A multi‐layered network model identifies Akt1 as a common modulator of neurodegeneration

Abstract The accumulation of misfolded and aggregated proteins is a hallmark of neurodegenerative proteinopathies. Although multiple genetic loci have been associated with specific neurodegenerative diseases (NDs), molecular mechanisms that may have a broader relevance for most or all proteinopathies remain poorly resolved. In this study, we developed a multi‐layered network expansion (MLnet) model to predict protein modifiers that are common to a group of diseases and, therefore, may have broader pathophysiological relevance for that group. When applied to the four NDs Alzheimer's disease (AD), Huntington's disease, and spinocerebellar ataxia types 1 and 3, we predicted multiple members of the insulin pathway, including PDK1, Akt1, InR, and sgg (GSK‐3β), as common modifiers. We validated these modifiers with the help of four Drosophila ND models. Further evaluation of Akt1 in human cell‐based ND models revealed that activation of Akt1 signaling by the small molecule SC79 increased cell viability in all models. Moreover, treatment of AD model mice with SC79 enhanced their long‐term memory and ameliorated dysregulated anxiety levels, which are commonly affected in AD patients. These findings validate MLnet as a valuable tool to uncover molecular pathways and proteins involved in the pathophysiology of entire disease groups and identify potential therapeutic targets that have relevance across disease boundaries. MLnet can be used for any group of diseases and is available as a web tool at http://ssbio.cau.ac.kr/software/mlnet.

Thank you again for submitting your work to Molecular Systems Biology.We have now heard back from the three reviewers who agreed to evaluate your study.As you will see below the reviewers raise substantial concerns, which preclude the publication of the manuscript in its current form.Given that the reviewers did have positive words about the potential relevance of the approach, we have decided to offer you a chance to address the issues raised in a major revision.
Without repeating all the points listed below, some of the more fundamental issues are the following: -The methodological advance and the performance of the method need to be better demonstrated.
-The biological conclusions need to be strengthened by including further experimental and computational analyses.
All issues raised by the reviewers need to be satisfactorily addressed.As you may already know, our editorial policy allows in principle a single round of major revision, so it is essential to provide responses to the reviewers' comments that are as complete as possible.I understand that the required revisions are substantive.Please feel free to contact me in case you would like to discuss in further detail any of the issues raised or if you would like to share your revision plan with me.I would be happy to schedule a call.
On a more editorial level, we would ask you to address the following points: -Please provide a .docversion of the manuscript text (including legends for the main figures) and individual production quality figure files for the main Figures (one file per figure).
-The keywords need to be reduced to 5.
-Please provide a "standfirst text" summarizing the study in one or two sentences (approximately 250 characters), three to four "bullet points" highlighting the main findings and a "synopsis image" (550px width and max 400px height, jpeg format) to highlight the paper on our homepage.
-All Materials and Methods need to be described in the main text.We would ask you to use 'Structured Methods', our new Materials and Methods format, which is mandatory for Methods and Articles with a strong methodological focus.According to this format, the Materials and Methods section should include a Reagents and Tools Table (listing key reagents, experimental models, software and relevant equipment and including their sources and relevant identifiers) followed by a Methods and Protocols section in which we encourage the authors to describe their methods using a step-by-step protocol format with bullet points, to facilitate the adoption of the methodologies across labs.More information on how to adhere to this format as well as downloadable templates (.doc or .xls)for the Reagents and Tools Table can be found in our author guidelines: .An example of a Method paper with Structured Methods can be found here: .
-Please include a "Disclosure and Competing Interests Statement" in the main text.
-The code and the datasets should be made available (we do not allow data available upon request).
-Please include a Data availability section describing how the data, code etc. have been made available.This section needs to be formatted according to the example below: The datasets and computer code produced in this study are available in the following databases: -Chip-Seq data: Gene Expression Omnibus GSE46748 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE46748) -Modeling computer scripts: GitHub (https://github.com/SysBioChalmers/GECKO/releases/tag/v1.-Molecular Systems Biology supports formal data citations in the Reference list, to cite previously published datasets.In addition to citing the original papers that reported the data, we encourage you to also cite the relevant datasets directly in the Reference list.In the text, references to datasets are included as "Data ref: Smith et al, 2001" or "Data ref: NCBI Sequence Read Archive PRJNA342805, 2017".In the Reference list, data citations are very similar to normal literature references but must be labeled with "[DATASET]" at the end of the reference.For detailed instructions please refer to our Author Guidelines . -When you resubmit your manuscript, please download our CHECKLIST (https://bit.ly/EMBOPressAuthorChecklist)and include the completed form in your submission.*Please note* that the Author Checklist will be published alongside the paper as part of the transparent process Overall, it seems this could be a nice program.They demonstrate that it seems superior to other existing programs, and provide functional data of success.There is certainly a great need for this approach and using more global data to get experimental directions.It is also nice to see integration of model organism results to help predict pathways.
My suggestions have to do with clarity for a general biologist.
1-In many cases they discuss specific genes, but they do not include supplemental tables that list them.This would be most transparent and helpful, especially for others looking for interactions.For example, page 6 bottom they list the number of modifier genes they used for the different diseases, but don't include specifically what they are.They include reference to what they are, but it would be more helpful to include the specific list they used here.Similarly for top of page 7. Can provide a supplemental table of the top 100 seeds?I think more transparency will lead to great adoption of this approach.There are other examples of this, that they could include.2-For the Drosophila studies, can they organize the genes by biology as they discuss them in the discussion top of page 15?This would make the data easier to see trends.
3-With these fly genetic interaction data, one wonders if one selected 4-6 random genes NOT on the list, would one also see interactions?If they could do such an experiment, it would be highly compelling for the power of their approach.4-It is striking that the direction of effect in the fly is rather random given the disease.Also, they select an activator for the Akt pathway to use in mammalian cells (and then there it gave the same phenotype for everything, unlike in Drosophila).Can they include more rationale for choosing an activator rather than knocking the pathway down.5-Were the fly, mouse and cell studies performed in blinded manner?They should have been and it would be more compelling if they were.
6-It might be nice to give an example where they following one gene all the way through their analysis and where it is and what its PPI are, etc.This may allow more clarity for basic biologists.
Reviewer #2: Na & Lim et al. investigates the common gene modifiers across four different neurodegenerative diseases (NDs): Alzheimer's disease (AD), Huntington's disease, and Spinocerebellar ataxia types 1 and 3.The authors used a model they developed, the multi-layered network expansion (MLnet), to predict protein modifiers common to these diseases, identifying several members of the insulin pathway, including PDK1, Akt1, InR, and GSK-3β.Validation was carried out using Drosophila ND models and human cell-based ND models.They showed that the activation of Akt1 signaling by a small molecule, SC79, increased cell viability and improved memory and anxiety levels in AD model mice.Overall the study nicely combined computational, Drosophila, and in vivo evidence and was well-formulated/written.

General Remarks
The paper provides some compelling results and insights about the common molecular mechanisms across neurodegenerative diseases, particularly pointing towards the insulin pathway.It appears to place the work in the broader context of investigating common proteinopathies, although the reviewer is not entirely convinced that the identified modifier mediates these diseases through proteinopathy-related pathways (could it be general cognition mediated through the insulin pathway?) Overall, this study could be seen as a significant advance, specifically with the validation of the MLnet tool for computational biologists-the reviewer applauds the authors' efforts to create the webtool & think the authors should highlight it in the Abstract to increase impact.This research might also be of interest to neurobiologist albeit the reviewer does not have much expertise to judge from that front.

Major Points
• The work is predicated on the accuracy and validity of the MLnet model to generate good disease modifiers, which based on the authors assessment seem robust.But garbage-in-garbage-out, and the reviewers would like the authors to include lists of seed genes used.Also do you get better modifiers if you include just for example GWAS genes?
• There are implications that the insulin pathway could be a common pathway for these NDs, yet the mechanistic link between the insulin pathway and protein aggregation in these etiologies remains vague.Is it possible that the insulin pathway is linked to cognitive functions in general?This connection should be further explored and clarified both computationally and in experiments.

Minor Points
The effectiveness of SC79 is intriguing, particularly with in vitro and in vivo data.The reviewers caution that in cancer the PI3Kakt signaling is often activated and the authors will likely want to monitor the possible tumorigenic effects.Even if the compound can be made brain-specific, glioblastoma could be a concern as well.
Reviewer #3: MSB-2023-11801 "Multi-layered network expansion model identifies Akt1 as common modulator of neurodegeneration" by Na et al.In this m/s, the authors present MLnet, a new methodology (based on a multi-layered network expansion strategy) to identify protein modifiers common to several neurodegenerative diseases.In particular, the authors apply it to Alzheimer´s (AD) and Huntington (HD) diseases as well as spinocerebellar ataxias type 1 and 3 (SC1, 3).The authors tested the 12 top-scoring proteins on fly disease models for the AD, HD, SC1 and SC3, finding that 4 of them showed a phenotypic effect.They also found that these genes were related to the insulin pathway, and selected Akt1 (as a central gene in this pathway) for further validation on HEK293 cells and the 5xFAD mouse model.The m/s covers an interesting area of research, and it is wellstructured and clearly written.However, there are a few conceptual and methodological aspects that need to be clarified and/or corrected.Without these clarifications, it is difficult a assess the added value of the MLnet method and the novelty of the results presented in the paper.In its present forms, I fail to see the rationale behind the experiment and the advance (methodologic and/or scientific) provided by the m/s.Please, find below some specific comments that might help the authors to improve the quality and clarity of their m/s.I understand the appeal of finding a common mechanism (modulator) for all neurodegenerative diseases, but I fail to see any support for this hypothesis.May be the authors could elaborate a bit more on the evidence leading to this research.I also fail to see the rationale, or advantage, in identifying a common drug target.Do these diseases manifest together?And, in case that there are common mechanisms/modulators, are these expected to be unique for neurodegenerative diseases, or are also shared in other complex diseases?
To report the MLnet performance results, the authors always refer to the area under the curve (AUC), but they do not specify to which curve they refer.Given the high values reported, I assume it is the AUROC curve, but they should clarify it.If this is correct, ROC curves are seriously affected by data imbalance (i.e. if the number of positives is small, you can get high AUROC values predicting that everything is a negative).The authors should clearly describe the ratio between modulators and nonmodulators in every set, and also report individual values of precision and recall (or the AUPRC).
It is also not clear how the MLnet results compare to just using the common gene set identified for each individual ND.I understand that the network propagation might find more common genes, but it is unclear whether these would indeed be relevant (i.e.real modulators) for each disease.The authors should also compare if the MLnet identified common ND modulators have been reported to be associated to the individual diseases.Without this, it is difficult to assess the added value of the methodology.
The common ND modifiers identified by MLnet are enriched in apoptosis, autophagy and mitophagy.How about the seed modifiers used for each disease?If they are not, this is likely to be an artifact of just identifying very well connected general biological processes.To avoid this potential bias, the authors should also use randomized networks, preserving the connection degree of the proteins and shuffling the edges.Calculating enrichments using only the number of genes in a process identified might not be enough.It would also be interesting to see if these general processes are somehow specific for NDs, or they are found in other unrelated complex diseases which, again, could just be the result of their centrality.
4 out of the 12 potential modulator proteins tested in flies gave a signal.However, it is unclear what is the added value of MLnet in identifying common modulators.How close to the top are these 4 (or 12) proteins in the individual disease models?What would be the result if the authors would pick the top proteins in each individual ND and test them in all four fly models?Is the enrichment in the insulin pathway detected in the individual diseases (see point above)?What about other genes in the pathway not selected as common modifiers, would they show an effect?I do not really understand the experiment in HEK293 cells.Why are cells expressing ND-related genes expected to be less viable?Specific ND-related genes do not affect cell viability (and certainly not in HEK293 cells).The authors should clarify this experiment and the interpretation of the results.
In the last section, the authors kind of drop all the previous results and only focus on testing the effect of the insulin pathway in Alzheimer´s disease which, as they state, has been clearly established!I fail to see the added value of re-reporting that the modulation of the insulin pathway has an effect on the 5xFAD mouse model.If anything, they should test the effect of this modulation of models other that AD (i.e.HD or the SCs).The authors also say that GSK3b is the only identified AD-HD modifier not directly involved in the insulin pathway, but there is an infinity of literature relating GSK3b and insulin levels.Moreover, GSK3b inhibitors have been extensively tested to treat AD, showing very promising results in mice models, but failing the clinical trials.
Finally, the number of animals used in the Barnes maze test (4/5 mice) seems insufficient.The learning curve for the AD model is not really consistent in the days 3,4 and 5.The same effect is observed in the test days (15 and 16).Although, Fig 5G shows a mild significance when comparing days 5 and 16 between AD and AD+SC79, this difference should be supported by a robust learning curve, or any small change (i.e.adding one more animal or changing the days compared) could wipe it out.

Point-by-point responses
First of all, we would like to thank the reviewers for their encouraging comments and valuable suggestions.We believe that, by addressing the reviewers' concerns, a considerably stronger manuscript has emerged.Please find our point-by-point responses below.

Responses to Reviewer #1
1-In many cases they discuss specific genes, but they do not include supplemental tables that list them.This would be most transparent and helpful, especially for others looking for interactions.For example, page 6 bottom they list the number of modifier genes they used for the different diseases, but don't include specifically what they are.They include reference to what they are, but it would be more helpful to include the specific list they used here.Similarly for top of page 7. Can provide a supplemental table of the top 100 seeds?I think more transparency will lead to great adoption of this approach.There are other examples of this, that they could include.
We appreciate this valuable suggestion.We have now added three supplementary tables (Tables S1, S2 and S3) to the manuscript.Table S1 includes the genetic modifiers that we obtained from the NeuroGeM database and used to predict disease-specific modifiers in the first module.Table S2 includes the top 100 predicted disease-specific modifiers, i.e., seed genes used for the second module of MLnet.Table S3 contains the top100 common modifiers predicted by MLnet.2-For the Drosophila studies, can they organize the genes by biology as they discuss them in the discussion top of page 15?This would make the data easier to see trends.
We thank the reviewer for this great suggestion.We have rearranged the eye phenotype images in the revised with GMR-GAL4-driven misexpression of the ND-causing genes, along with GMR-GAL4-driven RNAi against each of the indicated genes in which mCherry served as a control.Compared with the wild-type (WT) compound eye with the ordered structure of ommatidia, flies with misexpression of individual ND-causing genes and RNAi against mCherry under the control of the GMR-GAL4 driver had rough eyes with the variation in phenotypic severity.Suppression or enhancement of these rough eye phenotypes caused by RNAi-mediated knockdown of the predicted common modifiers is indicated with + or -, respectively.As a negative control, two genes (Cyp6a18 and CG34372) randomly selected among low-ranked genes were also tested.
3-With these fly genetic interaction data, one wonders if one selected 4-6 random genes NOT on the list, would one also see interactions?If they could do such an experiment, it would be highly compelling for the power of their approach.This is an excellent suggestion.We agree that evaluating low-ranked genes as negative controls can provide further support for the power of our approach.Following this reviewer's suggestion, we selected two genes randomly from the bottom end of the ranking list (MLnet scores) and assessed their impact on the eye phenotype in the different ND models.As expected, these low-ranked genes did exhibit minimal to no modifier capability when altered in expression.We have included these results in Four of the 12 tested common modifier proteins (Akt1, InR, Pdk1, and sgg (GSK3β)) changed the eye phenotypes in all four ND models when down-regulated by RNAi.Interestingly, three of these (Akt1, InR, and Pdk1) are directly involved in insulin signaling and one of them (sgg) acts downstream of the insulin signaling pathway (Fig 3C).As a negative control, we also evaluated the effect of two randomly selected low-ranked genes (Cyp6a18 and CG34372) and found them to have little to no impact across the four Drosophila ND models (Fig 4).To validate the top 12 ranked proteins further, we searched the literature for evidence that supports the impact of them in specific NDs (Table S5) and, indeed, we could find evidence across diseases for many of these proteins.
4-It is striking that the direction of effect in the fly is rather random given the disease.Also, they select an activator for the Akt pathway to use in mammalian cells (and then there it gave the same phenotype for everything, unlike in Drosophila).Can they include more rationale for choosing an activator rather than knocking the pathway down.
The varying effects of modifiers on the eye phenotype is, indeed, an interesting finding.However, the Drosophila eye phenotype is determined by complex cellular processes, which are regulated differently in various contexts (diseases) due to distinct gene expression profiles and interactions.As we discuss in the manuscript (page 18), there exist multiple previous studies in which opposing effects of the same gene were reported.Effects on a specific phenotype (e.g., eye morphology, cell viability, etc.) can vary depending on the method and level of gene expression modulation and the disease under investigation.In our previous study in which we assembled a database of genetic modifiers of neurodegenerative diseases and analyzed the collected data, we found many instances in which opposite effects had been reported for the same gene (Na et al, 2013).Importantly, the consensus among experts who use genetic screens (particularly in D. melanogaster) is that it is hard to decide based on the screen results whether a gene is an enhancer or suppressor.Therefore, the results of the D. melanogaster experiments (Fig 4) should be interpreted as evidence for the ability of the positively-tested genes to act as disease modifiers rather than enhancers or suppressors.
Regarding Akt1, there is increasing evidence that increased activity of insulin signaling can enhance the protection of neuronal cells.Since decreased activity of or resistance in insulin signaling is often found in the patients of AD, there has been clinical attempts/trials to treat AD with nasal administration of insulin.Therefore, we hypothesized that activation of the insulin signaling pathway could alleviate neurodegenerative phenotypes.As there is an activator for Akt1, we chose Akt1 as a target to modulate.We updated the main text to reflect our motivation in choosing an Akt1 activator.

(Page 18)
Consistent with these previous studies, we find that all 12 top-ranked proteins, when suppressed in expression, modulate disease phenotype in at least two of the tested models.However, only proteins that are part of the insulin pathway affect phenotypes in all of the tested D. melanogaster disease models.Interestingly, supressing the expression of these modifiers enhances the phenotype in some ND models, while it reduces it in others.These observations are consistent with previous studies where overexpression of the same gene can have opposing effects on the phenotypic of different ND when tested in fly models (Branco et al, 2008).These differences are explained by the fact that the impact of genetic modulation on ND phenotypes is highly dependent on the method and level of modulation and the complexity of the pathophysiology of the individual ND (Na et al, 2013).Therefore, the results of the D. melanogaster experiments (Fig 4) should be interpreted as evidence for the ability of the positively-tested genes to act as disease modifiers rather than enhancers or suppressors.
(Page 15) Motivated by these findings, we aimed to test the disease-modifying impact of insulin signaling in mammalian models of ND.Since decreased activity of or resistance in insulin signaling is commonly found in the patients of AD, we hypothesized that activation of the insulin signaling pathway could alleviate neurodegenerative phenotypes.We chose Akt1 as a target for insulin signaling modulation because of its central position in this pathway.Akt1 is not ranked very high in the disease-specific modifier lists, with the exception of SCA1 (Table S5), but is second in the final ranking of common modifiers due to its interaction with many proteins that are themselves disease modifiers, i.e. partners that are highly ranked in the disease-specific modifier lists of module 1 (see Appendix text for details).Moreover, the availability of an activator of this kinase enables induction of downstream insulin signaling (Jo et al, 2012).
5-Were the fly, mouse and cell studies performed in blinded manner?They should have been and it would be more compelling if they were.
The Drosophila experiments were performed in a blinded manner.Following the acquisition of eye images, the levels of rough eyes were scored without knowledge of the modifiers' names.For cell and mice experiments, all in vitro assays and behavioral tests were also conducted in a blinded manner.In addition, the mouse behavior was analyzed with the ANYmaze software, a behavioral test analysis program that aims to reduce the risk of bias.This is now mentioned in Materials and Methods.We thank the reviewer for this comment.The optical density (OD) of each well was measured using a microplate reader at 450 nm (Molecular Devices, CA, USA), and the OD values were reported as % cell viability (mean ± standard error, n = 4 -8 per group).The in vitro assays were performed in a blinded manner.
(Page 30) All trials were recorded and analyzed by ANY-maze 6.0 Software.All behavioral tests were conducted in a blinded manner and the ANY-maze software was used to avoid any bias in the analysis.
6-It might be nice to give an example where they following one gene all the way through their analysis and where it is and what its PPI are, etc.This may allow more clarity for basic biologists.
Akt1 has been tested in models of HD and SCA1, but not of AD and SCA3 (Table S7, below).Akt1's assigned confidence score for the experimental results in HD and SCA1 models are 0.60 and 0.72, respectively.The high confidence scores compared with other modifiers (half of HD modifiers have a confidence score lower than 0.2) are due to the fact that Akt1 was identified as a modifier in several low-throughput experiments.
Table S7.Akt1 as a modifier and its confidence score in NDs

Ranks of Akt1 in the predicted disease-specific modifiers
Among disease-specific modifiers predicted by the first module of MLnet, Akt1 ranks only within the top 100 in SCA1 (Table S8, see below).However, Akt1 ranks second among common modifiers, which shows that the multi-layered networks approach of the second module of MLnet significantly changes the ranks.For comparison, two additional Drosophila genes are listed that are ranked very low in the individual diseases (not tested experimentally), don't get a high common modifier ranking and have a similar number of protein interaction partners as Akt1 in the network.
Table S8.Ranks of Akt1, CG7470, and CG10749 within predicted disease-specific modifiers 3. Ranks of Akt1's interaction partners in the predicted disease-specific modifiers.
In MLnet, the common modifier score (and, thus, the rank) of a protein generated in module 2 depends on its own seed score (from module 1) as well as the seed scores (from module 1) and the updated common modifier score of its interaction partners in the network (see Fig S2).As Akt1 acquires a high common modifier score but is initially not top ranked in individual diseases, it can be assumed that it gets this high final score because the scores of Akt1's interaction partners are high across the different diseases, higher than those of proteins that don't get pushed to the top of the common modifier list by module 2. Therefore, we investigated the distribution of the ranks of interacting partners of Akt1 and two Drosophila genes (CG7470 and CG10749) that are not ranked high among predicted common modifiers (they are ranked 1102 and 4827, respectively).Consequently, Akt1's rank in the list of predicted common modifiers is high in contrast to two other genes that have a similar number of interaction partners as Akt1 but much fewer interaction partners that are top ranked in individual disease and used as seeds.

Appendix Figure S14. Histogram of interaction partner ranks.
Interaction partners of Akt1 (blue), CG7470 (red), and CG10749 (green) in the predicted diseasespecific modifier lists are binned according to their ranks and the count per bin is provided.Ten bins are used for the top 100 ranks of each ND.
Please Appendix Text S1 for the updated information.

Responses to Reviewer #2:
Overall, this study could be seen as a significant advance, specifically with the validation of the MLnet tool for computational biologists-the reviewer applauds the authors' efforts to create the webtool & think the authors should highlight it in the Abstract to increase impact.This research might also be of interest to neurobiologist albeit the reviewer does not have much expertise to judge from that front.
We appreciate the reviewer's encouraging comment.As suggested, we revised the Abstract to emphasize the availability of the new web-tool.

Abstract
The accumulation of misfolded and aggregated proteins is a hallmark of neurodegenerative proteinopathies.
Although multiple genetic loci have been associated with specific neurodegenerative diseases (NDs), molecular mechanisms that may have a broader relevance for most or all proteinopathies remain poorly resolved.In this study, we developed a multi-layered network expansion (MLnet) model to predict protein modifiers that are common to a group of diseases and, therefore, may have broader pathophysiological relevance for that group.
When applied to the four NDs, Alzheimer's disease (AD), Huntington's disease, and Spinocerebellar ataxia types 1 and 3, we predicted multiple members of the insulin pathway, including PDK1, Akt1, InR, and sgg (GSK-3β), as common modifiers.We validated these modifiers with the help of four Drosophila ND models.Further evaluation of Akt1 in human cell-based ND models revealed that activation of Akt1 signaling by the small molecule SC79 increased cell viability in all models.Moreover, treatment of AD model mice with SC79 enhanced their long-term memory and ameliorated dysregulated anxiety levels, which are common symptoms of AD patients.These findings validate MLnet as a valuable tool to uncover molecular pathways and proteins involved in the pathophysiology of entire disease groups and identify potential therapeutic targets that have relevance cross disease boundaries.MLnet can be used for any group of diseases, and is available as a web tool at http://ssbio.cau.ac.kr/software/mlnet.

Major Points
• The work is predicated on the accuracy and validity of the MLnet model to generate good disease modifiers, which based on the authors assessment seem robust.But garbage-ingarbage-out, and the reviewers would like the authors to include lists of seed genes used.
We agree with the reviewer's concern about the reliability of the input data.We collected the known genetic modifiers from the NeuroGeM database that we have built previously, which contained a manually curated list of experimentally confirmed genetic modifiers of Drosophila.Importantly, we developed a confidence score that is based on the type and number of experiments showing that a specific gene is a genetic modifier of a specific neurodegenerative disease.We only used high-confidence modifiers as input to MLnet.We added three supplementary tables (Tables S1, S2 and S3) to the manuscript.Table S1 provides the list of known, high-confidence genetic modifiers of the four NDs obtained from the NeuroGeM and used as initial input to MLnet.Table S2 provides the 100 top-ranked predicted disease-specific modifiers from module 1 of MLnet that were used as seed genes for the prediction of common genetic modifiers in module 2. Table S3 contains the top100 common modifiers predicted by MLnet.
• Also do you get better modifiers if you include just for example GWAS genes?This is a great suggestion.We actually used GWAS genes when testing MLnet on human data.We have predicted human modifier genes common to AD and HD using the genes deposited in Neurocarta.Neurocarta compiles not only genes identified using animal and cell-based models but also GWAS genes associated with neurological disorders.As shown in Fig S11, the PI3K-Akt signaling (i.e., the insulin signaling) is most highly enriched within the predicted human AD/HD common genes.Notably, GWAS genes should be carefully included because of the high number of false positives.
neurodegenerative phenotype in the four Drosophila models.The newly added results confirm that it is a common modifier.
Appendix Figure S13.Effect of Pi3K92E knockdown on Drosophila eye phenotype in four ND models.
We now discuss the involvement of insulin in brain plasticity and cognition in much more detail.We highlight our finding that amyloid-b in the brain of AD mice treated with SC79 is not decreased compared to mice that were not treated, which may suggest that, besides reduced apoptosis, insulin pathway-mediated effects on brain plasticity contribute to the improved cognitive test results seen for the SC79-treated mice.To provide further support to this idea, we also analysed genes that have recently been associated with human adult cognitive function (Chen et al, 2023).Of the eight genes that have been associated with human adult cognitive function through rare coding variants with large effects, four had previously been shown to affect insulin and the insulin pathway, although this links has been established with peripheral and not cerebral insulin (Backe et al, 2019;Giovannone et al, 2003;Hamming et al, 2010;Gonzalez et al, 2022).
Concluding, we now discuss the possibility of changes in cognition in more detail and provide additional data in support of this idea.However, as much as we are curious to know more about this possible explanation for the importance of the insulin pathway, as is this reviewer, we feel that more experiments in that direction are outside the scope of this study, because the main focus is on introducing MLnet and not on revealing the intricate molecular details of the interplay between insulin signaling, brain plasticity and cognitive decline in neurodegeneration.

References:
Backe We updated the text accordingly: Given the positive testing of all four genes related to the insulin pathway, we decided to assess the impact of another insulin pathway protein on the disease phenotypes.We evaluated Pi3K92E (PI3K in Fig 3C ) because it is within the top 20 of the predicted common modifiers.Moreover, PI3K is of particular interest because it is one of the key mediators of the insulin pathway's impact on brain plasticity and neurogenesis.For instance, PI3-kinase is essential for glutamate receptor insertion at plasma membranes during synaptic plasticity (Man et al, 2003).Downregulation of Pi3k92E changed the eye phenotype in all four ND models (Fig S13), confirming the significance of insulin signaling for the model phenotypes investigated here.

(Page 19)
Specifically, insulin regulates glucose homeostasis and maintains energy requirements for different neuronal functions.It is vital for neuronal growth and differentiation as well as neuroprotection by modulating autophagy, mitochondrial function, ER stress, and apoptosis (Pomytkin et al, 2018;Burillo et al, 2021).Thus, dysfunction of insulin signaling makes neuronal cells vulnerable to metabolic and cellular stresses (Kim & Feldman, 2015).Moreover, the insulin signaling pathway plays key roles in brain plasticity, impacting cognitive functions such as learning and memory (Spinelli et al, 2019).In the hippocampus, for instance, insulin positively impacts synaptic and structural plasticity.Recently, eight genes have been associated with human adult cognitive function through rare coding variants with large effects.Four of these eight genes had previously been shown to affect insulin and the insulin pathway, although this links has been established with peripheral and not cerebral insulin (Chen et al, 2023;Backe et al, 2019;Giovannone et al, 2003;Hamming et al, 2010;González et al, 2022).
(Page 20) When we investigated the levels of amyloid-b in the brain of AD mice treated with or without SC79, there were no significant changes in amyloid-b levels whether insulin signaling was activated or not (Fig S15).As activated insulin signaling improved cell viability in in-vitro assays (Fig 5B-E), this may suggest that activation of Akt1 in our experiments may have enhanced anti-apoptotic effects while impacting autophagy to a lesser extent.Insulin actually experts anti-apoptotic effects via Akt1, which reduces the mitochondrial release of cytochrome c (Li et al, 2009;Kang et al, 2003).Alternatively, activation of Akt1 may be beneficial via its regulatory impact on cognitive functions (Spinelli et al, 2019).

Minor Points
The effectiveness of SC79 is intriguing, particularly with in vitro and in vivo data.The reviewers caution that in cancer the PI3K-akt signaling is often activated and the authors will likely want to monitor the possible tumorigenic effects.Even if the compound can be made brain-specific, glioblastoma could be a concern as well.
We fully agree with the reviewer.We already mentioned this caveat in the original manuscript.There are actually Akt1 inhibitors in clinical trials for the treatment of various cancers.Intriguingly though, there exists an inverse correlation between AD and cancer incidence (Ospina-Romero et al, 2020).AD patients are less prone to cancer.Notably, this inverse relationship between AD and cancer incidence becomes more pronounced with age.As cerebral insulin and PI3K-Akt activities have commonly been found reduced in AD patients, one may speculate that there exists a therapeutic window for controlled activation, if a brain-specific Akt1 activator could be developed.
Ospina-Romero M, Glymour MM, Hayes-Larson E, Mayeda ER, Graff RE, Brenowitz WD, Ackley SF, Witte JS & Kobayashi LC (2020) Association between Alzheimer disease and cancer with evaluation of study biases: A systematic review and meta-analysis.JAMA Netw Open 3: e2025515 In any case, we now make the potential danger clearer: (Page 20) Moreover, although our experiments suggest a modulatory role of Akt1 for multiple NDs, it may not be an ideal target for ND treatment development because of its involvement in numerous cellular processes and the fact that its enhanced activation can lead to cancerous cell transformation (Wang et al, 2017), which would require very close monitoring for tumorigenic effects when activated via a therapeutic agent.
1.I understand the appeal of finding a common mechanism (modulator) for all neurodegenerative diseases, but I fail to see any support for this hypothesis.May be the authors could elaborate a bit more on the evidence leading to this research.I also fail to see the rationale, or advantage, in identifying a common drug target.Do these diseases manifest together?
We are sorry that we did not make this point clear enough in the initial submission.Intracellular protein misfolding and aggregation are features that are common to essentially all late-onset neurodegenerative diseases, called proteinopathies.These include Alzheimer's disease, Parkinson's disease, tauopathies, forms of motor neuron disease (e.g., TDP-43 mutations), and the nine polyglutamine expansion diseases exemplified by Huntington's disease (HD), spinocerebellar ataxia type 1 (SCA1), and SCA3.Protein homeostasis (proteostasis) is crucial to the prevention of protein aggregation and has been demonstrated to decline with age and in proteinopathies.In this context, it is important to note that only a minority of proteinopathies occur as familial forms, i.e., they are caused by disease-causing mutations of high penetrance that are familial and can be directly linked to the proteins that misfold and aggregate.The majority of these diseases are idiopathic in nature.Even for the case of familial proteinopathies, genetic variation has been shown to affect the phenotype.Indeed, only between 40 and 70 % of the variance in the age of onset of HD and SCA can be accounted for by CAG repeat number in the disease-causing allele.Thus, significant research has been carried out to identify genes that increase the likelihood of proteinopathies, lower the age of onset and enhance the disease severity (called modifiers), predominantly via genetic screen in model organisms.Maybe not surprisingly, functional enrichment analyses of identified modifiers revealed that many of them are associate with proteostasis, i.e., defects in proteostasis may enhance generally the likelihood for protein aggregation.Importantly, modifiers of one proteinopathy can influence another -e.g., a significant fraction of SCA3 modifiers in Drosophila had similar effects in tau (Alzheimer) models (please check Figure 2).Therefore, we hypothesised that there is a subset of genetic modifiers that affects the severity of one of these diseases that has broader relevance and may modify several other or even all proteinopathies.One may expect that at least some of these common modifiers have functions associated with proteostasis, given the importance of proteostasis in preventing aggregation.
We believe that identifying such an "Achilles heel" is not only relevant for a better understanding of the pathophysiology of proteinopathies but may also be useful from a disease monitoring and therapeutic points of view.Monitoring activity of common modifiers may serve as biomarkers of proteostatic or other relevant cellular activity and as indicators for disease risk across multiple proteinopathies.Moreover, altering the activity of the common modifiers directly or indirectly may slow disease progression or delay the age of disease onset independent of the type of proteinopathy.
We revised our manuscript to clarify this important point: NDs belong to the ever-growing group of diseases called proteinopathies (Hipp et al, 2014), because intracellular protein misfolding and aggregation are common to these diseases.Protein homeostasis (proteostasis) is crucial to the prevention of protein aggregation and has been demonstrated to decline with age and in proteinopathies.(Labbadia & Morimoto, 2015;Balch et al, 2008;Hipp et al, 2019).Given the fact that protein misfolding and aggregation is common to proteinopathies and modifiers of one proteinopathy can influence another, e.g., a significant fraction of SCA3 modifiers in Drosophila had similar effects in Alzheimer models, we hypothesized that there may exist a subset of genetic modifiers that has broader relevance and may modify several or even all proteinopathies.Such common or generic modifiers may be central hubs in proteostatic control or key regulators of the cellular stress response.A bioinformatics analysis that we carried out on existing modifier sets previously revealed, however, only a small and incoherent set of modifiers that were identified in multiple ND models (Na et al, 2013), which may be due to the limited power and coverage of high-throughput screens for modifiers.Therefore, we set out to develop a robust computational framework that, with the help of data integration, predicts protein modifiers common to multiple diseases.We believe that identifying modifiers is not only relevant for a better understanding of the pathophysiology of proteinopathies but may also be useful from a disease monitoring and therapeutic points of view.Common modifiers may serve as biomarkers, and monitoring their activity indicate disease risk across multiple proteinopathies.Moreover, altering the activity of the common modifiers directly or indirectly may slow disease progression or delay the age of disease onset independent of the type of proteinopathy.
There are several reviewer comments that relate to the same or related issues, so we merged them.
2-1.And, in case that there are common mechanisms/modulators, are these expected to be unique for neurodegenerative diseases, or are also shared in other complex diseases?2-2.Calculating enrichments using only the number of genes in a process identified might not be enough.It would also be interesting to see if these general processes are somehow specific for NDs, or they are found in other unrelated complex diseases which, again, could just be the result of their centrality.2-3.Is the enrichment in the insulin pathway detected in the individual diseases (see point above)?
We appreciate these important comments.Our MLnet does not show any significant bias toward network hub proteins, which are known to play roles across multiple diseases (please refer to our response to Point 5).Nevertheless, to investigate this point further and follow the reviewer's suggestion, we investigated whether the insulin signaling pathway would automatically come up as common to another disease group because of its centrality in cellular metabolism.To the end, we collected disease-related genes of three inflammatory diseases (gastroenteritis, hepatitis, and dermatitis).We hypothesized that inflammatory diseases have a set of commonly associated genes that are distinct from the common modifiers of proteinopathies and, more importantly, that these genes have different functions.
Consistent with this hypothesis, we find inflammation-and immune response-related pathways enriched among the predicted common modulators of inflammatory diseases and no enrichment for the insulin or related pathways.
Regarding the enrichment of insulin pathway genes as modifiers of individual disease, such an enrichment may not generally be found for all NDs.It should be so if the genetic screens carried out for different disease cover the same set of genes.However, this is generally not the case.If only few insulin pathway-related genes were tested in screens of a specific disease, it is less likely that a GO enrichment analysis will show the insulin pathway as significant.This difference in data coverage is one of the reasons why we developed MLnet.
Integrating the data across disease should help reduce those imbalances and help reweight genes that were not or only rarely tested in some diseases.
(Page 13-14) As the insulin signaling pathway plays a central role in metabolism and the proteins that are part of it interact with many partners, we investigated whether this pathway would automatically come up in our network-based approach even when using genes associated with diseases not related to neurodegeneration.To this end, we collected disease-related genes of three inflammatory diseases (gastroenteritis, hepatitis, and dermatitis) and tested for annotations enriched among the top-ranked proteins predicted by MLnet to be common to these diseases.Specifically, we collected 265, 146 and 442 genes associated with gastroenteritis, hepatitis, and dermatitis, respectively, from Neurocarta (Portales-Casamar et al, 2013) and submitted them to MLnet.Among the top 100 proteins predicted to be associated with all three inflammatory diseases, pathways related to the immune response and inflammation are significantly enriched (Fig S12 and Table S6) The performances of MLnet in the prediction of modifiers common to different disease groups was assessed using different numbers of seed proteins.Diseases were grouped as indicated and common modifiers for that group were predicted with MLnet.As ground truth served the intersection of high-confidence genetic modifiers that were identified experimentally for each disease in that group.The total number of highconfidence modifiers of each ND are: 113 for AD, 209 for HD, 36 for SCA1, and 59 for SCA3.The number of experimentally found common modifiers for each group is given in parenthesis.AUROC was calculated by leave-one-out-cross-validation and the bars are mean ± standard error (n = 3).The results for different numbers of seeds are shown as blue bars.As controls, we also assessed the performance of a simple gene prioritization approach (N), GeneMania (G), and Endeavour (E).
4. It is also not clear how the MLnet results compare to just using the common gene set identified for each individual ND.I understand that the network propagation might find more common genes, but it is unclear whether these would indeed be relevant (i.e.real modulators) for each disease.The authors should also compare if the MLnet identified common ND modulators have been reported to be associated to the individual diseases.Without this, it is difficult to assess the added value of the methodology.
We appreciate this comment.There are several aspects to unpack here.There are the diseasespecific modifiers that were identified experimentally and those that we predict with the first module of MLnet.
Regarding the experimentally determined ones, we compared them when we assembled the NeuroGeM database (Na et al. 2013) and found very little overlap across multiple NDs.Indeed, we found only four genes that were identified as modifiers across five NDs and these genes are involved in very different cellular processes (some related to proteostasis).One key reason for this limited overlap is that, as alluded to in the response to question 2, the set of genes screened in different experiments for different diseases are quite different and never cover the entire genome.In other words, not all genes have been tested for all NDs, often only a subset of the entire genome.This is one of the main reasons why we developed module 1 of MLnet, which gives a disease-specific modifier score to all genes.Regarding predicted disease-specific modifiers, one can, as suggested by the reviewer, just use a prioritization approach as in module 1 to predict disease-specific modifiers and then search for overlap across diseases.We did that in Fig 2 (column N).This analysis revealed that this approach is less effective than the network-expansion approach implemented in module 2 of MLnet.To illustrate this aspect further, we traced the disease-specific ranks of the top 12 common modifiers shown in Fig 4 .As revealed in the table below (Table S5), all of these genes are ranked much lower in some of the specific NDs.Thus, taking just the top ranked 200 genes of each disease and select common ones would not provide the result we achieved with MLnet.
We agree with the reviewer that it is important to provide orthogonal evidence for the validity of predicted common modifiers in each disease; that is why we tested them in the Drosophila models.In addition, we now searched the literature for evidence that supports the impact of individual genes in specific diseases (see references for each gene and disease in the table below; Table S5).We could find evidence across diseases for many of the genes.However, it needs to be stressed that not all genes may have been assessed for their impact on models of each of the diseases in the focus here.Again, this is a reason why we developed MLnet.

1584
* Predicted ranks and their evidence (PubMed IDs) for disease association are shown.Green color, experimentally discovered modifiers; Grey color, potential association; white color, no evidence.
We updated the text as follows: (Page 12) Second, we traced the disease-specific ranks of the top 12 common modifiers predicted by MLnet, which we will discuss and experimentally validate in the following sections.Most of these genes are not ranked in top 200 of at least one of the four NDs (Table S5).Thus, taking just the top ranked 200 genes of each disease and select common ones would not provide the result we achieve with MLnet.

(Page 14)
Four of the 12 tested common modifier proteins (Akt1, InR, Pdk1, and sgg (GSK3β)) changed the eye phenotypes in all four ND models when down-regulated by RNAi.Interestingly, three of these (Akt1, InR, and Pdk1) are directly involved in insulin signaling and one of them (sgg) acts downstream of the insulin signaling pathway (Fig 3C).As a negative control, we also evaluated the effect of two randomly selected low-ranked genes (Cyp6a18 and CG34372) and found them to have little to no impact across the four Drosophila ND models (Fig 4).To validate the top 12 ranked proteins further, we searched the literature for evidence that supports the impact of them in specific NDs (Table S5) and, indeed, we could find evidence across diseases for many of these proteins.
5. The common ND modifiers identified by MLnet are enriched in apoptosis, autophagy and mitophagy.How about the seed modifiers used for each disease?If they are not, this is likely to be an artifact of just identifying very well connected general biological processes.To avoid this potential bias, the authors should also use randomized networks, preserving the connection degree of the proteins and shuffling the edges.
We agree with the reviewer that these terms should appear enriched in at least some of the lists of disease-specific modifiers.If they appear "de novo" that would be of concern.We analysed disease-specific modifiers that are generated by module 1.Specifically, we performed KEGG pathway enrichment analysis with the disease-specific modifiers used as seeds (Fig S10).As can be seen from the Fig S10 (see also below), the terms autophagy, mitophagy, and effectors of the insulin pathway (FoxO, mTOR) are all found enriched in at least two diseases.They do not appear in each singe disease, maybe because the corresponding or related genes had not been tested in models for each disease.Importantly, as mentioned in response to point 1 of this reviewer, our central hypothesis is that there is a subset of genetic modifiers that affect the severity of one of these diseases that have broader relevance and which may modify several other or even all proteinopathies.In other words, there may be genes of a specific pathway that have been tested positive in models of one disease but for which the relevance in other diseases may not yet be known.We designed MLnet to reweight individual genes so that such candidates are prioritized.
We also agree with the reviewer that a protein interaction network-based approach may suffer from artifacts related to highly connected proteins (hubs).We include a normalization factor into the MLnet calculation to account for protein "hubiness".We show in Fig S7 that MLnet does not suffer from a strong correlation between prediction rank and protein degree ("hubiness").It is well established that highly connected and annotated proteins/genes rank often high in prioritization/gene-prediction approaches, just as a result of their involvement in and connection to numerous cellular processes.Thus, a sort of normalization is necessary.However, one should not a priori exclude these highly connected/annotated genes/proteins because they are still relevant to many processes.So, what is the right approach?In the original submission, we tested two methods with the one implemented in MLnet coming out on top.We now followed the reviewer's suggestion and also tested randomized networks for normalization.Specifically, we randomized the interaction network while maintaining interaction degree, used randomized network data for normalization and assessed the ability of this approach to predict experimentally established modifiers common to different disease combinations.As shown in Fig S6 (see below), though the network randomization approach showed a better performance than seed randomization, MLnet outperformed the other approaches for all disease combinations.10.In the last section, the authors kind of drop all the previous results and only focus on testing the effect of the insulin pathway in Alzheimer´s disease which, as they state, has been clearly established!I fail to see the added value of re-reporting that the modulation of the insulin pathway has an effect on the 5xFAD mouse model.If anything, they should test the effect of this modulation of models other that AD (i.e.HD or the SCs).
We appreciate this pertinent objection by the reviewer.
We tested the activation of Akt1 initially in four ND cell models (Fig 5B-E).As our resources are limited, we were not able to test activation of Akt1 in mice models of all four NDs in the focus here, which would be ideal.We decided to go for AD models for two main reasons.
(1) AD is the ND that is the most common among the four studied here and the one with the biggest socio-economical burden.
(2) We agree that the importance of the insulin pathway is much better elucidated in Alzheimer than in the other ND pathologies.However, scientific results can be conflicting and not always easily reproducible.For example, treatment for restoring the insulin and PI3K-Akt signaling pathways, which is reduced in AD patients, showed mixed results in clinical trials.Specifically, as this reviewer is certainly aware of, administration of insulin to healthy individuals and AD patients improved memory function in small-scale studies, but in large scale clinical trials the results were mixed (Hallschmid, 2021;Morris & Burns, 2012).Furthermore, GSK3b inhibitors showed positive improvement in animal models, but not in clinical studies (Rippin & Eldar-Finkelman, 2021;Arciniegas Ruiz & Eldar-Finkelman, 2022).These contradictory results might be due to the fact that insulin is involved in a variety of cellular processes including cell growth, survival, and metabolism, as well as a variety of tissues including the brain and pancreatic tissues.Thus, we wanted to test insulin activation in a mouse AD model by ourselves because further clarification in this area is needed.In addition, previous studies revealed that activated Akt1 level are decreased in mouse AD models, but its level is up-regulated in AD patients (Griffin et al., 2005;Kommaddi et al., 2023;Rickle et al., 2004).Therefore, we wanted to test the effect of an Akt1 activator.
In hindside, we agree that testing in a HD mouse model may have been more impactful.This said, the focus of this paper is on the introduction of a novel computational method, and we already validate predictions in Drosophila and cell models of four NDs.
We thank the reviewer for these clarifying remarks.We were not expressing things clearly in our initial submission.We have now clarified these points and highlight that GSK3β is involved in the extended insulin signaling pathway as shown in Fig 3C (see below).We now discuss that insulin and GSK3b inhibitors have been tested to treat AD but failed or showed mixed results in large clinical trials (see below).(Page 20-21) Moreover, although our experiments suggest a modulatory role of Akt1 for multiple NDs, it may not be an ideal target for ND treatment development because of its involvement in numerous cellular processes and the fact that its enhanced activation can lead to cancerous cell transformation (Wang et al, 2017), which would require very close monitoring for tumorigenic effects when activated via a therapeutic agent.Other proteins in the insulin pathway such as the downstream effector GSK3β are already actively targeted for ND therapy development.GSK3β inhibitors showed positive improvement in animal models, but unfortunately failed in AD patients (Rippin & Eldar-Finkelman, 2021;Arciniegas Ruiz & Eldar-Finkelman, 2022).Moreover, direct insulin administration to healthy individuals and AD patients improved memory performance in small studies, but mixed results were reported for larger clinical trials (Hallschmid, 2021;Morris & Burns, 2012).It is clear that more research is required to fully understand the roles of insulin signaling in NDs and whether activation of specific elements of this signaling pathway may benefit patients.
12. Finally, the number of animals used in the Barnes maze test (4/5 mice) seems insufficient.The learning curve for the AD model is not really consistent in the days 3,4 and 5.The same effect is observed in the test days (15 and 16).Although, Fig 5G shows a mild significance when comparing days 5 and 16 between AD and AD+SC79, this difference should be supported by a robust learning curve, or any small change (i.e.adding one more animal or changing the days compared) could wipe it out.
We thank the reviewer for the comment.We have tested another 16 mice (four mice of WT, five mice of WT + SC79, three mice of AD, and four mice of AD + SC79) after submitting our manuscript and have added now the new results to Thank you for sending us your revised manuscript.We have now heard back from the two reviewers who agreed to evaluate your revised study.As you will see below, the reviewers are satisfied with the performed revisions and support publication.As such, I am glad to inform you that we can soon accept your manuscript for publication, pending some minor revisions listed below, all related to editorial issues.
-The funding information (NRF-2022M3A9B6082687, and NRF-2018R1A5A1025077; 2021R1A2C1006353; Korea University grant; CIHR) needs to be entered in the submission system.
-Please indicate the statistical test used for data analysis in the legends of Figures 3a-b.
-The "Tool Availability" section should be renamed to "Data Availability".Full source codes etc. should be made available on GitHub.
-Tables S1-S8 should be uploaded individually (one xls file per Dataset) as EV Datasets EV1-EV8.The callouts in the text should be corrected accordingly.Please include a description of each dataset in the xls file as a separate sheet.
-Please include callouts for Appendix Figure S14 and Datasets EV7-EV8 in the main text.
-The callouts to Appendix Figures should be Appendix -Please make sure that no text remains labelled in red in the Appendix.The Appendix will appear online as it is provided (it is not going to be typeset).
-The source data files need to be reorganized.Please provide one file (or zip folder) per main figure .A single ZIP folder should be provided for all the source data for the Appendix figures.
-The synopsis image does not display well at the final size required.Please resupply the image as a jpg or png at the required final size (it needs to be exactly 550 px wide, and the height ideally < 500 px), ensuring that all labels are legible.Reorganizing the image so that not all image elements are displayed on "one line" might work better.
Please resubmit your revised manuscript online, with a covering letter listing amendments and responses to each point raised by the referees.Please resubmit the paper **within one month** and ideally as soon as possible.If we do not receive the revised manuscript within this time period, the file might be closed and any subsequent resubmission would be treated as a new manuscript.Please use the Manuscript Number (above) in all correspondence.
Click on the link below to submit your revised paper.https://msb.msubmit.net/cgi-bin/main.plexAs a matter of course, please make sure that you have correctly followed the instructions for authors as given on the submission website.
Please click the link below to modify this ORCID: Link Not Available The system will prompt you to fill in your funding and payment information.This will allow Wiley to send you a quote for the article processing charge (APC) in case of acceptance.This quote takes into account any reduction or fee waivers that you may be eligible for.Authors do not need to pay any fees before their manuscript is accepted and transferred to the publisher.*** PLEASE NOTE *** As part of the EMBO Press transparent editorial process initiative (see our Editorial at https://dx.doi.org/10.1038/msb.2010.72 , Molecular Systems Biology will publish online a Review Process File to accompany accepted manuscripts.When preparing your letter of response, please be aware that in the event of acceptance, your cover letter/point-by-point document will be included as part of this File, which will be available to the scientific community.More information about this initiative is available in our Instructions to Authors.If you have any questions about this initiative, please contact the editorial office (msb@embo.org).----------------------------------------------------------------------------Reviewer #2: The authors have satisfactorily addressed my major concerns.
Suggest correction to the sentence: A multi-layered network expansion model is introduced that finds (replace with "to find"?or rewrite) proteins with pathophysiological relevance for groups of diseases.
Reviewer #3: I was not at all convinced by the first version of the m/s.However, in my opinion, the authors have done a great job during revision, addressing all the comments raised by the Reviewers.My opinion is tat now the m/s is ready for publication.I am still not convinced of some of the points/decision made by the authors, but they are all now properly discussed and should be clear to the reader.
Finally, I would like to thank the authors for their effort in tackling all the points raised.Molecular Systems Biology charges an article processing charge (APC) to cover the publication costs.You, as the corresponding author for this manuscript, should have already received a quote with the article processing fee separately.Please let us know in case this quote has not been received.
Once your article is at Wiley for editorial production, you will receive an email from Wiley's Author Services system, which will ask you to log in and will present you with the publication license form for completion.Within the same system the publication fee can be paid by credit card, an invoice or pro forma can be requested.
Payment of the publication charge and the signed Open Access Agreement form must be received before the article can be published online.
Upon acceptance it is mandatory for you to return the completed payment form.Failure to send back the form may result in a delay in the publication of your paper.
Molecular Systems Biology articles are published under the Creative Commons licence CC BY, which facilitates the sharing of scientific information by reducing legal barriers, while mandating attribution of the source in accordance to standard scholarly practice.
Proofs will be forwarded to you within the next 2-3 weeks.

Plants and microbes Information included in the manuscript?
In which section is the information available?
(Reagents and Tools ). a statement of how many times the experiment shown was independently replicated in the laboratory.
-common tests, such as t-test (please specify whether paired vs. unpaired), simple χ2 tests, Wilcoxon and Mann-Whitney tests, can be unambiguously identified by name only, but more complex techniques should be described in the methods section; Please complete ALL of the questions below.Select "Not Applicable" only when the requested information is not relevant for your study.
if n<5, the individual data points from each experiment should be plotted.Any statistical test employed should be justified.Source Data should be included to report the data underlying figures according to the guidelines set out in the authorship guidelines on Data Each figure caption should contain the following information, for each panel where they are relevant: a specification of the experimental system investigated (eg cell line, species name).the assay(s) and method(s) used to carry out the reported observations and measurements.an explicit mention of the biological and chemical entity(ies) that are being measured.an explicit mention of the biological and chemical entity(ies) that are altered/varied/perturbed in a controlled manner.

Study protocol
Information included in the manuscript?
In which section is the information available?
(Reagents and Tools

Reporting
Adherence to community standards Information included in the manuscript?
In which section is the information available?
(Reagents and Tools Have primary datasets been deposited according to the journal's guidelines (see 'Data Deposition' section) and the respective accession numbers provided in the Data Availability Section?

Data Availability
Were human clinical and genomic datasets deposited in a public access-controlled repository in accordance to ethical obligations to the patients and to the applicable consent agreement?

Not Applicable
Are computational models that are central and integral to a study available without restrictions in a machine-readable form?Were the relevant accession numbers or links provided?

Data Availability
If publicly available data were reused, provide the respective data citations in the reference list.

Yes Reagents and Tools Table
The MDAR framework recommends adoption of discipline-specific guidelines, established and endorsed through community initiatives.Journals have their own policy about requiring specific guidelines and recommendations to complement MDAR.
-layered network expansion model identifies Akt1 as common modulator of neurodegeneration Dear Dr Gsponer, 0) -[data type]: [full name of the resource] [accession number/identifier] ([doi or URL or identifiers.org/DATABASE:ACCESSION])-For data quantification: please specify the name of the statistical test used to generate error bars and P values, the number (n) of independent experiments (specify technical or biological replicates) underlying each data point and the test used to calculate p-values in each figure legend.The figure legends should contain a basic description of n, P and the test applied.Graphs must include a description of the bars and the error bars (s.d., s.e.m.).

Fig 4 ,Figure 4 .
Figure 4. Changes in the rough eye phenotype of Drosophila models for AD, HD, SCA1, and SCA3 due to knockdown of predicted common modifiers.Representative bright-field microscope images of fly eyeswith GMR-GAL4-driven misexpression of the ND-causing genes, along with GMR-GAL4-driven RNAi against each of the indicated genes in which mCherry served as a control.Compared with the wild-type (WT) compound eye with the ordered structure of ommatidia, flies with misexpression of individual ND-causing genes and RNAi against mCherry under the control of the GMR-GAL4 driver had rough eyes with the variation in phenotypic severity.Suppression or enhancement of these rough eye phenotypes caused by RNAi-mediated knockdown of the predicted common modifiers is indicated with + or -, respectively.As a negative control, two genes (Cyp6a18 and CG34372) randomly selected among low-ranked genes were also tested.
Fig 4 and the Results section of the revised manuscript.(Page 14)

(
Page 28) The fly eyes were photographed under the same adjustment setting of I-MEASURE software for capturing images.The Drosophila experiments were performed in a blinded manner.(Page 29) Fig S14 shows the histogram of the ranks of interaction partners of these proteins in the individual diseases.As only top 100 proteins from each disease get a seed score, only the number of interaction partners that are part of these 100 seeds are shown.As can be seen from Fig S14, Akt1 has, consistently across all diseases, more interaction partners that are top ranked, i.e., that are part of the 100 seed proteins.

Figure 3 .
Figure 3. KEGG pathway and GO enrichment analyses with predicted common modifiers.A KEGG pathway enrichment analysis of the top 100 predicted common modifiers.B GO enrichment analysis of the top 100 predicted common modifiers.C A simplified schematic diagram of the insulin signaling pathway and downstream functions.
Figs 5F-H (please see below).Though there are variations in the results because of the intrinsic variations of animal behavior, trends are confirmed and p-values still indicate statistical significance (p-value < 0.05).Importantly, the learning curves at days 3,4 and 5 are much more consistent now.-layered network expansion model identifies Akt1 as common modulator of neurodegeneration Dear Dr Gsponer, Figure S1, Appendix Figure S2 etc. (instead of Fig S1, Fig S2 etc.).
-layered network expansion model identifies Akt1 as common modulator of neurodegeneration Dear Dr Gsponer, Thank you again for sending us your revised manuscript.We are now satisfied with the modifications made and I am pleased to inform you that your paper has been accepted for publication.*** PLEASE NOTE *** As part of the EMBO Publications transparent editorial process initiative (see our Editorial at https://dx.doi.org/10.103/msb.2010.72),Molecular Systems Biology publishes online a Review Process File with each accepted manuscripts.This file will be published in conjunction with your paper and will include the anonymous referee reports, your point-by-point response and all pertinent correspondence relating to the manuscript.If you do NOT want this File to be published, please inform the editorial office at msb@embo.orgwithin 14 days upon receipt of the present letter.LICENSE AND PAYMENT:All articles published in Molecular Systems Biology are fully open access: immediately and freely available to read, download and share.

Table S5 .
Ranks of top 12 predicted common modifiers in predicted disease-specific modifiers and their disease association evidence *

Please note that a copy of this checklist will be published alongside your article.
Thank you very much for submitting your work to Molecular Systems Biology.

In which section is the information available?
(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)

In which section is the information available?
(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)

Cell materials Information included in the manuscript? In which section is the information available?
(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)Cell lines: Provide species information, strain.Provide accession number in repository OR supplier name, catalog number, clone number, and/OR RRID.

In which section is the information available?
Provide species, strain, sex, age, genetic modification status.Provide accession number in repository OR supplier name, catalog number, clone number, OR RRID.
(Reagents and ToolsTable, Materials and Methods, Figures, Data Availability Section) Laboratory animals or Model organisms:

In which section is the information available?
Table, Materials and Methods, Figures, Data Availability Section) (Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)If collected and within the bounds of privacy constraints report on age, sex and gender or ethnicity for all study participants.

In which section is the information available?
This checklist is adapted from Materials Design Analysis Reporting (MDAR) Checklist for Authors.MDAR establishes a minimum set of requirements in transparent reporting in the life sciences (see Statement of Task: 10.31222/osf.io/9sm4x).Please follow the journal's guidelines in preparing your the data were obtained and processed according to the field's best practice and are presented to reflect the results of the experiments in an accurate and unbiased manner.
(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)If your work benefited from core facilities, was their service mentioned

Checklist for Life Science Articles (updated January ideally
, figure panels should include only measurements that are directly comparable to each other and obtained with the same assay.plots include clearly labeled error bars for independent experiments and sample sizes.Unless justified, error bars should not be shown for technical the exact sample size (n) for each experimental group/condition, given as a number, not a range; a description of the sample collection allowing the reader to understand whether the samples represent technical or biological replicates (including how many animals, litters, cultures, etc.

In which section is the information available?
Table, Materials and Methods, Figures, Data Availability Section) If study protocol has been pre-registered, provide DOI in the manuscript.For clinical trials, provide the trial registration number OR cite DOI.(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section) Provide DOI OR other citation details if external detailed step-by-step protocols are available.Not Applicable

Experimental study design and statistics Information included in the manuscript? In which section is the information available?
(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)Include a statement about sample size estimate even if no statistical methods were used.YesMaterials and Methods, FiguresWere any steps taken to minimize the effects of subjective bias when allocating animals/samples to treatment (e.g.randomization procedure)?If yes, have they been described?YesMaterials and MethodsInclude a statement about blinding even if no blinding was done.Yes Materials and MethodsDescribe inclusion/exclusion criteria if samples or animals were excluded from the analysis.Were the criteria pre-established?If sample or data points were omitted from analysis, report if this was due to attrition or intentional exclusion and provide justification.Not ApplicableFor every figure, are statistical tests justified as appropriate?Do the data meet the assumptions of the tests (e.g., normal distribution)?Describe any methods used to assess it.Is there an estimate of variation within each group of data?Is the variance similar between the groups that are being statistically compared?

definition and in-laboratory replication Information included in the manuscript? In which section is the information available?
(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)In the figure legends: state number of times the experiment was replicated in laboratory.

In which section is the information available?
(Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section)Include a statement confirming that informed consent was obtained from all subjects and that the experiments conformed to the principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report.State details of authority granting ethics approval (IRB or equivalent committee(s), provide reference number for approval.Include a statement of compliance with ethical regulations.
Studies involving human participants: State details of authority granting ethics approval (IRB or equivalent committee(s), provide reference number for approval.Not ApplicableStudies involving human participants:Not ApplicableStudies involving human participants: For publication of patient photos, include a statement confirming that consent to publish was obtained.Not Applicable Studies involving experimental animals:

Use Research of Concern (DURC) Information included in the manuscript? In which section is the information available?
(Reagents and ToolsTable, Materials and Methods, Figures, Data Availability Section) Could your study fall under dual use research restrictions?Please check biosecurity documents and list of select agents and toxins (CDC): https://www.selectagents.gov/sat/list.htmNot Applicable If you used a select agent, is the security level of the lab appropriate and reported in the manuscript?Not Applicable If a study is subject to dual use research of concern regulations, is the name of the authority

granting approval and reference number for
the regulatory approval provided in the manuscript?

and III randomized controlled trials
Table, Materials and Methods, Figures, Data Availability Section) State if relevant guidelines or checklists (e.g., ICMJE, MIBBI, ARRIVE, PRISMA) have been followed or provided.Not Applicable For tumor marker prognostic studies, we recommend that you follow the REMARK reporting guidelines (see link list at top right).See author guidelines, under 'Reporting Guidelines'.Please confirm you have followed these guidelines., please refer to the CONSORT flow diagram (see link list at top right) and submit the CONSORT checklist (see link list at top right) with your submission.See author guidelines, under 'Reporting Guidelines'.Please confirm you have submitted this list.Reagents and Tools Table, Materials and Methods, Figures, Data Availability Section) (