Review Reports - Machine Learning-Driven Multi-Omics Analysis Identifies CHP2 as a Key PANoptosis-Related Dual-Function Biomarker in Colorectal Cancer

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript has a few technological highlights, such as the use of multiple algorithms converging on the same candidate gene sets (Figure 3), and it seems that downregulation of CHP2 is consistent in these data sets - even independent datasets. They use a nice immune correlation analysis - but I also think they are overinterpreting this as "mechanism" when its only correlation. The only basic validation done in the paper actually shows that overexpression of CHP2 reduces proliferation and migration of cells. But otherwise - there are many things less convincing.

This manuscript makes a few bold statements, or "conclusions" which arent really supported by the data. And many assumptions lack any validation and therefore cannot be considered as convincing.

This begins with the exclusive focus on "PANoptosis," but it's nowhere shown that any of this happens in the cells and tissues investigated. The authors show cell death (by Annexin V staining) but thats a very general measure for cell death - and not specific for what they consider as "PANoptosis". Different sets of markers or methods/assays should have been used here, including Caspase-1 activation for pyroptosis, RIPK3 phosphorylation for necroptosis, Caspases 3 and 8 cleavage for "classic" apoptosis, and ideally, the authors should look if the "PANoptosme" is formed or not when they treat their cells. Interestingly, the authors actually acknowledge this limitation themselves (lines 738 - 740): "Our flow cytometry results confirmed that the overexpression of CHP2 induced apoptosis; however, we did not investigate the specific modes of PANoptosis. Thus, this is the priority for our future work".... but that doesn't stop them from talking about PANoptosis throughout the entire manuscript. This is scientifically not justified and misleading.

Then, there are many other claims that remain unvalidated by any means. The data is entirely computational. That includes also the "drug data" and none of these have seen any experimental validation - not any of the 18 drugs investigated here (Fig. 9). So this is largely speculation and there is now way the link that is claimed that tumors with high expression levels of CHP2 would show something like "selective resistance" agaist the standard chemotherapies is also entirely speculation. And so are all the claims about certain biomarkers.

I think it is a key problem that the authors consistently use correlation for "mechanism" - that isn't necessarily the case. For example, wherever CIBERSORT is used and not independently validated, this is slippery found and inferred 2mechanisms" are nothing but correlations. These are useful for forming hypotheses, but here they're used to claim mechanisms. In fact, its nothing but unconfirmed hypotheses. They may hold up, but most likely they won't, if challenged by validation. And on top of this, many of the correlations are rather waek and therefore, not convincing (r = 0.2 to 0.3).

There is also some puzzling inconsistency in the manuscript: CHP2 may be downregulated in colorectal cancers, and it's likely to act as a tumor suppressor, which is also kind of confirmed by the observation that overexpression of CHP2 blocks tumor cell proliferation. But as the manuscript moves on, CHP2 is considered a "high risk" factor if it's overexpressed - see for example lines 563 - 564: " In this study, the low-risk group was defined by low CHP2 expression, whereas the high-risk group was defined by high CHP2 expression. The high-risk groups exhibited significant resistance to 12 agents and hypersensitivity to the other 6 agents". It is the HIGH CHP2 tumors that are resistant? When has CHP2 suddenly changed sides and become an oncogene? Thats not logical and lacks any explanation.

There are a few other things that go completely unvalidated, like the "ceRNA Network" analysis: This is again entirely relying on computational correlations, nothing else. The authors claim that diverse LNC RNAs and miRNAs cooperate with CHP2 and form a "regulatory network" but as said before - it lacks validation and as such is nothing more than a nice hypothesis or even speculation. Not even some RT-PCTR experiments were done here. No correlation was shown in any actual cells, or patient samples etc. But that doesn't stop the authors from introducing this hypothetical "network" as a "multimodal ceRNA axis" (line 709-711): "we identified a novel, multimodal ceRNA axis. The axis represents a complex LINC00689-mediated miRNA feedback loop (CHP2--hsa-miR-146a-3p--LINC00689--hsa-miR-1236-3p) and is directly related to the lncRNA FAM74A7, which might regulate the expression of CHP2." AT least they use the word "might" when it comes to regulation of CHP2, as this - as so much - not validated by any means here. It's nothing but speculation, in my opinion.

Author Response

Comments 1: This manuscript makes a few bold statements, or "conclusions" which arent really supported by the data. And many assumptions lack any validation and therefore cannot be considered as convincing.

Response 1: We accept this criticism. In the revised manuscript, we have refined our language:

We changed all descriptions based solely on computational analysis (e.g., immune infiltration, ceRNA networks) from "mechanism" to "association" or "hypothesis". Mention exactly where in the revised manuscript this change can be found in Page 13, lines 465-507; Page 15, lines 545-548; Page 26, lines 740-766.

Comments 2:

Response 2:

We fully agree. To confirm the induction of PANoptosis by CHP2, we performed Western blot analysis to detect a triad of markers covering apoptosis, pyroptosis, and necroptosis:

Experimental Results: Overexpression of CHP2 concurrently activated p-MLKL (necroptosis), N-GSDMD (pyroptosis), and Cleaved-Caspase3 (apoptosis).Page 22, lines 686-691.

Rescue Experiment: We utilized siRNA-3 in stable CHP2-overexpressing cell lines, which significantly reversed the expression of these three death markers, confirming the selective regulatory role of CHP2 in the PANoptosis network. These data are now included in Figure 12F–J. The efficiency validation experiment of siRNA is shown in Figure S3.

Comments 3: Then, there are many other claims that remain unvalidated by any means. The data is entirely computational. That includes also the "drug data" and none of these have seen any experimental validation - not any of the 18 drugs investigated here (Fig. 9). So this is largely speculation and there is now way the link that is claimed that tumors with high expression levels of CHP2 would show something like "selective resistance" agaist the standard chemotherapies is also entirely speculation. And so are all the claims about certain biomarkers.

Response 3:

This was a vital suggestion. We performed in vitro drug sensitivity assays (CCK-8 validation) to confirm this "selective sensitivity/resistance" pattern (Page 22, lines 673-685, 801-813):

Standard Chemotherapy: Validation confirmed that low CHP2 expression (High-risk group) exhibited significant resistance to 5-FU, Oxaliplatin, and Irinotecan, while restoring CHP2 expression sensitized the cells to these regimens (Figure 12A–C).

Targeted Therapy: Interestingly, the low CHP2 group showed higher sensitivity to Ribociclib and Lapatinib (Figure 12D–E). This demonstrates that CHP2 is a dual-function biomarker that identifies patients resistant to chemotherapy but potentially responsive to targeted inhibitors.

Comments 4:

Response 4:

We fully agree with the reviewer’s professional assessment that correlation coefficients in the range of 0.2 to 0.3 should be interpreted as associations rather than definitive mechanisms. To address this and improve the scientific rigor of our manuscript, we have made the following substantial revisions:

(1) We have deleted Table 1 from the manuscript, which contained the preliminary correlation data that the reviewer identified as less convincing. By removing these weak correlations, we ensure the manuscript remains focused on the most robust findings (Lines 465-507).

(2) We have systematically replaced the term "mechanism" with "association," "correlation," or "hypothesis" throughout the text where findings were based solely on bioinformatic inferences (e.g., CIBERSORT and ceRNA analysis).

(3) We have clarified in the revised manuscript that the ceRNA network and immune cell infiltration are not the primary focuses of this study. These analyses are now presented strictly as supplementary evidence intended to provide heuristic clues for future research.

(4) By prioritizing the experimentally validated role of CHP2 in PANoptosis and its predictive value for drug sensitivity, we believe the revised manuscript now offers a more balanced and scientifically justified interpretation of the data.

Comments 5: There is also some puzzling inconsistency in the manuscript: CHP2 may be downregulated in colorectal cancers, and it's likely to act as a tumor suppressor, which is also kind of confirmed by the observation that overexpression of CHP2 blocks tumor cell proliferation. But as the manuscript moves on, CHP2 is considered a "high risk" factor if it's overexpressed - see for example lines 563 - 564: " In this study, the low-risk group was defined by low CHP2 expression, whereas the high-risk group was defined by high CHP2 expression. The high-risk groups exhibited significant resistance to 12 agents and hypersensitivity to the other 6 agents". It is the HIGH CHP2 tumors that are resistant? When has CHP2 suddenly changed sides and become an oncogene? Thats not logical and lacks any explanation.

Response 5:

We appreciate the reviewer pointing out this core logical error. We sincerely apologize for the confusion. We have corrected the entire logic of the manuscript: CHP2 functions as a tumor suppressor in CRC, and its downregulation (loss) is the key factor driving poor prognosis and chemoresistance.

We have redefined the risk model: Low CHP2 expression = High-risk group, and High CHP2 expression = Low-risk group. This correction ensures that our bioinformatic predictions are now fully consistent with our experimental observation that CHP2 inhibits cell proliferation and migration. Page 19, lines 611--618.

Comments 6: There are a few other things that go completely unvalidated, like the "ceRNA Network" analysis: This is again entirely relying on computational correlations, nothing else. The authors claim that diverse LNC RNAs and miRNAs cooperate with CHP2 and form a "regulatory network" but as said before - it lacks validation and as such is nothing more than a nice hypothesis or even speculation. Not even some RT-PCTR experiments were done here. No correlation was shown in any actual cells, or patient samples etc. But that doesn't stop the authors from introducing this hypothetical "network" as a "multimodal ceRNA axis" (line 709-711): "we identified a novel, multimodal ceRNA axis. The axis represents a complex LINC00689-mediated miRNA feedback loop (CHP2--hsa-miR-146a-3p--LINC00689--hsa-miR-1236-3p) and is directly related to the lncRNA FAM74A7, which might regulate the expression of CHP2." AT least they use the word "might" when it comes to regulation of CHP2, as this - as so much - not validated by any means here. It's nothing but speculation, in my opinion.

Response 6:

We acknowledge the reviewer’s concern that the ceRNA network remains a computational prediction. In the revised manuscript, we have addressed this by:

We explicitly clarify that the primary objective of this study is to establish CHP2 as a biomarker for PANoptosis regulation. The ceRNA analysis is now presented as a secondary, exploratory supplement intended to provide heuristic clues rather than a primary focus of validation. We have removed the term "multimodal ceRNA axis" and replaced it with "potential regulatory association." We have also ensured that terms such as "might" or "predicted" are used consistently to reflect its hypothetical nature. We have revised the "bold" conclusions and acknowledged that this predicted network requires future experimental verification. Page 15, lines 534, 545-548 and Page 27, lines 793-797.

Reviewer 2 Report

Comments and Suggestions for Authors

Zetian Zhang et al. submitted a manuscript to Cells investigating the role of CHP2 as a regulator of regulated cell death programs and its potential relevance to prognosis, the tumor immune microenvironment, and therapeutic response in colorectal cancer. The topic broadly aligns with the scope of Cells, particularly in relation to cancer cell death pathways and tumor–immune interactions. However, in its current form, the manuscript relies heavily on bioinformatic inference and correlative analyses, while the central biological and mechanistic claims are not sufficiently supported by experimental evidence.

I therefore recommend major revision before the manuscript can be considered for acceptance. The following concerns must be carefully addressed.

1. The manuscript frames CHP2 as a regulator of PANoptosis; however, no direct experimental evidence supporting PANoptosis is provided. Cell death analyses rely primarily on Annexin V/7-AAD flow cytometry and general viability assays, which are insufficient to distinguish apoptosis, necroptosis, and pyroptosis. Canonical PANoptosis markers, including caspase-1 activation, GSDMD/GSDME cleavage, RIPK3/MLKL phosphorylation, inflammasome components, or pathway-specific inhibitor and rescue experiments, are not examined. As acknowledged by the authors, specific modes of regulated cell death were not dissected. In the absence of mechanistic validation, the PANoptosis-related conclusions remain speculative and should either be experimentally substantiated or substantially toned down to reflect association rather than mechanism.

2. The conclusions regarding selective drug resistance or sensitivity are derived entirely from computational IC50 predictions using public pharmacogenomic datasets. No experimental drug response validation is performed in colorectal cancer cell lines or patient-derived models. Given the strength of the therapeutic claims, experimental validation is essential. At a minimum, drug sensitivity assays across multiple CRC cell lines stratified by CHP2 expression should be included. Without such validation, the therapeutic implications should be presented as hypothesis-generating rather than predictive.

3. The manuscript reports that higher CHP2 expression is associated with improved overall and recurrence-free survival, yet subsequently defines a “high-risk” group characterized by high CHP2 expression in the context of drug response analyses. This apparent contradiction is not adequately explained and may confuse readers. The authors should clarify the conceptual framework distinguishing prognostic associations from treatment response predictions or revise the terminology and interpretation to ensure internal consistency.

4. The reported predictive performance, with AUC values approaching 0.98, raises concerns regarding possible overfitting or data leakage, particularly given the integration of multiple GEO datasets. It is not sufficiently clear whether model training and testing were conducted using fully independent cohorts or whether preprocessing steps may have introduced bias. Stronger evidence of model robustness is required, including clear cohort separation, independent external validation, and appropriate reporting of calibration and multivariate analyses.

5. Experimental validation is restricted to CHP2 overexpression in two colorectal cancer cell lines, without complementary loss-of-function or rescue experiments. Given that CHP2 is reported to be downregulated in colorectal cancer, knockdown or CRISPR-based loss-of-function approaches would be more biologically relevant and are necessary to support causality. While zebrafish xenograft assays provide preliminary in vivo insight, the mechanistic depth of these experiments remains limited.

6. Finally, the tumor immune microenvironment analyses rely on deconvolution algorithms and correlation-based approaches without orthogonal validation, such as immunohistochemistry, multiplex immunofluorescence, or single-cell analyses. While such analyses are acceptable as exploratory, the manuscript currently overinterprets these findings as definitive immune phenotypes.

Author Response

Comments 1:

The manuscript frames CHP2 as a regulator of PANoptosis; however, no direct experimental evidence supporting PANoptosis is provided. Cell death analyses rely primarily on Annexin V/7-AAD flow cytometry and general viability assays, which are insufficient to distinguish apoptosis, necroptosis, and pyroptosis. Canonical PANoptosis markers, including caspase-1 activation, GSDMD/GSDME cleavage, RIPK3/MLKL phosphorylation, inflammasome components, or pathway-specific inhibitor and rescue experiments, are not examined. As acknowledged by the authors, specific modes of regulated cell death were not dissected. In the absence of mechanistic validation, the PANoptosis-related conclusions remain speculative and should either be experimentally substantiated or substantially toned down to reflect association rather than mechanism.

Response:

We appreciate this professional critique. We agree that Annexin V staining alone is insufficient. We have added Western blot data for p-MLKL (necroptosis), N-GSDMD (pyroptosis), and Cleaved-Caspase3 (apoptosis). Overexpression of CHP2 concurrently activated these markers. siRNA-mediated rescue experiments confirmed that silencing CHP2 reversed these effects, providing strong evidence for its role in the PANoptosis network (Figure 12F–J), page 25.

Comments 2: The conclusions regarding selective drug resistance or sensitivity are derived entirely from computational IC50 predictions using public pharmacogenomic datasets. No experimental drug response validation is performed in colorectal cancer cell lines or patient-derived models. Given the strength of the therapeutic claims, experimental validation is essential. At a minimum, drug sensitivity assays across multiple CRC cell lines stratified by CHP2 expression should be included. Without such validation, the therapeutic implications should be presented as hypothesis-generating rather than predictive.

Response: Thank you for pointing this out. We have now performed in vitro drug sensitivity assays in HCT116 and SW480 cells. To ensure a robust proof-of-concept without redundant testing, we strategically prioritized five agents representing the two main pillars of CRC treatment: standard chemotherapy cornerstones (5-FU, Oxaliplatin, and Irinotecan) and diverse targeted mechanisms (Ribociclib for cell-cycle and Lapatinib for RTK signaling). Cells with low CHP2 expression (high-risk group) exhibited significant resistance to these standard chemotherapy agents, while showing selective sensitivity to Ribociclib and Lapatinib. These findings elevate our therapeutic claims from computational predictions to experimentally validated hypotheses (Figure 12A–E, Page 25), demonstrating the model's bidirectional predictive power across clinically relevant drug classes.

Comments 3: The manuscript reports that higher CHP2 expression is associated with improved overall and recurrence-free survival, yet subsequently defines a “high-risk” group characterized by high CHP2 expression in the context of drug response analyses. This apparent contradiction is not adequately explained and may confuse readers. The authors should clarify the conceptual framework distinguishing prognostic associations from treatment response predictions or revise the terminology and interpretation to ensure internal consistency.

Response:

We thank the reviewer for identifying this contradiction. We have synchronized the entire framework: CHP2 is a tumor suppressor, and its low expression is associated with poor prognosis. Accordingly, we have redefined the high-risk group as characterized by low CHP2 expression. This correction ensures internal consistency across survival analysis, risk modeling, and experimental observations.Page 19, lines 611-634.

Comments 4:

The reported predictive performance, with AUC values approaching 0.98, raises concerns regarding possible overfitting or data leakage, particularly given the integration of multiple GEO datasets. It is not sufficiently clear whether model training and testing were conducted using fully independent cohorts or whether preprocessing steps may have introduced bias. Stronger evidence of model robustness is required, including clear cohort separation, independent external validation, and appropriate reporting of calibration and multivariate analyses.

Response:

We thank the Editor for raising an important concern regarding the very high AUC and the potential risk of overfitting or information leakage when integrating multiple GEO datasets. We agree that the original wording did not sufficiently clarify the independence of model evaluation and may have led to misunderstanding.

To address this concern without adding additional figures or tables, we made the following revisions to improve methodological transparency and to present the results more conservatively:

(1) Clarified the scope of cohort merging and batch adjustment. We revised the Methods to specify that the merged, batch-adjusted expression matrix was used for descriptive transcriptome-wide analyses (e.g., differential expression and enrichment), and we clarified the evaluation framework for the diagnostic model to avoid potential leakage concerns.

(2) Explicitly stated cross-validated, out-of-fold performance reporting. We revised the ML Methods and the corresponding Results text to clearly indicate that AUCs (including the XGBoost AUC = 0.981) were obtained from cross-validated out-of-fold predictions (internal validation), rather than from a fully independent external cohort, and we adjusted the wording to avoid overstating generalizability based solely on internal estimates.

(3) Added an explicit limitation statement. In the Discussion, we now explicitly acknowledge that evaluation using fully independent external cohorts, as well as calibration and multivariable analyses, would further strengthen clinical generalizability; however, these analyses were limited here by incomplete and inconsistent clinical annotations in public GEO datasets.

We believe these revisions directly address the concern about possible leakage/overfitting by clarifying the evaluation procedure and by presenting the performance estimates and limitations in a more transparent and cautious manner. Mention exactly where in the revised manuscript this change can be found in Page 5, lines 219-224; Page 16, lines 556-562; Page 19, lines 588-589.

Comments 5: Experimental validation is restricted to CHP2 overexpression in two colorectal cancer cell lines, without complementary loss-of-function or rescue experiments. Given that CHP2 is reported to be downregulated in colorectal cancer, knockdown or CRISPR-based loss-of-function approaches would be more biologically relevant and are necessary to support causality. While zebrafish xenograft assays provide preliminary in vivo insight, the mechanistic depth of these experiments remains limited.

Response:

We highly value the reviewer’s perspective on the biological relevance of loss-of-function (LoF) studies. Since CHP2 is downregulated in CRC, demonstrating that its depletion can reverse anti-tumor phenotypes is indeed essential for establishing causality. We have addressed this with the following substantial experimental additions:

(1) To move beyond simple gain-of-function, we conducted siRNA-mediated rescue experiments cells with stable CHP2 overexpression. This "overexpression followed by knockdown" approach is a rigorous way to confirm that the observed induction of PANoptosis is specifically driven by CHP2 levels. As shown in Figure 12F–J, the knockdown of CHP2 significantly attenuated the activation of p-MLKL, N-GSDMD, and Cleaved-Caspase3, directly linking CHP2 to the regulation of these death pathways.

(2) To ensure the specificity of the LoF effect, we synthesized and screened three candidate siRNAs (siRNA-1, 2, and 3). All screening and validation experiments were performed in triplicate. siRNA-3 was selected for its superior and consistent silencing efficiency (Figure S3). This systematic approach minimizes potential off-target effects and strengthens the evidence for causality.

(3) While we acknowledge the inherent limitations of the zebrafish xenograft model in terms of mechanistic depth compared to mammalian models, it serves as a crucial macroscopic bridge. The zebrafish assays provided real-time, transparent observation of how CHP2 levels influence tumor cell proliferation and metastatic migration in a complex systemic environment. These in vivo results functionally corroborate our in vitro findings, demonstrating that the CHP2-induced phenotypes are not cell-culture artifacts but are reproducible in a living organism.

(4) By performing these experiments in stable cell lines where we first restored the physiologically "normal" levels of CHP2 and then silenced it, we have created a model that more accurately reflects the dynamic role of this tumor suppressor in CRC progression.

Comments 6: Finally, the tumor immune microenvironment analyses rely on deconvolution algorithms and correlation-based approaches without orthogonal validation, such as immunohistochemistry, multiplex immunofluorescence, or single-cell analyses. While such analyses are acceptable as exploratory, the manuscript currently overinterprets these findings as definitive immune phenotypes.

Response:

We appreciate the reviewer’s critical insight into the inherent limitations of computational immune deconvolution. We agree that while algorithms like CIBERSORT are powerful for landscape estimation, they provide inferred associations rather than direct physical evidence of immune cell infiltration. To address the concern of overinterpretation, we have made the following revisions:

(1) We have systematically revised the text to clarify that the TIME analyses are exploratory and hypothesis-generating. We have replaced definitive terms such as "immune mechanism" or "definitive phenotype" with more cautious descriptors like "predicted immune landscape" or "suggested associations." Page 13, lines 465-507; Page 15, lines 545-548; Page 26, lines 740-766.

(2) We have refocused the discussion to show how these predicted immune profiles (e.g., the "cold" tumor phenotype in low-CHP2/high-risk patients) are logically consistent with our experimentally validated observations of chemoresistance and poor prognosis. This alignment provides a more coherent, albeit preliminary, biological context for CHP2’s role.

(3) In response to the reviewer's concern about weak correlations, we have removed Table 1 and simplified the relevant sections to avoid over-reliance on low-coefficient correlations (r = 0.2 to 0.3). This ensures that the manuscript's narrative is driven by the most robust data points.

(4) We explicitly acknowledge in the Discussion that our current findings represent a "computational blueprint" of the CRC immune microenvironment. We state that orthogonal validation using techniques such as IHC, multiplex immunofluorescence (mIF), or spatial transcriptomics is the intended focus of our follow-up studies to confirm these immune cell spatial distributions and functional states.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors must update the paper considering the following suggestions.

The abstract is concise, yet the background and objectives of the study must be clearly mentioned.
Elaborate the results in the abstract as well. Put a summary of the results obtained.
Does implementation of shap models provide a "rationale" for the predictions? How it is helping clinicians understand which specific biological pathways are driving the resistance?
Minor: References in the whole document are not in MDPI template. Correct them.

Author Response

Comments 1 & 2 :

The abstract is concise, yet the background and objectives of the study must be clearly mentioned.

Elaborate the results in the abstract as well. Put a summary of the results obtained.

Response:

We agree that the abstract needed a more comprehensive structure to accurately reflect the study's scope.

Background & Objectives: We have added a clear background section highlighting the challenge of CRC heterogeneity and our objective to identify dual-function biomarkers.

Results Summary: The abstract now includes a detailed summary of our findings, specifically mentioning the activation of p-MLKL, N-GSDMD, and Cleaved-Caspase3 during PANoptosis and the selective sensitivity to Ribociclib and Lapatinib.

Formatting: The revised abstract is organized into Background, Methods, Results, and Conclusion sections while remaining within the 200-word limit.

Mention exactly where in the revised manuscript this change can be found in Page 1, lines 27-39.

Comments 3:

Does implementation of shap models provide a "rationale" for the predictions? How it is helping clinicians understand which specific biological pathways are driving the resistance?

Response:

We thank the reviewer for this important question. We agree that SHAP does not provide a causal biological ‘rationale’ in the mechanistic sense, nor does it directly identify which signaling pathways drive resistance. Rather, SHAP is a post-hoc interpretability method that explains the model’s rationale by attributing each prediction to contributions from the input features (genes), at both the global (cohort) level and the local (individual prediction) level.

To avoid overinterpretation and to better align with clinical needs, we revised the manuscript text (Methods and Results, Page 5, lines 229-234) to explicitly clarify the role and limitation of SHAP. Specifically, we now state that SHAP helps clinicians understand which genes/features most strongly influence the predicted probability and how those gene expression values push the prediction toward ‘resistant’ versus ‘sensitive’.We also explicitly state that pathway-level interpretation should be derived from independent functional analyses (e.g., GO/KEGG enrichment of resistance-associated genes) and that SHAP alone should be considered hypothesis-generating rather than mechanistic proof.

These revisions improve transparency and ensure that SHAP is presented appropriately as an interpretability tool for model predictions, while biological pathway inferences are supported by orthogonal analyses.

Comments 4:

Minor: References in the whole document are not in MDPI template. Correct them.

Response: We apologize for the inconsistencies in the reference formatting. We understand that some initial adjustments were made during the preliminary processing stages by the assistant editor. To ensure full compliance, we have now performed a comprehensive manual review of the entire document. All references have been strictly updated and verified according to the official MDPI template guidelines (numbered citation style and specific bibliographic formatting).

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

1) The authors have addressed the overfitting issues raised by reviewers. They now state that the AUC values shown are derived from cross-validated data the authors have addressed that in the materials and methods.

2) the data allegedly showing PANoptosis have now been strongly improved: there is now a new figure 12 that shows Western blots of necroptosis markers like MLKL and pyoptosis markers like GSDMD or cleaved CASP3 for apoptosis; this is much more precise and detailed.

3) The authors have added a new section that describes real drug chemosenitivity assays - not just computational predictions - via oncoPredict... at least 5 drugs have been used now on real cells. this is a very good addition.

4) the cell lines have been justified and their use much better explained

but there are still a few issues that remain clarification:

5) in section 3.8, you still have placeholders in the text ("Figure X") and that should be fixed. No big deal.

6) there is still no external validation cohort for any of the findings. The authors use cross-validation on the same set of merged GEO data sets. Thats not the ideal situation, which would make use of external, completely independent data sets.

7) What the authors call a "ceRNA" network also still remains entirely non-validated. Now the authors have changed the wording and consider this a "hypothesis-generating approach," but for me, it's still highly speculative. Such maps are generated in a short amount of time, but rarely add much to a biological paper. Vice versa, it's very different: A small number of validation experiments can massively increase the quality of a bioinformatics-based manuscript.

At least ONE validation step would be highly appreciated - otherwise, it's entirely hypothetical or "predicted" . Specifically, this includes lncRNAs such as the one mentioned repeatedly here: LINC006789 is prominently featured in the discussion but not in the results and discussion sections, and we know close to nothing about it. THose candidate "genes" are always and notoriously difficult to validate... personally, I wouldnt even touch them.

8) I think we still don't know how exactly CHP2 connects to the PANoptosis machinery? That also remains unvalidated and unproven, and unclear.

Author Response

Comments 1: in section 3.8, you still have placeholders in the text ("Figure X") and that should be fixed. No big deal.

Response 1: We sincerely apologize for the oversight regarding the formatting of figure citations. We have replaced it. Page 22, line 705.

Comments 2: there is still no external validation cohort for any of the findings. The authors use cross-validation on the same set of merged GEO data sets. Thats not the ideal situation, which would make use of external, completely independent data sets.

Response 2:

We appreciate the reviewer's professional concern regarding the data validation strategy.

In this study, we integrated five independent GEO cohorts (GSE14333, GSE17536, GSE20916, GSE39582, and GSE9348) alongside TCGA (COAD+READ) data to capture the biological heterogeneity across different international cohorts.

To complement this, we prioritized biological validation in functional models rather than purely computational datasets. Our findings were rigorously validated in human CRC cell lines (HCT116 and SW480) and an in vivo Zebrafish xenograft model.We believe these experimental validations provide robust evidence for the diagnostic and therapeutic relevance of CHP2.

Furthermore, we fully recognize the significance of clinical validation. In our subsequent experiments, we will strive to collect clinical samples as much as possible to further verify our findings and their clinical applicability.

Comments 3:

What the authors call a "ceRNA" network also still remains entirely non-validated. Now the authors have changed the wording and consider this a "hypothesis-generating approach," but for me, it's still highly speculative. Such maps are generated in a short amount of time, but rarely add much to a biological paper. Vice versa, it's very different: A small number of validation experiments can massively increase the quality of a bioinformatics-based manuscript.

Response 3:

We sincerely appreciate the reviewer’s professional critique. We fully agree that ceRNA networks are often "notoriously difficult to validate" and that overemphasizing such "hypothesis-generating" maps can lead readers to perceive them as a way of "stacking workload" without adding significant biological value.

(1)We have now performed qRT-PCR to provide the requested validation.Our results confirm that LINC00689 and FAM74A7 are significantly downregulated in CRC cells. Conversely, inhibitory miRNAs such as hsa-miR-1236-3p and hsa-miR-146a-3p were markedly upregulated.

(2)To avoid distracting from our core findings—the diagnostic model of CHP2 and its role in PANoptosis—we put the qRT-PCR resluts to Supplementary Materials (Figure S10). This ensures the main manuscript remains focused and concise.

(3)This reciprocal expression pattern supports a model where the loss of protective lncRNA "sponges" allows for the overabundance of inhibitory miRNAs, leading to the robust post-transcriptional repression of CHP2. By providing these data in the supplements, we show that our bioinformatic inferences have a tangible experimental basis without cluttering the main text.

Page 15, lines 547-559.

Comments 4: I think we still don't know how exactly CHP2 connects to the PANoptosis machinery? That also remains unvalidated and unproven, and unclear.

Response 4:

We appreciate the reviewer’s critical insight into the mechanistic link between CHP2 and PANoptosis. We speculate that the reviewer's concern may stem from two possibilities, which we have addressed as follows:

(1) Imprecision in Description:We recognize that our previous description of the connection might have been insufficiently clear. We have now significantly revised and expanded this section in the manuscript to better articulate the logical flow from CHP2 expression to the activation of PANoptotic executioners. Page 22, lines 697-708.

(2) Depth of Mechanistic Study:If the reviewer is seeking a deep, structural exploration of how CHP2 triggers the PANoptosome, we would like to clarify the defined scope of this work. The primary objective of this study is the identification of a robust diagnostic and prognostic biomarker for CRC—CHP2—derived from the intersection of three distinct cell death gene sets of PANoptosis.

Furthermore, this manuscript represents the preliminary research of our PhD candidate, Zetian Zhang, focusing specifically on the screening and functional validation of biomarkers. Our research group is currently conducting higher-level mechanistic studies, including molecular docking and Co-immunoprecipitation (Co-IP), to explore the direct interactions of CHP2. While our preliminary results confirm that CHP2 interacts with a key protein within the PANoptosis complex, this data remains unpublished and falls outside the scope of the current biomarker-focused study. Therefore, we unfortunately cannot provide these specific datasets at this stage. We kindly ask for your forbearance and hope you will appreciate that the current validation of terminal markers (p-MLKL, N-GSDMD, and Cleaved-Caspase3) and rescue assays provide sufficient evidence to support CHP2’s role in the PANoptic process for the purposes of this paper.

Reviewer 2 Report

Comments and Suggestions for Authors

The revision has substantially strengthened the manuscript. The newly added experiments and clarifications address several of the major concerns raised in the initial review and significantly improve the overall rigor and clarity of the study.

While some limitations remain, particularly regarding the depth of mechanistic dissection and the breadth of validation, the authors have responded constructively and made meaningful efforts to support their conclusions.

Provided that the claims are appropriately moderated and consistently framed in line with the current level of evidence, the manuscript is now considerably closer to being suitable for publication.

Author Response

Comments 1: While some limitations remain, particularly regarding the depth of mechanistic dissection and the breadth of validation, the authors have responded constructively and made meaningful efforts to support their conclusions.

Response 1: We sincerely accept your observation regarding the limitations in the depth of mechanistic dissection.

As this study primarily focuses on the identification and functional validation of CHP2 as a diagnostic and prognostic biomarker, we have centered our mechanistic validation on the terminal executioners of PANoptosis: p-MLKL, N-GSDMD, and Cleaved-Caspase3. While more exhaustive structural mechanism studies—such as molecular docking and Co-IP—are currently yielding preliminary progress in our laboratory, they represent the next phase of our PhD candidate Zetian Zhang’s research and fall outside the current scope.

We kindly ask for your forbearance and hope you will appreciate that the current validation of terminal markers and recsue assays provide sufficient evidence to support CHP2’s role in the PANoptic process for the purposes of this paper.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thanks for addressing the comments and suggestions.

Author Response

Dear Reviewer,

We are very pleased to receive your positive feedback. We sincerely thank you for the time and effort you dedicated to reviewing our manuscript. Your constructive suggestions during the initial review process were instrumental in helping us refine our work and strengthen the overall quality and clarity of the paper.

We are gratified that the revisions have addressed your comments and suggestions to your satisfaction. Thank you again for your professional support.