Review Reports - Stage-Associated Cellular and Molecular Signatures in Diabetic Retinopathy Identified Through Integrated Bulk and Single-Cell Transcriptomic Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In the study entitled “Stage-Specific Cellular and Molecular Signatures in Diabetic Retinopathy Revealed by Integrated Bulk and Single-Cell Transcriptomics.” by Li et al. the authors present a comprehensive analysis of stage-specific molecular and cellular changes in diabetic retinopathy (DR) by integrating human bulk RNA-seq with mouse single-cell RNA-seq from two complementary models (db/db and STZ-induced). The multi-platform, multi-species approach offers novel insights into the dynamic progression of DR.

This study assembles and analyzes a large amount of transcriptomic data and provides a potentially useful stage-oriented framework for exploring DR biology. However, many of the current conclusions extend beyond what can be supported by cross-species, computational integration alone. By tempering causal and translational claims, more clearly acknowledging dataset limitations, and reframing key findings as associations and predictions, the manuscript would be substantially strengthened.

The current title accurately reflects the datasets but overstates mechanistic insights. A more cautious title (e.g.,“Stage-Associated Cellular and Molecular Signatures in Diabetic Retinopathy Identified Through Integrated Bulk and Single-Cell Transcriptomic Analysis” would better reflect the predictive rather than confirmatory nature of the study.

Briefly discuss limitations of comparing human peripheral retina with mouse models and how these differences may affect interpretation. In addition, the rationale for integrating two distinct mouse models should be explicitly justified and limitations acknowledged. The text should more explicitly acknowledge that the mouse data provide model-based approximations rather than direct representations of human stage-specific DR biology. In general, the cross-species, cross-model integration introduces significant uncertainty, and the manuscript must explicitly acknowledge that mechanistic conclusions for human DR remain speculative. Thus, mechanistic interpretations (e.g., pathway signaling) could be framed as hypotheses for future validation.

Bulk-derived WGCNA modules may reflect changing cell-type proportions rather than true stage-specific programs. When mapped to single-cell data, the authors should clarify that they identify cells enriched for these genes, avoiding causal language like “define a stage.”

CellChat predictions of intercellular signaling (e.g., LAMININ, ANGPTL) is an interesting component of the study but it is based on ligand–receptor co-expression and do not prove functional activity. These findings should be framed as predicted communication changes that warrant functional validation. The Discussion should explicitly note that transcript abundance does not necessarily reflect protein levels, secretion, receptor engagement, or pathway activation.

The manuscript currently implies that ribosomal dysregulation in photoreceptors precedes and may contribute to vascular injury. This is an interesting hypothesis but remains speculative based on the present data. The wording should be revised to emphasize that these are associations observed in early diabetic conditions, not demonstrated drivers of subsequent vascular pathology.

The manuscript frequently refers to the identification of “novel therapeutic targets” and “personalized intervention strategies.” While the datasets provide valuable descriptive insight, they remain correlative and largely computational. No functional perturbation experiments are performed to demonstrate that modulating the highlighted pathways (e.g., ANGPTL4, LAMININ components, macroglial genes) alters DR phenotypes. These translational claims should therefore be moderated and framed as hypothesis-generating rather than target-validating.

The Discussion occasionally conflates correlation with causation; wording should emphasize associative findings.

Minor Issue

Figures are complex; clearer labeling for species, models, cell types, and datasets is needed.

Author Response

Comments and Suggestions for Authors

Response：

We sincerely appreciate your valuable suggestions regarding the title of our manuscript. We agree that a more cautious title better reflects the descriptive and predictive nature of the analyses presented in this study. Accordingly, we have revised the manuscript title following the reviewer’s recommendation. The new title now reads:

“Stage-Associated Cellular and Molecular Signatures in Diabetic Retinopathy Identified Through Integrated Bulk and Single-Cell Transcriptomic Analysis.”（Line 1-4）

We believe that this revised title more accurately reflects the scope of the study and avoids overstating mechanistic conclusions.

Response:

We thank the reviewer for this important comment. It is important to acknowledge the inherent limitations of translating findings from diabetic mouse models to the human peripheral retina. Fundamental anatomical and cellular differences exist between species(PMID:32555229,30712875). Notably, the human retina contains a specialized macula and fovea characterized by high cone density and a distinct vascular architecture, whereas the mouse retina is rod-dominant and lacks a macular structure. However, Lu et al. demonstrated that humans and mice exhibit broadly similar retinal cellular expression patterns(PMID:32386599), providing a rationale for cross-species data integration. Furthermore, given that the macula is a developmentally advanced region not present in mice, we elected to compare the human peripheral retina—rather than macular tissue—with the whole mouse retina. This approach is biologically justified, as the mouse retina more closely resembles the human peripheral retina in the absence of a macular equivalent, thereby minimizing confounding related to macular specialization and enabling evaluation of conserved retinal responses to diabetic stress.

Then, We agree that cross-species comparisons and the integration of different diabetic mouse models introduce inherent limitations that should be clearly acknowledged. In the revised manuscript, we have expanded the Discussion to explicitly address these issues.we have further justified the rationale for integrating the db/db and STZ mouse models(PMID:28836097). These models represent complementary experimental systems that capture distinct aspects of diabetic pathology (genetic type 2 diabetes versus chemically induced hyperglycemia). Despite differences in systemic metabolic context, both models recapitulate key retinal alterations associated with diabetic retinopathy, including microvascular dysfunction, inflammation, oxidative stress, and extracellular matrix remodeling (PMID:26297071,33603228,28942458).Their combined analysis was therefore intended to improve the robustness of identifying diabetes-associated transcriptional signatures .Importantly, we have revised the manuscript to explicitly emphasize that the mouse datasets should be interpreted as model-based approximations of disease progression(stage-associated), rather than direct representations of human stage-specific DR biology. Accordingly, mechanistic interpretations for human DR derived from these cross-species analyses should be considered hypothesis-generating and require future experimental validation in human tissue or functional models.

Cross-species comparisons between human and mouse retinas should be inter-preted cautiously. In this study, transcriptomic data from human peripheral retina were integrated with datasets derived from mouse whole retina, which differ in retinal architecture, vascular organization, metabolic demand, and cellular composition. In addition, the two mouse models used (db/db and STZ) represent distinct forms of dia-betic pathology and do not fully recapitulate the complexity of human diabetic reti-nopathy. Therefore, these mouse datasets should be considered model-based approx-imations of disease-associated transcriptional changes, and mechanistic interpretations for human DR remain hypothesis-generating and require further validation.

We thank the reviewer for this important comment. We agree that cross-species comparisons and the integration of different diabetic mouse models introduce inherent limitations that should be explicitly acknowledged. In the revised manuscript, we have expanded the Discussion to address these issues more clearly.(Line:277-297, 405-408, 409-410)

Response:

Thank you for your insightful comment. We agree that bulk-derived WGCNA modules may reflect changes in cell-type proportions rather than pure stage-specific programs, and that language should be appropriately cautious when mapping these modules to single-cell data.

We now explicitly acknowledge that bulk-derived modules may capture both compositional shifts and transcriptional changes, and that single-cell mapping serves to deconvolve cellular origins. We have replaced language suggesting modules "define" stages with more precise descriptors . We now state that we identify cell types enriched for these genes, and refer to " stage- associated" rather than "stage-defining programs." In the manuscript, we note the inherent challenges of bulk deconvolution and emphasize that our findings represent associations rather than definitions.（Line 85-108 and other concerning stages）

These revisions ensure our interpretation aligns with the resolution of our analytical approach. Thank you for helping us improve precision and rigor.

Response:

We thank the reviewer for this helpful comment. We fully agree that the CellChat analysis is based on ligand–receptor co-expression inferred from transcriptomic data and does not directly demonstrate protein abundance, spatial localization, or functional signaling activity. Therefore, the CellChat results in this study should be interpreted as hypothesis-generating predictions that highlight candidate intercellular signaling pathways requiring further experimental validation at the protein and functional levels.

Accordingly, we have revised the manuscript to consistently frame LAMININ- and ANGPTL-related interactions as predicted communication changes rather than confirmed signaling events(Line 326-329).

Specifically, statements referring to ANGPTL and LAMININ pathways as “central drivers” have been revised to describe them as predicted communication signals identified by computational analysis that may contribute to stage-associated cellular interactions in DR. We now explicitly note in the Discussion that protein-level validation and spatial confirmation will be required to determine their functional relevance in future studies(Line 236-“suggesting” ,Line265-“contributing to”)

Response:

We thank the reviewer for this insightful comment. We agree that the present transcriptomic data do not provide direct evidence that ribosomal dysregulation in photoreceptors is a causal driver of subsequent vascular injury. To avoid overinterpretation, we have revised the manuscript to clarify that the observed ribosomal response represent associations detected under early diabetic conditions（Line 322-329）.

Specifically, the wording in the Results and Discussion sections has been modified to emphasize that these findings suggest a potential link between early neuronal metabolic stress and later vascular alterations..Ribosome-related transcriptional changes observed in photoreceptors under early diabetic conditions may reflect an early neuronal stress response; however, whether these alterations contribute to subsequent microvascular pathology remains to be determined.

In the Discussion sections has been modified to emphasize that these findings suggest a potential link between early neuronal metabolic stress and later vascular alterations, but do not establish a causal relationship. We now frame ribosomal response as a candidate early molecular feature of diabetic retinal stress that may precede microvascular changes and warrants further functional investigation（Line 305, 322-329）.

Response:

We thank the reviewer for this important comment. We agree that the present analyses are primarily transcriptomic and computational, and therefore cannot establish causal roles for the highlighted pathways. In the revised manuscript, we have moderated the translational language and replaced statements referring to “novel therapeutic targets” or “personalized intervention strategies” with wording that more appropriately describes these findings as candidate or potential pathways and hypothesis-generating observations derived from integrative transcriptomic analyses.

We now explicitly state that signals involving ANGPTL4, LAMININ components, and macroglial-associated genes represent predicted disease-associated pathways whose functional relevance requires validation through future perturbation studies. These revisions have been incorporated throughout to ensure that the conclusions remain appropriately cautious and aligned with the descriptive nature of the datasets(Line 374-380).

The Discussion occasionally conflates correlation with causation; wording should emphasize associative findings.

Response:

We thank the reviewer for this important comment. In the revised manuscript, we have carefully reviewed and refined the wording throughout the Discussion section to avoid implying causal relationships where the data support only associative observations. Statements that could be interpreted as causal have been revised to emphasize that the findings represent correlations or predicted associations derived from transcriptomic analyses. These modifications ensure that the interpretation of the results remains scientifically cautious and consistent with the descriptive nature of the study.

Minor Issue

Figures are complex; clearer labeling for species, models, cell types, and datasets is needed.

Response:

Thank you for this helpful suggestion. In response, we have thoroughly reviewed and revised all figures to improve clarity. Species, models, cell types, and dataset identifiers are now clearly labeled on each relevant panel.Annotations have been standardized across all figures to ensure consistency in formatting and terminology.Figure legends have been updated to provide clear descriptions of the sample information and experimental design for each panel.

These revisions enhance the readability and interpretability of the figures. Thank you for helping us improve figure accessibility.

Reviewer 2 Report

Comments and Suggestions for Authors

The article “Stage-Specific Cellular and Molecular Signatures in Diabetic 2 Retinopathy Revealed by Integrated Bulk and Single-Cell 3 Transcriptomics” by Li et al., is a well-designed integrative transcriptomic analysis combining human bulk RNA-seq and mouse single-cell RNA-seq to identify stage-specific cellular and molecular changes in diabetic retinopathy (DR). The work is timely, methodologically comprehensive, and biologically meaningful. It provides a stage-resolved framework of DR progression, which is valuable for both mechanistic insight and therapeutic targeting. However, several conceptual, methodological, and interpretational issues should be addressed before publication.

The assignment of mouse datasets to NDR, NPDR, and PDR stages is based largely on literature assumptions rather than direct phenotypic validation for example STZ 29 weeks = PDR, db/db 13 weeks = NDR, but there was no retinal imaging, vascular leakage quantification and no neovascularization confirmation. It would be better to include histological validation (isolectin staining, fluorescein angiography, or literature citations clearly demonstrating these exact timepoints correspond to these stages).
While ortholog conversion was performed, species differences may influence ribosomal gene regulation, ECM signaling and glial responses. Authors should provide a quantitative summary of percentage of ortholog retention, genes lost in conversion and pathway conservation rate.
The conclusion that ribosomal pathway enrichment indicates dysfunction is speculative. Upregulation of ribosomal genes could indicate compensatory response, increased translation, stress response. No direct functional evidence of ribosomal impairment is shown.
The Cell Chat results are interesting, but no protein validation, no spatial confirmation. Claims that ANGPTL and LAMININ are “central drivers” are somewhat overstated.
Monocle regression is well executed, but hyperglycemia duration as a continuous variable is not clearly defined across datasets. Different mouse models were integrated — time interpretation may be confounded.
Multiple testing correction thresholds should be explicitly stated for each analysis. Exact sample numbers for each dataset should be listed in Methods.
Some figure legends refer to incorrect dataset numbers (e.g., GSE26299 vs GSE160306). Improve consistency.

Author Response

Comments and Suggestions for Authors

The assignment of mouse datasets to NDR, NPDR, and PDR stages is based largely on literature assumptions rather than direct phenotypic validation for example STZ 29 weeks = PDR, db/db 13 weeks = NDR, but there was no retinal imaging, vascular leakage quantification and no neovascularization confirmation. It would be better to include histological validation (isolectin staining, fluorescein angiography, or literature citations clearly demonstrating these exact timepoints correspond to these stages).

Response:

We thank the reviewer for this important comment.

Retinal neovascularization was confirmed in 16-week STZ-induced diabetic mice after the onset of hyperglycemia, as demonstrated by IB4 staining of retinal vascular whole mounts and CD105 immunohistochemistry.[20 /28836097、23032068]. Given that hy-perglycemia lasted for 25 weeks in the GSE178121 dataset, this dataset was considered associated with the PDR stage.

Studies have shown that 10-week-old db/db mice show increased Müller cell re-activity, elevated apoptosis across retinal layers, and decreased ERG amplitudes 26384381/ 27114552 ,while 13-week-old db/db mice exhibit reduced electroretinogra-phy (ERG) responses and retinal thinning 37653465. These researches support the GSE205123 dataset (13-week-old mice) corresponds to the NDR group. In db/db mice at 18 weeks of age, reactive gliosis and pericyte loss have been reported[20]. Pericyte loss contributes to microaneurysms and retinal hemorrhages[21], hallmark features of ear-ly-stage DR[1]. Therefore, we defined 18 weeks as the cutoff point between the “diabetic” and NPDR groups. Consistently, the 19-21-week db/db are observed features of VEGF upregulation and enhanced BRB permeability 20332345. Based on these prior findings, the GSE204880 (21-week-old mice) relevants to NPDR.However, it is important to acknowledge that most murine models do not fully capture the complete pathological spectrum of human PDR, including preretinal neovascularization and fibrovascular membrane formation. Indeed, no single diabetic mouse model entirely replicates the full disease course of human DR. Therefore, while the stage assignments of the three mouse datasets used in this study are grounded in existing literature and established pathological features, they remain approximations and carry inherent limitations. As such, these classifications should be interpreted with due caution(Line 93-107, Line289-297).

While ortholog conversion was performed, species differences may influence ribosomal gene regulation, ECM signaling and glial responses. Authors should provide a quantitative summary of percentage of ortholog retention, genes lost in conversion and pathway conservation rate

Response:

We thank the reviewer for raising this important point regarding cross-species ortholog conversion and pathway conservation.

In the revised manuscript, we have now provided a quantitative summary of the ortholog mapping process.（Line 86-92） A total of 3806 human WGCNA genes were subjected to ortholog conversion, among which 3344 (87.9%) were successfully mapped to mouse orthologs. Notably, the majority of unmapped genes consisted of long non-coding RNAs (lncRNAs), snoRNAs, pseudogenes, and T cell receptor segments, which are known to lack direct one-to-one orthologs between human and mouse due to species-specific annotation differences and evolutionary divergence. Protein-coding genes, including those involved in ribosomal function and ECM signaling, were largely retained.

Regarding pathway conservation analysis, we respectfully note that the human gene modules were derived from bulk transcriptomic data, whereas the validation step was performed using mouse single-cell RNA sequencing data. Pathway enrichment derived from bulk WGCNA modules reflects tissue-level co-expression structure, while enrichment from single-cell data is inherently cell-type–specific and dependent on cellular composition. Because of these fundamental differences in data modality and analytical framework, direct quantitative comparison of pathway enrichment results between species would not be methodologically equivalent and may lead to misleading estimates of “pathway conservation.”

Therefore, while we agree that cross-species conservation is an important consideration, we believe that formal pathway conservation rate analysis under these heterogeneous data structures would not provide a statistically robust or biologically interpretable measure. We have clarified this rationale in the revised manuscript.

We hope this explanation adequately addresses the reviewer’s concern.

The conclusion that ribosomal pathway enrichment indicates dysfunction is speculative. Upregulation of ribosomal genes could indicate compensatory response, increased translation, stress response. No direct functional evidence of ribosomal impairment is shown.

Response:

A ribosome-associated response in photoreceptors appears to be detectable in the early diabetic retina We thank the reviewer for this insightful comment. We agree that enrichment of ribosome-related pathways does not necessarily indicate ribosomal dysfunction. In the revised manuscript, we have carefully refined the wording to avoid implying functional impairment. The description has been revised to refer more cautiously to a ribosome-associated transcriptional response rather than ribosomal dysfunction.

We have also added clarifying text in the Discussion section acknowledging that the observed upregulation of ribosomal genes may reflect several biological processes, including compensatory responses, increased translational demand, or cellular stress responses under early diabetic conditions. Importantly, we now explicitly state that no direct functional evidence of ribosomal impairment is provided in the present study, and that further experimental investigations will be required to determine the functional significance of these transcriptional changes(Line 326-329).

The Cell Chat results are interesting, but no protein validation, no spatial confirmation. Claims that ANGPTL and LAMININ are “central drivers” are somewhat overstated

Response:

We thank the reviewer for this helpful comment. We agree that the CellChat analysis is based on ligand–receptor co-expression inferred from transcriptomic data and does not directly demonstrate protein abundance, spatial localization, or functional signaling activity. Accordingly, we have revised the manuscript to moderate the interpretation of these results.

Specifically, statements referring to ANGPTL and LAMININ pathways as “central drivers” have been revised to describe them as suggestion or predicted communication signals identified by computational analysis that may contribute to stage-associated cellular interactions in DR. We now explicitly note in the Discussion that protein-level validation and spatial confirmation will be required to determine their functional relevance in future studies(Line 374-380).

Monocle regression is well executed, but hyperglycemia duration as a continuous variable is not clearly defined across datasets. Different mouse models were integrated — time interpretation may be confounded.

Response:

This study utilized four publicly available transcriptomic datasets. Specifically, GSE160306 includes human retinal samples from individuals with non-diabetic retinopathy (NDR), non-proliferative diabetic retinopathy (NPDR), and proliferative diabetic retinopathy (PDR). The mouse datasets comprise GSE205123 (13-week db/db mice, corresponding to 7 weeks of hyperglycemia), GSE204880 (21-week db/db mice, corresponding to 15 weeks of hyperglycemia), and GSE178121 (29-week streptozotocin-induced diabetic mice, corresponding to 25 weeks of hyperglycemia). Detailed characteristics of these datasets are summarized in Table 1. In addition, the order of Figure 1 and Figure 6 has been corrected as per the review comments(table 1).

Multiple testing correction thresholds should be explicitly stated for each analysis. Exact sample numbers for each dataset should be listed in Methods.

Response:

We thank the reviewer for this helpful suggestion. In the revised manuscript, we have explicitly specified the multiple testing correction methods and significance thresholds applied in each analysis in the Methods section to improve transparency and reproducibility. In addition, we have carefully reviewed the datasets used in this study and have listed the exact sample numbers for each dataset in the Methods section. These revisions ensure clearer reporting of the statistical criteria and dataset composition(Line 154,255-256,545-546).

Some figure legends refer to incorrect dataset numbers (e.g., GSE26299 vs GSE160306). Improve consistency.

Response:

We thank the reviewer for noting this issue. All figures and figure legends have been carefully checked, and dataset identifiers have been corrected and standardized throughout the manuscript to ensure consistency between the figures, legends, and the main text.

Thank you for your meticulous attention to this detail, which has helped improve the overall quality and reliability of our manuscript.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have successfully responded to all my requests. Therefore, I have no further comments.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have addressed all the comments. I really appreciate their efforts.