a much smaller number of groups” is important because it “contributes to growing literature suggesting that the phenotypic basis of adaptation is not as diverse as the genetic basis (lines 161 - 162).”

As the authors briefly remark, this will bias their datasets for lineages with high fitness in all 12 environments, as all these strains must be fit enough to maintain a high abundance.

The word “briefly” feels a bit unfair because we discuss this bias on 3 separate occasions (on lines 146 - 147, 260 - 264, and in more detail on 706 - 714). We even walk through an example of a class of mutants that our study misses. We say, “our study is underpowered to detect adaptive lineages that have low fitness in any of the 12 environments. This is bound to exclude large numbers of adaptive mutants. For example, previous work has shown some FLU resistant mutants have strong tradeoffs in RAD (Cowen and Lindquist 2005). Perhaps we are unable to detect these mutants because their barcodes are at too low a frequency in RAD environments, thus they are excluded from our collection of 774.”

In our revised version, we will add more text to the first mention of these missing mutants (lines 146 - 147) so that the implications are more immediately made apparent.

While we “miss” some classes of mutants, we “catch” other classes that may have been missed in previous studies of convergence. For example, we observe a unique class of FLU-resistant mutants that primarily emerged in evolution experiments that lack FLU (Figure 3). Thus, we think that the unique design of our study, surveying 12 environments, allows us to make a novel contribution to the study of phenotypic convergence.

One of the main observations of the authors is phenotypic space is constrained to a few clusters of roughly similar relative fitness patterns, giving hope that such clusters could be enumerated and considered to design antimicrobial treatment strategies. However, by excluding all lineages that fit in only one or a few environments, they conceal much of the diversity that might exist in terms of trade-offs and set up an inclusion threshold that might present only a small fraction of phenotypic space with characteristics consistent with generalist resistance mechanisms or broadly increased fitness. This has important implications regarding the general conclusions of the authors regarding the evolution of trade-offs.

We discussed these implications in some detail in the 16 lines mentioned above (146 - 147, 260 - 264, 706 - 714). To add to this discussion, we will also add the following sentence to the end of the paragraph on lines 697 - 714: “This could complicate (or even make impossible) endeavors to design antimicrobial treatment strategies that thwart resistance”.

We will also add a new paragraph that discusses these implications earlier in our manuscript. This paragraph will highlight the strengths of our method (e.g., that we “catch” classes of mutants that are often overlooked) while being transparent about the weaknesses of our approach (e.g., that we “miss” mutants with strong tradeoffs).

(2) Most large-scale pooled competition assays using barcodes are usually stopped after ~25 to avoid noise due to the emergence of secondary mutations.

The rate at which new mutations enter a population is driven by various factors such as the mutation rate and population size, so choosing an arbitrary threshold like 25 generations is difficult.

We conducted our fitness competition following previous work using the Levy/Blundell yeast barcode system, in which the number of generations reported varies from 32 to 40 (PMID33263280, PMID27594428, PMID37861305, see PMID27594428 for detailed calculation of the fraction of lineages biased by secondary mutations in this system).

The authors measure fitness across ~40 generations, which is almost the same number of generations as in the evolution experiment. This raises the possibility of secondary mutations biasing abundance values, which would not have been detected by the whole genome sequencing as it was performed before the competition assay.

We understand how the reviewer came to this misunderstanding and will adjust our revised manuscript accordingly. Previous work has demonstrated that, in this particular evolution platform, most of the mutations actually occur during the transformation that introduces the DNA barcodes (PMID25731169). In other words, these mutations do not accumulate during the 40 generations of evolution, they are already there. So the observation that we collect a genetically diverse pool of adaptive mutants after 40 generations of evolution is not evidence that 40 generations is enough time for secondary mutations to bias abundance values.

(3) The approach used by the authors to identify and visualize clusters of phenotypes among lineages does not seem to consider the uncertainty in the measurement of their relative fitness. As can be seen from Figure S4, the inter-replicate difference in measured fitness can often be quite large. From these graphs, it is also possible to see that some of the fitness measurements do not correlate linearly (ex.: Med Flu, Hi Rad Low Flu), meaning that taking the average of both replicates might not be the best approach.

This concern, and all subsequent concerns, seem to be driven by either (a) general concerns about the noisiness of fitness measurements obtained from large-scale barcode fitness assays or (b) general concerns about whether the clusters obtained from our dimensional reduction approach capture this noise as opposed to biologically meaningful differences.

We will respond to each concern point-by-point, but want to start by generally stating that (a) our particular large-scale barcode fitness assay has several features that diminish noise, and (b) we devote 4 figures and 200 lines of text to demonstrating that these clusters capture biologically meaningful differences between mutants (and not noise).

In terms of this specific concern, we performed an analysis of noise in the submitted manuscript: Our noisiest fitness measurements correspond to barcodes that are the least abundant and thus suffer the most from stochastic sampling noise. These are also the barcodes that introduce the nonlinearity the reviewer mentions. We removed these from our dataset by increasing our coverage threshold from 500 reads to 5,000 reads. The clusters did not collapse, which suggests that they were not capturing noise (Figure S7 panel B). But we agree with the reviewer that this analysis alone is not sufficient to conclude that the clusters distinguish groups of mutants with unique fitness tradeoffs.

Because the clustering approach used does not seem to take this variability into account, it becomes difficult to evaluate the strength of the clustering, especially because the UMAP projection does not include any representation of uncertainty around the position of lineages.

To evaluate the strength of the clustering, we performed numerous analyses including whole genome sequencing, growth experiments, reclustering, and tracing the evolutionary origins of each cluster (Figures 5 - 8). All of these analyses suggested that our clusters capture groups of mutants that have different fitness tradeoffs. We will adjust our revised manuscript to make clear that we do not rely on the results of a clustering algorithm alone to draw conclusions about phenotypic convergence.

We are also grateful to the reviewer for helping us realize that, as written, our manuscript is not clear with regard to how we perform clustering. We are not using UMAP to decide which mutant belongs to which cluster. Recent work highlights the importance of using an independent clustering method (PMID37590228). Although this recent work addresses the challenge of clustering much higher dimensional data than we survey here, we did indeed use an independent clustering method (gaussian mixture model). In other words, we use UMAP for visualization but not clustering. We also confirm our clustering results using a second independent method (hierarchical clustering; Figure S8). And in our revised manuscript, will confirm with a third method (PCA, see below). We will adjust the main text and the methods section to make these choices clearer.

This might paint a misleading picture where clusters appear well separate and well defined but are in fact much fuzzier, which would impact the conclusion that the phenotypic space is constricted.

The salient question is whether the clusters are so “fuzzy” that they are not meaningful. That interpretation seems unreasonable. Our clusters group mutants with similar genotypes, evolutionary histories, and fitness tradeoffs (Figures 5 - 8). Clustering mutants with similar behaviors is important and useful. It improves phenotypic prediction by revealing which mutants are likely to have at least some phenotypic effects in common. And it also suggests that the phenotypic space is constrained, at least to some degree, which previous work suggests is helpful in predicting evolution (PMID33263280, PMID37437111, PMID22282810, PMID25806684).

(4) The authors make the decision to use UMAP and a gaussian mixed model to cluster and represent the different fitness landscapes of their lineages of interest. Their approach has many caveats. First, compared to PCA, the axis does not provide any information about the actual dissimilarities between clusters. Using PCA would have allowed a better understanding of the amount of variance explained by components that separate clusters, as well as more interpretable components.

The components derived from PCA are often not interpretable. It’s not obvious that each one, or even the first one, will represent some intuitive phenotype, like resistance to fluconazole.

Moreover, we see many non-linearities in our data. For example, fitness in a double drug environment is not predicted by adding up fitness in the relevant single drug environments. Also, there are mutants that have high fitness when fluconazole is absent or abundant, but low fitness when mild concentrations are present. These types of nonlinearities can make the axes in PCA very difficult to interpret, plus these nonlinearities can be missed by PCA, thus we prefer other clustering methods.

We will adjust our revised manuscript to explain these reasons why we chose UMAP and GMM over PCA.

Also, we will include PCA in the supplement of our revised manuscript. Please find below PC1 vs PC2, with points colored according to the cluster assignment in figure 4 (i.e. using a gaussian mixture model). It appears the clusters are largely preserved.

Author response image 1.

Second, the advantages of dimensional reduction are not clear. In the competition experiment, 11/12 conditions (all but the no drug, no DMSO conditions) can be mapped to only three dimensions: concentration of fluconazole, concentration of radicicol, and relative fitness. Each lineage would have its own fitness landscape as defined by the plane formed by relative fitness values in this space, which can then be examined and compared between lineages.

We worry that the idea stems from apriori notions of what the important dimensions should be. It also seems like this would miss important nonlinearities such as our observation that low fluconazole behaves more like a novel selection pressure than a dialed down version of high fluconazole.

Also, we believe the reviewer meant “fitness profile” and not “fitness landscape”. A fitness landscape imagines a walk where every “step” is a mutation. Most lineages in barcoded evolution experiments possess only a single adaptive mutation. A single-step walk is not enough to build a landscape, though others are expanding barcoded evolution experiments beyond the first step (PMID34465770, PMID31723263), so maybe one day this will be possible.

Third, the choice of 7 clusters as the cutoff for the multiple Gaussian model is not well explained. Based on Figure S6A, BIC starts leveling off at 6 clusters, not 7, and going to 8 clusters would provide the same reduction as going from 6 to 7. This choice also appears arbitrary in Figure S6B, where BIC levels off at 9 clusters when only highly abundant lineages are considered.

We agree. We did not rely on the results of BIC alone to make final decisions about how many clusters to include. We thank the reviewer for pointing out this gap in our writing. We will adjust our revised manuscript to explain that we ultimately chose to describe 6 clusters that we were able to validate with follow-up experiments. In figures 5, 6, 7, and 8, we use external information to validate the clusters that we report in figure 4. And in lines 697 – 714, we explain that there are may be additional clusters beyond those we tease apart in this study.

This directly contradicts the statement in the main text that clusters are robust to noise, as more a stringent inclusion threshold appears to increase and not decrease the optimal number of clusters. Additional criteria to BIC could have been used to help choose the optimal number of clusters or even if mixed Gaussian modeling is appropriate for this dataset.

We are under the following impression: If our clustering method was overfitting, i.e. capturing noise, the optimal number of clusters should decrease when we eliminate noise. It increased. In other words, the observation that our clusters did not collapse (i.e. merge) when we removed noise suggests these clusters were not capturing noise.

More generally, our validation experiments, described below, provide additional evidence that our clusters capture meaningful differences between mutants (and not noise).

(5) Large-scale barcode sequencing assays can often be noisy and are generally validated using growth curves or competition assays.

Some types of bar-seq methods, in particular those that look at fold change across two time points, are noisier than others that look at how frequency changes across multiple timepoints (PMID30391162). Here, we use the less noisy method. We also reduce noise by using a stricter coverage threshold than previous work (e.g., PMID33263280), and by excluding batch effects by performing all experiments simultaneously (PMID37237236).

The main assay we use to measure fitness has been previously validated (PMID27594428). No subsequent study using this assay validates using the methods suggested by the reviewer (see PMID37861305, PMID33263280, PMID31611676, PMID29429618, PMID37192196, PMID34465770, PMID33493203).

More to the point, bar-seq has been used, without the reviewer’s suggested validation, to demonstrate that the way some mutant’s fitness changes across environments is different from other mutants (PMID33263280, PMID37861305, PMID31611676, PMID33493203, PMID34596043). This is the same thing that we use bar-seq to demonstrate.

For all of these reasons, we are hesitant to confirm bar-seq itself as a valid way to infer fitness. It seems this is already accepted as a standard in our field.

Having these types of results would help support the accuracy of the main assay in the manuscript and thus better support the claims of the authors.

We don’t agree that fitness measurements obtained from this bar-seq assay generally require validation. But we do agree that it is important to validate whether the mutants in each of our 6 clusters indeed are different from one another in meaningful ways, in particular, in that they have different fitness tradeoffs. We have four figures (5 - 8) and 200 lines of text dedicated to validating whether our clusters capture reproducible and biologically meaningful differences between mutants. Happily, one of these figures (Fig 7) includes growth curves, which are exactly the type of validation experiment asked for by the reviewer.

Below, we walk through the different types of validation experiments that are present in our original manuscript, and additional validation experiments that we plan to include in the revised version. We are hopeful that these validation experiments are sufficient, or at the very least, that this list empowers reviewers to point out where more work is needed.

(1) Mutants from different clusters have different growth curves: In our original manuscript, we measured growth curves corresponding to a fitness tradeoff that we thought was surprising. Mutants in clusters 4 and 5 both have fitness advantages in single drug conditions. While mutants from cluster 4 also are advantageous in the double drug conditions, mutants from cluster 5 are not! We validated these different behaviors by studying growth curves for a mutant from each cluster (Figures 7 and S10).

(2) Mutants from different clusters have different evolutionary origins: In our original manuscript, we came up with a novel way to ask whether the clusters capture different types of adaptive mutants. We asked whether the mutants in each cluster originate from different evolution experiments. Indeed they often do (see pie charts in Figures 6, 7, 8). This method also provides evidence supporting each cluster’s differing fitness tradeoffs.

For example, mutants in cluster 5 appear to have a tradeoff in a double drug condition (described above). They rarely originate from that evolution condition, unlike mutants in nearby cluster 4 (see Figure 7).

(3) Mutants from each cluster often fall into different genes: In our original manuscript, we sequenced many of these mutants and show that mutants in the same gene are often found in the same cluster. For example, all 3 IRA1 mutants are in cluster 6 (Fig 8), both GPB2 mutants are in cluster 4 (Figs 7 & 8), and 35/36 PDR mutants are in either cluster 2 or 3 (Figs 5 & 6).

(4) Mutants from each cluster have behaviors previously observed in the literature: In our original manuscript, we compared our sequencing results to the literature and found congruence. For example, PDR mutants are known to provide a fitness benefit in fluconazole and are found in clusters that have high fitness in fluconazole (lines 457 - 462). Previous work suggests that some mutations to PDR have different tradeoffs than others, which is what we see (lines 540 - 542). IRA1 mutants were previously observed to have high fitness in our “no drug” condition, and are found in the cluster that has the highest fitness in the “no drug” condition (lines 642 - 646). Previous work even confirms the unusual fitness tradeoff we observe where IRA1 and other cluster 6 mutants have low fitness only in low concentrations of fluconazole (lines 652 - 657).

(5) Mutants largely remain in their clusters when we use alternate clustering methods: In our original manuscript, we performed various different reclustering and/or normalization approaches on our data (Fig 6, S5, S7, S8, S9). The clusters of mutants that we observe in figure 4 do not change substantially when we recluster the data. We will add PCA (see above) to these analyses in our revised manuscript.

(6) We will include additional data showing that mutants in different clusters have different evolutionary origins: Cluster 1 is defined by high fitness in low fluconazole that declines with increasing fluconazole (see Fig 4E and Fig 5C). In our revised manuscript, we will show that cluster 1 lineages were overwhelmingly sampled from evolutions conducted in our lowest concentration of fluconazole (see figure panel A below). No other cluster’s evolutionary history shows this pattern (figures 6, 7, and 8).

(7) We will include additional data showing that mutants in different clusters have different growth curves: Cluster 1 lineages are unique in that their fitness advantage is specific to low flu and trades off in higher concentrations of fluconazole. We obtained growth curves for three cluster 1 mutants (2 SUR1 mutants and 1 UPC2 mutant). We compared them to growth curves for three PDR mutants (from clusters 2 and 3). Cluster 1 mutants appear to have the highest growth rates and reach the higher carrying capacity in low fluconazole (see red and green lines in Author response image 2 panel B below). But the cluster 1 mutants are negatively affected by higher concentrations of fluconazole, much more so than the mutants from clusters 2 and 3 (see Author response image 2 panel C below). This is consistent with the different fitness tradeoffs we observe for each cluster (figures 4 and 5). We will include a more detailed version of this analysis and the figures below in our revised manuscript.

Author response image 2.

Validation experiments demonstrate that cluster 1 mutants have uniquely high fitness in only the lowest concentration of fluconazole. (A) The mutant lineages in cluster 1 were largely sampled from evolution experiments performed in low flu. This is not true of other clusters (see pie charts in main manuscript). (B) In low flu (4 𝜇g/ml), Cluster 1 lineages (red/UPC2 and green/SUR1) grow faster and achieve higher density than lineages from clusters 2 and 3 (blue/PDR). This is consistent with barseq measurements demonstrating that cluster 1 mutants have the highest fitness in low flu. (C) Cluster 1 lineages are sensitive to increasing flu concentrations (SUR1 and UPC2 mutants, middle and rightmost graphs). This is apparent in that the gray (8 𝜇g/ml flu) and light blue (32 𝜇g/ml flu) growth curves rise more slowly and reach lower density than the dark blue curves (4 𝜇g/ml flu). But this is not the case for the PDR mutants from clusters 2 and 3 (leftmost graph). These observations are consistent with the bar-seq fitness data presented in the main manuscript (Fig 4E).

With all of these validation efforts combined, we are hopeful that the reviewer is now more convinced that our clusters capture groups of mutants with different fitness tradeoffs (as opposed to noise). We want to conclude by saying that we are grateful to the reviewer for making us think deeply about areas where we can include additional validation efforts as well as areas where we can make our manuscript clearer.

Reviewer #2 (Public Review):

Summary:

Schmidlin & Apodaca et al. aim to distinguish mutants that resist drugs via different mechanisms by examining fitness tradeoffs across hundreds of fluconazole-resistant yeast strains. They barcoded a collection of fluconazole-resistant isolates and evolved them in different environments with a view to having relevance for evolutionary theory, medicine, and genotypephenotype mapping.

Strengths:

There are multiple strengths to this paper, the first of which is pointing out how much work has gone into it; the quality of the experiments (the thought process, the data, the figures) is excellent. Here, the authors seek to induce mutations in multiple environments, which is a really large-scale task. I particularly like the attention paid to isolates with are resistant to low concentrations of FLU. So often these are overlooked in favour of those conferring MIC values >64/128 etc. What was seen is different genotype and fitness profiles. I think there's a wealth of information here that will actually be of interest to more than just the fields mentioned (evolutionary medicine/theory).

We are very grateful for this positive review. This was indeed a lot of work! We are happy that the reviewer noted what we feel is a unique strength of our manuscript: that we survey adaptive isolates across multiple environments, including low drug concentrations.

Weaknesses:

Not picking up low fitness lineages - which the authors discuss and provide a rationale as to why. I can completely see how this has occurred during this research, and whilst it is a shame I do not think this takes away from the findings of this paper. Maybe in the next one!

We thank the reviewer for these words of encouragement and will work towards catching more low fitness lineages in our next project.

In the abstract the authors focus on 'tradeoffs' yet in the discussion they say the purpose of the study is to see how many different mechanisms of FLU resistance may exist (lines 679-680), followed up by "We distinguish mutants that likely act via different mechanisms by identifying those with different fitness tradeoffs across 12 environments". Whilst I do see their point, and this is entirely feasible, I would like a bit more explanation around this (perhaps in the intro) to help lay-readers make this jump. The remainder of my comments on 'weaknesses' are relatively fixable, I think:

We think that phrasing the “jump” as a question might help lay readers get from point A to point B. So, in the introduction of our revised manuscript, we will add a paragraph roughly similar to this one: “If two groups of drug-resistant mutants have different fitness tradeoffs, does it mean that they provide resistance through different underlying mechanisms? Alternatively, it could mean that both provide drug resistance via the same mechanism, but some mutations come with a cost that others don’t pay. However, another way to phrase this alternative is to say that both groups of mutants affect fitness through different suites of mechanisms that are only partially overlapping. And so, by identifying groups of mutants with different fitness tradeoffs, we argue that we will be uncovering sets of mutations that impact fitness through different underlying mechanisms. The ability to do so would be useful for genotype-phenotype mapping endeavors.”

In the introduction I struggle to see how this body of research fits in with the current literature, as the literature cited is a hodge-podge of bacterial and fungal evolution studies, which are very different! So example, the authors state "previous work suggests that mutants with different fitness tradeoffs may affect fitness through different molecular mechanisms" (lines 129-131) and then cite three papers, only one of which is a fungal research output. However, the next sentence focuses solely on literature from fungal research. Citing bacterial work as a foundation is fine, but as you're using yeast for this I think tailoring the introduction more to what is and isn't known in fungi would be more appropriate. It would also be great to then circle back around and mention monotherapy vs combination drug therapy for fungal infections as a rationale for this study. The study seems to be focused on FLU-resistant mutants, which is the first-line drug of choice, but many (yeast) infections have acquired resistance to this and combination therapy is the norm.

In our revised manuscript, we will carefully review all citations. The issue may stem from our attempt to reach two different groups of scientists. We ourselves are broadly interested in the structure of the genotype-phenotype-fitness map (PMID33263280, PMID32804946). Though the 3 papers the reviewer mentions on lines 132 - 133 all pertain to yeast, we cite them because they are studies about the complexity of this map. Their conclusions, in theory, should apply broadly, beyond yeast. Similarly, the reason we cite papers from yeast, as well as bacteria and cancer, is that we believe general conclusions about the genotype-phenotype-fitness map should apply broadly. For example, the sentence the reviewer highlights, “previous work suggests that mutants with different fitness tradeoffs may affect fitness through different molecular mechanisms” is a general observation about the way genotype maps to fitness. So we cited papers from across the tree of life to support this sentence.

On the other hand, because we study drug resistant mutations, we also hope that our work is of use to scientists studying the evolution of resistance. We agree with the reviewer that in this regard, some of our findings may be especially pertinent to the evolution of resistance to antifungal drugs. We will consider this when reviewing the citations in our revised manuscript and add some text to clarify these points.

Methods: Line 769 - which yeast? I haven't even seen mention of which species is being used in this study; different yeast employ different mechanisms of adaptation for resistance, so could greatly impact the results seen. This could help with some background context if the species is mentioned (although I assume S. cerevisiae).

In the revised manuscript, we will make clear that we study S. cerevisiae.

In which case, should aneuploidy be considered as a mechanism? This is mentioned briefly on line 556, but with all the sequencing data acquired this could be checked quickly?

We like this idea and we are working on it, but it is not straightforward. The reviewer is correct in that we can use the sequencing data that we already have. But calling aneuploidy with certainty is tough because its signal can be masked by noise. In other words, some regions of the genome may be sequenced more than others by chance. Given this is not straightforward, at least not for us, this analysis will likely have to wait for a subsequent paper.

I think the authors could be bolder and try and link this to other (pathogenic) yeasts. What are the implications of this work on say, Candida infections?

Perhaps because our background lies in general study of the genotype-phenotype map, we did not want to make bold assertions about how our work might apply to pathogenic yeasts. But we see how this could be helpful and will add some discussion points about this. Specifically, we will discuss which of the genes and mutants we observe are also found in Candida. We will also investigate whether our observation that low fluconazole represents a seemingly unique challenge, not just a milder version of high fluconazole, has any corollary in the Candida literature.

Author Response

Reviewer 1 (Public Review):

1. With respect to the predictions, the authors propose that the subjects, depending on their linguistic background and the length of the tone in a trial, can put forward one or two predictions. The first is a short-term prediction based on the statistics of the previous stimuli and identical for both groups (i.e. short tones are expected after long tones and vice versa). The second is a long-term prediction based on their linguistic background. According to the authors, after a short tone, Basque speakers will predict the beginning of a new phrasal chunk, and Spanish speakers will predict it after a long tone.

In this way, when a short tone is omitted, Basque speakers would experience the violation of only one prediction (i.e. the short-term prediction), but Spanish speakers will experience the violation of two predictions (i.e. the short-term and long-term predictions), resulting in a higher amplitude MMN. The opposite would occur when a long tone is omitted. So, to recap, the authors propose that subjects will predict the alternation of tone durations (short-term predictions) and the beginning of new phrasal chunks (long-term predictions).

The problem with this is that subjects are also likely to predict the completion of the current phrasal chunk. In speech, phrases are seldom left incomplete. In Spanish is very unlikely to hear a function-word that is not followed by a content-word (and the opposite happens in Basque). On the contrary, after the completion of a phrasal chunk, a speaker might stop talking and a silence might follow, instead of the beginning of a new phrasal chunk.

Considering that the completion of a phrasal chunk is more likely than the beginning of a new one, the prior endowed to the participants by their linguistic background should make us expect a pattern of results actually opposite to the one reported here.

Response: We acknowledge the plausibility of the hypothesis advanced by Reviewer #1. We would like to further clarify the rationale that led us to predict that the hypothesized long-term predictions should manifest at the onset of (and not within) a “phrasal chunk”. The hypothesis does not directly concern the probability of a short event to follow a long one (or the other way around), which to our knowledge has not been systematically quantified in previous cross-linguistic studies. Rather, it concerns how the auditory system forms higher-level auditory chunks based on the rhythmic properties of the native language, which is what the previous behavioral studies on perceptual grouping have addressed (e.g., Iversen 2008; Molnar et al. 2014; Molnar et al. 2016). When presented with sequences of two tones alternating in duration, Spanish speakers typically report perceiving the auditory stream as a repetition of short-long chunks separated by a pause, while speakers of Basque usually report the opposite long-short grouping bias. These results suggest that the auditory system performs a chunking operation by grouping pairs of tones into compressed, higher-level auditory units (often perceived as a single event). The way two constituent tones are combined depends on linguistic experience. Based on this background, we hypothesized the presence of (i) a short-term system that merely encodes a repetition of alternations rule and predicts transitions from one constituent tone to the other (a → b → a → b, etc.); (ii) a long-term system that encodes a repetition of concatenated alternations rule and predicts transitions from one high-level unit to the other (ab → ab, etc.). Under this view, we expect predictions based on the long-term system to be stronger at the onset of (rather than within) high-level units and therefore omissions of the first constituent tone to elicit larger responses than omissions of the second constituent tone.

In other words, the omission of the onset tone would reflect the omission of the whole chunk. On the other hand, the omission of the internal tone would be better handled by the short-term system, involved in processing the low-level structure of our sequences.

A similar concern was also raised by Reviewer #2. We will include the view proposed by Reviewer #1 and Reviewer #2 in the updated version of the manuscript.

1. The authors report an interaction effect that modulates the amplitude of the omission response, but caveats make the interpretation of this effect somewhat uncertain. The authors report a widespread omission response, which resembles the classical mismatch response (in MEG) with strong activations in sensors over temporal regions. Instead, the interaction found is circumscribed to four sensors that do not overlap with the peaks of activation of the omission response.

Response: We appreciate that all three reviewers agreed on the robustness of the data analysis pipeline. The approach employed to identify the presence of an interaction effect was indeed conservative, using a non-parametric test on combined gradiometers data, no a priori assumptions regarding the location of the effect, and small cluster thresholds (cfg.clusteralpha = 0.05) to enhance the likelihood of detecting highly localized clusters with large effect sizes. This approach led to the identification of the cluster illustrated in Figure 2c, where the interaction effect is evident. The fact that this interaction effect arises in a relatively small cluster of sensors does not alter its statistical robustness. The only partial overlap of the cluster with the activation peaks might simply reflect the fact that distinct sources contribute to the generation of the omission-MMN, which has been demonstrated in numerous prior studies (e.g., Zhang et al., 2018; Ross & Hamm, 2020).

Furthermore, the boxplot in Figure 2E suggests that part of the interaction effect might be due to the presence of two outliers (if removed, the effect is no longer significant). Overall, it is possible that the reported interaction is driven by a main effect of omission type which the authors report, and find consistently only in the Basque group (showing a higher amplitude omission response for long tones than for short tones). Because of these points, it is difficult to interpret this interaction as a modulation of the omission response.

Response: The two participants mentioned by Reviewer #1, despite being somewhat distant from the rest of the group, are not outliers according to the standard Tukey’s rule. As shown in Author response image 1 below, no participant fell outside the upper (Q3+1.5xIQR) and lower whiskers (Q1-1.5xIQR) of the boxplot.

Author response image 1.

The presence of a main effect of omission type does not impact the interpretation of the interaction, especially considering that these effects emerge over distinct clusters of channels.

The code to generate Author response image 1 and the corresponding statistics have been added to the script “analysis_interaction_data.R” in the OSF folder (https://osf.io/6jep8/).

It should also be noted that in the source analysis, the interaction only showed a trend in the left auditory cortex, but in its current version the manuscript does not report the statistics of such a trend.

Response: Our interpretation of the results for the present study is mainly driven by the effect observed on sensor-level data, which is statistically robust. The source modeling analyses (in non-invasive electrophysiology) provide a possible model of the candidate brain sources driving the effect observed at the sensor level. The source showing the interactive effect in our study is the left auditory cortex. More details and statistics will be provided in the reviewed version of the manuscript.

Reviewer #2 (Public Review):

1. Despite the evidence provided on neural responses, the main conclusion of the study reflects a known behavioral effect on rhythmic sequence perceptual organization driven by linguistic background (Molnar et al. 2016, particularly). Also, the authors themselves provide a good review of the literature that evidences the influence of long-term priors in neural responses related to predictive activity. Thus, in my opinion, the strength of the statements the authors make on the novelty of the findings may be a bit far-fetched in some instances.

Response: We will consider the suggestion of reviewer #2 for the new version of the manuscript. Overall, we believe that the novelty of the current study lies in bridging together findings from two research fields - basic auditory neuroscience and cross-linguistic research - to provide evidence for a predictive coding model in the auditory that uses long-term priors to make perceptual inferences.

1. Albeit the paradigm is well designed, I fail to see the grounding of the hypotheses laid by the authors as framed under the predictive coding perspective. The study assumes that responses to an omission at the beginning of a perceptual rhythmic pattern will be stronger than at the end. I feel this is unjustified. If anything, omission responses should be larger when the gap occurs at the end of the pattern, as that would be where stronger expectations are placed: if in my language a short sound occurs after a long one, and I perceptually group tone sequences of alternating tone duration accordingly, when I hear a short sound I will expect a long one following; but after a long one, I don't necessarily need to expect a short one, as something else might occur.

Response: A similar point was advanced by Reviewer #1. We tried to clarify our hypothesis (see above). We will consider including this interpretation in the updated version of the manuscript.

1. In this regard, it is my opinion that what is reflected in the data may be better accounted for (or at least, additionally) by a different neural response to an omission depending on the phase of an underlying attentional rhythm (in terms of Large and Jones rhythmic attention theory, for instance) and putative underlying entrained oscillatory neural activity (in terms of Lakatos' studies, for instance). Certainly, the fact that the aligned phase may differ depending on linguistic background is very interesting and would reflect the known behavioral effect.

Response: We thank the reviewer for this comment, which is indeed very pertinent. Below are some comments highlighting our thoughts on this.

1) We will explore in more detail the possibility that the aligned phase may differ depending on linguistic background, which is indeed very interesting. However, we believe that even if a phase modulation by language experience is found, it would not negate the possibility that the group differences in the MMN are driven by different long-term predictions. Rather, since the hypothesized phase differences would be driven by long-term linguistic experience, phase entrainment may reflect a mechanism through which long-term predictions are carried. On this point, we agree with the Reviewer when says that “this view would not change the impact of the results but add depth to their interpretation”.

2) Related to the point above: Despite evoked responses and oscillations are often considered distinct electrophysiological phenomena, current evidence suggests that these phenomena are interconnected (e.g., Studenova et al., 2023). In our view, the hypotheses that the MMN reflects differences in phase alignment and long-term prediction errors are not mutually exclusive.

3) Despite the plausibility of the view proposed by reviewer #2, many studies in the auditory neuroscience literature putatively consider the MMN as an index of prediction error (e.g., Bendixen et al., 2012; Heilbron and Chait, 2018). There are good reasons to believe that also in our study the MMN reflects, at least in part, an error response.

In the updated version of the manuscript, we will include a paragraph discussing the possibility that the reported group differences in the omission MMN might be partially accounted for by differences in neural entrainment to the rhythmic sound sequences.

Reviewer #3 (Public Review):

The main weaknesses are the strength of the effects and generalisability. The sample size is also relatively small by today's standards, with N=20 in each group. Furthermore, the crucial effects are all mostly in the .01>P<.05 range, such as the crucial interaction P=.03. It would be nice to see it replicated in the future, with more participants and other languages. It would also have been nice to see behavioural data that could be correlated with neural data to better understand the real-world consequences of the effect.

Response: We appreciate the positive feedback from Reviewer #3. Concerning this weakness highlighted: we agree with Reviewer #3 that it would be nice to see this study replicated in the future with larger sample sizes and a behavioral counterpart. Overall, we hope this work will lead to more studies using cross-linguistic/cultural comparisons to assess the effect of experience on neural processing. In the context of the present study, we believe that the lack of behavioral data does not undermine the main findings of this study, given the careful selection of the participants and the well-known robustness of the perceptual grouping effect (e.g., Iversen 2008; Yoshida et al., 2010; Molnar et al. 2014; Molnar et al. 2016). As highlighted by Reviewer #2, having Spanish and Basque dominant “speakers as a sample equates that in Molnar et al. (2016), and thus overcomes the lack of direct behavioral evidence for a difference in rhythmic grouping across linguistic groups. Molnar et al. (2016)'s evidence on the behavioral effect is compelling, and the evidence on neural signatures provided by the present study aligns with it.”

References

1. Bendixen, A., SanMiguel, I., & Schröger, E. (2012). Early electrophysiological indicators for predictive processing in audition: a review. International Journal of Psychophysiology, 83(2), 120-131.

2. Heilbron, M., & Chait, M. (2018). Great expectations: is there evidence for predictive coding in auditory cortex?. Neuroscience, 389, 54-73.

3. Iversen, J. R., Patel, A. D., & Ohgushi, K. (2008). Perception of rhythmic grouping depends on auditory experience. The Journal of the Acoustical Society of America, 124(4), 2263-2271.

4. Molnar, M., Lallier, M., & Carreiras, M. (2014). The amount of language exposure determines nonlinguistic tone grouping biases in infants from a bilingual environment. Language Learning, 64(s2), 45-64.

5. Molnar, M., Carreiras, M., & Gervain, J. (2016). Language dominance shapes non-linguistic rhythmic grouping in bilinguals. Cognition, 152, 150-159.

6. Ross, J. M., & Hamm, J. P. (2020). Cortical microcircuit mechanisms of mismatch negativity and its underlying subcomponents. Frontiers in Neural Circuits, 14, 13.

7. Simon, J., Balla, V., & Winkler, I. (2019). Temporal boundary of auditory event formation: An electrophysiological marker. International Journal of Psychophysiology, 140, 53-61.

8. Studenova, A. A., Forster, C., Engemann, D. A., Hensch, T., Sander, C., Mauche, N., ... & Nikulin, V. V. (2023). Event-related modulation of alpha rhythm explains the auditory P300 evoked response in EEG. bioRxiv, 2023-02.

9. Yoshida, K. A., Iversen, J. R., Patel, A. D., Mazuka, R., Nito, H., Gervain, J., & Werker, J. F. (2010). The development of perceptual grouping biases in infancy: A Japanese-English cross-linguistic study. Cognition, 115(2), 356-361.

10. Zhang, Y., Yan, F., Wang, L., Wang, Y., Wang, C., Wang, Q., & Huang, L. (2018). Cortical areas associated with mismatch negativity: A connectivity study using propofol anesthesia. Frontiers in Human Neuroscience, 12, 392.

Author response:

Reviewer #1 (Public review):

Summary:

This study presents a new Bayesian approach to estimate importation probabilities of malaria, combining epidemiological data, travel history, and genetic data through pairwise IBD estimates. Importation is an important factor challenging malaria elimination, especially in low-transmission settings. This paper focuses on Magude and Matutuine, two districts in southern Mozambique with very low malaria transmission. The results show isolation-by-distance in Mozambique, with genetic relatedness decreasing with distances larger than 100 km, and no spatial correlation for distances between 10 and 100 km. But again, strong spatial correlation in distances smaller than 10 km. They report high genetic relatedness between Matutuine and Inhambane, higher than between Matutuine and Magude. Inhambane is the main source of importation in Matutuine, accounting for 63.5% of imported cases. Magude, on the other hand, shows smaller importation and travel rates than Matutuine, as it is a rural area with less mobility. Additionally, they report higher levels of importation and travel in the dry season, when transmission is lower. Also, no association with importation was found for occupation, sex, and other factors. These data have practical implications for public health strategies aiming for malaria elimination, for example, testing and treating travelers from Matutuine in the dry season.

Strengths:

The strength of this study lies in the combination of different sources of data - epidemiological, travel, and genetic data - to estimate importation probabilities, and the statistical analyses.

Weaknesses:

The authors recognize the limitations related to sample size and the biases of travel reports.

Thank you for your review and consideration. As mentioned, we state in the manuscript the limitations related to sample sizes and travel reports. We aim to continue this study with new prospective data, aiming to address these limitations.

Reviewer #2 (Public review):

Summary:

Based on a detailed dataset, the authors present a novel Bayesian approach to classify malaria cases as either imported or locally acquired.

Strengths:

The proposed Bayesian approach for case classification is simple, well justified, and allows the integration of parasite genomics, travel history, and epidemiological data. The work is well-written, very organized, and brings important contributions both to malaria control efforts in Mozambique and to the scientific community. Understanding the origin of cases is essential for designing more effective control measures and elimination strategies.

Weakness:

While the authors aim to classify cases as imported or locally acquired, the work lacks a quantification of the contribution of each case type to overall transmission.

The Bayesian rationale is sound and well justified; however, the formulation appears to present an inconsistency that is replicated in both the main text and the Supplementary Material.

In fact, one of the questions that remains unanswered is the overall contribution of importation events to transmission in the areas. While the Bayesian classifier does not quantify this, our future analysis will focus on combining outbreak detection, genetic clustering and importation classification to quantify the contribution of imported cases to outbreak resurgence and to the overall transmission.

Thank you for pointing out the inconsistency in the final formula. In fact, the final formula corresponds to P(I_A | G), instead to i>P(I_A), so:

instead of

We will correct this error in a new version of the manuscript.

Reviewer #3 (Public review):

The authors present an important approach to identify imported P. falciparum malaria cases, combining genetic and epidemiological/travel data. This tool has the potential to be expanded to other contexts. The data was analyzed using convincing methods, including a novel statistical model; although some recognized limitations can be improved. This study will be of interest to researchers in public health and infectious diseases.

Strengths:

The study has several strengths, mainly the development of a novel Bayesian model that integrates genomic, epidemiological, and travel data to estimate importation probabilities. The results showed insights into malaria transmission dynamics, particularly identifying importation sources and differences in importation rates in Mozambique. Finally, the relevance of the findings is to suggest interventions focusing on the traveler population to help efforts for malaria elimination.

Weaknesses:

The study also has some limitations. The sample collection was not representative of some provinces, and not all samples had sufficient metadata for risk factor analysis, which can also be affected by travel recall bias. Additionally, the authors used a proxy for transmission intensity and assumed some conditions for the genetic variable when calculating the importation probability for specific scenarios. The weaknesses were assessed by the authors.

We acknowledge the limitations commented by the reviewer. We have the following plans to address the limitations. We will repeat the study for our data collected in 2023, which this time contains a good representation of all the provinces of Mozambique, and completeness of the metadata collection was ensured by implementing a new protocol in January 2023. Regarding the proxy for transmission intensity, we will refine the model by integrating monthly estimates of malaria incidence (previously calibrated to address testing and reporting rates) from the DHIS2 data, taking also into account the date of the reported cases in the analysis.

Author Response

We are grateful to the editors for considering our manuscript and facilitating the peer review process. Importantly, we would like to express our gratitude to reviewers for their constructive comments. Given eLife’s publishing format, we provide an initial author response now, which will be followed by a revised manuscript in the near future. Please find our responses below.

eLife Assessment

This study presents a valuable insight into a computational mechanism of pain perception. The evidence supporting the authors’ claims is solid, although the inclusion of 1) more diverse candidate computational models, 2) more systematic analysis of the temporal regularity effects on the model fit, and 3) tests on clinical samples would have strengthened the study. The work will be of interest to pain researchers working on computational models and cognitive mechanisms of pain in a Bayesian framework.

Thank you very much again for considering the manuscript and judging it as a valuable contribution to understanding mechanisms of pain perception. We recognise the above-mentioned points of improvement and elaborate on them in the initial response to the reviewers.

Reviewer 1

Reviewer Comment 1.1 — Selection of candidate computational models: While the paper juxtaposes the simple model-free RL model against a Kalman Filter model in the context of pain perception, the rationale behind this choice remains ambiguous. It prompts the question: could other RL-based models, such as model-based RL or hierarchical RL, offer additional insights? A more detailed explanation of their computational model selection would provide greater clarity and depth to the study.

Thank you for this point. Our models were selected a-priori, following the modelling strategy from Jepma et al. (2018) and hence considered the same set of core models for clear extension of the analysis to our non-cue paradigm. The key question for us was whether expectations were used to weight the behavioural estimates, so our main interest was to compare expectation vs non-expectation weighted models.

Model-based and hierarchical RL are very broad terms that can be used to refer to many different models, and we are not clear about which specific models the reviewer is referring to. Our Bayesian models are generative models, i.e. they learn the generative statistics of the environment (which is characterised by inherent stochasticity and volatility) and hence operate model-based analyses of the stimulus dynamics. In our case, this happened hierarchically and it was combined with a simple RL rule.

Reviewer Comment 1.2 — Effects of varying levels of volatility and stochasticity: The study commendably integrates varying levels of volatility and stochasticity into its experimental design. However, the depth of analysis concerning the effects of these variables on model fit appears shallow. A looming concern is whether the superior performance of the expectation-weighted Kalman Filter model might be a natural outcome of the experimental design. While the non-significant difference between eKF and eRL for the high stochasticity condition somewhat alleviates this concern, it raises another query: Would a more granular analysis of volatility and stochasticity effects reveal fine-grained model fit patterns?

We are sorry that the reviewer finds shallow ”the depth of analysis concerning the effects of these variables on model fit”. We are not sure which analysis the reviewer has in mind when suggesting a ”more granular analysis of volatility and stochasticity effects” to ”reveal fine-grained model fit patterns”. Therefore, we find it difficult to improve our manuscript in this regard. We are happy to add analyses to our paper but we would be greatful for some specific pointers. We have already provided:

• Analysis of model-naive performance across different levels of stochasticity and volatility (section 2.3, figure 3, supplementary information section 1.1 and tables S1-2)

• Model fitting for each stochasticity/volatility condition (section 2.4.1, figure 4, supplementary table S5)

• Group-level and individual-level differences of each model parameter across stochasticity/volatility conditions (supplementary information section 7, figures S4-S5).

• Effect of confidence on scaling factor for each stochasticity/volatility condition (figure 5)

Reviewer Comment 1.3 — Rating instruction: According to Fig. 1A, participants were prompted to rate their responses to the question, ”How much pain DID you just feel?” and to specify their confidence level regarding their pain. It is difficult for me to understand the meaning of confidence in this context, given that they were asked to report their subjective feelings. It might have been better to query participants about perceived stimulus intensity levels. This per- spective is seemingly echoed in lines 100-101, ”the primary aim of the experiment was to determine whether the expectations participants hold about the sequence inform their perceptual beliefs about the intensity of the stimuli.”

Thank you for raising this question, which allows us to clarify our paradigm. On half of the trials, participants were asked to report the perceived intensity of the previous stimulus; on the remaining trials, participants were requested to predict the intensity of the next stimulus. Therefore, we did query ”participants about perceived stimulus intensity levels”, as described at lines 49-55, 296-303, and depicted in figure 1.

The confidence refers to the level of confidence that participants have regarding their rating - how sure they are. This is done in addition to their perceived stimulus intensity and it has been used in a large body of previous studies in any sensory modality.

Reviewer Comment 1.4 — Relevance to clinical pain: While the authors underscore the rele- vance of their findings to chronic pain, they did not include data pertaining to clinical pain. Notably, their initial preprint seemed to encompass data from a clinical sample (https://www.medrxiv.org /content/10.1101/2023.03.23.23287656v1), which, for reasons unexplained, has been omitted in the current version. Clarification on this discrepancy would be instrumental in discerning the true relevance of the study’s findings to clinical pain scenarios.

The preprint that the Reviewer is referring to was an older version of the manuscript in which we combined two different experiments, which were initially born as separate studies: the one that we submitted to eLife (done in the lab, with noxious stimuli in healthy participants) and an online study with a different statistical learning paradigm (without noxious stimuli, in chronic back pain participants). Unfortunately, the paradigms were different and not directly comparable. Indeed, following submission to a different journal, the manuscript was criticised for this reason. We therefore split the paper in two, and submitted the first study to eLife. We are now planning to perform the same lab-based experiment with noxious stimuli on chronic back pain participants. Progress on this front has been slowed down by the fact that I (Flavia Mancini) am on maternity leave, but it remains top priority once back to work.

Reviewer Comment 1.5 — Paper organization: The paper’s organization appears a little bit weird, possibly due to the removal of significant content from their initial preprint. Sections 2.1- 2.2 and 2.4 seem more suitable for the Methods section, while 2.3 and 2.4.1 are the only parts that present results. In addition, enhancing clarity through graphical diagrams, especially for the experimental design and computational models, would be quite beneficial. A reference point could be Fig. 1 and Fig. 5 from Jepma et al. (2018), which similarly explored RL and KF models.

Thank you for these suggestions. We will consider restructuring the paper in the revised version.

Reviewer 2

Reviewer Comment 2.1 — This is a highly interesting and novel finding with potential impli- cations for the understanding and treatment of chronic pain where pain regulation is deficient. The paradigm is clear, the analysis is state-of-the-art, the results are convincing, and the interpretation is adequate.

Thank you very much for these positive comments.

Reviewer 3

We are really grateful for reviewer’s insightful comments and for providing useful guidance regarding our methodology. We are also thankful for highlighting the strengths of our manuscript. Below we respond to individual weakness mentioned in the reviews report.

Reviewer Comment 3.1 — In Figure 1, panel C, the authors illustrate the stimulation intensity, perceived intensity, and prediction intensity on the same scale, facilitating a more direct comparison. It appears that the stimulation intensity has been mathematically transformed to fit a scale from 0 to 100, aligning it with the intensity ratings corresponding to either past or future stimuli. Given that the pain threshold is specifically marked at 50 on this scale, one could logically infer that all ratings falling below this value should be deemed non-painful. However, I find myself uncertain about this interpretation, especially in relation to the term ”arbitrary units” used in the figure. I would greatly appreciate clarification on how to accurately interpret these units, as well as an explanation of the relationship between these values and the definition of pain threshold in this experiment.

Indeed, as detailed in the Methods section 4.3, the stimulation intensity was originally trans- formed from the 1-13 scale to 0-100 scale to match the scales in the participant response screens. Following the method used to establish the pain threshold, we set the stimulus intensity of 7 as the threshold on the original 1-13 scale. However, during the rating part of the experiment, several of the participants never or very rarely selected a value above 50 (their individually defined pain threshold), despite previously indicating a moment during pain threshold procedure when a stimulus becomes painful. This then results in the re-scaled intensity values as well the perception rating, both on the same 0-100 scale of arbitrary units, to never go above the pain threshold. Please see all participant ratings and inputs in the Figure below. We see that it would be more illustrative to re-plot Figure 1 with a different exemplary participant, whose ratings go above the pain threshold, perhaps with an input intensity on the 1-13 scale on the additional right-hand-side y-axis. We will add this in the revised version as well as highlight the fact above.

Importantly, while values below 50 are deemed non-painful by participants, the thermal stimulation still activates C-fibres involved in nociception, and we would argue that the modelling framework and analysis still applies in this case.

Reviewer Comment 3.2 — The method of generating fluctuations in stimulation temperatures, along with the handling of perceptual uncertainty in modelling, requires further elucidation. The current models appear to presume that participants perceive each stimulus accurately, introducing noise only at the response stage. This assumption may fail to capture the inherent uncertainty in the perception of each stimulus intensity, especially when differences in consecutive temperatures are as minimal as 1°C.

We agree with the reviewer that there are multiple sources of uncertainty involved in the process of rating the intensity of thermal stimuli - including the perception uncertainty. In order to include an account of inaccurate perception, one would have to consider different sources that contribute to this, which there may be many. In our approach, we consider one, which is captured in the expectation weighted model, more clearly exemplified in the expectation-weighted Kalman-Filter model (eKF). The model assumes participants perception of input as an imperfect indicator of the true level of pain. In this case, it turns out that perception is corrupted as a result of the expectation participants hold about the upcoming stimuli. The extent of this effect is partly governed by a subjective level of noise ϵ, which may also subsume other sources of uncertainty beyond the expectation effect. Moreover, the response noise ξ, could also subsume any other unexplained sources of noise.

Author response image 1.

Stimulis intensity transformation

Reviewer Comment 3.3 — A key conclusion drawn is that eKF is a better model than eRL. However, a closer examination of the results reveals that the two models behave very similarly, and it is not clear that they can be readily distinguished based on model recovery and model comparison results.

While, the eKF appears to rank higher than the eRL in terms of LOOIC and sigma effects, we don’t wish to make make sweeping statements regarding significance of differences between eRL and eKF, but merely point to the trend in the data. We shall make this clearer in the revised version of the manuscript. However, the most important result is that the models involving expectation-weighing are arguably better capturing the data.

Reviewer Comment 3.4 — Regarding model recovery, the distinction between the eKF and eRL models seems blurred. When the simulation is based on the eKF, there is no ability to distinguish whether either eKF or eRL is better. When the simulation is based on the eRL, the eRL appears to be the best model, but the difference with eKF is small. This raises a few more questions. What is the range of the parameters used for the simulations?

We agree that the distinction between eKF and eRL in the model recovery is not that clean-cut, which may in turn point to the similarity between the two models. To simulate the data for the model and parameter recovery analysis, we used the group means and variances estimated on the participant data to sample individual parameter values.

Reviewer Comment 3.5 — Is it possible that either eRL or eKF are best when different parameters are simulated? Additionally, increasing the number of simulations to at least 100 could provide more convincing model recovery results.

It could be a possibility, but would require further investigation and comparison of fits for different bins/ranges of parameters to see if there is any consistent advantage of one model over another is each bin. We will consider adding this analysis, and provide an additional 50 simulations to paint a more convincing picture.

Reviewer Comment 3.6 — Regarding model comparison, the authors reported that ”the expectation-weighted KF model offered a better fit than the eRL, although in conditions of high stochasticity, this difference was short of significance against the eRL model.” This interpretation is based on a significance test that hinges on the ratio between the ELPD and the surrounding standard error (SE). Unfortunately, there’s no agreed-upon threshold of SEs that determines sig- nificance, but a general guideline is to consider ”several SEs,” with a higher number typically viewed as more robust. However, the text lacks clarity regarding the specific number of SEs applied in this test. At a cursory glance, it appears that the authors may have employed 2 SEs in their interpretation, while only depicting 1 SE in Figure 4.

Indeed, we considered 2 sigma effect as a threshold, however we recognise that there is no agreed-upon threshold, and shall make this and our interpretation clearer regarding the trend in the data, in the revision.

Reviewer Comment 3.7 — With respect to parameter recovery, a few additional details could be included for completeness. Specifically, while the range of the learning rate is understandably confined between 0 and 1, the range of other simulated parameters, particularly those without clear boundaries, remains ambiguous. Including scatter plots with the simulated parameters on the x- axis and the recovered parameters on the y-axis would effectively convey this missing information. Furthermore, it would be beneficial for the authors to clarify whether the same priors were used for both the modelling results presented in the main paper and the parameter recovery presented in the supplementary material.

Thank for this comment and for the suggestions. To simulate the data for the model and parameter recovery analysis, we used the group means and variances estimated on the participant data to sample individual parameter values. The priors on the group and individual-level parameters in the recovery analysis where the same as in the fitting procedure. We will include the requested scatter plots in the next iteration of the manuscript.

Reviewer Comment 3.8 — While the reliance on R-hat values for convergence in model fitting is standard, a more comprehensive assessment could include estimates of the effective sample size (bulk ESS and/or tail ESS) and the Estimated Bayesian Fraction of Missing Information (EBFMI), to show efficient sampling across the distribution. Consideration of divergences, if any, would further enhance the reliability of the results.

Thank you very much for this suggestion, we will aim to include these measures in the revised version.

Reviewer Comment 3.9 — The authors write: ”Going beyond conditioning paradigms based in cuing of pain outcomes, our findings offer a more accurate description of endogenous pain regula- tion.” Unfortunately, this statement isn’t substantiated by the results. The authors did not engage in a direct comparison between conditioning and sequence-based paradigms. Moreover, even if such a comparison had been made, it remains unclear what would constitute the gold standard for quantifying ”endogenous pain regulation.”

This is valid point, indeed we do not compare paradigms in our study, and will remove this statement in the future version.

Author response:

Reviewer #1 (Public Review):  

Weaknesses:  

The weakness of this study lies in the fact that many of the genomic datasets originated from novel methods that were not validated with orthogonal approaches, such as DNA-FISH. Therefore, the detailed correlations described in this work are based on methodologies whose efficacy is not clearly established. Specifically, the authors utilized two modified protocols of TSA-seq for the detection of NADs (MKI67IP TSA-seq) and LADs (LMNB1-TSA-seq). Although these methods have been described in a bioRxiv manuscript by Kumar et al., they have not yet been published. Moreover, and surprisingly, Kumar et al., work is not cited in the current manuscript, despite its use of all TSA-seq data for NADs and LADs across the four cell lines. Moreover, Kumar et al. did not provide any DNA-FISH validation for their methods. Therefore, the interesting correlations described in this work are not based on robust technologies.    

An attempt to validate the data was made for SON-TSA-seq of human foreskin fibroblasts (HFF) using multiplexed FISH data from IMR90 fibroblasts (from the lung) by the Zhuang lab (Su et al., 2020). However, the comparability of these datasets is questionable. It might have been more reasonable for the authors to conduct their analyses in IMR90 cells, thereby allowing them to utilize MERFISH data for validating the TSA-seq method and also for mapping NADs and LADs. 

We disagree with the statement that the TSA-seq approach and data has not been validated by orthogonal approaches and with the conclusion that the TSA-seq approach is not robust as summarized here and detailed below in “Specific Comments”.  TSA-seq is robust because it is based only on the original immunostaining specificity provided by the primary and secondary antibodies plus the diffusion properties of the tyramide-free radical. TSA-seq has been extensively validated by microscopy and by the orthogonal genomic measurements provided by LMNB1 DamID and NAD-seq.  This includes: a) the initial validation by FISH of both nuclear speckle (to an accuracy of ~50 nm) and nuclear lamina TSA-seq  and the cross-validation of nuclear lamina TSA-seq with lamin B1 DamID in a first publication (Chen et al, JCB 2018, doi: 10.1083/jcb.201807108); b) the further validation of SON TSA-seq by FISH in a second publication ((Zhang et al, Genome Research 2021, doi:10.1101/gr.266239.120); c) the cross-validation of nucleolar TSA-seq using NAD-seq and the validation by light microscopy of the predictions of differences in the relative distributions of centromeres, nuclear speckles, and nucleoli made from nuclear speckle, nucleolar, and pericentric heterochromatin TSA-seq in the Kumar et al, bioRxiv preprint (which is in a last revision stage involving additional formatting for the journal requirements) doi:https://doi.org/10.1101/2023.10.29.564613; d) the extensive validation of nuclear speckle, LMNB1, and nucleolar TSA-seq generated in HFF human fibroblasts using published light microscopy distance measurements of hundreds of probes generated by multiplexed immuno-FISH MERFISH data (Su et al, Cell 2020, https://doi.org/10.1016/j.cell.2020.07.032), as we described for nucleolar TSA-seq in the Kumar et al, bioRxiv preprint and to some extent for LMNB1 and SON TSA-seq in the current manuscript version (see Specific Comments with attached Author response image 2).

Reviewer 1 raised concerns regarding this FISH validation given that the HFF TSA-seq and DamID data was compared to IMR90 MERFISH measurements.  The Su et al, Cell 2020 MERFISH paper came out well after the 4D Nucleome Consortium settled on HFF as one of the two main “Tier 1” cell lines.  We reasoned that the nuclear genome organization in a second fibroblast cell line would be sufficiently similar to justify using IMR90 FISH data as a proxy for our analysis of our HFF data. Indeed, there is a high correlation between the HFF TSA-seq and distances measured by MERFISH to nuclear lamina, nucleoli, and nuclear speckles (Author response image 1).  Comparing HFF SON-TSA-seq data with published IMR90 SON TSA-seq data (Alexander et al, Mol Cell 2021, doi.org/10.1016/j.molcel.2021.03.006), the HFF SON TSA-seq versus MERFISH scatterplot is very similar to the IMR90 SON TSA-seq versus MERFISH scatterplot.  We acknowledge the validation provided by the IMR90 MERFISH is limited by the degree to which genome organization relative to nuclear locales is similar in IMR90 and HFF fibroblasts. However, the correlation between measured microscopic distances from nuclear lamina, nucleoli, and nuclear speckles and TSA-seq scores is already quite high. We anticipate the conclusions drawn from such comparisons are solid and will only become that much stronger with future comparisons within the same cell line.

Author response image 1.

Scatterplots showing the correlation between TSA-seq and MERFISH microscopic distances. Top: IMR90 SON TSA-seq (from Alexander et al, Mol Cell 2021) (left) and HFF SON TSA-seq (right) (x-axis) versus distance to nuclear speckles (y-axis). Bottom: HFF Lamin B1 TSA-seq (x-axis) versus distance to nuclear lamina (y-axis) (left) and HFF MKI67IP (nucleolar) TSA-seq (x-axis) versus distance to nucleolus (y-axis) (right).

In our revision, we will add justification of the use of IMR90 fibroblasts as a proxy for HFF fibroblasts through comparison of available data sets. 

Reviewer #2 (Public Review):  

Weaknesses:  

The experiments are largely descriptive, and it is difficult to draw many cause-and-effect relationships. Similarly, the paper would be very much strengthened if the authors provided additional summary statements and interpretation of their results (especially for those not as familiar with 3D genome organization). The study would benefit from a clear and specific hypothesis.

We acknowledge that this study was hypothesis-generating rather than hypothesis-testing in its goal. This research was funded through the NIH 4D-Nucleome Consortium, which had as its initial goal the development, benchmarking, and validation of new genomic technologies.  Our Center focused on the mapping of the genome relative to different nuclear locales and the correlation of this intranuclear positioning of the genome with functions- specifically gene expression and DNA replication timing. By its very nature, this project has taken a discovery-driven versus hypothesis-driven scientific approach.  Our question fundamentally was whether we could gain new insights into nuclear genome organization through the integration of genomic and microscopic measurements of chromosome positioning relative to multiple different nuclear compartments/bodies and their correlation with functional assays such as RNA-seq and Repli-seq.

Indeed, as described in this manuscript, this study resulted in multiple new insights into nuclear genome organization as summarized in our last main figure.  We believe our work and conclusions will be of general interest to scientists working in the fields of 3D genome organization and nuclear cell biology.  We anticipate that each of these new insights will prompt future hypothesis-driven science focused on specific questions and the testing of cause-and-effect relationships. 

Given the extensive scope of this manuscript, we were limited in the extent that we could describe and summarize the background, data, analysis, and significance for every new insight. In our editing to reach the eLife recommended word count, we removed some of the explanations and summaries that we had originally included. 

As suggested by Reviewer 2, in our revision we will add back additional summary and interpretation statements to help readers unfamiliar with 3D genome organization.

Specific Comments in response to Reviewer 1:

(1)  We disagree with the comment that TSA-seq has not been cross-validated by other orthogonal genomic methods.  In the first TSA-seq paper (Chen et al, JCB 2018, doi: 10.1083/jcb.201807108), we showed a good correlation between the identification of iLADs and LADs by nuclear lamin and nuclear speckle TSA-seq and the orthogonal genomic method of lamin B1 DamID, which is reproduced using our new TSA-seq 2.0 protocol in this manuscript.  Similarly, in the Kumar et al, bioRxiv preprint (doi:https://doi.org/10.1101/2023.10.29.564613), we showed a general agreement between the identification of NADs by nucleolar TSA-seq and the orthogonal genomic method of NAD-seq.  (We expect this preprint to be in press soon; it is now undergoing a last revision involving only reformatting for journal requirements.) Additionally, we also showed a high correlation between Hi-C compartments and subcompartments and TSA-seq in the Chen et al, JCB 2018 paper. Specifically, there is an excellent correlation between the A1 Hi-C subcompartment and Speckle Associated Domains as detected by nuclear speckle TSA-seq.  Additionally, the A2 Hi-C subcompartment correlated well with iLAD regions with intermediate nuclear speckle TSA-seq scores, and the B2 and B3 Hi-C subcompartments with LADs detected by both LMNB TSA-seq and LMNB1 DamID.  More generally, Hi-C A and B compartment identity correlated well with predictions of iLADs versus LADs from nuclear speckle and nuclear lamina TSA-seq.

(2)  In the Chen et al, JCB 2018 paper we also qualitatively and quantitatively validated TSA-seq using FISH.  Qualitatively, we showed that both nuclear speckle and nuclear lamin TSA-seq correlated well with distances to nuclear speckles versus the nuclear lamina, respectively, measured by immuno-FISH.

Quantitatively, we showed that SON TSA-seq could be used to estimate the microscopic mean distance to nuclear speckles with mean and median residuals of ~50 nm.  First, we used light microscopy to show that the spreading of tyramide-biotin signal from a point-source of TSA staining fits well with the exponential decay predicted theoretically by reaction-diffusion equations assuming a steady rate of tyramide-biotin free radical generation by the HRP enzyme and a constant probability throughout the nucleus of free-radical quenching (through reaction with protein tyrosine residues and nucleic acids).  Second, we used the exponential decay constant measured by light microscopy together with FISH measurements of mean speckle distance for several genomic regions to fit an exponential function and to predict distance to nuclear speckles genome-wide directly from SON TSA-seq sequencing reads.  Third, we used this approach to test the predictions against a new set of FISH measurements, demonstrating an accuracy of these predictions of ~50 nm.

(3)  The importance of the quantitative validation by immuno-FISH of using TSA-seq to estimate mean distance to nuclear speckles is that it demonstrates the robustness of the TSA-seq approach.  Specifically, it shows how the TSA-seq signal is predicted to depend only on the specificity of the primary and secondary antibody staining and the diffusion properties of the tyramide-biotin free radicals produced by the HRP peroxidase.  This is fundamentally different from the significant dependence on antibodies and choice of marker proteins for molecular proximity assays such as DamID, ChIP-seq, and Cut and Run/Tag which depend on molecular proximity for labeling and/or pulldown of DNA.

This robustness leads to specific predictions.  First, it predicts similar TSA-seq signals will be produced using antibodies against different marker proteins against the same nuclear compartment.  This is because the exponential decay constant (distance at which the signal drops by one half) for the spreading of the TSA is in the range of several hundred nm, as measured by light microscopy for several TSA staining conditions.  Indeed, we showed in the Chen et al, JCB 2018 paper that antibodies against two different nuclear speckle proteins produced very similar TSA-seq signals while antibodies against LMNB versus LMNA also produced very similar TSA-seq signals.  Similarly, we showed in the Kumar et al preprint that antibodies against four different nucleolar proteins showed similar TSA-seq signals, with the highest correlation coefficients for the TSA-seq signals produced by the antibodies against two GC nucleolar marker proteins and the TSA-seq signals produced by the antibodies against two FC/DFC nucleolar marker proteins.

Author response image 2.

Comparison of TSA-seq data from different cell lines versus IMR90 MERFISH.  The observed correlation between SON (nuclear speckle) TSA-seq versus MERFISH is nearly as high for TSA-seq data from HFF as it is for TSA-seq data from the IMR90 cell line (Alexander et al, Mol Cell 2021) in which the MERFISH was performed. The correlations for SON, LMNB1 (nuclear lamina) and MKI67IP (nucleolus) versus MERFISH are highest for HFF TSA-seq data as compared to TSA-seq data from other cell lines (H1, K562, HCT116).  Comparison of measured distances to nuclear locale (y-axis) versus TSA-seq scores (x-axis) from different cell lines labeled in red. Left to right: SON, LMNB1, and MKI67IP.  Top to bottom: SON TSA-seq versus MERFISH for two TSA-seq replicates; TSA-seq from HFF, H1, K562, and HCT116 versus MERFISH.

Second, it predicts that the quantitative relationship between TSA-seq signal and mean distance from a nuclear compartment will depend on the convolution of the predicted exponential decay of spreading of the TSA signal produced by a point source with the more complicated staining distribution of nuclear compartments such as the nuclear lamina or nucleoli.  We successfully used this concept to explain the differences emerging between LMNB1 DamID and TSA-seq signals for flat nuclei and to recognize the polarized distribution of different LADs over the nuclear periphery.

(4)  After our genomic data production and during our data analysis, a valuable resource from the Zhuang lab was published, using MERFISH to visualize hundreds of genomic loci in IMR90 cells. We acknowledge that the much more extensive validation of TSA-seq by the multiplexed immuno-FISH MERFISH data is dependent on the degree to which the nuclear genome organization is similar between IMR90 and HFF fibroblasts.  However, the correlation between distances to nuclear speckles, nucleoli, and the nuclear lamina measured in IMR90 fibroblasts and the nuclear speckle, nucleolar, and nuclear lamina TSA-seq measured in HFF fibroblasts is already striking (See Author response image 1).  With regard to SON TSA-seq, the MERFISH versus HFF TSA-seq correlation is close to what we observe using published IMR90 SON TSA-seq data (correlation coefficients of 0.89 (IMR90 TSA-seq) versus 0.86 (HFF TSA-seq).  Moreover, this correlation is highest using TSA-seq data from HFF cells as compared to the three other cell lines. (see Author response image 2).  We believe these correlations can be considered a lower bound on the actual correlations between the FISH distances and TSA-seq that we would have observed if we had performed both assays on the same cell line. 

(5)  Currently, we still require tens of millions of cells to perform each TSA-seq assay.  This requires significant expansion of cells and a resulting increase in passage numbers of the IMR90 cells before we can perform the TSA-seq. During this expansion we observe a noticeable slowing of the IMR90 cell growth as expected for secondary cell lines as we approach the Hayflick limit.  We still do not know to what degree nuclear organization relative to nuclear locales may change as a function of cell cycle composition (ie percentage of cycling versus quiescent cells) and cell age.  Thus, even if we performed TSA-seq on IMR90 cells we would be comparing MERFISH from lower passages with a higher percentage of actively proliferating cells with TSA-seq from higher passages with a higher percentage of quiescent cells. 

We are currently working on a new TSA-seq protocol that will work with thousands of cells.  We believe it is better investment of time and resources to wait until this new protocol is optimized before we repeat TSA-seq in IMR90 cells for a better comparison with multiplexed FISH data. 

Specific Comments in response to Reviewer 2:

(1)  As we acknowledge in our Response summary, we were limited in the degree to which we could actually follow-up our findings with experiments designed to test specific hypotheses generated by our data.  However, we do want to point out that our comparison of wild-type K562 cells with the LMNA/LBR double knockout was designed to test the long-standing model that nuclear lamina association of genomic loci contributes to gene silencing.  This experiment was motivated by our surprising result that gene expression differences between cell lines correlated strongly with differences in positioning relative to nuclear speckles rather than the nuclear lamina.  Despite documenting in these double knockout cells a decreased nuclear lamina association of most LADs, and an increased nuclear lamina association of the “p-w-v” fiLADs identified in this manuscript, we saw no significant change in gene expression in any of these regions as compared to wild-type K562 cells.  Meanwhile, distances to nuclear speckles as measured by TSA-seq remained nearly constant.

We would argue that this represents a specific example in which new insights generated by our genomics comparison of cell lines led to a clear and specific hypothesis and the experimental testing of this hypothesis.

In response to Reviewer 2, we are modifying the text to make this clearer and to explicitly describe how we were testing the hypothesis that distance to nuclear lamina is correlated with but not causally linked to gene expression and how to test this hypothesis we used a DKO of LMNA and LBR to change distances relative to the nuclear lamina and to test the effect on gene expression.

Author response:

We thank the reviewers for their thorough reading and thoughtful feedback. Below, we provisionally address each of the concerns raised in the public reviews, and outline our planned revision that aims to further clarify and strengthen the manuscript.

In our response, we clarify our conceptualization of elasticity as a dimension of controllability, formalizing it within an information-theoretic framework, and demonstrating that controllability and its elasticity are partially dissociable. Furthermore, we provide clarifications and additional modeling results showing that our experimental design and modeling approach are well-suited to dissociating elasticity inference from more general learning processes, and are not inherently biased to find overestimates of elasticity. Finally, we clarify the advantages and disadvantages of our canonical correlation analysis (CCA) approach for identifying latent relationships between multidimensional data sets, and provide additional analyses that strengthen the link between elasticity estimation biases and a specific psychopathology profile.

Reviewer 1:

This research takes a novel theoretical and methodological approach to understanding how people estimate the level of control they have over their environment, and how they adjust their actions accordingly. The task is innovative and both it and the findings are well-described (with excellent visuals). They also offer thorough validation for the particular model they develop. The research has the potential to theoretically inform the understanding of control across domains, which is a topic of great importance.

We thank the reviewer for their favorable appraisal and valuable suggestions, which have helped clarify and strengthen the study’s conclusion. 

An overarching concern is that this paper is framed as addressing resource investments across domains that include time, money, and effort, and the introductory examples focus heavily on effort-based resources (e.g., exercising, studying, practicing). The experiments, though, focus entirely on the equivalent of monetary resources - participants make discrete actions based on the number of points they want to use on a given turn. While the same ideas might generalize to decisions about other kinds of resources (e.g., if participants were having to invest the effort to reach a goal), this seems like the kind of speculation that would be better reserved for the Discussion section rather than using effort investment as a means of introducing a new concept (elasticity of control) that the paper will go on to test.

We thank the reviewer for pointing out a lack of clarity regarding the kinds of resources tested in the present experiment. Investing additional resources in the form of extra tickets did not only require participants to pay more money. It also required them to invest additional time – since each additional ticket meant making another attempt to board the vehicle, extending the duration of the trial, and attentional effort – since every attempt required precisely timing a spacebar press as the vehicle crossed the screen. Given this involvement of money, time, and effort resources, we believe it would be imprecise to present the study as concerning monetary resources in particular. That said, we agree with the Reviewer that results might differ depending on the resource type that the experiment or the participant considers most. Thus, in our revision of the manuscript, we will make sure to clarify the kinds of resources the experiment involved, and highlight the open question of whether inferences concerning the elasticity of control generalize across different resource domains.

Setting aside the framing of the core concepts, my understanding of the task is that it effectively captures people's estimates of the likelihood of achieving their goal (Pr(success)) conditional on a given investment of resources. The ground truth across the different environments varies such that this function is sometimes flat (low controllability), sometimes increases linearly (elastic controllability), and sometimes increases as a step function (inelastic controllability). If this is accurate, then it raises two questions.

First, on the modeling front, I wonder if a suitable alternative to the current model would be to assume that the participants are simply considering different continuous functions like these and, within a Bayesian framework, evaluating the probabilistic evidence for each function based on each trial's outcome. This would give participants an estimate of the marginal increase in Pr(success) for each ticket, and they could then weigh the expected value of that ticket choice (Pr(success)*150 points) against the marginal increase in point cost for each ticket. This should yield similar predictions for optimal performance (e.g., opt-out for lower controllability environments, i.e., flatter functions), and the continuous nature of this form of function approximation also has the benefit of enabling tests of generalization to predict changes in behavior if there was, for instance, changes in available tickets for purchase (e.g., up to 4 or 5) or changes in ticket prices. Such a model would of course also maintain a critical role for priors based on one's experience within the task as well as over longer timescales, and could be meaningfully interpreted as such (e.g., priors related to the likelihood of success/failure and whether one's actions influence these). It could also potentially reduce the complexity of the model by replacing controllability-specific parameters with multiple candidate functions (presumably learned through past experience, and/or tuned by experience in this task environment), each of which is being updated simultaneously.

Second, if the reframing above is apt (regardless of the best model for implementing it), it seems like the taxonomy being offered by the authors risks a form of "jangle fallacy," in particular by positing distinct constructs (controllability and elasticity) for processes that ultimately comprise aspects of the same process (estimation of the relationship between investment and outcome likelihood). Which of these two frames is used doesn't bear on the rigor of the approach or the strength of the findings, but it does bear on how readers will digest and draw inferences from this work. It is ultimately up to the authors which of these they choose to favor, but I think the paper would benefit from some discussion of a common-process alternative, at least to prevent too strong of inferences about separate processes/modes that may not exist. I personally think the approach and findings in this paper would also be easier to digest under a common-construct approach rather than forcing new terminology but, again, I defer to the authors on this.

We thank the reviewer for suggesting this interesting alternative modeling approach. We agree that a Bayesian framework evaluating different continuous functions could offer advantages, particularly in its ability to generalize to other ticket quantities and prices. We will attempt to implement this as an alternative model and compare it with the current model.  

We also acknowledge the importance of avoiding a potential "jangle fallacy". We entirely agree with the Reviewer that elasticity and controllability inferences are not distinct processes. Specifically, we view resource elasticity as a dimension of controllability, hence the name of our ‘elastic controllability’ model. In response to this and other Reviewers’ comments, we now offer a formal definition of elasticity as the reduction in uncertainty about controllability due to knowing the amount of resources the agent is able and willing to invest (see further details in response to Reviewer 3 below).  

With respect to how this conceptualization is expressed in the modelling, we note that the representation in our model of maximum controllability and its elasticity via different variables is analogous to how a distribution may be represented by separate mean and variance parameters. Ultimately, even in the model suggested by the Reviewer, there would need to be a dedicated variable representing elasticity, such as the probability of sloped controllability functions. A single-process account thus allows that different aspects of this process would be differently biased (e.g., one can have an accurate estimate of the mean of a distribution but overestimate its variance). Therefore, our characterization of distinct elasticity and controllability biases (or to put it more accurately, ‘elasticity of controllability bias’ and ‘maximum controllability bias’) is consistent with a common construct account. 

That said, given the Reviewer’s comments, we believe that some of the terminology we used may have been misleading. In our planned revision, we will modify the text to clarify that we view elasticity as a dimension of controllability that can only be estimated in conjunction with controllability. 

Reviewer 2:

This research investigates how people might value different factors that contribute to controllability in a creative and thorough way. The authors use computational modeling to try to dissociate "elasticity" from "overall controllability," and find some differential associations with psychopathology. This was a convincing justification for using modeling above and beyond behavioral output and yielded interesting results. Interestingly, the authors conclude that these findings suggest that biased elasticity could distort agency beliefs via maladaptive resource allocation. Overall, this paper reveals some important findings about how people consider components of controllability.

We appreciate the Reviewer's positive assessment of our findings and computational approach to dissociating elasticity and overall controllability.

The primary weakness of this research is that it is not entirely clear what is meant by "elastic" and "inelastic" and how these constructs differ from existing considerations of various factors/calculations that contribute to perceptions of and decisions about controllability. I think this weakness is primarily an issue of framing, where it's not clear whether elasticity is, in fact, theoretically dissociable from controllability. Instead, it seems that the elements that make up "elasticity" are simply some of the many calculations that contribute to controllability. In other words, an "elastic" environment is inherently more controllable than an "inelastic" one, since both environments might have the same level of predictability, but in an "elastic" environment, one can also partake in additional actions to have additional control overachieving the goal (i.e., expend effort, money, time).

We thank the reviewer for highlighting the lack of clarity in our concept of elasticity. We first clarify that elasticity cannot be entirely dissociated from controllability because it is a dimension of controllability. If no controllability is afforded, then there cannot be elasticity or inelasticity. This is why in describing the experimental environments, we only label high-controllability, but not low-controllability, environments as ‘elastic’ or ‘inelastic’. For further details on this conceptualization of elasticity, and a planned revision of the text, see our response above to Reviewer 1. 

Second, we now clarify that controllability can also be computed without knowing the amount of resources the agent is able and willing to invest, for instance by assuming infinite resources available or a particular distribution of resource availabilities. However, knowing the agent’s available resources often reduces uncertainty concerning controllability. This reduction in uncertainty is what we define as elasticity. Since any action requires some resources, this means that no controllable environment is entirely inelastic if we also consider agents that do not have enough resources to commit any action. However, even in this case environments can differ in the degree to which they are elastic. For further details on this formal definition, see our response to Reviewer 3 below. We will make these necessary clarifications in the revised manuscript. 

Importantly, whether an environment is more or less elastic does not determine whether it is more or less controllable. In particular, environments can be more controllable yet less elastic. This is true even if we allow that investing different levels of resources (i.e., purchasing 0, 1, 2, or 3 tickets) constitute different actions, in conjunction with participants’ vehicle choices. Below, we show this using two existing definitions of controllability. 

Definition 1, reward-based controllability¹: If control is defined as the fraction of available reward that is controllably achievable, and we assume all participants are in principle willing and able to invest 3 tickets, controllability can be computed in the present task as:

where P(S' \= goal ∣ 𝑆, 𝐴, 𝐶 ) is the probability of reaching the treasure from present state 𝑆 when taking action A and investing C resources in executing the action. In any of the task environments, the probability of reaching the goal is maximized by purchasing 3 tickets (𝐶 = 3) and choosing the vehicle that leads to the goal (𝐴 = correct vehicle). Conversely, the probability of reaching the goal is minimized by purchasing 3 tickets (𝐶 = 3) and choosing the vehicle that does not lead to the goal (𝐴 = wrong vehicle). This calculation is thus entirely independent of elasticity, since it only considers what would be achieved by maximal resource investment, whereas elasticity consists of the reduction in controllability that would arise if the maximal available 𝐶 is reduced. Consequently, any environment where the maximum available control is higher yet varies less with resource investment would be more controllable and less elastic. 

Note that if we also account for ticket costs in calculating reward, this will only reduce the fraction of achievable reward and thus the calculated control in elastic environments.   

Definition 2, information-theoretic controllability²: Here controllability is defined as the reduction in outcome entropy due to knowing which action is taken:

I(S'; A, C | S) = H(S'|S) - H(S'|S, A, C)

where H(S'|S) is the conditional entropy of the distribution of outcomes S' given the present state 𝑆, and H(S'|S, A, C) is the conditional entropy of the outcome given the present state, action, and resource investment. 

To compare controllability, we consider two environments with the same maximum control:

• Inelastic environment: If the correct vehicle is chosen, there is a 100% chance of reaching the goal state with 1, 2, or 3 tickets. Thus, out of 7 possible action-resource investment combinations, three deterministically lead to the goal state (≥1 tickets and correct vehicle choice), three never lead to it (≥1 tickets and wrong vehicle choice), and one (0 tickets) leads to it 20% of the time (since walking leads to the treasure on 20% of trials).

• Elastic Environment: If the correct vehicle is chosen, the probability of boarding it is 0% with 1 ticket, 50% with 2 tickets, and 100% with 3 tickets. Thus, out of 7 possible actionresource investment combinations, one deterministically leads to the goal state (3 tickets and correct vehicle choice), one never leads to it (3 tickets and wrong vehicle choice), one leads to it 60% of the time (2 tickets and correct vehicle choice: 50% boarding + 50% × 20% when failing to board), one leads to it 10% of time (2 ticket and wrong vehicle choice), and three lead to it 20% of time (0-1 tickets).

Here we assume a uniform prior over actions, which renders the information-theoretic definition of controllability equal to another definition termed ‘instrumental divergence’3,4. We note that changing the uniform prior assumption would change the results for the two environments, but that would not change the general conclusion that there can be environments that are more controllable yet less elastic. 

Step 1: Calculating H(S'|S)

For the inelastic environment:

P(goal) = (3 × 100% + 3 × 0% + 1 × 20%)/7 = .46, P(non-goal) = .54  H(S'|S) = – [.46 × log₂(.46) + .54 × log₂(.54)] \= 1 bit

For the elastic environment:

P(goal) \= (1 × 100% + 1 × 0% + 1 × 60% + 1 × 10% + 3 × 20%)/7 \= .33, P(non-goal) \= .67  H(S'|S) = – [.33 × log₂(.33) + .67 × log₂(.67)] \= .91 bits

Step 2: Calculating H(S'|S, A, C)

Inelastic environment: Six action-resource investment combinations have deterministic outcomes entailing zero entropy, whereas investing 0 tickets has a probabilistic outcome (20%). The entropy for 0 tickets is: H(S'|C \= 0) \= -[.2 × log₂(.2) + 0.8 × log₂ (.8)] = .72 bits. Since this actionresource investment combination is chosen with probability 1/7, the total conditional entropy is approximately .10 bits

Elastic environment: 2 actions have deterministic outcomes (3 tickets with correct/wrong vehicle), whereas the other 5 actions have probabilistic outcomes:

2 tickets and correct vehicle (60% success): 

H(S'|A = correct, C = 2) = – [.6 × log₂(.6) + .4 × log₂(.4)] \= .97 bits 2 tickets and wrong vehicle (10% success): 

H(S'|A = wrong, C = 2) = – [.1 × ₂(.1) + .9 × ₂(.9)] \= .47 bits 0-1 tickets (20% success):

H(S'|C = 0-1) = – [.2 × ₂(.2) + .8 × ₂ .8)] \= .72 bits

Thus the total conditional entropy of the elastic environment is: H(S'|S, A, C) = (1/7) × .97 + (1/7) × .47 + (3/7) × .72 \= .52 bits

Step 3: Calculating I(S' | A, S)  

Inelastic environment: I(S'; A, C | S) = H(S'|S) – H(S'|S, A, C) = 1 – 0.1 = .9 bits 

Elastic environment: I(S'; A, C | S) = H(S'|S) – H(S'|S, A, C) = .91 – .52 = .39 bits

Thus, the inelastic environment offers higher information-theoretic controllability (.9 bits) compared to the elastic environment (.39 bits). 

Of note, even if each combination of cost and goal reaching is defined as a distinct outcome, then information-theoretic controllability is higher for the inelastic (2.81 bits) than for the elastic (2.30 bits) environment. 

In sum, for both definitions of controllability, we see that environments can be more elastic yet less controllable. We will amend the manuscript to clarify this distinction between controllability and its elasticity.

Reviewer 3:

A bias in how people infer the amount of control they have over their environment is widely believed to be a key component of several mental illnesses including depression, anxiety, and addiction. Accordingly, this bias has been a major focus in computational models of those disorders. However, all of these models treat control as a unidimensional property, roughly, how strongly outcomes depend on action. This paper proposes---correctly, I think---that the intuitive notion of "control" captures multiple dimensions in the relationship between action and outcome is multi-dimensional. In particular, the authors propose that the degree to which outcome depends on how much *effort* we exert, calling this dimension the "elasticity of control". They additionally propose that this dimension (rather than the more holistic notion of controllability) may be specifically impaired in certain types of psychopathology. This idea thus has the potential to change how we think about mental disorders in a substantial way, and could even help us better understand how healthy people navigate challenging decision-making problems.

Unfortunately, my view is that neither the theoretical nor empirical aspects of the paper really deliver on that promise. In particular, most (perhaps all) of the interesting claims in the paper have weak empirical support.

We appreciate the Reviewer's thoughtful engagement with our research and recognition of the potential significance of distinguishing between different dimensions of control in understanding psychopathology. We believe that all the Reviewer’s comments can be addressed with clarifications or additional analyses, as detailed below.  

Starting with theory, the elasticity idea does not truly "extend" the standard control model in the way the authors suggest. The reason is that effort is simply one dimension of action. Thus, the proposed model ultimately grounds out in how strongly our outcomes depend on our actions (as in the standard model). Contrary to the authors' claims, the elasticity of control is still a fixed property of the environment. Consistent with this, the computational model proposed here is a learning model of this fixed environmental property. The idea is still valuable, however, because it identifies a key dimension of action (namely, effort) that is particularly relevant to the notion of perceived control. Expressing the elasticity idea in this way might support a more general theoretical formulation of the idea that could be applied in other contexts. See Huys & Dayan (2009), Zorowitz, Momennejad, & Daw (2018), and Gagne & Dayan (2022) for examples of generalizable formulations of perceived control.

We thank the Reviewer for the suggestion that we formalize our concept of elasticity to resource investment, which we agree is a dimension of action. We first note that we have not argued against the claim that elasticity is a fixed property of the environment. We surmise the Reviewer might have misread our statement that “controllability is not a fixed property of the environment”. The latter statement is motivated by the observation that controllability is often higher for agents that can invest more resources (e.g., a richer person can buy more things). We will clarify this in our revision of the manuscript.

To formalize elasticity, we build on Huys & Dayan’s definition of controllability(1) as the fraction of reward that is controllably achievable, 𝜒 (though using information-theoretic definitions(2,3) would work as well). To the extent that this fraction depends on the amount of resources the agent is able and willing to invest (max 𝐶), this formulation can be probabilistically computed without information about the particular agent involved, specifically, by assuming a certain distribution of agents with different amounts of available resources. This would result in a probability distribution over 𝜒. Elasticity can thus be defined as the amount of information obtained about controllability due to knowing the amount of resources available to the agent: I(𝜒; max 𝐶). We will add this formal definition to the manuscript.  

Turning to experiment, the authors make two key claims: (1) people infer the elasticity of control, and (2) individual differences in how people make this inference are importantly related to psychopathology. Starting with claim 1, there are three sub-claims here; implicitly, the authors make all three. (1A) People's behavior is sensitive to differences in elasticity, (1B) people actually represent/track something like elasticity, and (1C) people do so naturally as they go about their daily lives. The results clearly support 1A. However, 1B and 1C are not supported. Starting with 1B, the experiment cannot support the claim that people represent or track elasticity because the effort is the only dimension over which participants can engage in any meaningful decision-making (the other dimension, selecting which destination to visit, simply amounts to selecting the location where you were just told the treasure lies). Thus, any adaptive behavior will necessarily come out in a sensitivity to how outcomes depend on effort. More concretely, any model that captures the fact that you are more likely to succeed in two attempts than one will produce the observed behavior. The null models do not make this basic assumption and thus do not provide a useful comparison.

We appreciate the reviewer's critical analysis of our claims regarding elasticity inference, which as detailed below, has led to an important new analysis that strengthens the study’s conclusions. However, we respectfully disagree with two of the Reviewer’s arguments. First, resource investment was not the only meaningful decision dimension in our task, since participant also needed to choose the correct vehicle to get to the right destination. That this was not trivial is evidenced by our exclusion of over 8% of participants who made incorrect vehicle choices more than 10% of the time. Included participants also occasionally erred in this choice (mean error rate = 3%, range [0-10%]). 

Second, the experimental task cannot be solved well by a model that simply tracks how outcomes depend on effort because 20% of the time participants reached the treasure despite failing to board their vehicle of choice. In such cases, reward outcomes and control were decoupled. Participants could identify when this was the case by observing the starting location, which was revealed together with the outcome (since depending on the starting location, the treasure location was automatically reached by walking). To determine whether participants distinguished between control-related and non-control-related reward, we have now fitted a variant of our model to the data that allows learning from each of these kinds of outcomes by means of a different free parameter. The results show that participants learned considerably more from control-related outcomes. They were thus not merely tracking outcomes, but specifically inferred when outcomes can be attributed to control. We will include this new analysis in the revised manuscript.

Controllability inference by itself, however, still does not suffice to explain the observed behavior. This is shown by our ‘controllability’ model, which learns to invest more resources to improve control, yet still fails to capture key features of participants’ behavior, as detailed in the manuscript. This means that explaining participants’ behavior requires a model that not only infers controllability—beyond merely outcome probability—but also assumes a priori that increased effort could enhance control. Building these a priori assumption into the model amounts to embedding within it an understanding of elasticity – the idea that control over the environment may be increased by greater resource investment. 

That being said, we acknowledge the value in considering alternative computational formulations of adaptation to elasticity. Thus, in our revision of the manuscript, we will add a discussion concerning possible alternative models.  

For 1C, the claim that people infer elasticity outside of the experimental task cannot be supported because the authors explicitly tell people about the two notions of control as part of the training phase: "To reinforce participants' understanding of how elasticity and controllability were manifested in each planet, [participants] were informed of the planet type they had visited after every 15 trips." (line 384).

We thank the reviewer for highlighting this point. We agree that our experimental design does not test whether people infer elasticity spontaneously. Our research question was whether people can distinguish between elastic and inelastic controllability. The results strongly support that they can, and this does have potential implications for behavior outside of the experimental task. Specifically, to the extent that people are aware that in some contexts additional resource investment improve control, whereas in other contexts it does not, then our results indicate that they would be able to distinguish between these two kinds of contexts through trial-and-error learning. That said, we agree that investigating whether and how people spontaneously infer elasticity is an interesting direction for future work. We will clarify the scope of the present conclusions in the revised manuscript.

Finally, I turn to claim 2, that individual differences in how people infer elasticity are importantly related to psychopathology. There is much to say about the decision to treat psychopathology as a unidimensional construct. However, I will keep it concrete and simply note that CCA (by design) obscures the relationship between any two variables. Thus, as suggestive as Figure 6B is, we cannot conclude that there is a strong relationship between Sense of Agency and the elasticity bias---this result is consistent with any possible relationship (even a negative one). The fact that the direct relationship between these two variables is not shown or reported leads me to infer that they do not have a significant or strong relationship in the data.

We agree that CCA is not designed to reveal the relationship between any two variables. However, the advantage of this analysis is that it pulls together information from multiple variables. Doing so does not treat psychopathology as unidimensional. Rather, it seeks a particular dimension that most strongly correlates with different aspects of task performance. This is especially useful for multidimensional psychopathology data because such data are often dominated by strong correlations between dimensions, whereas the research seeks to explain the distinctions between the dimensions. Similar considerations hold for the multidimensional task parameters, which although less correlated, may still jointly predict the relevant psychopathological profile better than each parameter does in isolation. Thus, the CCA enabled us to identify a general relationship between task performance and psychopathology that accounts for different symptom measures and aspects of controllability inference. 

Using CCA can thus reveal relationships that do not readily show up in two-variable analyses. Indeed, the direct correlation between Sense of Agency (SOA) and elasticity bias was not significant – a result that, for completeness, we will now report in the supplementary materials along with all other direct correlations. We note, however, that the CCA analysis was preregistered and its results were replicated. Furthermore, an auxiliary analysis specifically confirmed the contributions of both elasticity bias (Figure 6D, bottom plot) and, although not reported in the original paper, of the Sense of Agency score (SOA; p\=.03 permutation test) to the observed canonical correlation. Participants scoring higher on the psychopathology profile also overinvested resources in inelastic environments but did not futilely invest in uncontrollable environments (Figure 6A), providing external validation to the conclusion that the CCA captured meaningful variance specific to elasticity inference. The results thus enable us to safely conclude that differences in elasticity inferences are significantly associated with a profile of controlrelated psychopathology to which SOA contributed significantly.  

Finally, whereas interpretation of individual CCA loadings that were not specifically tested remains speculative, we note that the pattern of loadings largely replicated across the initial and replication studies (see Figure 6B), and aligns with prior findings. For instance, the positive loadings of SOA and OCD match prior suggestions that a lower sense of control leads to greater compensatory effort(7), whereas the negative loading for depression scores matches prior work showing reduced resource investment in depression(5-6).

We will revise the text to better clarify the advantageous and disadvantageous of our analytical approach, and the conclusions that can and cannot be drawn from it.

There is also a feature of the task that limits our ability to draw strong conclusions about individual differences in elasticity inference. As the authors clearly acknowledge, the task was designed "to be especially sensitive to overestimation of elasticity" (line 287). A straightforward consequence of this is that the resulting *empirical* estimate of estimation bias (i.e., the gamma_elasticity parameter) is itself biased. This immediately undermines any claim that references the directionality of the elasticity bias (e.g. in the abstract). Concretely, an undirected deficit such as slower learning of elasticity would appear as a directed overestimation bias. When we further consider that elasticity inference is the only meaningful learning/decisionmaking problem in the task (argued above), the situation becomes much worse. Many general deficits in learning or decision-making would be captured by the elasticity bias parameter. Thus, a conservative interpretation of the results is simply that psychopathology is associated with impaired learning and decision-making.

We apologize for our imprecise statement that the task was ‘especially sensitive to overestimation of elasticity’, which justifiably led to Reviewer’s concern that slower elasticity learning can be mistaken for elasticity bias. To make sure this was not the case, we made use of the fact that our computational model explicitly separates bias direction (λ) from the rate of learning through two distinct parameters, which initialize the prior concentration and mean of the model’s initial beliefs concerning elasticity (see Methods pg. 22). The higher the concentration of the initial beliefs (𝜖), the slower the learning. Parameter recovery tests confirmed that our task enables acceptable recovery of both the bias λ_elasticity (r=.81) and the concentration 𝝐_elasticity (r=.59) parameters. And importantly, the level of confusion between the parameters was low (confusion of 0.15 for 𝝐_elasticity→ λ_elasticity and 0.04 for λ_elasticity→ 𝝐_elasticity). This result confirms that our task enables dissociating elasticity biases from the rate of elasticity learning. 

Moreover, to validate that the minimal level of confusion existing between bias and the rate of learning did not drive our psychopathology results, we re-ran the CCA while separating concentration from bias parameters. The results (Author response image 1) demonstrate that differences in learning rate (𝜖) had virtually no contribution to our CCA results, whereas the contribution of the pure bias (𝜆) was preserved. 

We will incorporate these clarifications and additional analysis in our revised manuscript.

Author response image 1.

Showing that a model parameter correlates with the data it was fit to does not provide any new information, and cannot support claims like "a prior assumption that control is likely available was reflected in a futile investment of resources in uncontrollable environments." To make that claim, one must collect independent measures of the assumption and the investment.

We apologize if this and related statements seemed to be describing independent findings. They were merely meant to describe the relationship between model parameters and modelindependent measures of task performance. It is inaccurate, though, to say that they provide no new information, since results could have been otherwise. For instance, instead of a higher controllability bias primarily associating with futile investment of resources in uncontrollable environments, it could have been primarily associated with more proper investment of resources in high-controllability environments. Additionally, we believe these analyses are of value to readers who seek to understand the role of different parameters in the model. In our planned revision, we will clarify that the relevant analyses are merely descriptive. 

Did participants always make two attempts when purchasing tickets? This seems to violate the intuitive model, in which you would sometimes succeed on the first jump. If so, why was this choice made? Relatedly, it is not clear to me after a close reading how the outcome of each trial was actually determined.

We thank the reviewer for highlighting the need to clarify these aspects of the task in the revised manuscript. 

When participants purchased two extra tickets, they attempted both jumps, and were never informed about whether either of them succeeded. Instead, after choosing a vehicle and attempting both jumps, participants were notified where they arrived at. This outcome was determined based on the cumulative probability of either of the two jumps succeeding. Success meant that participants arrived at where their chosen vehicle goes, whereas failure meant they walked to the nearest location (as determined by where they started from). 

Though it is unintuitive to attempt a second jump before seeing whether the first succeed, this design choice ensured two key objectives. First, that participants would consistently need to invest not only more money but also more effort and time in planets with high elastic controllability. Second, that the task could potentially generalize to the many real-world situations where the amount of invested effort has to be determined prior to seeing any outcome, for instance, preparing for an exam or a job interview. 

It should be noted that the model is heuristically defined and does not reflect Bayesian updating. In particular, it overestimates control by not using losses with less than 3 tickets (intuitively, the inference here depends on your beliefs about elasticity). I wonder if the forced three-ticket trials in the task might be historically related to this modeling choice.

We apologize for not making this clear, but in fact losing with less than 3 tickets does reduce the model’s estimate of available control. It does so by increasing the elasticity estimates

(a_elastic≥1, a_elastic2 parameters), signifying that more tickets are needed to obtain the maximum available level of control, thereby reducing the average controllability estimate across ticket investment options. 

It would be interesting to further develop the model such that losing with less than 3 tickets would also impact inferences concerning the maximum available control, depending on present beliefs concerning elasticity, but the forced three-ticket purchases already expose participants to the maximum available control, and thus, the present data may not be best suited to test such a model. These trials were implemented to minimize individual differences concerning inferences of maximum available control, thereby focusing differences on elasticity inferences. We will discuss the Reviewer’s suggestion for a potentially more accurate model in the revised manuscript. 

References

(1) Huys, Q. J. M., & Dayan, P. (2009). A Bayesian formulation of behavioral control. Cognition, 113(3), 314– 328.

(2) Ligneul, R. (2021). Prediction or causation? Towards a redefinition of task controllability. Trends in Cognitive Sciences, 25(6), 431–433.

(3) Mistry, P., & Liljeholm, M. (2016). Instrumental divergence and the value of control. Scientific Reports, 6, 36295.

(4) Lin, J. (1991). Divergence measures based on the Shannon entropy. IEEE Transactions on Information Theory, 37(1), 145–151

(5) Cohen RM, Weingartner H, Smallberg SA, Pickar D, Murphy DL. Effort and cognition in depression. Arch Gen Psychiatry. 1982 May;39(5):593-7. doi: 10.1001/archpsyc.1982.04290050061012. PMID: 7092490.

(6) Bi R, Dong W, Zheng Z, Li S, Zhang D. Altered motivation of effortful decision-making for self and others in subthreshold depression. Depress Anxiety. 2022 Aug;39(8-9):633-645. doi: 10.1002/da.23267. Epub 2022 Jun 3. PMID: 35657301; PMCID: PMC9543190.

(7) Tapal, A., Oren, E., Dar, R., & Eitam, B. (2017). The Sense of Agency Scale: A measure of consciously perceived control over one's mind, body, and the immediate environment. Frontiers in Psychology, 8, 1552

Author response: 

We thank the reviewers for their feedback on our paper. We have taken all their comments into account in revising the manuscript. We provide a point-by-point response to their comments, below.

Reviewer #1:

Major comments:

The manuscript is clearly written with a level of detail that allows others to reproduce the imaging and cell-tracking pipeline. Of the 22 movies recorded one was used for cell tracking. One movie seems sufficient for the second part of the manuscript, as this manuscript presents a proof-of-principle pipeline for an imaging experiment followed by cell tracking and molecular characterisation of the cells by HCR. In addition, cell tracking in a 5-10 day time-lapse movie is an enormous time commitment.

My only major comment is regarding "Suppl_data_5_spineless_tracking". The image file does not load.

It looks like the wrong file is linked to the mastodon dataset. The "Current BDV dataset path" is set to "Beryl_data_files/BLB mosaic cut movie-02.xml", but this file does not exist in the folder. Please link it to the correct file.

We have corrected the file path in the updated version of Suppl. Data 5.

Minor comments:

The authors state that their imaging settings aim to reduce photo damage. Do they see cell death in the regenerating legs? Is the cell death induced by the light exposure or can they tell if the same cells die between the movies? That is, do they observe cell death in the same phases of regeneration and/or in the same regions of the regenerating legs?

Yes, we observe cell death during Parhyale leg regeneration. We have added the following sentence to explain this in the revised manuscript: "During the course of regeneration some cells undergo apoptosis (reported in Alwes et al., 2016). Using the H2B-mRFPruby marker, apoptotic cells appear as bright pyknotic nuclei that break up and become engulfed by circulating phagocytes (see bright specks in Figure 2F)."

We now also document apoptosis in regenerated legs that have not been subjected to live imaging in a new supplementary figure (Suppl. Figure 3),  and we refer to these observations as follows: "While some cell death might be caused by photodamage, apoptosis can also be observed in similar numbers in regenerating legs that have not been subjected to live imaging (Suppl. Figure 3)."

Based on 22 movies, the authors divide the regeneration process into three phases and they describe that the timing of leg regeneration varies between individuals. Are the phases proportionally the same length between regenerating legs or do the authors find differences between fast/slow regenerating legs? If there is a difference in the proportions, why might this be?

Both early and late phases contribute to variation in the speed of regeneration, but there is no clear relationship between the relative duration of each phase and the speed of regeneration. We now present graphs supporting these points in a new supplementary figure (Suppl. Figure 2).  

To clarify this point, we have added the following sentence in the manuscript: "We find that the overall speed of leg regeneration is determined largely by variation in the speed of the early (wound closure) phase of regeneration, and to a lesser extent by variation in later phases when leg morphogenesis takes place (Suppl. Figure 2 A,B). There is no clear relationship between the relative duration of each phase and the speed of regeneration (Suppl. Figure 2 A',B')."

Based on their initial cell tracing experiment, could the authors elaborate more on what kind of biological information can be extracted from the cell lineages, apart from determining which is the progenitor of a cell? What does it tell us about the cell population in the tissue? Is there indication of multi- or pluripotent stem cells? What does it say about the type of regeneration that is taking place in terms of epimorphosis and morphallaxis, the old concepts of regeneration?

In the first paragraph of Future Directions we describe briefly the kind of biological information that could be gained by applying our live imaging approach with appropriate cell-type markers (see below). We do not comment further, as we do not currently have this information at hand. Regarding the concepts of epimorphosis and morphallaxis, as we explain in Alwes et al. 2016, these terms describe two extreme conditions that do not capture what we observe during Parhyale leg regeneration. Our current work does not bring new insights on this topic.

Page 5. The authors mention the possibility of identifying the cell ID based on transcriptomic profiling data. Can they suggest how many and which cell types they expect to find in the last stage based on their transcriptomic data?

We have added this sentence: "Using single-nucleus transcriptional profiling, we have identified approximately 15 transcriptionally-distinct cell types in adult Parhyale legs (Almazán et al., 2022), including epidermis, muscle, neurons, hemocytes, and a number of still unidentified cell types."

Page 6. Correction: "..molecular and other makers.." should be "..molecular and other markers.."

Corrected

Page 8. The HCR in situ protocol probably has another important advantage over the conventional in situ protocol, which is not mentioned in this study. The hybridisation step in HCR is performed at a lower temperature (37˚C) than in conventional in situ hybridisation (65˚C, Rehm et al., 2009). In other organisms, a high hybridisation temperature affects the overall tissue morphology and cell location (tissue shrinkage). A lower hybridisation temperature has less impact on the tissue and makes manual cell alignment between the live imaging movie and the fixed HCR in situ stained specimen easier and more reliable. If this is also the case in Parhyale, the authors must mention it.

This may be correct, but all our specimens were treated at 37˚C, so we cannot assess whether hybridisation temperature affects morphological preservation in our specimens.

Page 9. The authors should include more information on the spineless study. What been is spineless? What do the cell lineages tell about the spineless progenitors, apart from them being spread in the tissue at the time of amputation? Do spineless progenitors proliferate during regeneration? Do any spineless expressing cells share a common progenitor cell?

We now point out that spineless encodes a transcription factor. We provide a summary of the lineages generating spineless-expressing cells in Suppl. Figure 6, and we explain that "These epidermal progenitors undergo 0, 1 or 2 cell divisions, and generate mostly spineless-expressing cells (Suppl. Figure 5)."

Page 10. Regarding the imaging temperature, the Materials and Methods state "... a temperature control chamber set to 26 or 27˚C..."; however, in Suppl. Data 1, 26˚C and 29˚C are indicated as imaging temperatures. Which is correct?

We corrected the Methods by adding "with the exception of dataset li51, imaged at 29°C"

Page 10. Regarding the imaging step size, the Materials and Methods state "...step size of 1-2.46 µm..."; however, Suppl. Data 1 indicate a step size between 1.24 - 2.48 µm. Which is correct?

We corrected the Methods.

Page 11. Correct "...as the highest resolution data..." to "...at the highest resolution data..."

The original text is correct ("standardised to the same dimensions as the highest resolution data").

Page 11. Indicate which supplementary data set is referred to: "Using Mastodon, we generated ground truth annotations on the original image dataset, consisting of 278 cell tracks, including 13,888 spots and 13,610 links across 55 time points (see Supplementary Data)."

Corrected

p. 15. Indicate which supplementary data set is referred to: "In this study we used HCR probes for the Parhyale orthologues of futsch (MSTRG.441), nompA (MSTRG.6903) and spineless (MSTRG.197), ordered from Molecular Instruments (20 oligonucleotides per probe set). The transcript sequences targeted by each probe set are given in the Supplementary Data."

Corrected

Figure 3. Suggestion to the overview schematics: The authors might consider adding "molting" as the end point of the red bar (representing differentiation).

The time of molting is not known in the majority of these datasets, because the specimens were fixed and stained prior to molting. We added the relevant information in the figure legend: "Datasets li-13 and li-16 were recorded until the molt; the other recordings were stopped before molting."

Figure 4B': Please indicate that the nuclei signal is DAPI.

Corrected

Supplementary figure 1A. Word is missing in the figure legend: ...the image also shows weak…

Corrected

Supplementary Figure 2: Please indicate the autofluorescence in the granular cells. Does it correspond to the yellow cells?

Corrected

Video legend for video 1 and 2. Please correct "H2B-mREFruby" to "H2B-mRFPruby".

Corrected

Reviewer #2:

Major comments:

MC 1. Given that most of the technical advances necessary to achieve the work described in this manuscript have been published previously, it would be helpful for the authors to more clearly identify the primary novelty of this manuscript. The abstract and introduction to the manuscript focus heavily on the technical details of imaging and analysis optimization and some additional summary of the implications of these advances should be included here to aid the reader.

This paper describes a technical advance. While previous work (Alwes et al. 2016) established some key elements of our live imaging approach, we were not at that time able to record the entire time course of leg regeneration (the longest recordings were 3.5 days long). Here we present a method for imaging the entire course of leg regeneration (up to 10 days of imaging), optimised to reduce photodamage and to improve cell tracking. We also develop a method of in situ staining in cuticularised adult legs (an important technical breakthrough in this experimental system), which we combine with live imaging to determine the fate of tracked cells. We have revised the abstract and introduction of the paper to point out these novelties, in relation to our previous publications.

In the abstract we explain: "Building on previous work that allowed us to image different parts of the process of leg regeneration in the crustacean Parhyale hawaiensis, we present here a method for live imaging that captures the entire process of leg regeneration, spanning up to 10 days, at cellular resolution. Our method includes (1) mounting and long-term live imaging of regenerating legs under conditions that yield high spatial and temporal resolution but minimise photodamage, (2) fixing and in situ staining of the regenerated legs that were imaged, to identify cell fates, and (3) computer-assisted cell tracking to determine the cell lineages and progenitors of identified cells. The method is optimised to limit light exposure while maximising tracking efficiency."

The introduction includes the following text: "Our first systematic study using this approach presented continuous live imaging over periods of 2-3 days, capturing key events of leg regeneration such as wound closure, cell proliferation and morphogenesis of regenerating legs with single-cell resolution (Alwes et al., 2016). Here, we extend this work by developing a method for imaging the entire course of leg regeneration, optimised to reduce photodamage and to improve cell tracking. We also develop a method of in situ staining of gene expression in cuticularised adult legs, which we combine with live imaging to determine the fate of tracked cells."

MC 2. The description of the regeneration time course is nicely detailed but also very qualitative. A major advantage of continuous recording and automated cell tracking in the manner presented in this manuscript would be to enable deeper quantitative characterization of cellular and tissue dynamics during regeneration. Rather than providing movies and manually annotated timelines, some characterization of the dynamics of the regeneration process (the heterogeneity in this is very very interesting, but not analyzed at all) and correlating them against cellular behaviors would dramatically increase the impact of the work and leverage the advances presented here. For example, do migration rates differ between replicates? Division rates? Division synchrony? Migration orientation? This seems to be an incredibly rich dataset that would be fascinating to explore in greater detail, which seems to me to be the primary advance presented in this manuscript. I can appreciate that the authors may want to segregate some biological findings from the method, but I believe some nominal effort highlighting the quantitative nature of what this method enables would strengthen the impact of the paper and be useful for the reader. Selecting a small number of simple metrics (eg. Division frequency, average cell migration speed) and plotting them alongside the qualitative phases of the regeneration timeline that have already been generated would be a fairly modest investment of effort using tools that already exist in the Mastodon interface, I would roughly estimate on the order of an hour or two per dataset. I believe that this effort would be well worth it and better highlight a major strength of the approach.

The primary goal of this work was to establish a robust method for continuous long-term live imaging of regeneration, but we do appreciate that a more quantitative analysis would add value to the data we are presenting. We tried to address this request in three steps:

First, we examined whether clear temporal patterns in cell division, cell movements or other cellular features can be observed in an accurately tracked dataset (li13-t4, tracked in Sugawara et al. 2022). To test this we used the feature extraction functions now available on the Mastodon platform (see link). We could discern a meaningful temporal pattern for cell divisions (see below); the other features showed no interpretable pattern of variation.

Second, we asked whether we could use automated cell tracking to analyse the patterns of cell division in all our datasets. Using an Elephant deep learning model trained on the tracks of the li13-t4 dataset, we performed automated cell tracking in the same dataset, and compared the pattern of cell divisions from the automated cell track predictions with those coming from manually validated cell tracks. We observed that the automated tracks gave very imprecise results, with a high background of false positives obscuring the real temporal pattern (see images below, with validated data on the left, automated tracking on the right). These results show that the automated cell tracking is not accurate enough to provide a meaningful picture on the pattern of cell divisions.

Third, we tried to improve the accuracy of detection of dividing cells by additional training of Elephant models on each dataset (to lower the rate of false positives), followed by manual proofreading. Given how labour intensive this is, we could only apply this approach to 4 additional datasets. The results of this analysis are presented in Figure 4.

Author response image 1.

MC 3. The authors describe the challenges faced by their described approach:

Using this mode of semi-automated and manual cell tracking, we find that most cells in the upper slices of our image stacks (top 30 microns) can be tracked with a high degree of confidence. A smaller proportion of cell lineages are trackable in the deeper layers.

Given that the authors quantify this in Table 1, it would aid the reader to provide metrics in the manuscript text at this point. Furthermore, the metrics provided in Table 1 appear to be for overall performance, but the text describes that performance appears to be heavily depth dependent. Segregating the performance metrics further, for example providing DET, TRA, precision and recall for superficial layers only and for the overall dataset, would help support these arguments and better highlight performance a potential adopter of the method might expect.

In the revised manuscript we have added data on the tracking performance of Elephant in relation to imaging depth in Suppl. Figure 3. These data confirm our original statement (which was based on manual tracking) that nuclei are more challenging to track in deeper layers.

We point to these new results in two parts of the paper, as follows: "A smaller proportion of cells are trackable in the deeper layers (see Suppl. Figure 3)", and "Our results, summarised in Table 1A, show that the detection of nuclei can be enhanced by doubling the z resolution at the expense of xy resolution and image quality. This improvement is particularly evident in the deeper layers of the imaging stacks, which are usually the most challenging to track (Suppl. Figure 3)."

MC 4. Performance characterization in Table 1 appears to derive from a single dataset that is then subsampled and processed in different ways to assess the impact of these changes on cell tracking and detection performance. While this is a suitable strategy for this type of optimization it leaves open the question of performance consistency across datasets. I fully recognize that this type of quantification can be onerous and time consuming, but some attempt to assess performance variability across datasets would be valuable. Manual curation over a short time window over a random sampling of the acquired data would be sufficient to assess this.

We think that similar trade-offs will apply to all our datasets because tracking performance is constrained by the same features, which are intrinsic to our system; e.g. by the crowding of nuclei in relation to axial resolution, or the speed of mitosis in relation to the temporal resolution of imaging. We therefore do not see a clear rationale for repeating this analysis. On a practical level, our existing image datasets could not be subsampled to generate the various conditions tested in Table 1, so proving this point experimentally would require generating new recordings, and tracking these to generate ground truth data. This would require months of additional work.

A second, related question is whether Elephant would perform equally well in detecting and tracking nuclei across different datasets. This point has been addressed in the Sugawara et al. 2022 paper, where the performance of Elephant was tested on diverse fluorescence datasets.

Reviewer #3:

Major comments:

• The authors should clearly specify what are the key technical improvements compared to their previous studies (Alwes et al. 2016, Elife; Konstantinides & Averof 2014, Science). There, the approaches for mounting, imaging, and cell tracking are already introduced, and the imaging is reported to run for up to 7 days in some cases.

In Konstantinides and Averof (2014) we did not present any live imaging at cellular resolution. In Alwes et al. (2016) we described key elements of our live imaging approach, but we were never able to record the entire time course of leg regeneration. The longest recordings in that work were 3.5 days long.

We have revised the abstract and introduction to clarify the novelty of this work, in relation to our previous publications. Please see our response to comment MC1 of reviewer 2.

• While the authors mention testing the effect of imaging parameters (such as scanning speed and line averaging) on the imaging/tracking outcome, very little or no information is provided on how this was done beyond the parameters that they finally arrived to.

Scan speed and averaging parameters were determined by measuring contrast and signal-to-noise ratios in images captured over a range of settings. We have now added these data in Supplementary Figure 1.

• The authors claim that, using the acquired live imaging data across entire regeneration time course, they are now able to confirm and extend their description of leg regeneration. However, many claims about the order and timing of various cellular events during regeneration are supported only by references to individual snapshots in figures or supplementary movies. Presenting a more quantitative description of cellular processes during regeneration from the acquired data would significantly enhance the manuscript and showcase the usefulness of the improved workflow.

The events we describe can be easily observed in the maximum projections, available in Suppl. Data 2. Regarding the quantitative analysis, please see our response to comment MC2 of reviewer 2.  

• Table 1 summarizes the performance of cell tracking using simulated datasets of different quality. However only averages and/or maxima are given for the different metrics, which makes it difficult to evaluate the associated conclusions. In some cases, only 1 or 2 test runs were performed.

The metrics extracted from each of the three replicates, per dataset, are now included in Suppl. Data 4.

We consistently used 3 replicates to measure tracking performance with each of the datasets. The "replicates" column label in Table 1 referred to the number of scans that were averaged to generate the image, not to the replicates used for estimating the tracking performance. To avoid confusion, we changed that label to "averaging".

• OPTIONAL: An imaging approach that allows using the current mounting strategy but could help with some of the tradeoffs is using a spinning-disk confocal microscope instead of a laser scanning one. If the authors have such a system available, it could be interesting to compare it with their current scanning confocal setup.

Preliminary experiments that we carried out several years ago on a spinning disk confocal (with a 20x objective and the CSU-W1 spinning disk) were not very encouraging, and we therefore did not pursue this approach further. The main problem was bad image quality in deeper tissue layers.

Minor comments:

• The presented imaging protocol was optimized for one laser wavelength only (561 nm) - this should be mentioned when discussing the technical limitations since animals tend to react differently to different wavelengths. Same settings might thus not be applicable for imaging a different fluorescent protein.

In the second paragraph of the Results section, we explain that we perform the imaging at long wavelengths in order to minimise photodamage. It should be clear to the readers that changing the excitation wavelength will have an impact for long-term live imaging.

• For transferability, it would be useful if the intensity of laser illumination was measured and given in the Methods, instead of just a relative intensity setting from the imaging software. Similarly,more details of the imaging system should be provided where appropriate (e.g., detector specifications).

We have now measured the intensity of the laser illumination and added this information in the

Methods: "Laser power was typically set to 0.3% to 0.8%, which yields 0.51 to 1.37 µW at 561 nm (measured with a ThorLabs Microscope Slide Power Sensor, #S170C)."

Regarding the imaging system and the detector, we provide all the information that is available to us on the microscope's technical sheets.

• The versions of analysis scripts associated with the manuscript should be uploaded to an online repository that permanently preserves the respective version.

The scripts are now available on gitbub and online repositories. The relevant links are included in the revised manuscript.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The image analysis pipeline is tested in analysing microscopy imaging data of gastruloids of varying sizes, for which an optimised protocol for in toto image acquisition is established based on whole mount sample preparation using an optimal refractive index matched mounting media, opposing dual side imaging with two-photon microscopy for enhanced laser penetration, dual view registration, and weighted fusion for improved in toto sample data representation. For enhanced imaging speed in a two-photon microscope, parallel imaging was used, and the authors performed spectral unmixing analysis to avoid issues of signal cross-talk.

In the image analysis pipeline, different pre-treatments are done depending on the analysis to be performed (for nuclear segmentation - contrast enhancement and normalisation; for quantitative analysis of gene expression - corrections for optical artifacts inducing signal intensity variations). Stardist3D was used for the nuclear segmentation. The study analyses into properties of gastruloid nuclear density, patterns of cell division, morphology, deformation, and gene expression.

Strengths:

The methods developed are sound, well described, and well-validated, using a sample challenging for microscopy, gastruloids. Many of the established methods are very useful (e.g. registration, corrections, signal normalisation, lazy loading bioimage visualisation, spectral decomposition analysis), facilitate the development of quantitative research, and would be of interest to the wider scientific community.

We thank the reviewer for this positive feedback.

Weaknesses:

A recommendation should be added on when or under which conditions to use this pipeline.

We thank the reviewer for this valuable feedback, which will be addressed in the revision. In general, the pipeline is applicable to any tissue, but it is particularly useful for large and dense 3D samples—such as organoids, embryos, explants, spheroids, or tumors—that are typically composed of multiple cell layers and have a thickness greater than 50 µm.

The processing and analysis pipeline are compatible with any type of 3D imaging data (e.g. confocal, 2 photon, light-sheet, live or fixed).

- Spectral unmixing to remove signal cross-talk of multiple fluorescent targets is typically more relevant in two-photon imaging due to the broader excitation spectra of fluorophores compared to single-photon imaging. In confocal or light-sheet microscopy, alternating excitation wavelengths often circumvents the need for unmixing. Spectral decomposition performs even better with true spectral detectors; however, these are usually not non-descanned detectors, which are more appropriate for deep tissue imaging. Our approach demonstrates that simultaneous cross-talk-free four-color two-photon imaging can be achieved in dense 3D specimen with four non-descanned detectors and co-excitation by just two laser lines. Depending on the dispersion in optically dense samples, depth-dependent apparent emission spectra need to be considered.

- Nuclei segmentation using our trained StarDist3D model is applicable to any system under two conditions: (1) the nuclei exhibit a star-convex shape, as required by the StarDist architecture, and (2) the image resolution is sufficient in XYZ to allow resampling. The exact sampling required is object- and system-dependent, but the goal is to achieve nearly isotropic objects with diameters of approximately 15 pixels while maintaining image quality. In practice, images containing objects that are natively close to or larger than 15 pixels in diameter should segment well after resampling. Conversely, images with objects that are significantly smaller along one or more dimensions will require careful inspection of the segmentation results.

- Normalization is broadly applicable to multicolor data when at least one channel is expected to be ubiquitously expressed within its domain. Wavelength-dependent correction requires experimental calibration using either an ubiquitous signal at each wavelength. Importantly, this calibration only needs to be performed once for a given set of experimental conditions (e.g., fluorophores, tissue type, mounting medium).

- Multi-scale analysis of gene expression and morphometrics is applicable to any 3D multicolor image. This includes both the 3D visualization tools (Napari plugins) and the various analytical plots (e.g., correlation plots, radial analysis). Multi-scale analysis can be performed even with imperfect segmentation, as long as segmentation errors tend to cancel out when averaged locally at the relevant spatial scale. However, systematic errors—such as segmentation uncertainty along the Z-axis due to strong anisotropy—may accumulate and introduce bias in downstream analyses. Caution is advised when analyzing hollow structures (e.g., curved epithelial monolayers with large cavities), as the pipeline was developed primarily for 3D bulk tissues, and appropriate masking of cavities would be needed.

Reviewer #2 (Public review):

Summary:

This study presents an integrated experimental and computational pipeline for high-resolution, quantitative imaging and analysis of gastruloids. The experimental module employs dual-view two-photon spectral imaging combined with optimized clearing and mounting techniques to image whole-mount immunostained gastruloids. This approach enables the acquisition of comprehensive 3D images that capture both tissue-scale and single-cell level information.

The computational module encompasses both pre-processing of acquired images and downstream analysis, providing quantitative insights into the structural and molecular characteristics of gastruloids. The pre-processing pipeline, tailored for dual-view two-photon microscopy, includes spectral unmixing of fluorescence signals using depth-dependent spectral profiles, as well as image fusion via rigid 3D transformation based on content-based block-matching algorithms. Nuclei segmentation was performed using a custom-trained StarDist3D model, validated against 2D manual annotations, and achieving an F1 score of 85+/-3% at a 50% intersection-over-union (IoU) threshold. Another custom-trained StarDist3D model enabled accurate detection of proliferating cells and the generation of 3D spatial maps of nuclear density and proliferation probability. Moreover, the pipeline facilitates detailed morphometric analysis of cell density and nuclear deformation, revealing pronounced spatial heterogeneities during early gastruloid morphogenesis.

All computational tools developed in this study are released as open-source, Python-based software.

Strengths:

The authors applied two-photon microscopy to whole-mount deep imaging of gastruloids, achieving in toto visualization at single-cell resolution. By combining spectral imaging with an unmixing algorithm, they successfully separated four fluorescent signals, enabling spatial analysis of gene expression patterns.

The entire computational workflow, from image pre-processing to segmentation with a custom-trained StarDist3D model and subsequent quantitative analysis, is made available as open-source software. In addition, user-friendly interfaces are provided through the open-source, community-driven Napari platform, facilitating interactive exploration and analysis.

We thank the reviewer for this positive feedback.

Weaknesses:

The computational module appears promising. However, the analysis pipeline has not been validated on datasets beyond those generated by the authors, making it difficult to assess its general applicability.

We agree that applying our analysis pipeline to published datasets—particularly those acquired with different imaging systems—would be valuable. However, only a few high-resolution datasets of large organoid samples are publicly available, and most of these either lack multiple fluorescence channels or represent 3D hollow structures. Our computational pipeline consists of several independent modules: spectral filtering, dual-view registration, local contrast enhancement, 3D nuclei segmentation, image normalization based on a ubiquitous marker, and multiscale analysis of gene expression and morphometrics.

Spectral filtering has already been applied in other systems (e.g. [7] and [8]), but is here extended to account for imaging depth-dependent apparent emission spectra of the different fluorophores. In our pipeline, we provide code to run spectral filtering on multichannel images, integrated in Python. In order to apply the spectral filtering algorithm utilized here, spectral patterns of each fluorophore need to be calibrated as a function of imaging depth, which depend on the specific emission windows and detector settings of the microscope.

Image normalization using a wavelength-dependent correction also requires calibration on a given imaging setup to measure the difference in signal decay among the different fluorophores species. To our knowledge, the calibration procedures for spectral-filtering and our image-normalization approach have not been performed previously in 3D samples, which is why validation on published datasets is not readily possible. Nevertheless, they are described in detail in the Methods section, and the code used—from the calibration measurements to the corrected images—is available open-source at the Zenodo link in the manuscript.

Dual-view registration, local contrast enhancement, and multiscale analysis of gene expression and morphometrics are not limited to organoid data or our specific imaging modalities. If we identify suitable datasets to validate these modules, we will include them in the revised manuscript.

To evaluate our 3D nuclei segmentation model, we plan to test it on diverse systems, including gastruloids stained with the nuclear marker Draq5 from Moos et al. [1]; breast cancer spheroids; primary ductal adenocarcinoma organoids; human colon organoids and HCT116 monolayers from Ong et al. [2]; and zebrafish tissues imaged by confocal microscopy from Li et al [3]. These datasets were acquired using either light-sheet or confocal microscopy, with varying imaging parameters (e.g., objective lens, pixel size, staining method).

Preliminary results are promising (see Author response image 1). We will provide quantitative comparisons of our model’s performance on these datasets, using annotations or reference predictions provided by the original authors where available.

Author response image 1.

Qualitative comparison of our custom Stardist3D segmentation strategy on diverse published 3D nuclei datasets. We show one slice from the XY plane for simplicity. (a) Gastruloid stained with the nuclear marker DRAQ5 imaged with an open-top dual-view and dual-illumination LSM [1]. (b) Breast cancer spheroid [2]. (c) Primary pancreatic ductal adenocarcinoma organoids imaged with confocal microscopy[2]. (d) Human colon organoid imaged with LSM laser scanning confocal microscope [2]. (e) Monolayer HCT116 cells imaged with LSM laser scanning confocal microscope [2]. (f) Fixed zebrafish embryo stained for nuclei and imaged with a Zeiss LSM 880 confocal microscopy [3].

Besides, the nuclei segmentation component lacks benchmarking against existing methods.

We agree with the reviewer that a benchmark against existing segmentation methods would be very useful. We tried different pre-trained models:

- CellPose, which we tested in a previous paper ([4]) and which showed poor performances compared to our trained StarDist3D model.

- DeepStar3D ([2]) is only available in the software 3DCellScope. We could not benchmark the model on our data, because the free and accessible version of the software is limited to small datasets. An image of a single whole-mount gastruloid with one channel, having dimensions (347,467,477) was too large to be processed, see screenshot below. The segmentation model could not be extracted from the source code and tested externally because the trained DeepStar3D weights are encrypted.

Author response image 2.

Screenshot of the 3DCellScore software. We could not perform 3D nuclei segmentation of a whole-mount gastruloids because the image size was too large to be processed.

- AnyStar ([5]), which is a model trained from the StarDist3D architecture, was not performing well on our data because of the heterogeneous stainings. Basic pre-processing such as median and gaussian filtering did not improve the results and led to wrong segmentation of touching nuclei. AnyStar was demonstrated to segment well colon organoids in Ong et al, 2025 ([2]), but the nuclei were more homogeneously stained. Our Hoechst staining displays bright chromatin spots that are incorrectly labeled as individual nuclei.

- Cellos ([6]), another model trained from StarDist3D, was also not performing well. The objects used for training and to validate the results are sparse and not touching, so the predicted segmentation has a lot of false negatives even when lowering the probability threshold to detect more objects. Additionally, the network was trained with an anisotropy of (9,1,1), based on images with low z resolution, so it performed poorly on almost isotropic images. Adapting our images to the network’s anisotropy results in an imprecise segmentation that can not be used to measure 3D nuclei deformations.

We tried both Cellos and AnyStar predictions on a gastruloid image from Fig. S2 of our main manuscript. Author response image 3 displays the results qualitatively compared to our trained model Stardist-tapenade. For the revision of the paper, we will perform a comprehensive benchmark of these state-of-the-art routines, including quantitative assessment of the performance.

Author response image 3.

Qualitative comparison of two published segmentation models versus our model. We show one slice from the XY plane for simplicity. Segmentations are displayed with their contours only. (Top left) Gastruloid stained with Hoechst, image extracted from Fig S2 of our manuscript. (Top right) Same image overlayed with the prediction from the Cellos model, showing many false negatives. (Bottom left) Same image overlayed with the prediction from our Stardist-tapenade model. (Bottom right) Same image overlayed with the prediction from the AnyStar model, false positives are indicated with a red arrow.

Appraisal:

The authors set out to establish a quantitative imaging and analysis pipeline for gastruloids using dual-view two-photon microscopy, spectral unmixing, and a custom computational framework for 3D segmentation and gene expression analysis. This aim is largely achieved. The integration of experimental and computational modules enables high-resolution in toto imaging and robust quantitative analysis at the single-cell level. The data presented support the authors' conclusions regarding the ability to capture spatial patterns of gene expression and cellular morphology across developmental stages.

Impact and utility:

This work presents a compelling and broadly applicable methodological advance. The approach is particularly impactful for the developmental biology community, as it allows researchers to extract quantitative information from high-resolution images to better understand morphogenetic processes. The data are publicly available on Zenodo, and the software is released on GitHub, making them highly valuable resources for the community.

We thank the reviewer for these positive feedbacks.

Reviewer #3 (Public review):

Summary

The paper presents an imaging and analysis pipeline for whole-mount gastruloid imaging with two-photon microscopy. The presented pipeline includes spectral unmixing, registration, segmentation, and a wavelength-dependent intensity normalization step, followed by quantitative analysis of spatial gene expression patterns and nuclear morphometry on a tissue level. The utility of the approach is demonstrated by several experimental findings, such as establishing spatial correlations between local nuclear deformation and tissue density changes, as well as the radial distribution pattern of mesoderm markers. The pipeline is distributed as a Python package, notebooks, and multiple napari plugins.

Strengths

The paper is well-written with detailed methodological descriptions, which I think would make it a valuable reference for researchers performing similar volumetric tissue imaging experiments (gastruloids/organoids). The pipeline itself addresses many practical challenges, including resolution loss within tissue, registration of large volumes, nuclear segmentation, and intensity normalization. Especially the intensity decay measurements and wavelength-dependent intensity normalization approach using nuclear (Hoechst) signal as reference are very interesting and should be applicable to other imaging contexts. The morphometric analysis is equally well done, with the correlation between nuclear shape deformation and tissue density changes being an interesting finding. The paper is quite thorough in its technical description of the methods (which are a lot), and their experimental validation is appropriate. Finally, the provided code and napari plugins seem to be well done (I installed a selected list of the plugins and they ran without issues) and should be very helpful for the community.

We thank the reviewer for his positive feedback and appreciation of our work.

Weaknesses

I don't see any major weaknesses, and I would only have two issues that I think should be addressed in a revision:

(1) The demonstration notebooks lack accompanying sample datasets, preventing users from running them immediately and limiting the pipeline's accessibility. I would suggest to include (selective) demo data set that can be used to run the notebooks (e.g. for spectral unmixing) and or provide easily accessible demo input sample data for the napari plugins (I saw that there is some sample data for the processing plugin, so this maybe could already be used for the notebooks?).

We thank the reviewer for this relevant suggestion. The 7 notebooks were updated to automatically download sample tests. The different parts of the pipeline can now be run immediately: https://github.com/GuignardLab/tapenade/tree/chekcs_on_notebooks/src/tapenade/notebooks

(2) The results for the morphometric analysis (Figure 4) seem to be only shown in lateral (xy) views without the corresponding axial (z) views. I would suggest adding this to the figure and showing the density/strain/angle distributions for those axial views as well.

We agree with the reviewer that a morphometric analysis based on the axial views would be informative and plan to perform this analysis for the revision.

(1) Moos, F., Suppinger, S., de Medeiros, G., Oost, K.C., Boni, A., Rémy, C., Weevers, S.L., Tsiairis, C., Strnad, P. and Liberali, P., 2024. Open-top multisample dual-view light-sheet microscope for live imaging of large multicellular systems. Nature Methods, 21(5), pp.798-803.

(2) Ong, H.T., Karatas, E., Poquillon, T., Grenci, G., Furlan, A., Dilasser, F., Mohamad Raffi, S.B., Blanc, D., Drimaracci, E., Mikec, D. and Galisot, G., 2025. Digitalized organoids: integrated pipeline for high-speed 3D analysis of organoid structures using multilevel segmentation and cellular topology. Nature Methods, 22(6), pp.1343-1354.

(3) Li, L., Wu, L., Chen, A., Delp, E.J. and Umulis, D.M., 2023. 3D nuclei segmentation for multi-cellular quantification of zebrafish embryos using NISNet3D. Electronic Imaging, 35, pp.1-9.

(4) Vanaret, J., Dupuis, V., Lenne, P. F., Richard, F., Tlili, S., & Roudot, P. (2023). A detector-independent quality score for cell segmentation without ground truth in 3D live fluorescence microscopy. IEEE Journal of Selected Topics in Quantum Electronics, 29(4: Biophotonics), 1-12.

(5) Dey, N., Abulnaga, M., Billot, B., Turk, E. A., Grant, E., Dalca, A. V., & Golland, P. (2024). AnyStar: Domain randomized universal star-convex 3D instance segmentation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 7593-7603).

(6) Mukashyaka, P., Kumar, P., Mellert, D. J., Nicholas, S., Noorbakhsh, J., Brugiolo, M., ... & Chuang, J. H. (2023). High-throughput deconvolution of 3D organoid dynamics at cellular resolution for cancer pharmacology with Cellos. Nature Communications, 14(1), 8406.

(7) Rakhymzhan, A., Leben, R., Zimmermann, H., Günther, R., Mex, P., Reismann, D., ... & Niesner, R. A. (2017). Synergistic strategy for multicolor two-photon microscopy: application to the analysis of germinal center reactions in vivo. Scientific reports, 7(1), 7101.

(8) Dunsing, V., Petrich, A., & Chiantia, S. (2021). Multicolor fluorescence fluctuation spectroscopy in living cells via spectral detection. Elife, 10, e69687.

Author Response:

Reviewer #1 (Public Review):

Summary:

This manuscript reports the substrate-bound structure of SiaQM from F. nucleatum, which is the membrane component of a Neu5Ac-specific Tripartite ATP-dependent Periplasmic (TRAP) transporter. Until recently, there was no experimentally derived structural information regarding the membrane components of the TRAP transporter, limiting our understanding of the transport mechanism. Since 2022, there have been 3 different studies reporting the structures of the membrane components of Neu5Ac-specific TRAP transporters. While it was possible to narrow down the binding site location by comparing the structures to proteins of the same fold, a structure with substrate bound has been missing. In this work, the authors report the Na+-bound state and the Na+ plus Neu5Ac state of FnSiaQM, revealing information regarding substrate coordination. In previous studies, 2 Na+ ion sites were identified. Here, the authors also tentatively assign a 3rd Na+ site. The authors reconstitute the transporter to assess the effects of mutating the binding site residues they identified in their structures. Of the 2 positions tested, only one of them appears to be critical to substrate binding.

Strengths:

The main strength of this work is the capture of the substrate-bound state of SiaQM, which provides insight into an important part of the transport cycle.

Weaknesses:

The main weakness is the lack of experimental validation of the structural findings. The authors identified the Neu5Ac binding site, but only tested 2 residues for their involvement in substrate interactions, which was very limited. The authors tentatively identified a 3rd Na+ binding site, which if true would be an impactful finding, but this site was not tested for its contribution to Na+ dependent transport, and the authors themselves report that the structural evidence is not wholly convincing. This lack of experimental validation undermines the confidence of the findings. However, the reporting of these new data is important as it will facilitate follow-up studies by the authors or other researchers.

The main concern, also mentioned by other reviewers, is the lack of mutational data and functional studies on the identified binding sites. Two other structures of TRAP transporters have been determined, one from Haemophilus influenzae (Hi) and the other from Photobacterium profundum (Pp). We will refer to the references in this paper as [1], Peter et al. as [2], and Davies et al. as [3]. The table below lists all the mutations made in the Neu5Ac binding site, including direct polar interactions between Neu5Ac and the side chains, as well as the newly identified metal sites.

The structure of Fusobacterium nucleatum (Fn) that we have reported shows a significant sequence identity with the previously reported Hi structure. When we superimpose the Pp and Fn structures, we observe that nearly all the residues that bind to the Neu5Ac and the third metal site are conserved. This suggests that mutagenesis and functional studies from other research can be related to the structure presented in our work.

The table below shows that all three residues that directly interact with Neu5Ac have been tested by site-directed mutagenesis for their role in Neu5Ac transport. Both D521 and S300 are critical for transport, while S345 is not. We do not believe that a mutation of D521A in Fn, followed by transport studies, will provide any new information.

However, Peter et al. have mutated only one of the 5 residues near the newly identified metal binding site, which resulted in no transport. The rest of the residues have not been functionally tested. We propose to mutate these residues into Ala, express and purify the proteins, and then carry out transport assays on those that show expression. We will include this information in the revised manuscript.

Reviewer #2 (Public Review):

In this exciting new paper from the Ramaswamy group at Purdue, the authors provide a new structure of the membrane domains of a tripartite ATP-independent periplasmic (TRAP) transporter for the important sugar acid, N-acetylneuraminic acid or sialic acid (Neu5Ac). While there have been a number of other structures in the last couple of years (the first for any TRAP-T) this is the first to trap the structure with Neu5Ac bound to the membrane domains. This is an important breakthrough as in this system the ligand is delivered by a substrate-binding protein (SBP), in this case, called SiaP, where Neu5Ac binding is well studied but the 'hand over' to the membrane component is not clear. The structure of the membrane domains, SiaQM, revealed strong similarities to other SBP-independent Na+-dependent carriers that use an elevator mechanism and have defined Na+ and ligand binding sites. Here they solve the cryo-EM structure of the protein from the bacterial oral pathogen Fusobacterium nucleatum and identify a potential third (and theoretically predicted) Na+ binding site but also locate for the first time the Neu5Ac binding site. While this sits in a region of the protein that one might expect it to sit, based on comparison to other transporters like VcINDY, it provides the first molecular details of the binding site architecture and identifies a key role for Ser300 in the transport process, which their structure suggests coordinates the carboxylate group of Neu5Ac. The work also uses biochemical methods to confirm the transporter from F. nucleatum is active and similar to those used by selected other human and animal pathogens and now provides a framework for the design of inhibitors of these systems.

The strengths of the paper lie in the locating of Neu5Ac bound to SiaQM, providing important new information on how TRAP transporters function. The complementary biochemical analysis also confirms that this is not an atypical system and that the results are likely true for all sialic acid-specific TRAP systems.

The main weakness is the lack of follow-up on the identified binding site in terms of structure-function analysis. While Ser300 is shown to be important, only one other residue is mutated and a much more extensive analysis of the newly identified binding site would have been useful.

Please see the comments above.

Reviewer #3 (Public Review):

The manuscript by Goyal et al reports substrate-bound and substrate-free structures of a tripartite ATP-independent periplasmic (TRAP) transporter from a previously uncharacterized homolog, F. nucleatum. This is one of the most mechanistically fascinating transporter families, by means of its QM domain (the domain reported in his manuscript) operating as a monomeric 'elevator', and its P domain functioning as a substrate-binding 'operator' that is required to deliver the substrate to the QM domain; together, this is termed an 'elevator with an operator' mechanism. Remarkably, previous structures had not demonstrated the substrate Neu5Ac bound. In addition, they confirm the previously reported Na+ binding sites and report a new metal binding site in the transporter, which seems to be mechanistically relevant. Finally, they mutate the substrate binding site and use proteoliposomal uptake assays to show the mechanistic relevance of the proposed substrate binding residues.

The structures are of good quality, the functional data is robust, the text is well-written, and the authors are appropriately careful with their interpretations. Determination of a substrate-bound structure is an important achievement and fills an important gap in the 'elevator with an operator' mechanism. Nevertheless, I have concerns with the data presentation, which in its current state does not intuitively demonstrate the discussed findings. Furthermore, the structural analysis appears limited, and even slight improvements in data processing and resulting resolution would greatly improve the authors' claims. I have several suggestions to hopefully improve the clarity and quality of the manuscript.

We appreciate your feedback and will make the necessary modifications to the manuscript incorporating most of the suggestions. We will submit the revised version once the experiments are completed. We are also working on improving the quality of the figures and have made several attempts to enhance the resolution using CryoSPARC or RELION, but without success. We will continue to explore newer methods in an effort to achieve higher resolution and to model more lipids, particularly in the binding pocket.

Author Response

Joint Public Review

Strengths

Overall, the idea that the PAG interacts with the BLA via the midline thalamus during a predator vs. foraging test is new and quite interesting. The authors have used appropriate tools to address their questions. The major impact in the field would be to add evidence to claims that the BLA can be downstream of the dPAG to evoke defensive behaviors. The study also adds to a body of evidence that the PAG mediates primal fear responses.

Weaknesses

(Anatomical concerns)

1) The authors claim that the recordings were performed in the dorsal PAG (dPAG), but the histological images in Fig. 1B and Supplementary S2 for example show the tip of the electrode in a different subregion of PAG (ventral/lateral). They should perform a more careful histological analysis of the recording sites and explain the histological inclusion and exclusion criteria. Diagrams showing the sites of all PAG and BLA recordings, as well as all fiber optics, would be helpful.

The PAG is composed of dorsomedial (dm), dorsolateral (dl), lateral (l), and ventrolateral (vl) columns that extend along the rostro-caudal axis of the aqueduct. The term “dorsal PAG” (dPAG) generally encompasses dmPAG, dlPAG, and lPAG, as substantiated by track-tracing, neurochemical, and immunohistochemical techniques (e.g., Bandler et al., 1991; Bandler & Keay, 1996; Carrive, 1993). As Bandler and Shipley (1994) summarized, “These findings suggest that what has been traditionally called the 'dorsal PAG' (a collective term for regions dorsal and lateral to the aqueduct), consists of three anatomically distinct longitudinal columns: dorsomedial and lateral columns…and a dorsolateral column…" Similarly, Schenberg et al. (2005) clarified in their review that, “According to this parcellation...the defensive behaviors (freezing, flight or fight) and aversion-related responses (switchoff behavior) were ascribed to the DMPAG, DLPAG, and LPAG (usually named the ‘dorsal’ PAG).” In our study, all recordings were conducted within the dPAG. Also, Figures 1B and S2 in our manuscript correspond to the -6.04 mm template from Paxinos & Watson’s atlas (1998), which is shown in the left panel in Author response image 1 and is considerably anterior to the location where the vlPAG emerges, as shown in the right panel. In our revised manuscript, we will provide a detailed definition of the dPAG, inclusive of dmPAG, dlPAG, and lPAG, and support this with the referenced literature.

Author response image 1.

2) Prior studies investigating the role of BLA neurons during a foraging vs. robot test similar to the one used in this study should be also cited and discussed (e.g., Amir et al 2019; Amir et al 2015). These two studies demonstrated that most neurons in the basal portion of the BLA exhibit inhibitory activity during foraging behavior and only a small fraction of neurons (~4%) display excitatory activity in response to the robot (in contrast to the 25% reported in the present study). A very accurate histological analysis of BLA recording sites should be performed to clarify whether distinct subregions of the BLA encode foraging and predator-related information, as previously shown in the two described studies.

In the revised manuscript, we will discuss papers by Amir et al. (2015) and Amir et al. (2019) that utilized a similar 'approach food-avoid predator' paradigm. These studies found a correlation between the neuronal activities in the basolateral amygdala (BL) and the velocity of animal movement during foraging, regardless of the presence or absence of predators. Specifically, the majority of BL neurons were inhibited in both conditions, with only 4.5% being responsive to predators. Consequently, Amir et al. posited that amygdala activity predominantly aligns with behavioral output such as foraging, rather than with responses to threats.

In contrast, our body of work (Kim et al., 2018; Kong et al., 2021; the present study) reveals that the majority of neurons in the BA/BLA displayed distinct responses in pre-robot and robot sessions. Kong et al. (2021) discussed in depth several factors that may account for this discrepancy, given that both Amir et al. and our research used similar behavioral paradigms. Differences in apparatus features, experimental procedures, and data analysis methodologies (refer to Amir et al., 2019) could be contributing to the conflicting results and interpretations concerning the significance of amygdalar neuronal activities.

Additionally, our studies uniquely monitored the same set of amygdalar neurons during pre-robot and robot sessions, affording us the opportunity for a direct comparison of neuronal activities under different threat conditions.

Another salient difference lines in the foraging success rates, which were markedly higher in Amir et al (~80%) compared to our studies (<3-4%). We hypothesize that there may be an inverse relationship between the pellet procurement rate and the intensity of fear. The high foraging success rate in Amir et al., which correlates with subdued amygdalar activity, stands in contrast to our findings of heightened amygdalar activity associated with a lower foraging success rate. Supporting this notion, optogeneticallyinduced amygdalar activity led naïve rats to abandon foraging and escape to the nest (Kong et al., 2021, the present study).

3) An important claim of this study that the PAG sends predator-related signals to BLA via the PVT (Fig. 4). The authors stated that PVT neurons labeled by intra-BLA injection of the retrograde tracer CTB were activated by the predator, but a proper immunohistochemical quantification with a control group was not provided to support this claim. To provide better support for their claim, the authors should quantify the doublelabeled PVT neurons (cFos plus CTB positive neurons) during the robot test.

As recommended, we will include a revised Fig. 4 in the manuscript to present the quantification of neurons that are double-labeled with c-Fos and CTB in the PVT. This updated figure will provide a more rigorous analysis and visual representation of the data.

4) The AVV anterograde tracer deposit spread to a large part of the PAG, including dorsolateral and lateral PAG, and supraoculomotor regions (Fig. 4B). Is the projection to the PVT from the dPAG or other regions of the PAG?

As previously addressed in response to Comment #1, the dPAG comprises the dmPAG, dlPAG, and lPAG. In the revised manuscript, we will acknowledge the diffusion of the AAV to the adjacent deep gray layer of the superior colliculus. Additionally, we are considering conducting more restricted AAV injections into the dPAG to verify terminal expressions in the PVT.

(Concerns about the strength of the evidence supporting a role for the PVT)

5) The authors conclude in the discussion section that the dPAG-amygdala pathway is involved in generating antipredatory defensive behavior. However, the current results are entirely based on correlational analyses of neural firing rate and there is no direct demonstration that the PAG provides information about the robot to the BLA. Therefore, the authors should tone down their interpretation or provide more evidence to support it by performing experiments applying inhibitory tools in the dPAG > PVT > BLA pathway and examining the impact on behavior and downstream neural firing.

As suggested, we will moderate the assertions about the functional implications of the PVT, based on the data from anterograde and retrograde tracers, to present a more measured interpretation in the manuscript.

(Other concerns)

6) One of the main findings of this study is the observation that BLA neurons that are responsive to PAG photostimulation are preferentially recruited during the foraging vs. robot test (Fig. 3). However, the experimental design used to address this question is problematic because the laser photostimulation of PAG neurons preceded the foraging vs. robot test. Prior photoactivation of PAG may have caused indirect shortterm synaptic plasticity in BLA cells, which would favor the response of these cells to the robot. Please see Oishi et al, 2019 PMID: 30621738, which demonstrated that 10 trains of 20Hz photoactivation (300 pulses each) was sufficient to induce LTP in brain slices.

After approximately eight photostimulation trials of the dPAG, with 40 pulses each, the animals entered a post-photostimulation testing phase (referred to as "Post"; Fig. 3C), lasting 10-15 minutes over an average of eight trials before robot testing. Although the PAG does not directly project to the BLA, the remote possibility of trans-synaptic plasticity in the BLA cannot be completely excluded and will be acknowledged. Additionally, it is noteworthy that Oishi et al's (2019) study applied a total of 3,000 pulses (i.e., 10 15-s trains of 20-Hz pulses) and investigated CA3-CA3 synaptic plasticity, as opposed to a total of 320 pulses (i.e., 8 2-s trains of 20-Hz pulses) in our study.

7) The authors should perform a longitudinal analysis of the behavioral responses of the rats across the trials to clarify whether the animals habituate to the robot or not. In Figure 1E, it appears that PAG neurons fire less across the trials, which could be associated with behavioral habituation to the predator robot. If that is the case, the activity of many other PAG and BLA neurons will also most likely vary according to the trial number, which would impact the current interpretation of the results.

In Figure 1E, the y-axis represents the Z scores of individual dPAG neurons, instead of representing repeated tests of the same neuron across multiple trials. The raster plot in Figure 1F clearly depicts that the same dPAG neurons consistently display heightened neural activity in response to the approaching robot across successive trials.

8) In Figure 1, it is unclear why the authors compared the activity of neurons that respond to the robot activation against the activity of the neurons during the retrieval of the food pellets in the pre-robot and postrobot sessions. The best comparison would be aligning the cells that were responsive to the activation of the robot with the moment in which the animals run back to the nest after consuming the pellets during the prerobot or post-robot sessions. This would enable the authors to demonstrate that the PAG responses are directly associated with the expression of escaping behavior in the presence of the robot rather than associated with the onset of goal-directed movement in direction to the next during the pre- and post-robot sessions. A graphic showing the correlation between PAG firing rate and escape response would be also informative.

Figure 1E compares the dPAG neural activity when animals enter a designated pellet zone (time-stamped by camera tracking) during both pre-robot and post-robot trials to the dPAG neural activity when entering the robot trigger zone (time-stamped by robot activation). We wish to clarify that rats carry the large (0.5 g) pellet back to the nest for consumption rather than consume it in the open arena before returning to the nest.

In our study, we aimed to investigate the direct response of dPAG neurons to the looming predator and explore the communication between dPAG and BLA in relation to antipredatory defensive responses. To build upon our previous research that suggests a potential role of dPAG in conveying such responses to the BLA (Kim et al., 2013) and the immediate firing of BLA neurons in response to predatory threats (Kim et al., 2018; Kong et al., 2021), we chose to narrow our testing window to a short latency period (< 500 ms) following robot activations. This specific time window allowed us to focus on the initial stages of the threat stimulus processing and minimize potential confounding factors such as the presence of residual firing activity triggered by the robot during the animals’ escape or any activity changes induced by the animals' behavior.

Furthermore, Figure S1C clearly demonstrates that (i) increased activity of dPAG robot cells preceded the animals’ actual turning and fleeing behavior toward the nest, as indicated by the peak values of movement speed (dark yellow), and (ii) the presence of pellets did not affect activity changes of the robot cells during pre- and post-robot sessions. These observations suggest that the heightened activity of dPAG robot cells was not due to movement changes or pellet motivation.

Lastly, as stated in the original manuscript, the vast majority of robot cells (90.9%) did not show significant correlations between movement speed and firing rates, lending further support to the interpretation that the dPAG activity observed was not merely a reflection of movement changes.

References

Bandler, R., Carrive, P., & Depaulis, A. (1991). Emerging principles of organization of the midbrain periaqueductal gray matter. The midbrain periaqueductal gray matter: functional, anatomical, and neurochemical organization, 1-8.

Bandler, R. & Keay, K. A. (1996). Columnar organization in the midbrain periaqueductal gray and the integration of emotional expression. Progress in brain research, 107, 285-300.

Bandler, R. & Shipley, M. T. (1994) Columnar organization in the midbrain periaqueductal gray: modules for emotional expression? Trends in Neurosciences, 17(9), 379-89.

Carrive, P. (1993). The periaqueductal gray and defensive behavior: functional representation and neuronal organization. Behavioural brain research, 58(1-2), 27-47.

Oishi, N., Nomoto, M., Ohkawa, N., Saitoh, Y., Sano, Y., Tsujimura, S., ... & Inokuchi, K. (2019). Artificial association of memory events by optogenetic stimulation of hippocampal CA3 cell ensembles. Molecular brain, 12, 1-10.

Paxinos, G. & Watson, C. (1998). The Rat Brain in Stereotaxic Coordinates. Academic Press, San Diego. Schenberg, L. C., Póvoa, R. M. F., Costa, A. L. P., Caldellas, A. V., Tufik, S., & Bittencourt, A. S. (2005). Functional specializations within the tectum defense systems of the rat. Neuroscience & Biobehavioral Reviews, 29(8), 1279-1298.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

This work by Ding et al uses agent-based simulations to explore the role of the structure of molecular motor myosin filaments in force generation in cytoskeletal structures. The focus of the study is on disordered actin bundles which can occur in the cell cytoskeleton and have also been investigated with in vitro purified protein experiments.

Strengths:

The key finding is that cooperative effects between multiple myosin filaments can enhance both total force and the efficiency of force generation (force per myosin). These trends were possible to obtain only because the detailed structure of the motor filaments with multiple heads is represented in the model.

We appreciate your comments about the strength of our study.

Weaknesses:

It is not clearly described what scientific/biological questions about cellular force production the work answers. There should be more discussion of how their simulation results compare with existing experiments or can be tested in future experiments.

Thank you for the comment. First, our study explains why non-muscle myosin II in stress fibers shows focal distributions rather than uniform distributions; if they stay closely, they can generate much larger forces in the stress fibers via the cooperative overlap. Our study also predicts a difference between bipolar structures (found in skeletal muscle myosins and non-muscle myosins) and side polar structures (found in smooth muscle myosins) in terms of the likelihood of the cooperative overlap. As shown below, myosin filaments with the bipolar structure can add up their forces better than those with the side polar structure when their overlap level is the same. We will add discussion about these in the revised manuscript.

Author response image 1.

As the reviewer noticed, our results were briefly compared with prior observations in Ref. 4 (Thoresen et al., Biophys J, 2013) where different myosin isoforms were used for in vitro actin bundles. We will add more quantitative comparisons between the in vitro study and our results.

In addition, at the end of the conclusion section, we suggested future experiments that can be used for verifying our results. In particular, experiments with synthetic myosin filaments with tunable geometry seem to be suitable for verifying our computational predictions and observations.

The model assumptions and scientific context need to be described better.

We apologize for the insufficient descriptions about the model. We will revise those parts to better explain model assumptions and scientific context.

The network contractility seems to be a mere appendix to the bundle contractility which is presented in much more detail.

We included some cases run with the two-dimensional network in this study to prove the generality of our conclusions. We included minimal preliminary results in this study because we are currently working on a follow-up study with network structures. I hope that the reviewer would understand our intention and situation.

Reviewer #2 (Public review):

Summary:

In this study, the authors use a mechanical model to investigate how the geometry and deformations of myosin II filaments influence their force generation. They introduce a force generation efficiency that is defined as the ratio of the total generated force and the maximal force that the motors can generate. By changing the architecture of the myosin II filaments, they study the force generation efficiency in different systems: two filaments, a disorganized bundle, and a 2D network. In the simple two-filament systems, they found that in the presence of actin cross-linking proteins motors cannot add up their force because of steric hindrances. In the disorganized bundle, the authors identified a critical overlap of motors for cooperative force generation. This overlap is also influenced by the arrangement of the motor on the filaments and influenced by the length of the bare zone between the motor heads.

Strengths:

The strength of the study is the identification of organizational principles in myosin II filaments that influence force generation. It provides a complementary mechanistic perspective on the operation of these motor filaments. The force generation efficiency and the cooperative overlap number are quantitative ways to characterize the force generation of molecular motors in clusters and between filaments. These quantities and their conceptual implications are most likely also applicable in other systems.

Thank you for the comments about the strength of our study.

Weaknesses:

The detailed model that the authors present relies on over 20 numerical parameters that are listed in the supplement. Because of this vast amount of parameters, it is not clear how general the findings are. On the other hand, it was not obvious how specific the model is to myosin II, meaning how well it can describe experimental findings or make measurable predictions. The model seems to be quantitative, but the interpretation and connection to real experiments are rather qualitative in my point of view.

As the reviewer mentioned, all agent-based computational models for simulating the actin cytoskeleton are inevitably involved with such a large number of parameters. Some of the parameter values are not known well, so we have tuned our parameter values carefully by comparing our results with experimental observations in our previous studies since 2009. 

We were aware of the importance of rigorous representation of unbinding and walking rates of myosin motors, so we implemented the parallel cluster model, which can predict those rates with consideration of the mechanochemical rates of myosin II, into our model. Thus, we are convincing that our motors represent myosin II.

In our manuscript, our results were compared with prior observations in Ref. 4 (Thoresen et al., Biophys J, 2013) several times. In particular, larger force generation with more myosin heads per thick filament was consistent between the experiment and our simulations.

Our study can make various predictions. First, our study explains why non-muscle myosin II in stress fibers shows focal distributions rather than uniform distributions; if they stay closely, they can generate much larger forces in the stress fibers via the cooperative overlap. Our study also predicts a difference between bipolar structures (found in skeletal muscle myosins and non-muscle myosins) and side polar structures (found in smooth muscle myosins) in terms of the likelihood of the cooperative overlap. As shown in Author response image 1, myosin filaments with the bipolar structure can add up their forces better than those with the side polar structure when their overlap level is the same. We will add discussion about these in the revised manuscript.

We will add more discussion about these in the revised manuscript.

It was often difficult for me to follow what parameters were changed and what parameters were set to what numerical values when inspecting the curve shown in the figures. The manuscript could be more specific by explicitly giving numbers. For example, in the caption for Figure 6, instead of saying "is varied by changing the number of motor arms, the bare zone length, the spacing between motor arms", the authors could be more specific and give the ranges: ""is varied by changing the number of motor arms form ... to .., the bare zone length from .. to..., and the spacing between motor arms from .. to ..".

This unspecificity is also reflected in the text: "We ran simulations with a variation in either L_sp or L_bz" What is the range of this variation? "When L_M was similar" similar to what? "despite different N_M." What are the different values for N_M? These are only a few examples that show that the text could be way more specific and quantitative instead of qualitative descriptions.

We appreciate the comment. We will specify the range of the variation in each parameter in the revised manuscript.

In the text, after equation (2) the authors discuss assumptions about the binding of the motor to the actin filament. I think these model-related assumptions and explanations should be discussed not in the results section but rather in the "model overview" section.

Thank you for pointing this out. We will reorganize the text in the revised manuscript.

The lines with different colors in Figure 2A are not explained. What systems and parameters do they represent?

The different colors used in Fig. 2A were used for distinguishing 20 cases. We will add explanation about the colors in the figure caption in the revised manuscript.

Author response:

We thank the reviewers for their support of this work and insightful recommendations for how to improve it. We have provided specific responses to each reviewer comment below. To summarize how we intend to address the requested revisions:

Many of the reviewers’ comments requested additional technical or quality details about the DMS libraries or assays (e.g., number of cells tested, number of sequencing reads, assay replication, assay sensitivity, library balance), and we provide additional information and analyses that we can incorporate into the relevant portions of the text, supplementary tables, and supplementary figures to address these questions.

Some comments asked to clarify nomenclature/wording or provide additional labels to images, and we will make these changes as requested.

A few questions would require additional experimental data to address. Where experiments have already been performed, we will incorporate those results or cite relevant work previously reported in the literature.

Reviewer 1:

Summary

Howard et al. performed deep mutational scanning on the MC4R gene, using a reporter assay to investigate two distinct downstream pathways across multiple experimental conditions. They validated their findings with ClinVar data and previous studies. Additionally, they provided insights into the application of DMS results for personalized drug therapy and differential ligand responses across variant types.

Strengths

They captured over 99% of variants with robust signals and investigated subtle functionalities, such as pathway-specific activities and interactions with different ligands, by refining both the experimental design and analytical methods.

Weaknesses

While the study generated informative results, it lacks a detailed explanation regarding the input library, replicate correlation, and sequencing depth for a given number of cells.

Additionally, there are several questions that it would be helpful for authors to clarify.

(1) It would be helpful to clarify the information regarding the quality of the input library and experimental replicates. Are variants evenly represented in the library? Additionally, have the authors considered using long-read sequencing to confirm the presence of a single intended variant per construct? Finally, could the authors provide details on the correlation between experimental replicates under each condition?

Are variants evenly represented in the library?

We strive to achieve as evenly balanced library as possible at every stage of the DMS process (e.g., initial cloning in E. coli through integration into human cells). Below is a representative plot showing the number of barcodes per amino acid variant at each position in a given ~60 amino acid subregion of MC4R, which highlights how evenly variants are represented at the E. coli cloning stage.

Author response image 1.

We also make similar measurements after the library is integrated into HEK293T cell lines, and see similarly even coverage across all variants, as shown in the plot below.

Author response image 2.

Additionally, have the authors considered using long-read sequencing to confirm the presence of a single intended variant per construct?

We agree long-read sequencing would be an excellent way to confirm that our constructs contain a single intended variant. However, we elected for an alternate method (outlined in more detail in Jones et al. 2020) that leverages multiple layers of validation. First, the oligo chip-synthesized portions of the protein containing the variants are cloned into a sequence-verified plasmid backbone, which greatly decreases the chances of spuriously generating a mutation in a different portion of the protein. We then sequence both the oligo portion and random barcode using overlapping paired end reads during barcode mapping to avoid sequencing errors and to help detect DNA synthesis errors. At this stage, we computationally reject any constructs that have more than one variant. Given this, the vast majority of remaining unintended variants would come from somatic mutations introduced by the E. coli cloning or replication process, which should be low frequency. We have used our in-house full plasmid sequencing method, OCTOPUS, to sample and spot check this for several other DMS libraries we have generated using the same cloning methods. We have found variants in the plasmid backbone in only ~1% of plasmids in these libraries. Our statistical model also helps correct for this by accounting for barcode-specific variation. Finally we believe this provides further motivation for having multiple barcodes per variant, which dilutes the effect of any unintended additional variants.

Finally, could the authors provide details on the correlation between experimental replicates under each condition?

Certainly! In general, the Gs reporter had higher correlation between replicates than the Gq system (r ~ 0.5 vs r ~ 0.4). The plots below show two representative correlations at the RNA-seq stage of read counts for barcodes between the low a-MSH conditions. One important advantage of our statistical model is that it’s able to leverage information from barcodes regardless of the number of replicates they appear in.

Author response image 3.

Since the functional readout of variants is conducted through RNA sequencing, it seems crucial to sequence a sufficient number of cells with adequate sequencing saturation. Could the authors clarify the coverage depth used for each RNA-seq experiment and how this depth was determined? Additionally, how many cells were sequenced in each experiment?

This will be addressed by incorporating the following details into the manuscript:

We seeded 17 million cells per replicate at the start of each assay and, with a doubling of ~1.5x over the course of the assay, harvested ~25.5 million cells per replicate for RNA extraction and sequencing. We found this sufficient to get at least ~30-60x cellular coverage per amino acid variant.

Total mapped reads per replicate at RNA-seq stage

- Gs/CRE: 9.1-18.2 million mapped reads, median=12.3

- Gq/UAS: 8.6-24.1 million mapped reads, median=14.5

- Gs/CRE+Chaperone: 6.4-9.5 million mapped reads, median=7.5

Reads per barcode distribution

- Median read counts of 8, 10, and 6 reads per sample per barcode for Gs/CRE, Gq/UAS, and Gs/CRE+Chaperone assays, respectively.

Barcodes per variant distribution

- As reported, the median number of barcodes per variant across samples (the “median of medians”) is 56 for Gs/CRE and 28 for Gq/UAS

- Additionally, it is 44 for Gs/CRE+Chaperone

It appears that the frequencies of individual RNA-seq barcode variants were used as a proxy for MR4C activity. Would it be important to also normalize for heterogeneity in RNA-seq coverage across different cells in the experiment? Variability in cell representation (i.e., the distribution of variants across cells) could lead to misinterpretation of variant effects. For example, suppose barcode_a1 represents variant A and barcode_b1 represents variant B. If the RNA-seq results show 6 reads for barcode_a1 and 7 reads for barcode_b1, it might initially appear that both variants have similar effect sizes. However, if these reads correspond to 6 separate cells each containing 1 copy of barcode_a1, and only 1 cell containing 7 copies of barcode_b1, the interpretation changes significantly. Additionally, if certain variants occupy a larger proportion of the cell population, they are more likely to be overrepresented in RNA sequencing.

We account for this heterogeneity in several ways. First, as shown above (Response to Reviewer 1, Question 1), we aim to have even representation of variants within our libraries. Second, we utilize compositional control conditions like forskolin or unstimulated conditions to obtain treatment-independent measurements of barcode abundance and, consequently, of mutant-vs-WT effects that are due to compositional rather than biological variability. We expect that variability observed under these controls is due to subtle effects of molecular cloning, gene expression, and stochasticity. Using these controls, we observe that mutant-vs-WT effects are generally close to zero in these normalization conditions (e.g., in untreated Gq, see Supplementary Figure 3) as compared to drug-treated conditions. For example, pre-mature stops behave similar to WT in normalization conditions. This indicates that mutant abundance is relatively homogenous. Where there are barcode-dependent effects on abundance, we can use information from these conditions to normalize that effect. Finally, our mixed-effect model accounts for barcode-specific deviations from the expected mutant effect (e.g. a “high count” barcode consistently being high relative to the mean).

Although the assay system appears to effectively represent MC4R functionality at the molecular level, we are curious about the potential disparity between the DMS score system and physiological relevance. How do variants reported in gnomAD distribute within the DMS scoring system?

Figure 2D shows DMS scores (variant effect on Gs signaling) relative to human population frequency for all MC4R variants reported in gnomAD as of January 8, 2024.

To measure Gq signaling, the authors used the GAL4-VPR relay system. Is there additional experimental data to support that this relay system accurately represents Gq signaling?

The full Gq reporter uses an NFAT response element from the IL-2 promoter to regulate the expression of the GAL4-VPR relay. In this system, the activation of Gq signaling results in the activation of the NFAT response element, and this signal is then amplified by the GAL4-VPR relay. The NFAT response element has been previously well-validated to respond to the activation of Gq signaling (e.g., PMID: 8631834). We will add this reference to the text to further support the use of the Gq assay.

Identifying the variants responsive to the corrector was impressive. However, we are curious about how the authors confirmed that the restoration of MC4R activity was due to the correction of the MC4R protein itself. Is there a possibility that the observed effect could be influenced by other factors affected by the corrector? When the corrector was applied to the cells, were any expected or unexpected differential gene expression changes observed?

While we do not directly measure whether Ipsen-17 has effects on other signaling processes, previous work has shown that Ipsen-17 treatment does not indirectly alter signaling kinetics such as receptor internalization (Wang et al., 2014). Furthermore, our analysis methods inherently account for this by normalizing variant effects to WT signaling levels. Any observed rescue of a given variant inherently means that the variant is specifically more responsive to Ipsen-17 than WT, and the fact that different variants exhibit different levels of rescue is reassuring that the mechanism is on target to MC4R. Lastly, Ipsen-17 is known to be an antagonist of alpha-MSH activity and is thought to bind directly to the same site on MC4R (Wang et al., 2014).

As mentioned in the introduction, gain-of-function (GoF) variants are known to be protective against obesity. It would be interesting to see further studies on the observed GoF variants. Do the authors have any plans for additional research on these variants?

We agree this would be an excellent line of inquiry, but due to changes in company priorities we unfortunately do not have any plans for additional research on these variants.

Reviewer 2:

Overview

In this manuscript, the authors use deep mutational scanning to assess the effect of ~6,600 protein-coding variants in MC4R, a G protein-coupled receptor associated with obesity. Reasoning that current deep mutational scanning approaches are insufficiently precise for some drug development applications, they focus on articulating new, more precise approaches. These approaches, which include a new statistical model and innovative reporter assay, enable them to probe molecular phenotypes directly relevant to the development of drugs that target this receptor with high precision and statistical rigor.

They use the resulting data for a variety of purposes, including probing the relationship between MC4R's sequence and structure, analyzing the effect of clinically important variants, identifying variants that disrupt downstream MC4R signaling via one but not both pathways, identifying loss of function variants are amenable to a corrector drug and exploring how deep mutational scanning data could guide small molecule drug optimization.

Strengths

The analysis and statistical framework developed by the authors represent a significant advance. In particular, the study makes use of barcode-level internally replicated measurements to more accurately estimate measurement noise.

The framework allows variant effects to be compared across experimental conditions, a task that is currently hard to do with rigor. Thus, this framework will be applicable to a large number of existing and future deep mutational scanning experiments.

The authors refine their existing barcode transcription-based assay for GPCR signaling, and develop a clever "relay" new reporter system to boost signaling in a particular pathway. They show that these reporters can be used to measure both gain of function and loss of function effects, which many deep mutational scanning approaches cannot do.

The use of systematic approaches to integrate and then interrogate high-dimensional deep mutational scanning data is a big strength. For example, the authors applied PCA to the variant effect results from reporters for two different MC4R signaling pathways and were able to discover variants that biased signaling through one or the other pathway. This approach paves the way for analyses of higher dimensional deep mutational scans.

The authors use the deep mutational scanning data they collect to map how different variants impact small molecule agonists activate MC4R signaling. This is an exciting idea, because developing small-molecule protein-targeting therapeutics is difficult, and this manuscript suggests a new way to map small-molecule-protein interactions.

Weaknesses

The authors derive insights into the relationship between MC4R signaling through different pathways and its structure. While these make sense based on what is already known, the manuscript would be stronger if some of these insights were validated using methods other than deep mutational scanning.

Likewise, the authors use their data to identify positions where variants disrupt MC4R activation by one small molecule agonist but not another. They hypothesize these effects point to positions that are more or less important for the binding of different small molecule agonists. The manuscript would be stronger if some of these insights were explored further.

Impact

In this manuscript, the authors present new methods, including a statistical framework for analyzing deep mutational scanning data that will have a broad impact. They also generate MC4R variant effect data that is of interest to the GPCR community.

Author response:

Reviewer 1:

There are no significant weaknesses to signal in the manuscript. However, in order to fully conclude that there is no obvious advantage for the linguistic dimension in neonates, it would have been most useful to test a third condition in which the two dimensions were pitted against each other, that is, in which they provide conflicting information as to the boundaries of the words comprised in the artificial language. This last condition would have allowed us to determine whether statistical learning weighs linguistic and non-linguistic features equally, or whether phonetic content is preferentially processed.

We appreciate the reviewers' suggestion that a stream with conflicting information would provide valuable insights. In the present study, we started with a simpler case involving two orthogonal features (i.e., phonemes and voices), with one feature being informative and the other uninformative, and we found similar learning capacities for both. Future work should explore whether infants—and humans more broadly—can simultaneously track regularities in multiple speech features. However, creating a stream with two conflicting statistical structures is challenging. To use neural entrainment, the two features must lead to segmentation at different chunk sizes so that their effects lead to changes in power/PLV at different frequencies—for instance, using duplets for the voice dimension and triplets for the linguistic dimension  (or vice versa). Consequently, the two dimensions would not be directly comparable within the same participant in terms of the number of distinguishable syllables/voices, memory demand, or SNR given the 1/F decrease in amplitude of background EEG activity. This would involve comparisons between two distinct groups counter-balancing chunk size and linguistic non-linguistic dimension. Considering the test phase, words for one dimension would have been part-words for the other dimension. As we are measuring differences and not preferences, interpreting the results would also have been difficult. Additionally, it may be difficult to find a sufficient number of clearly discriminable voices for such a design (triplets imply 12 voices). Therefore, an entirely different experimental paradigm would need to be developed.

If such a design were tested, one possibility is that the regularities for the two dimensions are calculated in parallel, in line with the idea that the calculation of statistical regularities is a ubiquitous implicit mechanism (see Benjamin et al., 2024, for a proposed neural mechanism). Yet, similar to our present study, possibly only phonetic features would be used as word candidates. Another possibility is that only one informative feature would be explicitly processed at a time due to the serial nature of perceptual awareness, which may prioritise one feature over the other.

Note: The reviewer’s summary contains a typo: syllabic rate (4 Hz) –not 2 Hz, and word rate (2 Hz) –not 4 Hz.

Reviewer 2:

N400: I am skeptical regarding the interpretation of the phoneme-specific ERP effect as a precursor of the N400 and would suggest toning it down. While the authors are correct in that infant ERP components are typically slower and more posterior compared to adult components, and the observed pattern is hence consistent with an adult N400, at the same time, it could also be a lot of other things. On a functional level, I can't follow the author's argument as to why a violation in phoneme regularity should elicit an N400, since there is no evidence for any semantic processing involved. In sum, I think there is just not enough evidence from the present paradigm to confidently call it an N400.

The reviewer is correct that we cannot definitively determine the type of processing reflected by the ERP component that appears when neonates hear a triplet after exposure to a stream with phonetic regularities. We interpreted this component as a precursor to the N400, based on prior findings in speech segmentation tasks without semantic content, where a ~400 ms component emerged when adult participants recognised pseudowords (Sander et al., 2002) or during structured streams of syllables (Cunillera et al., 2006, 2009). Additionally, the component we observed had a similar topography and timing to those labelled as N400 in infant studies, where semantic processing was involved (Parise et al., 2010; Friedrich & Friederici, 2011).

Given our experimental design, the difference we observed must be related to the type of regularity during familiarisation (either phonemes or voices). Thus, we interpreted this component as reflecting lexical search— a process which could be triggered by a linguistic structure but which would not be relevant to a non-linguistic regularity such as voices. However, we are open to alternative interpretations. In any case, this difference between the two streams reveals that computing regularities based on phonemes versus voices does not lead to the same processes. We will revise and tone down the corresponding part of the discussion to clarify that it is just a possible interpretation of the results.  

Female and male voices: Why did the authors choose to include male and female voices? While using both female and male stimuli of course leads to a higher generalizability, it also introduces a second dimension for one feature that is not present for this other (i.e., phoneme for Experiment 1 and voice identity plus gender for Experiment 2). Hence, couldn't it also be that the infants extracted the regularity with which one gender voice followed the other? For instance, in List B, in the words, one gender is always followed by the other (M-F or F-M), while in 2/3 of the part-words, the gender is repeated (F-F and M-M). Wouldn't you expect the same pattern of results if infants learned regularities based on gender rather than identity?

We used three female and three male voices to maximise acoustic variability. The streams were synthesised using MBROLA, which provides a limited set of artificial voices. Indeed, there were not enough French voices of acceptable quality, so we also used two Italian voices (the phonemes used existed in both Italian and French).

Voices differ in timbre, and female voices tend to be higher pitched. However, it is sometimes difficult to categorise low-pitched female voices and high-pitched male voices. Given that gender may be an important factor in infants' speech perception (newborns, for instance, prefer female voices at birth), we conducted tests to assess whether this dimension could have influenced our results.  

We first quantified the transitional probabilities matrices during the structured stream of Experiment 2, considering that there are only two types of voices: Female and Male.  

For List A, all transition probabilities are equal to 0.5 (P(M|F), P(F|M), P(M|M), P(F|F)), resulting in flat TPs throughout the stream (see Author response image 1, top). Therefore, we would not expect neural entrainment at the word rate (2 Hz), nor would we anticipate ERP differences between the presented duplets in the test phase.

For List B, P(M|F)=P(F|M)=0.66 while P(M|M)=P(F|F)=0.33. However, this does not produce a regular pattern of TP drops throughout the stream (see Author response image 1, bottom). As a result, strong neural entrainment at 2 Hz was unlikely, although some degree of entrainment might have occasionally occurred due to some drops occurring at a 2 Hz frequency. Regarding the test phase, all three Words and only one Part-word presented alternating patterns (TP=0.6). Therefore, the difference in the ERPs between Words and Partwords in List B might be attributed to gender alternation.  

However, it seems unlikely that gender alternation alone explains the entire pattern of results, as the effect is inconsistent and appears in only one of the lists. To rule out this possibility, we analysed the effects in each list separately.

Author response image 1.

Transition probabilities (TPs) across the structured stream in Experiment 2, considering voices processed by gender (Female or Male). Top: List A. Bottom: List B.

We computed the mean activation within the time windows and electrodes of interest and compared the effects of word type and list using a two-way ANOVA. For the difference between Words and Part-words over the positive cluster, we observed a main effect of word type (F(1,31) = 5.902, p = 0.021), with no effects of list or interactions (p > 0.1). Over the negative cluster, we again observed a main effect of word type (F(1,31) = 10.916, p = 0.0016), with no effects of list or interactions (p > 0.1). See Author response image 2.  

Author response image 2.

Difference in ERP voltage (Words – Part-words) for the two lists (A and B); W=Words; P=Part-Words, 

We conducted a similar analysis for neural entrainment during the structured stream on voices. A comparison of entrainment at 2 Hz between participants who completed List A and List B showed no significant differences (t(30) = -0.27, p = 0.79). A test against zero for each list indicated significant entrainment in both cases (List A: t(17) = 4.44, p = 0.00036; List B: t(13) = 3.16, p = 0.0075). See Author response image 3.

Author response image 3.

Neural entrainment at 2Hz during the structured stream of Experiment 2 for Lists A and B.

Words entrainment over occipital electrodes: Do you have any idea why the duplet entrainment effect occurs over the electrodes it does, in particular over the occipital electrodes (which seems a bit unintuitive given that this is a purely auditory experiment with sleeping neonates).

Neural entrainment might be considered as a succession of evoked response induced by the stream. After applying an average reference in high-density EEG recordings, the auditory ERP in neonates typically consists of a central positivity and a posterior negativity with a source located at the electrical zero in a single-dipole model (i.e. approximately in the superior temporal region (Dehaene-Lambertz & Dehaene, 1994). In adults, because of the average reference (i.e. the sum of voltages is equal to zero at each time point) and because the electrodes cannot capture the negative pole of the auditory response, the negativity is distributed around the head. In infants, however, the brain is higher within the skull, allowing for a more accurate recording of the negative pole of the auditory ERP (see Author response image 4 for the location of electrodes in an infant head model).  

Besides the posterior electrodes, we can see some entrainment on more anterior electrodes that probably corresponds to the positive pole of the auditory ERP.

Author response image 4.

International 10–20 sensors' location on the skull of an infant template, with the underlying 3-D reconstruction of the grey-white matter interface and projection of each electrode to the cortex. Computed across 16 infants (from Kabdebon et al, Neuroimage, 2014). The O1, O2, T5, and T6 electrodes project lower than in adults.

Reviewer 3:

(1) While it's true that voice is not essential for language (i.e., sign languages are implemented over gestures; the use of voices to produce non-linguistic sounds, like laughter), it is a feature of spoken languages. Thus I'm not sure if we can really consider this study as a comparison between linguistic and non-linguistic dimensions. In turn, I'm not sure that these results show that statistical learning at birth operates on non-linguistic features, being voices a linguistic dimension at least in spoken languages. I'd like to hear the authors' opinions on this.

On one hand, it has been shown that statistical learning (SL) operates across multiple modalities and domains in human adults and animals. On the other hand, SL is considered essential for infants to begin parsing speech. Therefore, we aimed to investigate whether SL capacities at birth are more effective on linguistic dimensions of speech, potentially as a way to promote language learning.

We agree with the reviewer that voices play an important role in communication (e.g., for identifying who is speaking); however, they do not contribute to language structure or meaning, and listeners are expected to normalize across voices to accurately perceive phonemes and words. Thus, voices are speech features but not linguistic features. Additionally, in natural speech, there are no abrupt voice changes within a word as in our experiment; instead, voice changes typically occur on a longer timescale and involve only a limited number of voices, such as in a dialogue. Therefore, computing regularities based on voice changes would not be useful in real-life language learning. We considered that contrasting syllables and voices was an elegant way to test SL beyond its linguistic dimension, as the experimental paradigm is identical in both experiments.  

Along the same line, in the Discussion section, the present results are interpreted within a theoretical framework showing statistical learning in auditory non-linguistic (string of tones, music) and visual domains as well as visual and other animal species. I'm not sure if that theoretical framework is the right fit for the present results.

(2) I'm not sure whether the fact that we see parallel and independent tracking of statistics in the two dimensions of speech at birth indicates that newborns would be able to do so in all the other dimensions of the speech. If so, what other dimensions are the authors referring to?

The reviewer is correct that demonstrating the universality of SL requires testing additional modalities and acoustic dimensions. However, we postulate that SL is grounded in a basic mechanism of long-term associative learning, as proposed in Benjamin et al. (2024), which relies on a slow decay in the representation of a given event. This simple mechanism, capable of operating on any representational output, accounts for many types of sequence learning reported in the literature (Benjamin et al., in preparation). We will revise the discussion section to clarify this theoretical framework.

(3) Lines 341-345: Statistical learning is an evolutionary ancient learning mechanism but I do not think that the present results are showing it. This is a study on human neonates and adults, there are no other animal species involved therefore I do not see a connection with the evolutionary history of statistical learning. It would be much more interesting to make claims on the ontogeny (rather than philogeny) of statistical learning, and what regularities newborns are able to detect right after birth. I believe that this is one of the strengths of this work.

We did not intend to make claims about the phylogeny of SL. Since SL appears to be a learning mechanism shared across species, we use it as a framework to suggest that SL may arise from general operational principles applicable to diverse neural networks. Thus, while it is highly useful for language acquisition, it is not specific to it. We will revise this section to tone down our claims.  

(4) The description of the stimuli in Lines 110-113 is a bit confusing. In Experiment 1, e.g., "pe" and "tu" are both uttered by the same voice, correct? ("random voice each time" is confusing). Whereas in Experiment 2, e.g., "pe" and "tu" are uttered by different voices, for example, "pe" by yellow voice and "tu" by red voice. If this is correct, then I recommend the authors to rephrase this section to make it more clear.

To clarify, in Experiment 1, the voices were randomly assigned to each syllable, with the constraint that no voice was repeated consecutively. This means that syllables within the same word were spoken by different voices, and each syllable was heard with various voices throughout the stream. As a result, neonates had to retrieve the words based solely on syllabic patterns, without relying on consistent voice associations or specific voice relationships.

In Experiment 2, the design was orthogonal: while the syllables were presented in a random order, the voices followed a structured pattern. Similar to Experiment 1, each syllable (e.g., “pe” and “tu”) was spoken by different voices. The key difference is that in Experiment 2, the structured regularities were applied to the voices rather than the syllables. In other words, the “green” voice was always followed by the “red” voice for example but uttered different syllables.

We will revise the methods section to clarify these important points.

(5) Line 114: the sentence "they should compute a 36 x 36 TPs matrix relating each acoustic signal, with TPs alternating between 1/6 within words and 1/12 between words" is confusing as it seems like there are different acoustic signals. Can the authors clarify this point?

Thank you for highlighting this point. To clarify, our suggestion is that neonates might not track regularities between phonemes and voices as separate features. Instead, they may treat each syllable-voice combination as a distinct item—for example, "pe" spoken by the "yellow" voice is one item, while "pe" spoken by the "red" voice is another. Under this scenario, there would be a total of 36 unique items (6 syllables × 6 voices), and infants would need to track regularities between these 36 combinations.

We will rephrase this sentence in the manuscript to make it clearer.

Author response:

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

Reviewer #1 (Public review):

(1) I was surprised at the effect of BicD2 knockdown on LAMP (and VPS41) localization, which really suggests that in HeLa and Cos7 cells, BicD2 regulation of Kinesin-1 (rather than dynein) is the primary driver of lysosome localization. The KIF5B-knockout rescue of the BicD2overexpression phenotype was a very powerful result that supports this conclusion. Have the authors looked at other cargos, eg, Golgi or centrosomes in G2? Can the authors include more discussion about what this result means or how they imagine dynein and kinesin-1's interaction with BicD2 is regulated? 

We have performed this experiment as requested by the reviewer. The BICD2 siRNA also resulted in Golgi fragmentation and localization defects of the centrosome in cells that are in G2 phase of the cell cycle (Supplemental Fig. 2E-H).

We have also added additional discussion related to how BICD2 might couple cargos to opposite polarity motors (lines 440-447). Interestingly, the lysosome motility defect we observe upon BICD2 knock down has similarity to the RAB6A trafficking phenotype. In both cases, what one sees is a sharp reduction in the number of motile particles rather than a reversal in the direction of motility. This suggests that both motors are involved in the steady state distribution of these cargoes.

(2) Have the authors examined if the SMALED mutants show diminished or increased binding to KIF5B? While the authors are correct that the mutations could hyperactivate dynein because they reduce BicD2 autoinhibition, it is possible that the SMALED mutants hyperactivate dynein because they no longer bind kinesin. This would be particularly interesting, given the complex relationship between BicD2 regulation of dynein and kinesin that the authors show in Figure 3. 

Thank you for this suggestion. We had not considered this. We have added this experiment in the revised manuscript (Supplemental Fig. 3H, I). We find that the interaction between wild-type BICD2 and KIF5B is only slightly above the control. This is consistent with published findings that indicate that although the isolated CC2 domain of BICD2 is able to interact with KIF5B, the binding is lower for the full-length protein. This is most likely due to the intramolecular interaction between the N and C-termini of BICD2 partially blocking the binding site. Interestingly, however, all three mutants display a reduced interaction with KIF5B, with the reduction being most severe for the cargo domain binding mutants. Thus, as we discuss in the revised manuscript, dynein hyperactivity likely results from increased binding to dynein and a concurrent reduction in binding to KIF5B.

(3) What is already known about the protein GRAMD1A? Did the authors choose to focus on GRAMD1A because it was the only novel interaction found in the SMALED mutant interactomes, or was this protein interesting for a different reason? Does the known function of GRAMD1A explain the potential dysfunction of cells expressing BICD2_R747C or patients who have this mutation? More discussion of this protein and why the authors focused on it would really strengthen the manuscript. 

We chose to focus on GRAMD1A for a few reasons. The protein that displayed the highest gain of function interaction with BICD2_R747C in our proteomic analysis was Plastin. However, using at least one antibody against Plastin, we were not able to validate this result. In addition, we had previously performed a proteomic screen using a BICD2_R747A (arginine to alanine) mutation and had compared the interactome of this mutant to the wild-type protein. Plastin was not recovered in that screen but the top hit was GRAMD1A. Given that we isolated GRAMD1A in two separate screens as a gain of function interaction, we believed the result was worth focusing on for followup studies. 

GRAMD1A (as well as its paralogs GRAMD1B and C) function in non-vesicle transport of accessible cholesterol from the plasma membrane to the ER. We have added additional discussion on GRAMD1A (lines 484-495). While we observe a relocalization of GRAMD1A in mutant expressing cells, we do not know whether this is sufficient to result in cholesterol transport defects. There are several routes for cholesterol uptake, with the GRAMD1A pathway representing just one these routes. 

Reviewer #2 (Public review):

(1) The authors use cells that have been engineered to express the different BICD2 constructs. As shown in Figure 4B, the authors see wide expression of BICD2_WT throughout the cell. However, WT BICD2 usually localizes to the TGN. This widespread localization introduces some uncertainty about the interactome data. The authors should either try to verify the interaction data (specifically with the HOPS complex and GRAMD1A) by immunoprecipitating endogenous BICD2 or by repeating their interactome experiment in Figure 1 using BICD2 knockout cells that express the BICD2_WT construct. This should also be done to verify the immunoprecipitation and microscopy data shown in Figure 7. 

The localization of our exogenous BICD2-mNeon constructs is similar to what others have seen using GFP tagged versions of the protein (for example Peeters et al., 2013). In addition, in the experiment shown in the initial version of the paper, we were focusing on the centrosomal localization of BICD2. However, our BICD2-mNeon construct is also observed at the Golgi, in addition to its localization throughout the cell (Supplemental Fig. 3C). 

We attempted to perform a co-immunoprecipitation experiment using endogenous proteins as suggested by the reviewer. Although a rabbit polyclonal antibody was able to coimmunoprecipitate RANBP2 with BICD2, the antibody complex of heavy and light chains comigrated with the VPS41 band and was abundantly detected by the secondary antibody used in the western blot. Thus, we were not able to make a conclusion regarding whether or not VPS41 was present in the co-immunoprecipitate. We attempted the experiment using a mouse monoclonal antibody against BICD2. However, this antibody failed in the immunoprecipitation experiment and we could not detect either RANBP2 (a validated cargo) or VPS41. Although the VPS41 antibody we used in the paper works for western blot, it does not recognize the native protein. Thus, despite our best efforts, we are not able to draw a valid conclusion from these coip experiments.

It is beyond the scope of the revision to perform the entire experiment in a BICD2 KO cell line.  A BICD2 KO cell line does not exist and it would take several months to make such a knock out in the FLP IN HEK cells that were used in this manuscript. However, we have validated the interaction between BICD2 and VPS41 in cells that have been depleted of endogenous BICD2 (Supplemental Fig. 1B). The transgenic constructs contain silent mutations that make them refractory to bicD2 siRNA1. Thus, although endogenous BICD2 is depleted by the siRNA treatment, wild-type and mutant BICD2_TurboID is not. A similar approach was also used to demonstrate the gain of function interaction between BICD2_R747C and GRAMD1A in cells depleted of endogenous BICD2 (Supplemental Fig. 5A).

(2) The authors conclude that cargo transport defects resulting from BICD2 mutations may contribute to SMALED2 symptoms. However, the authors are unable to determine if BICD2 directly binds to the potential new cargo, the HOPS complex. To address this, the authors could purify full-length WT BICD2 and perform in vitro experiments. Furthermore, the authors were unable to identify the minimal region of BICD2 needed for HOPS interaction. The authors could expand on the experiment attempted with the extended BICD2 C-terminal using a deltaCC1 construct, which could also be used for in vitro experiments. 

We have not been successful in purifying full length BICD2 in bacteria, perhaps due to solubility issues. However, we have added several experiments to further examine the nature of the BICD2-HOPS complex interaction.

We have performed the experiment as requested. We find that BICD2_delCC1 is able to bind VPS41, but not as efficiently as the full length protein. However, unlike the CC3 cargo binding construct, the BICD2_delCC1 construct also displays reduced binding to RANBP2 (Supplemental Fig. 1D). We attribute this defect to either the intramolecular BICD2 interaction blocking cargo binding or potentially to a folding defect in the BICD2_delCC1 construct. Thus, although we performed this experiment as suggested by the reviewer, we are not able to make a solid conclusion.

Based on the fact that VPS41 was the most abundantly detected HOPS component in the BICD2 interactome, we hypothesized that it was the point of direct contact between BICD2 and the HOPS complex. However, contrary to our hypothesis, depletion of VPS41 did not compromise the association between BICD2 and VPS16 and VPS18 (Supplemental Fig. 1E). Thus, we conclude that there are multiple points of contact between BICD2 and the HOPS complex, with BICD2 perhaps recognizing a common motif or domain present in these proteins.

We next attempted to map the interaction site using Alphafold2 multimer. Although we were able to use this platform to predict a high confidence interaction between BICD2 and RAB6A (consistent with published results), this did not yield a high confidence prediction for the BICD2HOPS complex interaction.

Ultimately although we added several new experiments, we were not able to determine the minimal region for binding, nor whether the interaction is direct or indirect. These caveats are clearly stated in the revised manuscript. Regardless of whether the interaction is direct or indirect however, it is noteworthy that the association between BICD2 and the HOPS complex is reduced by the R747C SMALED2 mutation.

(3) Again, the authors conclude that BICD2 mutants cause cargo transport defects that are likely to lead to SMALED2 symptoms. This would be better supported if the authors are able to find a protein relevant to SMALED2 and examine if/how its localization is changed under expression of the BICD2 mutants. The authors currently use the HOPS complex and GRAMD1A as indicators of cargo transport defects, but it is unclear if these are relevant to SMALED2 symptoms. 

This point was addressed in the general discussion. Given the complexity of SMALED2 (autosomal dominant disorder; variable phenotypic severity; adult onset disorder in many instances, etc.) it is very hard to model in a cell line. One of the reasons we focused our studies on the HOPS complex and VPS41 in particular was because mutations in VPS41 are associated with spinocerebellar ataxia, a neurodevelopment disorder. However, we cannot conclude whether the reduction/loss of interaction of BICD2 with the HOPS complex is causative for disease symptoms. We also cannot conclude at present whether the mis-targeting of GRAMD1A is causative for disease symptoms. We have discussed these caveats in the revised manuscript and have included a section in the discussion that specifically lists the limitations of our study (lines 511-530).

With that said, we can conclude that mutations in the cargo binding domain of BICD2 result in dynein hyperactivity, altered BICD2 localization in hippocampal neurons, and reduced neurite growth. Given that we observe interactome changes in HEK cells, it is plausible that interactome changes also exist in motor neurons. However, even in the absence of interactome changes, hyperactivation of dynein alone can result in cargo trafficking defects; the same cargos can be excessively localized in the soma vs the axon. As noted previously, however, a thorough examination of these points will require the use of genetically engineered motor neurons and is beyond the scope of the current study.

Reviewer #3 (Public review):

Strengths: 

Extensive interactomes are presented for both WT BicD2 as well as the disease mutants, which will be valuable for the community. The HOPS complex was identified as a novel interactor of BicD2, which is important for fusion of late endosomes and lysosomes, which is of interest, since some of the BicD2 disease mutations result in Golgi-fragmentation phenotypes. The interaction with the HOPS complex is affected by the R747C mutation, which also results in a gain-of-function interaction with GRAMD1A. 

Weaknesses: 

The manuscript should be strengthened by further evidence of the BicD2/HOPS complex interaction and the functional implications for spinal muscular atrophy by changes in the interactome through mutations. Which functional implications does the loss of the BicD2/HOPS complex interaction and the gain of function interaction with GRAMD1A have in the context of the R747C mutant? 

(1) In the biotin proximity ligation assay, a large number of targets were identified, but it is not clear why only the HOPS complex was chosen for further verification. Immunoprecipitation was used for target verification, but due to the very high number of targets identified in the screen, and the fact that the HOPS complex is a membrane protein that could potentially be immunoprecipitated along with lysosomes or dynein, additional experiments to verify the interaction of BicD2 with the HOPS complex (reconstitution of a complex in vitro, GST-pull down of a complex from cell extracts or other approaches) are needed to strengthen the manuscript. 

As discussed for reviewer 2 (point 2), we have added several experiments to better characterize the BICD2-HOPS complex interaction.

We chose to focus on the HOPS complex for a few reasons. The list of interactions that displayed a >2 fold enrichment vs control was actually not that large (66 proteins). Within this list, we identified 4 out of 6 HOPS components and VPS41 was the 5th most enriched protein in the BICD2 interactome (RANBP2 by contrast was #16 on this list). Furthermore, the BICD2_R747C mutation resulted in greatly reduced interaction of BICD2 with the HOPS complex, whereas its interaction with dynein was increased. These results indicate that these proteins are not simply immunoprecipitating with the BICD2/dynein complex. Apart from the HOPS complex, lysosomal proteins were not present in the interactome, making it unlikely that they were identified due to non-specific interactions between BICD2 and co-precipitating lysosomes.

(2) In the biotin proximity ligation assay, a large number of Bi cD2 interactions were identified that are distinct between the mutant and the WT, but it was not clear why, particularly GRAMD1A was chosen as a gain-of-function interaction, and what the functional role of a BicD2/GRAMD1A interaction may be. A Western blot shows a strengthened interaction with the R747C mutant, but GRAMD1A also interacts with WT BicD2. 

Please see the above discussion on GRAMD1A (reviewer 1, point 3). GRAMD1A comes down non-specifically with the binding control as well as BICD2_wt. We therefore conclude that wildtype BICD2 does not specifically interact with GRAMD1A above background levels (Fig. 7, compare the control lane vs BICD2-wt).

(3) Furthermore, the functional implications of changed interactions with HOPS and GRAMD1A in the R747C mutant are unclear. Additional experiments are needed to establish the functional implication of the loss of the BicD2/HOPS interaction in the BicD2/R747C mutant. For the GRAMD1A gain of function interaction, according to the authors, a significant amount of the protein localized with BicD2/R747C at the centrosomal region. This changed localization is not very clear from the presented images (no centrosomal or other markers were used, and the changed localization could also be an effect of dynein hyperactivation in the mutant). Furthermore, the functional implication of a changed localization of GRAMD1A is unclear from the presented data. 

We have performed the experiment as requested by the reviewer. The re-localized GRAMD1A localizes adjacent to Pericentrin, a centrosomal marker (Supplemental Fig. 5B-F). GRAMD1A and BICD2 appear to co-localize in a ring around the Pericentrin marked centrosome.

The re-localization of GRAMD1A to the centrosomal area by BICD2_R747C appears to be unique to this mutant, and not simply an issue of dynein hyperactivity. The other two mutants tested, BICD2_N188T and BICD2_R694C also hyperactivate dynein. However, they do not result in the same type of dramatic re-localization of GRAMD1A as we observe with the BICD2_R747C mutant. We conclude that this altered localization results from a gain of function interaction with BICD2_R747C as well as dynein hyperactivity.

Reviewer #1 (Recommendations for the authors): 

Please add a discussion about how the authors calculated the Cell Body enrichment shown in 5E. Is this a ratio of the BicD2 intensity in the cell body:axon? Did the authors normalize for potential differences in BicD2 variant expression? 

Yes, it is a ratio of the intensity between the cell body and axon. This is described in the Methods section under quantification (lines 725-728). We attempted to image cells expressing similar amounts of protein.  

Reviewer #2 (Recommendations for the authors): 

(1) The paper would benefit from an explanation of why the authors chose to follow up on the HOPS complex out of all proteins identified in the interactome experiment. 

This discussion has been included in the revised manuscript.  

(2) In panel B of Supplementary Figure 1, RFP mTurbo has a significant amount of non-specific binding to VPS18. The authors note that in the initial interactome experiment, there was a twofold enrichment of this protein in BICD2 pulldown versus control. Do the authors have a co-IP that has a similar enrichment?

VPS18 occasionally comes down non-specifically with our RFP-TurboID control. However, the interaction is specific, because very little VPS18 comes down with the BICD2 construct lacking the cargo binding domain (Fig. 2B). An additional example of the VPS18 binding result is shown in Supplemental Fig. 1E.

(3) In Figure 2B, there seems to be less Vps18 in the input for BICD2 delCC3-mTrbo. Do the authors have a blot where there is equal input across all conditions? This may increase the slight signal seen in the pulldown.

The blot shown in Supplemental Fig. 1C has equivalent load for VPS18 across all lanes. Minimal binding of VPS18 is observed with the BICD2_delCC3 sample.

(4) In Figure 3, can the authors show representative images of GFP-VPS-41 and LAMP1 localization that are at the same magnification? It currently looks as if the localization pattern differs between the two under control siRNA. Alternatively, the authors should show colocalization of the two, as the authors note both are localized to late endosomes/lysosomes. 

We have provided additional images that are at the same magnification (Supplemental Fig. 2IK). Co-localization between GFP-VPS41 (rabbit polyclonal antibody against GFP) and LAMP1 (rabbit polyclonal antibody) is not possible. However, published studies have shown that a subset of V5 tagged VPS41 vesicles are positive for LAMP1. We have cited this study.

(5) In Supplementary Figure 2, the authors should show the knockdown efficiency of both BICD2 siRNAs. The VPS41 staining in panel B looks like there is less perinuclear localization than with BICD2 siRNA 1. Is the because of knockdown efficiency? 

We have included this data (Supplemental Fig. 2B). Both siRNAs are capable of depleting BICD2. However, we do see slightly more effective knock down with siRNA-1.

(6) The data in Figure 4A would be more striking with quantification. 

Quantifications have been provided (Supplemental Fig. 3A,B). Using a one-way Anova analysis, BICD2_R747C is the only mutant that shows significance. Variability in the binding experiment resulted in the other two mutants not showing a statistically significant change. However, the additional assays that are provided (centrosomal enrichment of BICD2 and peroxisome tethering) clearly demonstrate that the R694C mutant also results in dynein hyperactivation. It should be noted that the analysis done by Huynh et al., 2017 also showed a binding increase between BICD2 disease mutants and dynein. However, due to binding variability, their results were not not statistically significant.

(7) Can the authors explain how centrosome enrichment is calculated in Figure 4F? The intensity of colocalization with the centrosome between mutant constructs visually does not look significantly different. Is this a ratio of centrosome localization to cell body localization? 

We apologize for this omission. This has been added to the quantification section of the Methods (lines 721-723). Yes, it is a ratio of mean signal at the centrosome vs mean signal in the rest of the cell.

(8) The current input blot in Supplementary Figure 4A shows increasing amounts of importin beta across the lanes. Do the authors have a blot of panel A in which the input level of importin beta is the same between constructs? Does this change the level of importin beta that is pulled down?

Another replicate of this experiment has been shown. We have retained the original experiment as well (Supplemental Figs. 4A, B).

Reviewer #3 (Recommendations for the authors): 

Minor points: 

(1) In the .pdf version of the supplemental tables, the text is often cropped. It is recommended to delete the .pdf versions and just retain the Excel versions of the tables. 

We are not sure why this occurred. Excel files were provided. In addition, the raw data from the mass spectrometry experiments will also be included with the final version of the manuscript.

(2) Line 367: For transport of Rab6, kinesin-1 is the dominant motor, but dynein is still active and engaging in a tug of war (Serra Marquez et al 2022). 

Thank you. We have revised our text to include this discussion. In this regard, LAMP1 vesicles are similar. Loss of BICD2 results in a greater number of stationary vesicles rather than vesicles that are excessively targeted towards the microtubules minus end.

(3) Line 371: BicD2 is required for the transport of RanBP2 from annulate lamellae to nuclear pore complexes.

Thank you. We have modified our text. 

(4) Yi et al., 2023 have previously shown changed interactions of the BicD2/R747C mutant, such as decreased binding to Nup358 and increased binding to Nesprin-2, as well as functional implications for the associated brain developmental pathways, which should be acknowledged.

We apologize for leaving this out. In the original version of the manuscript, we were attempting to keep the discussion more concise. We have added a discussion of these findings in the revised manuscript (lines 496-507).

Author response:

Reviewer #1 (Public review):

The study examines how pyruvate, a key product of glycolysis that influences TCA metabolism and gluconeogenesis, impacts cellular metabolism and cell size. It primarily utilizes the Drosophila liver-like fat body, which is composed of large post-mitotic cells that are metabolically very active. The study focuses on the key observations that over-expression of the pyruvate importer MPC complex (which imports pyruvate from the cytoplasm into mitochondria) can reduce cell size in a cell-autonomous manner. They find this is by metabolic rewiring that shunts pyruvate away from TCA metabolism and into gluconeogenesis. Surprisingly, mTORC and Myc pathways are also hyper-active in this background, despite the decreased cell size, suggesting a non-canonical cell size regulation signaling pathway. They also show a similar cell size reduction in HepG2 organoids. Metabolic analysis reveals that enhanced gluconeogenesis suppresses protein synthesis. Their working model is that elevated pyruvate mitochondrial import drives oxaloacetate production and fuels gluconeogenesis during late larval development, thus reducing amino acid production and thus reducing protein synthesis.

Strengths:

The study is significant because stem cells and many cancers exhibit metabolic rewiring of pyruvate metabolism. It provides new insights into how the fate of pyruvate can be tuned to influence Drosophila biomass accrual, and how pyruvate pools can influence the balance between carbohydrate and protein biosynthesis. Strengths include its rigorous dissection of metabolic rewiring and use of Drosophila and mammalian cell systems to dissect carbohydrate:protein crosstalk.

Weaknesses:

However, questions on how these two pathways crosstalk, and how this interfaces with canonical Myc and mTORC machinery remain. There are also questions related to how this protein:carbohydrate crosstalk interfaces with lipid biosynthesis. Addressing these will increase the overall impact of the study.

We thank the reviewer for recognizing the significance of our work and for providing constructive feedback. Our findings indicate that elevated pyruvate transport into mitochondria acts independently of canonical pathways, such as mTORC1 or Myc signaling, to regulate cell size. To investigate these pathways, we utilized immunofluorescence with well-validated surrogate measures (p-S6 and p-4EBP1) in clonal analyses of MPC expression, as well as RNA-seq analyses in whole fat body tissues expressing MPC. These methods revealed hyperactivation of mTORC1 and Myc signaling in fat body cells expressing MPC in Drosophila, which are dramatically smaller than control cells. One explanation of these seemingly contradictory observations could be an excess of nutrients that activate mTORC1 or Myc pathways. However, our data is inconsistent with a nutrient surplus that could explain this hyperactivation. Instead, we observed reduced amino acid abundance upon MPC expression, which is very surprising given the observed hyperactivation of mTORC1. This led us to hypothesize the existence of a feedback mechanism that senses inappropriate reductions in cell size and activates signaling pathways to promote cell growth. The best characterized “sizer” pathway for mammalian cells is the CycD/CDK4 complex which has been well studied in the context of cell size regulation of the cell cycle (PMID 10970848, 34022133). However, the mechanisms that sense cell size in post-mitotic cells, such as fat body cells and hepatocytes, remain poorly understood. Investigating the hypothesized size-sensing mechanisms at play here is a fascinating direction for future research.

For the current study, we conducted epistatic analyses with mTOR pathway members by overexpressing PI3K and knocking down the TORC1 inhibitor Tuberous Sclerosis Complex 1 (Tsc1). These manipulations increased the size of control fat body cells but not those over-expressing the MPC (Supplementary Fig. 3c, 3d). Regarding Myc, its overexpression increased the size of both control and MPC+ clones (Supplementary Fig. 3e), but Myc knockdown had no additional effect on cell size in MPC+ clones (Supplementary Fig. 3f). These results suggest that neither mTORC1, PI3K, nor Myc are epistatic to the cell size effects of MPC expression. Consequently, we shifted our focus to metabolic mechanisms regulating biomass production and cell size.

When analyzing cellular biomolecules contributing to biomass, we observed a significant impact on protein levels in Drosophila fat body cells and mammalian MPC-expressing HepG2 spheroids. TAG abundance in MPC-expressing HepG2 spheroids and whole fat body cells showed a statistically insignificant decrease compared to controls. Furthermore, lipid droplets in fat body cells were comparable in MPC-expressing clones when normalized to cell size.

Interestingly, RNA-seq analysis revealed increased expression of fatty acid and cholesterol biosynthesis pathways in MPC-expressing fat body cells. Upregulated genes included major SREBP targets, such as ATPCL (2.08-fold), FASN1 (1.15-fold), FASN2 (1.07-fold), and ACC (1.26-fold). Since mTOR promotes SREBP activation and MPC-expressing cells showed elevated mTOR activity and upregulation of SREBP targets, we hypothesize that SREBP is activated in these cells. Nonetheless, our data on amino acid abundance and its impact on protein synthesis activity suggest that protein abundance, rather than lipids, is likely to play a larger causal role in regulating cell size in response to increased pyruvate transport into mitochondria.

Reviewer #2 (Public review):

In this manuscript, the authors leverage multiple cellular models including the drosophila fat body and cultured hepatocytes to investigate the metabolic programs governing cell size. By profiling gene programs in the larval fat body during the third instar stage - in which cells cease proliferation and initiate a period of cell growth - the authors uncover a coordinated downregulation of genes involved in mitochondrial pyruvate import and metabolism. Enforced expression of the mitochondrial pyruvate carrier restrains cell size, despite active signaling of mTORC1 and other pathways viewed as traditional determinants of cell size. Mechanistically, the authors find that mitochondrial pyruvate import restrains cell size by fueling gluconeogenesis through the combined action of pyruvate carboxylase and phosphoenolpyruvate carboxykinase. Pyruvate conversion to oxaloacetate and use as a gluconeogenic substrate restrains cell growth by siphoning oxaloacetate away from aspartate and other amino acid biosynthesis, revealing a tradeoff between gluconeogenesis and provision of amino acids required to sustain protein biosynthesis. Overall, this manuscript is extremely rigorous, with each point interrogated through a variety of genetic and pharmacologic assays. The major conceptual advance is uncovering the regulation of cell size as a consequence of compartmentalized metabolism, which is dominant even over traditional signaling inputs. The work has implications for understanding cell size control in cell types that engage in gluconeogenesis but more broadly raise the possibility that metabolic tradeoffs determine cell size control in a variety of contexts.

We thank the reviewer for their thoughtful recognition of our efforts, and we are honored by the enthusiasm the reviewer expressed for the findings and the significance of our research. We share the reviewer’s opinion that our work might help to unravel metabolic mechanisms that regulate biomass gain independent of the well-known signaling pathways.

Reviewer #3 (Public review):

Summary:

In this article, Toshniwal et al. investigate the role of pyruvate metabolism in controlling cell growth. They find that elevated expression of the mitochondrial pyruvate carrier (MPC) leads to decreased cell size in the Drosophila fat body, a transformed human hepatocyte cell line (HepG2), and primary rat hepatocytes. Using genetic approaches and metabolic assays, the authors find that elevated pyruvate import into cells with forced expression of MPC increases the cellular NADH/NAD+ ratio, which drives the production of oxaloacetate via pyruvate carboxylase. Genetic, pharmacological, and metabolic approaches suggest that oxaloacetate is used to support gluconeogenesis rather than amino acid synthesis in cells over-expressing MPC. The reduction in cellular amino acids impairs protein synthesis, leading to impaired cell growth.

Strengths:

This study shows that the metabolic program of a cell, and especially its NADH/NAD+ ratio, can play a dominant role in regulating cell growth.

The combination of complementary approaches, ranging from Drosophila genetics to metabolic flux measurements in mammalian cells, strengthens the findings of the paper and shows a conservation of MPC effects across evolution.

Weaknesses:

In general, the strengths of this paper outweigh its weaknesses. However, some areas of inconsistency and rigor deserve further attention.

Thank you for reviewing our manuscript and offering constructive feedback. We appreciate your recognition of the significance of our work and your acknowledgment of the compelling evidence we have presented. We will carefully revise the manuscript in line with the reviewers' recommendations.

The authors comment that MPC overrides hormonal controls on gluconeogenesis and cell size (Discussion, paragraph 3). Such a claim cannot be made for mammalian experiments that are conducted with immortalized cell lines or primary hepatocytes.

We appreciate the reviewer’s insightful comment. Pyruvate is a primary substrate for gluconeogenesis, and our findings suggest that increased pyruvate transport into mitochondria increases the NADH-to-NAD+ ratio, and thereby elevates gluconeogenesis. Notably, we did not observe any changes in the expression of key glucagon targets, such as PC, PEPCK2, and G6PC, suggesting that the glucagon response is not activated upon MPC expression. By the statement referenced by the reviewer, we intended to highlight that excess pyruvate import into mitochondria drives gluconeogenesis independently of hormonal and physiological regulation.

It seems the reviewer might also have been expressing the sentiment that our in vitro models may not fully reflect the in vivo situation, and we completely agree.  Moving forward, we plan to perform similar analyses in mammalian models to test the in vivo relevance of this mechanism. For now, we will refine the language in the manuscript to clarify this point.

Nuclear size looks to be decreased in fat body cells with elevated MPC levels, consistent with reduced endoreplication, a process that drives growth in these cells. However, acute, ex vivo EdU labeling and measures of tissue DNA content are equivalent in wild-type and MPC+ fat body cells. This is surprising - how do the authors interpret these apparently contradictory phenotypes?

We thank the reviewer for raising this important issue. The size of the nucleus is regulated by DNA content and various factors, including the physical properties of DNA, chromatin condensation, the nuclear lamina, and other structural components (PMID 32997613). Additionally, cytoplasmic and cellular volume also impacts nuclear size, as extensively documented during development (PMID 17998401, PMID 32473090).

In MPC-expressing cells, it is plausible that the reduced cellular volume impacts chromatin condensation or the nuclear lamina in a way that slightly decreases nuclear size without altering DNA content. Specifically, in our whole fat body experiments using CG-Gal4 (as shown in Supplementary Figure 2a-c), we noted that after 12 hours of MPC expression, cell size was significantly reduced (Supplementary Figure 2c and Author response image 1A). However, the reduction in nuclear size became significant only after 36 hours of MPC expression (Author response image 1B), suggesting that the reduction in cell size is a more acute response to MPC expression, followed only later by effects on nuclear size.

In clonal analyses, this relationship was further clarified. MPC-expressing cells with a size greater than 1000 µm² displayed nuclear sizes comparable to control cells, whereas those with a drastic reduction in cell size (less than 1000 µm²) exhibited smaller nuclei (Author response image 1C and D). These observations collectively suggest that changes in nuclear size are more likely to be downstream rather than upstream of cell size reduction. Given that DNA content remains unaffected, we focused on investigating the rate of protein synthesis. Our findings suggest that protein synthesis might play a causal role in regulating cell size, thereby reinforcing the connection between cellular and nuclear size in this context.

Author response image 1.

Cell Size vs. Nuclear Size in MPC-Expressing Fat Body Cells. A. Cell size comparison between control (blue, ay-GFP) and MPC+ (red, ay-MPC) fat body cells over time, measured in hours after MPC expression induction. B. Nuclear area measurements from the same fat body cells in ay-GFP and ay-MPC groups. C. Scatter plot of nuclear area vs. cell area for control (ay-GFP) cells, including the corresponding R^² value. D. Scatter plot of nuclear area vs. cell area for MPC-expressing (ay-MPC) cells, with the respective R^² value.

This image highlights the relationship between nuclear and cell size in MPC-expressing fat body cells, emphasizing the distinct cellular responses observed following MPC induction.

In Figure 4d, oxygen consumption rates are measured in control cells and those over-expressing MPC. Values are normalized to protein levels, but protein is reduced in MPC+ cells. Is oxygen consumption changed by MPC expression on a per-cell basis?

As described in the manuscript, MPC-expressing cells are smaller in size. In this context, we felt that it was most appropriate to normalize oxygen consumption rates (OCR) to cellular mass to enable an accurate interpretation of metabolic activity. Therefore, we normalized OCR with protein content to account for variations in cellular size and (probably) mitochondrial mass.

Trehalose is the main circulating sugar in Drosophila and should be measured in addition to hemolymph glucose. Additionally, the units in Figure 4h should be related to hemolymph volume - it is not clear that they are.

We appreciate this valuable suggestion. In the revised manuscript, we will quantify trehalose abundance in circulation and within fat bodies. As described in the Methods section, following the approach outlined in Ugrankar-Banerjee et al., 2023, we bled 10 larvae (either control or MPC-expressing) using forceps onto parafilm. From this, 2 microliters of hemolymph were collected for glucose measurement. We will apply this methodology to include the trehalose measurements as part of our updated analysis.

Measurements of NADH/NAD ratios in conditions where these are manipulated genetically and pharmacologically (Figure 5) would strengthen the findings of the paper. Along the same lines, expression of manipulated genes - whether by RT-qPCR or Western blotting - would be helpful to assess the degree of knockdown/knockout in a cell population (for example, Got2 manipulations in Figures 6 and S8).

We appreciate this suggestion, which will provide additional rigor to our study. We have already quantified NADH/NAD+ ratios in HepG2 cells under UK5099, NMN, and Asp supplementation, as presented in Figure 6k. As suggested, we will quantify the expression of Got2 manipulations mentioned in Figure 6j using RT-qPCR and validate the corresponding data in Supplementary Figure 8f through western blot analysis.

Additionally, we will assess the efficiency of pcb, pdha, dlat, pepck2, and Got2 manipulations used to modulate the expression of these genes. These validations will ensure the robustness of our findings and strengthen the conclusions of our study.

Author response:

Reviewer #1:

Weaknesses:

(1) The crystal structure of HsIFT172c reveals a single globular domain formed by the last three TPR repeats and C-terminal residues of IFT172. However, the authors subdivide this globular domain into TPR, linker, and U-box-like regions that they treat as separate entities throughout the manuscript. This is potentially misleading as the U-box surface that is proposed to bind ubiquitin or E2 is not surface accessible but instead interacts with the TPR motifs. They justify this approach by speculating that the presented IFT172c structure represents an autoinhibited state and that the U-box-like domain can become accessible following phosphorylation. However, additional evidence supporting the proposed autoinhibited state and the potential accessibility of the U-box surface following phosphorylation is needed, as it is not tested or supported by the current data.

We thank the reviewer for this comment. IFT172C contains TPR region and Ubox-like region which are admittedly tightly bound to each other. While there is a possibility that this region functions and exists as one domain, below are the reasons why we chose to classify these regions as two different domains.

(1) TPR and Ubox-like regions are two different structural classes

(2) TPR region is linked to Ubox-like region via a long linker which seems poised to regulate the relative movement between these regions.

(3) Many ciliopathy mutations are mapped to the interface of TPR region and the Ubox region hinting at a regulatory mechanism governed by this interface.

(2) While in vitro ubiquitination of IFT172 has been demonstrated, in vivo evidence of this process is necessary to support its physiological relevance.

We thank the reviewer for this comment. We are currently working on identifying the substrates of IF172 to reveal the physiological relevant of its ubiquitination activity.

(3) The authors describe IFT172 as being autoubiquitinated. However, the identified E2 enzymes UBCH5A and UBCH5B can both function in E3-independent ubiquitination (as pointed out by the authors) and mediate ubiquitin chain formation in an E3-independent manner in vitro (see ubiquitin chain ladder formation in Figure 3A). In addition, point mutation of known E3-binding sites in UBCH5A or TPR/U-box interface residues in IFT172 has no effect on the mono-ubiquitination of IFT172c1. Together, these data suggest that IFT172 is an E3-independent substrate of UBCH5A in vitro. The authors should state this possibility more clearly and avoid terminology such as "autoubiquitination" as it implies that IFT172 is an E3 ligase, which is misleading. Similarly, statements on page 10 and elsewhere are not supported by the data (e.g. "the low in vitro ubiquitination activity exhibited by IFT172" and "ubiquitin conjugation occurring on HsIFT172C1 in the presence of UBCH5A, possibly in coordination with the IFT172 U-box domain").

We now consider this possibility and tone down our statements about the autoubiquitination activity of IFT172 in a revised version of the manuscript.

(4) Related to the above point, the conclusion on page 11, that mono-ubiquitination of IFT172 is U-box-independent while polyubiquitination of IFT172 is U-box-dependent appears implausible. The authors should consider that UBCH5A is known to form free ubiquitin chains in vitro and structural rearrangements in F1715A/C1725R variants could render additional ubiquitination sites or the monoubiquitinated form of IFT172 inaccessible/unfavorable for further processing by UBCH5A.

We now consider this possibility and tone down our statements about the autoubiquitination activity of IFT172 in the conclusion on pg. 11.

(5) Identification of the specific ubiquitination site(s) within IFT172 would be valuable as it would allow targeted mutation to determine whether the ubiquitination of IFT172 is physiologically relevant. Ubiquitination of the C1 but not the C2 or C3 constructs suggests that the ubiquitination site is located in TPRs ranging from residues 969-1470. Could this region of TPR repeats (lacking the IFT172C3 part) suffice as a substrate for UBCH5A in ubiquitination assays?

We thank the reviewer for raising this important point about ubiquitination site identification. While not included in our manuscript, we did perform mass spectrometry analysis of ubiquitination sites using wild-type IFT172 and several mutants (P1725A, C1727R, and F1715A). As shown in the figure below, we detected multiple ubiquitination sites across these constructs. The wild-type protein showed ubiquitination at positions K1022, K1237, K1271, and K1551, while the mutants displayed slightly different patterns of modification. However, we should note that the MS intensity signals for these ubiquitinated peptides were relatively low compared to unmodified peptides, making it difficult to draw strong conclusions about site specificity or physiological relevance.

Author response image 1.

These results align with the reviewer's suggestion that ubiquitination occurs within the TPR-containing region. However, given the technical limitations of the MS analysis and the potential for E3-independent ubiquitination by UBCH5A, we have taken a conservative approach in interpreting these findings.

(6) The discrepancy between the molecular weight shifts observed in anti-ubiquitin Western blots and Coomassie-stained gels is noteworthy. The authors show the appearance of a mono-ubiquitinated protein of ~108 kDa in anti-ubiquitin Western blots. However, this molecular weight shift is not observed for total IFT172 in the corresponding Coomassie-stained gels (Figures 3B, D, F). Surprisingly, this MW shift is visible in an anti-His Western blot of a ubiquitination assay (Fig 3C). Together, this raises the concern that only a small fraction of IFT172 is being modified with ubiquitin. Quantification of the percentage of ubiquitinated IFT172 in the in vitro experiments could provide helpful context.

We do acknowledge in the manuscript is that the conjugation of ubiquitins to IFT172C is weak (Page 16). Future experiments of identification of potential substrates and its implications in ciliary regulation will provide further context to our in vitro ubiquitination experiments.

(7) The authors propose that IFT172 binds ubiquitin and demonstrate that GST-tagged HsIFT172C2 or HsIFT172C3 can pull down tetra-ubiquitin chains. However, ubiquitin is known to be "sticky" and to have a tendency for weak, nonspecific interactions with exposed hydrophobic surfaces. Given that only a small proportion of the ubiquitin chains bind in the pull-down, specific point mutations that identify the ubiquitin-binding site are required to convincingly show the ubiquitin binding of IFT172.

(8) The authors generated structure-guided mutations based on the predicted Ub-interface and on the TPR/U-box interface and used these for the ubiquitination assays in Fig 3. These same mutations could provide valuable insights into ubiquitin binding assays as they may disrupt or enhance ubiquitin binding (by relieving "autoinhibition"), respectively. Surprisingly, two of these sites are highlighted in the predicted ubiquitin-binding interface (F1715, I1688; Figure 4E) but not analyzed in the accompanying ubiquitin-binding assays in Figure 4.

We agree that these mutations could provide insights into ubiquitin binding by IFT172. We are currently pursuing further mutagenesis studies on the IFT172-Ub interface based on the AF model. We however have evaluated the ubiquitin binding activity of the mutant F1715A using similar pulldowns, which showed no significant impact for the mutation on the ubiquitin binding activity of IFT172. We are yet to evaluate the impact of alternate amino acid substitutions at these positions. The I1688 mutants we cloned could not be expressed in soluble form, thus could not be used for testing in ubiquitination activity or ubiquitin binding assays.

(9) If IFT172 is a ubiquitin-binding protein, it might be expected that the pull-down experiments in Figure S1 would identify ubiquitin, ubiquitinated proteins, or E2 enzymes. These were not observed, raising doubt that IFT172 is a ubiquitin-binding protein.

It is likely that IFT172 only binds ubiquitin with low affinity as indicated by our in vitro pulldowns and the AF interface. In our pull down experiment performed using the Chlamy flagella extracts, we have used extensive washes to remove non-specific interactors. This might have also excluded the identification of weak but bona fide interactors of IFT172. Additionally, we have not used any ubiquitination preserving reagents such as NEM in our pulldown buffers, exposing the cellular ubiquitinated proteins to DUB mediated proteolysis further preventing their identification in our pulldown/MS experiment.

(10) The cell-based experiments demonstrate that the U-box-like region is important for the stability of IFT172 but does not demonstrate that the effect on the TGFb pathway is due to the loss of ubiquitin-binding or ubiquitination activity of IFT172.

We acknowledge that our current data cannot distinguish whether the TGFβ pathway defects arise from general protein instability or from specific loss of ubiquitin-related functions. Our experiments demonstrate that the U-box-like region is required for both IFT172 stability and proper TGFβ signaling, but we agree that establishing a direct mechanistic link between these phenomena would require additional evidence. We will revise our discussion to more clearly acknowledge this limitation in our current understanding of the relationship between IFT172's U-box region and TGFβ pathway regulation.

(11) The challenges in experimentally validating the interaction between IFT172 and the UBX-domain-containing protein are understandable. Alternative approaches, such as using single domains from the UBX protein, implementing solubilizing tags, or disrupting the predicted binding interface in Chlamydomonas flagella pull-downs, could be considered. In this context, the conclusion on page 7 that "The uncharacterized UBX-domain-containing protein was validated by AF-M as a direct IFT172 interactor" is incorrect as a prediction of an interaction interface with AF-M does not validate a direct interaction per se.

We agree with the reviewer that our AlphaFold-Multimer (AF-M) predictions alone do not constitute experimental validation of a direct interaction. We appreciate the reviewer's understanding of the technical challenges in validating this interaction experimentally. We will revise our text to more precisely state that "The uncharacterized UBX-domain-containing protein was validated by AF-M as a potential direct IFT172 interactor" and will discuss the AF-M predictions as computational evidence that suggests, but does not prove, a direct interaction. This more accurately reflects the current state of our understanding of this potential interaction.

Reviewer #3:

Weaknesses:

(1) Interaction studies were carried out by pulldown experiments, which identified more IFT172 interaction partners. Whether these interactions can be seen in living cells remains to be elucidated in subsequent studies.

We agree with the reviewer that validation of protein-protein interactions in living cells provides important physiological context. While our pulldown experiments have identified several promising interaction partners and the AF-M predictions provide computational support for these interactions, we acknowledge that demonstrating these interactions in vivo would strengthen our findings. However, we believe our current biochemical and structural analyses provide valuable insights into the molecular basis of IFT172's interactions, laying important groundwork for future cell-based studies.

(2) The cell culture-based experiments in the IFT172 mutants are exciting and show that the U-box domain is important for protein stability and point towards involvement of the U-box domain in cellular signaling processes. However, the characterization of the generated cell lines falls behind the very rigorous analysis of other aspects of this work.

We thank the reviewer for noting that the characterization of our cell lines could be more rigorous. In the revised manuscript, we will provide additional characterization of the cell lines, including detailed sequencing information and validation data for the IFT172 mutants. This will bring the documentation of our cell-based experiments up to the same standard as other aspects of our work.

Author response:

We thank the reviewers for their help and their suggestions to make this manuscript more rigorous. We would like to post provisional author responses when eLife publish the reviewed preprint, and the more detailed responses will be supplemented with the revised manuscript.

• There are questions about choices made in the computational approach (architecture and type of generative model, training set).

We will train a new generator model based on the current GAN architecture, but with ‘hybrid’ AMP/AVP training sets (Reviewer 1 and 3). Hence, we can directly compare the performances of two generators. Based on our preliminary data, providing GAN with more AVP sequences during training helped the designed peptides pass the AVP filter, at the cost of reducing the average AMPredicgtor scores. The new generator also elevated the diversity of designed sequences.

We also perturbed the detailed architecture of our deep learning models, including fully-connected graph edge encodings and different versions of ESM (e.g. esm1b_t33_650M_UR50S, esm2_t48_15B_UR50D, Reviewer 2). In the revised manuscript, we will report the effects of these modifications and suggest the overall construct of GCN and GAN are suitable for a light-weight sequence label model, as demonstrated in Author response table 1 and 2. For the generator, we suggest that using our approach, we may have reached a plateau for the GAN sampling (Author response table 3).

Author response table 1.

Results of AMPredictor with different graph edge encodings

Author response table 2.

Results of AMPredictor with different ESM versions

Author response table 3.

Evaluation of generated sequences with different sampling numbers

• There is an important concern about the small number of antimicrobial peptides tested, compared to other studies, and the origin of antiviral activities.

We will address this concern by increasing the number of peptides tested in anti-microbial and anti-viral experiments. As reported in current version of our manuscript, the first generation of GAN generated 128 unique designs and the top 2% (3 designs) was tested experimentally. The second generation of GAN will produce ~1024 designs (1-2 weeks) and the top 2% (~ 20 new sequences) will be tested. We are in the process of synthesize (2-3 weeks) and MIC measurement (1 week). The overall size of tested sample will reach 20-30 sequences. We will focus on sequences with low similarity (< 30%) to any known AMPs, thus expanding the universe functional peptides. We estimated the collection of these new data in 6 weeks.

Author response:

Reviewer #1 (Public Review):

(1) Figure 3: it is unclear what is the efficiency of Msi2 deletion shRNA - could you demonstrate it by at least two independent methods? (QPCR, Western, or IHC?) please quantitate the data.

In Figure 3, we did not delete Msi2 via shRNA. Instead, we utilized a genetic model in which the Msi2 gene was disrupted via gene trap mutagenesis. We have also used this model in previous publications to define the impact of Msi2 loss in other systems1.

(2) In Figure 4, similarly, it is unclear if Msi2 depletion was effective- and what is shRNA efficiency. Please test this by at least two independent methods (QPCR, Western, or IHC) and also please quantitate the data

We demonstrated that the efficiency of Msi2 depletion was ~83% (Figures 4A and 4C) via qPCR analysis for our in vitro and in vivo experiments, respectively, and verified the knockdown via bulk RNA-seq analysis. The shRNA hairpin used was previously validated and published by our lab2.

(3) the reason for impairment of cell growth demonstrated in Figs 3 and 4 is not clear: is it apoptosis? Necrosis? Cell cycle defects? Autophagy? Senescence? Please probe 2-3 possibilities and provide the data.

The basis of the cell growth impairment after Msi2 deletion/knockdown in this paper is certainly an important question, and future experiments will be performed to better delineate this. In previous publications loss of Msi2 in leukemia cells has been shown to inhibit growth via arrested cell cycle progression by increasing the expression of p213. Further, loss of Msi2 was also shown to promote apoptosis in part by upregulating Bax3. These data suggest that Msi2 can have an impact via multiple distinct mechanisms including by mediating cell cycle arrest and blocking apoptosis. While these specific genes were not detectably changed after loss of Msi2 in lung cancer cells, other genes in these and other pathways will be important to study in the future.

(4) Since Musashi-1 is a Musashi-2 paralogue that could compensate for Musashi-2 loss, please test Msi1 expression levels in matching Fig 3 and Fig 4 sections (in cells/ tumors with Msi2 deletion and in KP cells with Msi2 shRNA). One method could suffice here.

In our RNA-seq of cells following Msi2 knockdown, Msi1 expression was undetectable. The TPM values for Msi1 in control and knockdown cells were less than 0.01, suggesting that it did not compensate for the loss of Msi2.

(5) It is not exactly clear why RNA-seq (as opposed to proteomics) was done to investigate downstream Msi2 targets (since Msi2 is in first place, translational and not transcriptional regulator)- . RNA effects in Fig 5J are quite modest, 2-fold or so. It would be useful (if antibodies available) to test four targets in Fig 5J by Western blot, to see any impact of musashi-2 depletion on those target protein levels. Indeed, several papers - including Kudinov et al PNAS, PMID: 27274057, Makhov P et al PMID: 33723247 and PMID: 37173995 - used proteomics/ RIP approaches and found direct Musashi-2 targets in lung cancer, including EGFR, and others.

Previous published work from the lab showed that expression of Msi2 in the context of myeloid leukemia1can not only repress NUMB protein (I believe protein should be all caps?) (as has been previously demonstrated in the nervous system) but also Numb RNA. This indicated that as an RNA binding protein, Msi2 also can bind and destabilize direct binding targets such as Numb; this was the reason for pursuing transcriptomic analysis.  However as the reviewer suggests, proteomic studies are certainly very important to develop a complete picture of the impact of Musashi to determine which targets are controlled by Msi2 at the protein level.

Reviewer #2 (Public Review):

(1) It will be interesting to determine whether Msi2+ cells are a relatively stable subset or rather the Msi2+ cells in lung is a dynamic concept that is transient or interconvertible. This is relevant to the interpretation of what Msi2 positivity really means.

In previous unpublished work from our lab, we have found that Msi2+ cells from a GFP reporter KPf/fC mouse are readily able to become GFP negative (Msi2-), but the inverse is not true. Specifically, when Msi2+ KPf/fC pancreatic cells were transplanted into the flanks of NSG mice, Msi2+ cells formed tumors in all recipients; these tumors contained both GFP+ and GFP- cells (over 80%)  recapitulating the original heterogeneity and suggesting GFP+ cells can give rise to both GFP+ and GFP- cells (Lytle and Reya, unpublished observations). In contrast only a small subset of GFP- transplanted mice formed tumors. One of the rare GFP- derived tumors was isolated and found to contain largely GFP- cells, with ~0.1% GFP+ cells. The small frequency of GFP expression could be from contaminating cells or may suggest that GFP- cells retain some ability to switch on Msi under selective pressure, and that although they pose a lower risk of driving tumorigenesis than Msi+ cells, they may nonetheless bear latent potential to become higher risk. These data may offer a possible model for projecting the potential of Msi2+ cells in the lung, but is something that needs to be further studied in this tissue.

(2) Does Kras mutation and/or p53 loss upregulate Msi2? This point and the point above are related to whether Msi2+ cells are truly more susceptible to tumorigenesis, as the authors suggested.

In unpublished work from our lab, we have found that Kras mutation upregulates Msi2 over baseline and subsequent p53 loss upregulates Msi2 further in the context of pancreatic cells (Lytle and Reya unpublished results), therefore it is possible that the same is true for the lung. Specifically, we have observed that Msi2 increased from normal acinar cells to Kras-mutated acinar (e.g. pancreatic intraepithelial neoplasia (PanIN)).

To address whether Msi2+ cells are more susceptible to tumorigenesis, we have recently published data showing that the stabilization of the oncogenic MYC protein in lung Msi2+ cells drive the formation of small-cell lung cancer in a new inducible Msi2-CreERT2; CAG-LSL-MycT58A mice (Msi2-Myc)4 model. More importantly, this data provides the first evidence that normal Msi2+ cells are primed and highly sensitive to MYC-driven transformation across many organs and not just the lung4.

(3) The KO of Msi2 reducing tumor number and burden in the lung cancer initiation model is interesting. However, there are two alternative interpretations. First, it is possible that the Msi2 KO mice (without Kras activation and p53 loss) has reduced total lung cell numbers or altered percentage of stem cells. There is currently only one sentence citing data not shown on line 125, commenting that there is no difference in BASC and AT2 cell populations. It will be helpful that such data are shown and the effect of KO on overall lung mass or cellularity is clarified. Second, the phenotype may also be due to a difference in the efficiencies of cre on Kras and p53 in the Msi2 WT and KO mice.

We isolated the lungs of three Msi2 WT and three Msi2 KO mice and used immunofluorescence staining to stain for CC10 (BASC) and SPC (AT2) to determine if these cell populations were reduced after Msi2 loss alone. Below are representative images showing that the Msi2 KO mice did not have lower numbers of both BASC and AT2 cell populations. 

Author response image 1.

(4) All shRNA experiments (for both Msi2 KD and the KD of candidate genes) utilized a single shRNA. This approach cannot exclude off-target effects of the shRNA.

The shRNA hairpin used for Msi2 was previously validated and published by our lab2. Additionally, in this work we did develop and use a Msi2 genetic knockout mouse model that validates our shRNA knockdown data showing the specific impact of Msi2 on lung tumor growth.

(5) The technical details of the PDX experiment (Figure 4F) are not fully explained.

Due to space considerations, we were unable not put the specifics in the legend, but the details are in the methods section (Flank Transplant Assays). In brief, 500,000 cells/well were plated in a 6-well plate coated with Matrigel and 83,000 cells/well were plated in a 24-well plate coated with Matrigel for subsequent determination of transduction efficiency via FACS. 24 hours after transduction, media from the cells was collected and placed on ice. 1mL of 2mg/mL collagenase/dispase was then added to the well and incubated for 45 minutes at 37ºC to dissociate the remaining cells from Matrigel followed by subsequent washes. Cells were pelleted by centrifugation and an equivalent number of shControl and shMsi2 transduced cells were resuspended in full media, mixed at a 1:1 ratio with growth factor reduced Matrigel at a final volume of 100 μL, and transplanted subcutaneously into the flanks of NSG recipient mice.

Reviewer #3 (Public Review):

- In Figure 1, characterization of Msi2 expression in the normal mouse lung was carried out by using a Msi2-GFP Knock-in reporter and analyzed by flow cytometry followed by cytospins and immunostaining. Additional characterization of Msi2 expression by co-immunostaining with well-known markers of airway and alveolar cell types in intact lung tissue will strengthen the existing data and provide more specific information about Msi2 expression and abundancy in relevant cell types. It will be also interesting to know whether Msi2 is expressed or not in other abundant lung cell types such as ciliated and AT1 cells.

We performed co-staining of Msi2 and CC10 as well as Msi2 and SPC in Figure 1C. In the future we can include additional markers as well as markers for airway and other alveolar cell types.

- While this set of experiments provide strong evidence that Msi2 is required for tumor progression and growth in lung adenocarcinoma, it is unclear whether normal Msi2+ lung cells are more responsive to transformation or whether Msi2 is upregulated early during the process of tumorigenesis. Future lineage tracing experiments using Msi2-CreER and mouse models of chemically-induced lung carcinogenesis will provide additional data that will fully support this claim.

Recently, we published data showing that Msi2 is expressed in Clara cells at the bronchoalveolar junction in the lung of our new Msi2-CreERT2 knock-in mouse model4. Furthermore, stabilization of the oncogenic MYC protein in these specific cells to model Myc amplification was sufficient to drive the formation of small-cell lung cancer4. These data excitingly demonstrate that Msi2+ cells are more responsive to transformation after Myc stabilization.

- In Figure 4F, Patient-derived xenograft (PDX) assays were conducted in 2 patients only and the percentage of cells infected by shRNA-Msi2 is low in both PDX (30% and 10% for patient 1 and 2 respectively). It is surprising that Msi2 downregulation in a small percentage of tumor cells has such a dramatic effect on tumor growth and expansion. Confirmation of this finding with additional patient samples would suggest an important non-cell autonomous role for Msi2 in lung adenocarcinoma.

In the future we hope to collect more patient samples to further validate the data presented with the first 2 patients shown here. We are not certain about the reason behind the large impact of Msi2 inhibition, but as cancer stem cells drive the formation of the rest of the tumor and also drive the stromal microenvironment, it is possible that when Msi2 is deleted, Msi2- cells no longer form tumors? and also the ability to build the stromal microenvironment is impacted. This possibility needs to be further tested in future experiments.

References

(1) Ito, T. Kwon, H. Y., Zimdahl, B., Congdon, K. L., Blum, J., Lento, W. E., Zhao, C., Lagoo, A., Gerrard, G., Foroni, L., Goldman, J., Goh, H., Kim, S. H., Kim, D. W., Chuah, C., Oehler, V. G., Radich, J. P., Jordan, C. T., & Reya, T. Regulation of myeloid leukaemia by the cell-fate determinant Musashi. Nature 466, 765–768 (2010).

(2) Fox, R. G. Lytle, N. K., Jaquish, D. V., Park, F. D., Ito, T., Bajaj, J., Koechlein, C. S., Zimdahl, B., Yano, M., Kopp, J. L., Kritzik, M., Sicklick, J. K., Sander, M., Grandgenett, P. M., Hollingsworth, M. A., Shibata, S., Pizzo, D., Valasek, M. A., Sasik, R., Scadeng, M., Okano, H., Kim, Y., MacLeod, A. R., Lowy, A. M., & Reya, T. Image-based detection and targeting of therapy resistance in pancreatic adenocarcinoma. Nature 534, 407–411 (2016).

(3) Zhang, H. Tan, S., Wang, J., Chen, S., Quan, J., Xian, J., Zhang, Ss., He, J., & Zhang, L. Musashi2 modulates K562 leukemic cell proliferation and apoptosis involving the MAPK pathway. Exp Cell Res 320, 119-27 (2014).

(4) Rajbhandari, N., Hamilton, M., Quintero, C.M., Ferguson, L.P., Fox, R., Schürch, C.M., Wang, J., Nakamura, M., Lytle, N.K., McDermott, M., Diaz, E., Pettit, H., Kritzik, M., Han, H., Cridebring, D., Wen, K.W., Tsai, S., Goggins, M.G., Lowy, A.M., Wechsler-Reya, R.J., Von Hoff, D.D., Newman, A.M., & Reya, T. Single-cell mapping identifies MSI+ cells as a common origin for diverse subtypes of pancreatic cancer. Cancer Cell 41(11):1989-2005.e9 (2023).

Author Response

Reviewer #1 (Public Review):

1) “It is unclear whether new in vivo experiments were conducted for this study”.

All in vivo experiments shown were conducted independently by new researchers in the lab, using the original fly stocks. This will be more clearly stated in the revised supplement. The aim of repeating the experiments was to directly compare the consequences of impaired N- and C-terminal shedding side-by-side in two Hh-dependent developmental systems.

2) “A critical shortcoming of the study is that experiments showing Shh secretion/export do not include a Shh(-) control condition. Without demonstration that the bands analyzed are specific for Shh(+) conditions, these experiments cannot be appropriately evaluated”.

C9C5 antibody reactivity and specificity is shown below, and this control will be added to the revised manuscript. We established the C9C5 immunoblotting protocol – and generated the blot shown in Author Response Image 1 - before any of the experiments in the manuscript were started. The immunoblot clearly shows Shh specificity similar to that of R&D AF464 anti-Shh antibodies that were previously used in the lab. The immunoblot also shows that both antibodies detect the same Shh signals in media, that C9C5 is more sensitive, and that AF464 and C9C5 detect 5E1-IP’d dual-lipidated and monolipidated soluble Shh equally well. Also note that, in our hands, C9C5 is highly specific: this antibody detects N-truncated C25S;Δ26-35Shh of increased electrophoretic mobility, but does not cause unspecific signals above or below, even if the blot is strongly overexposed (as shown here). Specific Shh detection by C9C5 is also discussed in our response to editor’s comments below.

Cells were transfected with constructs encoding full-length C25SShh or truncated C25S;Δ26-35Shh, and proteins in serum-containing media were 5E1 immunoprecipitated or concentrated by heparin-sepharose pulldown. Dual-lipidated R&D 8908-SH was dissolved in the same medium and subjected to the same 5E1 immunoprecipitation or heparin pulldown. The blot was incubated with antibody AF464 and (after stripping) with antibody C9C5. Immunoblot analysis revealed high specificity of both antibodies and also revealed poor interactions of dual-lipidated 8908-SH with highly charged heparin.

3) “A stably expressing Shh/Hhat cell line would reduce condition to condition and experiment to experiment variability”.

We fully agree with this reviewer and therefore aimed to establish stable Hhat expressing cell lines several years ago. However, stable Hhat expression eliminated transfected cells after several passages, or cells gradually ceased to express Hhat, preventing us to establish a stable line despite several attempts and tried strategies. For this reason, we established transient co-expression of Shh/Hhat from the same mRNA to at least eliminate variability between relative Shh/Hhat expression levels and to assure complete Shh palmitoylation in our assays.

4) “Unusual normalization strategies are used for many experiments, and quantification/statistical analyses are missing for several experiments”.

This comment refers to data shown in Fig. 3 (here, no quantification of Scube2 function in Disp-/- cells had been conducted) and to qPCR data shown in Fig. 4 (here, Shh and C25AShh were compared only indirectly via dual-lipidated R&D 8908-SH, but not directly in a side-by-side experiment, and Shh variants with an N-terminal alanine or a serine were directly compared). We agree with the reviewer and therefore currently repeat qPCR assays and quantify blots to eliminate these technical shortcomings from the final manuscript.

5) “The study provides a modest advance in the understanding of the complex issue of Shh membrane extraction”

Our investigation identified unexpected links between Disp as a furin-activated Hh exporter, sheddase-mediated Shh release, Scube2-mediated Shh release and lipoprotein-mediated Hh transport – established modes indeed but with no previously established direct connections – that increase their relevance. We also identified a previously unknown N-processed Shh variant attached to lipoproteins and show that Disp/Scube2 function absolutely requires lipoproteins. Therefore, although we do agree that our findings are confirmatory for the above modes, they also provide new mechanistic insight and challenge the currently dominating model of Disp-mediated hand-over of dual-lipidated Hh to Scube2 chaperones (this model does not predict a role for lipoprotein particles but for both Shh lipids in signaling, for a recent discussion, see PMID 36932157). Our findings suggest an answer to the intensely debated question of whether Disp/Ptch extract cholesterol from the outer or inner plasma membrane leaflet, and suggest that N-palmitate is dispensable for signaling of lipoprotein-associated Shh to Ptch receptors. Finally, we note that previous in vivo studies in flies often relied on Hh overexpression in the fat body, raising questions on their physiological relevance. Our in vivo analyses of Hh function in wing- and eye discs are more physiologically relevant and can explain the previously reported presence of non-lipidated bioactive Hh in disc tissue (PMID: 23554573).

Reviewer #2 (Public Review):

1) “However, the results concerning the roles of lipoproteins and Shh lipid modifications are largely confirmatory of previous results, and molecular identity/physiological relevance of the newly identified Shh variant remain unclear”.

Regarding the confirmatory aspects of our work, please also refer to our response to reviewer 1. In addition, we would like to reply that our unbiased experimental approach was designed to challenge the model of Shh shedding by testing whether established Shh release regulators affect it (e.g. support it) or not. As described in our work, Disp, Scube2 and lipoproteins all contribute to increased shedding (which is new), that Disp function depends on lipoprotein presence (also new), and that lipoproteins modify the outcome of Shh shedding (dual Shh shedding versus N-shedding and lipoprotein association), which is also new.

Regarding physiological relevance, we would like to reply that our finding that artificially generated monolipidated variants (C25SShh and ShhN) solubilize in uncontrolled manner from producing cells can explain previously observed, highly variable gain-of-function or loss-of-function phenotypes upon their overexpression in vivo 1, 2, 3, 4, 5. Our data is also supported by the observed presence of variably lipidated Shh/Hh variants in vivo 6, and the in vivo observation that complete removal of Scube activity in zebrafish embryos phenocopies a complete loss of Hh function that is bypassed by increased ligand expression - and even results in wild-type-like ectopic Shh target gene expression 7. The in vivo observations are compatible with our data but are incompatible with proposed alternative models of Scube-mediated dual-lipidated Shh extraction and continued Shh/Scube association to allow for morphogen transport.

2) “Thus, it would be important to demonstrate key findings in cells that secrete Shh endogenously”.

Experimental data shown in Fig. S8B demonstrates that en-controlled expression of sheddase-resistant Hh variants blocks endogenous Hh function in the same wing disc compartment. To our knowledge, this assay is the most physiologically relevant test of the mechanism of Disp-mediated Hh release. Still, we have now started to analyze Hh from Drosophila disc tissue biochemically and hope that we can include our findings in the final manuscript.

3) “The authors could use an orthogonal approach, optimally a demonstration of physical interaction, or at least fractionation by a different parameter”.

We agree with this reviewer’s assessment and are currently in the process to establish co-IP and density gradient conditions to test physical HDL/Shh interactions. The results will be included in the final version of record.

Author Response

eLife assessment

This study presents potentially valuable results on glutamine-rich motifs in relation to protein expression and alternative genetic codes. The author's interpretation of the results is so far only supported by incomplete evidence, due to a lack of acknowledgment of alternative explanations, missing controls and statistical analysis and writing unclear to non experts in the field. These shortcomings could be at least partially overcome by additional experiments, thorough rewriting, or both.

We thank both the Reviewing Editor and Senior Editor for handling this manuscript and will submit our revised manuscript after the reviewed preprint is published by eLife.  

Reviewer #1 (Public Review):

Summary

This work contains 3 sections. The first section describes how protein domains with SQ motifs can increase the abundance of a lacZ reporter in yeast. The authors call this phenomenon autonomous protein expression-enhancing activity, and this finding is well supported. The authors show evidence that this increase in protein abundance and enzymatic activity is not due to changes in plasmid copy number or mRNA abundance, and that this phenomenon is not affected by mutants in translational quality control. It was not completely clear whether the increased protein abundance is due to increased translation or to increased protein stability.

In section 2, the authors performed mutagenesis of three N-terminal domains to study how protein sequence changes protein stability and enzymatic activity of the fusions. These data are very interesting, but this section needs more interpretation. It is not clear if the effect is due to the number of S/T/Q/N amino acids or due to the number of phosphorylation sites.

In section 3, the authors undertake an extensive computational analysis of amino acid runs in 27 species. Many aspects of this section are fascinating to an expert reader. They identify regions with poly-X tracks. These data were not normalized correctly: I think that a null expectation for how often poly-X track occur should be built for each species based on the underlying prevalence of amino acids in that species. As a result, I believe that the claim is not well supported by the data.

Strengths

This work is about an interesting topic and contains stimulating bioinformatics analysis. The first two sections, where the authors investigate how S/T/Q/N abundance modulates protein expression level, is well supported by the data. The bioinformatics analysis of Q abundance in ciliate proteomes is fascinating. There are some ciliates that have repurposed stop codons to code for Q. The authors find that in these proteomes, Q-runs are greatly expanded. They offer interesting speculations on how this expansion might impact protein function.

Weakness

At this time, the manuscript is disorganized and difficult to read. An expert in the field, who will not be distracted by the disorganization, will find some very interesting results included. In particular, the order of the introduction does not match the rest of the paper.

In the first and second sections, where the authors investigate how S/T/Q/N abundance modulates protein expression levels, it is unclear if the effect is due to the number of phosphorylation sites or the number of S/T/Q/N residues.

There are three reasons why the number of phosphorylation sites in the Q-rich motifs is not relevant to their autonomous protein expression-enhancing (PEE) activities:

First, we have reported previously that phosphorylation-defective Rad51-NTD (Rad51-3SA) and wild-type Rad51-NTD exhibit similar autonomous PEE activity. Mec1/Tel1-dependent phosphorylation of Rad51-NTD antagonizes the proteasomal degradation pathway, increasing the half-life of Rad51 from ∼30 min to ≥180 min (Ref 27; Woo, T. T. et al. 2020).

1. T. T. Woo, C. N. Chuang, M. Higashide, A. Shinohara, T. F. Wang, Dual roles of yeast Rad51 N-terminal domain in repairing DNA double-strand breaks. Nucleic Acids Res 48, 8474-8489 (2020).

Second, in our preprint manuscript, we have also shown that phosphorylation-defective Rad53-SCD1 (Rad51-SCD1-5STA) also exhibits autonomous PEE activity similar to that of wild-type Rad53-SCD (Figure 2D, Figure 4A and Figure 4C).

Third, as revealed by the results of our preprint manuscript (Figure 4), it is the percentages, and not the numbers, of S/T/Q/N residues that are correlated with the PEE activities of Q-rich motifs.

The authors also do not discuss if the N-end rule for protein stability applies to the lacZ reporter or the fusion proteins.

The autonomous PEE function of S/T/Q-rich NTDs is unlikely to be relevant to the N-end rule. The N-end rule links the in vivo half-life of a protein to the identity of its N-terminal residues. In S. cerevisiae, the N-end rule operates as part of the ubiquitin system and comprises two pathways. First, the Arg/N-end rule pathway, involving a single N-terminal amidohydrolase Nta1, mediates deamidation of N-terminal asparagine (N) and glutamine (Q) into aspartate (D) and glutamate (E), which in turn are arginylated by a single Ate1 R-transferase, generating the Arg/N degron. N-terminal R and other primary degrons are recognized by a single N-recognin Ubr1 in concert with ubiquitin-conjugating Ubc2/Rad6. Ubr1 can also recognize several other N-terminal residues, including lysine (K), histidine (H), phenylalanine (F), tryptophan (W), leucine (L) and isoleucine (I) (Bachmair, A. et al. 1986; Tasaki, T. et al. 2012; Varshavshy, A. et al. 2019). Second, the Ac/N-end rule pathway targets proteins containing N-terminally acetylated (Ac) residues. Prior to acetylation, the first amino acid methionine (M) is catalytically removed by Met-aminopeptides, unless a residue at position 2 is non-permissive (too large) for MetAPs. If a retained N-terminal M or otherwise a valine (V), cysteine (C), alanine (A), serine (S) or threonine (T) residue is followed by residues that allow N-terminal acetylation, the proteins containing these AcN degrons are targeted for ubiquitylation and proteasome-mediated degradation by the Doa10 E3 ligase (Hwang, C. S., 2019).

A. Bachmair, D. Finley, A. Varshavsky, In vivo half-life of a protein is a function of its amino-terminal residue. Science 234, 179-186 (1986).

T. Tasaki, S. M. Sriram, K. S. Park, Y. T. Kwon, The N-end rule pathway. Annu Rev Biochem 81, 261-289 (2012).

A. Varshavsky, N-degron and C-degron pathways of protein degradation. Proc Natl Acad Sci 116, 358-366 (2019).

C. S. Hwang, A. Shemorry, D. Auerbach, A. Varshavsky, The N-end rule pathway is mediated by a complex of the RING-type Ubr1 and HECT-type Ufd4 ubiquitin ligases. Nat Cell Biol 12, 1177-1185 (2010).

The PEE activities of these S/T/Q-rich domains are unlikely to arise from counteracting the N-end rule for two reasons. First, the first two amino acid residues of Rad51-NTD, Hop1-SCD, Rad53-SCD1, Sup35-PND, Rad51-ΔN, and LacZ-NVH are MS, ME, ME, MS, ME, and MI, respectively, where M is methionine, S is serine, E is glutamic acid and I is isoleucine. Second, Sml1-NTD behaves similarly to these N-terminal fusion tags, despite its methionine and glutamine (MQ) amino acid signature at the N-terminus.

The most interesting part of the paper is an exploration of S/T/Q/N-rich regions and other repetitive AA runs in 27 proteomes, particularly ciliates. However, this analysis is missing a critical control that makes it nearly impossible to evaluate the importance of the findings. The authors find the abundance of different amino acid runs in various proteomes. They also report the background abundance of each amino acid. They do not use this background abundance to normalize the runs of amino acids to create a null expectation from each proteome. For example, it has been clear for some time (Ruff, 2017; Ruff et al., 2016) that Drosophila contains a very high background of Q's in the proteome and it is necessary to control for this background abundance when finding runs of Q's.

We apologize for not explaining sufficiently well the topic eliciting this reviewer’s concern in our preprint manuscript. In the second paragraph of page 14, we cite six references to highlight that SCDs are overrepresented in yeast and human proteins involved in several biological processes (32, 74), and that polyX prevalence differs among species (43, 75-77).

1. Cheung HC, San Lucas FA, Hicks S, Chang K, Bertuch AA, Ribes-Zamora A. An S/T-Q cluster domain census unveils new putative targets under Tel1/Mec1 control. BMC Genomics. 2012;13:664.

2. Mier P, Elena-Real C, Urbanek A, Bernado P, Andrade-Navarro MA. The importance of definitions in the study of polyQ regions: A tale of thresholds, impurities and sequence context. Comput Struct Biotechnol J. 2020;18:306-13.

3. Cara L, Baitemirova M, Follis J, Larios-Sanz M, Ribes-Zamora A. The ATM- and ATR-related SCD domain is over-represented in proteins involved in nervous system development. Sci Rep. 2016;6:19050.

4. Kuspa A, Loomis WF. The genome of Dictyostelium discoideum. Methods Mol Biol. 2006;346:15-30.

5. Davies HM, Nofal SD, McLaughlin EJ, Osborne AR. Repetitive sequences in malaria parasite proteins. FEMS Microbiol Rev. 2017;41(6):923-40.

6. Mier P, Alanis-Lobato G, Andrade-Navarro MA. Context characterization of amino acid homorepeats using evolution, position, and order. Proteins. 2017;85(4):709-19.

We will cite the two references by Kiersten M. Ruff in our revised manuscript.

K. M. Ruff and R. V. Pappu, (2015) Multiscale simulation provides mechanistic insights into the effects of sequence contexts of early-stage polyglutamine-mediated aggregation. Biophysical Journal 108, 495a.

K. M. Ruff, J. B. Warner, A. Posey and P. S. Tan (2017) Polyglutamine length dependent structural properties and phase behavior of huntingtin exon1. Biophysical Journal 112, 511a.

The authors could easily address this problem with the data and analysis they have already collected. However, at this time, without this normalization, I am hesitant to trust the lists of proteins with long runs of amino acid and the ensuing GO enrichment analysis.

Ruff KM. 2017. Washington University in St.

Ruff KM, Holehouse AS, Richardson MGO, Pappu RV. 2016. Proteomic and Biophysical Analysis of Polar Tracts. Biophys J 110:556a.

We thank Reviewer #1 for this helpful suggestion and now address this issue by means of a different approach described below.

Based on a previous study (43; Palo Mier et al. 2020), we applied seven different thresholds to seek both short and long, as well as pure and impure, polyX strings in 20 different representative near-complete proteomes, including 4X (4/4), 5X (4/5-5/5), 6X (4/6-6/6), 7X (4/7-7/7), 8-10X (≥50%X), 11-10X (≥50%X) and ≥21X (≥50%X).

To normalize the runs of amino acids and create a null expectation from each proteome, we determined the ratios of the overall number of X residues for each of the seven polyX motifs relative to those in the entire proteome of each species, respectively. The results of four different polyX motifs are shown below, i.e., polyQ (Author response image 1), polyN (Author response image 2), polyS (Author response image 3) and polyT (Author response image 4).

Author response image 1.

Q contents in 7 different types of polyQ motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.  

Author response image 2.

N contents in 7 different types of polyN motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

Author response image 3.

S contents in 7 different types of polyS motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.  

Author response image 4.

T contents in 7 different types of polyT motifs in 20 near-complete proteomes. The five ciliates with reassigned stops codon (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

The results summarized in these four new figures support that polyX prevalence differs among species and that the overall X contents of polyX motifs often but not always correlate with the X usage frequency in entire proteomes (43; Palo Mier et al. 2020).

Most importantly, our results reveal that, compared to Stentor coeruleus or several non-ciliate eukaryotic organisms (e.g., Plasmodium falciparum, Caenorhabditis elegans, Danio rerio, Mus musculus and Homo sapiens), the five ciliates with reassigned TAAQ and TAGQ codons not only have higher Q usage frequencies, but also more polyQ motifs in their proteomes (Figure 1). In contrast, polyQ motifs prevail in Candida albicans, Candida tropicalis, Dictyostelium discoideum, Chlamydomonas reinhardtii, Drosophila melanogaster and Aedes aegypti, though the Q usage frequencies in their entire proteomes are not significantly higher than those of other eukaryotes (Figure 1). Due to their higher N usage frequencies, Dictyostelium discoideum, Plasmodium falciparum and Pseudocohnilembus persalinus have more polyN motifs than the other 23 eukaryotes we examined here (Figure 2). Generally speaking, all 26 eukaryotes we assessed have similar S usage frequencies and percentages of S contents in polyS motifs (Figure 3). Among these 26 eukaryotes, Dictyostelium discoideum possesses many more polyT motifs, though its T usage frequency is similar to that of the other 25 eukaryotes (Figure 4).

In conclusion, these new normalized results confirm that the reassignment of stop codons to Q indeed results in both higher Q usage frequencies and more polyQ motifs in ciliates.  

Reviewer #2 (Public Review):

Summary:

This study seeks to understand the connection between protein sequence and function in disordered regions enriched in polar amino acids (specifically Q, N, S and T). While the authors suggest that specific motifs facilitate protein-enhancing activities, their findings are correlative, and the evidence is incomplete. Similarly, the authors propose that the re-assignment of stop codons to glutamine-encoding codons underlies the greater user of glutamine in a subset of ciliates, but again, the conclusions here are, at best, correlative. The authors perform extensive bioinformatic analysis, with detailed (albeit somewhat ad hoc) discussion on a number of proteins. Overall, the results presented here are interesting, but are unable to exclude competing hypotheses.

Strengths:

Following up on previous work, the authors wish to uncover a mechanism associated with poly-Q and SCD motifs explaining proposed protein expression-enhancing activities. They note that these motifs often occur IDRs and hypothesize that structural plasticity could be capitalized upon as a mechanism of diversification in evolution. To investigate this further, they employ bioinformatics to investigate the sequence features of proteomes of 27 eukaryotes. They deepen their sequence space exploration uncovering sub-phylum-specific features associated with species in which a stop-codon substitution has occurred. The authors propose this stop-codon substitution underlies an expansion of ploy-Q repeats and increased glutamine distribution.

Weaknesses:

The preprint provides extensive, detailed, and entirely unnecessary background information throughout, hampering reading and making it difficult to understand the ideas being proposed. The introduction provides a large amount of detailed background that appears entirely irrelevant for the paper. Many places detailed discussions on specific proteins that are likely of interest to the authors occur, yet without context, this does not enhance the paper for the reader.

The paper uses many unnecessary, new, or redefined acronyms which makes reading difficult. As examples:

(1) Prion forming domains (PFDs). Do the authors mean prion-like domains (PLDs), an established term with an empirical definition from the PLAAC algorithm? If yes, they should say this. If not, they must define what a prion-forming domain is formally.

The N-terminal domain (1-123 amino acids) of S. cerevisiae Sup35 was already referred to as a “prion forming domain (PFD)” in 2006 (Tuite, M. F. 2006). Since then, PFD has also been employed as an acronym in other yeast prion papers (Cox, B.S. et al. 2007; Toombs, T. et al. 2011).

M. F., Tuite, Yeast prions and their prion forming domain. Cell 27, 397-407 (2005).

B. S. Cox, L. Byrne, M. F., Tuite, Protein Stability. Prion 1, 170-178 (2007).

J. A. Toombs, N. M. Liss, K. R. Cobble, Z. Ben-Musa, E. D. Ross, [PSI+] maintenance is dependent on the composition, not primary sequence, of the oligopeptide repeat domain. PLoS One 6, e21953 (2011).

(2) SCD is already an acronym in the IDP field (meaning sequence charge decoration) - the authors should avoid this as their chosen acronym for Serine(S) / threonine (T)-glutamine (Q) cluster domains. Moreover, do we really need another acronym here (we do not).

SCD was first used in 2005 as an acronym for the Serine (S)/threonine (T)-glutamine (Q) cluster domain in the DNA damage checkpoint field (Traven, A. and Heierhorst, J. 2005). Almost a decade later, SCD became an acronym for “sequence charge decoration” (Sawle, L. et al. 2015; Firman, T. et al. 2018).

A. Traven and J, Heierhorst, SQ/TQ cluster domains: concentrated ATM/ATR kinase phosphorylation site regions in DNA-damage-response proteins. Bioessays. 27, 397-407 (2005).

L. Sawle and K, Ghosh, A theoretical method to compute sequence dependent configurational properties in charged polymers and proteins. J. Chem Phys. 143, 085101(2015).

T. Firman and Ghosh, K. Sequence charge decoration dictates coil-globule transition in intrinsically disordered proteins. J. Chem Phys. 148, 123305 (2018).

(3) Protein expression-enhancing (PEE) - just say expression-enhancing, there is no need for an acronym here.

Thank you. Since we have shown that addition of Q-rich motifs to LacZ affects protein expression rather than transcription, we think it is better to use the “PEE” acronym.

The results suggest autonomous protein expression-enhancing activities of regions of multiple proteins containing Q-rich and SCD motifs. Their definition of expression-enhancing activities is vague and the evidence they provide to support the claim is weak. While their previous work may support their claim with more evidence, it should be explained in more detail. The assay they choose is a fusion reporter measuring beta-galactosidase activity and tracking expression levels. Given the presented data they have shown that they can drive the expression of their reporters and that beta gal remains active, in addition to the increase in expression of fusion reporter during the stress response. They have not detailed what their control and mock treatment is, which makes complete understanding of their experimental approach difficult. Furthermore, their nuclear localization signal on the tag could be influencing the degradation kinetics or sequestering the reporter, leading to its accumulation and the appearance of enhanced expression. Their evidence refuting ubiquitin-mediated degradation does not have a convincing control.

Based on the experimental results, the authors then go on to perform bioinformatic analysis of SCD proteins and polyX proteins. Unfortunately, there is no clear hypothesis for what is being tested; there is a vague sense of investigating polyX/SCD regions, but I did not find the connection between the first and section compelling (especially given polar-rich regions have been shown to engage in many different functions). As such, this bioinformatic analysis largely presents as many lists of percentages without any meaningful interpretation. The bioinformatics analysis lacks any kind of rigorous statistical tests, making it difficult to evaluate the conclusions drawn. The methods section is severely lacking. Specifically, many of the methods require the reader to read many other papers. While referencing prior work is of course, important, the authors should ensure the methods in this paper provide the details needed to allow a reader to evaluate the work being presented. As it stands, this is not the case.

Thank you. As described in detail below, we have now performed rigorous statistical testing using the GofuncR package.

Overall, my major concern with this work is that the authors make two central claims in this paper (as per the Discussion). The authors claim that Q-rich motifs enhance protein expression. The implication here is that Q-rich motif IDRs are special, but this is not tested. As such, they cannot exclude the competing hypothesis ("N-terminal disordered regions enhance expression").

In fact, “N-terminal disordered regions enhance expression” exactly summarizes our hypothesis.

On pages 12-13 and Figure 4 of our preprint manuscript, we explained our hypothesis in the paragraph entitled “The relationship between PEE function, amino acid contents, and structural flexibility”.

The authors also do not explore the possibility that this effect is in part/entirely driven by mRNA-level effects (see Verma Na Comms 2019).

As pointed out by the first reviewer, we show evidence that the increase in protein abundance and enzymatic activity is not due to changes in plasmid copy number or mRNA abundance (Figure 2), and that this phenomenon is not affected by translational quality control mutants (Figure 3).

As such, while these observations are interesting, they feel preliminary and, in my opinion, cannot be used to draw hard conclusions on how N-terminal IDR sequence features influence protein expression. This does not mean the authors are necessarily wrong, but from the data presented here, I do not believe strong conclusions can be drawn. That re-assignment of stop codons to Q increases proteome-wide Q usage. I was unable to understand what result led the authors to this conclusion.

My reading of the results is that a subset of ciliates has re-assigned UAA and UAG from the stop codon to Q. Those ciliates have more polyQ-containing proteins. However, they also have more polyN-containing proteins and proteins enriched in S/T-Q clusters. Surely if this were a stop-codon-dependent effect, we'd ONLY see an enhancement in Q-richness, not a corresponding enhancement in all polar-rich IDR frequencies? It seems the better working hypothesis is that free-floating climate proteomes are enriched in polar amino acids compared to sessile ciliates.

Thank you. These comments are not supported by the results in Figure 1.

Regardless, the absence of any kind of statistical analysis makes it hard to draw strong conclusions here.

We apologize for not explaining more clearly the results of Tables 5-7 in our preprint manuscript.

To address the concerns about our GO enrichment analysis by both reviewers, we have now performed rigorous statistical testing for SCD and polyQ protein overrepresentation using the GOfuncR package (https://bioconductor.org/packages/release/bioc/html/GOfuncR.html). GOfuncR is an R package program that conducts standard candidate vs. background enrichment analysis by means of the hypergeometric test. We then adjusted the raw p-values according to the Family-wise error rate (FWER). The same method had been applied to GO enrichment analysis of human genomes (Huttenhower, C., et al. 2009).

Curtis Huttenhower, C., Haley, E. M., Hibbs, M., A., Dumeaux, V., Barrett, D. R., Hilary A. Coller, H. A., and Olga G. Troyanskaya, O., G. Exploring the human genome with functional maps, Genome Research 19, 1093-1106 (2009).

The results presented in Author response image 5 and Author response image 6 support our hypothesis that Q-rich motifs prevail in proteins involved in specialized biological processes, including Saccharomyces cerevisiae RNA-mediated transposition, Candida albicans filamentous growth, peptidyl-glutamic acid modification in ciliates with reassigned stop codons (TAAQ and TAGQ), Tetrahymena thermophila xylan catabolism, Dictyostelium discoideum sexual reproduction, Plasmodium falciparum infection, as well as the nervous systems of Drosophila melanogaster, Mus musculus, and Homo sapiens (74). In contrast, peptidyl-glutamic acid modification and microtubule-based movement are not overrepresented with Q-rich proteins in Stentor coeruleus, a ciliate with standard stop codons.

1. Cara L, Baitemirova M, Follis J, Larios-Sanz M, Ribes-Zamora A. The ATM- and ATR-related SCD domain is over-represented in proteins involved in nervous system development. Sci Rep. 2016;6:19050.

Author response image 5.

Selection of biological processes with overrepresented SCD-containing proteins in different eukaryotes. The percentages and number of SCD-containing proteins in our search that belong to each indicated Gene Ontology (GO) group are shown. GOfuncR (Huttenhower, C., et al. 2009) was applied for GO enrichment and statistical analysis. The p values adjusted according to the Family-wise error rate (FWER) are shown. The five ciliates with reassigned stop codons (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

Author response image 6.

Selection of biological processes with overrepresented polyQ-containing proteins in different eukaryotes. The percentages and numbers of polyQ-containing proteins in our search that belong to each indicated Gene Ontology (GO) group are shown. GOfuncR (Huttenhower, C., et al. 2009) was applied for GO enrichment and statistical analysis. The p values adjusted according to the Family-wise error rate (FWER) are shown. The five ciliates with reassigned stops codons (TAAQ and TAGQ) are indicated in red. Stentor coeruleus, a ciliate with standard stop codons, is indicated in green.

Author Response

Reviewer #1 (Public Review):

Wang and all present an interesting body of work focused on the effects of high altitude and hypoxia on erythropoiesis, resulting in erythrocytosis. This work is specifically focused on the spleen, identifying splenic macrophages as central cells in this effect. This is logical since these cells are involved in erythrophagocytosis and iron recycling. The results suggest that hypoxia induces splenomegaly with decreased number of splenic macrophages. There is also evidence that ferroptosis is induced in these macrophages, leading to cell destruction. Finally, the data suggest that ferroptosis in splenic red pulp macrophages causes the decrease in RBC clearance, resulting in erythrocytosis aka lengthening the RBC lifespan. However, there are many issues with the presented results, with somewhat superficial data, meaning the conclusions are overstated and there is decreased confidence that the hypotheses and observed results are directly causally related to hypoxia.

Major points:

1) The spleen is a relatively poorly understood organ but what is known about its role in erythropoiesis especially in mice is that it functions both to clear as well as to generate RBCs. The later process is termed extramedullary hematopoiesis and can occur in other bones beyond the pelvis, liver, and spleen. In mice, the spleen is the main organ of extramedullary erythropoiesis. The finding of transiently decreased spleen size prior to splenomegaly under hypoxic conditions is interesting but not well developed in the manuscript. This is a shortcoming as this is an opportunity to evaluate the immediate effect of hypoxia separately from its more chronic effect. Based just on spleen size, no conclusions can be drawn about what happens in the spleen in response to hypoxia.

Thank you for your insightful comments and questions. The spleen is instrumental in both immune response and the clearance of erythrocytes, as well as serving as a significant reservoir of blood in the body. This organ, characterized by its high perfusion rate and pliability, constricts under conditions of intense stress, such as during peak physical exertion, the diving reflex, or protracted periods of apnea. This contraction can trigger an immediate release of red blood cells (RBCs) into the bloodstream in instances of substantial blood loss or significant reduction of RBCs. Moreover, elevated oxygen consumption rates in certain animal species can be partially attributed to splenic contractions, which augment hematocrit levels and the overall volume of circulating blood, thereby enhancing venous return and oxygen delivery (Dane et al. J Appl Physiol, 2006, 101:289-97; Longhurst et al. Am J Physiol, 1986, 251: H502-9). In our investigation, we noted a significant contraction of the spleen following exposure to hypoxia for a period of one day. We hypothesized that the body, under such conditions, is incapable of generating sufficient RBCs promptly enough to facilitate enhanced oxygen delivery. Consequently, the spleen reacts by releasing its stored RBCs through splenic constriction, leading to a measurable reduction in spleen size.

However, we agree with you that further investigation is required to fully understand the implications of these changes. Considering the comments, we propose to extend our research by incorporating more detailed examinations of spleen morphology and function during hypoxia, including the potential impact on extramedullary hematopoiesis. We anticipate that such an expanded analysis would not only help elucidate the initial response to hypoxia but also provide insights into the more chronic effects of this condition on spleen function and erythropoiesis.

2) Monocyte repopulation of tissue resident macrophages is a minor component of the process being described and it is surprising that monocytes in the bone marrow and spleen are also decreased. Can the authors conjecture why this is happening? Typically, the expectation would be that a decrease in tissue resident macrophages would be accompanied by an increase in monocyte migration into the organ in a compensatory manner.

We appreciate your insightful query regarding the observed decrease in monocytes in the bone marrow and spleen, particularly considering the typical compensatory increase in monocyte migration into organs following a decrease in tissue resident macrophages.

The observed decrease in monocytes within the bone marrow is likely attributable to the fact that monocytes and precursor cells for red blood cells (RBCs) both originate from the same hematopoietic stem cells within the bone marrow. It is well established that exposure to hypobaric hypoxia (HH) induces erythroid differentiation specifically within the bone marrow, originating from these hematopoietic stem cells. As such, we postulate that the differentiation into monocytes is reduced under hypoxic conditions, which may subsequently cause a decrease in migration to the spleen.

Furthermore, we hypothesize that an increased migration of monocytes to other tissues under HH exposure may also contribute to the decreased migration to the spleen. The liver, which partially contributes to the clearance of RBCs, may play a role in this process. Our investigations to date have indeed identified an increased monocyte migration to the liver. We were pleased to discover an elevation in CSF1 expression in the liver following HH exposure for both 7 and 14 days. This finding was corroborated through flow cytometry, which confirmed an increase in monocyte migration to the liver.

Consequently, we propose that under HH conditions, the liver requires an increased influx of monocytes, which in turn leads to a decrease in monocyte migration to the spleen. However, it is important to note that these findings will be discussed more comprehensively in our forthcoming publication, and as such, the data pertaining to these results have not been included in the current manuscript.

3) Figure 3 does not definitively provide evidence that cell death is specifically occurring in splenic macrophages and the fraction of Cd11b+ cells is not changed in NN vs HH. Furthermore, the IHC of F4/80 in Fig 3U is not definitive as cells can express F4/80 more or less brightly and no negative/positive controls are shown for this panel.

We appreciate your insightful comments and critiques regarding Figure 3. We acknowledge that the figure, as presented, does not definitively demonstrate that cell death is specifically occurring in splenic macrophages. While it is challenging to definitively determine the occurrence of cell death in macrophages based solely on Figure 3D-F, our single-cell analysis provides strong evidence that such an event occurs. We initially observed cell death within the spleen under hypobaric hypoxia (HH) conditions, and to discern the precise cell type involved, we conducted single-cell analyses. Regrettably, we did not articulate this clearly in our preliminary manuscript. In the revised version, we have modified the sequence of Figure 3A-C and Figure 3D-F for better clarity. Besides, we observed a significant decrease in the fraction of F4/80hiCD11bhi macrophages under HH conditions compared to NN. To make the changes more evident in CD86 and CD206, we have transformed these scatter plots into histograms in our revised manuscript.

Considering the limitations of F4/80 as a conclusive macrophage identifier, we have concurrently presented the immunohistochemical (IHC) analyses of heme oxygenase-1 (HO-1). Functioning as a macrophage marker, particularly in cells involved in iron metabolism, HO-1 offers additional diagnostic accuracy. Observations from both F4/80 and HO-1 staining suggested a primary localization of positively stained cells within the splenic red pulp. Following exposure to hypoxia-hyperoxia (HH) conditions, a decrease was noted in the expression of both F4/80 and HO-1. This decrease implies that HH conditions contribute to a reduction in macrophage population and impede the iron metabolism process. In the revised version of our manuscript, we have enhanced the clarity of Figure 3U to illustrate the presence of positive staining, with an emphasis on HO-1 staining, which is predominantly observed in the red pulp.

4) The phagocytic function of splenic red pulp macrophages relative to infection cannot be used directly to understand erythrophagocytosis. The standard approach is to use opsonized RBCs in vitro. Furthermore, RBC survival is a standard method to assess erythrophagocytosis function. In this method, biotin is injected via tail vein directly and small blood samples are collected to measure the clearance of biotinilation by flow; kits are available to accomplish this. Because the method is standard, Fig 4D is not necessary and Fig 4E needs to be performed only in blood by sampling mice repeatedly and comparing the rate of biotin decline in HH with NN (not comparing 7 d with 14 d).

We appreciate your insightful comments and suggestions. We concur that the phagocytic function of splenic red pulp macrophages in the context of infection may not be directly translatable to understanding erythrophagocytosis. Given our assessment that the use of cy5.5-labeled E.coli alone may not be sufficient to accurately evaluate the phagocytic function of macrophages, we extended our study to include the use of NHS-biotin-labeled RBCs to assess phagocytic capabilities. While the presence of biotin-labeled RBCs in the blood could provide an indication of RBC clearance, this measure does not exclusively reflect the spleen's role in the process, as it fails to account for the clearance activities of other organs.

Consequently, we propose that the remaining biotin-labeled RBCs in the spleen may provide a more direct representation of the organ's function in RBC clearance and sequestration. Our observations of diminished erythrophagocytosis at both 7 and 14 days following exposure to HH guided our subsequent efforts to quantify biotin-labeled RBCs in both the circulatory system and spleen. These measurements were conducted during the 7 to 14-day span following the confirmation of impaired erythrophagocytosis. Comparative evaluation of RBC clearance rates under NN and HH conditions provided further evidence supporting our preliminary observations, with the data revealing a decrease in the RBC clearance rate in the context of HH conditions. In response to feedback from other reviewers, we have elected to exclude the phagocytic results and the diagram of the erythrocyte labeling assay. These amendments will be incorporated into the revised manuscript. The reviewers' constructive feedback has played a crucial role in refining the methodological precision and coherence of our investigation.

5) It is unclear whether Tuftsin has a specific effect on phagocytosis of RBCs without other potential confounding effects. Furthermore, quantifying iron in red pulp splenic macrophages requires alternative readily available more quantitative methods (e.g. sorted red pulp macrophages non-heme iron concentration).

We appreciate your comments and questions regarding the potential effect of Tuftsin on the phagocytosis of RBCs and the quantification of iron in red pulp splenic macrophages. Regarding the role of Tuftsin, we concur that the literature directly associating Tuftsin with erythrophagocytosis is scant. The work of Gino Roberto Corazza et al. does suggest a link between Tuftsin and general phagocytic capacity, but it does not specifically address erythrophagocytosis (Am J Gastroenterol, 1999;94:391-397). We agree that further investigations are required to elucidate the potential confounding effects and to ascertain whether Tuftsin has a specific impact on the phagocytosis of RBCs. Concerning the quantification of iron in red pulp splenic macrophages, we acknowledge your suggestion to employ readily available and more quantitative methods. We have incorporated additional Fe2+ staining in the spleen at two time points: 7 and 14 days subsequent to HH exposure (refer to the following Figure). The resultant data reveal an escalated deposition of Fe2+ within the red pulp, as evidenced in Figures 5 (panels L and M) and Figure 7 (panels L and M).

6) In Fig 5, PBMCs are not thought to represent splenic macrophages and although of some interest, does not contribute significantly to the conclusions regarding splenic macrophages at the heart of the current work. The data is also in the wrong direction, namely providing evidence that PBMCs are relatively iron poor which is not consistent with ferroptosis which would increase cellular iron.

We appreciate your insightful critique regarding Figure 5 and the interpretation of our data on peripheral blood mononuclear cells (PBMCs) in relation to splenic macrophages. We understand that PBMCs do not directly represent splenic macrophages, and we agree that any conclusions drawn from PBMCs must be considered with caution when discussing the behavior of splenic macrophages.

The primary rationale for incorporating PBMCs into our study was to investigate the potential correspondence between their gene expression changes and those observed in the spleen after HH exposure. This was posited as a working hypothesis for further exploration rather than a conclusive statement. The gene expression in PBMCs was congruous with changes in the spleen's gene expression, demonstrating an iron deficiency phenotype, ostensibly due to the mobilization of intracellular iron for hemoglobin synthesis. Thus, it is plausible that NCOA4 may facilitate iron mobilization through the degradation of ferritin to store iron.

It remains ambiguous whether ferroptosis was initiated in the PBMCs during our study. Ferroptosis primarily occurs as a response to an increase in Fe2+ rather than an overall increase in intracellular iron. Our preliminary proposition was that relative changes in gene expression in PBMCs could potentially mirror corresponding changes in protein expression in the spleen, thereby potentially indicating alterations in iron processing capacity post-HH exposure. However, we fully acknowledge that this is a conjecture requiring further empirical substantiation or clinical validation.

7) Tfr1 increase is typically correlated with cellular iron deficiency while ferroptosis consistent with iron loading. The direction of the changes in multiple elements relevant to iron trafficking is somewhat confusing and without additional evidence, there is little confidence that the authors have reached the correct conclusion. Furthermore, the results here are analyses of total spleen samples rather than specific cells in the spleen.

We appreciate your astute comments and agree that the observed increase in transferrin receptor (TfR) expression, typically associated with cellular iron deficiency, appears contradictory to the expected iron-loading state associated with ferroptosis. We understand that this apparent contradiction might engender some uncertainty about our conclusions.

In our investigation, we evaluated total spleen samples as opposed to distinct cell types within the spleen, a factor that could have contributed to the seemingly discordant findings. An integral element to bear in mind is the existence of immature RBCs in the spleen, particularly within the hematopoietic island where these immature RBCs cluster around nurse macrophages. These immature RBCs contain abundant TfR which was needed for iron uptake and hemoglobin synthesis. These cells, which prove challenging to eliminate via perfusion, might have played a role in the observed upregulation in TfR expression, especially in the aftermath of HH exposure. Our further research revealed that the expression of TfR in macrophages diminished following hypoxic conditions, thereby suggesting that the elevated TfR expression in tissue samples may predominantly originate from other cell types, especially immature RBCs (refer to subsequent Figure).

Reviewer #2 (Public Review):

The authors aimed at elucidating the development of high altitude polycythemia which affects mice and men staying in the hypoxic atmosphere at high altitude (hypobaric hypoxia; HH). HH causes increased erythropoietin production which stimulates the production of red blood cells. The authors hypothesize that increased production is only partially responsible for exaggerated red blood cell production, i.e. polycythemia, but that decreased erythrophagocytosis in the spleen contributes to high red blood cells counts.

The main strength of the study is the use of a mouse model exposed to HH in a hypobaric chamber. However, not all of the reported results are convincing due to some smaller effects which one may doubt to result in the overall increase in red blood cells as claimed by the authors. Moreover, direct proof for reduced erythrophagocytosis is compromised due to a strong spontaneous loss of labelled red blood cells, although effects of labelled E. coli phagocytosis are shown. Their discussion addresses some of the unexpected results, such as the reduced expression of HO-1 under hypoxia but due to the above-mentioned limitations much of the discussion remains hypothetical.

Thank you for your valuable feedback and insight. We appreciate the recognition of the strength of our study model, the exposure of mice to hypobaric hypoxia (HH) in a hypobaric animal chamber. We also understand your concerns about the smaller effects and their potential impact on the overall increase in red blood cells (RBCs), as well as the apparent reduced erythrophagocytosis due to the loss of labelled RBCs.

Erythropoiesis has been predominantly attributed to the amplified production of RBCs under conditions of HH. The focus of our research was to underscore the potential acceleration of hypoxia-associated polycythemia (HAPC) as a result of compromised erythrophagocytosis. Considering the spontaneous loss of labelled RBCs in vivo, we assessed the clearance rate of RBCs at the stages of 7 and 14 days within the HH environment, and subsequently compared this rate within the period from 7 to 14 days following the clear manifestation of erythrophagocytosis impairment at the two aforementioned points identified in our study. This approach was designed to negate the effects of spontaneous loss of labelled RBCs in both NN and HH conditions. Correspondingly, the results derived from blood and spleen analyses corroborated a decline in the RBC clearance rate under HH when juxtaposed with NN conditions.

Apart from the E. coli phagocytosis and the labeled RBCs experiment (this part of the results was removed in the revision), the injection of Tuftsin further substantiated the impairment of erythrophagocytosis in the HH spleen, as evidenced by the observed decrease in iron within the red pulp of the spleen post-perfusion. Furthermore, to validate our findings, we incorporated RBCs staining in splenic cells at 7 and 14 days of HH exposure, which provided concrete confirmation of impaired erythrophagocytosis (new Figure 4E).

As for the reduced expression of heme oxygenase-1 (HO-1) under hypoxia, we agree that this was an unexpected result, and we are in the process of further exploring the underlying mechanisms. It is possible that there are other regulatory pathways at play that are yet to be identified. However, we believe that by offering possible interpretations of our data and potential directions for future research, we contribute to the ongoing scientific discourse in this area.

Reviewer #3 (Public Review):

The manuscript by Yang et al. investigated in mice how hypobaric hypoxia can modify the RBC clearance function of the spleen, a concept that is of interest. Via interpretation of their data, the authors proposed a model that hypoxia causes an increase in cellular iron levels, possibly in RPMs, leading to ferroptosis, and downregulates their erythrophagocytic capacity. However, most of the data is generated on total splenocytes/total spleen, and the conclusions are not always supported by the presented data. The model of the authors could be questioned by the paper by Youssef et al. (which the authors cite, but in an unclear context) that the ferroptosis in RPMs could be mediated by augmented erythrophagocytosis. As such, the loss of RPMs in vivo which is indeed clear in the histological section shown (and is a strong and interesting finding) can be not directly caused by hypoxia, but by enhanced RBC clearance. Such a possibility should be taken into account.

Thank you for your insightful comments and constructive feedback. In their research, Youssef et al. (2018) discerned that elevated erythrophagocytosis of stressed red blood cells (RBCs) instigates ferroptosis in red pulp macrophages (RPMs) within the spleen, as evidenced in a mouse model of transfusion. This augmentation of erythrophagocytosis was conspicuous five hours post-injection of RBCs. Conversely, our study elucidated the decrease in erythrophagocytosis in the spleen after both 7 and 14 days.

Typically, macrophages exhibit an enhanced phagocytic capacity in the immediate aftermath of stress or stimulation. Nonetheless, the temporal points of observation in our study were considerably extended (seven and fourteen days). It remains uncertain whether phagocytic capability was amplified during the acute phase of HH exposure—particularly within the first day, considering that splenoconstriction under HH for one day results in the release of stored RBCs into the bloodstream—and whether this initial response could precipitate ferroptosis and subsequently diminished erythrophagocytosis at the 7 or 14 day marks under continued HH conditions.

Major points:

1) The authors present data from total splenocytes and then relate the obtained data to RPMs, which are quantitatively a minor population in the spleen. Eg, labile iron is increased in the splenocytes upon HH, but the manuscript does not show that this occurs in the red pulp or RPMs. They also measure gene/protein expression changes in the total spleen and connect them to changes in macrophages, as indicated in the model Figure (Fig. 7). HO-1 and levels of Ferritin (L and H) can be attributed to the drop in RPMs in the spleen. Are any of these changes preserved cell-intrinsically in cultured macrophages? This should be shown to support the model (relates also to lines 487-88, where the authors again speculate that hypoxia decreases HO-1 which was not demonstrated). In the current stage, for example, we do not know if the labile iron increase in cultured cells and in the spleen in vivo upon hypoxia is the same phenomenon, and why labile iron is increased. To improve the manuscript, the authors should study specifically RPMs.

We express our gratitude for your perceptive remarks. In our initial manuscript, we did not evaluate labile iron within the red pulp and red pulp macrophages (RPMs). To address this oversight, we utilized the Lillie staining method, in accordance with the protocol outlined by Liu et al., (Chemosphere, 2021, 264(Pt 1):128413), to discern Fe2+ presence within these regions. The outcomes were consistent with our antecedent Western blot and flow cytometry findings in the spleen, corroborating an increment in labile iron specifically within the red pulp of the spleen.

However, we acknowledge the necessity for other supplementary experimental efforts to further validate these findings. Additionally, we scrutinized the expression of heme oxygenase-1 (HO-1) and iron-related proteins, including transferrin receptor (TfR), ferroportin (Fpn), ferritin (Ft), and nuclear receptor coactivator 4 (NCOA4) in primary macrophages subjected to 1% hypoxic conditions, both with and without hemoglobin treatment. Our results indicated that the expression of ferroptosis-related proteins was consistent with in vivo studies, however the expression of iron related proteins was not similar in vitro and in vivo. It suggesting that the increase in labile iron in cultured cells and the spleen in vivo upon hypoxia are not identical phenomena. However, the precise mechanism remains elusive.

In our study, we observed a decrease in HO-1 protein expression following 7 and 14 days of HH exposure, as shown in Figure 3U, 5A, and S1A. This finding contradicts previous research that identified HO-1 as a hypoxia-inducible factor (HIF) target under hypoxic conditions (P J Lee et al., 1997). Our discussion, therefore, addressed the potential discrepancy in HO-1 expression under HH. According to our findings, HO-1 regulation under HH appears to be predominantly influenced by macrophage numbers and the RBCs to be processed in the spleen or macrophages, rather than by hypoxia alone.

It is challenging to discern whether the increased labile iron observed in vitro accurately reflects the in vivo phenomenon, as replicating the iron requirements for RBCs production induced by HH in vitro is inherently difficult. However, by integrating our in vivo and in vitro studies, we determined that the elevated Fe2+ levels were not dependent on HO-1 protein expression, as HO-1 levels was increased in vitro while decreasing in vivo under hypoxic/HH exposure.

2) The paper uses flow cytometry, but how this method was applied is suboptimal: there are no gating strategies, no indication if single events were determined, and how cell viability was assessed, which are the parent populations when % of cells is shown on the graphs. How RBCs in the spleen could be analyzed without dedicated cell surface markers? A drop in splenic RPMs is presented as the key finding of the manuscript but Fig. 3M shows gating (suboptimal) for monocytes, not RPMs. RPMs are typically F4/80-high, CD11-low (again no gating strategy is shown for RPMs). Also, the authors used single-cell RNAseq to detect a drop in splenic macrophages upon HH, but they do not indicate in Fig. A-C which cluster of cells relates to macrophages. Cell clusters are not identified in these panels, hence the data is not interpretable).

Thank you for your comments and constructive critique regarding our flow cytometry methodology and presentation. We understand the need for greater transparency and detailed explanation of our procedures, and we acknowledge that the lack of gating strategies and other pertinent information in our initial manuscript may have affected the clarity of our findings.

In our initial report, we provided an overview of the decline in migrated macrophages (F4/80hiCD11bhi), including both M1 and M2 expression in migrated macrophages, as illustrated in Figure 3, but did not specifically address the changes in red pulp macrophages (RPMs). Based on previous results, it is difficult to identify CD11b- and CD11blo cells. We will repeat the results and attempt to identify F4/80hiCD11blo cells in the revised manuscript. The results of the reanalysis are now included (Figure 3M). However, single-cell in vivo analysis studies may more accurately identify specific cell types that decrease after exposure to HH.

Furthermore, we substantiated the reduction in red pulp, as evidenced by Figure 4J, given that iron processing primarily occurs within the red pulp. In Figure 3, our initial objective was merely to illustrate the reduction in total macrophages in the spleen following HH exposure.

To further clarify the characterization of various cell types, we conducted a single-cell analysis. Our findings indicated that clusters 0,1,3,4,14,18, and 29 represented B cells, clusters 2, 10, 12, and 28 represented T cells, clusters 15 and 22 corresponded to NK cells, clusters 5, 11, 13, and 19 represented NKT cells, clusters 6, 9, and 24 represented cell cycle cells, clusters 26 and 17 represented plasma cells, clusters 21 and 23 represented neutrophils, cluster 30 represented erythrocytes, and clusters 7, 8, 16, 20, 24, and 27 represented dendritic cells (DCs) and macrophages, as depicted in Figure 3E.

3) The authors draw conclusions that are not supported by the data, some examples: a) they cannot exclude eg the compensatory involvement of the liver in the RBCs clearance (the differences between HH sham and HH splenectomy is mild in Fig. 2 E, F and G).

Thank you for your insightful comments and for pointing out the potential involvement of other organs, such as the liver, in the RBC clearance under HH conditions. We concur with your observation that the differences between the HH sham and HH splenectomy conditions in Fig. 2 E, F, and G are modest. This could indeed suggest a compensatory role of other organs in RBC clearance when splenectomy is performed. Our intent, however, was to underscore the primary role of the spleen in this process under HH exposure.

In fact, after our initial investigations, we conducted a more extensive study examining the role of the liver in RBC clearance under HH conditions. Our findings, as illustrated in the figures submitted with this response, indeed support a compensatory role for the liver. Specifically, we observed an increase in macrophage numbers and phagocytic activity in the liver under HH conditions. Although the differences in RBC count between the HH sham and HH splenectomy conditions may seem minor, it is essential to consider the unit of this measurement, which is value*1012/ml. Even a small numerical difference can represent a significant biological variation at this scale.

b) splenomegaly is typically caused by increased extramedullary erythropoiesis, not RBC retention. Why do the authors support the second possibility? Related to this, why do the authors conclude that data in Fig. 4 G,H support the model of RBC retention? A significant drop in splenic RBCs (poorly gated) was observed at 7 days, between NN and HH groups, which could actually indicate increased RBC clearance capacity = less retention.

Prior investigations have predominantly suggested that spleen enlargement under hypoxic conditions stems from the spleen's extramedullary hematopoiesis. Nevertheless, an intriguing study conducted in 1994 by the General Hospital of Xizang Military Region reported substantial exaggeration and congestion of splenic sinuses in high altitude polycythemia (HAPC) patients. This finding was based on the dissection of spleens from 12 patients with HAPC (Zou Xunda, et al., Southwest Defense Medicine, 1994;5:294-296). Moreover, a recent study indicated that extramedullary erythropoiesis reaches its zenith between 3 to 7 days (Wang H et al., 2021).

Considering these findings, the present study postulates that hypoxia-induced inhibition of erythrophagocytosis may lead to RBC retention. However, we acknowledge that the manuscript in its current preprint form does not offer conclusive evidence to substantiate this hypothesis. To bridge this gap, we further conducted experiments where the spleen was perfused, and total cells were collected post HH exposure. These cells were then smeared onto slides and subjected to Wright staining. Our results unequivocally demonstrate an evident increase in deformation and retention of RBCs in the spleen following 7 and 14 days of HH exposure. This finding strengthens our initial hypothesis and contributes a novel perspective to the understanding of splenic responses under hypoxic conditions.

c) lines 452-54: there is no data for decreased phagocytosis in vivo, especially in the context of erythrophagocytosis. This should be done with stressed RBCs transfusion assays, very good examples, like from Youssef et al. or Threul et al. are available in the literature.

Thanks. In their seminal work, Youssef and colleagues demonstrated that the transfusion of stressed RBCs triggers erythrophagocytosis and subsequently incites ferroptosis in red pulp macrophages (RPMs) within a span of five hours. Given these observations, the applicability of this model to evaluate macrophage phagocytosis in the spleen or RPMs under HH conditions may be limited, as HH has already induced erythropoiesis in vivo. In addition, it was unclear whether the membrane characteristics of stress induced RBCs were similar to those of HH induced RBCs, as this is an important signal for in vivo phagocytosis. The ambiguity arises from the fact that we currently lack sufficient knowledge to discern whether the changes in phagocytosis are instigated by the presence of stressed RBCs or by changes of macrophages induced by HH in vivo. Nonetheless, we appreciate the potential value of this approach and intend to explore its utility in our future investigations. The prospect of distinguishing the effects of stressed RBCs from those of HH on macrophage phagocytosis is an intriguing line of inquiry that could yield significant insights into the mechanisms governing these physiological processes. We will investigate this issue in our further study.

d) Line 475 - ferritinophagy was not shown in response to hypoxia by the manuscript, especially that NCOA4 is decreased, at least in the total spleen.

Drawing on the research published in eLife in 2015, it was unequivocally established that ferritinophagy, facilitated by Nuclear Receptor Coactivator 4 (NCOA4), is indispensable for erythropoiesis. This process is modulated by iron-dependent HECT and RLD domain containing E3 ubiquitin protein ligase 2 (HERC2)-mediated proteolysis (Joseph D Mancias et al., eLife. 2015; 4: e10308). As is widely recognized, NCOA4 plays a critical role in directing ferritin (Ft) to the lysosome, where both NCOA4 and Ft undergo coordinated degradation.

In our study, we provide evidence that exposure to HH stimulates erythropoiesis (Figure 1). We propose that this, in turn, could promote ferritinophagy via NCOA4, resulting in a decrease in NCOA4 protein levels post-HH exposure. We will further increase experiments to verify this concern. This finding not only aligns with the established understanding of ferritinophagy and erythropoiesis but also adds a novel dimension to the understanding of cellular responses to hypoxic conditions.

4) In a few cases, the authors show only representative dot plots or histograms, without quantification for n>1. In Fig. 4B the authors write about a significant decrease (although with n=1 no statistics could be applied here; of note, it is not clear what kind of samples were analyzed here). Another example is Fig. 6I. In this case, it is even more important as the data are conflicting the cited article and the new one: PMCID: PMC9908853 which shows that hypoxia stimulates efferocytosis. Sometimes the manuscript claim that some changes are observed, although they are not visible in representative figures (eg for M1 and M2 macrophages in Fig. 3M)

We recognize that our initial portrayal of Figure 4B was lacking in precision, given that it did not include the corresponding statistical graph. While our results demonstrated a significant reduction in the ability to phagocytose E. coli, in line with the recommendations of other reviewers, we have opted to remove the results pertaining to E. coli phagocytosis in this revision, as they primarily reflected immune function. In relation to PMC9908853, which reported metabolic adaptation facilitating enhanced macrophage efferocytosis in limited-oxygen environments, it is worth noting that the macrophages investigated in this study were derived from ER-Hoxb8 macrophage progenitors following the removal of β-estradiol. Consequently, questions arise regarding the comparability between these cultured macrophages and primary macrophages obtained fresh from the spleen post HH exposure. The characteristics and functions of these two different macrophage sources may not align precisely, and this distinction necessitates further investigation.

5) There are several unclear issues in methodology:

• what is the purity of primary RPMs in the culture? RPMs are quantitatively poorly represented in splenocyte single-cell suspensions. This reviewer is quite skeptical that the processing of splenocytes from approx 1 mm3 of tissue was sufficient to establish primary RPM cultures. The authors should prove that the cultured cells were indeed RPMs, not monocyte-derived macrophages or other splenic macrophage subtypes.

Thank you for your thoughtful comments and inquiries. Firstly, I apologize if we did not make it clear in the original manuscript. The purity of the primary RPMs in our culture was found to be approximately 40%, as identified by F4/80hiCD11blo markers using flow cytometry. We recognize that RPMs are typically underrepresented in splenocyte single-cell suspensions, and the concern you raise about the potential for contamination by other cell types is valid.

We apologize for any ambiguities in the methodological description that may have led to misunderstandings during the review. Indeed, the entirety of the spleen is typically employed for splenic macrophage culture. The size of the spleen can vary dependent on the species and age of the animal, but in mice, it is commonly approximately 1 cm in length. The spleen is then dissected into minuscule fragments, each approximately 1 mm3 in volume, to aid in enzymatic digestion. This procedure does not merely utilize a single 1 mm3 tissue fragment for RPMs cultures. Although the isolation and culture of spleen macrophages can present considerable challenges, our method has been optimized to enhance the yield of this specific cell population.

• (around line 183) In the description of flow cytometry, there are several missing issues. In 1) it is unclear which type of samples were analyzed. In 2) it is not clear how splenocyte cell suspension was prepared.

1) Whole blood was extracted from the mice and collected into an anticoagulant tube, which was then set aside for subsequent thiazole orange (TO) staining. 2) Splenic tissue was procured from the mice and subsequently processed into a single-cell suspension using a 40 μm filter. The erythrocytes within the entire sample were subsequently lysed and eliminated, and the remaining cell suspension was resuspended in phosphate-buffered saline (PBS) in preparation for ensuing analyses.

We have meticulously revised these methodological details in the corresponding section of the manuscript to ensure clarity and precision.

• In line 192: what does it mean: 'This step can be omitted from cell samples'?

The methodology employed for the quantification of intracellular divalent iron content and lipid peroxidation level was executed as follows: Splenic tissue was first processed into a single cell suspension, subsequently followed by the lysis of RBCs. It should be noted that this particular stage is superfluous when dealing with isolated cell samples. Subsequently, a total of 1 × 106 cells were incubated with 100 μL of BioTracker Far-red Labile Fe2+ Dye (1 mM, Sigma, SCT037, USA) for a duration of 1 hour, or alternatively, C11-Bodipy 581/591 (10 μM, Thermo Fisher, D3861, USA) for a span of 30 minutes. Post incubation, cells were thoroughly washed twice with PBS. Flow cytometric analysis was subsequently performed, utilizing the FL6 (638 nm/660 nm) channel for the determination of intracellular divalent iron content, and the FL1 (488 nm/525 nm) channel for the quantification of the lipid peroxidation level.

• 'TO method' is not commonly used anymore and hence it was unclear to this Reviewer. Reticulocytes should be analyzed with proper gating, using cell surface markers.

We are appreciative of your astute observation pertaining to the methodology we employed to analyze reticulocytes in our study. We value your recommendation to utilize cell surface markers for effective gating, which indeed represents a more modern and accurate approach. However, as reticulocyte identification is not the central focus of our investigation, we opted for the TO staining method—due to its simplicity and credibility of results. In our initial exploration, we adopted the TO staining method in accordance with the protocol outlined (Sci Rep, 2018, 8(1):12793), primarily owing to its established use and demonstrated efficacy in reticulocyte identification.

• The description of 'phagocytosis of E. coli and RBCs' in the Methods section is unclear and incomplete. The Results section suggests that for the biotinylated RBCs, phagocytosis? or retention? Of RBCs was quantified in vivo, upon transfusion. However, the Methods section suggests either in vitro/ex vivo approach. It is vague what was indeed performed and how in detail. If RBC transfusion was done, this should be properly described. Of note, biotinylation of RBCs is typically done in vivo only, being a first step in RBC lifespan assay. The such assay is missing in the manuscript. Also, it is not clear if the detection of biotinylated RBCs was performed in permeablized cells (this would be required).

Thanks for the comments. In our initial methodology, we employed Cy5.5-labeled Escherichia coli to probe phagocytic function, albeit with the understanding that this may not constitute the most ideal model for phagocytosis detection within this context (in light of recommendations from other reviewers, we have removed the E. coli phagocytosis results from this revision, as they predominantly mirror immune function). Our fundamental aim was to ascertain whether HH compromises the erythrophagocytic potential of splenic macrophages. In pursuit of this, we subsequently analyzed the clearance of biotinylated RBCs in both the bloodstream and spleen to assess phagocytic functionality in vivo.

In the present study, instead of transfusing biotinylated RBCs into mice, we opted to inject N-Hydroxysuccinimide (NHS)-biotin into the bloodstream. NHS-biotin is capable of binding with cell membranes in vivo and can be recognized by streptavidin-fluorescein isothiocyanate (FITC) after cells are extracted from the blood or spleen in vitro. Consequently, biotin-labeled RBCs were detectable in both the blood and spleen following NHS-biotin injection for a duration of 21 days.

Ultimately, we employed flow cytometry to analyze the NHS-biotin labeled RBCs in the blood or spleen. This method facilitates the detection of live cells and is not applicable to permeabilized cells. We believe this approach better aligns with our investigative goals and offers a more robust evaluation of erythrophagocytic function under hypoxic conditions.

Author response:

Reviewer #1 (Public review):

Summary:

Jocher, Janssen, et al examine the robustness of comparative functional genomics studies in primates that make use of induced pluripotent stem cell-derived cells. Comparative studies in primates, especially amongst the great apes, are generally hindered by the very limited availability of samples, and iPSCs, which can be maintained in the laboratory indefinitely and defined into other cell types, have emerged as promising model systems because they allow the generation of data from tissues and cells that would otherwise be unobservable.

Undirected differentiation of iPSCs into many cell types at once, using a method known as embryoid body differentiation, requires researchers to manually assign all cell types in the dataset so they can be correctly analysed. Typically, this is done using marker genes associated with a specific cell type. These are defined a priori, and have historically tended to be characterised in mice and humans and then employed to annotate other species. Jocher, Janssen, et al ask if the marker genes and features used to define a given cell type in one species are suitable for use in a second species, and then quantify the degree of usefulness of these markers. They find that genes that are informative and cell type specific in a given species are less valuable for cell type identification in other species, and that this value, or transferability, drops off as the evolutionary distance between species increases.

This paper will help guide future comparative studies of gene expression in primates (and more broadly) as well as add to the growing literature on the broader challenges of selecting powerful and reliable marker genes for use in single-cell transcriptomics.

Strengths:

Marker gene selection and cell type annotation is a challenging problem in scRNA studies, and successful classification of cells often requires manual expert input. This can be hard to reproduce across studies, as, despite general agreement on the identity of many cell types, different methods for identifying marker genes will return different sets of genes. The rise of comparative functional genomics complicates this even further, as a robust marker gene in one species need not always be as useful in a different taxon. The finding that so many marker genes have poor transferability is striking, and by interrogating the assumption of transferability in a thorough and systematic fashion, this paper reminds us of the importance of systematically validating analytical choices. The focus on identifying how transferability varies across different types of marker genes (especially when comparing TFs to lncRNAs), and on exploring different methods to identify marker genes, also suggests additional criteria by which future researchers could select robust marker genes in their own data.

The paper is built on a substantial amount of clearly reported and thoroughly considered data, including EBs and cells from four different primate species - humans, orangutans, and two macaque species. The authors go to great lengths to ensure the EBs are as comparable as possible across species, and take similar care with their computational analyses, always erring on the side of drawing conservative conclusions that are robustly supported by their data over more tenuously supported ones that could be impacted by data processing artefacts such as differences in mappability, etc. For example, I like the approach of using liftoff to robustly identify genes in non-human species that can be mapped to and compared across species confidently, rather than relying on the likely incomplete annotation of the non-human primate genomes. The authors also provide an interactive data visualisation website that allows users to explore the dataset in depth, examine expression patterns of their own favourite marker genes and perform the same kinds of analyses on their own data if desired, facilitating consistency between comparative primate studies.

We thank the Reviewer for their kind assessment of our work.

Weaknesses and recommendations:

(1) Embryoid body generation is known to be highly variable from one replicate to the next for both technical and biological reasons, and the authors do their best to account for this, both by their testing of different ways of generating EBs, and by including multiple technical replicates/clones per species. However, there is still some variability that could be worth exploring in more depth. For example, the orangutan seems to have differentiated preferentially towards cardiac mesoderm whereas the other species seemed to prefer ectoderm fates, as shown in Figure 2C. Likewise, Supplementary Figure 2C suggests a significant unbalance in the contributions across replicates within a species, which is not surprising given the nature of EBs, while Supplementary Figure 6 suggests that despite including three different clones from a single rhesus macaque, most of the data came from a single clone. The manuscript would be strengthened by a more thorough exploration of the intra-species patterns of variability, especially for the taxa with multiple biological replicates, and how they impact the number of cell types detected across taxa, etc.

You are absolutely correct in pointing out that the large clonal variability in cell type composition is a challenge for our analysis. We also noted the odd behavior of the orangutan EBs, and their underrepresentation of ectoderm. There are many possible sources for these variable differentiation propensities: clone, sample origin (in this case urine) and individual. However, unfortunately for the orangutan, we have only one individual and one sample origin and thus cannot say whether this germ layer preference says something about the species or is due to our specific sample.

Because of this high variability from multiple sources, getting enough cell types with an appreciable overlap between species was limiting to analyses. In order to be able to derive meaningful conclusions from intra-species analyses and the impact of different sources of variation on cell type propensity, we would need to sequence many more EBs with an experimental design that balances possible sources of variation. This would go beyond the scope of this study.

Instead, here we control for intra-species variation in our analyses as much as possible: For the analysis of cell type specificity and conservation the comparison is relative for the different specificity degrees (Figure 3C).  For the analysis of marker gene conservation, we explicitly take intra-species variation into account (Figure 4D).

The same holds for the temporal aspect of the data, which is not really discussed in depth despite being a strength of the design. Instead, days 8 and 16 are analysed jointly, without much attention being paid to the possible differences between them.

Concerning the temporal aspect, indeed we knowingly omitted to include an explicit comparison of day 8 and day 16 EBs, because we felt that it was not directly relevant to our main message. Our pseudotime analysis showed that the differences of the two time points were indeed a matter of degree and not so much of quality. All major lineages were already present at day 8 and even though day 8 cells had on average earlier pseudotimes, there was a large overlap in the pseudotime distributions between the two sampling time points (Author response image 1). That is why we decided to analyse the data together.

Are EBs at day 16 more variable between species than at day 8? Is day 8 too soon to do these kinds of analyses?

When we started the experiment, we simply did not know what to expect. We were worried that cell types at day 8 might be too transient, but longer culture can also introduce biases. That is why we wanted to look at two time points, however as mentioned above the differences are in degree.

Concerning the cell type composition: yes, day 16 EBs are more heterogeneous than day 8 EBs. Firstly, older EBs have more distinguishable cell types and hence even if all EBs had identical composition, the sampling variance would be higher given that we sampled a similar number of cells from both time points. Secondly, in order to grow EBs for a longer time, we moved them from floating to attached culture on day 8 and it is unclear how much variance is added by this extra handling step.

Are markers for earlier developmental progenitors better/more transferable than those for more derived cell types?

We did not see any differences in the marker conservation between early and late cell types, but we have too little data to say whether this carries biological meaning.

Author response image 1.

Pseudotime analysis for a differentiation trajectory towards neurons. Single cells were first aggregated into metacells per species using SEACells (Persad et al. 2023). Pluripotent and ectoderm metacells were then integrated across all four species using Harmony and a combined pseudotime was inferred with Slingshot (Street et al. 2018), specifying iPSCs as the starting cluster. Here, lineage 3 is shown, illustrating a differentiation towards neurons. (A) PHATE embedding colored by pseudotime (Moon et al. 2019). (B) PHATE embedding colored by celltype. (C) Pseudotime distribution across the sampling timepoints (day 8 and day 16) in different species.

(2) Closely tied to the point above, by necessity the authors collapse their data into seven fairly coarse cell types and then examine the performance of canonical marker genes (as well as those discovered de novo) across the species. However some of the clusters they use are somewhat broad, and so it is worth asking whether the lack of specificity exhibited by some marker genes and driving their conclusions is driven by inter-species heterogeneity within a given cluster.

Author response image 2.

UMAP visualization for the Harmony-integrated dataset across all four species for the seven shared cell types, colored by cell type identity (A) and species (B).

Good point, if we understand correctly, the concern is that in our relatively broadly defined cell types, species are not well mixed and that this in turn is partly responsible for marker gene divergence. This problem is indeed difficult to address, because most approaches to evaluate this require integration across species which might lead to questionable results (see our Discussion).

Nevertheless, we attempted an integration across all four species. To this end, we subset the cells for the 7 cell types that we found in all four species and visualized cell types and species in the UMAPs above (Author response image 2).

We see that cardiac fibroblasts appear poorly integrated in the UMAP, but they still have very transferable marker genes across species. We quantified integration quality using the cell-specific mixing score (cms) (Lütge et al. 2021) and indeed found that the proportion of well integrated cells is lowest for cardiac fibroblasts (Author response image 3A). On the other end of the cms spectrum, neural crest cells appear to have the best integration across species, but their marker transferability between species is rather worse than for cardiac fibroblasts (Supplementary Figure 9). Cell-type wise calculated rank-biased overlap scores that we use for marker gene conservation show the same trends (Author response image 3B) as the F1 scores for marker gene transferability.  Hence, given our current dataset we do not see any indication that the low marker gene conservation is a result of too broadly defined cell types.

Author response image 3.

(A) Evaluation of species mixing per cell type in the Harmony-integrated dataset, quantified by the fraction of cells with an adjusted cell-specific mixing score (cms) above 0.05. (B) Summary of rank-biased overlap (RBO) scores per cell type to assess concordance of marker gene rankings for all species pairs.

Reviewer #2 (Public review):

Summary:

The authors present an important study on identifying and comparing orthologous cell types across multiple species. This manuscript focuses on characterizing cell types in embryoid bodies (EBs) derived from induced pluripotent stem cells (iPSCs) of four primate species, humans, orangutans, cynomolgus macaques, and rhesus macaques, providing valuable insights into cross-species comparisons.

Strengths:

To achieve this, the authors developed a semi-automated computational pipeline that integrates classification and marker-based cluster annotation to identify orthologous cell types across primates. This study makes a significant contribution to the field by advancing cross-species cell type identification.

We thank the reviewer for their positive and thoughtful feedback.

Weaknesses:

However, several critical points need to be addressed.

(1) Use of Liftoff for GTF Annotation

The authors used Liftoff to generate GTF files for Pongo abelii, Macaca fascicularis, and Macaca mulatta by transferring the hg38 annotation to the corresponding primate genomes. However, it is unclear why they did not use species-specific GTF files, as all these genomes have existing annotations. Why did the authors choose not to follow this approach?

As Reviewer 1 also points out, also we have observed that the annotation of non-human primates often has truncated 3’UTRs. This is especially problematic for 3’ UMI transcriptome data as the ones in the 10x dataset that we present here. To illustrate this we compared the Liftoff annotation derived from Gencode v32,  that we also used throughout our manuscript to the Ensembl gene annotation Macaca_fascicularis_6.0.111. We used transcriptomes from human and cynomolgus iPSC bulk RNAseq  (Kliesmete et al. 2024) using the Prime-seq protocol (Janjic et al. 2022) which is very similar to 10x in that it also uses 3’ UMIs. On average using Liftoff produces higher counts than the Ensembl annotation (Author response image 4A). Moreover, when comparing across species, using Ensembl for the macaque leads to an asymmetry in differentially expressed genes, with apparently many more up-regulated genes in humans. In contrast, when we use the Liftoff annotation, we detect fewer DE-genes and a similar number of genes is up-regulated in macaques as in humans (Author response image 4B). We think that the many more DE-genes are artifacts due to mismatched annotation in human and cynomolgus macaques. We illustrate this for the case of the transcription factor SALL4 in Author response image 4 C,D.  The Ensembl annotation reports 2 transcripts, while Liftoff from Gencode v32 suggests 5 transcripts, one of which has a longer 3’UTR. This longer transcript is also supported by Nanopore data from macaque iPSCs. The truncation of the 3’UTR in this case leads to underestimation of the expression of SALL4 in macaques and hence SALL4 is detected as up-regulated in humans (DESeq2: LFC= 1.34, p-adj<2e-9). In contrast, when using the Liftoff annotation SALL4 does not appear to be DE between humans and macaques (LFC=0.33, p.adj=0.20).

Author response image 4. 

(A) UMI-counts/ gene for the same cynomolgus macaque iPSC samples. On the x-axis the gtf file from Ensembl Macaca_fascicularis_6.0.111 was used to count and on the y-axis we used our filtered Liftoff annotation that transferred the human gene models from Gencode v32. (B) The # of DE-genes between human  and cynomolgus iPSCs detected with DESeq2. In Liftoff, we counted human samples using Gencode v32 and compared it to the Liftoff annotation of the same human gene models to macFas6. In Ensembl, we use Gencode v32 for the human and  Ensembl Macaca_fascicularis_6.0.111 for the Macaque. For both comparisons we subset the genes to only contain one to one orthologues as annotated in biomart. Up and down regulation is relative to human expression. C) Read counts for one example gene SALL4. Here we used in addition to the Liftoff and Ensembl annotation also transcripts derived from Nanopore cDNA sequencing of cynomolgus iPSCs. D) Gene models for SALL4 in the space of MacFas6 and a coverage for iPSC-Prime-seq bulk RNA-sequencing.

(2) Transcript Filtering and Potential Biases

The authors excluded transcripts with partial mapping (<50%), low sequence identity (<50%), or excessive length differences (>100 bp and >2× length ratio). Such filtering may introduce biases in read alignment. Did the authors evaluate the impact of these filtering choices on alignment rates?

We excluded those transcripts from analysis in both species, because they present a convolution of sequence-annotation differences and expression. The focus in our study is on regulatory evolution and we knowingly omit marker differences that are due to a marker being mutated away, we will make this clearer in the text of a revised version.

(3) Data Integration with Harmony

The methods section does not specify the parameters used for data integration with Harmony. Including these details would clarify how cross-species integration was performed.

We want to stress  that none of our conservation and marker gene analyses relies on cross-species integration. We only used the Harmony integrated data for visualisation in Figure 1 and the rough germ-layer check up in Supplementary Figure S3.  We will add a better description in the revised version.

References

Janjic, Aleksandar, Lucas E. Wange, Johannes W. Bagnoli, Johanna Geuder, Phong Nguyen, Daniel Richter, Beate Vieth, et al. 2022. “Prime-Seq, Efficient and Powerful Bulk RNA Sequencing.” Genome Biology 23 (1): 88.

Kliesmete, Zane, Peter Orchard, Victor Yan Kin Lee, Johanna Geuder, Simon M. Krauß, Mari Ohnuki, Jessica Jocher, Beate Vieth, Wolfgang Enard, and Ines Hellmann. 2024. “Evidence for Compensatory Evolution within Pleiotropic Regulatory Elements.” Genome Research 34 (10): 1528–39.

Lütge, Almut, Joanna Zyprych-Walczak, Urszula Brykczynska Kunzmann, Helena L. Crowell, Daniela Calini, Dheeraj Malhotra, Charlotte Soneson, and Mark D. Robinson. 2021. “CellMixS: Quantifying and Visualizing Batch Effects in Single-Cell RNA-Seq Data.” Life Science Alliance 4 (6): e202001004.

Moon, Kevin R., David van Dijk, Zheng Wang, Scott Gigante, Daniel B. Burkhardt, William S. Chen, Kristina Yim, et al. 2019. “Visualizing Structure and Transitions in High-Dimensional Biological Data.” Nature Biotechnology 37 (12): 1482–92.

Persad, Sitara, Zi-Ning Choo, Christine Dien, Noor Sohail, Ignas Masilionis, Ronan Chaligné, Tal Nawy, et al. 2023. “SEACells Infers Transcriptional and Epigenomic Cellular States from Single-Cell Genomics Data.” Nature Biotechnology 41 (12): 1746–57.

Street, Kelly, Davide Risso, Russell B. Fletcher, Diya Das, John Ngai, Nir Yosef, Elizabeth Purdom, and Sandrine Dudoit. 2018. “Slingshot: Cell Lineage and Pseudotime Inference for Single-Cell Transcriptomics.” BMC Genomics 19 (1): 477.

Author Response

We would like to thank the senior editor, reviewing editor and all the reviewers for taking out precious time to review our manuscript and appreciating our study. We are excited that all of you have found strength in our work and have provided comments to strengthen it further. We sincerely appreciate the valuable comments and suggestions, which we believe will help us to further improve the quality of our work.

Reviewer 1

The manuscript by Dubey et al. examines the function of the acetyltransferase Tip60. The authors show that (auto)acetylation of a lysine residue in Tip60 is important for its nuclear localization and liquid-liquid-phase-separation (LLPS). The main observations are: (i) Tip60 is localized to the nucleus, where it typically forms punctate foci. (ii) An intrinsically disordered region (IDR) within Tip60 is critical for the normal distribution of Tip60. (iii) Within the IDR the authors show that a lysine residue (K187), that is auto-acetylated, is critical. Mutation of that lysine residue to a non-acetylable arginine abolishes the behavior. (iv) biochemical experiments show that the formation of the punctate foci may be consistent with LLPS.

On balance, this is an interesting study that describes the role of acetylation of Tip60 in controlling its biochemical behavior as well as its localization and function in cells. The authors mention in their Discussion section other examples showing that acetylation can change the behavior of proteins with respect to LLPS; depending on the specific context, acetylation can promote (as here for Tip60) or impair LLPS.

Strengths:

The experiments are largely convincing and appear to be well executed.

Weaknesses:

The main concern I have is that all in vivo (i.e. in cells) experiments are done with overexpression in Cos-1 cells, in the presence of the endogenous protein. No attempt is made to use e.g. cells that would be KO for Tip60 in order to have a cleaner system or to look at the endogenous protein. It would be reassuring to know that what the authors observe with highly overexpressed proteins also takes place with endogenous proteins.

Response: The main reason to perform these experiments with overexpression system was to generate different point mutants and deletion mutants of TIP60 and analyse their effect on its properties and functions. To validate our observations with overexpression system, we also examined localization pattern of endogenous TIP60 by IFA and results depict similar kind of foci pattern within the nucleus as observed with overexpressed TIP60 protein (Figure 4A). However, we understand the reviewers concern and agree to repeat some of the overexpression experiments under endogenous TIP60 knockdown conditions using siRNA or shRNA against 3’ UTR region.

Also, it is not clear how often the experiments have been repeated and additional quantifications (e.g. of western blots) would be useful.

Response: The experiments were performed as independent biological replicates (n=3) and this is mentioned in the figure legends. Regarding the suggestion for quantifying Western blots, we want to bring into the notice that where ever required (for blots such as Figure 2F, 6H) that require quantitative estimation, graph representing quantitated value with p-value had already been added. However as suggested, in addition, quantitation for Figure 6D will be performed and added in the revised version.

In addition, regarding the LLPS description (Figure 1), it would be important to show the wetting behaviour and the temperature-dependent reversibility of the droplet formation.

Response: We appreciate the suggestion, and we will perform these assays and include the results in the revised version.

In Fig 3C the mutant (K187R) Tip60 is cytoplasmic, but still appears to form foci. Is this still reflecting phase separation, or some form of aggregation?

Response: TIP60 (K187R) mutant remains cytosolic with homogenous distribution as shown in Figure 2E. Also with TIP60 partners like PXR or p53, this mutant protein remains homogenously distributed in the cytosol. However, when co-expressed with TIP60 (Wild-type) protein, this mutant protein although still remain cytosolic some foci-like pattern is also observed at the nuclear periphery which we believe could be accumulated aggregates.

Reviewer 2

The manuscript "Autoacetylation-mediated phase separation of TIP60 is critical for its functions" by Dubey S. et al reported that the acetyltransferase TIP60 undergoes phase separation in vitro and cell nuclei. The intrinsically disordered region (IDR) of TIP60, particularly K187 within the IDR, is critical for phase separation and nuclear import. The authors showed that K187 is autoacetylated, which is important for TIP60 nuclear localization and activity on histone H4. The authors did several experiments to examine the function of K187R mutants including chromatin binding, oligomerization, phase separation, and nuclear foci formation. However, the physiological relevance of these experiments is not clear since TIP60 K187R mutants do not get into nuclei. The authors also functionally tested the cancer-derived R188P mutant, which mimics K187R in nuclear localization, disruption of wound healing, and DNA damage repair. However, similar to K187R, the R188P mutant is also deficient in nuclear import, and therefore, its defects cannot be directly attributed to the disruption of the phase separation property of TIP60. The main deficiency of the manuscript is the lack of support for the conclusion that "autoacetylation-mediated phase separation of TIP60 is critical for its functions".

This study offers some intriguing observations. However, the evidence supporting the primary conclusion, specifically regarding the necessity of the intrinsically disordered region (IDR) and K187ac of TIP60 for its phase separation and function in cells, lacks sufficient support and warrants more scrutiny. Additionally, certain aspects of the experimental design are perplexing and lack controls to exclude alternative interpretations. The manuscript can benefit from additional editing and proofreading to improve clarity.

Response: We understand the point raised by the reviewer, however we would like to draw his attention to the data where we clearly demonstrated that acetylation of lysine 187 within the IDR of TIP60 is required for its phase separation (Figure 2J). We would like to draw reviewer’s attention to other TIP60 mutants within IDR (R177H, R188H, K189R) which all enters the nucleus and make phase separated foci. Cancer-associated mutation at R188 behaves similarly because it also hampers TIP60 acetylation at the adjacent K187 residue. Our in vitro and in cellulo results clearly demonstrate that autoacetylation of TIP60 at K187 within its IDR is critical for multiple functions including its translocation inside the nucleus, its protein-protein interaction and oligomerization which are prerequisite for phase separation of TIP60.

There are two putative NLS sequences (NLS #1 from aa145; NLS #2 from aa184) in TIP60, both of which are within the IDR. Deletion of the whole IDR is therefore expected to abolish the nuclear localization of TIP60. Since K187 is within NLS #2, the cytoplasmic localization of the IDR and K187R mutants may not be related to the ability of TIP60 to phase separation.

Response: We are not disputing the presence of putative NLS within IDR region of TIP60, however our results through different mutations within IDR region (K76, K80, K148, K150, R177, R178, R188, K189) clearly demonstrate that only K187 residue acetylation is critical to shuttle TIP60 inside the nucleus while all other lysine mutants located within these putative NLS region exhibited no impact on TIP60’s nuclear shuttling. We have mentioned this in our discussion, that autoacetylation of TIP60’s K187 may induce local structural modifications in its IDR which is critical for translocating TIP60 inside the nucleus where it undergoes phase separation critical for its functions. A previous example of similar kind shows, acetylation of lysine within the NLS region of TyrRS by PCAF promote its nuclear localization (Cao X et al 2017, PNAS). IDR region (which also contains K187 site) is important for phase separation once the protein enters inside the nucleus. This could be the cell’s mechanism to prevent unwarranted action of TIP60 until it enters the nucleus and phase separate on chromatin at appropriate locations.

The chromatin-binding activity of TIP60 depends on HAT activity, but not phase-separation (Fig 1I), (Fig 2B). How do the authors reconcile the fact that the K187R mutant is able to bind to chromatin with lower activity than the HAT mutant (Fig 2F, 2I)?

Response: K187 acetylation is required for TIP60’s nuclear translocation but not critical for chromatin binding. When soluble fraction is prepared in fractionation experiment, nuclear membrane is disrupted and TIP60 (K187R) mutant has no longer hindrance in accessing the chromatin and thus can load on the chromatin (although not as efficient as Wild-type protein). For efficient chromatin binding auto-acetylation of other lysine residues in TIP60 is required which might be hampered due to reduced catalytic activity or not sufficient enough to maintain equilibrium with HDAC’s activity inside the nucleus. In case of K187R, the reduced auto-acetylation is captured when protein is the cytosol. During fractionation, once this mutant has access to chromatin, it might auto-acetylate other lysine residues critical for chromatin loading (remember catalytic domain is intact in this mutant). This is evident due to hyper auto-acetylation of Wild-type protein compared to K187R or HAT mutant proteins. We want to bring into notice that phase-separation occurs only after efficient chromatin loading of TIP60 that is the reason that under in-cellulo conditions, both K187R (which cannot enter the nucleus) and HAT mutant (which enters the nucleus but fails to efficiently binds onto the chromatin) fails to form phase separated nuclear punctate foci.

The DIC images of phase separation in Fig 2I need to be improved. The image for K187R showed the irregular shape of the condensates, which suggests particles in solution or on the slide. The authors may need to use fluorescent-tagged TIP60 in the in vitro LLPS experiments.

Response: We believe this comment is for figure 2J. The irregularly shaped condensates observed for TIP60 K187R are unique to the mutant protein and are not caused by particles on the slide. We would like to draw reviewer’s attention to supplementary figure S2A, where DIC images for TIP60 (Wild-type) protein tested under different protein and PEG8000 conditions are completely clear where protein did not made phase separated droplets ruling out the probability of particles in solution or slides.

The authors mentioned that the HAT mutant of TIP60 does not phase separate, which needs to be included.

Response: We have already added the image of RFP-TIP60 (HAT mutant) in supplementary Fig S4A (panel 2) in the manuscript.

Related to Point 3, the HAT mutant that doesn't form punctate foci by itself, can incorporate into WT TIP60 (Fig 5A). In vitro LLPS assay for WT, HAT, and K187R mutants with or without acetylation should be included. WT and mutant TIP can be labelled with GFP and RFP, respectively.

Response: We would like to draw reviewer’s attention towards our co-expression experiments performed in Figure 5 where Wild-type protein (both tagged and untagged condition) is able to phase separate and make punctate foci with co-expressed HAT mutant protein (with depleted autoacetylation capacity). We believe these in cellulo experiments are already able to answer the queries what reviewer is suggesting to acheive by in vitro experiments.

Fig 3A and 3B showed that neither K187 mutant nor HAT mutant could oligomerize. If both experiments were conducted in the absence of in vitro acetylation, how do the authors reconcile these results?

Response: We thank the reviewer for highlighting our oversight in omitting the mention of acetyl coenzyme A here. To induce acetylation under in vitro conditions, we have added 10 µM acetyl CoA into the reactions depicted in Figure 3A and 3B. The information for acetyl CoA for Figure 3B was already included in the GST-pull down assay (material and methods section). We will add the same in the oligomerization assay of material and methods in the revised manuscript.

In Fig 4, the colocalization images showed little overlap between TIP60 and nuclear speckle (NS) marker SC35, indicating that the majority of TIP60 localized in the nuclear structure other than NS. Have the authors tried to perturbate the NS by depleting the NS scaffold protein and examining TIP60 foci formation? Do PXR and TP53 localize to NS?

Response: Under normal conditions majority of TIP60 is not localized in nuclear speckles (NS) so we believe that perturbing NS will not have significant effect on TIP60 foci formation. Interestingly, recently a study by Shelly Burger group (Alexander KA et al Mol Cell. 2021 15;81(8):1666-1681) had shown that p53 localizes to NS to regulate subset of its targeted genes. We have mentioned about it in our discussion section. No information is available about localization of PXR in NS.

Were TIP60 substrates, H4 (or NCP), PXR, TP53, present inTIP60 condensates in vitro? It's interesting to see both PXR and TP53 had homogenous nuclear signals when expressed together with K187R, R188P (Fig 6E, 6G), or HAT (Suppl Fig S4A) mutants. Are PXR or TP53 nuclear foci dependent on their acetylation by TIP60? This can and should be tested.

Response: Both p53 and PXR are known to be acetylated by TIP60. In case of PXR, TIP60 acetylate PXR at lysine 170 and this TIP60-mediated acetylation of PXR at K170 is important for TIP60-PXR foci which now we know are formed by phase separation (Bakshi K et al Sci Rep. 2017 Jun 16;7(1):3635).

Since R188P mutant, like K187R, does not get into the nuclei, it is not suitable to use this mutant to examine the functional relevance of phase separation for TIP60. The authors need to find another mutant in IDR that retains nuclear localization and overall HAT activity but specifically disrupts phase separation. Otherwise, the conclusion needs to be restated. All cancer-derived mutants need to be tested for LLPS in vitro.

Response: We appreciate the reviewer’s point here, but it is important to note that the objective of these experiments is to understand the impact of K187R (critical in multiple aspects of TIP60 including phase separation) and R188P (a naturally occurring cancer-associated mutation and behaving similarly to K187R) on TIP60’s activities to determine their functional relevance. As suggested by the reviewer to test and find IDR mutant that fails to phase separate however retains nuclear localization and catalytic activity can be examined in future studies.

For all cellular experiments, it is not mentioned whether endogenous TIP60 was removed and absent in the cell lines used in this study. It's important to clarify this point because the localization and function of mutant TIP60 are affected by WT TIP60 (Fig 5).

Response: Endogenous TIP60 was present in in cellulo experiments, however as suggested by reviewer 1 we will perform some of the in cellulo experiments under endogenous TIP60 knockdown condition to validate our findings.

It is troubling that H4 peptide is used for in vitro HAT assay since TIP60 has much higher activity on nucleosomes and its preferred substrates include H2A.

Response: The purpose of using H4 peptide in the HAT assay is to determine the impact of mutations of TIP60’s catalytic activity. As H4 is one of the major histone substrate for TIP60, we believe it satisfy the objective of experiments.

Reviewer 3

This study presents results arguing that the mammalian acetyltransferase Tip60/KAT5 auto-acetylates itself on one specific lysine residue before the MYST domain, which in turn favors not only nuclear localization but also condensate formation on chromatin through LLPS. The authors further argue that this modification is responsible for the bulk of Tip60 autoacetylation and acetyltransferase activity towards histone H4. Finally, they suggest that it is required for association with txn factors and in vivo function in gene regulation and DNA damage response.

These are very wide and important claims and, while some results are interesting and intriguing, there is not really close to enough work performed/data presented to support them. In addition, some results are redundant between them, lack consistency in the mutants analyzed, and show contradiction between them. The most important shortcoming of the study is the fact that every single experiment in cells was done in over-expressed conditions, from transiently transfected cells. It is well known that these conditions can lead to non-specific mass effects, cellular localization not reflecting native conditions, and disruption of native interactome. On that topic, it is quite striking that the authors completely ignore the fact that Tip60 is exclusively found as part of a stable large multi-subunit complex in vivo, with more than 15 different proteins. Thus, arguing for a single residue acetylation regulating condensate formation and most Tip60 functions while ignoring native conditions (and the fact that Tip60 cannot function outside its native complex) does not allow me to support this study.

Response: We appreciate the reviewer’s point here, but it is important to note that the main purpose to use overexpression system in the study is to analyse the effect of different generated point/deletion mutations on TIP60. We have overexpressed proteins with different tags (GFP or RFP) or without tags (Figure 3C, Figure 5) to confirm the behaviour of protein which remains unperturbed due to presence of tags. To validate we have also examined localization of endogenous TIP60 protein which also depict similar localization behaviour as overexpressed protein. We would like to draw attention that there are several reports in literature where similar kind of overexpression system are used to determine functions of TIP60 and its mutants. Also nuclear foci pattern observed for TIP60 in our studies is also reported by several other groups.

Sun, Y., et. al. (2005) A role for the Tip60 histone acetyltransferase in the acetylation and activation of ATM. Proc Natl Acad Sci U S A, 102(37):13182-7.

Kim, C.-H. et al. (2015) ‘The chromodomain-containing histone acetyltransferase TIP60 acts as a code reader, recognizing the epigenetic codes for initiating transcription’, Bioscience, Biotechnology, and Biochemistry, 79(4), pp. 532–538.

Wee, C. L. et al. (2014) ‘Nuclear Arc Interacts with the Histone Acetyltransferase Tip60 to Modify H4K12 Acetylation(1,2,3).’, eNeuro, 1(1). doi: 10.1523/ENEURO.0019-14.2014.

However, as a caution and suggested by other reviewers also we will perform some of these overexpression experiments in absence of endogenous TIP60 by using 3’ UTR specific siRNA/shRNA.

We thank the reviewer for his comment on muti-subunit complex proteins and we would like to expand our study by determining the interaction of some of the complex subunits with TIP60 ((Wild-type) that forms nuclear condensates), TIP60 ((HAT mutant) that enters the nucleus but do not form condensates) and TIP60 ((K187R) that do not enter the nucleus and do not form condensates). We will include the result of these experiments in the revised manuscript.

• It is known that over-expression after transient transfection can lead to non-specific acetylation of lysines on the proteins, likely in part to protect from proteasome-mediated degradation. It is not clear whether the Kac sites targeted in the experiments are based on published/public data. In that sense, it is surprising that the K327R mutant does not behave like a HAT-dead mutant (which is what exactly?) or the K187R mutant as this site needs to be auto-acetylated to free the catalytic pocket, so essential for acetyltransferase activity like in all MYST-family HATs. In addition, the effect of K187R on the total acetyl-lysine signal of Tip60 is very surprising as this site does not seem to be a dominant one in public databases.

Response: We have chosen autoacetylation sites based on previously published studies where LC-MS/MS and in vitro acetylation assays were used to identified autoacetylation sites in TIP60 which includes K187. We have already mentioned about it in the manuscript and have quoted the references (1. Yang, C., et al (2012). Function of the active site lysine autoacetylation in Tip60 catalysis. PloS one 7, e32886. 10.1371/journal.pone.0032886. 2. Yi, J., et al (2014). Regulation of histone acetyltransferase TIP60 function by histone deacetylase 3. The Journal of biological chemistry 289, 33878–33886. 10.1074/jbc.M114.575266.). We would like to emphasize that both these studies have identified K187 as autoacetylation site in TIP60. Since TIP60 HAT mutant (with significantly reduced catalytic activity) can also enter nucleus, it is not surprising that K327 could also enter the nucleus.

• As the physiological relevance of the results is not clear, the mutants need to be analyzed at the native level of expression to study real functional effects on transcription and localization (ChIP/IF). It is not clear the claim that Tip60 forms nuclear foci/punctate signals at physiological levels is based on what. This is certainly debated because in part of the poor choice of antibodies available for IF analysis. In that sense, it is not clear which Ab is used in the Westerns. Endogenous Tip60 is known to be expressed in multiple isoforms from splice variants, the most dominant one being isoform 2 (PLIP) which lacks a big part (aa96-147) of the so-called IDR domain presented in the study. Does this major isoform behave the same?

Response: TIP60 antibody used in the study is from Santa Cruz (Cat. No.- sc-166323). This antibody is widely used for TIP60 detection by several methods and has been cited in numerous publications. Cat. No. will be mentioned in the manuscript. Regarding isoforms, three isoforms are known for TIP60 among which isoform 2 is majorly expressed and used in our study. Isoform and 1 and 2 have same length of IDR (150 amino acids) while isoform 3 has IDR of 97 amino acids. Interestingly, the K187 is present in all the isoforms (already mentioned in the manuscript) and missing region (96-147 amino acid) in isoform 3 has less propensity for disordered region (marked in blue circle). This clearly shows that all the isoforms of TIP60 has the tendency to phase separate.

Author response image 1.

• It is extremely strange to show that the K187R mutant fails to get in the nuclei by cell imaging but remains chromatin-bound by fractionation... If K187 is auto-acetylated and required to enter the nucleus, why would a HAT-dead mutant not behave the same?

Response: We would like to draw attention that both HAT mutant and K187R mutant are not completely catalytically dead. As our data shows both these mutants have catalytic activity although at significantly decreased levels. We believe that K187 acetylation is critical for TIP60 to enter the nucleus and once TIP60 shuttles inside the nucleus autoacetylation of other sites is required for efficient chromatin binding of TIP60. In fractionation assay, nuclear membrane is dissolved while preparing the soluble fraction so there is no hindrance for K187R mutant in accessing the chromatin. While in the case of HAT mutant, it can acetylate the K187 site and thus is able to enter the nucleus however this residual catalytic activity is either not able to autoacetylate other residues required for its efficient chromatin binding or to counter activities of HDAC’s deacetylating the TIP60.

• If K187 acetylation is key to Tip60 function, it would be most logical (and classical) to test a K187Q acetyl-mimic substitution. In that sense, what happens with the R188Q mutant? That all goes back to the fact that this cluster of basic residues looks quite like an NLS.

Response: As suggested we will generate acetylation mimicking mutant for K187 site and examine it. Result will be added in the revised manuscript.

• The effect of the mutant on the TIP60 complex itself needs to be analyzed, e.g. for associated subunits like p400, ING3, TRRAP, Brd8...

Response: As suggested we will examine the effect of mutations on TIP60 complex

Author Response:

Reviewer #1:

Summary:

This research study utilizes a realistic motoneuron model to explore the potential to trace back the appropriate levels of excitation, inhibition, and neuromodulation in the firing patterns of motoneurons observed in in-vitro and in-vivo experiments in mammals. The research employs high-performance computing power to achieve its objectives. The work introduces a new framework that enhances understanding of the neural inputs to motoneuron pools, thereby opening up new avenues for hypothesis testing research.

Strengths: The significance of the study holds relevance for all neuroscientists. Motoneurons are a unique class of neurons with known distribution of outputs for a wide range of voluntary and involuntary motor commands, and their physiological function is precisely understood. More importantly, they can be recorded in-vivo using minimally invasive methods, and they are directly impacted by many neurodegenerative diseases at the spinal cord level. The computational framework developed in this research offers the potential to reverse engineer the synaptic input distribution when assessing motor unit activity in humans, which holds particular importance. Overall, the strength of the findings focuses on providing a novel framework for studying and understanding the inputs that govern motoneuron behavior, with broad applications in neuroscience and potential implications for understanding neurodegenerative diseases. It highlights the significance of the study for various research domains, making it valuable to the scientific community.

Weaknesses: The exact levels of inhibition, excitation, and neuromodulatory inputs to neural networks are unknown. Therefore the work is based on fine-tuned measures that are indirectly based on experimental results. However, obtaining such physiological information is challenging and currently impossible. From a computational perspective it is a challenge that in theory can be solved. Thus, although we have no ground-truth evidence, this framework can provide compelling evidence for all hypothesis testing research and potentially solve this physiological problem with the use of computers.

We agree with the reviewer. This work was intended to determine the feasibility of reverse engineering motor unit firing patterns, using neuron models with a high degree realism. Given the results support this feasibility, our model and technique will therefore serve to construct new hypotheses as well as testing them.

Reviewer #2:

The study presents an extensive computational approach to identify the motor neuron input from the characteristics of single motor neuron discharge patterns during a ramp up/down contraction. This reverse engineering approach is relevant due to limitations in our ability to estimate this input experimentally. Using well-established models of single motor neurons, a (very) large number of simulations were performed that allowed identification of this relation. In this way, the results enable researchers to measure motor neuron behavior and from those results determine the underlying neural input scheme. Overall, the results are very convincing and represent an important step forward in understanding the neural strategies for controlling movement.

Nevertheless, I would suggest that the authors consider the following recommendations to strengthen the message further. First, I believe that the relation between individual motor neuron behavioral characteristics (delta F, brace height etc.) and the motor neuron input properties can be illustrated more clearly. Although this is explained in the text, I believe that this is not optimally supported by figures. Figure 6 to some extent shows this, but figures 8 and 9 as well as Table 1 shows primarily the goodness of fit rather than the actual fit.

We agree with the reviewer that showing the relationship between the motor neuron behavioral characteristics (delta F, brace height etc.) and the motor neuron input properties would be a great addition to the manuscript. Because the regression models have multiple dimensions (7 inputs and 3 outputs) it is difficult to show the relationship in a static image. We thought it best to show the goodness of fit even though it is more abstract and less intuitive. We added a supplemental diagram to Figure 8 to show the structure of the reverse engineered model that was fit (see Figure 8D).

Author response image 1: Figure 8. Residual plots showing the goodness of fit of the different predicted values: (A) Inhibition, (B) Neuromodulation and (C) excitatory Weight Ratio. The summary plots are for the models showing highest 𝑅2 results in Table 1. The predicted values are calculated using the features extracted from the firing rates (see Figure 7, section Machine learning inference of motor pool characteristics and Regression using motoneuron outputs to predict input organization). Diagram (D) shows the multidimensionality of the RE models (see Model fits) which have 7 feature inputs (see Feature Extraction) predicting 3 outputs (Inhibition, Neuromodulation and Weight Ratio).

Second, I would have expected the discussion to have addressed specifically the question of which of the two primary schemes (push-pull, balanced) is the most prevalent. This is the main research question of the study, but it is to some degree left unanswered. Now that the authors have identified the relation between the characteristics of motor neuron behaviors (which has been reported in many previous studies), why not exploit this finding by summarizing the results of previous studies (at least a few representative ones) and discuss the most likely underlying input scheme? Is there a consistent trend towards one of the schemes, or are both strategies commonly used?

We agree with the reviewer that our discussion should have addressed which of the two primary schemes – push-pull or balanced – is the most prevalent. At first glance, the upper right of Figure 6 looks the most realistic when compared to real data. We thus would expect that the push-pull scheme to dominate for the given task. We added a brief section (Push-Pull vs Balance Motor Command) in the discussion to address the reviewer’s comments. This section is not exhaustive but frames the debate using relevant literature. We are also now preparing to deploy these techniques on real data.

In addition, it seems striking to me that highly non-linear excitation profiles are necessary to obtain a linear CST ramp in many model configurations. Although somewhat speculative, one may expect that an approximately linear relation is desired for robust and intuitive motor control. It seems to me that humans generally have a good ability to accurately grade the magnitude of the motor output, which implies that either a non-linear relation has been learnt (complex task), or that the central nervous system can generally rely on a somewhat linear relation between the neural drive to the muscle and the output (simpler task).

We agree with the reviewer, and we were surprised by these results. Our motoneuron pool is equipped with persistent inward currents (PICs) which are nonlinear. Therefore, for the motoneuron to produce a linear output the central nervous system would have to incorporate these nonlinearities into its commands.

Following this reasoning, it could be interesting to report also for which input scheme, the excitation profile is most linear. I understand that this is not the primary aim of the study, but it may be an interesting way to elaborate on the finding that in many cases non-linear excitation profiles were needed to produce the linear ramp.

This is a very interesting point. The most realistic firing patterns – with respect to human data – are found in the parameter regions in the upper right in Figure 6, which in fact produce the most nonlinear input (see push-pull pattern in Figure 4C). However, in future studies we hope to separate the total motor command illustrated here into descending and feedback commands. This may result in a more linear descending drive.

Author response:

Public Reviews:

Reviewer #1 (Public Review):

(1) It will be interesting to monitor the levels of another MIM insertase namely, OXA1. This will help to understand whether some of the observed changes in levels of OXPHOS subunits are related to alterations in the amounts of this insertase.

OXA1 was not detected in the untargeted mass spectrometry analysis, most likely due to the fact that it is a polytopic membrane protein, spanning the membrane five times (1,2). Consequently, we measured OXA1 levels with immunoblotting, comparing patient fibroblast cells to the HC. No significant change in OXA1 steady state levels was observed. 

See the results below. These results will be added and discussed in the revised manuscript.

Author response image 1.

(2) Figure 3: How do the authors explain that although TIMM17 and TIMM23 were found to be significantly reduced by Western analysis they were not detected as such by the Mass Spec. method?

The untargeted mass spectrometry in the current study failed to detect the presence of TIMM17 for both, patient fibroblasts and mice neurons, while TIMM23 was detected only for mice neurons and a decrease was observed for this protein but was not significant. This is most likely due to the fact that TIMM17 and TIMM23 are both polytopic membrane proteins, spanning the membrane four times, which makes it difficult to extract them in quantities suitable for MS detection (2,3).

(3) How do the authors explain the higher levels of some proteins in the TIMM50 mutated cells?

The levels of fully functional TIM23 complex are deceased in patients' fibroblasts. Therefore, the mechanism by which the steady state level of some TIM23 substrate proteins is increased, can only be explained relying on events that occur outside the mitochondria. This could include increase in transcription, translation or post translation modifications, all of which may increase their steady state level albite the decrease in the steady state level of the import complex.

(4) Can the authors elaborate on why mutated cells are impaired in their ability to switch their energetic emphasis to glycolysis when needed?

Cellular regulation of the metabolic switch to glycolysis occurs via two known pathways: 1) Activation of AMP-activated protein kinase (AMPK) by increased levels of AMP/ADP (4). 2) Inhibition of pyruvate dehydrogenase (PDH) complexes by pyruvate dehydrogenase kinases (PDK) (5). Therefore, changes in the steady state levels of any of these regulators could push the cells towards anaerobic energy production, when needed. In our model systems, we did not observe changes in any of the AMPK, PDH or PDK subunits that were detected in our untargeted mass spectrometry analysis (see volcano plots below, no PDK subunits were detected in patient fibroblasts). Although this doesn’t directly explain why the cells have an impaired ability to switch their energetic emphasis, it does possibly explain why the switch did not occur de facto.

Author response image 2.

Reviewer #2 (Public Review):

(1) The authors claim in the abstract, the introduction, and the discussion that TIMM50 and the TIM23 translocase might not be relevant for mitochondrial protein import in mammals. This is misleading and certainly wrong!!!

Indeed, it was not in our intention to claim that the TIM23 complex might not be relevant. We have now rewritten the relevant parts to convey the correct message:

Abstract – 

Line 25 - “Strikingly, TIMM50 deficiency had no impact on the steady state levels of most of its putative substrates, suggesting that even low levels of a functional TIM23 complex are sufficient to maintain the majority of complex-dependent mitochondrial proteome.”

Introduction – 

Line 87 - Surprisingly, functional and physiological analysis points to the possibility that low levels of TIM23 complex core subunits (TIMM50, TIMM17 and TIMM23) are sufficient for maintaining steady-state levels of most presequence-containing proteins. However, the reduced TIM23CORE component levels do affect some critical mitochondrial properties and neuronal activity.

Discussion – 

Line 339 – “…surprising, as normal TIM23 complex levels are suggested to be indispensable for the translocation of presequence-containing mitochondrial proteins…”

Line 344 – “…it is possible that unlike what occurs in yeast, normal levels of mammalian TIMM50 and TIM23 complex are mainly essential for maintaining the steady state levels of intricate complexes/assemblies.”

Line 396 – “In summary, our results suggest that even low levels of TIMM50 and TIM23CORE components suffice in maintaining the majority of mitochondrial matrix and inner membrane proteome. Nevertheless, reductions in TIMM50 levels led to a decrease of many OXPHOS and MRP complex subunits, which indicates that normal TIMM50 levels might be mainly essential for maintaining the steady state levels and assembly of intricate complex proteins.”

(1) Homberg B, Rehling P, Cruz-Zaragoza LD. The multifaceted mitochondrial OXA insertase. Trends Cell Biol. 2023;33(9):765–72. 

(2) Carroll J, Altman MC, Fearnley IM, Walker JE. Identification of membrane proteins by tandem mass spectrometry of protein ions. Proc Natl Acad Sci U S A.

2007;104(36):14330–5. 

(3) Dekker PJT, Keil P, Rassow J, Maarse AC, Pfanner N, Meijer M. Identification of MIM23, a putative component of the protein import machinery of the mitochondrial inner membrane. FEBS Lett. 1993;330(1):66–70. 

(4) Trefts E, Shaw RJ. AMPK: restoring metabolic homeostasis over space and time. Mol Cell [Internet]. 2021;81(18):3677–90. Available from:

https://doi.org/10.1016/j.molcel.2021.08.015

(5) Zhang S, Hulver MW, McMillan RP, Cline MA, Gilbert ER. The pivotal role of pyruvate dehydrogenase kinases in metabolic flexibility. Nutr Metab. 2014;11(1):1–9.

Author response:

Public Reviews:

Reviewer #1 (Public review):

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

Summary:

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

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

Strengths:

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

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

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

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

Weaknesses:

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

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

We appreciate the reviewer's comment and the relevant references. We have revised the manuscript accordingly to clarify the distinction between studies that incorporate limited interactions and those that explicitly analyze sensory constraints and interference. We have refined our statement to acknowledge these contributions while maintaining our focus on sensory challenges beyond limited neighbor detection, such as signal degradation, occlusion effects, and multimodal sensory integration (see lines 61-64):

While collective movement has been extensively studied in various species, including insect swarming, fish schooling, and bird murmuration (Pitcher, Partridge and Wardle, 1976; Partridge, 1982; Strandburg-Peshkin et al., 2013; Pearce et al., 2014; Rosenthal, Twomey, Hartnett, Wu, Couzin, et al., 2015; Bastien and Romanczuk, 2020; Davidson et al., 2021; Aidan, Bleichman and Ayali, 2024), as well as in swarm robotics agents performing tasks such as coordinated navigation and maze-solving (Faria Dias et al., 2021; Youssefi and Rouhani, 2021; Cheraghi, Shahzad and Graffi, 2022), most studies have focused on movement algorithms , often assuming full detection of neighbors (Parrish and Edelstein-Keshet, 1999; Couzin et al., 2002, 2005; Sumpter et al., 2008; Nagy et al., 2010; Bialek et al., 2012; Gautrais et al., 2012; Attanasi et al., 2014). Some models have incorporated limited interaction rules where individuals respond to one or a few neighbors due to sensory constraints (Bode, Franks and Wood, 2011; Jhawar et al., 2020). However, fewer studies explicitly examine how sensory interference, occlusion, and noise shape decision-making in collective systems (Rosenthal et al., 2015).

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

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

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

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

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

We have updated the manuscript to use these terms consistently and explicitly define them in relevant sections (see lines 87-94 and 329-330). This distinction ensures that the reader can differentiate between interference as an acoustic phenomenon and its broader implications in navigation.

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

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

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

· Call Intensity Variation: The tested call intensity levels (100, 110, 120, 130 dB SPL) are presented in Table 1 under the “Call Level” parameter. The effect of increasing call intensity was analyzed in relation to exit probability, jamming probability, and collision rate. This is now explicitly referenced in the Discussion.

We have revised the manuscript to explicitly reference these aspects in the Results and Discussion sections.

Reviewer #2 (Public review):

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

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

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

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

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

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

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

To clarify this in the manuscript, we have updated the text to explicitly state that directionality follows phase-dependent frequency changes rather than being a direct function of velocity, see lines 460-465.

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

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

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

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

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

These definitions have been explicitly stated in the revised manuscript to prevent any ambiguity (lines 467-468). Additionally, a Supplementary figure demonstrating the geometrical relations has been added to the manuscript.

Author response image 1.

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

In nature, minor collisions between bats are common and typically do not result in significant disruptions to flight (Boerma et al., 2019; Roy et al., 2019; Goldstein et al., 2024).Given this, our model does not explicitly simulate the physical impact of a collision event. Instead, during the collision event the bat keeps decreasing its velocity and changing its flight direction until the distance between bats is above the threshold (0.4 m). We assume that the primary cost of such interactions arises from the effort required to avoid collisions, rather than from the collision itself. This assumption aligns with observations of bat behavior in dense flight environments, where individuals prioritize collision avoidance rather than modeling post-collision dynamics.

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

The number of repetitions for each scenario is detailed in Table 1, but we included it in a more prominent location in the text for clarity. Specifically, we now state (Lines 274-275):

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

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

We clarified in the revised text (Lines 534-535 in Statistical Analysis)

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

Reviewer #3 (Public review):

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

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

Summary:

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

Strengths:

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

Weaknesses:

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

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

To improve clarity, we have updated the text to explicitly state that the phase transition distances are empirically grounded and apply equally to both modeled species (lines 430-447).

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

In echolocating bats, call parameters are primarily shaped by the target distance and echo strength. Accordingly, there is little difference in call structure between prey capture and obstacles-related maneuvers, aside from intensity adjustments based on target strength (Hagino et al., 2007; Hiryu et al., 2008; Surlykke, Ghose and Moss, 2009; Kothari et al., 2014). In our study, due to the dense cave environment, the bats are found to operate in the approach phase nearly all the time, which is consistent with natural cave emergence, where they are navigating through a cluttered environment rather than engaging in open-space search. For one of the species (Rhinopoma M.), we also have empirical recordings of individuals flying under similar conditions (Goldstein et al., 2024). Our model was designed to remain as simple as possible while relying on conservative assumptions that may underestimate bat performance. If, in reality, bats fine-tune their echolocation calls even earlier or more precisely during navigation than assumed, our model would still conservatively reflect their actual capabilities.

We actually used logarithmically frequency modulated (FM) chirps, generated using the MATLAB built-in function chirp(t, f0, t1, f1, 'logarithmic'). This method aligns with the nonlinear FM characteristics of Pipistrellus kuhlii (PK) and Rhinopoma microphyllum (RM) and provides a realistic approximation of their echolocation signals. We acknowledge that this was not sufficiently emphasized in the original text, and we have now explicitly highlighted this in the revised version to ensure clarity (sell Lines 447-449 in Methods).

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

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

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

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

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

We have now explicitly highlighted this in the revised version (see Lines 468-470).

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

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

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

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

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

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

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

See lines 518-523 in the revied version.

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

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

· Higher Effective Density: A 2D model forces all bats into a single plane rather than distributing them through a 3D volume, increasing the likelihood of overlap in calls and echoes and making jamming more severe. As described in the text: the average distance to the nearest bat in our simulation is 0.27m (with 100 bats), whereas reported distances in very dense colonies are 0.5m, as observed in Myotis grisescens and Tadarida brasiliensis (Fujioka et al., 2021; Sabol and Hudson, 1995; Betke et al., 2008; Gillam et al, 2010)

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

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

To address the reviewer’s concern, we have clarified this justification in the revised text and will provide supporting references where applicable: (see Methods lines 407-412)

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

There is substantial literature supporting the assumption that bats can recognize their own echoes and distinguish them from conspecific signals (Schnitzler and Bioscience, 2001‏; Kazial, Burnett and Masters, 2001; Burnett and Masters, 2002; Kazial, Kenny and Burnett, 2008; Chili, Xian and Moss, 2009; Yovel et al., 2009; Beetz and Hechavarría, 2022). However, we acknowledge that false echoes may present a major challenge in dense groups. To address this, we explicitly tested the impact of the self-echo identification assumption in our study see Results Figure 4: The impact of confusion on performance, and lines 345-355 in the Discussion.

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

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

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

The avoidance distances (1.5 m from walls and 0.4 m from other bats) were selected as internal parameters to ensure coherent flight trajectories while maintaining a reasonable collision rate. These distances provide a balance between maneuverability and stability, preventing erratic flight patterns while still enabling effective obstacle avoidance. In the revised paper, we have added supplementary figures illustrating the effect of model parameters on performance, specifically focusing on the avoidance distance.

The 15-second exit limit was determined as described in the text (Lines 403-404): “A 15-second window was chosen because it is approximately twice the average exit time for 40 bats and allows for a second corrective maneuver if needed.” In other words, it allowed each bat to circle the ‘cave’ twice to exit even in the most crowded environment. This threshold was set to keep simulation time reasonable while allowing sufficient time for most bats to exit successfully.

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

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

The "average exit time for 40 bats" is also confusing and not well explained. Was this determined empirically? From the simulation? If the latter, what are the conditions? Does it include masking, no masking, or which species?

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

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

Regarding the average exit time for 40 bats, this value was determined from our simulations, where it represents the mean time for successful exits under standard conditions with masking.

We have revised the text to clarify these details see, lines 466.

References:

Aidan, Y., Bleichman, I. and Ayali, A. (2024) ‘Pausing to swarm: locust intermittent motion is instrumental for swarming-related visual processing’, Biology letters, 20(2), p. 20230468. Available at: https://doi.org/10.1098/rsbl.2023.0468.

Attanasi, A. et al. (2014) ‘Collective Behaviour without Collective Order in Wild Swarms of Midges’. Edited by T. Vicsek, 10(7). Available at: https://doi.org/10.1371/journal.pcbi.1003697.

Bastien, R. and Romanczuk, P. (2020) ‘A model of collective behavior based purely on vision’, Science Advances, 6(6). Available at: https://doi.org/10.1126/sciadv.aay0792.

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

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

Bialek, W. et al. (2012) ‘Statistical mechanics for natural flocks of birds’, Proceedings of the National Academy of Sciences, 109(13), pp. 4786–4791. Available at: https://doi.org/10.1073/PNAS.1118633109.

Bode, N.W.F., Franks, D.W. and Wood, A.J. (2011) ‘Limited interactions in flocks: Relating model simulations to empirical data’, Journal of the Royal Society Interface, 8(55), pp. 301–304. Available at: https://doi.org/10.1098/RSIF.2010.0397.

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

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

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

Cheraghi, A.R., Shahzad, S. and Graffi, K. (2022) ‘Past, Present, and Future of Swarm Robotics’, in Lecture Notes in Networks and Systems. Available at: https://doi.org/10.1007/978-3-030-82199-9_13.

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

Couzin, I.D. et al. (2002) ‘Collective Memory and Spatial Sorting in Animal Groups’, Journal of Theoretical Biology, 218(1), pp. 1–11. Available at: https://doi.org/10.1006/jtbi.2002.3065.

Couzin, I.D. et al. (2005) ‘Effective leadership and decision-making in animal groups on the move’, Nature, 433(7025), pp. 513–516. Available at: https://doi.org/10.1038/nature03236.

Davidson, J.D. et al. (2021) ‘Collective detection based on visual information in animal groups’, Journal of the Royal Society, 18(180), p. 2021.02.18.431380. Available at: https://doi.org/10.1098/rsif.2021.0142.

Faria Dias, P.G. et al. (2021) ‘Swarm robotics: A perspective on the latest reviewed concepts and applications’, Sensors. Available at: https://doi.org/10.3390/s21062062.

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

Gautrais, J. et al. (2012) ‘Deciphering Interactions in Moving Animal Groups’, PLOS Computational Biology, 8(9), p. e1002678. Available at: https://doi.org/10.1371/JOURNAL.PCBI.1002678.

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

Goldstein, A. et al. (2024) ‘Collective Sensing – On-Board Recordings Reveal How Bats Maneuver Under Severe 4 Acoustic Interference’, Under Review, pp. 1–25.

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

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

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

Jakobsen, L. et al. (2024) ‘Velocity as an overlooked driver in the echolocation behavior of aerial hawking vespertilionid bats’. Available at: https://doi.org/10.1016/j.cub.2024.12.042.

Jakobsen, L., Brinkløv, S. and Surlykke, A. (2013) ‘Intensity and directionality of bat echolocation signals’, Frontiers in Physiology, 4 APR(April), pp. 1–9. Available at: https://doi.org/10.3389/fphys.2013.00089.

Jakobsen, L. and Surlykke, A. (2010) ‘Vespertilionid bats control the width of their biosonar sound beam dynamically during prey pursuit’, 107(31). Available at: https://doi.org/10.1073/pnas.1006630107.

Jhawar, J. et al. (2020) ‘Noise-induced schooling of fish’, Nature Physics 2020 16:4, 16(4), pp. 488–493. Available at: https://doi.org/10.1038/s41567-020-0787-y.

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

Kazial, K.A., Burnett, S.C. and Masters, W.M. (2001) ‘ Individual and Group Variation in Echolocation Calls of Big Brown Bats, Eptesicus Fuscus (Chiroptera: Vespertilionidae) ’, Journal of Mammalogy, 82(2), pp. 339–351. Available at: https://doi.org/10.1644/1545-1542(2001)082<0339:iagvie>2.0.co;2.

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

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

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

Nagy, M. et al. (2010) ‘Hierarchical group dynamics in pigeon flocks’, Nature 2010 464:7290, 464(7290), pp. 890–893. Available at: https://doi.org/10.1038/nature08891.

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

Parrish, J.K. and Edelstein-Keshet, L. (1999) ‘Complexity, Pattern, and Evolutionary Trade-Offs in Animal Aggregation’, Science, 284(5411), pp. 99–101. Available at: https://doi.org/10.1126/SCIENCE.284.5411.99.

Partridge, B.L. (1982) ‘The Structure and Function of Fish Schools’, 246(6), pp. 114–123. Available at: https://doi.org/10.2307/24966618.

Pearce, D.J.G. et al. (2014) ‘Role of projection in the control of bird flocks’, Proceedings of the National Academy of Sciences of the United States of America, 111(29), pp. 10422–10426. Available at: https://doi.org/10.1073/pnas.1402202111.

Pitcher, T.J., Partridge, B.L. and Wardle, C.S. (1976) ‘A blind fish can school’, Science, 194(4268), pp. 963–965. Available at: https://doi.org/10.1126/science.982056.

Rosenthal, S.B., Twomey, C.R., Hartnett, A.T., Wu, H.S., Couzin, I.D., et al. (2015) ‘Revealing the hidden networks of interaction in mobile animal groups allows prediction of complex behavioral contagion’, Proceedings of the National Academy of Sciences of the United States of America, 112(15), pp. 4690–4695. Available at: https://doi.org/10.1073/pnas.1420068112.

Rosenthal, S.B., Twomey, C.R., Hartnett, A.T., Wu, H.S. and Couzin, I.D. (2015) ‘Revealing the hidden networks of interaction in mobile animal groups allows prediction of complex behavioral contagion’, Proceedings of the National Academy of Sciences of the United States of America, 112(15), pp. 4690–4695. Available at: https://doi.org/10.1073/PNAS.1420068112/-/DCSUPPLEMENTAL/PNAS.1420068112.SAPP.PDF.

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

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

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

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

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

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

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

Simmons, J.A. and Kick, S.A. (1983) ‘Interception of Flying Insects by Bats’, Neuroethology and Behavioral Physiology, pp. 267–279. Available at: https://doi.org/10.1007/978-3-642-69271-0_20.

Strandburg-Peshkin, A. et al. (2013) ‘Visual sensory networks and effective information transfer in animal groups’, Current Biology. Cell Press. Available at: https://doi.org/10.1016/j.cub.2013.07.059.

Sumpter, D.J.T. et al. (2008) ‘Consensus Decision Making by Fish’, Current Biology, 18(22), pp. 1773–1777. Available at: https://doi.org/10.1016/J.CUB.2008.09.064.

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

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

Ulanovsky, N. and Moss, C.F. (2008) ‘What the bat’s voice tells the bat’s brain’, Proceedings of the National Academy of Sciences of the United States of America, 105(25), pp. 8491–8498. Available at: https://doi.org/10.1073/pnas.0703550105.

Vanderelst, D. and Peremans, H. (2018) ‘Modeling bat prey capture in echolocating bats : The feasibility of reactive pursuit’, Journal of theoretical biology, 456, pp. 305–314.

Youssefi, K.A.R. and Rouhani, M. (2021) ‘Swarm intelligence based robotic search in unknown maze-like environments’, Expert Systems with Applications, 178. Available at: https://doi.org/10.1016/j.eswa.2021.114907.

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

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

Author response:

We thank the reviewers for their thorough evaluation and constructive feedback on our manuscript.

We think that their valuable suggestions will strengthen the manuscript and help us clarify several important points.

All reviewers acknowledged the importance of our theoretical results and network classification in making pattern formation analysis a more tractable problem. At the same time, they have also raised a number of important concerns that we shall carefully consider.

A. A major clarification that the reviewers found important concerns the definition of non-trivial pattern transformations and its generalization to higher dimensions. In this regard, the reviewers’ comments are:

Reviewer #1:

(on non-trivial pattern transformations):

(3) All modelling is confined to one spatial dimension, and the very definition of a "non-trivial" transformation is framed in terms of peak positions along a line, which clearly must be reformulated for higher dimensions. It's well-known that diffusions in 1, 2, and 3 dimensions are also dramatically different, so the relevance of the three-class taxonomy to real multicellular tissues remains unclear, or at least should be explained in more detail. Reviewer #2 (on non-trivial pattern transformations):

(5) The definition of non-trivial pattern formation is provided only in the Supplementary Information, despite its central importance for interpreting the main results. It would significantly improve clarity if this definition were included and explained in the main text. Additionally, it remains unclear how the definition is consistently applied across the different initial conditions. In particular, the authors should clarify how slope-based measures are determined for both the random noise and sharp peak/step function initial states. Furthermore, the authors do not specify how the sign function is evaluated at zero. If the standard mathematical definition sgn(0)=0 is used, then even a simple widening of a peak could fulfill the criterion for nontrivial pattern transformation.

We agree with Reviewer #2 that including a more detailed definition of non-trivial pattern transformation in the main text would enhance the clarity of the paper. The one-dimensional (1D) definition currently provided in the Supplementary Information was chosen because all computations presented therein involve exclusively one-dimensional patterns. However, we acknowledge that this definition, as it was, did not have a totally unambiguous generalization  to higher dimensions. Therefore, in a revised version of the manuscript, we will incorporate an expanded definition applicable to higher-dimensional cases.

This general definition of a non-trivial pattern transformation should make no reference to the sign of spatial derivatives of either the initial or resulting patterns. Specifically, a pattern transformation is considered non-trivial if it satisfies the following criteria:

- It is heterogeneous: The resulting pattern is heterogeneous in space.

- It is rearranging: The arrangement of critical points (i.e. peaks, valleys and saddle points in a gene product concentration) along the domain in the resulting pattern of a gene product is different to the arrangement of critical points in its initial pattern. This includes the emergence of new critical points, the disappearance of existing ones, or the spatial displacement of critical points from one location to another.

- It is non-replicating: The spatial arrangement of critical points in the pattern of one gene product must differ from that of any other upstream gene product.

Nonetheless, our two initial patterns are spatially discontinuous functions: in homogeneous initial patterns, the white noise is discontinuous by definition; and for the spike and spike+homogeneous initial patterns, we use sharp spikes defined by the rectangular function, which is discontinuous at the spike boundaries. Therefore, the aforementioned definition should be supplemented with the following two ad hoc assumptions:

- Homogeneous initial patterns do not comprise any critical point. White noise in this type of initial patterns represents small thermodynamic fluctuations around the steady state and, for the purpose of pattern transformation, this is equivalent to a constant concentration along the domain.

- Spike and spike+homogeneous initial patterns each contain a single critical point located at the center of the spike. The sharp spikes, modeled using the rectangular function, serve as a theoretical idealization to facilitate mathematical analysis. Once diffusion begins to act, these sharp boundaries are smoothed into differentiable gradients, maintaining a unique critical point at the center of the initial spike, which is the most relevant information for pattern transformation.

Finally, it is worth recalling that our gene network classification is fundamentally based on an analysis of the dispersion relation associated with the gene network, and the construction of this dispersion relation is independent of the spatial dimensionality of the domain (i.e. it does not require assuming any specific number of dimensions). The fact that the description of this dispersion relation was in the SI may have been non-ideal for the understandability of the article and will, consequently, be moved to the main text in an upcoming version of the article. Thus, the gene networks that can lead to pattern transformation are the same in 1D, 2D or 3D. As for the resulting patterns, the broad description we provide also applies to any number of dimensions; these would be periodic, non periodic as in the amplified noise patterns or non periodic as in the hierarchic networks. For the latter notice that, except for boundary effects that we later discuss, the spike initial condition is radially symmetric and thus, the patterns resulting from it will also be radially symmetric. We will make this point more explicit in a revised version of the article, especially since, as suggested, this important portion of the Supplementary Information will be incorporated into the main text.

Reviewer 2 suggests that with our definition of non-trivial pattern transformation, the simple widening of a concentration peak would constitute a non-trivial pattern transformation. This is not the case, as already shown in the figures as a example, since in a widening there is no change in the position of the critical point. A different situation applies if a wide and completely flat concentration peak (i.e. a plateau) forms. As we will explain in the coming version this is not possible because of requirement R5.

We think that this clarification of the definition of non-trivial pattern transformation will also help clarify the next point (B below) since it would make it clearer that this article does not intend to explain which specific resulting pattern would arise from any given gene network.

B. The main concern among these relates to the validity of our linearization of the model equations and the extension of the results obtained for the linear system to the fully nonlinear system. In this regard, the reviewers’ comments are:

Reviewer #1:

(on linearization):

(2) A central step in the model formulation is the linearisation of the reaction term around a homogeneous steady state; higher-order kinetics, including ubiquitous bimolecular sinks such as A + B → AB, are simply collapsed into the Jacobian without any stated amplitude bound on the perturbations. Because the manuscript never analyses how far this assumption can be relaxed, the robustness of the three-class taxonomy under realistic nonlinear reactions or large spike amplitudes remains uncertain.

Reviewer #2:

(on linearization):

(2) Most of the proofs presented in the Supplementary Information rely on linearized versions of the governing equations, and it remains unclear how these results extend to the fully nonlinear system. We are concerned that the generality of the conclusions drawn from the linear analysis may be overstated in the main text. For example, in Section S3, the authors introduce the concept of dynamic equivalence of transitive chains (Proposition S3.1) and intracellular transitive M-branching (Proposition S3.2), which pertains to the system's steady-state behavior. However, the proof is based solely on the linearized equations, without additional justification for why the result should hold in the presence of nonlinearities. Moreover, the linearized system is used to analyze the response to a "spike initial pattern of arbitrary height C" (SI Chapter S5.1), yet it is not clear how conclusions derived from the linear regime can be valid for large perturbations, where nonlinear effects are expected to play a significant role. We encourage the authors to clarify the assumptions under which the linearized analysis remains valid and to discuss the potential limitations of applying these results to the nonlinear regime.

In this article, we address two main questions: first, which gene network topologies can give rise to non-trivial pattern transformations; and second, which broad types of resulting patterns can these gene network topologies give rise to resulting pattern. Thus, we are not intending to explain which exact resulting patterns would arise from any given gene network (i.e. a gene network topology with specific functions and interaction strengths or weights), a question for which non-linearities do indeed matter.

For most known gene regulatory networks, available empirical information is typically limited to the nature of gene product regulations -indicating whether they act as activators or inhibitors- while details about the specific functional form of these regulations are rare. For instance, given two gene products, i and j, the network may indicate that i acts as an activator of j, implying that the concentration of j increases with that of i. However, this increase could follow a variety of functional forms: it may be quadratic (e.g., ), cubic (e.g., ), or any other function f j(gi). As we explain in the description of our model, we restrict our study to functions with a monotonicity constraint: higher concentrations of i lead to increased production of j (i.e., ).  In other words, a given gene interaction is always inhibitory or activatory, it does not change of sign. This monotonicity constraint corresponds to requirement (R5) in our main text. This requirement it is based on the biologically plausible idea that the complexity of gene regulation in development stems more from the topology of gene networks than from the complexity of the regulation by which a gene product may regulate another (i.e. we use simple monotonic functions).

Question 1: A critical part to understand question 1 is in the dispersion relation that was explained in SI. From the reviewers’ comments it is clear that having this crucial part in the main text of an upcoming version of the article would improve understandability, specially for question 1.

In brief, any pattern transformation requires the initial pattern to change. The trigger of such change is a change in the concentration of some gene product, either conceptualized as a noise fluctuation (in the homogeneous initial pattern) or a regulated change in a specific point (in the spike initial pattern). Mathematically, both can be conceptualized as perturbations and, for pattern transformation to be possible, such perturbation should grow so that the initial pattern becomes unstable and can change to another resulting pattern.

If the perturbation is small, one can use the standard linear perturbation analysis in S6.2 of our Supplementary Information. In other words, the linear analysis is enough to ascertain if a small perturbation would grow or not. A gene network in which this will not happen would be unable to lead to pattern transformation, whichever the nonlinear part of f(g). In that sense, the linear approximation provides a necessary condition that any gene network needs to fulfill to lead to pattern transformation.

However, the linear analysis would not ascertain whether a specific gene network will actually lead to pattern transformation (i.e., the condition is not sufficient). This, as well as the shape of the specific resulting pattern, may actually depend on the non-linear parts too. As we discuss, based on the dispersion relation, and other complementing arguments along the article, we can also get some insights on the possible patterns from the linear approximation alone (question 2). This arguments hold thanks to the imposition of requirements (R1-R5) on function f(g), which prevent strange behaviors stemming from the nonlinear part of the equation.

The amplitude bound of perturbations mentioned by Reviewer #1 is addressed by requirements (R2) and (R4). Although the solution to the linear system predicts unbounded growth of unstable eigenmodes, the assume functions f(g) on which the nonlinear terms  eventually halt this growth, thereby ensuring the boundedness of solutions as imposed by (R4). This assumption on the nonlinear part is literally requirement R2 on f(g) in the main text.

The transitive chains and branchings in section S3 of the Supplementary Information mentioned by the Reviewer #2 are topological properties of gene networks and therefore they influence only the linear part of the reaction-diffusion equations. This is why the proofs in that section are based on the linearized equations. We agree that clarifying this point in the text, as suggested by the reviewer, would improve the reader’s understanding of the section.

Regarding Reviewer #2’s concerns about large perturbations, we acknowledge that the phrasing using “arbitrary height” may be confusing. For the homogeneous initial conditions these perturbations are assumed to be small because they are actually molecular noise (otherwise the initial condition could not be considered homogenous in the classical sense of developmental biology models). In the spike initial conditions in hierarchic networks the perturbation is not necessarily small. For the analysis provided in the SI we indeed assume that the perturbations are small enough for the linear approximation to be possible. Notice, however, that since these networks require an intracellular self-activating loop upstream of the first extracellular signal, the effective perturbation would rapidly grow to a value determined by such loop.

In general the height of the initial spike does not affect the fact that hierarchic networks can lead to non-trivial pattern transformation. By definition these networks require the secretion of an extracellular signal from the cells in the spike (otherwise no change in gene product concentrations can occur over space). By definition this signal is not produced by any other cells and, thus, its concentration is governed by diffusion from the spike and its production in the cells in the spike. Thus, whichever the initial height of the spike and whichever the non-linearities in f(g), the signal’s concentration would decrease with the distance from the spike. As explained in the main text, this would lead to non-trivial pattern transformations if other general conditions are met. In general, the height of the initial perturbation can affect which specific pattern transformation would arise from a specific gene network but not which gene network topologies can lead to pattern transformation. This will be more clearly stated in an upcoming version of the article. C. In the following, we respond to the remaining concerns raised by the reviewers:

Reviewer #1:

(1) The Results section is difficult to follow. Key logical steps and network configurations are described shortly in prose, which constantly require the reader to address either SI or other parts of the text (see numerous links on the requirements R1-R5 listed at the beginning of the paper) to gain minimal understanding. As a result, a scientifically literate but non-specialist reader may struggle to grasp the argument with a reasonable time invested.

We acknowledge that the current version of the main text may not be as clear as we intended. Initially, we believed that placing the more technical mathematical passages in the Supplementary Information would make the main text more accessible to readers. However, we agree with the reviewer that including some of these computations in the main text could improve clarity. We also believe that adding a summary table outlining all the model’s requirements would further contribute to that goal.

Reviewer #2:

(1) We have serious concerns regarding the validity of the simulation results presented in the manuscript. Rather than simulating the full nonlinear system described by Equation (1), the authors base their results on a truncated expansion (Equation S.8.2) that captures only the time evolution of small deviations around a spatially homogeneous steady state. However, it remains unclear how this reduced system is derived from the full equations specifically, which terms are retained or neglected and why- and how the expansion of the nonlinear function can be steady-state independent, as claimed. Additionally, in simulations involving the spike plus homogeneous initial condition, it is not evident -or, where equations are provided, it is not correct- that the assumed global homogeneous background actually corresponds to a steady state of the full dynamics. We elaborate on these concerns in the following:

We believe there has been a misunderstanding regarding the presentation of the model equations (S8.2) used throughout our simulations. Accordingly, we agree that this relevant section of the Supplementary Information should be rewritten in a revised version of the manuscript to clarify this issue. Below, we address all the concerns raised by the reviewer.

Equation (S8.2) represents the full nonlinear system described in Equation (1). While we recognize that the model may oversimplify real biological processes, its purpose is to illustrate our general statements about pattern formation rather than to capture any specific or detailed mechanism. In this context, model (S8.2) offers three key advantages for our goals: it allows rapid manipulation of gene network topology simply by modifying the matrix J, making it ideal for illustrating pattern formation across different network classes; it accommodates gene networks of arbitrary size -unlike other models, such as the classical Gierer-Meinhardt model, which are limited to two-element Turing or noise-amplifying networks-; and, due to the simplicity of its nonlinear terms, this model involves relatively few free parameters, facilitating the fine-tuning needed to identify parameter regions where non-trivial pattern transformations occur.

Indeed, we find that the ability of model (S8.2) to illustrate our results despite having such simple nonlinear terms -bearing in mind that at least some nonlinearity is always necessary for selforganization- strongly supports the claim that the capacity of a gene network to produce pattern transformations is fully determined by the linear part of Equation (1). In this sense, nonlinear terms primarily influence the precise parameter values at which these transformations occur and contribute to shaping specific features of the resulting patterns.

Model (S8.2) has been successfully employed in pattern formation studies elsewhere in the literature; accordingly, we provide relevant bibliographic references to support its widespread use.

We believe the misunderstanding arises from our explanation of the biological interpretation of the model. As noted in the accompanying bibliography, the model is based on a general reactiondiffusion mechanism assuming the existence of a steady state. However, this conceptual reactiondiffusion framework is not the same as our Equation (1); rather, it was introduced by the original proponents of the model in the seminal paper cited in our text. In this context, Equation (S8.2) describes small concentration perturbations around that steady state, where the variables represent deviations in concentration relative to the general steady state.

The aforementioned general steady state corresponds to the trivial equilibrium point g≡0 in equations (S8.2). Consequently, all our simulations based on model (S8.2) start from this steady state, to which we add white noise to generate homogeneous initial patterns or a sharp spike for the two types of spike initial patterns.

It is also worth noting that Equations (S8.2) represent a non-dimensional model.

It is assumed that the homogeneous steady states are given by g_i=0 and g_i=c_i, where 1/c_i = \mu_i or \hat{\mu}_i, independently of the specific network structure. However, the basis for this assumption is unclear, especially since some of the functions do not satisfy this condition -for example, f5 as defined below Eq. S8.10.5. Moreover, if g_i=c_i does not correspond to a true steady state, then the time evolution of deviations from this state is not correctly described by Eq. S8.2, as the zeroth-order terms do not vanish in that case.

From the explanations above, it is important to distinguish two scales in the process: the scale of small perturbations, where equations (S8.2) apply; and the global scale, where the conceptual general reaction-diffusion system operates. Since the specific form of this general system does not affect equations (S8.2), we assume that it follows any of the models cited in the text, which yield a non-zero steady state at .

In this sense, Equation (S8.2) represent a small concentration deviation of such global system and g(t ,x) is a relative concentration where g≡0 represents the steady-state at are concentrations above , and g<0 are concentrations below .

As previously mentioned, simulations are performed using Equations (S8.2) on the basis of the equilibrium point g≡0. The result of these simulations is then superimposed on the non-zero steady state and presented in the figures along the article.

Using the full model instead of the simplified Equations (S8.2) may result in slightly different resulting patterns, but it does not affect the gene network’s ability to produce pattern transformations, nor does it alter the main structural properties of the patterns—for example, the periodic nature of patterns generated by Turing networks.

Additionally, the equations used contain only linear terms and a cubic degradation term for each species g_i, while neglecting all quadratic terms and cubic terms involving cross-species interactions (i≠j). An explanation for this selective truncation is not provided, and without knowledge of the full equation (f), it is impossible to assess whether this expansion is mathematically justified. If, as suggested in the Supplementary Information, the linear and cubic terms are derived from f, then at the very least, the Jacobian matrix should depend on the background steady-state concentration. However, the equations for the small deviation around a steady state (including the Jacobian matrix) used in the simulations appear to be independent of the particular steady state concentration.

The Jacobian of Equation (S8.2) is independent of g because g represents a small perturbation around a steady state of a general reaction-diffusion system. Consequently, the matrix J corresponds to the Jacobian of the general system evaluated at that steady state. Evaluating the Jacobian of equations (S8.2) at the equilibrium point g≡0 -which represents the general steady state- recovers the matrix J.

This is why we believe that the differences observed between the spike-only initial condition and the spike superimposed on a homogeneous background are not due to the initial conditions themselves, but rather result from a modified reaction scheme introduced through a questionable cutoff.

"In simulations with spike initial patterns, the reference value g≡0 represents an actual concentration of 0 and therefore, we must add to (S8.2) a Heaviside function Φ acting of f (i.e., Φ(f(g))=f(g) if f(g)>0 , Φ(f(g))=0 if f(g){less than or equal to}0 ) to prevent the existence of negative concentrations for any gene product (i.e., g_i<0 for some i )." (SI chapter S8).

This cutoff alters the dynamics (no inhibition) and introduces a different reaction scheme between the two simulations. The need for this correction may itself reflect either a problem in the original equations (which should fulfill the necessary conditions and prevent negative concentrations (R4 in main text)) or the inappropriateness of using an expanded approximation which assumes independence on the steady state concentration. It is already questionable if the linearized equations with a cubic degradation term are valid for the spike initial conditions (with different background concentration values), as the amplitude of this perturbation seems rather large.

For homogeneous and spike+homogeneous initial conditions, we interpret equations (S8.2) as small perturbations around a non-zero steady state of a general reaction-diffusion system. For spike-only initial conditions, that steady state is zero. As we mention before, g≡0 will then represent such steady-state of zero concentration, g>0 are positive concentrations of the general system, and g<0 would represent unfeasible negative concentrations of the general system. Therefore, the use of a cutoff function to handle such initial conditions is justified. Moreover, this cutoff function is the same as the one employed in the reference general system cited in our paper.

We acknowledge that the cutoff influences the simulations and accounts for the differences observed between spike and spike+homogeneous initial conditions. However, this distinction reflects what occurs in real biological systems, which is precisely why we differentiate these two types of initial states. For instance, the emergence of a periodic pattern in a noise-amplifying network depends critically on the formation of regions with concentrations below the steady state near the initial spike. Such regions can form in spike-plus-homogeneous initial patterns but not in spike-only initial patterns, where concentrations below the steady state would correspond to biologically unfeasible negative values.

Lastly, we note that under the current simulation scheme, it is not possible to meaningfully assess criteria RH2a and RH2b, as they rely on nonlinear interactions that are absent from the implemented dynamics.

It is explicitly stated in the relevant subsections of Section S7 in the Supplementary Information that, for the simulations involving RH2a and RH2b, the function f(g) in equation (S8.2) is modified by adding an ad hoc quadratic term to enable the assessment of these criteria.

(3) Several statements in the main text are presented without accompanying proof or sufficient explanation, which makes it difficult to assess their validity. In some cases, the lack of justification raises serious doubts about whether the claims are generally true. Examples are:

"For the purpose of clarity we will explain our results as if these cells have a simple arrangement in space (e.g., a 1D line or a 2D square lattice) but, as we will discuss, our results shall apply with the same logic to any distribution of cells in space." (Main text l.145-l.148).

We believe that the confusion in this statement arises from the ambiguous use of the phrase “our results”. We will revise the text to provide a more precise description. Specifically, by “our results,” we refer to the conclusion that it is possible to determine whether a gene network leads to nontrivial pattern transformations based solely on its topology. This conclusion is independent of the dimensionality of space, as none of our arguments rely on assumptions specific to spatial dimensions. While one-dimensional examples are used for clarity and illustration, the underlying reasoning applies generally. In an improved version of the article, we will clarify this point explicitly and move relevant arguments from the Supplementary Information into the main text.

Critically, our classification of gene networks is ultimately based on an argument concerning the dispersion relation associated with the network, and the construction of this dispersion relation is independent of the spatial dimensionality of the domain. In this sense, the networks identified in the text as capable of producing pattern transformations will be able to generate non-trivial pattern transformations in any spatial domain and in any number of dimensions. While the specific parameter values that permit such transformations may vary depending on the geometry and dimensionality of the domain, the existence of at least one such parameter set remains unaffected.

The geometry of the domain can influence the specific form of the resulting patterns, but it does not alter the broader class of patterns (e.g., periodic patterns, peaks emerging around a spike, etc.) that a given gene network topology can produce. One such geometric influence, commonly observed in simulations, involves boundary effects. For example, structures such as peaks or rings forming near the boundaries may appear higher, broader, or spatially shifted compared to those arising in the central regions of the domain. However, we think a pattern consisting of a periodic train of peaks where only those near the boundary are slightly different can still be classified as a periodic pattern.

"For any non-trivial pattern transformation (as long as it is symmetric around the initial spike), there exists an H gene network capable of producing it from a spike initial pattern." (Main text l.366f).

A justification for this statement is provided shortly after the claim, although we acknowledge that the current explanation is somewhat cumbersome and would benefit from a clearer presentation in a revised version of the main text.

A more detailed justification is provided in the Supplementary Information, based on three key ideas. First, any pattern (provided it is symmetric with respect to the initial spike) can be described as an arrangement of peaks with varying heights and spatial positions along a one-dimensional domain. Second, there exists a simple gene network—the diamond network—that, through parameter tuning, can produce two peaks of arbitrary height and symmetric position relative to the initial spike. Third, by placing multiple diamond networks positively upstream of a common gene product, that gene product can express peaks at each location where the upstream diamond networks induce them. Under mild additional conditions, this mechanism allows the formation of essentially any symmetric pattern. These mild conditions, along with a detailed analysis of the diamond network’s ability to generate peaks with controllable height and position, are discussed in the Supplementary Information.

"In 2D there are no peaks but concentric rings of high gene product concentration centered around the spike, while in 3D there are concentric spherical shells." (Main text l. 447ff).

This result pertains specifically to pattern transformations arising from spike initial patterns. As defined in the text, spike initial patterns are radially symmetric. Since diffusion preserves radial symmetry, pattern transformations from spike initial patterns in two or three dimensions reduce to effectively one-dimensional transformations along each radial direction. In this framework, each pair of concentration peaks symmetric with respect to the spike in one dimension corresponds to a ring surrounding the spike in two dimensions, and each ring in two dimensions becomes a hollow spherical shell around the spike in three dimensions.

We agree that including a brief section in the Supplementary Information to clarify these subtleties would be helpful for readers to better understand the generalization of certain patterns to higher dimensions.

(4) The study identifies one-signal networks and examines how combinations of these structures can give rise to minimal pattern-forming subnetworks. However, the analysis of the combinations of these minimal pattern-forming subnetworks remains relatively brief, and the manuscript does not explore how the results might change if the subnetworks were combined in upstream and downstream configurations. In our view, it is not evident that all possible gene regulatory networks can be fully characterized by these categories, nor that the resulting patterns can be reliably predicted. Rather, the approach appears more suited to identifying which known subnetworks are present within a larger network, without necessarily capturing the full dynamics of more complex configurations.

We acknowledge that our explanation regarding the combination of sub-networks was relatively brief, and we intend to address this in a revised version. Our argument that combining sub-networks does not produce qualitatively new types of pattern transformations -beyond those already described- is based on the dispersion relation. Although this relation was only detailed in the Supplementary Information, it is central to our argument and will therefore be moved to the main text. Below, we provide an outline of this argument:

Our study identifies two distinct behaviors of the principal branch of the dispersion relation at large wavenumbers. Based on this, gene networks capable of pattern formation can be classified into two categories: networks of the first kind, where the real part of the principal branch diverges to infinity as the wavenumber increases; and networks of the second kind, where the real part of the principal branch converges to a positive finite value for large wavenumbers. Naturally this argument applies to any gene network irrespectively of which, or how many, sub-networks are used to built it.

Any gene regulatory network capable of pattern formation falls into one of these two categories. We identified that networks of the first kind contain at least one Turing sub-network, whereas networks of the second kind include either an H sub-network or a noise-amplifying sub-network. In this way, the primary objective of our study -namely, achieving a topological classification of gene regulatory networks capable of pattern formation- is fulfilled. It is important to note that while the dispersion relation provides broad information about the possible resulting patterns a gene network topology can produce (e.g., periodic versus noisy), it does not specify the exact patterns that emerge for each particular set of parameter values.

Finally, regarding the shape of the resulting patterns, Figure S10 in the Supplementary Information exemplifies the notion that the behavior of combined networks can be understood as a combination of the individual behaviors of each constituent sub-network (note that the contribution of each type of sub-network in the resulting pattern is readily distinguishable). Consequently, we focus our detailed analysis on the patterning properties of the fundamental classes.

(6) The manuscript lacks a clear and detailed explanation of the underlying model and its assumptions. In particular, it is not well-defined what constitutes a "cell" in the context of the model, nor is it justified why spatial features of cells -such as their size or boundaries- can be neglected. Furthermore, the concept of the extracellular space in the one-dimensional model remains ambiguous, making it unclear which gene products are assumed to diffuse.

The size of cells is ignored in our model because we assume that they are small enough with respect to the total size of the domain that the space continuous reaction-diffusion equation (equation (1) in the main text) holds. Conceptually, one could understand cells in our model each of the pieces in an even partition of the domain into small subdomains surrounding each position x. This is anyway the standard procedure in most models of pattern formation by reaction-diffusion in embryonic development.

For extracellular signals, we assume that g(t ,x) corresponds to the concentration of the signal in the extracellular space surrounding the cell located at position x. The extracellular space is any fluid medium for which Fick Laws apply and, therfore, the Fickian diffusion term in equation (1) is valid.

For intracellular gene products, we assume that g(t ,x) corresponds to the concentration of such gene product within the cell at position x (if the gene product in hand is a transcription factor, for example), or on its surface (if it is a membrane-bound receptor). When collapsed in the continuous equations there is not such difference between being strictly within the cell or on its boundary. The only important fact is that these gene products cannot diffuse.

Regarding cell boundaries, let us consider an extracellular signal s that regulates a transcriptor factor i within cells (in our model, i is an intracellular gene product). Such regulation shall be mediated by a membrane-bound receptor, which corresponds to intracellular gene product j. In terms of the gene regulatory network this is s→ j→i. Cell boundary effects mentioned by the reviewer should be encapsulated in the specific functional form of the regulation function f(g), but they have no effect in the actual topology of the network. Consequently, they are out of the scope of this study: as we mentioned before, considering different non-linear terms for f(g) will affect the parameter range for which a gene network is capable of producing non-trivial pattern transformations, but not their overall ability to produce non-trivial pattern transformations (i.e., the existence of at least one choice of model parameters for which such transformations take place).

Finally, we would like to once again express our sincere gratitude to all reviewers for their insightful and constructive feedback. We are confident that the thorough peer review process will significantly enhance both the clarity and depth of our work. We greatly value the detailed comments provided and will carefully incorporate them in the preparation of a revised manuscript, which we intend to submit in the coming months.

Author Response

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Given knowledge of the amino acid sequence and of some version of the 3D structure of two monomers that are expected to form a complex, the authors investigate whether it is possible to accurately predict which residues will be in contact in the 3D structure of the expected complex. To this effect, they train a deep learning model that takes as inputs the geometric structures of the individual monomers, per-residue features (PSSMs) extracted from MSAs for each monomer, and rich representations of the amino acid sequences computed with the pre-trained protein language models ESM-1b, MSA Transformer, and ESM-IF. Predicting inter-protein contacts in complexes is an important problem. Multimer variants of AlphaFold, such as AlphaFold-Multimer, are the current state of the art for full protein complex structure prediction, and if the three-dimensional structure of a complex can be accurately predicted then the inter-protein contacts can also be accurately determined. By contrast, the method presented here seeks state-of-the-art performance among models that have been trained end-to-end for inter-protein contact prediction.

Strengths:

The paper is carefully written and the method is very well detailed. The model works both for homodimers and heterodimers. The ablation studies convincingly demonstrate that the chosen model architecture is appropriate for the task. Various comparisons suggest that PLMGraph-Inter performs substantially better, given the same input than DeepHomo, GLINTER, CDPred, DeepHomo2, and DRN-1D2D_Inter. As a byproduct of the analysis, a potentially useful heuristic criterion for acceptable contact prediction quality is found by the authors: namely, to have at least 50% precision in the prediction of the top 50 contacts.

We thank the reviewer for recognizing the strengths of our work!

Weaknesses:

My biggest issue with this work is the evaluations made using bound monomer structures as inputs, coming from the very complexes to be predicted. Conformational changes in protein-protein association are the key element of the binding mechanism and are challenging to predict. While the GLINTER paper (Xie & Xu, 2022) is guilty of the same sin, the authors of CDPred (Guo et al., 2022) correctly only report test results obtained using predicted unbound tertiary structures as inputs to their model. Test results using experimental monomer structures in bound states can hide important limitations in the model, and thus say very little about the realistic use cases in which only the unbound structures (experimental or predicted) are available. I therefore strongly suggest reducing the importance given to the results obtained using bound structures and emphasizing instead those obtained using predicted monomer structures as inputs.

We thank the reviewer for the suggestion! We evaluated PLMGraph-Inter with the predicted monomers and analyzed the result in details (see the “Impact of the monomeric structure quality on contact prediction” section and Figure 3). To mimic the real cases, we even deliberately reduced the performance of AF2 by using reduced MSAs (see the 2nd paragraph in the ““Impact of the monomeric structure quality on contact prediction” section). We leave some of the results in the supplementary of the current manuscript (Table S2). We will move these results to the main text to emphasize the performance of PLMGraph-Inter with the predicted monomers in the revision.

In particular, the most relevant comparison with AlphaFold-Multimer (AFM) is given in Figure S2, not Figure 6. Unfortunately, it substantially shrinks the proportion of structures for which AFM fails while PLMGraph-Inter performs decently. Still, it would be interesting to investigate why this occurs. One possibility would be that the predicted monomer structures are of bad quality there, and PLMGraph-Inter may be able to rely on a signal from its language model features instead. Finally, AFM multimer confidence values ("iptm + ptm") should be provided, especially in the cases in which AFM struggles.

We thank the reviewer for the suggestion! Yes! The performance of PLMGraph-Inter drops when the predicted monomers are used in the prediction. However, it is difficult to say which is a fairer comparison, Figure 6 or Figure S2, since AFM also searched monomer templates (see the third paragraph in 7. Supplementary Information : 7.1 Data in the AlphaFold-Multimer preprint: https://www.biorxiv.org/content/10.1101/2021.10.04.463034v2.full) in the prediction. When we checked our AFM runs, we found that 99% of the targets in our study (including all the targets in the four datasets: HomoPDB, HeteroPDB, DHTest and DB5.5) employed at least 20 templates in their predictions, and 87.8% of the targets employed the native templates. We will provide the AFM confidence values of the AFM predictions in the revision.

Besides, in cases where any experimental structures - bound or unbound - are available and given to PLMGraph-Inter as inputs, they should also be provided to AlphaFold-Multimer (AFM) as templates. Withholding these from AFM only makes the comparison artificially unfair. Hence, a new test should be run using AFM templates, and a new version of Figure 6 should be produced. Additionally, AFM's mean precision, at least for top-50 contact prediction, should be reported so it can be compared with PLMGraph-Inter's.

We thank the reviewers for the suggestion! We would like to notify that AFM also searched monomer templates (see the third paragraph in 7. Supplementary Information : 7.1 Data in the AlphaFold-Multimer preprint: https://www.biorxiv.org/content/10.1101/2021.10.04.463034v2.full) in the prediction. When we checked our AFM runs, we found that 99% of the targets in our study (including all the targets in the four datasets: HomoPDB, HeteroPDB, DHTest and DB5.5) employed at least 20 templates in their predictions, and 87.8% of the targets employed the native template.

It's a shame that many of the structures used in the comparison with AFM are actually in the AFM v2 training set. If there are any outside the AFM v2 training set and, ideally, not sequence- or structure-homologous to anything in the AFM v2 training set, they should be discussed and reported on separately. In addition, why not test on structures from the "Benchmark 2" or "Recent-PDB-Multimers" datasets used in the AFM paper?

We thank the reviewer for the suggestion! The biggest challenge to objectively evaluate AFM is that as far as we known, AFM does not release the PDB ids of its training set and the “Recent-PDB-Multimers” dataset. “Benchmark 2” only includes 17 heterodimer proteins, and the number can be further decreased after removing targets redundant to our training set. We think it is difficult to draw conclusions from such a small number of targets. In the revision, we will analyze the performance of AFM on targets released after the date cutoff of the AFM training set, but with which we cannot totally remove the redundancy between the training and the test sets of AFM.

It is also worth noting that the AFM v2 weights have now been outdated for a while, and better v3 weights now exist, with a training cutoff of 2021-09-30.

We thank the reviewer for reminding the new version of AFM. The only difference between AFM V3 and V2 is the cutoff date of the training set. Our test set would have more overlaps with the training set of AFM V3, which is one reason that we think AFM V2 is more appropriate to be used in the comparison.

Another weakness in the evaluation framework: because PLMGraph-Inter uses structural inputs, it is not sufficient to make its test set non-redundant in sequence to its training set. It must also be non-redundant in structure. The Benchmark 2 dataset mentioned above is an example of a test set constructed by removing structures with homologous templates in the AF2 training set. Something similar should be done here.

We agree with the reviewer that testing whether the model can keep its performance on targets with no templates (i.e. non-redundant in structure) is important. We will perform the analysis in the revision.

Finally, the performance of DRN-1D2D for top-50 precision reported in Table 1 suggests to me that, in an ablation study, language model features alone would yield better performance than geometric features alone. So, I am puzzled why model "a" in the ablation is a "geometry-only" model and not a "LM-only" one.

Using the protein geometric graph to integrate multiple protein language models is the main idea of PLMGraph-Inter. Comparing with our previous work (DRN-1D2D_Inter), we consider the building of the geometric graph as one major contribution of this work. To emphasize the efficacy of this geometric graph, we chose to use the “geometry-only” model as the base model. We will further clarity this in the revision.

Reviewer #2 (Public Review):

This work introduces PLMGraph-Inter, a new deep-learning approach for predicting inter-protein contacts, which is crucial for understanding protein-protein interactions. Despite advancements in this field, especially driven by AlphaFold, prediction accuracy and efficiency in terms of computational cost) still remains an area for improvement. PLMGraph-Inter utilizes invariant geometric graphs to integrate the features from multiple protein language models into the structural information of each subunit. When compared against other inter-protein contact prediction methods, PLMGraph-Inter shows better performance which indicates that utilizing both sequence embeddings and structural embeddings is important to achieve high-accuracy predictions with relatively smaller computational costs for the model training.

The conclusions of this paper are mostly well supported by data, but test examples should be revisited with a more strict sequence identity cutoff to avoid any potential information leakage from the training data. The main figures should be improved to make them easier to understand.

We thank the reviewer for recognizing the significance of our work! We will revise the manuscript carefully to address the reviewer’s concerns.

1. The sequence identity cutoff to remove redundancies between training and test set was set to 40%, which is a bit high to remove test examples having homology to training examples. For example, CDPred uses a sequence identity cutoff of 30% to strictly remove redundancies between training and test set examples. To make their results more solid, the authors should have curated test examples with lower sequence identity cutoffs, or have provided the performance changes against sequence identities to the closest training examples.

We thank the reviewer for the valuable suggestion! Using different thresholds to reduce the redundancy between the test set and the training set is a very good suggestion, and we will perform the analysis in the revision. In the current version of the manuscript, the 40% sequence identity is used as the cutoff for many previous studies used this cutoff (e.g. the Recent-PDB-Multimers used in AlphaFold-Multimer (see: 7.8 Datasets in the AlphaFold-Multimer paper); the work of DSCRIPT: https://www.cell.com/action/showPdf?pii=S2405-4712%2821%2900333-1 (see: the PPI dataset paragraph in the METHODS DETAILS section of the STAR METHODS)). One reason for using the relatively higher threshold for PPI studies is that PPIs are generally not as conserved as protein monomers.

We performed a preliminary analysis using different thresholds to remove redundancy when preparing this provisional response letter:

Author response table 1.

Table1. The performance of PLMGraph-Inter on the HomoPDB and HeteroPDB test sets using native structures(AlphaFold2 predicted structures).

Method:

To remove redundancy, we clustered 11096 sequences from the training set and test sets (HomoPDB, HeteroPDB) using MMSeq2 with different sequence identity threshold (40%, 30%, 20%, 10%) (the lowest cutoff for CD-HIT is 40%, so we switched to MMSeq2). Each sequence is then uniquely labeled by the cluster (e.g. cluster 0, cluster 1, …) to which it belongs, from which each PPI can be marked with a pair of clusters (e.g. cluster 0-cluster 1). The PPIs belonging to the same cluster pair (note: cluster n - cluster m and cluster n-cluster m were considered as the same pair) were considered as redundant. For each PPI in the test set, if the pair cluster it belongs to contains the PPI belonging to the training set, we remove that PPI from the test set.

We will perform more detailed analyses in the revised manuscript.

1. Figures with head-to-head comparison scatter plots are hard to understand as scatter plots because too many different methods are abstracted into a single plot with multiple colors. It would be better to provide individual head-to-head scatter plots as supplementary figures, not in the main figure.

We thank the reviewer for the suggestion! We will include the individual head-to-head scatter plots as supplementary figures in the revision.

3) The authors claim that PLMGraph-Inter is complementary to AlphaFold-multimer as it shows better precision for the cases where AlphaFold-multimer fails. To strengthen the point, the qualities of predicted complex structures via protein-protein docking with predicted contacts as restraints should have been compared to those of AlphaFold-multimer structures.

We thank the reviewer for the suggestion! We will add this comparison in the revision.

4) It would be interesting to further analyze whether there is a difference in prediction performance depending on the depth of multiple sequence alignment or the type of complex (antigen-antibody, enzyme-substrates, single species PPI, multiple species PPI, etc).

We thank the reviewer for the suggestion! We will perform such analysis in the revision.

Author response:

eLife Assessment 

This valuable study investigates how the neural representation of individual finger movements changes during the early period of sequence learning. By combining a new method for extracting features from human magnetoencephalography data and decoding analyses, the authors provide incomplete evidence of an early, swift change in the brain regions correlated with sequence learning, including a set of previously unreported frontal cortical regions. The addition of more control analyses to rule out that head movement artefacts influence the findings, and to further explain the proposal of offline contextualization during short rest periods as the basis for improvement performance would strengthen the manuscript. 

We appreciate the Editorial assessment on our paper’s strengths and novelty.  We have implemented additional control analyses to show that neither task-related eye movements nor increasing overlap of finger movements during learning account for our findings, which are that contextualized neural representations in a network of bilateral frontoparietal brain regions actively contribute to skill learning.  Importantly, we carried out additional analyses showing that contextualization develops predominantly during rest intervals.

Public Reviews:

We thank the Reviewers for their comments and suggestions, prompting new analyses and additions that strengthened our report.

Reviewer #1 (Public review): 

Summary: 

This study addresses the issue of rapid skill learning and whether individual sequence elements (here: finger presses) are differentially represented in human MEG data. The authors use a decoding approach to classify individual finger elements and accomplish an accuracy of around 94%. A relevant finding is that the neural representations of individual finger elements dynamically change over the course of learning. This would be highly relevant for any attempts to develop better brain machine interfaces - one now can decode individual elements within a sequence with high precision, but these representations are not static but develop over the course of learning. 

Strengths: The work follows a large body of work from the same group on the behavioural and neural foundations of sequence learning. The behavioural task is well established and neatly designed to allow for tracking learning and how individual sequence elements contribute. The inclusion of short offline rest periods between learning epochs has been influential because it has revealed that a lot, if not most of the gains in behaviour (ie speed of finger movements) occur in these so-called micro-offline rest periods. The authors use a range of new decoding techniques, and exhaustively interrogate their data in different ways, using different decoding approaches. Regardless of the approach, impressively high decoding accuracies are observed, but when using a hybrid approach that combines the MEG data in different ways, the authors observe decoding accuracies of individual sequence elements from the MEG data of up to 94%. 

We have previously showed that neural replay of MEG activity representing the practiced skill correlated with micro-offline gains during rest intervals of early learning, 1 consistent with the recent report that hippocampal ripples during these offline periods predict human motor sequence learning2.  However, decoding accuracy in our earlier work1 needed improvement.  Here, we reported a strategy to improve decoding accuracy that could benefit future studies of neural replay or BCI using MEG.

Weaknesses: 

There are a few concerns which the authors may well be able to resolve. These are not weaknesses as such, but factors that would be helpful to address as these concern potential contributions to the results that one would like to rule out. Regarding the decoding results shown in Figure 2 etc, a concern is that within individual frequency bands, the highest accuracy seems to be within frequencies that match the rate of keypresses. This is a general concern when relating movement to brain activity, so is not specific to decoding as done here. As far as reported, there was no specific restraint to the arm or shoulder, and even then it is conceivable that small head movements would correlate highly with the vigor of individual finger movements. This concern is supported by the highest contribution in decoding accuracy being in middle frontal regions - midline structures that would be specifically sensitive to movement artefacts and don't seem to come to mind as key structures for very simple sequential keypress tasks such as this - and the overall pattern is remarkably symmetrical (despite being a unimanual finger task) and spatially broad. This issue may well be matching the time course of learning, as the vigor and speed of finger presses will also influence the degree to which the arm/shoulder and head move. This is not to say that useful information is contained within either of the frequencies or broadband data. But it raises the question of whether a lot is dominated by movement "artefacts" and one may get a more specific answer if removing any such contributions. 

Reviewer #1 expresses concern that the combination of the low-frequency narrow-band decoder results, and the bilateral middle frontal regions displaying the highest average intra-parcel decoding performance across subjects is suggestive that the decoding results could be driven by head movement or other artefacts.

Head movement artefacts are highly unlikely to contribute meaningfully to our results for the following reasons. First, in addition to ICA denoising, all “recordings were visually inspected and marked to denoise segments containing other large amplitude artifacts due to movements” (see Methods). Second, the response pad was positioned in a manner that minimized wrist, arm or more proximal body movements during the task. Third, while head position was not monitored online for this study, the head was restrained using an inflatable air bladder, and head position was assessed at the beginning and at the end of each recording. Head movement did not exceed 5mm between the beginning and end of each scan for all participants included in the study. Fourth, we agree that despite the steps taken above, it is possible that minor head movements could still contribute to some remaining variance in the MEG data in our study. The Reviewer states a concern that “it is conceivable that small head movements would correlate highly with the vigor of individual finger movements”. However, in order for any such correlations to meaningfully impact decoding performance, such head movements would need to: (A) be consistent and pervasive throughout the recording (which might not be the case if the head movements were related to movement vigor and vigor changed over time); and (B) systematically vary between different finger movements, and also between the same finger movement performed at different sequence locations (see 5-class decoding performance in Figure 4B). The possibility of any head movement artefacts meeting all these conditions is extremely unlikely.

Given the task design, a much more likely confound in our estimation would be the contribution of eye movement artefacts to the decoder performance (an issue appropriately raised by Reviewer #3 in the comments below). Remember from Figure 1A in the manuscript that an asterisk marks the current position in the sequence and is updated at each keypress. Since participants make very few performance errors, the position of the asterisk on the display is highly correlated with the keypress being made in the sequence. Thus, it is possible that if participants are attending to the visual feedback provided on the display, they may move their eyes in a way that is systematically related to the task.  Since we did record eye movements simultaneously with the MEG recordings (EyeLink 1000 Plus; Fs = 600 Hz), we were able to perform a control analysis to address this question. For each keypress event during trials in which no errors occurred (which is the same time-point that the asterisk position is updated), we extracted three features related to eye movements: 1) the gaze position at the time of asterisk position update (or keyDown event), 2) the gaze position 150ms later, and 3) the peak velocity of the eye movement between the two positions. We then constructed a classifier from these features with the aim of predicting the location of the asterisk (ordinal positions 1-5) on the display. As shown in the confusion matrix below (Author response image 1), the classifier failed to perform above chance levels (Overall cross-validated accuracy = 0.21817):

Author response image 1.

Confusion matrix showing that three eye movement features fail to predict asterisk position on the task display above chance levels (Fold 1 test accuracy = 0.21718; Fold 2 test accuracy = 0.22023; Fold 3 test accuracy = 0.21859; Fold 4 test accuracy = 0.22113; Fold 5 test accuracy = 0.21373; Overall cross-validated accuracy = 0.2181). Since the ordinal position of the asterisk on the display is highly correlated with the ordinal position of individual keypresses in the sequence, this analysis provides strong evidence that keypress decoding performance from MEG features is not explained by systematic relationships between finger movement behavior and eye movements (i.e. – behavioral artefacts).

In fact, inspection of the eye position data revealed that a majority of participants on most trials displayed random walk gaze patterns around a center fixation point, indicating that participants did not attend to the asterisk position on the display. This is consistent with intrinsic generation of the action sequence, and congruent with the fact that the display does not provide explicit feedback related to performance. A similar real-world example would be manually inputting a long password into a secure online application. In this case, one intrinsically generates the sequence from memory and receives similar feedback about the password sequence position (also provided as asterisks), which is typically ignored by the user. The minimal participant engagement with the visual task display observed in this study highlights another important point – that the behavior in explicit sequence learning motor tasks is highly generative in nature rather than reactive to stimulus cues as in the serial reaction time task (SRTT).  This is a crucial difference that must be carefully considered when designing investigations and comparing findings across studies.

We observed that initial keypress decoding accuracy was predominantly driven by contralateral primary sensorimotor cortex in the initial practice trials before transitioning to bilateral frontoparietal regions by trials 11 or 12 as performance gains plateaued.  The contribution of contralateral primary sensorimotor areas to early skill learning has been extensively reported in humans and non-human animals. 1,3-5  Similarly, the increased involvement of bilateral frontal and parietal regions to decoding during early skill learning in the non-dominant hand is well known.  Enhanced bilateral activation in both frontal and parietal cortex during skill learning has been extensively reported6-11, and appears to be even more prominent during early fine motor skill learning in the non-dominant hand12,13.  The frontal regions identified in these studies are known to play crucial roles in executive control14, motor planning15, and working memory6,8,16-18 processes, while the same parietal regions are known to integrate multimodal sensory feedback and support visuomotor transformations6,8,16-18, in addition to working memory19. Thus, it is not surprising that these regions increasingly contribute to decoding as subjects internalize the sequential task.  We now include a statement reflecting these considerations in the revised Discussion.

A somewhat related point is this: when combining voxel and parcel space, a concern is whether a degree of circularity may have contributed to the improved accuracy of the combined data, because it seems to use the same MEG signals twice - the voxels most contributing are also those contributing most to a parcel being identified as relevant, as parcels reflect the average of voxels within a boundary. In this context, I struggled to understand the explanation given, ie that the improved accuracy of the hybrid model may be due to "lower spatially resolved whole-brain and higher spatially resolved regional activity patterns".

We strongly disagree with the Reviewer’s assertion that the construction of the hybrid-space decoder is circular. To clarify, the base feature set for the hybrid-space decoder constructed for all participants includes whole-brain spatial patterns of MEG source activity averaged within parcels. As stated in the manuscript, these 148 inter-parcel features reflect “lower spatially resolved whole-brain activity patterns” or global brain dynamics. We then independently test how well spatial patterns of MEG source activity for all voxels distributed within individual parcels can decode keypress actions. Again, the testing of these intra-parcel spatial patterns, intended to capture “higher spatially resolved regional brain activity patterns”, is completely independent from one another and independent from the weighting of individual inter-parcel features. These intra-parcel features could, for example, provide additional information about muscle activation patterns or the task environment. These approximately 1150 intra-parcel voxels (on average, within the total number varying between subjects) are then combined with the 148 inter-parcel features to construct the final hybrid-space decoder. In fact, this varied spatial filter approach shares some similarities to the construction of convolutional neural networks (CNNs) used to perform object recognition in image classification applications. One could also view this hybrid-space decoding approach as a spatial analogue to common time-frequency based analyses such as theta-gamma phase amplitude coupling (PAC), which combine information from two or more narrow-band spectral features derived from the same time-series data.

We directly tested this hypothesis – that spatially overlapping intra- and inter-parcel features portray different information – by constructing an alternative hybrid-space decoder (HybridAlt) that excluded average inter-parcel features which spatially overlapped with intra-parcel voxel features, and comparing the performance to the decoder used in the manuscript (HybridOrig). The prediction was that if the overlapping parcel contained similar information to the more spatially resolved voxel patterns, then removing the parcel features (n=8) from the decoding analysis should not impact performance. In fact, despite making up less than 1% of the overall input feature space, removing those parcels resulted in a significant drop in overall performance greater than 2% (78.15% ± SD 7.03% for HybridOrig vs. 75.49% ± SD 7.17% for HybridAlt; Wilcoxon signed rank test, z = 3.7410, p = 1.8326e-04) (Author response image 2).

Author response image 2.

Comparison of decoding performances with two different hybrid approaches. HybridAlt: Intra-parcel voxel-space features of top ranked parcels and inter-parcel features of remaining parcels. HybridOrig:  Voxel-space features of top ranked parcels and whole-brain parcel-space features (i.e. – the version used in the manuscript). Dots represent decoding accuracy for individual subjects. Dashed lines indicate the trend in performance change across participants. Note, that HybridOrig (the approach used in our manuscript) significantly outperforms the HybridAlt approach, indicating that the excluded parcel features provide unique information compared to the spatially overlapping intra-parcel voxel patterns.

Firstly, there will be a relatively high degree of spatial contiguity among voxels because of the nature of the signal measured, i.e. nearby individual voxels are unlikely to be independent. Secondly, the voxel data gives a somewhat misleading sense of precision; the inversion can be set up to give an estimate for each voxel, but there will not just be dependence among adjacent voxels, but also substantial variation in the sensitivity and confidence with which activity can be projected to different parts of the brain. Midline and deeper structures come to mind, where the inversion will be more problematic than for regions along the dorsal convexity of the brain, and a concern is that in those midline structures, the highest decoding accuracy is seen. 

We definitely agree with the Reviewer that some inter-parcel features representing neighboring (or spatially contiguous) voxels are likely to be correlated. This has been well documented in the MEG literature20,21 and is a particularly important confound to address in functional or effective connectivity analyses (not performed in the present study). In the present analysis, any correlation between adjacent voxels presents a multi-collinearity problem, which effectively reduces the dimensionality of the input feature space. However, as long as there are multiple groups of correlated voxels within each parcel (i.e. - the effective dimensionality is still greater than 1), the intra-parcel spatial patterns could still meaningfully contribute to the decoder performance. Two specific results support this assertion.

First, we obtained higher decoding accuracy with voxel-space features [74.51% (± SD 7.34%)] compared to parcel space features [68.77% (± SD 7.6%)] (Figure 3B), indicating individual voxels carry more information in decoding the keypresses than the averaged voxel-space features or parcel-space features.  Second, Individual voxels within a parcel showed varying feature importance scores in decoding keypresses (Author response image 3). This finding supports the Reviewer’s assertion that neighboring voxels express similar information, but also shows that the correlated voxels form mini subclusters that are much smaller spatially than the parcel they reside in.

Author response image 3.

Feature importance score of individual voxels in decoding keypresses: MRMR was used to rank the individual voxel space features in decoding keypresses and the min-max normalized MRMR score was mapped to a structural brain surface. Note that individual voxels within a parcel showed different contribution to decoding.

Some of these concerns could be addressed by recording head movement (with enough precision) to regress out these contributions. The authors state that head movement was monitored with 3 fiducials, and their time courses ought to provide a way to deal with this issue. The ICA procedure may not have sufficiently dealt with removing movement-related problems, but one could eg relate individual components that were identified to the keypresses as another means for checking. An alternative could be to focus on frequency ranges above the movement frequencies. The accuracy for those still seems impressive and may provide a slightly more biologically plausible assessment. 

We have already addressed the issue of movement related artefacts in the first response above. With respect to a focus on frequency ranges above movement frequencies, the Reviewer states the “accuracy for those still seems impressive and may provide a slightly more biologically plausible assessment”. First, it is important to note that cortical delta-band oscillations measured with local field potentials (LFPs) in macaques is known to contain important information related to end-effector kinematics22,23 muscle activation patterns24 and temporal sequencing25 during skilled reaching and grasping actions. Thus, there is a substantial body of evidence that low-frequency neural oscillatory activity in this range contains important information about the skill learning behavior investigated in the present study. Second, our own data shows (which the Reviewer also points out) that significant information related to the skill learning behavior is also present in higher frequency bands (see Figure 2A and Figure 3—figure supplement 1). As we pointed out in our earlier response to questions about the hybrid space decoder architecture (see above), it is likely that different, yet complimentary, information is encoded across different temporal frequencies (just as it is encoded across different spatial frequencies). Again, this interpretation is supported by our data as the highest performing classifiers in all cases (when holding all parameters constant) were always constructed from broadband input MEG data (Figure 2A and Figure 3—figure supplement 1).  

One question concerns the interpretation of the results shown in Figure 4. They imply that during the course of learning, entirely different brain networks underpin the behaviour. Not only that, but they also include regions that would seem rather unexpected to be key nodes for learning and expressing relatively simple finger sequences, such as here. What then is the biological plausibility of these results? The authors seem to circumnavigate this issue by moving into a distance metric that captures the (neural network) changes over the course of learning, but the discussion seems detached from which regions are actually involved; or they offer a rather broad discussion of the anatomical regions identified here, eg in the context of LFOs, where they merely refer to "frontoparietal regions". 

The Reviewer notes the shift in brain networks driving keypress decoding performance between trials 1, 11 and 36 as shown in Figure 4A. The Reviewer questions whether these substantial shifts in brain network states underpinning the skill are biologically plausible, as well as the likelihood that bilateral superior and middle frontal and parietal cortex are important nodes within these networks.

First, previous fMRI work in humans performing a similar sequence learning task showed that flexibility in brain network composition (i.e. – changes in brain region members displaying coordinated activity) is up-regulated in novel learning environments and explains differences in learning rates across individuals26.  This work supports our interpretation of the present study data, that brain networks engaged in sequential motor skills rapidly reconfigure during early learning.

Second, frontoparietal network activity is known to support motor memory encoding during early learning27,28. For example, reactivation events in the posterior parietal29 and medial prefrontal30,31 cortex (MPFC) have been temporally linked to hippocampal replay, and are posited to support memory consolidation across several memory domains32, including motor sequence learning1,33,34.  Further, synchronized interactions between MPFC and hippocampus are more prominent during early learning as opposed to later stages27,35,36, perhaps reflecting “redistribution of hippocampal memories to MPFC” 27.  MPFC contributes to very early memory formation by learning association between contexts, locations, events and adaptive responses during rapid learning37. Consistently, coupling between hippocampus and MPFC has been shown during, and importantly immediately following (rest) initial memory encoding38,39.  Importantly, MPFC activity during initial memory encoding predicts subsequent recall40. Thus, the spatial map required to encode a motor sequence memory may be “built under the supervision of the prefrontal cortex” 28, also engaged in the development of an abstract representation of the sequence41.  In more abstract terms, the prefrontal, premotor and parietal cortices support novice performance “by deploying attentional and control processes” 42-44 required during early learning42-44. The dorsolateral prefrontal cortex DLPFC specifically is thought to engage in goal selection and sequence monitoring during early skill practice45, all consistent with the schema model of declarative memory in which prefrontal cortices play an important role in encoding46,47.  Thus, several prefrontal and frontoparietal regions contributing to long term learning 48 are also engaged in early stages of encoding. Altogether, there is strong biological support for the involvement of bilateral prefrontal and frontoparietal regions to decoding during early skill learning.  We now address this issue in the revised manuscript.

If I understand correctly, the offline neural representation analysis is in essence the comparison of the last keypress vs the first keypress of the next sequence. In that sense, the activity during offline rest periods is actually not considered. This makes the nomenclature somewhat confusing. While it matches the behavioural analysis, having only key presses one can't do it in any other way, but here the authors actually do have recordings of brain activity during offline rest. So at the very least calling it offline neural representation is misleading to this reviewer because what is compared is activity during the last and during the next keypress, not activity during offline periods. But it also seems a missed opportunity - the authors argue that most of the relevant learning occurs during offline rest periods, yet there is no attempt to actually test whether activity during this period can be useful for the questions at hand here. 

We agree with the Reviewer that our previous “offline neural representation” nomenclature could be misinterpreted. In the revised manuscript we refer to this difference as the “offline neural representational change”. Please, note that our previous work did link offline neural activity (i.e. – 16-22 Hz beta power and neural replay density during inter-practice rest periods) to observed micro-offline gains49.

Reviewer #2 (Public review): 

Summary 

Dash et al. asked whether and how the neural representation of individual finger movements is "contextualized" within a trained sequence during the very early period of sequential skill learning by using decoding of MEG signal. Specifically, they assessed whether/how the same finger presses (pressing index finger) embedded in the different ordinal positions of a practiced sequence (4-1-3-2-4; here, the numbers 1 through 4 correspond to the little through the index fingers of the non-dominant left hand) change their representation (MEG feature). They did this by computing either the decoding accuracy of the index finger at the ordinal positions 1 vs. 5 (index_OP1 vs index_OP5) or pattern distance between index_OP1 vs. index_OP5 at each training trial and found that both the decoding accuracy and the pattern distance progressively increase over the course of learning trials. More interestingly, they also computed the pattern distance for index_OP5 for the last execution of a practice trial vs. index_OP1 for the first execution in the next practice trial (i.e., across the rest period). This "off-line" distance was significantly larger than the "on-line" distance, which was computed within practice trials and predicted micro-offline skill gain. Based on these results, the authors conclude that the differentiation of representation for the identical movement embedded in different positions of a sequential skill ("contextualization") primarily occurs during early skill learning, especially during rest, consistent with the recent theory of the "micro-offline learning" proposed by the authors' group. I think this is an important and timely topic for the field of motor learning and beyond. <br /> Strengths 

The specific strengths of the current work are as follows. First, the use of temporally rich neural information (MEG signal) has a large advantage over previous studies testing sequential representations using fMRI. This allowed the authors to examine the earliest period (= the first few minutes of training) of skill learning with finer temporal resolution. Second, through the optimization of MEG feature extraction, the current study achieved extremely high decoding accuracy (approx. 94%) compared to previous works. As claimed by the authors, this is one of the strengths of the paper (but see my comments). Third, although some potential refinement might be needed, comparing "online" and "offline" pattern distance is a neat idea. 

Weaknesses 

Along with the strengths I raised above, the paper has some weaknesses. First, the pursuit of high decoding accuracy, especially the choice of time points and window length (i.e., 200 msec window starting from 0 msec from key press onset), casts a shadow on the interpretation of the main result. Currently, it is unclear whether the decoding results simply reflect behavioral change or true underlying neural change. As shown in the behavioral data, the key press speed reached 3~4 presses per second already at around the end of the early learning period (11th trial), which means inter-press intervals become as short as 250-330 msec. Thus, in almost more than 60% of training period data, the time window for MEG feature extraction (200 msec) spans around 60% of the inter-press intervals. Considering that the preparation/cueing of subsequent presses starts ahead of the actual press (e.g., Kornysheva et al., 2019) and/or potential online planning (e.g., Ariani and Diedrichsen, 2019), the decoder likely has captured these future press information as well as the signal related to the current key press, independent of the formation of genuine sequential representation (e.g., "contextualization" of individual press). This may also explain the gradual increase in decoding accuracy or pattern distance between index_OP1 vs. index_OP5 (Figure 4C and 5A), which co-occurred with performance improvement, as shorter inter-press intervals are more favorable for the dissociating the two index finger presses followed by different finger presses. The compromised decoding accuracies for the control sequences can be explained in similar logic. Therefore, more careful consideration and elaborated discussion seem necessary when trying to both achieve high-performance decoding and assess early skill learning, as it can impact all the subsequent analyses.

The Reviewer raises the possibility that (given the windowing parameters used in the present study) an increase in “contextualization” with learning could simply reflect faster typing speeds as opposed to an actual change in the underlying neural representation. The issue can essentially be framed as a mixing problem. As correct sequences are generated at higher and higher speeds over training, MEG activity patterns related to the planning, execution, evaluation and memory of individual keypresses overlap more in time. Thus, increased overlap between the “4” and “1” keypresses (at the start of the sequence) and “2” and “4” keypresses (at the end of the sequence) could artefactually increase contextualization distances even if the underlying neural representations for the individual keypresses remain unchanged (assuming this mixing of representations is used by the classifier to differentially tag each index finger press). If this were the case, it follows that such mixing effects reflecting the ordinal sequence structure would also be observable in the distribution of decoder misclassifications. For example, “4” keypresses would be more likely to be misclassified as “1” or “2” keypresses (or vice versa) than as “3” keypresses. The confusion matrices presented in Figures 3C and 4B and Figure 3—figure supplement 3A in the previously submitted manuscript do not show this trend in the distribution of misclassifications across the four fingers.

Moreover, if the representation distance is largely driven by this mixing effect, it’s also possible that the increased overlap between consecutive index finger keypresses during the 4-4 transition marking the end of one sequence and the beginning of the next one could actually mask contextualization-related changes to the underlying neural representations and make them harder to detect. In this case, a decoder tasked with separating individual index finger keypresses into two distinct classes based upon sequence position might show decreased performance with learning as adjacent keypresses overlapped in time with each other to an increasing extent. However, Figure 4C in our previously submitted manuscript does not support this possibility, as the 2-class hybrid classifier displays improved classification performance over early practice trials despite greater temporal overlap.

We also conducted a new multivariate regression analysis to directly assess whether the neural representation distance score could be predicted by the 4-1, 2-4 and 4-4 keypress transition times observed for each complete correct sequence (both predictor and response variables were z-score normalized within-subject). The results of this analysis affirmed that the possible alternative explanation put forward by the Reviewer is not supported by our data (Adjusted R2 = 0.00431; F = 5.62). We now include this new negative control analysis result in the revised manuscript.

Overall, we do strongly agree with the Reviewer that the naturalistic, self-paced, generative task employed in the present study results in overlapping brain processes related to planning, execution, evaluation and memory of the action sequence. We also agree that there are several tradeoffs to consider in the construction of the classifiers depending on the study aim. Given our aim of optimizing keypress decoder accuracy in the present study, the set of trade-offs resulted in representations reflecting more the latter three processes, and less so the planning component. Whether separate decoders can be constructed to tease apart the representations or networks supporting these overlapping processes is an important future direction of research in this area. For example, work presently underway in our lab constrains the selection of windowing parameters in a manner that allows individual classifiers to be temporally linked to specific planning, execution, evaluation or memory-related processes to discern which brain networks are involved and how they adaptively reorganize with learning. Results from the present study (Figure 4—figure supplement 2) showing hybrid-space decoder prediction accuracies exceeding 74% for temporal windows spanning as little as 25ms and located up to 100ms prior to the keyDown event strongly support the feasibility of such an approach.

Related to the above point, testing only one particular sequence (4-1-3-2-4), aside from the control ones, limits the generalizability of the finding. This also may have contributed to the extremely high decoding accuracy reported in the current study. 

The Reviewer raises a question about the generalizability of the decoder accuracy reported in our study. Fortunately, a comparison between decoder performances on Day 1 and Day 2 datasets does provide some insight into this issue. As the Reviewer points out, the classifiers in this study were trained and tested on keypresses performed while practicing a specific sequence (4-1-3-2-4). The study was designed this way as to avoid the impact of interference effects on learning dynamics. The cross-validated performance of classifiers on MEG data collected within the same session was 90.47% overall accuracy (4-class; Figure 3C). We then tested classifier performance on data collected during a separate MEG session conducted approximately 24 hours later (Day 2; see Figure 3—supplement 3). We observed a reduction in overall accuracy rate to 87.11% when tested on MEG data recorded while participants performed the same learned sequence, and 79.44% when they performed several previously unpracticed sequences. Both changes in accuracy are important with regards to the generalizability of our findings. First, 87.11% performance accuracy for the trained sequence data on Day 2 (a reduction of only 3.36%) indicates that the hybrid-space decoder performance is robust over multiple MEG sessions, and thus, robust to variations in SNR across the MEG sensor array caused by small differences in head position between scans.  This indicates a substantial advantage over sensor-space decoding approaches. Furthermore, when tested on data from unpracticed sequences, overall performance dropped an additional 7.67%. This difference reflects the performance bias of the classifier for the trained sequence, possibly caused by high-order sequence structure being incorporated into the feature weights. In the future, it will be important to understand in more detail how random or repeated keypress sequence training data impacts overall decoder performance and generalization. We strongly agree with the Reviewer that the issue of generalizability is extremely important and have added a new paragraph to the Discussion in the revised manuscript highlighting the strengths and weaknesses of our study with respect to this issue.

In terms of clinical BCI, one of the potential relevance of the study, as claimed by the authors, it is not clear that the specific time window chosen in the current study (up to 200 msec since key press onset) is really useful. In most cases, clinical BCI would target neural signals with no overt movement execution due to patients' inability to move (e.g., Hochberg et al., 2012). Given the time window, the surprisingly high performance of the current decoder may result from sensory feedback and/or planning of subsequent movement, which may not always be available in the clinical BCI context. Of course, the decoding accuracy is still much higher than chance even when using signal before the key press (as shown in Figure 4 Supplement 2), but it is not immediately clear to me that the authors relate their high decoding accuracy based on post-movement signal to clinical BCI settings.

The Reviewer questions the relevance of the specific window parameters used in the present study for clinical BCI applications, particularly for paretic patients who are unable to produce finger movements or for whom afferent sensory feedback is no longer intact. We strongly agree with the Reviewer that any intended clinical application must carefully consider these specific input feature constraints dictated by the clinical cohort, and in turn impose appropriate and complimentary constraints on classifier parameters that may differ from the ones used in the present study.  We now highlight this issue in the Discussion of the revised manuscript and relate our present findings to published clinical BCI work within this context.

One of the important and fascinating claims of the current study is that the "contextualization" of individual finger movements in a trained sequence specifically occurs during short rest periods in very early skill learning, echoing the recent theory of micro-offline learning proposed by the authors' group. Here, I think two points need to be clarified. First, the concept of "contextualization" is kept somewhat blurry throughout the text. It is only at the later part of the Discussion (around line #330 on page 13) that some potential mechanism for the "contextualization" is provided as "what-and-where" binding. Still, it is unclear what "contextualization" actually is in the current data, as the MEG signal analyzed is extracted from 0-200 msec after the keypress. If one thinks something is contextualizing an action, that contextualization should come earlier than the action itself. 

The Reviewer requests that we: 1) more clearly define our use of the term “contextualization” and 2) provide the rationale for assessing it over a 200ms window aligned to the keyDown event. This choice of window parameters means that the MEG activity used in our analysis was coincident with, rather than preceding, the actual keypresses.  We define contextualization as the differentiation of representation for the identical movement embedded in different positions of a sequential skill. That is, representations of individual action elements progressively incorporate information about their relationship to the overall sequence structure as the skill is learned. We agree with the Reviewer that this can be appropriately interpreted as “what-and-where” binding. We now incorporate this definition in the Introduction of the revised manuscript as requested.

The window parameters for optimizing accurate decoding individual finger movements were determined using a grid search of the parameter space (a sliding window of variable width between 25-350 ms with 25 ms increments variably aligned from 0 to +100ms with 10ms increments relative to the keyDown event). This approach generated 140 different temporal windows for each keypress for each participant, with the final parameter selection determined through comparison of the resulting performance between each decoder.  Importantly, the decision to optimize for decoding accuracy placed an emphasis on keypress representations characterized by the most consistent and robust features shared across subjects, which in turn maximize statistical power in detecting common learning-related changes. In this case, the optimal window encompassed a 200ms epoch aligned to the keyDown event (t0 = 0 ms).  We then asked if the representations (i.e. – spatial patterns of combined parcel- and voxel-space activity) of the same digit at two different sequence positions changed with practice within this optimal decoding window.  Of course, our findings do not rule out the possibility that contextualization can also be found before or even after this time window, as we did not directly address this issue in the present study.  Ongoing work in our lab, as pointed out above, is investigating contextualization within different time windows tailored specifically for assessing sequence skill action planning, execution, evaluation and memory processes.

The second point is that the result provided by the authors is not yet convincing enough to support the claim that "contextualization" occurs during rest. In the original analysis, the authors presented the statistical significance regarding the correlation between the "offline" pattern differentiation and micro-offline skill gain (Figure 5. Supplement 1), as well as the larger "offline" distance than "online" distance (Figure 5B). However, this analysis looks like regressing two variables (monotonically) increasing as a function of the trial. Although some information in this analysis, such as what the independent/dependent variables were or how individual subjects were treated, was missing in the Methods, getting a statistically significant slope seems unsurprising in such a situation. Also, curiously, the same quantitative evidence was not provided for its "online" counterpart, and the authors only briefly mentioned in the text that there was no significant correlation between them. It may be true looking at the data in Figure 5A as the online representation distance looks less monotonically changing, but the classification accuracy presented in Figure 4C, which should reflect similar representational distance, shows a more monotonic increase up to the 11th trial. Further, the ways the "online" and "offline" representation distance was estimated seem to make them not directly comparable. While the "online" distance was computed using all the correct press data within each 10 sec of execution, the "offline" distance is basically computed by only two presses (i.e., the last index_OP5 vs. the first index_OP1 separated by 10 sec of rest). Theoretically, the distance between the neural activity patterns for temporally closer events tends to be closer than that between the patterns for temporally far-apart events. It would be fairer to use the distance between the first index_OP1 vs. the last index_OP5 within an execution period for "online" distance, as well. 

The Reviewer suggests that the current data is not convincing enough to show that contextualization occurs during rest and raises two important concerns: 1) the relationship between online contextualization and micro-online gains is not shown, and 2) the online distance was calculated differently from its offline counterpart (i.e. - instead of calculating the distance between last IndexOP5 and first IndexOP1 from a single trial, the distance was calculated for each sequence within a trial and then averaged).

We addressed the first concern by performing individual subject correlations between 1) contextualization changes during rest intervals and micro-offline gains; 2) contextualization changes during practice trials and micro-online gains, and 3) contextualization changes during practice trials and micro-offline gains (Author response image 4). We then statistically compared the resulting correlation coefficient distributions and found that within-subject correlations for contextualization changes during rest intervals and micro-offline gains were significantly higher than online contextualization and micro-online gains (t = 3.2827, p = 0.0015) and online contextualization and micro-offline gains (t = 3.7021, p = 5.3013e-04). These results are consistent with our interpretation that micro-offline gains are supported by contextualization changes during the inter-practice rest period.

Author response image 4.

Distribution of individual subject correlation coefficients between contextualization changes occurring during practice or rest with  micro-online and micro-offline performance gains. Note that, the correlation distributions were significantly higher for the relationship between contextualization changes during rest and micro-offline gains than for contextualization changes during practice and either micro-online or offline gain.

With respect to the second concern highlighted above, we agree with the Reviewer that one limitation of the analysis comparing online versus offline changes in contextualization as presented in the reviewed manuscript, is that it does not eliminate the possibility that any differences could simply be explained by the passage of time (which is smaller for the online analysis compared to the offline analysis). The Reviewer suggests an approach that addresses this issue, which we have now carried out.   When quantifying online changes in contextualization from the first IndexOP1 the last IndexOP5 keypress in the same trial we observed no learning-related trend (Author response image 5, right panel). Importantly, offline distances were significantly larger than online distances regardless of the measurement approach and neither predicted online learning (Author response image 6).

Author response image 5.

Trial by trial trend of offline (left panel) and online (middle and right panels) changes in contextualization. Offline changes in contextualization were assessed by calculating the distance between neural representations for the last IndexOP5 keypress in the previous trial and the first IndexOP1 keypress in the present trial. Two different approaches were used to characterize online contextualization changes. The analysis included in the reviewed manuscript (middle panel) calculated the distance between IndexOP1 and IndexOP5 for each correct sequence, which was then averaged across the trial. This approach is limited by the lack of control for the passage of time when making online versus offline comparisons. Thus, the second approach controlled for the passage of time by calculating distance between the representations associated with the first IndexOP1 keypress and the last IndexOP5 keypress within the same trial. Note that while the first approach showed an increase online contextualization trend with practice, the second approach did not.

Author response image 6.

Relationship between online contextualization and online learning is shown for both within-sequence (left; note that this is the online contextualization measure used in the reviewd manuscript) and across-sequence (right) distance calculation. There was no significant relationship between online learning and online contextualization regardless of the measurement approach.

A related concern regarding the control analysis, where individual values for max speed and the degree of online contextualization were compared (Figure 5 Supplement 3), is whether the individual difference is meaningful. If I understood correctly, the optimization of the decoding process (temporal window, feature inclusion/reduction, decoder, etc.) was performed for individual participants, and the same feature extraction was also employed for the analysis of representation distance (i.e., contextualization). If this is the case, the distances are individually differently calculated and they may need to be normalized relative to some stable reference (e.g., 1 vs. 4 or average distance within the control sequence presses) before comparison across the individuals. 

The Reviewer makes a good point here. We have now implemented the suggested normalization procedure in the analysis provided in the revised manuscript.

Reviewer #3 (Public review): 

Summary: 

One goal of this paper is to introduce a new approach for highly accurate decoding of finger movements from human magnetoencephalography data via dimension reduction of a "multi-scale, hybrid" feature space. Following this decoding approach, the authors aim to show that early skill learning involves "contextualization" of the neural coding of individual movements, relative to their position in a sequence of consecutive movements. Furthermore, they aim to show that this "contextualization" develops primarily during short rest periods interspersed with skill training and correlates with a performance metric which the authors interpret as an indicator of offline learning. <br /> Strengths: 

A clear strength of the paper is the innovative decoding approach, which achieves impressive decoding accuracies via dimension reduction of a "multi-scale, hybrid space". This hybrid-space approach follows the neurobiologically plausible idea of the concurrent distribution of neural coding across local circuits as well as large-scale networks. A further strength of the study is the large number of tested dimension reduction techniques and classifiers (though the manuscript reveals little about the comparison of the latter). 

We appreciate the Reviewer’s comments regarding the paper’s strengths.

A simple control analysis based on shuffled class labels could lend further support to this complex decoding approach. As a control analysis that completely rules out any source of overfitting, the authors could test the decoder after shuffling class labels. Following such shuffling, decoding accuracies should drop to chance level for all decoding approaches, including the optimized decoder. This would also provide an estimate of actual chance-level performance (which is informative over and beyond the theoretical chance level). Furthermore, currently, the manuscript does not explain the huge drop in decoding accuracies for the voxel-space decoding (Figure 3B). Finally, the authors' approach to cortical parcellation raises questions regarding the information carried by varying dipole orientations within a parcel (which currently seems to be ignored?) and the implementation of the mean-flipping method (given that there are two dimensions - space and time - what do the authors refer to when they talk about the sign of the "average source", line 477?). 

The Reviewer recommends that we: 1) conduct an additional control analysis on classifier performance using shuffled class labels, 2) provide a more detailed explanation regarding the drop in decoding accuracies for the voxel-space decoding following LDA dimensionality reduction (see Fig 3B), and 3) provide additional details on how problems related to dipole solution orientations were addressed in the present study.  

In relation to the first point, we have now implemented a random shuffling approach as a control for the classification analyses. The results of this analysis indicated that the chance level accuracy was 22.12% (± SD 9.1%) for individual keypress decoding (4-class classification), and 18.41% (± SD 7.4%) for individual sequence item decoding (5-class classification), irrespective of the input feature set or the type of decoder used. Thus, the decoding accuracy observed with the final model was substantially higher than these chance levels.  

Second, please note that the dimensionality of the voxel-space feature set is very high (i.e. – 15684). LDA attempts to map the input features onto a much smaller dimensional space (number of classes-1; e.g. –  3 dimensions, for 4-class keypress decoding). Given the very high dimension of the voxel-space input features in this case, the resulting mapping exhibits reduced accuracy. Despite this general consideration, please refer to Figure 3—figure supplement 3, where we observe improvement in voxel-space decoder performance when utilizing alternative dimensionality reduction techniques.

The decoders constructed in the present study assess the average spatial patterns across time (as defined by the windowing procedure) in the input feature space.  We now provide additional details in the Methods of the revised manuscript pertaining to the parcellation procedure and how the sign ambiguity problem was addressed in our analysis.

Weaknesses: 

A clear weakness of the paper lies in the authors' conclusions regarding "contextualization". Several potential confounds, described below, question the neurobiological implications proposed by the authors and provide a simpler explanation of the results. Furthermore, the paper follows the assumption that short breaks result in offline skill learning, while recent evidence, described below, casts doubt on this assumption. 

We thank the Reviewer for giving us the opportunity to address these issues in detail (see below).

The authors interpret the ordinal position information captured by their decoding approach as a reflection of neural coding dedicated to the local context of a movement (Figure 4). One way to dissociate ordinal position information from information about the moving effectors is to train a classifier on one sequence and test the classifier on other sequences that require the same movements, but in different positions50. In the present study, however, participants trained to repeat a single sequence (4-1-3-2-4). As a result, ordinal position information is potentially confounded by the fixed finger transitions around each of the two critical positions (first and fifth press). Across consecutive correct sequences, the first keypress in a given sequence was always preceded by a movement of the index finger (=last movement of the preceding sequence), and followed by a little finger movement. The last keypress, on the other hand, was always preceded by a ring finger movement, and followed by an index finger movement (=first movement of the next sequence). Figure 4 - Supplement 2 shows that finger identity can be decoded with high accuracy (>70%) across a large time window around the time of the key press, up to at least +/-100 ms (and likely beyond, given that decoding accuracy is still high at the boundaries of the window depicted in that figure). This time window approaches the keypress transition times in this study. Given that distinct finger transitions characterized the first and fifth keypress, the classifier could thus rely on persistent (or "lingering") information from the preceding finger movement, and/or "preparatory" information about the subsequent finger movement, in order to dissociate the first and fifth keypress. Currently, the manuscript provides no evidence that the context information captured by the decoding approach is more than a by-product of temporally extended, and therefore overlapping, but independent neural representations of consecutive keypresses that are executed in close temporal proximity - rather than a neural representation dedicated to context. 

Such temporal overlap of consecutive, independent finger representations may also account for the dynamics of "ordinal coding"/"contextualization", i.e., the increase in 2-class decoding accuracy, across Day 1 (Figure 4C). As learning progresses, both tapping speed and the consistency of keypress transition times increase (Figure 1), i.e., consecutive keypresses are closer in time, and more consistently so. As a result, information related to a given keypress is increasingly overlapping in time with information related to the preceding and subsequent keypresses. The authors seem to argue that their regression analysis in Figure 5 - Figure Supplement 3 speaks against any influence of tapping speed on "ordinal coding" (even though that argument is not made explicitly in the manuscript). However, Figure 5 - Figure Supplement 3 shows inter-individual differences in a between-subject analysis (across trials, as in panel A, or separately for each trial, as in panel B), and, therefore, says little about the within-subject dynamics of "ordinal coding" across the experiment. A regression of trial-by-trial "ordinal coding" on trial-by-trial tapping speed (either within-subject or at a group-level, after averaging across subjects) could address this issue. Given the highly similar dynamics of "ordinal coding" on the one hand (Figure 4C), and tapping speed on the other hand (Figure 1B), I would expect a strong relationship between the two in the suggested within-subject (or group-level) regression. Furthermore, learning should increase the number of (consecutively) correct sequences, and, thus, the consistency of finger transitions. Therefore, the increase in 2-class decoding accuracy may simply reflect an increasing overlap in time of increasingly consistent information from consecutive keypresses, which allows the classifier to dissociate the first and fifth keypress more reliably as learning progresses, simply based on the characteristic finger transitions associated with each. In other words, given that the physical context of a given keypress changes as learning progresses - keypresses move closer together in time and are more consistently correct - it seems problematic to conclude that the mental representation of that context changes. To draw that conclusion, the physical context should remain stable (or any changes to the physical context should be controlled for). 

The issues raised by Reviewer #3 here are similar to two issues raised by Reviewer #2 above and agree they must both be carefully considered in any evaluation of our findings.

As both Reviewers pointed out, the classifiers in this study were trained and tested on keypresses performed while practicing a specific sequence (4-1-3-2-4). The study was designed this way as to avoid the impact of interference effects on learning dynamics. The cross-validated performance of classifiers on MEG data collected within the same session was 90.47% overall accuracy (4-class; Figure 3C). We then tested classifier performance on data collected during a separate MEG session conducted approximately 24 hours later (Day 2; see Figure 3—supplement 3). We observed a reduction in overall accuracy rate to 87.11% when tested on MEG data recorded while participants performed the same learned sequence, and 79.44% when they performed several previously unpracticed sequences. This classification performance difference of 7.67% when tested on the Day 2 data could reflect the performance bias of the classifier for the trained sequence, possibly caused by mixed information from temporally close keypresses being incorporated into the feature weights.

Along these same lines, both Reviewers also raise the possibility that an increase in “ordinal coding/contextualization” with learning could simply reflect an increase in this mixing effect caused by faster typing speeds as opposed to an actual change in the underlying neural representation. The basic idea is that as correct sequences are generated at higher and higher speeds over training, MEG activity patterns related to the planning, execution, evaluation and memory of individual keypresses overlap more in time. Thus, increased overlap between the “4” and “1” keypresses (at the start of the sequence) and “2” and “4” keypresses (at the end of the sequence) could artefactually increase contextualization distances even if the underlying neural representations for the individual keypresses remain unchanged (assuming this mixing of representations is used by the classifier to differentially tag each index finger press). If this were the case, it follows that such mixing effects reflecting the ordinal sequence structure would also be observable in the distribution of decoder misclassifications. For example, “4” keypresses would be more likely to be misclassified as “1” or “2” keypresses (or vice versa) than as “3” keypresses. The confusion matrices presented in Figures 3C and 4B and Figure 3—figure supplement 3A in the previously submitted manuscript do not show this trend in the distribution of misclassifications across the four fingers.

Following this logic, it’s also possible that if the ordinal coding is largely driven by this mixing effect, the increased overlap between consecutive index finger keypresses during the 4-4 transition marking the end of one sequence and the beginning of the next one could actually mask contextualization-related changes to the underlying neural representations and make them harder to detect. In this case, a decoder tasked with separating individual index finger keypresses into two distinct classes based upon sequence position might show decreased performance with learning as adjacent keypresses overlapped in time with each other to an increasing extent. However, Figure 4C in our previously submitted manuscript does not support this possibility, as the 2-class hybrid classifier displays improved classification performance over early practice trials despite greater temporal overlap.

As noted in the above replay to Reviewer #2, we also conducted a new multivariate regression analysis to directly assess whether the neural representation distance score could be predicted by the 4-1, 2-4 and 4-4 keypress transition times observed for each complete correct sequence (both predictor and response variables were z-score normalized within-subject). The results of this analysis affirmed that the possible alternative explanation put forward by the Reviewer is not supported by our data (Adjusted R2 = 0.00431; F = 5.62). We now include this new negative control analysis result in the revised manuscript.

Finally, the Reviewer hints that one way to address this issue would be to compare MEG responses before and after learning for sequences typed at a fixed speed. However, given that the speed-accuracy trade-off should improve with learning, a comparison between unlearned and learned skill states would dictate that the skill be evaluated at a very low fixed speed. Essentially, such a design presents the problem that the post-training test is evaluating the representation in the unlearned behavioral state that is not representative of the acquired skill. Thus, this approach would not address our experimental question: “do neural representations of the same action performed at different locations within a skill sequence contextually differentiate or remain stable as learning evolves”.

A similar difference in physical context may explain why neural representation distances ("differentiation") differ between rest and practice (Figure 5). The authors define "offline differentiation" by comparing the hybrid space features of the last index finger movement of a trial (ordinal position 5) and the first index finger movement of the next trial (ordinal position 1). However, the latter is not only the first movement in the sequence but also the very first movement in that trial (at least in trials that started with a correct sequence), i.e., not preceded by any recent movement. In contrast, the last index finger of the last correct sequence in the preceding trial includes the characteristic finger transition from the fourth to the fifth movement. Thus, there is more overlapping information arising from the consistent, neighbouring keypresses for the last index finger movement, compared to the first index finger movement of the next trial. A strong difference (larger neural representation distance) between these two movements is, therefore, not surprising, given the task design, and this difference is also expected to increase with learning, given the increase in tapping speed, and the consequent stronger overlap in representations for consecutive keypresses. Furthermore, initiating a new sequence involves pre-planning, while ongoing practice relies on online planning (Ariani et al., eNeuro 2021), i.e., two mental operations that are dissociable at the level of neural representation (Ariani et al., bioRxiv 2023). 

The Reviewer argues that the comparison of last finger movement of a trial and the first in the next trial are performed in different circumstances and contexts. This is an important point and one we tend to agree with. For this task, the first sequence in a practice trial (which is pre-planned offline) is performed in a somewhat different context from the sequence iterations that follow, which involve temporally overlapping planning, execution and evaluation processes.  The Reviewer is particularly concerned about a difference in the temporal mixing effect issue raised above between the first and last keypresses performed in a trial. However, in contrast to the Reviewers stated argument above, findings from Korneysheva et. al (2019) showed that neural representations of individual actions are competitively queued during the pre-planning period in a manner that reflects the ordinal structure of the learned sequence.  Thus, mixing effects are likely still present for the first keypress in a trial. Also note that we now present new control analyses in multiple responses above confirming that hypothetical mixing effects between adjacent keypresses do not explain our reported contextualization finding. A statement addressing these possibilities raised by the Reviewer has been added to the Discussion in the revised manuscript.

In relation to pre-planning, ongoing MEG work in our lab is investigating contextualization within different time windows tailored specifically for assessing how sequence skill action planning evolves with learning.

Given these differences in the physical context and associated mental processes, it is not surprising that "offline differentiation", as defined here, is more pronounced than "online differentiation". For the latter, the authors compared movements that were better matched regarding the presence of consistent preceding and subsequent keypresses (online differentiation was defined as the mean difference between all first vs. last index finger movements during practice).  It is unclear why the authors did not follow a similar definition for "online differentiation" as for "micro-online gains" (and, indeed, a definition that is more consistent with their definition of "offline differentiation"), i.e., the difference between the first index finger movement of the first correct sequence during practice, and the last index finger of the last correct sequence. While these two movements are, again, not matched for the presence of neighbouring keypresses (see the argument above), this mismatch would at least be the same across "offline differentiation" and "online differentiation", so they would be more comparable. 

This is the same point made earlier by Reviewer #2, and we agree with this assessment. As stated in the response to Reviewer #2 above, we have now carried out quantification of online contextualization using this approach and included it in the revised manuscript. We thank the Reviewer for this suggestion.

A further complication in interpreting the results regarding "contextualization" stems from the visual feedback that participants received during the task. Each keypress generated an asterisk shown above the string on the screen, irrespective of whether the keypress was correct or incorrect. As a result, incorrect (e.g., additional, or missing) keypresses could shift the phase of the visual feedback string (of asterisks) relative to the ordinal position of the current movement in the sequence (e.g., the fifth movement in the sequence could coincide with the presentation of any asterisk in the string, from the first to the fifth). Given that more incorrect keypresses are expected at the start of the experiment, compared to later stages, the consistency in visual feedback position, relative to the ordinal position of the movement in the sequence, increased across the experiment. A better differentiation between the first and the fifth movement with learning could, therefore, simply reflect better decoding of the more consistent visual feedback, based either on the feedback-induced brain response, or feedback-induced eye movements (the study did not include eye tracking). It is not clear why the authors introduced this complicated visual feedback in their task, besides consistency with their previous studies.

We strongly agree with the Reviewer that eye movements related to task engagement are important to rule out as a potential driver of the decoding accuracy or contextualization effect. We address this issue above in response to a question raised by Reviewer #1 about the impact of movement related artefacts in general on our findings.

First, the assumption the Reviewer makes here about the distribution of errors in this task is incorrect. On average across subjects, 2.32% ± 1.48% (mean ± SD) of all keypresses performed were errors, which were evenly distributed across the four possible keypress responses. While errors increased progressively over practice trials, they did so in proportion to the increase in correct keypresses, so that the overall ratio of correct-to-incorrect keypresses remained stable over the training session. Thus, the Reviewer’s assumptions that there is a higher relative frequency of errors in early trials, and a resulting systematic trend phase shift differences between the visual display updates (i.e. – a change in asterisk position above the displayed sequence) and the keypress performed is not substantiated by the data. To the contrary, the asterisk position on the display and the keypress being executed remained highly correlated over the entire training session. We now include a statement about the frequency and distribution of errors in the revised manuscript.

Given this high correlation, we firmly agree with the Reviewer that the issue of eye movement-related artefacts is still an important one to address. Fortunately, we did collect eye movement data during the MEG recordings so were able to investigate this. As detailed in the response to Reviewer #1 above, we found that gaze positions and eye-movement velocity time-locked to visual display updates (i.e. – a change in asterisk position above the displayed sequence) did not reflect the asterisk location above chance levels (Overall cross-validated accuracy = 0.21817; see Author response image 1). Furthermore, an inspection of the eye position data revealed that a majority of participants on most trials displayed random walk gaze patterns around a center fixation point, indicating that participants did not attend to the asterisk position on the display. This is consistent with intrinsic generation of the action sequence, and congruent with the fact that the display does not provide explicit feedback related to performance. As pointed out above, a similar real-world example would be manually inputting a long password into a secure online application. In this case, one intrinsically generates the sequence from memory and receives similar feedback about the password sequence position (also provided as asterisks), which is typically ignored by the user. Notably, the minimal participant engagement with the visual task display observed in this study highlights an important difference between behavior observed during explicit sequence learning motor tasks (which is highly generative in nature) with reactive responses to stimulus cues in a serial reaction time task (SRTT).  This is a crucial difference that must be carefully considered when comparing findings across studies. All elements pertaining to this new control analysis are now included in the revised manuscript.

The authors report a significant correlation between "offline differentiation" and cumulative micro-offline gains. However, it would be more informative to correlate trial-by-trial changes in each of the two variables. This would address the question of whether there is a trial-by-trial relation between the degree of "contextualization" and the amount of micro-offline gains - are performance changes (micro-offline gains) less pronounced across rest periods for which the change in "contextualization" is relatively low? Furthermore, is the relationship between micro-offline gains and "offline differentiation" significantly stronger than the relationship between micro-offline gains and "online differentiation"? 

In response to a similar issue raised above by Reviewer #2, we now include new analyses comparing correlation magnitudes between (1) “online differention” vs micro-online gains, (2) “online differention” vs micro-offline gains and (3) “offline differentiation” and micro-offline gains (see Author response images 4, 5 and 6 above). These new analyses and results have been added to the revised manuscript. Once again, we thank both Reviewers for this suggestion.

The authors follow the assumption that micro-offline gains reflect offline learning.

This statement is incorrect. The original Bonstrup et al (2019) 49 paper clearly states that micro-offline gains must be carefully interpreted based upon the behavioral context within which they are observed, and lays out the conditions under which one can have confidence that micro-offline gains reflect offline learning.  In fact, the excellent meta-analysis of Pan & Rickard (2015) 51, which re-interprets the benefits of sleep in overnight skill consolidation from a “reactive inhibition” perspective, was a crucial resource in the experimental design of our initial study49, as well as in all our subsequent work. Pan & Rickard stated:

“Empirically, reactive inhibition refers to performance worsening that can accumulate during a period of continuous training (Hull, 1943). It tends to dissipate, at least in part, when brief breaks are inserted between blocks of training. If there are multiple performance-break cycles over a training session, as in the motor sequence literature, performance can exhibit a scalloped effect, worsening during each uninterrupted performance block but improving across blocks52,53. Rickard, Cai, Rieth, Jones, and Ard (2008) and Brawn, Fenn, Nusbaum, and Margoliash (2010) 52,53 demonstrated highly robust scalloped reactive inhibition effects using the commonly employed 30 s–30 s performance break cycle, as shown for Rickard et al.’s (2008) massed practice sleep group in Figure 2. The scalloped effect is evident for that group after the first few 30 s blocks of each session. The absence of the scalloped effect during the first few blocks of training in the massed group suggests that rapid learning during that period masks any reactive inhibition effect.”

Crucially, Pan & Rickard51 made several concrete recommendations for reducing the impact of the reactive inhibition confound on offline learning studies. One of these recommendations was to reduce practice times to 10s (most prior sequence learning studies up until that point had employed 30s long practice trials). They stated:

“The traditional design involving 30 s-30 s performance break cycles should be abandoned given the evidence that it results in a reactive inhibition confound, and alternative designs with reduced performance duration per block used instead 51. One promising possibility is to switch to 10 s performance durations for each performance-break cycle Instead 51. That design appears sufficient to eliminate at least the majority of the reactive inhibition effect 52,53.”

We mindfully incorporated recommendations from Pan and Rickard51  into our own study designs including 1) utilizing 10s practice trials and 2) constraining our analysis of micro-offline gains to early learning trials (where performance monotonically increases and 95% of overall performance gains occur), which are prior to the emergence of the “scalloped” performance dynamics that are strongly linked to reactive inhibition effects. 

However, there is no direct evidence in the literature that micro-offline gains really result from offline learning, i.e., an improvement in skill level.

We strongly disagree with the Reviewer’s assertion that “there is no direct evidence in the literature that micro-offline gains really result from offline learning, i.e., an improvement in skill level.”  The initial Bönstrup et al. (2019) 49 report was followed up by a large online crowd-sourcing study (Bönstrup et al., 2020) 54. This second (and much larger) study provided several additional important findings supporting our interpretation of micro-offline gains in cases where the important behavioral conditions clarified above were met (see Author response image 7 below for further details on these conditions).

Author response image 7.

Micro-offline gains observed in learning and non-learning contexts are attributed to different underlying causes. (A) Micro-offline and online changes relative to overall trial-by-trial learning. This figure is based on data from Bönstrup et al. (2019) 49. During early learning, micro-offline gains (red bars) closely track trial-by-trial performance gains (green line with open circle markers), with minimal contribution from micro-online gains (blue bars). The stated conclusion in Bönstrup et al. (2019) is that micro-offline gains only during this Early Learning stage reflect rapid memory consolidation (see also 54). After early learning, about practice trial 11, skill plateaus. This plateau skill period is characterized by a striking emergence of coupled (and relatively stable) micro-online drops and micro-offline increases. Bönstrup et al. (2019) as well as others in the literature 55-57, argue that micro-offline gains during the plateau period likely reflect recovery from inhibitory performance factors such as reactive inhibition or fatigue, and thus must be excluded from analyses relating micro-offline gains to skill learning.  The Non-repeating groups in Experiments 3 and 4 from Das et al. (2024) suffer from a lack of consideration of these known confounds.

Evidence documented in that paper54 showed that micro-offline gains during early skill learning were: 1) replicable and generalized to subjects learning the task in their daily living environment (n=389); 2) equivalent when significantly shortening practice period duration, thus confirming that they are not a result of recovery from performance fatigue (n=118);  3) reduced (along with learning rates) by retroactive interference applied immediately after each practice period relative to interference applied after passage of time (n=373), indicating stabilization of the motor memory at a microscale of several seconds consistent with rapid consolidation; and 4) not modified by random termination of the practice periods, ruling out a contribution of predictive motor slowing (N = 71) 54.  Altogether, our findings were strongly consistent with the interpretation that micro-offline gains reflect memory consolidation supporting early skill learning. This is precisely the portion of the learning curve Pan and Rickard51 refer to when they state “…rapid learning during that period masks any reactive inhibition effect”.

This interpretation is further supported by brain imaging evidence linking known memory-related networks and consolidation mechanisms to micro-offline gains. First, we reported that the density of fast hippocampo-neocortical skill memory replay events increases approximately three-fold during early learning inter-practice rest periods with the density explaining differences in the magnitude of micro-offline gains across subjects1. Second, Jacobacci et al. (2020) independently reproduced our original behavioral findings and reported BOLD fMRI changes in the hippocampus and precuneus (regions also identified in our MEG study1) linked to micro-offline gains during early skill learning. 33 These functional changes were coupled with rapid alterations in brain microstructure in the order of minutes, suggesting that the same network that operates during rest periods of early learning undergoes structural plasticity over several minutes following practice58. Third, even more recently, Chen et al. (2024) provided direct evidence from intracranial EEG in humans linking sharp-wave ripple events (which are known markers for neural replay59) in the hippocampus (80-120 Hz in humans) with micro-offline gains during early skill learning. The authors report that the strong increase in ripple rates tracked learning behavior, both across blocks and across participants. The authors conclude that hippocampal ripples during resting offline periods contribute to motor sequence learning. 2

Thus, there is actually now substantial evidence in the literature directly supporting the assertion “that micro-offline gains really result from offline learning”.  On the contrary, according to Gupta & Rickard (2024) “…the mechanism underlying RI [reactive inhibition] is not well established” after over 80 years of investigation60, possibly due to the fact that “reactive inhibition” is a categorical description of behavioral effects that likely result from several heterogenous processes with very different underlying mechanisms.

On the contrary, recent evidence questions this interpretation (Gupta & Rickard, npj Sci Learn 2022; Gupta & Rickard, Sci Rep 2024; Das et al., bioRxiv 2024). Instead, there is evidence that micro-offline gains are transient performance benefits that emerge when participants train with breaks, compared to participants who train without breaks, however, these benefits vanish within seconds after training if both groups of participants perform under comparable conditions (Das et al., bioRxiv 2024). 

It is important to point out that the recent work of Gupta & Rickard (2022,2024) 55 does not present any data that directly opposes our finding that early skill learning49 is expressed as micro-offline gains during rest breaks. These studies are essentially an extension of the Rickard et al (2008) paper that employed a massed (30s practice followed by 30s breaks) vs spaced (10s practice followed by 10s breaks) to assess if recovery from reactive inhibition effects could account for performance gains measured after several minutes or hours. Gupta & Rickard (2022) added two additional groups (30s practice/10s break and 10s practice/10s break as used in the work from our group). The primary aim of the study was to assess whether it was more likely that changes in performance when retested 5 minutes after skill training (consisting of 12 practice trials for the massed groups and 36 practice trials for the spaced groups) had ended reflected memory consolidation effects or recovery from reactive inhibition effects. The Gupta & Rickard (2024) follow-up paper employed a similar design with the primary difference being that participants performed a fixed number of sequences on each trial as opposed to trials lasting a fixed duration. This was done to facilitate the fitting of a quantitative statistical model to the data.  To reiterate, neither study included any analysis of micro-online or micro-offline gains and did not include any comparison focused on skill gains during early learning. Instead, Gupta & Rickard (2022), reported evidence for reactive inhibition effects for all groups over much longer training periods. Again, we reported the same finding for trials following the early learning period in our original Bönstrup et al. (2019) paper49 (Author response image 7). Also, please note that we reported in this paper that cumulative micro-offline gains over early learning did not correlate with overnight offline consolidation measured 24 hours later49 (see the Results section and further elaboration in the Discussion). Thus, while the composition of our data is supportive of a short-term memory consolidation process operating over several seconds during early learning, it likely differs from those involved over longer training times and offline periods, as assessed by Gupta & Rickard (2022).

In the recent preprint from Das et al (2024) 61,  the authors make the strong claim that “micro-offline gains during early learning do not reflect offline learning” which is not supported by their own data.   The authors hypothesize that if “micro-offline gains represent offline learning, participants should reach higher skill levels when training with breaks, compared to training without breaks”.  The study utilizes a spaced vs. massed practice group between-subjects design inspired by the reactive inhibition work from Rickard and others to test this hypothesis. Crucially, the design incorporates only a small fraction of the training used in other investigations to evaluate early skill learning1,33,49,54,57,58,62.  A direct comparison between the practice schedule designs for the spaced and massed groups in Das et al., and the training schedule all participants experienced in the original Bönstrup et al. (2019) paper highlights this issue as well as several others (Author response image 8):

Author response image 8.

(A) Comparison of Das et al. Spaced & Massed group training session designs, and the training session design from the original Bönstrup et al. (2019) 49 paper. Similar to the approach taken by Das et al., all practice is visualized as 10-second practice trials with a variable number (either 0, 1 or 30) of 10-second-long inter-practice rest intervals to allow for direct comparisons between designs. The two key takeaways from this comparison are that (1) the intervention differences (i.e. – practice schedules) between the Massed and Spaced groups from the Das et al. report are extremely small (less than 12% of the overall session schedule) and (2) the overall amount of practice is much less than compared to the design from the original Bönstrup report 49  (which has been utilized in several subsequent studies). (B) Group-level learning curve data from Bönstrup et al. (2019) 49 is used to estimate the performance range accounted for by the equivalent periods covering Test 1, Training 1 and Test 2 from Das et al (2024). Note that the intervention in the Das et al. study is limited to a period covering less than 50% of the overall learning range.

First, participants in the original Bönstrup et al. study 49 experienced 157.14% more practice time and 46.97% less inter-practice rest time than the Spaced group in the Das et al. study (Author response image 8).  Thus, the overall amount of practice and rest differ substantially between studies, with much more limited training occurring for participants in Das et al.  

Second, and perhaps most importantly, the actual intervention (i.e. – the difference in practice schedule between the Spaced and Massed groups) employed by Das et al. covers a very small fraction of the overall training session. Identical practice schedule segments for both the Spaced & Massed groups are indicated by the red shaded area in Author response image 8. Please note that these identical segments cover 94.84% of the Massed group training schedule and 88.01% of the Spaced group training schedule (since it has 60 seconds of additional rest). This means that the actual interventions cover less than 5% (for Massed) and 12% (for Spaced) of the total training session, which minimizes any chance of observing a difference between groups.

Also note that the very beginning of the practice schedule (during which Figure R9 shows substantial learning is known to occur) is labeled in the Das et al. study as Test 1.  Test 1 encompasses the first 20 seconds of practice (alternatively viewed as the first two 10-second-long practice trials with no inter-practice rest). This is immediately followed by the Training 1 intervention, which is composed of only three 10-second-long practice trials (with 10-second inter-practice rest for the Spaced group and no inter-practice rest for the Massed group). Author response image 8 also shows that since there is no inter-practice rest after the third Training practice trial for the Spaced group, this third trial (for both Training 1 and 2) is actually a part of an identical practice schedule segment shared by both groups (Massed and Spaced), reducing the magnitude of the intervention even further.

Moreover, we know from the original Bönstrup et al. (2019) paper49 that 46.57% of all overall group-level performance gains occurred between trials 2 and 5 for that study. Thus, Das et al. are limiting their designed intervention to a period covering less than half of the early learning range discussed in the literature, which again, minimizes any chance of observing an effect.

This issue is amplified even further at Training 2 since skill learning prior to the long 5-minute break is retained, further constraining the performance range over these three trials. A related issue pertains to the trials labeled as Test 1 (trials 1-2) and Test 2 (trials 6-7) by Das et al. Again, we know from the original Bönstrup et al. paper 49 that 18.06% and 14.43% (32.49% total) of all overall group-level performance gains occurred during trials corresponding to Das et al Test 1 and Test 2, respectively. In other words, Das et al averaged skill performance over 20 seconds of practice at two time-points where dramatic skill improvements occur. Pan & Rickard (1995) previously showed that such averaging is known to inject artefacts into analyses of performance gains.

Furthermore, the structure of the Test in Das et. al study appears to have an interference effect on the Spaced group performance after the training intervention.  This makes sense if you consider that the Spaced group is required to now perform the task in a Massed practice environment (i.e., two 10-second-long practice trials merged into one long trial), further blurring the true intervention effects. This effect is observable in Figure 1C,E of their pre-print. Specifically, while the Massed group continues to show an increase in performance during test relative to the last 10 seconds of practice during training, the Spaced group displays a marked decrease. This decrease is in stark contrast to the monotonic increases observed for both groups at all other time-points.

Interestingly, when statistical comparisons between the groups are made at the time-points when the intervention is present (as opposed to after it has been removed) then the stated hypothesis, “If micro-offline gains represent offline learning, participants should reach higher skill levels when training with breaks, compared to training without breaks”, is confirmed.

The data presented by Gupta and Rickard (2022, 2024) and Das et al. (2024) is in many ways more confirmatory of the constraints employed by our group and others with respect to experimental design, analysis and interpretation of study findings, rather than contradictory. Still, it does highlight a limitation of the current micro-online/offline framework, which was originally only intended to be applied to early skill learning over spaced practice schedules when reactive inhibition effects are minimized49. Extrapolation of this current framework to post-plateau performance periods, longer timespans, or non-learning situations (e.g. – the Non-repeating groups from Experiments 3 & 4 in Das et al. (2024)), when reactive inhibition plays a more substantive role, is not warranted. Ultimately, it will be important to develop new paradigms allowing one to independently estimate the different coincident or antagonistic features (e.g. - memory consolidation, planning, working memory and reactive inhibition) contributing to micro-online and micro-offline gains during and after early skill learning within a unifying framework.

References

(1) Buch, E. R., Claudino, L., Quentin, R., Bonstrup, M. & Cohen, L. G. Consolidation of human skill linked to waking hippocampo-neocortical replay. Cell Rep 35, 109193 (2021). https://doi.org:10.1016/j.celrep.2021.109193

(2) Chen, P.-C., Stritzelberger, J., Walther, K., Hamer, H. & Staresina, B. P. Hippocampal ripples during offline periods predict human motor sequence learning. bioRxiv, 2024.2010.2006.614680 (2024). https://doi.org:10.1101/2024.10.06.614680

(3) Classen, J., Liepert, J., Wise, S. P., Hallett, M. & Cohen, L. G. Rapid plasticity of human cortical movement representation induced by practice. J Neurophysiol 79, 1117-1123 (1998).

(4) Karni, A. et al. Functional MRI evidence for adult motor cortex plasticity during motor skill learning. Nature 377, 155-158 (1995). https://doi.org:10.1038/377155a0

(5) Kleim, J. A., Barbay, S. & Nudo, R. J. Functional reorganization of the rat motor cortex following motor skill learning. J Neurophysiol 80, 3321-3325 (1998).

(6) Shadmehr, R. & Holcomb, H. H. Neural correlates of motor memory consolidation. Science 277, 821-824 (1997).

(7) Doyon, J. et al. Experience-dependent changes in cerebellar contributions to motor sequence learning. Proc Natl Acad Sci U S A 99, 1017-1022 (2002).

(8) Toni, I., Ramnani, N., Josephs, O., Ashburner, J. & Passingham, R. E. Learning arbitrary visuomotor associations: temporal dynamic of brain activity. Neuroimage 14, 1048-1057 (2001).

(9) Grafton, S. T. et al. Functional anatomy of human procedural learning determined with regional cerebral blood flow and PET. J Neurosci 12, 2542-2548 (1992).

(10) Kennerley, S. W., Sakai, K. & Rushworth, M. F. Organization of action sequences and the role of the pre-SMA. J Neurophysiol 91, 978-993 (2004). https://doi.org:10.1152/jn.00651.2003 00651.2003 [pii]

(11) Hardwick, R. M., Rottschy, C., Miall, R. C. & Eickhoff, S. B. A quantitative meta-analysis and review of motor learning in the human brain. Neuroimage 67, 283-297 (2013). https://doi.org:10.1016/j.neuroimage.2012.11.020

(12) Sawamura, D. et al. Acquisition of chopstick-operation skills with the non-dominant hand and concomitant changes in brain activity. Sci Rep 9, 20397 (2019). https://doi.org:10.1038/s41598-019-56956-0

(13) Lee, S. H., Jin, S. H. & An, J. The difference in cortical activation pattern for complex motor skills: A functional near- infrared spectroscopy study. Sci Rep 9, 14066 (2019). https://doi.org:10.1038/s41598-019-50644-9

(14) Battaglia-Mayer, A. & Caminiti, R. Corticocortical Systems Underlying High-Order Motor Control. J Neurosci 39, 4404-4421 (2019). https://doi.org:10.1523/JNEUROSCI.2094-18.2019

(15) Toni, I., Thoenissen, D. & Zilles, K. Movement preparation and motor intention. Neuroimage 14, S110-117 (2001). https://doi.org:10.1006/nimg.2001.0841

(16) Wolpert, D. M., Goodbody, S. J. & Husain, M. Maintaining internal representations: the role of the human superior parietal lobe. Nat Neurosci 1, 529-533 (1998). https://doi.org:10.1038/2245

(17) Andersen, R. A. & Buneo, C. A. Intentional maps in posterior parietal cortex. Annu Rev Neurosci 25, 189-220 (2002). https://doi.org:10.1146/annurev.neuro.25.112701.142922 112701.142922 [pii]

(18) Buneo, C. A. & Andersen, R. A. The posterior parietal cortex: sensorimotor interface for the planning and online control of visually guided movements. Neuropsychologia 44, 2594-2606 (2006). https://doi.org:S0028-3932(05)00333-7 [pii] 10.1016/j.neuropsychologia.2005.10.011

(19) Grover, S., Wen, W., Viswanathan, V., Gill, C. T. & Reinhart, R. M. G. Long-lasting, dissociable improvements in working memory and long-term memory in older adults with repetitive neuromodulation. Nat Neurosci 25, 1237-1246 (2022). https://doi.org:10.1038/s41593-022-01132-3

(20) Colclough, G. L. et al. How reliable are MEG resting-state connectivity metrics? Neuroimage 138, 284-293 (2016). https://doi.org:10.1016/j.neuroimage.2016.05.070

(21) Colclough, G. L., Brookes, M. J., Smith, S. M. & Woolrich, M. W. A symmetric multivariate leakage correction for MEG connectomes. NeuroImage 117, 439-448 (2015). https://doi.org:10.1016/j.neuroimage.2015.03.071

(22) Mollazadeh, M. et al. Spatiotemporal variation of multiple neurophysiological signals in the primary motor cortex during dexterous reach-to-grasp movements. J Neurosci 31, 15531-15543 (2011). https://doi.org:10.1523/JNEUROSCI.2999-11.2011

(23) Bansal, A. K., Vargas-Irwin, C. E., Truccolo, W. & Donoghue, J. P. Relationships among low-frequency local field potentials, spiking activity, and three-dimensional reach and grasp kinematics in primary motor and ventral premotor cortices. J Neurophysiol 105, 1603-1619 (2011). https://doi.org:10.1152/jn.00532.2010

(24) Flint, R. D., Ethier, C., Oby, E. R., Miller, L. E. & Slutzky, M. W. Local field potentials allow accurate decoding of muscle activity. J Neurophysiol 108, 18-24 (2012). https://doi.org:10.1152/jn.00832.2011

(25) Churchland, M. M. et al. Neural population dynamics during reaching. Nature 487, 51-56 (2012). https://doi.org:10.1038/nature11129

(26) Bassett, D. S. et al. Dynamic reconfiguration of human brain networks during learning. Proc Natl Acad Sci U S A 108, 7641-7646 (2011). https://doi.org:10.1073/pnas.1018985108

(27) Albouy, G., King, B. R., Maquet, P. & Doyon, J. Hippocampus and striatum: dynamics and interaction during acquisition and sleep-related motor sequence memory consolidation. Hippocampus 23, 985-1004 (2013). https://doi.org:10.1002/hipo.22183

(28) Albouy, G. et al. Neural correlates of performance variability during motor sequence acquisition. Neuroimage 60, 324-331 (2012). https://doi.org:10.1016/j.neuroimage.2011.12.049

(29) Qin, Y. L., McNaughton, B. L., Skaggs, W. E. & Barnes, C. A. Memory reprocessing in corticocortical and hippocampocortical neuronal ensembles. Philos Trans R Soc Lond B Biol Sci 352, 1525-1533 (1997). https://doi.org:10.1098/rstb.1997.0139

(30) Euston, D. R., Tatsuno, M. & McNaughton, B. L. Fast-forward playback of recent memory sequences in prefrontal cortex during sleep. Science 318, 1147-1150 (2007). https://doi.org:10.1126/science.1148979

(31) Molle, M. & Born, J. Hippocampus whispering in deep sleep to prefrontal cortex--for good memories? Neuron 61, 496-498 (2009). https://doi.org:S0896-6273(09)00122-6 [pii] 10.1016/j.neuron.2009.02.002

(32) Frankland, P. W. & Bontempi, B. The organization of recent and remote memories. Nat Rev Neurosci 6, 119-130 (2005). https://doi.org:10.1038/nrn1607

(33) Jacobacci, F. et al. Rapid hippocampal plasticity supports motor sequence learning. Proc Natl Acad Sci U S A 117, 23898-23903 (2020). https://doi.org:10.1073/pnas.2009576117

(34) Albouy, G. et al. Maintaining vs. enhancing motor sequence memories: respective roles of striatal and hippocampal systems. Neuroimage 108, 423-434 (2015). https://doi.org:10.1016/j.neuroimage.2014.12.049

(35) Gais, S. et al. Sleep transforms the cerebral trace of declarative memories. Proc Natl Acad Sci U S A 104, 18778-18783 (2007). https://doi.org:0705454104 [pii] 10.1073/pnas.0705454104

(36) Sterpenich, V. et al. Sleep promotes the neural reorganization of remote emotional memory. J Neurosci 29, 5143-5152 (2009). https://doi.org:10.1523/JNEUROSCI.0561-09.2009

(37) Euston, D. R., Gruber, A. J. & McNaughton, B. L. The role of medial prefrontal cortex in memory and decision making. Neuron 76, 1057-1070 (2012). https://doi.org:10.1016/j.neuron.2012.12.002

(38) van Kesteren, M. T., Fernandez, G., Norris, D. G. & Hermans, E. J. Persistent schema-dependent hippocampal-neocortical connectivity during memory encoding and postencoding rest in humans. Proc Natl Acad Sci U S A 107, 7550-7555 (2010). https://doi.org:10.1073/pnas.0914892107

(39) van Kesteren, M. T., Ruiter, D. J., Fernandez, G. & Henson, R. N. How schema and novelty augment memory formation. Trends Neurosci 35, 211-219 (2012). https://doi.org:10.1016/j.tins.2012.02.001

(40) Wagner, A. D. et al. Building memories: remembering and forgetting of verbal experiences as predicted by brain activity. Science (New York, N.Y.) 281, 1188-1191 (1998).

(41) Ashe, J., Lungu, O. V., Basford, A. T. & Lu, X. Cortical control of motor sequences. Curr Opin Neurobiol 16, 213-221 (2006).

(42) Hikosaka, O., Nakamura, K., Sakai, K. & Nakahara, H. Central mechanisms of motor skill learning. Curr Opin Neurobiol 12, 217-222 (2002).

(43) Penhune, V. B. & Steele, C. J. Parallel contributions of cerebellar, striatal and M1 mechanisms to motor sequence learning. Behav. Brain Res. 226, 579-591 (2012). https://doi.org:10.1016/j.bbr.2011.09.044

(44) Doyon, J. et al. Contributions of the basal ganglia and functionally related brain structures to motor learning. Behavioural brain research 199, 61-75 (2009). https://doi.org:10.1016/j.bbr.2008.11.012

(45) Schendan, H. E., Searl, M. M., Melrose, R. J. & Stern, C. E. An FMRI study of the role of the medial temporal lobe in implicit and explicit sequence learning. Neuron 37, 1013-1025 (2003). https://doi.org:10.1016/s0896-6273(03)00123-5

(46) Morris, R. G. M. Elements of a neurobiological theory of hippocampal function: the role of synaptic plasticity, synaptic tagging and schemas. The European journal of neuroscience 23, 2829-2846 (2006). https://doi.org:10.1111/j.1460-9568.2006.04888.x

(47) Tse, D. et al. Schemas and memory consolidation. Science 316, 76-82 (2007). https://doi.org:10.1126/science.1135935

(48) Berlot, E., Popp, N. J. & Diedrichsen, J. A critical re-evaluation of fMRI signatures of motor sequence learning. Elife 9 (2020). https://doi.org:10.7554/eLife.55241

(49) Bonstrup, M. et al. A Rapid Form of Offline Consolidation in Skill Learning. Curr Biol 29, 1346-1351 e1344 (2019). https://doi.org:10.1016/j.cub.2019.02.049

(50) Kornysheva, K. et al. Neural Competitive Queuing of Ordinal Structure Underlies Skilled Sequential Action. Neuron 101, 1166-1180 e1163 (2019). https://doi.org:10.1016/j.neuron.2019.01.018

(51) Pan, S. C. & Rickard, T. C. Sleep and motor learning: Is there room for consolidation? Psychol Bull 141, 812-834 (2015). https://doi.org:10.1037/bul0000009

(52) Rickard, T. C., Cai, D. J., Rieth, C. A., Jones, J. & Ard, M. C. Sleep does not enhance motor sequence learning. J Exp Psychol Learn Mem Cogn 34, 834-842 (2008). https://doi.org:10.1037/0278-7393.34.4.834

53) Brawn, T. P., Fenn, K. M., Nusbaum, H. C. & Margoliash, D. Consolidating the effects of waking and sleep on motor-sequence learning. J Neurosci 30, 13977-13982 (2010). https://doi.org:10.1523/JNEUROSCI.3295-10.2010

(54) Bonstrup, M., Iturrate, I., Hebart, M. N., Censor, N. & Cohen, L. G. Mechanisms of offline motor learning at a microscale of seconds in large-scale crowdsourced data. NPJ Sci Learn 5, 7 (2020). https://doi.org:10.1038/s41539-020-0066-9

(55) Gupta, M. W. & Rickard, T. C. Dissipation of reactive inhibition is sufficient to explain post-rest improvements in motor sequence learning. NPJ Sci Learn 7, 25 (2022). https://doi.org:10.1038/s41539-022-00140-z

(56) Jacobacci, F. et al. Rapid hippocampal plasticity supports motor sequence learning. Proceedings of the National Academy of Sciences 117, 23898-23903 (2020).

(57) Brooks, E., Wallis, S., Hendrikse, J. & Coxon, J. Micro-consolidation occurs when learning an implicit motor sequence, but is not influenced by HIIT exercise. NPJ Sci Learn 9, 23 (2024). https://doi.org:10.1038/s41539-024-00238-6

(58) Deleglise, A. et al. Human motor sequence learning drives transient changes in network topology and hippocampal connectivity early during memory consolidation. Cereb Cortex 33, 6120-6131 (2023). https://doi.org:10.1093/cercor/bhac489

(59) Buzsaki, G. Hippocampal sharp wave-ripple: A cognitive biomarker for episodic memory and planning. Hippocampus 25, 1073-1188 (2015). https://doi.org:10.1002/hipo.22488

(60) Gupta, M. W. & Rickard, T. C. Comparison of online, offline, and hybrid hypotheses of motor sequence learning using a quantitative model that incorporate reactive inhibition. Sci Rep 14, 4661 (2024). https://doi.org:10.1038/s41598-024-52726-9

(61) Das, A., Karagiorgis, A., Diedrichsen, J., Stenner, M.-P. & Azanon, E. “Micro-offline gains” convey no benefit for motor skill learning. bioRxiv, 2024.2007.2011.602795 (2024). https://doi.org:10.1101/2024.07.11.602795

(62) Mylonas, D. et al. Maintenance of Procedural Motor Memory across Brief Rest Periods Requires the Hippocampus. J Neurosci 44 (2024). https://doi.org:10.1523/JNEUROSCI.1839-23.2024

Author Response

Reviewer #1 (Public Review):

Summary:

By examining the prevalence of interactions with ancient amino acids of coenzymes in ancient versus recent folds, the authors noticed an increased interaction propensity for ancient interactions. They infer from this that coenzymes might have played an important role in prebiotic proteins.

Strengths:

(1) The analysis, which is very straightforward, is technically correct. However, the conclusions might not be as strong as presented.

(2) This paper presents an excellent summary of contemporary thought on what might have constituted prebiotic proteins and their properties.

(3) The paper is clearly written.

We are grateful for the kind comments of the reviewer on our manuscript. However, we would like to clarify a possible misunderstanding in the summary of our study. Specifically, analysis of "ancient versus recent folds" was not really reported in our results. Our analysis concerned "coenzyme age" rather than the "protein folds age" and was focused mainly on interaction with early vs. late amino acids in protein sequence. While structural propensities of the coenzyme binding sites were also analyzed, no distinction on the level of ancient vs. recent folds was assumed and this was only commented on in the discussion, based on previous work of others.

Weaknesses:

(1) The conclusions might not be as strong as presented. First of all, while ancient amino acids interact less frequently in late with a given coenzyme, maybe this just reflects the fact that proteins that evolved later might be using residues that have a more favorable binding free energy.

We would like to point out that there was no distinction to proteins that evolved early or late in our dataset of coenzyme-binding proteins. The aim of our analysis was purely to observe trends in the age of amino acids vs. age of coenzymes. While no direct inference can be made from this about early life as all the proteins are from extant life (as highlighted in the discussion of our work), our goal was to look for intrinsic propensities of early vs. late amino acids in binding to the different coenzyme entities. Indeed, very early interactions would be smeared by the eons of evolutionary history (perhaps also towards more favourable binding free energy, as pointed out also by the reviewer). Nevertheless, significant trends have been recorded across the PDB dataset, pointing to different propensities and mechanistic properties of the binding events. Rather than to a specific evolutionary past, our data therefore point to a “capacity” of the early amino acids to bind certain coenzymes and we believe that this is the major (and standing) conclusion of our work, along with the properties of such interactions. In our revised version, we will carefully go through all the conclusions and make sure that this message stands out but we are confident that the following concluding sentences copied from the abstract and the discussion of our manuscript fully comply with our data:

“These results imply the plausibility of a coenzyme-peptide functional collaboration preceding the establishment of the Central Dogma and full protein alphabet evolution”

“While no direct inferences about distant evolutionary past can be drawn from the analysis of extant proteins, the principles guiding these interactions can imply their potential prebiotic feasibility and significance.”

“This implies that late amino acids would not be necessarily needed for the sovereignty of coenzyme-peptide interplay.”

We would also like to add that proteins that evolved later might not always have higher free energy of binding. Musil et al., 2021 (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8294521/) showed in their study on the example of haloalkane dehalogenase Dha A that the ancestral sequence reconstruction is a powerful tool for designing more stable, but also more active proteins. Ancestral sequence reconstruction relies on finding ancient states of protein families to suggest mutations that will lead to more stable proteins than are currently existing proteins. Their study did not explore the ligand-protein interactions specifically, but showed that ancient states often show more favourable properties than modern proteins.

(2) What about other small molecules that existed in the probiotic soup? Do they also prefer such ancient amino acids? If so, this might reflect the interaction propensity of specific amino acids rather than the inferred important role of coenzymes.

We appreciate the comment of the reviewer towards other small molecules, which we assume points mainly towards metal ions (i.e. inorganic cofactors). We completely agree with the reviewer that such interactions are of utmost importance to the origins of life. Intentionally, they were not part of our study, as these have already been studied previously by others (e.g. Bromberg et al., 2022; and reviewed in Frenkel-Pinter et al., 2020) and also us (Fried et al., 2022). For example, it is noteworthy that prebiotically relevant metal binding sites (e.g. of Mg2+) exhibit enrichment in early amino acids such as Asp and Glu while more recent metal (e.g. Cu and Zn) site in the late amino acids His and Cys (Fried et al., 2022). At the same time, comparable analyses of amino acid - coenzyme trends were not available.

Nevertheless, involvement of metal ions in the coenzyme binding sites was also studied here and pointed to their bigger involvement with the Ancient coenzymes. In the revised version of the manuscript, we will be happy to enlarge the discussion of the studies concerning inorganic cofactors.

(3) Perhaps the conclusions just reflect the types of active sites that evolved first and nothing more.

We partly agree on this point with the reviewer but not on the fact why it is listed as the weakness of our study and on the “nothing more” notion. Understanding what the properties of the earliest binding sites is key to merging the gap between prebiotic chemistry and biochemistry. The potential of peptides preceding ribosomal synthesis (and the full alphabet evolution) along with prebiotically plausible coenzymes addresses exactly this gap, which is currently not understood.

Reviewer #2 (Public Review):

I enjoyed reading this paper and appreciate the careful analysis performed by the investigators examining whether 'ancient' cofactors are preferentially bound by the first-available amino acids, and whether later 'LUCA' cofactors are bound by the late-arriving amino acids. I've always found this question fascinating as there is a contradiction in inorganic metal-protein complexes (not what is focused on here). Metal coordination of Fe, Ni heavily relies on softer ligands like His and Cys - which are by most models latecomer amino acids. There are no traces of thiols or imidazoles in meteorites - although work by Dvorkin has indicated that could very well be due to acid degradation during extraction. Chris Dupont (PNAS 2005) showed that metal speciation in the early earth (such as proposed by Anbar and prior RJP Williams) matched the purported order of fold emergence.

As such, cofactor-protein interactions as a driving force for evolution has always made sense to me and I admittedly read this paper biased in its favor. But to make sure, I started to play around with the data that the authors kindly and importantly shared in the supplementary files. Here's what I found:

Point 1: The correlation between abundance of amino acids and protein age is dominated by glycine. There is a small, but visible difference in old vs new amino acid fractional abundance between Ancient and LUCA proteins (Figure 3, Supplementary Table 3). However, the bias is not evenly distributed among the amino acids - which Figure 4A shows but is hard to digest as presented. So instead I used the spreadsheet in Supplement 3 to calculate the fractional difference FDaa = F(old aa)-F(new aa). As expected from Figure 3, the mean FD for Ancient is greater than the mean FD for LUCA. But when you look at the same table for each amino acid FDcofactor = F(ancient cofactor) - F(LUCA cofactor), you now see that the bias is not evenly distributed between older and newer amino acids at all. In fact, most of the difference can be explained by glycine (FDcofactor = 3.8) and the rest by also including tryptophan (FDcofactor = -3.8). If you remove these two amino acids from the analysis, the trend seen in Figure 3 all but disappears.

Troubling - so you might argue that Gly is the oldest of the old and Trp is the newest of the new so the argument still stands. Unfortunately, Gly is a lot of things - flexible, small, polar - so what is the real correlation, age, or chemistry? This leads to point 2.

We truly acknowledge the effort that the reviewer made in the revision of the data and for the thoughtful, deeper analysis. We agree that this deserves further discussion of our data. As invited by the reviewer, we indeed repeated the analysis on the whole dataset. First, we would like to point out that the reviewer was most probably referring to the Supplementary Fig. 2 (and not 3, which concerns protein folds). While the difference between Ancient and LUCA coenzyme binding is indeed most pronounced for Gly and Trp, we failed to confirm that the trend disappears if those two amino acids are removed from the analysis (additional FDcofactors of 3.2 and -3.2 are observed for the early and late amino acids, resp.), as seen in Table I below. The main additional contributors to this effect are Asp (FD of 2.1) and Ser (FD of 1.8) from the early amino acids and Arg (FD of -2.6) and Cys (FD of -1.7) of the late amino acids. Hence, while we agree with the reviewer that Gly and Trp (the oldest and the youngest) contribute to this effect the most, we disagree that the trend reduces to these two amino acids.

In addition, the most recent coenzyme temporality (the Post-LUCA) was neglected in the reviewer’s analysis. The difference between F (old) and F (new) is even more pronounced in PostLUCA than in LUCA, vs. Ancient (Table II) and depends much less on Trp. Meanwhile, Asp, Ser, Leu, Phe, and Arg dominate the observed phenomenon (Table I). This further supports our lack of agreement with the reviewer’s point. Nevertheless, we remain grateful for this discussion and we will happily include this additional analysis in the Supplementary Material of our revised manuscript.

Author response table 1.

Amino acid fractional difference of all coenzymes at residue level

Author response table 2.

Amino acid fractional difference of all coenzymes

Point 2 - The correlation is dominated by phosphate.

In the ancient cofactor list, all but 4 comprise at least one phosphate (SAM, tetrahydrofolic acid, biopterin, and heme). Except for SAM, the rest have very low Gly abundance. The overall high Gly abundance in the ancient enzymes is due to the chemical property of glycine that can occupy the right-hand side of the Ramachandran plot. This allows it to make the alternating alphaleftalpharight conformation of the P-loop forming Milner-White's anionic nest. If you remove phosphate binding folds from the analysis the trend in Figure 3 vanishes.

Likewise, Trp is an important functional residue for binding quinones and tuning its redox potential. The LUCA cofactor set is dominated by quinone and derivatives, which likely drives up the new amino acid score for this class of cofactors.

Once again, we are thankful to the reviewer for raising this point. The role of Gly in the anionic nests proposed by Milner-White and Russel, as well as the Trp role in quinone binding are important points that we would be happy to highlight more in the discussion of the revised manuscript.<br /> Nevertheless, we disagree that the trends reduce only to the phosphate-containing coenzymes and importantly, that “the trend in Figure 3 vanishes” upon their removal. Table III and IV (below) show the data for coenzymes excluding those with phosphate moiety and the trend in Fig. 3 remains, albeit less pronounced.

Author response table 3.

Amino acid fractional difference of non-phosphate containing coenzymes

Author response table 4.

Amino acid fractional difference of non-phosphate containing coenzymes at residue level

In summary, while I still believe the premise that cofactors drove the shape of peptides and the folds that came from them - and that Rossmann folds are ancient phosphate-binding proteins, this analysis does not really bring anything new to these ideas that have already been stated by Tawfik/Longo, Milner-White/Russell, and many others.

I did this analysis ad hoc on a slice of the data the authors provided and could easily have missed something and I encourage the authors to check my work. If it holds up it should be noted that negative results can often be as informative as strong positive ones. I think the signal here is too weak to see in the noise using the current approach.

We are grateful to the reviewer for encouraging further look at our data. While we hope that the analysis on the whole dataset (listed in Tables I - IV) will change the reviewer’s standpoint on our work, we would still like to comment on the questioned novelty of our results. In fact, the extraordinary works by Tawfik/Longo and Milner-While/Russel (which were cited in our manuscript multiple times) presented one of the motivations for this study. We take the opportunity to copy the part of our discussion that specifically highlights the relevance of their studies, and points out the contribution of our work with respect to theirs.

“While all the coenzymes bind preferentially to protein residue sidechains, more backbone interactions appear in the ancient coenzyme class when compared to others. This supports an earlier hypothesis that functions of the earliest peptides (possibly of variable compositions and lengths) would be performed with the assistance of the main chain atoms rather than their sidechains (Milner-White and Russel 2011). Longo et al., recently analyzed binding sites of different phosphate-containing ligands which were arguably of high relevance during earliest stages of life, connecting all of today’s core metabolism (Longo et al., 2020 (b)). They observed that unlike the evolutionary younger binding motifs (which rely on sidechain binding), the most ancient lineages indeed bind to phosphate moieties predominantly via the protein backbone. Our analysis assigns this phenomenon primarily to interactions via early amino acids that (as mentioned above) are generally enriched in the binding interface of the ancient coenzymes. This implies that late amino acids would not be necessarily needed for the sovereignty of coenzymepeptide interplay.”

Unlike any other previous work, our study involves all the major coenzymes (not just the phosphate-containing ones) and is based on their evolutionary age, as well as age of amino acids. It is the first PDB-wide systematic evolutionary analysis of coenzyme-amino acid binding. Besides confirming some earlier theoretical assertions (such as role of backbone interactions in early peptide-coenzyme evolution) and observations (such as occurrence of the ancient phosphatecontaining coenzymes in the oldest protein folds), it uncovers substantial novel knowledge. For example, (i) enrichment of early amino acids in the binding of ancient coenzymes, vs. enrichment of late amino acids in the binding of LUCA and Post-LUCA coenzymes, (ii) the trends in secondary structure content of the binding sites of coenzyme of different temporalities, (iii) increased involvement of metal ions in the ancient coenzyme binding events, and (iv) the capacity of only early amino acids to bind ancient coenzymes. In our humble opinion, all of these points bring important contributions in the peptide-coenzyme knowledge gap which has been discussed in a number of previous studies.

Author response:

eLife assessment

This potentially useful study involves neuro-imaging and electrophysiology in a small cohort of congenital cataract patients after sight recovery and age-matched control participants with normal sight. It aims to characterize the effects of early visual deprivation on excitatory and inhibitory balance in the visual cortex. While the findings are taken to suggest the existence of persistent alterations in Glx/GABA ratio and aperiodic EEG signals, the evidence supporting these claims is incomplete. Specifically, small sample sizes, lack of a specific control cohort, and other methodological limitations will likely restrict the usefulness of the work, with relevance limited to scientists working in this particular subfield.

As pointed out in the public reviews, there are only very few human models which allow for assessing the role of early experience on neural circuit development. While the prevalent research in permanent congenital blindness reveals the response and adaptation of the developing brain to an atypical situation (blindness), research in sight restoration addresses the question of whether and how atypical development can be remediated if typical experience (vision) is restored. The literature on the role of visual experience in the development of E/I balance in humans, assessed via Magnetic Resonance Spectroscopy (MRS), has been limited to a few studies on congenital permanent blindness. Thus, we assessed sight recovery individuals with a history of congenital blindness, as limited evidence from other researchers indicated that the visual cortex E/I ratio might differ compared to normally sighted controls.

Individuals with total bilateral congenital cataracts who remained untreated until later in life are extremely rare, particularly if only carefully diagnosed patients are included in a study sample. A sample size of 10 patients is, at the very least, typical of past studies in this population, even for exclusively behavioral assessments. In the present study, in addition to behavioral assessment as an indirect measure of sensitive periods, we investigated participants with two neuroimaging methods (Magnetic Resonance Spectroscopy and electroencephalography) to directly assess the neural correlates of sensitive periods in humans. The electroencephalography data allowed us to link the results of our small sample to findings documented in large cohorts of both, sight recovery individuals and permanently congenitally blind individuals. As pointed out in a recent editorial recommending an “exploration-then-estimation procedure,” (“Consideration of Sample Size in Neuroscience Studies,” 2020), exploratory studies like ours provide crucial direction and specific hypotheses for future work.

We included an age-matched sighted control group recruited from the same community, measured in the same scanner and laboratory, to assess whether early experience is necessary for a typical excitatory/inhibitory (E/I) ratio to emerge in adulthood. The present findings indicate that this is indeed the case. Based on these results, a possible question to answer in future work, with individuals who had developmental cataracts, is whether later visual deprivation causes similar effects. Note that even if visual deprivation at a later stage in life caused similar effects, the current results would not be invalidated; by contrast, they are essential to understand future work on late (permanent or transient) blindness.

Thus, we think that the present manuscript has far reaching implications for our understanding of the conditions under which E/I balance, a crucial characteristic of brain functioning, emerges in humans.

Finally, our manuscript is one of the first few studies which relates MRS neurotransmitter concentrations to parameters of EEG aperiodic activity. Since present research has been using aperiodic activity as a correlate of the E/I ratio, and partially of higher cognitive functions, we think that our manuscript additionally contributes to a better understanding of what might be measured with aperiodic neurophysiological activity.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this human neuroimaging and electrophysiology study, the authors aimed to characterize the effects of a period of visual deprivation in the sensitive period on excitatory and inhibitory balance in the visual cortex. They attempted to do so by comparing neurochemistry conditions ('eyes open', 'eyes closed') and resting state, and visually evoked EEG activity between ten congenital cataract patients with recovered sight (CC), and ten age-matched control participants (SC) with normal sight.

First, they used magnetic resonance spectroscopy to measure in vivo neurochemistry from two locations, the primary location of interest in the visual cortex, and a control location in the frontal cortex. Such voxels are used to provide a control for the spatial specificity of any effects because the single-voxel MRS method provides a single sampling location. Using MR-visible proxies of excitatory and inhibitory neurotransmission, Glx and GABA+ respectively, the authors report no group effects in GABA+ or Glx, no difference in the functional conditions 'eyes closed' and 'eyes open'. They found an effect of the group in the ratio of Glx/GABA+ and no similar effect in the control voxel location. They then performed multiple exploratory correlations between MRS measures and visual acuity, and reported a weak positive correlation between the 'eyes open' condition and visual acuity in CC participants.

The same participants then took part in an EEG experiment. The authors selected only two electrodes placed in the visual cortex for analysis and reported a group difference in an EEG index of neural activity, the aperiodic intercept, as well as the aperiodic slope, considered a proxy for cortical inhibition. They report an exploratory correlation between the aperiodic intercept and Glx in one out of three EEG conditions.

The authors report the difference in E/I ratio, and interpret the lower E/I ratio as representing an adaptation to visual deprivation, which would have initially caused a higher E/I ratio. Although intriguing, the strength of evidence in support of this view is not strong. Amongst the limitations are the low sample size, a critical control cohort that could provide evidence for a higher E/I ratio in CC patients without recovered sight for example, and lower data quality in the control voxel.

Strengths of study:

How sensitive period experience shapes the developing brain is an enduring and important question in neuroscience. This question has been particularly difficult to investigate in humans. The authors recruited a small number of sight-recovered participants with bilateral congenital cataracts to investigate the effect of sensitive period deprivation on the balance of excitation and inhibition in the visual brain using measures of brain chemistry and brain electrophysiology. The research is novel, and the paper was interesting and well-written.

Limitations:

(1.1) Low sample size. Ten for CC and ten for SC, and a further two SC participants were rejected due to a lack of frontal control voxel data. The sample size limits the statistical power of the dataset and increases the likelihood of effect inflation.

Applying strict criteria, we only included individuals who were born with no patterned vision in the CC group. The population of individuals who have remained untreated past infancy is small in India, despite a higher prevalence of childhood cataract than Germany. Indeed, from the original 11 CC and 11 SC participants tested, one participant each from the CC and SC group had to be rejected, as their data had been corrupted, resulting in 10 participants in each group.

It was a challenge to recruit participants from this rare group with no history of neurological diagnosis/intake of neuromodulatory medications, who were able and willing to undergo both MRS and EEG. For this study, data collection took more than 1.5 years.

We took care of the validity of our results with two measures; first, assessed not just MRS, but additionally, EEG measures of E/I ratio. The latter allowed us to link results to a larger population of CC individuals, that is, we replicated the results of a larger group of 38 individuals (Ossandón et al., 2023) in our sub-group.

Second, we included a control voxel. As predicted, all group effects were restricted to the occipital voxel.

(1.2) Lack of specific control cohort. The control cohort has normal vision. The control cohort is not specific enough to distinguish between people with sight loss due to different causes and patients with congenital cataracts with co-morbidities. Further data from more specific populations, such as patients whose cataracts have not been removed, with developmental cataracts, or congenitally blind participants, would greatly improve the interpretability of the main finding. The lack of a more specific control cohort is a major caveat that limits a conclusive interpretation of the results.

The existing work on visual deprivation and neurochemical changes, as assessed with MRS, has been limited to permanent congenital blindness. In fact, most of the studies on permanent blindness included only congenitally blind or early blind humans (Coullon et al., 2015; Weaver et al., 2013), or, in separate studies, only late-blind individuals (Bernabeu et al., 2009). Thus, accordingly, we started with the most “extreme” visual deprivation model, sight recovery after congenital blindness. If we had not observed any group difference compared to normally sighted controls, investigating other groups might have been trivial. Based on our results, subsequent studies in late blind individuals, and then individuals with developmental cataracts, can be planned with clear hypotheses.

(1.3) MRS data quality differences. Data quality in the control voxel appears worse than in the visual cortex voxel. The frontal cortex MRS spectrum shows far broader linewidth than the visual cortex (Supplementary Figures). Compared to the visual voxel, the frontal cortex voxel has less defined Glx and GABA+ peaks; lower GABA+ and Glx concentrations, lower NAA SNR values; lower NAA concentrations. If the data quality is a lot worse in the FC, then small effects may not be detectable.

Worse data quality in the frontal than the visual cortex has been repeatedly observed in the MRS literature, attributable to magnetic field distortions (Juchem & Graaf, 2017) resulting from the proximity of the region to the sinuses (recent example: (Rideaux et al., 2022)). Nevertheless, we chose the frontal control region rather than a parietal voxel, given the potential  neurochemical changes in multisensory regions of the parietal cortex due to blindness. Such reorganization would be less likely in frontal areas associated with higher cognitive functions. Further, prior MRS studies of the visual cortex have used the frontal cortex as a control region as well (Pitchaimuthu et al., 2017; Rideaux et al., 2022).

In the present study, we checked that the frontal cortex datasets for Glx and GABA+ concentrations were of sufficient quality: the fit error was below 8.31% in both groups (Supplementary Material S3). For reference, Mikkelsen et al. reported a mean GABA+ fit error of 6.24 +/- 1.95% from a posterior cingulate cortex voxel across 8 GE scanners, using the Gannet pipeline. No absolute cutoffs have been proposed for fit errors. However, MRS studies in special populations (I/E ratio assessed in narcolepsy (Gao et al., 2024), GABA concentration assessed in Autism Spectrum Disorder (Maier et al., 2022)) have used frontal cortex data with a fit error of <10% to identify differences between cohorts (Gao et al., 2024; Pitchaimuthu et al., 2017). Based on the literature, MRS data from the frontal voxel of the present study would have been of sufficient quality to uncover group differences.

In the revised manuscript, we will add the recently published MRS quality assessment form to the supplementary materials. Additionally, we would like to allude to our apriori prediction of group differences for the visual cortex, but not for the frontal cortex voxel.

(1.4) Because of the direction of the difference in E/I, the authors interpret their findings as representing signatures of sight improvement after surgery without further evidence, either within the study or from the literature. However, the literature suggests that plasticity and visual deprivation drive the E/I index up rather than down. Decreasing GABA+ is thought to facilitate experience-dependent remodelling. What evidence is there that cortical inhibition increases in response to a visual cortex that is over-sensitised due to congenital cataracts? Without further experimental or literature support this interpretation remains very speculative.

Indeed, higher inhibition was not predicted, which we attempt to reconcile in our discussion section. We base our discussion mainly on the non-human animal literature, which has shown evidence of homeostatic changes after prolonged visual deprivation in the adult brain (Barnes et al., 2015). It is also interesting to note that after monocular deprivation in adult humans, resting GABA+ levels decreased in the visual cortex (Lunghi et al., 2015). Assuming that after delayed sight restoration, adult neuroplasticity mechanisms must be employed, these studies would predict a “balancing” of the increased excitatory drive following sight restoration by a commensurate increase in inhibition (Keck et al., 2017). Additionally, the EEG results of the present study allowed for speculation regarding the underlying neural mechanisms of an altered E/I ratio. The aperiodic EEG activity suggested higher spontaneous spiking (increased intercept) and increased inhibition (steeper aperiodic slope between 1-20 Hz) in CC vs SC individuals (Ossandón et al., 2023).

In the revised manuscript, we will more clearly indicate that these speculations are based primarily on non-human animal work, due to the lack of human studies on the subject.

(1.5) Heterogeneity in the patient group. Congenital cataract (CC) patients experienced a variety of duration of visual impairment and were of different ages. They presented with co-morbidities (absorbed lens, strabismus, nystagmus). Strabismus has been associated with abnormalities in GABAergic inhibition in the visual cortex. The possible interactions with residual vision and confounds of co-morbidities are not experimentally controlled for in the correlations, and not discussed.

The goal of the present study was to assess whether we would observe changes in E/I ratio after restoring vision at all. We would not have included patients without nystagmus in the CC group of the present study, since it would have been unlikely that they experienced congenital patterned visual deprivation. Amongst diagnosticians, nystagmus or strabismus might not be considered genuine “comorbidities” that emerge in people with congenital cataracts. Rather, these are consequences of congenital visual deprivation, which we employed as diagnostic criteria. Similarly, absorbed lenses are clear signs that cataracts were congenital. As in other models of experience dependent brain development (e.g. the extant literature on congenital permanent blindness, including anophthalmic individuals (Coullon et al., 2015; Weaver et al., 2013), some uncertainty remains regarding whether the (remaining, in our case) abnormalities of the eye, or the blindness they caused, are the factors driving neural changes. In case of people with reversed congenital cataracts, at least the retina is considered to be intact, as they would otherwise not receive cataract removal surgery.

However, we consider it unlikely that strabismus caused the group differences, because the present study shows group differences in the Glx/GABA+ ratio at rest, regardless of eye opening or eye closure, for which strabismus would have caused distinct effects. By contrast, the link between GABA concentration and, for example, interocular suppression in strabismus, have so far been documented during visual stimulation (Mukerji et al., 2022; Sengpiel et al., 2006), and differed in direction depending on the amblyopic vs. non-amblyopic eye. Further, one MRS study did not find group differences in GABA concentration between the visual cortices of 16 amblyopic individuals and sighted controls (Mukerji et al., 2022), supporting that the differences in Glx/GABA+ concentration which we observed were driven by congenital deprivation, and not amblyopia-associated visual acuity or eye movement differences.  

In the revised manuscript, we will discuss the inclusion criteria in more detail, and the aforementioned reasons why our data remains interpretable.

(1.6) Multiple exploratory correlations were performed to relate MRS measures to visual acuity (shown in Supplementary Materials), and only specific ones were shown in the main document. The authors describe the analysis as exploratory in the 'Methods' section. Furthermore, the correlation between visual acuity and E/I metric is weak, and not corrected for multiple comparisons. The results should be presented as preliminary, as no strong conclusions can be made from them. They can provide a hypothesis to test in a future study.

In the revised manuscript, we will clearly indicate that the exploratory correlation analyses are reported to put forth hypotheses for future studies.

(1.7) P.16 Given the correlation of the aperiodic intercept with age ("Age negatively correlated with the aperiodic intercept across CC and SC individuals, that is, a flattening of the intercept was observed with age"), age needs to be controlled for in the correlation between neurochemistry and the aperiodic intercept. Glx has also been shown to negatively correlate with age.

The correlation between chronological age and aperiodic intercept was observed across groups, but the correlation between Glx and the intercept of the aperiodic EEG activity was seen only in the CC group, even though the SC group was matched for age. Thus, such a correlation was very unlikely to  be predominantly driven by an effect of chronological age.

In the revised manuscript, we will add the linear regressions with age as a covariate included below, for the relationship between aperiodic intercept and Glx concentration in the CC group. 

a. A linear regression was conducted within the CC group to predict the intercept during visual stimulation, based on age and visual cortex Glx concentration. The results of the regression analysis indicated that the model explained a significant proportion of the variance in the aperiodic intercept, 𝑅2\=0.82_, t_(2,7)=16.1_, 𝑝=0.0024._ Note that the coefficient for age was not significant, 𝛽=0.007, t(7)=0.82, 𝑝=0.439. The regression coefficients and their respective statistics are presented in Author response table 1.

Author response table 1.

Regression Analysis Summary for Predicting Aperiodic Intercept (Visual Stimulation) in the CC group

b. A linear regression was conducted to predict the intercept during eye opening at rest, based on age and visual cortex Glx concentration. The results of the regression analysis indicated that the model explained a significant proportion of the variance in the aperiodic intercept, 𝑅2\=0.842_, t_(2,7)=18.6,  𝑝=0.00159_._ Note that the coefficient for age was not significant, 𝛽=−0.005, t(7)=−0.90, 𝑝=0.400. The regression coefficients and their respective statistics are presented in Author response table 2.

Author response table 2.

Regression Analysis Summary for Predicting Aperiodic Intercept (Eyes Open) in the CC group

c. Given that the Glx coefficient is significant in both models and age does not significantly predict either outcome, it can be concluded that Glx independently predicts the intercept of the aperiodic intercept.

(1.8) Multiple exploratory correlations were performed to relate MRS to EEG measures (shown in Supplementary Materials), and only specific ones were shown in the main document. Given the multiple measures from the MRS, the correlations with the EEG measures were exploratory, as stated in the text, p.16, and in Figure 4. Yet the introduction said that there was a prior hypothesis "We further hypothesized that neurotransmitter changes would relate to changes in the slope and intercept of the EEG aperiodic activity in the same subjects." It would be great if the text could be revised for consistency and the analysis described as exploratory.

In the revised manuscript, we will improve the phrasing. We consider the correlation analyses as exploratory due to our sample size and the absence of prior work. However, we did hypothesize that both MRS and EEG markers would concurrently be altered in CC vs SC individuals.

(1.9) The analysis for the EEG needs to take more advantage of the available data. As far as I understand, only two electrodes were used, yet far more were available as seen in their previous study (Ossandon et al., 2023). The spatial specificity is not established. The authors could use the frontal cortex electrode (FP1, FP2) signals as a control for spatial specificity in the group effects, or even better, all available electrodes and correct for multiple comparisons. Furthermore, they could use the aperiodic intercept vs Glx in SC to evaluate the specificity of the correlation to CC.

The aperiodic intercept and slope did not differ between CC and SC individuals for Fp1 and Fp2, suggesting the spatial specificity of the results. In the revised manuscript, we will add this analysis to the supplementary material.

Author response image 1.

Aperiodic intercept (top) and slope (bottom) for congenital cataract-reversal (CC, red) and age-matched normally sighted control (SC, blue) individuals. Distributions of these parameters are displayed as violin plots for three conditions; at rest with eyes closed (EC), at rest with eyes open (EO) and during visual stimulation (LU). Aperiodic parameters were calculated across electrodes Fp1 and Fp2. Solid black lines indicate mean values, dotted black lines indicate median values. Coloured lines connect values of individual participants across conditions.

Further, Glx concentration in the visual cortex did not correlate with the aperiodic intercept in the SC group (Figure 4), suggesting that this relationship was indeed specific to the CC group.

The data from all electrodes has been analyzed and published in other studies as well (Pant et al., 2023; Ossandón et al., 2023).

Reviewer #2 (Public Review):

Summary:

The manuscript reports non-invasive measures of activity and neurochemical profiles of the visual cortex in congenitally blind patients who recovered vision through the surgical removal of bilateral dense cataracts. The declared aim of the study is to find out how restoring visual function after several months or years of complete blindness impacts the balance between excitation and inhibition in the visual cortex.

Strengths:

The findings are undoubtedly useful for the community, as they contribute towards characterising the many ways this special population differs from normally sighted individuals. The combination of MRS and EEG measures is a promising strategy to estimate a fundamental physiological parameter - the balance between excitation and inhibition in the visual cortex, which animal studies show to be heavily dependent upon early visual experience. Thus, the reported results pave the way for further studies, which may use a similar approach to evaluate more patients and control groups.

Weaknesses:

(2.1) The main issue is the lack of an appropriate comparison group or condition to delineate the effect of sight recovery (as opposed to the effect of congenital blindness). Few previous studies suggested an increased excitation/Inhibition ratio in the visual cortex of congenitally blind patients; the present study reports a decreased E/I ratio instead. The authors claim that this implies a change of E/I ratio following sight recovery. However, supporting this claim would require showing a shift of E/I after vs. before the sight-recovery surgery, or at least it would require comparing patients who did and did not undergo the sight-recovery surgery (as common in the field).

Longitudinal studies would indeed be the best way to test the hypothesis that the lower E/I ratio in the CC group observed by the present study is a consequence of sight restoration. However, longitudinal studies involving neuroimaging are an effortful challenge, particularly in research conducted outside of major developed countries and dedicated neuroimaging research facilities. Crucially, however, had CC and SC individuals, as well as permanently congenitally blind vs SC individuals (Coullon et al., 2015; Weaver et al., 2013), not differed on any neurochemical markers, such a longitudinal study might have been trivial. Thus, in order to justify and better tailor longitudinal studies, cross-sectional studies are an initial step.

(2.2) MR Spectroscopy shows a reduced GLX/GABA ratio in patients vs. sighted controls; however, this finding remains rather isolated, not corroborated by other observations. The difference between patients and controls only emerges for the GLX/GABA ratio, but there is no accompanying difference in either the GLX or the GABA concentrations. There is an attempt to relate the MRS data with acuity measurements and electrophysiological indices, but the explorative correlational analyses do not help to build a coherent picture. A bland correlation between GLX/GABA and visual impairment is reported, but this is specific to the patients' group (N=10) and would not hold across groups (the correlation is positive, predicting the lowest GLX/GABA ratio values for the sighted controls - the opposite of what is found). There is also a strong correlation between GLX concentrations and the EEG power at the lowest temporal frequencies. Although this relation is intriguing, it only holds for a very specific combination of parameters (of the many tested): only with eyes open, only in the patient group.

We interpret these findings differently, that is, in the context of experiments from non-human animals and the larger MRS literature.

Homeostatic control of E/I balance assumes that the ratio of excitation (reflected here by Glx) and inhibition (reflected here by GABA+) is regulated. Like prior work (Gao et al., 2024, 2024; Narayan et al., 2022; Perica et al., 2022; Steel et al., 2020; Takado et al., 2022; Takei et al., 2016), we assumed that the ratio of Glx/GABA+ is indicative of E/I balance rather than solely the individual neurotransmitter levels. One of the motivations for assessing the ratio vs the absolute concentration is that as per the underlying E/I balance hypothesis, a change in excitation would cause a concomitant change in inhibition, and vice versa, which has been shown in non-human animal work (Fang et al., 2021; Haider et al., 2006; Tao & Poo, 2005) and modeling research (Vreeswijk & Sompolinsky, 1996; Wu et al., 2022). Importantly, our interpretation of the lower E/I ratio is not just from the Glx/GABA+ ratio, but additionally, based on the steeper EEG aperiodic slope (1-20 Hz).  

As in the discussion section and response 1.4, we did not expect to see a lower Glx/GABA+ ratio in CC individuals. We discuss the possible reasons for the direction of the correlation with visual acuity and aperiodic offset during passive visual stimulation, and offer interpretations and (testable) hypotheses.

We interpret the direction of the  Glx/GABA+ correlation with visual acuity to imply that patients with highest (compensatory) balancing of the consequences of congenital blindness (hyperexcitation), in light of visual stimulation, are those who recover best. Note, the sighted control group was selected based on their “normal” vision. Thus, clinical visual acuity measures are not expected to sufficiently vary, nor have the resolution to show strong correlations with neurophysiological measures. By contrast, the CC group comprised patients highly varying in visual outcomes, and thus were ideal to investigate such correlations.

This holds for the correlation between Glx and the aperiodic intercept, as well. Previous work has suggested that the intercept of the aperiodic activity is associated with broadband spiking activity in neural circuits (Manning et al., 2009). Thus, an atypical increase of spiking activity during visual stimulation, as indirectly suggested by “old” non-human primate work on visual deprivation (Hyvärinen et al., 1981) might drive a correlation not observed in healthy populations.

In the revised manuscript, we will more clearly indicate in the discussion that these are possible post-hoc interpretations. We argue that given the lack of such studies in humans, it is all the more important that extant data be presented completely, even if the direction of the effects are not as expected.

(2.3) For these reasons, the reported findings do not allow us to draw firm conclusions on the relation between EEG parameters and E/I ratio or on the impact of early (vs. late) visual experience on the excitation/inhibition ratio of the human visual cortex.

Indeed, the correlations we have tested between the E/I ratio and EEG parameters were exploratory, and have been reported as such. The goal of our study was not to compare the effects of early vs. late visual experience. The goal was to study whether early visual experience is necessary for a typical E/I ratio in visual neural circuits. We provided clear evidence in favor of this hypothesis. Thus, the present results suggest the necessity of investigating the effects of late visual deprivation. In fact, such research is missing in permanent blindness as well.

Reviewer #3 (Public Review):

This manuscript examines the impact of congenital visual deprivation on the excitatory/inhibitory (E/I) ratio in the visual cortex using Magnetic Resonance Spectroscopy (MRS) and electroencephalography (EEG) in individuals whose sight was restored. Ten individuals with reversed congenital cataracts were compared to age-matched, normally sighted controls, assessing the cortical E/I balance and its interrelationship to visual acuity. The study reveals that the Glx/GABA ratio in the visual cortex and the intercept and aperiodic signal are significantly altered in those with a history of early visual deprivation, suggesting persistent neurophysiological changes despite visual restoration.

My expertise is in EEG (particularly in the decomposition of periodic and aperiodic activity) and statistical methods. I have several major concerns in terms of methodological and statistical approaches along with the (over)interpretation of the results. These major concerns are detailed below.

(3.1) Variability in visual deprivation:

- The document states a large variability in the duration of visual deprivation (probably also the age at restoration), with significant implications for the sensitivity period's impact on visual circuit development. The variability and its potential effects on the outcomes need thorough exploration and discussion.

We work with a rare, unique patient population, which makes it difficult to systematically assess the effects of different visual histories while maintaining stringent inclusion criteria such as complete patterned visual deprivation at birth. Regardless, we considered the large variance in age at surgery and time since surgery as supportive of our interpretation: group differences were found despite the large variance in duration of visual deprivation. Moreover, the existing variance was used to explore possible associations between behavior and neural measures, as well as neurochemical and EEG measures.

In the revised manuscript, we will detail the advantages and disadvantages of our CC sample, with respect to duration of congenital visual deprivation.

(3.2) Sample size:

- The small sample size is a major concern as it may not provide sufficient power to detect subtle effects and/or overestimate significant effects, which then tend not to generalize to new data. One of the biggest drivers of the replication crisis in neuroscience.

We address the small sample size in our discussion, and make clear that small sample sizes were due to the nature of investigations in special populations. It is worth noting that our EEG results fully align  with those of a larger sample of CC individuals (Ossandón et al., 2023), providing us confidence about their validity and reproducibility. Moreover, our MRS results and correlations of those with EEG parameters were spatially specific to occipital cortex measures, as predicted.

The main problem with the correlation analyses between MRS and EEG measures is that the sample size is simply too small to conduct such an analysis. Moreover, it is unclear from the methods section that this analysis was only conducted in the patient group (which the reviewer assumed from the plots), and not explained why this was done only in the patient group. I would highly recommend removing these correlation analyses.

We marked the correlation analyses as exploratory; note that we do not base most of our discussion on the results of these analyses. As indicated by Reviewer 1, reporting them allows for deriving more precise hypothesis for future studies. It has to be noted that we investigate an extremely rare population, tested outside of major developed economies and dedicated neuroimaging research facilities. In addition to being a rare patient group, these individuals come from poor communities. Therefore, we consider it justified to report these correlations as exploratory, providing direction for future research.

(3.3) Statistical concerns:

- The statistical analyses, particularly the correlations drawn from a small sample, may not provide reliable estimates (see https://www.sciencedirect.com/science/article/pii/S0092656613000858, which clearly describes this problem).

It would undoubtedly be better to have a larger sample size. We nonetheless think it is of value to the research community to publish this dataset, since 10 multimodal data sets from a carefully diagnosed, rare population, representing a human model for the effects of early experience on brain development, are quite a lot.  Sample sizes in prior neuroimaging studies in transient blindness have most often ranged from n = 1 to n = 10. They nevertheless provided valuable direction for future research, and integration of results across multiple studies provides scientific insights.  

Identifying possible group differences was the goal of our study, with the correlations being an exploratory analysis, which we have clearly indicated in the methods, results and discussion.

- Statistical analyses for the MRS: The authors should consider some additional permutation statistics, which are more suitable for small sample sizes. The current statistical model (2x2) design ANOVA is not ideal for such small sample sizes. Moreover, it is unclear why the condition (EO & EC) was chosen as a predictor and not the brain region (visual & frontal) or neurochemicals. Finally, the authors did not provide any information on the alpha level nor any information on correction for multiple comparisons (in the methods section). Finally, even if the groups are matched w.r.t. age, the time between surgery and measurement, the duration of visual deprivation, (and sex?), these should be included as covariates as it has been shown that these are highly related to the measurements of interest (especially for the EEG measurements) and the age range of the current study is large.

In our ANOVA models, the neurochemicals were the outcome variables, and the conditions were chosen as predictors based on prior work suggesting that Glx/GABA+ might vary with eye closure (Kurcyus et al., 2018). The study was designed based on a hypothesis of group differences localized to the occipital cortex, due to visual deprivation. The frontal cortex voxel was chosen to indicate whether these differences were spatially specific. Therefore, we conducted separate ANOVAs based on this study design.

In the revised manuscript, we will add permutation analyses for our outcomes, as well as multiple regression models investigating whether the variance in visual history might have driven these results. Note that in the supplementary materials (S6, S7), we have reported the correlations between visual history metrics and MRS/EEG outcomes.

The alpha level used for the ANOVA models specified in the methods section was 0.05. The alpha level for the exploratory analyses reported in the main manuscript was 0.008, after correcting for (6) multiple comparisons using the Bonferroni correction, also specified in the methods. Note that the p-values following correction are expressed as multiplied by 6, due to most readers assuming an alpha level of 0.05 (see response regarding large p-values).

We used a control group matched for age and sex. Moreover, the controls were recruited and tested in the same institutes, using the same setup. We feel that we followed the gold standards for recruiting a healthy control group for a patient group.

- EEG statistical analyses: The same critique as for the MRS statistical analyses applies to the EEG analysis. In addition: was the 2x3 ANOVA conducted for EO and EC independently? This seems to be inconsistent with the approach in the MRS analyses, in which the authors chose EO & EC as predictors in their 2x2 ANOVA.

The 2x3 ANOVA was not conducted independently for the eyes open/eyes closed condition, the ANOVA conducted on the EEG metrics was 2x3 because it had group (CC, SC) and condition (eyes open (EO), eyes closed (EC) and visual stimulation (LU)) as predictors.

- Figure 4: The authors report a p-value of >0.999 with a correlation coefficient of -0.42 with a sample size of 10 subjects. This can't be correct (it should be around: p = 0.22). All statistical analyses should be checked.

As specified in the methods and figure legend, the reported p values in Figure 4 have been corrected using the Bonferroni correction, and therefore multiplied by the number of comparisons, leading to the seemingly large values.

Additionally, to check all statistical analyses, we put the manuscript through an independent Statistics Check (Nuijten & Polanin, 2020) (https://michelenuijten.shinyapps.io/statcheck-web/) and will upload the consistency report with the revised supplementary material.

- Figure 2c. Eyes closed condition: The highest score of the *Glx/GABA ratio seems to be ~3.6. In subplot 2a, there seem to be 3 subjects that show a Glx/GABA ratio score > 3.6. How can this be explained? There is also a discrepancy for the eyes-closed condition.

The three subjects that show the Glx/GABA+ ratio > 3.6 in subplot 2a are in the SC group, whereas the correlations plotted in figure 2c are only for the CC group, where the highest score is indeed ~3.6.

(3.4) Interpretation of aperiodic signal:

- Several recent papers demonstrated that the aperiodic signal measured in EEG or ECoG is related to various important aspects such as age, skull thickness, electrode impedance, as well as cognition. Thus, currently, very little is known about the underlying effects which influence the aperiodic intercept and slope. The entire interpretation of the aperiodic slope as a proxy for E/I is based on a computational model and simulation (as described in the Gao et al. paper).

Apart from the modeling work from Gao et al., multiple papers which have also been cited which used ECoG, EEG and MEG and showed concomitant changes in aperiodic activity with pharmacological manipulation of the E/I ratio (Colombo et al., 2019; Molina et al., 2020; Muthukumaraswamy & Liley, 2018). Further, several prior studies have interpreted changes in the aperiodic slope as reflective of changes in the E/I ratio, including studies of developmental groups (Favaro et al., 2023; Hill et al., 2022; McSweeney et al., 2023; Schaworonkow & Voytek, 2021) as well as patient groups (Molina et al., 2020; Ostlund et al., 2021).

In the revised manuscript, we will cite those studies not already included in the introduction.

- Especially the aperiodic intercept is a very sensitive measure to many influences (e.g. skull thickness, electrode impedance...). As crucial results (correlation aperiodic intercept and MRS measures) are facing this problem, this needs to be reevaluated. It is safer to make statements on the aperiodic slope than intercept. In theory, some of the potentially confounding measures are available to the authors (e.g. skull thickness can be computed from T1w images; electrode impedances are usually acquired alongside the EEG data) and could be therefore controlled.

All electrophysiological measures indeed depend on parameters such as skull thickness and electrode impedance. As in the extant literature using neurophysiological measures to compare brain function between patient and control groups, we used a control group matched in age/ sex, recruited in the same region, tested with the same devices, and analyzed with the same analysis pipeline. For example, impedance was kept below 10 kOhm for all subjects. There is no evidence available suggesting that congenital cataracts are associated with changes in skull thickness that would cause the observed pattern of group results. Moreover, we cannot think of how any of the exploratory correlations between neurophysiological measures and MRS measures could be accounted for by a difference e.g. in skull thickness.

- The authors wrote: "Higher frequencies (such as 20-40 Hz) have been predominantly associated with local circuit activity and feedforward signaling (Bastos et al., 2018; Van Kerkoerle et al., 2014); the increased 20-40 Hz slope may therefore signal increased spontaneous spiking activity in local networks. We speculate that the steeper slope of the aperiodic activity for the lower frequency range (1-20 Hz) in CC individuals reflects the concomitant increase in inhibition." The authors confuse the interpretation of periodic and aperiodic signals. This section refers to the interpretation of the periodic signal (higher frequencies). This interpretation cannot simply be translated to the aperiodic signal (slope).

Prior work has not always separated the aperiodic and periodic components, making it unclear what might have driven these effects in our data. The interpretation of the higher frequency range was intended to contrast with the interpretations of lower frequency range, in order to speculate as to why the two aperiodic fits might go in differing directions. We will clarify our interpretation in the revised manuscript. Note that Ossandon et al. reported highly similar results (group differences for CC individuals and for permanently congenitally blind humans) for the aperiodic activity between 20-40 Hz and oscillatory activity in the gamma range. We will allude to these findings in the revised manuscript.

- The authors further wrote: We used the slope of the aperiodic (1/f) component of the EEG spectrum as an estimate of E/I ratio (Gao et al., 2017; Medel et al., 2020; Muthukumaraswamy & Liley, 2018). This is a highly speculative interpretation with very little empirical evidence. These papers were conducted with ECoG data (mostly in animals) and mostly under anesthesia. Thus, these studies only allow an indirect interpretation by what the 1/f slope in EEG measurements is actually influenced.

Note that Muthukumaraswamy et al. (2018) used different types of pharmacological manipulations and analyzed periodic and aperiodic MEG activity in addition to monkey ECoG (Medel et al., 2020) (now published as (Medel et al., 2023)) compared EEG activity in addition to ECoG data after propofol administration. The interpretation of our results are in line with a number of recent studies in developing (Hill et al., 2022; Schaworonkow & Voytek, 2021) and special populations using EEG. As mentioned above, several prior studies have used the slope of the 1/f component/aperiodic activity as an indirect measure of the E/I ratio (Favaro et al., 2023; Hill et al., 2022; McSweeney et al., 2023; Molina et al., 2020; Ostlund et al., 2021; Schaworonkow & Voytek, 2021), including studies using scalp-recorded EEG. We will make more clear in the introduction of the revised manuscript that this metric is indirect.

While a full understanding of aperiodic activity needs to be provided, some convergent ideas have emerged . We think that our results contribute to this enterprise, since our study is, to the best of our knowledge, the first which assessed MRS measured neurotransmitter levels and EEG aperiodic activity.

(3.5) Problems with EEG preprocessing and analysis:

- It seems that the authors did not identify bad channels nor address the line noise issue (even a problem if a low pass filter of below-the-line noise was applied).

As pointed out in the methods and Figure 1, we only analyzed data from two channels, O1 and O2, neither of which were rejected for any participant. Channel rejection was performed for the larger dataset, published elsewhere (Ossandón et al., 2023; Pant et al., 2023).

In both published works, we did not consider frequency ranges above 40 Hz to avoid any possible contamination with line noise. Here, we focused on activity between 0 and 20 Hz, definitely excluding line noise contaminations. The low pass filter (FIR, 1-45 Hz) guaranteed that any spill-over effects of line noise would be restricted to frequencies just below the upper cutoff frequency.

Additionally, a prior version of the analysis used the cleanline.m function to remove line noise before filtering, and the group differences remained stable. We will report this analysis in the supplementary version of the revised manuscript. Further, both groups were measured in the same lab, making line noise as an account for the observed group effects highly unlikely. Finally, any of the exploratory MRS-EEG correlations would be hard to explain if the EEG parameters would be contaminated with line noise.

- What was the percentage of segments that needed to be rejected due to the 120μV criteria? This should be reported specifically for EO & EC and controls and patients.

The mean percentage of 1 second segments rejected for each resting state condition is below. Mean percentage of 6.25 long segments rejected in each group for the visual stimulation condition are also included, and will be added to the revised manuscript:

Author response table 3.

- The authors downsampled the data to 60Hz to "to match the stimulation rate". What is the intention of this? Because the subsequent spectral analyses are conflated by this choice (see Nyquist theorem).

This data were collected as part of a study designed to evoke alpha activity with visual white-noise, which ranged in luminance with equal power at all frequencies from 1-60 Hz, restricted by the refresh rate of the monitor on which stimuli were presented (Pant et al., 2023). This paradigm and method was developed by VanRullen and colleagues (Schwenk et al., 2020; Vanrullen & MacDonald, 2012), wherein the analysis requires the same sampling rate between the presented frequencies and the EEG data. The downsampling function used here automatically applies an anti-aliasing filter (EEGLAB 2019) .

- "Subsequently, baseline removal was conducted by subtracting the mean activity across the length of an epoch from every data point." The actual baseline time segment should be specified.

The time segment was the length of the epoch, that is, 1 second for the resting state conditions and 6.25 seconds for the visual stimulation conditions. This will be explicitly stated in the revised manuscript.

- "We excluded the alpha range (8-14 Hz) for this fit to avoid biasing the results due to documented differences in alpha activity between CC and SC individuals (Bottari et al., 2016; Ossandón et al., 2023; Pant et al., 2023)." This does not really make sense, as the FOOOF algorithm first fits the 1/f slope, for which the alpha activity is not relevant.

We did not use the FOOOF algorithm/toolbox in this manuscript. As stated in the methods, we used a 1/f fit to the 1-20 Hz spectrum in the log-log space, and subtracted this fit from the original spectrum to obtain the corrected spectrum. Given the pronounced difference in alpha power between groups (Bottari et al., 2016; Ossandón et al., 2023; Pant et al., 2023), we were concerned it might drive differences in the exponent values.  Our analysis pipeline had been adapted from previous publications of our group and other labs (Ossandón et al., 2023; Voytek et al., 2015; Waschke et al., 2017).

We have conducted the analysis with and without the exclusion of the alpha range, as well as using the FOOOF toolbox both in the 1-20 Hz and 20-40 Hz ranges (Ossandón et al., 2023); The findings of a steeper slope in the 1-20 Hz range as well as lower alpha power in CC vs SC individuals remained stable. In Ossandón et al., the comparison between the piecewise fits and FOOOF fits led the authors to use the former as it outperformed the FOOOF algorithm for their data.

- The model fits of the 1/f fitting for EO, EC, and both participant groups should be reported.

In Figure 3 of the manuscript, we depicted the mean spectra and 1/f fits for each group. We will add the fit quality metrics and show individual subjects’ fits in the revised manuscript.

(3.6) Validity of GABA measurements and results:

- According the a newer study by the authors of the Gannet toolbox (https://analyticalsciencejournals.onlinelibrary.wiley.com/doi/abs/10.1002/nbm.5076), the reliability and reproducibility of the gamma-aminobutyric acid (GABA) measurement can vary significantly depending on acquisition and modeling parameter. Thus, did the author address these challenges?

We took care of data quality while acquiring MRS data by ensuring appropriate voxel placement and linewidth prior to scanning. Acquisition as well as modeling parameters were constant for both groups, so they cannot have driven group differences.

The linked article compares the reproducibility of GABA measurement using Osprey, which was released in 2020 and uses linear combination modeling to fit the peak as opposed to Gannet’s simple peak fitting (Hupfeld et al., 2024). The study finds better test-retest reliability for Osprey compared to Gannet’s method.

As the present work was conceptualized in 2018, we used Gannet 3.0, which was the state-of-the-art edited spectral analysis toolbox at the time, and still is widely used. In the revised manuscript, we will include a supplementary section reanalyzing the main findings with Osprey.

- Furthermore, the authors wrote: "We confirmed the within-subject stability of metabolite quantification by testing a subset of the sighted controls (n=6) 2-4 weeks apart. Looking at the supplementary Figure 5 (which would be rather plotted as ICC or Blant-Altman plots), the within-subject stability compared to between-subject variability seems not to be great. Furthermore, I don't think such a small sample size qualifies for a rigorous assessment of stability.

Indeed, we did not intend to provide a rigorous assessment of within-subject stability. Rather, we aimed to confirm that data quality/concentration ratios did not systematically differ between the same subjects tested longitudinally; driven, for example, by scanner heating or time of day. As with the phantom testing, we attempted to give readers an idea of the quality of the data, as they were collected from a primarily clinical rather than a research site.

In the revised manuscript we will remove the statement regarding stability, and add the Blant-Altman plot.

- "Why might an enhanced inhibitory drive, as indicated by the lower Glx/GABA ratio" Is this interpretation really warranted, as the results of the group differences in the Glx/GABA ratio seem to be rather driven by a decreased Glx concentration in CC rather than an increased GABA (see Figure 2).

We used the Glx/GABA+ ratio as a measure, rather than individual Glx or GABA+ concentration, which did not significantly differ between groups. As detailed in Response 2.2, we think this metric aligns better with an underlying E/I balance hypothesis and has been used in many previous studies (Gao et al., 2024; Liu et al., 2015; Narayan et al., 2022; Perica et al., 2022).

Our interpretation of an enhanced inhibitory drive additionally comes from the combination of aperiodic EEG (1-20 Hz) and MRS measures, which, when considered together, are consistent with a decreased E/I ratio.

In the revised manuscript, we will rephrase this sentence accordingly. 

- Glx concentration predicted the aperiodic intercept in CC individuals' visual cortices during ambient and flickering visual stimulation. Why specifically investigate the Glx concentration, when the paper is about E/I ratio?

As stated in the methods, we exploratorily assessed the relationship between all MRS parameters (Glx, GABA+ and Glx/GABA+ ratio) with the aperiodic parameters (slope, offset), and corrected for multiple comparisons accordingly. We think this is a worthwhile analysis considering the rarity of the dataset/population (see 1.2, 1.6, 2.1 and reviewer 1’s comments about future hypotheses). We only report the Glx – aperiodic intercept correlation in the main manuscript as it survived correction for multiple comparisons.

(3.7) Interpretation of the correlation between MRS measurements and EEG aperiodic signal:

- The authors wrote: "The intercept of the aperiodic activity was highly correlated with the Glx concentration during rest with eyes open and during flickering stimulation (also see Supplementary Material S11). Based on the assumption that the aperiodic intercept reflects broadband firing (Manning et al., 2009; Winawer et al., 2013), this suggests that the Glx concentration might be related to broadband firing in CC individuals during active and passive visual stimulation." These results should not be interpreted (or with very caution) for several reasons (see also problem with influences on aperiodic intercept and small sample size). This is a result of the exploratory analyses of correlating every EEG parameter with every MRS parameter. This requires well-powered replication before any interpretation can be provided. Furthermore and importantly: why should this be specifically only in CC patients, but not in the SC control group?

We indicate clearly in all parts of the manuscript that these correlations are presented as exploratory. Further, we interpret the Glx-aperiodic offset correlation, and none of the others, as it survived the Bonferroni correction for multiple comparisons. We offer a hypothesis in the discussion section as to why such a correlation might exist in the CC but not the SC group (see response 2.2), and do not speculate further.

(3.8) Language and presentation:

- The manuscript requires language improvements and correction of numerous typos. Over-simplifications and unclear statements are present, which could mislead or confuse readers (see also interpretation of aperiodic signal).

In the revision, we will check that speculations are clearly marked and typos are removed.

- The authors state that "Together, the present results provide strong evidence for experience-dependent development of the E/I ratio in the human visual cortex, with consequences for behavior." The results of the study do not provide any strong evidence, because of the small sample size and exploratory analyses approach and not accounting for possible confounding factors.

We disagree with this statement and allude to convergent evidence of both MRS and neurophysiological measures. The latter link to corresponding results observed in a larger sample of CC individuals (Ossandón et al., 2023).

- "Our results imply a change in neurotransmitter concentrations as a consequence of *restoring* vision following congenital blindness." This is a speculative statement to infer a causal relationship on cross-sectional data.

As mentioned under 2.1, we conducted a cross-sectional study which might justify future longitudinal work. In order to advance science, new testable hypotheses were put forward at the end of a manuscript.

In the revised manuscript we will add “might imply” to better indicate the hypothetical character of this idea.

- In the limitation section, the authors wrote: "The sample size of the present study is relatively high for the rare population , but undoubtedly, overall, rather small." This sentence should be rewritten, as the study is plein underpowered. The further justification "We nevertheless think that our results are valid. Our findings neurochemically (Glx and GABA+ concentration), and anatomically (visual cortex) specific. The MRS parameters varied with parameters of the aperiodic EEG activity and visual acuity. The group differences for the EEG assessments corresponded to those of a larger sample of CC individuals (n=38) (Ossandón et al., 2023), and effects of chronological age were as expected from the literature." These statements do not provide any validation or justification of small samples. Furthermore, the current data set is a subset of an earlier published paper by the same authors "The EEG data sets reported here were part of data published earlier (Ossandón et al., 2023; Pant et al., 2023)." Thus, the statement "The group differences for the EEG assessments corresponded to those of a larger sample of CC individuals (n=38) " is a circular argument and should be avoided.

Our intention was not to justify having a small sample, but to justify why we think the results might be valid as they align with/replicate existing literature.

In the revised manuscript, we will add a figure showing that the EEG results of the 10 subjects considered here correspond to those of the 28 other subjects of Ossandon et al. We will adapt the text accordingly, clearly stating that the pattern of EEG results of the ten subjects reported here replicate those of the 28 additional subjects of Ossandon et al. (2023).

References

Barnes, S. J., Sammons, R. P., Jacobsen, R. I., Mackie, J., Keller, G. B., & Keck, T. (2015). Subnetwork-specific homeostatic plasticity in mouse visual cortex in vivo. Neuron, 86(5), 1290–1303. https://doi.org/10.1016/J.NEURON.2015.05.010

Bernabeu, A., Alfaro, A., García, M., & Fernández, E. (2009). Proton magnetic resonance spectroscopy (1H-MRS) reveals the presence of elevated myo-inositol in the occipital cortex of blind subjects. NeuroImage, 47(4), 1172–1176. https://doi.org/10.1016/j.neuroimage.2009.04.080

Bottari, D., Troje, N. F., Ley, P., Hense, M., Kekunnaya, R., & Röder, B. (2016). Sight restoration after congenital blindness does not reinstate alpha oscillatory activity in humans. Scientific Reports. https://doi.org/10.1038/srep24683

Colombo, M. A., Napolitani, M., Boly, M., Gosseries, O., Casarotto, S., Rosanova, M., Brichant, J. F., Boveroux, P., Rex, S., Laureys, S., Massimini, M., Chieregato, A., & Sarasso, S. (2019). The spectral exponent of the resting EEG indexes the presence of consciousness during unresponsiveness induced by propofol, xenon, and ketamine. NeuroImage, 189(September 2018), 631–644. https://doi.org/10.1016/j.neuroimage.2019.01.024

Consideration of Sample Size in Neuroscience Studies. (2020). Journal of Neuroscience, 40(21), 4076–4077. https://doi.org/10.1523/JNEUROSCI.0866-20.2020

Coullon, G. S. L., Emir, U. E., Fine, I., Watkins, K. E., & Bridge, H. (2015). Neurochemical changes in the pericalcarine cortex in congenital blindness attributable to bilateral anophthalmia. Journal of Neurophysiology. https://doi.org/10.1152/jn.00567.2015

Fang, Q., Li, Y. T., Peng, B., Li, Z., Zhang, L. I., & Tao, H. W. (2021). Balanced enhancements of synaptic excitation and inhibition underlie developmental maturation of receptive fields in the mouse visual cortex. Journal of Neuroscience, 41(49), 10065–10079. https://doi.org/10.1523/JNEUROSCI.0442-21.2021

Favaro, J., Colombo, M. A., Mikulan, E., Sartori, S., Nosadini, M., Pelizza, M. F., Rosanova, M., Sarasso, S., Massimini, M., & Toldo, I. (2023). The maturation of aperiodic EEG activity across development reveals a progressive differentiation of wakefulness from sleep. NeuroImage, 277. https://doi.org/10.1016/J.NEUROIMAGE.2023.120264

Gao, Y., Liu, Y., Zhao, S., Liu, Y., Zhang, C., Hui, S., Mikkelsen, M., Edden, R. A. E., Meng, X., Yu, B., & Xiao, L. (2024). MRS study on the correlation between frontal GABA+/Glx ratio and abnormal cognitive function in medication-naive patients with narcolepsy. Sleep Medicine, 119, 1–8. https://doi.org/10.1016/j.sleep.2024.04.004

Haider, B., Duque, A., Hasenstaub, A. R., & McCormick, D. A. (2006). Neocortical network activity in vivo is generated through a dynamic balance of excitation and inhibition. Journal of Neuroscience. https://doi.org/10.1523/JNEUROSCI.5297-05.2006

Hill, A. T., Clark, G. M., Bigelow, F. J., Lum, J. A. G., & Enticott, P. G. (2022). Periodic and aperiodic neural activity displays age-dependent changes across early-to-middle childhood. Developmental Cognitive Neuroscience, 54, 101076. https://doi.org/10.1016/J.DCN.2022.101076

Hupfeld, K. E., Zöllner, H. J., Hui, S. C. N., Song, Y., Murali-Manohar, S., Yedavalli, V., Oeltzschner, G., Prisciandaro, J. J., & Edden, R. A. E. (2024). Impact of acquisition and modeling parameters on the test–retest reproducibility of edited GABA+. NMR in Biomedicine, 37(4), e5076. https://doi.org/10.1002/nbm.5076

Hyvärinen, J., Carlson, S., & Hyvärinen, L. (1981). Early visual deprivation alters modality of neuronal responses in area 19 of monkey cortex. Neuroscience Letters, 26(3), 239–243. https://doi.org/10.1016/0304-3940(81)90139-7

Juchem, C., & Graaf, R. A. de. (2017). B0 magnetic field homogeneity and shimming for in vivo magnetic resonance spectroscopy. Analytical Biochemistry, 529, 17–29. https://doi.org/10.1016/j.ab.2016.06.003

Keck, T., Hübener, M., & Bonhoeffer, T. (2017). Interactions between synaptic homeostatic mechanisms: An attempt to reconcile BCM theory, synaptic scaling, and changing excitation/inhibition balance. Current Opinion in Neurobiology, 43, 87–93. https://doi.org/10.1016/J.CONB.2017.02.003

Kurcyus, K., Annac, E., Hanning, N. M., Harris, A. D., Oeltzschner, G., Edden, R., & Riedl, V. (2018). Opposite Dynamics of GABA and Glutamate Levels in the Occipital Cortex during Visual Processing. Journal of Neuroscience, 38(46), 9967–9976. https://doi.org/10.1523/JNEUROSCI.1214-18.2018

Liu, B., Wang, G., Gao, D., Gao, F., Zhao, B., Qiao, M., Yang, H., Yu, Y., Ren, F., Yang, P., Chen, W., & Rae, C. D. (2015). Alterations of GABA and glutamate-glutamine levels in premenstrual dysphoric disorder: A 3T proton magnetic resonance spectroscopy study. Psychiatry Research - Neuroimaging, 231(1), 64–70. https://doi.org/10.1016/J.PSCYCHRESNS.2014.10.020

Lunghi, C., Berchicci, M., Morrone, M. C., & Russo, F. D. (2015). Short‐term monocular deprivation alters early components of visual evoked potentials. The Journal of Physiology, 593(19), 4361. https://doi.org/10.1113/JP270950

Maier, S., Düppers, A. L., Runge, K., Dacko, M., Lange, T., Fangmeier, T., Riedel, A., Ebert, D., Endres, D., Domschke, K., Perlov, E., Nickel, K., & Tebartz van Elst, L. (2022). Increased prefrontal GABA concentrations in adults with autism spectrum disorders. Autism Research, 15(7), 1222–1236. https://doi.org/10.1002/aur.2740

Manning, J. R., Jacobs, J., Fried, I., & Kahana, M. J. (2009). Broadband shifts in local field potential power spectra are correlated with single-neuron spiking in humans. The Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 29(43), 13613–13620. https://doi.org/10.1523/JNEUROSCI.2041-09.2009

McSweeney, M., Morales, S., Valadez, E. A., Buzzell, G. A., Yoder, L., Fifer, W. P., Pini, N., Shuffrey, L. C., Elliott, A. J., Isler, J. R., & Fox, N. A. (2023). Age-related trends in aperiodic EEG activity and alpha oscillations during early- to middle-childhood. NeuroImage, 269, 119925. https://doi.org/10.1016/j.neuroimage.2023.119925

Medel, V., Irani, M., Crossley, N., Ossandón, T., & Boncompte, G. (2023). Complexity and 1/f slope jointly reflect brain states. Scientific Reports, 13(1), 21700. https://doi.org/10.1038/s41598-023-47316-0

Medel, V., Irani, M., Ossandón, T., & Boncompte, G. (2020). Complexity and 1/f slope jointly reflect cortical states across different E/I balances. bioRxiv, 2020.09.15.298497. https://doi.org/10.1101/2020.09.15.298497

Molina, J. L., Voytek, B., Thomas, M. L., Joshi, Y. B., Bhakta, S. G., Talledo, J. A., Swerdlow, N. R., & Light, G. A. (2020). Memantine Effects on Electroencephalographic Measures of Putative Excitatory/Inhibitory Balance in Schizophrenia. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 5(6), 562–568. https://doi.org/10.1016/j.bpsc.2020.02.004