Review Reports
- Teagan Prouse 1,
- Rinku Majumder 1,* and
- Samarpan Majumder 2,*
Reviewer 1: Chirayu Mohindroo Reviewer 2: Anonymous Reviewer 3: Anonymous
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsIn this study Prouse et al have examined the role of PROS1 expression in PDAC. Although the concept the findings are interesting I have major concerns :-
- The survival data "2.1. High Intratumoral PROS1 expression is associated with improved survival in PDAC" in the section has not been adjusted for any clinical covariate. What stage were these patients? What treatment did they receive what stage were they diagnosed with ? These are all important points at the very least a demographics table and adjustment of multiple demographic factors, treatment regimens different databases need to be adjusted
- For the expression data was there batch correction done between the three cohorts ? If not this needs to be done and the all the analysis needs to be revised accordingly
- Simply looking at certain genes is not helpful please conduct the analysis between High and low PROS1 group first do DEGs followed by hallmark or GOBP pathways the genes can only act as an adjunct.
Author Response
Comment 1:
The survival data "2.1. High Intratumoral PROS1 expression is associated with improved survival in PDAC" in the section has not been adjusted for any clinical covariate. What stage were these patients? What treatment did they receive at the stage they were diagnosed with? These are all important points. At the very least, a demographics table and adjustment of multiple demographic factors, treatment regimens, and different databases need to be adjusted.
Response 1:
Thank you for raising this very important concern. Unfortunately, these three cohorts from the cBioportal database offer limited, incomplete, or altogether absent information regarding clinical or pathological patient stage. Furthermore, there is no information in each of these three cBioportal cohorts regarding treatment type and when treatment began after diagnosis. Even basic demographic factors, such as race or molecular subtype, are missing or have severely incomplete data, which attenuates the statistical significance of evaluating these factors. Each of these three cohorts does have sex information on each patient and the age of each patient. However, the impact of sex and age on the probability of pancreatic cancer and/or treatment success is multifaceted, and findings vary significantly between studies (see PMIDs 40425248, 35402271, 38781719). This prevents us from drawing potential conclusions from evaluating age and sex as factors in this study, given the inconclusive and mixed evidence from previous larger studies. However, the complete demographic information for the age range and the total number of male and female patients in each cohort used in the analysis is shown below, along with the overall age range of the cohort (See Supplemental Table 3).
Given this limitation of the data set, a note has been added to the Limitations section regarding the absence of treatment for each patient and the time treatment began, as well as incomplete and inconsistent data on patient stage. This will clarify that these survival ratios cannot be adjusted due to insufficient data.
Comment 2:
For the expression data, was batch correction done between the three cohorts? If not, this needs to be done, and all the analysis needs to be revised accordingly.
Response 2:
Because analyses were restricted to a single cancer type and did not involve integration of external transcriptomic datasets, additional batch correction was not applied. All stratified analyses were conducted strictly within-cohort rather than on pooled transcriptomic data across the four cohorts, thereby limiting susceptibility to cross-study technical confounding. Only standardized summary measures were combined across datasets for comparative purposes. This clarification has been incorporated into the Methods section to ensure methodological transparency.
Comment 3:
Simply looking at certain genes is not helpful. Please analyze the High and low PROS1 group first, do DEGs, followed by hallmark or GOBP pathways, and the genes can only act as an adjunct.
Response 3:
Thank you for raising these concerns. However, focusing on a particular set of genes as stratification markers is not inherently invalid. Biological phenotypes of tumors often arise from complex interactions between many genes and regulatory elements. The concept that meaningful variation must be identified through DEG lists ignores that some genes, even driver or regulatory genes, don’t show large fold-changes across conditions but still have functional significance due to network-level effects (PMID 28622505). Single-gene strategies, or sets of genes, can be biologically meaningful when the genes listed in the study have known functional roles or clinical relevance (PMID 32571208). In this study, these genes and their signaling pathways have been well-studied and characterized in several cancers, including pancreatic cancer.
The genes in this study have been well studied in vivo and/or in vitro. Thus, this study serves as a purely exploratory analysis to evaluate these key genes under low or high PROS1 expression conditions intratumorally. Future in vivo and in vitro studies conducted by our group, involving spatial transcriptomics of pancreatic cancer tissue and other RNA-sequencing methods, of high- and low-PROS1-expressing tumors or tissues, are our next step to clarify which specific gene signaling pathways are impacted by PROS1 expression.
Reviewer 2 Report
Comments and Suggestions for AuthorsThis manuscript by Prouse et al. investigates the prognostic and molecular significance of intratumoral PROS1 expression in pancreatic ductal adenocarcinoma (PDAC) through pooled analyses of publicly available cBioPortal cohorts. The authors report that high PROS1 expression is associated with improved overall survival (pooled HR = 0.581) and coordinated attenuation of multiple pro-oncogenic pathways, including EMT, hypoxia adaptation, metabolic reprogramming, and invasion-associated genes such as MMP2 and SNAI2. The topic is biologically compelling and clinically relevant, particularly in light of the well-established association between PDAC and hypercoagulability. The manuscript is clearly structured and integrates preclinical evidence with bioinformatic analysis. However, several methodological, statistical, and translational aspects require clarification and strengthening prior to publication.
Major Comments
- Fixed-effects meta-analysis and heterogeneity assessment.
The pooled hazard ratio was calculated using an inverse-variance weighted fixed-effects model; however, the manuscript does not report heterogeneity metrics such as Cochran’s Q or I² statistics. Given that the pooled cohorts (TCGA GDC, TCGAPanCancer Atlas, CPTAC Cell 2021) differ in sequencing platforms, patient composition, and clinical annotation, between-study heterogeneity is plausible. The absence of heterogeneity reporting limits the interpretability of the fixed-effects assumption that a single true effect size underlies all cohorts. The authors should report heterogeneity statistics and justify the choice of a fixed-effects model. If heterogeneity is present, a random-effects sensitivity analysis would strengthen the robustness of the survival findings. - Truncation of survival dataat30 months.
The removal of patients with overall survival greater than 30 months is unconventional and may introduce bias, particularly in PDAC where long-term survivors represent a biologically meaningful subgroup. Excluding these patients potentially attenuates effect estimates and limits generalizability. The manuscript should clearly justify this decision and ideally provide a sensitivity analysis including the full survival range to demonstrate that the association between PROS1 expression and outcome remains consistent. - Stratification of PROS1 expression based on cohortmean.
High versus low PROS1 expression was defined relative to the cohort mean after outlier trimming. While the authors justify this by approximate normal distribution, mean-based dichotomization is sensitive to distributional shifts and may not be stable across cohorts. Alternative stratification approaches, such as median split, tertiles, or maximally selected rank statistics, could provide more robust classification. The manuscript would benefit from either additional justification of the chosen cutoff or demonstration that the findings are not dependent on the specific stratification strategy. - Interpretation of correlation-based attenuation categories.
The manuscript introduces categorical labels such as “moderately attenuated,” “strongly attenuated,” and “severely attenuated” based onΔρ thresholds. While statistically defined, these descriptors risk implying biological causality or magnitude beyond what correlation analyses can support. Since the analyses are purely associative and derived from stratified expression comparisons, causal language in the Discussion should be moderated. The authors should clarify that these findings represent correlative network behavior rather than direct regulatory effects of PROS1. - Expansion of translational perspective in the Discussion.
While the manuscript effectively positions PROS1 as a potential prognostic biomarker and molecular regulator in PDAC, the translational implicationsremain underdeveloped. Given that PROS1 encodes a secreted anticoagulant protein and that PDAC exhibits systemic biological effects, it would be valuable to expand the Discussion to address how PROS1-associated molecular signatures could be validated in a clinically applicable manner. In particular, recent advances in non-invasive biomarker platforms combining mid-infrared spectroscopic analysis with artificial intelligence–based modeling of extracellular vesicle signatures have demonstrated the feasibility of detecting tumor-associated molecular alterations in liquid biopsy samples. Such approaches could, in principle, enable dynamic monitoring of PROS1-related pathway modulation or broader coagulation–tumor signaling axes in PDAC patients. Incorporating a forward-looking paragraph outlining how emerging AI-driven liquid biopsy technologies may complement transcriptomic findings would significantly strengthen the translational impact of the study (https://doi.org/10.3390/cells14231909). - Clarification of the proposed dual role of PROS1 in tumor progression and thrombosis.
The manuscript proposes a dual-function model linking PROS1 to both tumor suppression and hypercoagulability modulation. While conceptually attractive, the present study does not include clinical VTE data or circulating Protein S measurements. Intratumoral PROS1 expression does not necessarily reflect systemic anticoagulant activity. Therefore, the Discussion should more clearly delineate that the thrombotic implicationsremain hypothetical and that the current findings pertain strictly to tumor transcriptomic associations.
Minor Comments
- Clarify whether pooled p-values were two-tailed and provide heterogeneity metrics for both hazard ratio andcorrelation meta-analyses. Additionally, indicate whether proportional hazards assumptions were assessed in the individual cohorts.
- It would be helpful to specify whether multivariable adjustments (stage, grade, treatment status) were possible or unavailable in the selected datasets. The absence of adjustment should be explicitly acknowledged as a limitation.
- The repeated use of “Independent expression” for statisticallysignificant but modest correlations may be confusing. Consider revising the terminology to reflect weak-to-moderate association rather than independence.
- “Maental-Haenzel” should be corrected to “Mantel-Haenszel” in Table 1. Minor formatting inconsistencies in gene names should be addressed.
- The limitations should explicitly mention the absence of protein-level validation, lack of immune infiltration quantification beyond selected markers, and absence of experimental validation of the inferred signaling network.
- In the Results section, the manuscript refers to four cohorts overall, while hazard ratio pooling was performed on three cohorts due to lack of survival data in one dataset. This distinction should be made clearer early in the Results to avoid confusion.
- Gene symbols should be consistently formatted in italics (e.g.,PROS1, MMP2, SNAI2, NOTCH1) throughout the manuscript according to journal guidelines. In several instances formatting appears inconsistent.
- Although most abbreviations are defined, ensure that all gene symbols and technical terms (e.g., EMT,Δρ) are defined at first appearance in the Results section for clarity.
Author Response
Major Comments
Comment 1:
Fixed-effects meta-analysis and heterogeneity assessment.
The pooled hazard ratio was calculated using an inverse-variance-weighted fixed-effects model; however, the manuscript does not report heterogeneity metrics such as Cochran’s Q or I². Given that the pooled cohorts (TCGA GDC, TCGAPanCancer Atlas, CPTAC Cell 2021) differ in sequencing platforms, patient composition, and clinical annotation, between-study heterogeneity is plausible. The absence of heterogeneity reporting limits the interpretability of the fixed-effects assumption that a single true effect size underlies all cohorts. The authors should report statistics on heterogeneity and justify the choice of a fixed-effects model. If heterogeneity is present, a random-effects sensitivity analysis would strengthen the robustness of the survival findings.
Response 1:
Thank you for your suggestion. We have conducted a heterogeneity assessment and have found the following results, which have been added to the Results section.
Between-cohort heterogeneity was evaluated with Cochran’s Q statistic and p-value and quantified with the I² metric for each gene. Across 49 genes analyzed, all Q tests were nonsignificant (p <0.05), indicative that no variability exists beyond expected sampling error. Only six genes demonstrated Q values exceeding degrees of freedom (df=3), and the I² values exceeded 26% for just five genes, consistent with low-to-moderate heterogeneity by conventional thresholds (Supplemental Table 2). As between-cohort variability was minimal and consistent with sampling variation, a fixed-effects model was considered appropriate for pooled analyses.
A fixed-effects model is justified for the survival analysis. When the number of studies is small (k ≤ 5), estimation of between-study variance in random-effects models is statistically unstable and may yield imprecise or biased results (PMID: 19381330, 24548571), and heterogeneity metrics such as I² are known to perform poorly in small meta-analyses (PMIDs: 25880989). The Cochrane Handbook cautions that random-effects variance estimates are unreliable when few studies are available (Cochrane Handbook for Systematic Reviews of Interventions, 2022).
Conceptually, random-effects models assume that each cohort estimates a different underlying true effect (PMID 26061376). In the present study, all cohorts represent the same disease entity and applied consistent survival modeling, supporting the assumption of a common underlying effect. Given the absence of statistically significant heterogeneity, low I² estimates, and the known instability of between-study variance estimation with only three cohorts, a fixed-effects meta-analytic framework was selected as the most statistically appropriate and methodologically justified approach. This has been added to the Methods section as further clarification.
Comment 2:
Truncation of survival data at 30 months.
The removal of patients with overall survival greater than 30 months is unconventional and may introduce bias, particularly in PDAC, where long-term survivors represent a biologically meaningful subgroup. Excluding these patients may attenuate effect estimates and limit generalizability. The manuscript should clearly justify this decision and ideally provide a sensitivity analysis, including the full survival range, to demonstrate that the association between PROS1 expression and outcome remains consistent.
Response 2:
Thank you for raising this important point. Notably, each patient cohort had few patients with survival greater than 30 months. The TCGA GDC cohort had a total of 158 patients, of whom 15 survived for more than 30 months after PDAC diagnosis. The TCGA PanCancer analysis had 159 patients, of whom 17 survived longer than 30 months after PDAC diagnosis. The CPTAC Cell 2021 cohort only had 2 patients of the initially 114 patients who survived longer than 30 months. The truncation of survival data to 30 months was thus done to remove small outliers that drastically affect data skew and unjustly impact overall statistical significance.
Furthermore, PDAC has a mortality rate of nearly 80% within the first year of diagnosis (PMID 40971231), and another study showed that the median survival time after PDAC diagnosis is about 4 months, with only 19% of patients surviving past one-year on average (PMID 23589669). This clarification has been added to the manuscript under Section 4.3 to further support the 30-month truncation.
Comment 3:
Stratification of PROS1 expression based on cohort mean.
High versus low PROS1 expression was defined relative to the cohort mean after outlier trimming. While the authors justify this by approximate normal distribution, mean-based dichotomization is sensitive to distributional shifts and may not be stable across cohorts. Alternative stratification approaches, such as median split, tertiles, or maximally selected rank statistics, could provide more robust classification. The manuscript would benefit from either additional justification of the chosen cutoff or demonstration that the findings are not dependent on the specific stratification strategy.
Response 3:
Patient samples were dichotomized into high- and low-PROS1 expression groups based on the cohort mean expression after outlier trimming. Absolute expression thresholds are not established for PROS1, and several prognostic biomarker studies similarly rely on study-specific distribution-based thresholds to stratify continuous expression values into discrete groups for survival and differential analyses. The mean is an accepted cohort-based approach when data are normally distributed, as in the case of the PROS1 expression data after outlier removal (PMIDs 26873370, 20500820). This statement has been added to the manuscript as further justification of the chosen cut-off.
Comment 4:
Interpretation of correlation-based attenuation categories.
The manuscript introduces categorical labels such as “moderately attenuated,” “strongly attenuated,” and “severely attenuated” based onΔρ thresholds. While statistically defined, these descriptors risk implying biological causality or magnitude beyond what correlation analyses can support. Since the analyses are purely associative and derived from stratified expression comparisons, causal language in the Discussion should be moderated. The authors should clarify that these findings represent correlative network behavior rather than direct regulatory effects of PROS1.
Response 4:
We thank the reviewer for this thoughtful comment. We agree that the original descriptors (“moderately,” “strongly,” and “severely attenuated”) could imply biological causality beyond what our correlation-based analyses can support. To address this concern, we have revised the manuscript to remove language that implies magnitude. Specifically, we replaced the term “attenuation” with statistical terminology, using instead lower-, intermediate-, and higher-magnitude Δρ categories throughout the manuscript. In addition, we have clarified in the Discussion that these findings reflect changes in co-expression network behavior rather than direct regulatory effects of PROS1. We believe these revisions more accurately frame the results within the limits of correlation-based analyses.
Comment 5:
Expansion of translational perspective in the Discussion.
While the manuscript effectively positions PROS1 as a potential prognostic biomarker and molecular regulator in PDAC, the translational implications remain underdeveloped. Given that PROS1 encodes a secreted anticoagulant protein and that PDAC exhibits systemic biological effects, it would be valuable to expand the Discussion to address how PROS1-associated molecular signatures could be validated in a clinically applicable manner. In particular, recent advances in non-invasive biomarker platforms combining mid-infrared spectroscopic analysis with artificial intelligence–based modeling of extracellular vesicle signatures has demonstrated the feasibility of detecting tumor-associated molecular alterations in liquid biopsy samples. Such approaches could, in principle, enable dynamic monitoring of modulation of the PROS1-related pathway or broader coagulation–tumor signaling axes in PDAC patients. Incorporating a forward-looking paragraph outlining how emerging AI-driven liquid biopsy technologies may complement transcriptomic findings would significantly strengthen the study's translational impact (https://doi.org/10.3390/cells14231909).
Response 5:
We appreciate the reviewer for this thoughtful suggestion to strengthen the manuscript. We have now incorporated a full forward-looking paragraph (as suggested) that directly references Di Santo et al. (2025, Cells), which provides proof-of-concept for AI-driven analysis of extracellular vehicles (EVs) in cancer using mid-infrared (Fourier-transform infrared, FTIR) spectroscopy, while noting its broader applicability (though not yet PDAC-specific).
Comment 6:
Clarification of the proposed dual role of PROS1 in tumor progression and thrombosis.
The manuscript proposes a dual-function model linking PROS1 to both tumor suppression and hypercoagulability modulation. While conceptually attractive, the present study does not include clinical VTE data or circulating Protein S measurements. Intratumoral PROS1 expression does not necessarily reflect systemic anticoagulant activity. Therefore, the Discussion should more clearly delineate that the thrombotic implications remain hypothetical and that the current findings pertain strictly to tumor transcriptomic associations.
Response 6:
We appreciate the reviewer's critical insight. Our findings pertain primarily to tumor-intrinsic transcriptomic associations and potential local regulatory effects of PROS1 within the PDAC microenvironment, rather than systemic anticoagulant activity. Although no published longitudinal study has been objectively designed to assess free or bound Protein S (PS) levels in PDAC patients, Lindahl et al. (PMID 1394036) observed decreasing PS levels with increasing disease severity. We have now incorporated this reference into the revised manuscript. We are presently generating data on free PS levels in patients using longitudinal blood draws; however, the premise is beyond the scope of the present manuscript. Future investigations integrating circulating PS levels, functional coagulation assays, and VTE incidence with tumor PROS1 expression could elucidate whether PROS1 contributes to a dual (tumor-suppressive yet thrombosis-modulating) role in PDAC biology, potentially informing precision strategies targeting coagulation-tumor crosstalk.
Minor Comments
Comment 1:
Clarify whether pooled p-values were two-tailed and provide heterogeneity metrics for both hazard ratio and correlation meta-analyses. Additionally, indicate whether proportional hazards assumptions were assessed in the individual cohorts.
Response 1:
Thank you for your insight – we have clarified that pooled p-values were two-tailed. Heterogeneity metrics have been added for the correlation meta-analyses; however, heterogeneity metrics were not appropriate for the hazard ratio meta-analyses due to the small sample size, as discussed in response 1.
We did not formally test proportional hazards assumptions, as our primary objective was to estimate overall associations between PROS1 expression and survival rather than to model time-dependent effects. Given the use of single-covariate Cox models and consistent directional findings across cohorts, we considered the Cox model an appropriate summary measure. We have clarified this in the Methods and moderated interpretation accordingly.
Comment 2:
It would be helpful to specify whether multivariable adjustments (stage, grade, treatment status) were possible or unavailable in the selected datasets. The absence of adjustment should be explicitly acknowledged as a limitation.
Response 2:
Thank you – we have included this in the Limitations section accordingly.
Comment 3:
The repeated use of “Independent expression” for statistically significant but modest correlations may be confusing. Consider revising the terminology to reflect weak-to-moderate association rather than independence.
Response 3:
Thank you for this insight. We have changed the terminology to “independent correlation” instead to highlight the correlative nature of the conclusions.
Comment 4:
“Maental-Haenzel” should be corrected to “Mantel-Haenszel” in Table 1. Minor formatting inconsistencies in gene names should be addressed.
Response 4:
Thank you for highlighting this – we have corrected these issues accordingly.
Comment 5:
The limitations should explicitly mention the absence of protein-level validation, lack of immune infiltration quantification beyond selected markers, and absence of experimental validation of the inferred signaling network.
Response 5:
Thank you, this has been included in the limitations section.
Comment 6:
In the Results section, the manuscript refers to four cohorts overall, while hazard ratio pooling was performed on three cohorts due to a lack of survival data in one dataset. This distinction should be made clearer early in the Results to avoid confusion.
Response 6:
We have made this distinction, as recommended.
Comment 7:
Gene symbols should be consistently formatted in italics (e.g., PROS1, MMP2, SNAI2, NOTCH1) throughout the manuscript according to journal guidelines. In several instances, formatting appears inconsistent.
Response 7:
Formatting has been corrected to be consistent.
Comment 8:
Although most abbreviations are defined, ensure that all gene symbols and technical terms (e.g., EMT, Δρ) are defined at first appearance in the Results section for clarity.
Response 8:
Thank you, these terms have been clarified.
Reviewer 3 Report
Comments and Suggestions for AuthorsTeagan Prouse and colleagues’ original article investigates the clinical and molecular significance of intratumoral PROS1 expression in pancreatic ductal adenocarcinoma (PDAC). Through pooled analyses of publicly available transcriptomic datasets from cBioPortal, the authors report that high PROS1 expression is associated with improved overall survival and with attenuation of several genes implicated in EMT, invasion, hypoxia, metabolic reprogramming, and oncogenic signaling. PDAC remains one of the most lethal malignancies, and the biological interface between tumor progression and cancer-associated thrombosis represents an emerging translational field. Furthermore, a molecule such as Protein S, potentially influencing both coagulation and tumor biology, is conceptually attractive as a dual-function biomarker or therapeutic target. Therefore, the topic dealt in this original article is timely and of interest to the readership of the journal. However, although the study presents intriguing associations, several methodological and interpretative issues limit the robustness of the conclusions. Additional analytical refinement and focused experimental validation are required before the mechanistic implications can be considered sufficiently substantiated.
Major issues
- As for the front matter:
- Despite appropriately reflecting the association between PROS1 expression and survival, the title implies mechanistic causation. Given that the present study is correlative, this phrasing should be moderated to reflect association rather than demonstrated regulatory activity.
- Similarly, in the abstract, statements suggesting that PROS1 “demonstrates” regulatory roles should be softened unless supported by direct functional validation in human PDAC samples. Including a brief mention of the dichotomization strategy and the fixed-effects meta-analytic approach would improve transparency.
- Keyword selection could be improved and expanded (to include other pivotal aspects of the study) to increase data dissemination in public online platforms.
- The Introduction would benefit from broader contextualization within current efforts targeting PDAC metabolic reprogramming, EMT plasticity, and survival signaling. Several recent mechanistic and pharmacological studies have identified vulnerabilities in PDAC linked to these same biological axes. Incorporating recent literature addressing EMT modulation in PDAC, metabolic rewiring and glycolytic dependence, therapeutic targeting of survival and invasion pathways, would strengthen the translational positioning of the study and situate PROS1 within the evolving molecular landscape of PDAC research. This would also allow clearer integration of how PROS1-associated attenuation of EMT, metabolic, and invasive programs aligns with ongoing mechanistic and drug-development efforts in pancreatic cancer.
- A central claim of the manuscript is that high PROS1 expression is associated with improved survival. However, the analyses appear to be univariable. Is PROS1 independently prognostic, or does it reflect underlying clinicopathologic or molecular features (e.g., stage, subtype, stromal content)? To address this concern, the authors should perform multivariable Cox regression analyses within each cohort adjusting for available clinical variables (age, stage, grade, treatment, molecular subtype where available). Demonstrating independent prognostic value is essential before proposing PROS1 as a candidate biomarker. In addition, modeling PROS1 as a continuous variable (rather than dichotomized at the mean) would help determine whether the survival association reflects a dose–response relationship. The exclusion of patients with >30 months survival also requires justification, as this may bias hazard estimation. A sensitivity analysis including all patients would strengthen the robustness of the findings.
- Because PROS1 may be expressed in stromal or immune compartments, an unresolved question is whether the observed transcriptomic associations reflect tumor cell–intrinsic biology or differences in tumor microenvironment composition. This is particularly relevant given the strong attenuation observed in EMT- and invasion-related genes. To address this, the authors should (a) adjust analyses using tumor purity estimates (e.g., ESTIMATE scores or similar algorithms); (b) examine correlations between PROS1 expression and stromal/immune signatures. Without this step, it is difficult to conclude that PROS1 expression directly influences tumor-intrinsic signaling.
- The manuscript reports strong attenuation of SNAI2, MMP2, and other EMT/invasion-associated genes in high PROS1 tumors. Does PROS1 directly regulate EMT and invasion programs in PDAC cells, or is it simply associated with a less aggressive molecular subtype? Because the current study is purely correlative, functional validation would be necessary to substantiate the mechanistic implications. To directly address this concern, the following experimental approaches would be appropriate and coherent with the authors’ framework: (i) validation of PROS1 protein expression in human PDAC tissue (e.g., IHC on tumor microarrays) to confirm that RNA-level differences translate into tumor-cell protein expression rather than stromal localization; (ii) correlation of PROS1 protein levels with EMT markers (e.g., CDH2, SNAI2, MMP2) at the protein level in patient samples. This would provide orthogonal validation of the transcriptomic associations. Furthermore, functional modulation of PROS1 in PDAC cell models (overexpression or knockdown) followed by EMT marker assessment (CDH2, SNAI2, VIM), invasion assays (e.g., transwell/Matrigel), metabolic readouts (e.g., glycolytic flux or HK2 expression). These experiments would directly answer whether PROS1 functionally suppresses EMT/invasion programs, thereby converting correlative findings into mechanistically supported conclusions. Importantly, these are not unrelated biochemical expansions, but rather targeted validations of the precise pathways implicated by the bioinformatic results.
- The claim of “coordinated suppression of oncogenic pathways” is currently based on a curated gene panel. To support this more robustly, the authors should perform unbiased pathway enrichment analysis (e.g., GSEA) comparing high vs. low PROS1 tumors, reporting of enrichment scores for EMT, hypoxia, glycolysis, and invasion signatures. This would provide systems-level confirmation that the observed gene-level associations reflect broader biological programs.
Minor Issues
- Terminology such as “severely attenuated” may overstate conclusions derived from differential correlation analysis and could be moderated.
- Clarification of outlier removal and truncation of survival data is needed.
- Minor formatting inconsistencies in tables should be corrected.
Overall recommendation: Major revisions
This study addresses a biologically meaningful and clinically relevant question. The association between high intratumoral PROS1 expression, improved survival, and reduced expression of EMT/invasion/metabolic genes is intriguing and potentially impactful. However, the current manuscript remains largely correlative. Critical analytical refinements (multivariable modeling, tumor purity adjustment, pathway enrichment) and focused experimental validation addressing EMT/invasion regulation are necessary to substantiate the mechanistic implications. With strengthened analytical rigor and targeted experimental validation directly addressing the biological questions raised above, the manuscript could make a meaningful contribution to the field of PDAC biology and translational oncology.
Author Response
Major Comments
Comment 1a-c:
As for the front matter:
- Despite appropriately reflecting the association between PROS1 expression and survival, the title implies mechanistic causation. Given that the present study is correlative, this phrasing should be adjusted to reflect association rather than to demonstrate regulatory activity.
Response 1a:
The title has been changed accordingly to reflect association rather than demonstrate regulatory activity.
- Similarly, in the abstract, statements suggesting that PROS1 “demonstrates” regulatory roles should be softened unless supported by direct functional validation in human PDAC samples. Including a brief mention of the dichotomization strategy and the fixed-effects meta-analytic approach would improve transparency.
Response 1b:
These changes have been made as suggested.
- Keyword selection could be improved and expanded (to include other pivotal aspects of the study) to increase data dissemination in public online platforms.
Response 1c:
We have added three additional keywords to expand data dissemination.
Comment 2:
The Introduction would benefit from broader contextualization within current efforts targeting PDAC metabolic reprogramming, EMT plasticity, and survival signaling. Several recent mechanistic and pharmacological studies have identified vulnerabilities in PDAC linked to these same biological axes. Incorporating recent literature addressing EMT modulation in PDAC, metabolic rewiring, and glycolytic dependence, therapeutic targeting of survival and invasion pathways would strengthen the translational positioning of the study and situate PROS1 within the evolving molecular landscape of PDAC research. This would also allow clearer integration of how PROS1-associated attenuation of EMT, metabolic, and invasive programs aligns with ongoing mechanistic and drug-development efforts in pancreatic cancer.
Response 2:
Thank you for this insight. We have added a paragraph to the introduction to discuss these points.
Comment 3:
A central claim of the manuscript is that high PROS1 expression is associated with improved survival. However, the analyses appear to be univariable. Is PROS1 independently prognostic, or does it reflect underlying clinicopathologic or molecular features (e.g., stage, subtype, stromal content)? To address this concern, the authors should perform multivariable Cox regression analyses within each cohort, adjusting for available clinical variables (age, stage, grade, treatment, molecular subtype where available). Demonstrating independent prognostic value is essential before proposing PROS1 as a candidate biomarker. In addition, modeling PROS1 as a continuous variable (rather than dichotomized at the mean) would help determine whether the survival association reflects a dose–response relationship. The exclusion of patients with >30 months survival also requires justification, as this may bias hazard estimation. A sensitivity analysis including all patients would strengthen the robustness of the findings.
Response 3:
Thank you for raising these very important concerns. Unfortunately, these three cohorts from the cBioportal database offer limited, incomplete, or altogether absent information regarding clinical or pathological patient stage. Furthermore, there is no information in each of these three cBioportal cohorts regarding treatment type and when treatment began after diagnosis. Even basic demographic factors, such as race or molecular subtype, are missing or have severely incomplete data, which attenuates statistical significance. Each of these three cohorts does have sex information on each patient and the age of each patient. However, the impact of sex and age on the probability of pancreatic cancer and/or treatment success is multifaceted, and findings vary significantly between studies (PMIDs 40425248, 35402271, 38781719).
Thank you for your suggestion to model PROS1 as a continuous variable to determine whether the survival association reflects a dose-response relationship. We have conducted a Cox regression analysis of each of the cohorts used in the survival analysis and found the following results, shown in Supplemental Table 1.
When modeled as a continuous variable, PROS1 expression was not significantly associated with overall survival in all three cohorts in a dose-dependent manner (HR per unit increase = ~1 for TCGA and GDC, and HR per unit increase ~0.9 for CPTAC Cell 2021). This suggests that the dichotomized survival difference may not reflect a linear dose-response relationship on PROS1 expression. This information has been added to the manuscript accordingly to provide further clarification.
Thank you for raising this important point about the 30-month truncation. Notably, each patient cohort had few patients with survival greater than 30 months. The TCGA GDC cohort had a total of 158 patients, of whom 15 survived for more than 30 months after PDAC diagnosis. The TCGA PanCancer analysis had 159 patients, of whom 17 survived longer than 30 months after PDAC diagnosis. The CPTAC Cell 2021 cohort only had 2 patients of the initially 114 patients who survived longer than 30 months. The truncation of survival data to 30 months was thus done to remove small outliers that drastically affect data skew and unjustly impact overall statistical significance. Thus, a sensitivity analysis would not be useful.
Furthermore, PDAC has a mortality rate of nearly 80% within the first year of diagnosis (PMID 40971231), and another study showed that the median survival time after PDAC diagnosis is about 4 months, with only 19% of patients surviving past one-year on average (PMID 23589669). This clarification has been added to the manuscript under Section 4.3 to further support the 30-month truncation.
Comment 4:
Because PROS1 may be expressed in stromal or immune compartments, an unresolved question is whether the observed transcriptomic associations reflect tumor cell–intrinsic biology or differences in tumor microenvironment composition. This is particularly relevant given the strong attenuation observed in EMT- and invasion-related genes. To address this, the authors should (a) adjust analyses using tumor purity estimates (e.g., ESTIMATE scores or similar algorithms); (b) examine correlations between PROS1 expression and stromal/immune signatures. Without this step, it is difficult to conclude that PROS1 expression directly influences tumor-intrinsic signaling.
Response 4:
We appreciate the reviewer’s concern regarding potential microenvironmental confounding in bulk transcriptomic analyses. We elected not to adjust our models using ESTIMATE or similar tumor purity scores because these metrics are themselves derived from overlapping gene expression signatures and could introduce analytical circularity (PMIDs 23975157, 32747523). Furthermore, bulk deconvolution approaches do not resolve compartment-specific expression (PMIDs 26192919, 25665023), and the current study is framed as characterizing tumor-level transcriptomic associations, recognizing the limits of bulk RNA-seq in resolving cellular origin (PMIDs 29795070, 28985577). The attenuation we observe in EMT/invasion programs is unlikely to be explained solely by stromal dilution, given the complexity of EMT signatures in mixed cell populations (PMIDs 24777541, 23817591). We agree that single-cell or spatial transcriptomic data would enhance resolution of cell-type expression, but this is beyond the scope of the current paper (PMIDs 34558315, 28252000).
Comment 5:
The manuscript reports strong attenuation of SNAI2, MMP2, and other EMT/invasion-associated genes in high PROS1 tumors. Does PROS1 directly regulate EMT and invasion programs in PDAC cells, or is it simply associated with a less aggressive molecular subtype? Because the current study is purely correlative, functional validation would be necessary to substantiate the mechanistic implications. To directly address this concern, the following experimental approaches would be appropriate and coherent with the authors’ framework: (i) validation of PROS1 protein expression in human PDAC tissue (e.g., IHC on tumor microarrays) to confirm that RNA-level differences translate into tumor-cell protein expression rather than stromal localization; (ii) correlation of PROS1 protein levels with EMT markers (e.g., CDH2, SNAI2, MMP2) at the protein level in patient samples. This would provide orthogonal validation of the transcriptomic associations. Furthermore, functional modulation of PROS1 in PDAC cell models (overexpression or knockdown) followed by EMT marker assessment (CDH2, SNAI2, VIM), invasion assays (e.g., transwell/Matrigel), metabolic readouts (e.g., glycolytic flux or HK2 expression). These experiments would directly answer whether PROS1 functionally suppresses EMT/invasion programs, thereby converting correlative findings into mechanistically supported conclusions. Importantly, these are not unrelated biochemical expansions, but rather targeted validations of the precise pathways implicated by the bioinformatic results.
Response 5:
We thank the reviewer for this insightful and constructive suggestion, which is well taken. We fully agree that a comprehensive high vs. low PROS1 differential expression analysis (followed by Hallmark and GO BP pathway enrichment) would provide valuable additional context, and that the genes highlighted in our manuscript (including SNAI2, MMP2, and other EMT/invasion-associated transcripts) ultimately serve as part of a broader molecular program rather than isolated readouts.
However, these deeper bioinformatic expansions—as well as the functional validation experiments proposed fall beyond the scope of the present manuscript. Our study is designed as a purely correlative, tumor-intrinsic transcriptomic analysis of PROS1 associations in publicly available and in-house PDAC cohorts. The observed strong attenuation of SNAI2, MMP2, and related EMT/invasion programs in PROS1-high tumors is presented explicitly as an association with a less aggressive molecular subtype, not as direct mechanistic evidence of regulation by PROS1. We have clearly stated this correlative nature throughout the manuscript (Results, Discussion, and Limitations sections) and have avoided any claims of causality.
We believe that the functional and mechanistic studies outlined by the reviewer represent a natural and important next phase of investigation that would best be addressed in a dedicated follow-up study. Such work would require substantial new experimental resources (patient data collection, multiple PDAC cell line models with stable genetic modulation, orthogonal functional assays, etc.) that are outside the current manuscript’s focus and timeline.
We have therefore retained our current presentation of the transcriptomic associations (including the attenuation of SNAI2/MMP2 and other EMT genes in PROS1-high tumors) while reinforcing in the revised Discussion that these observations are correlative and that functional validation will be required to establish direct regulatory roles. We have also cross-referenced this point to the newly added forward-looking paragraph on liquid-biopsy/EV-based monitoring of PROS1-associated signatures, which we believe appropriately frames the translational potential without overstepping the correlative nature of the present work.
We hope this clarification adequately addresses the reviewer’s concern while preserving the intended scope and focus of the manuscript.
Comment 6:
The claim of “coordinated suppression of oncogenic pathways” is currently based on a curated gene panel. To support this more robustly, the authors should perform unbiased pathway enrichment analysis (e.g., GSEA) comparing high vs. low PROS1 tumors, reporting of enrichment scores for EMT, hypoxia, glycolysis, and invasion signatures. This would provide systems-level confirmation that the observed gene-level associations reflect broader biological programs.
Response 6:
Thank you for raising these concerns. However, focusing on a particular set of genes as stratification markers is not inherently invalid. Biological phenotypes of tumors often arise from complex interactions between many genes and regulatory elements. The concept that meaningful variation must be identified through DEG lists ignores that some genes, even driver or regulatory genes, don’t show large fold-changes across conditions but still have functional significance due to network-level effects (PMID 28622505). Single-gene strategies, or sets of genes, can be biologically meaningful when the genes listed in the study have known functional roles or clinical relevance (PMID 32571208). In this study, these genes and their signaling pathways have been well-studied and characterized in several cancers, including pancreatic cancer.
The genes in this study have been well studied in vivo and/or in vitro. Thus, this study serves as a purely exploratory analysis to evaluate these key genes under low or high PROS1 expression conditions within the tumor. Future in vivo and in vitro studies conducted by our group, involving spatial transcriptomics of pancreatic cancer tissue and other RNA-sequencing methods, of high- and low-PROS1-expressing tumors or tissues, are our next step to clarify which specific gene signaling pathways are impacted by PROS1 expression. We will revise the statements regarding “coordinated suppression of oncogenic pathways” to discuss our findings more conservatively in order to adequately address this concern.
Minor Comments
Comment 1:
Terminology such as “severely attenuated” may overstate conclusions derived from differential correlation analysis and could be moderated.
Response 1:
We thank the reviewer for this thoughtful comment. We agree that the original descriptors (“moderately,” “strongly,” and “severely attenuated”) could imply biological causality beyond what our correlation-based analyses can support. To address this concern, we have revised the manuscript to remove language that implies magnitude. Specifically, we replaced the term “attenuation” with statistical terminology, using instead lower-, intermediate-, and higher-magnitude Δρ categories throughout the manuscript. In addition, we have clarified in the Discussion that these findings reflect changes in co-expression network behavior rather than direct regulatory effects of PROS1. We believe these revisions more accurately frame the results within the methodological limits of correlation-based analyses.
Comment 2:
Clarification of outlier removal and truncation of survival data is needed.
Response 2:
Thank you for raising this important point. Notably, each patient cohort had few patients with survival greater than 30 months. The TCGA GDC cohort had a total of 158 patients, of whom 15 survived for more than 30 months after PDAC diagnosis. The TCGA PanCancer analysis had 159 patients, of whom 17 survived longer than 30 months after PDAC diagnosis. The CPTAC Cell 2021 cohort only had 2 patients of the initially 114 patients who survived longer than 30 months. The truncation of survival data to 30 months was thus done to remove small outliers that drastically affect data skew and unjustly impact overall statistical significance.
Furthermore, PDAC has a mortality rate of nearly 80% within the first year of diagnosis (PMID 40971231), and another study showed that the median survival time after PDAC diagnosis is about 4 months, with only 19% of patients surviving past one-year on average (PMID 23589669). This clarification has been added to the manuscript under Section 4.3 to further support the 30-month truncation.
Comment 3:
Minor formatting inconsistencies in tables should be corrected.
Response 3:
Thank you for this suggestion – we have fixed the minor formatting inconsistencies in each table.
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe article, as revised, addresses all my comments and is suitable for publication. I would only recommend checking the text for possible spelling errors and verifying the formatting of the references.
Author Response
Comment 1: The article, as revised, addresses all my comments and is suitable for publication. I would only recommend checking the text for possible spelling errors and verifying the formatting of the references.
Response 1: Thank you for your valuable insight. We checked the text for potential spelling errors and verified the reference format to be compatible with IJMS guidelines.
Reviewer 3 Report
Comments and Suggestions for AuthorsI would like to thank the authors for their careful and constructive responses to the reviewer comments. The revisions have substantially improved the clarity and interpretative framing of the manuscript. These revisions strengthen the manuscript and better align the conclusions with the scope of the presented data. Most of the concerns raised in the initial review have therefore been adequately addressed. Overall, the authors responded respectfully and made some useful changes, but several of the most important methodological concerns were only partially addressed and would benefit from additional clarification. I believe that there are two major scientific concerns remain insufficiently addressed and require further clarification to ensure methodological transparency: (1) survival analysis methodology; (2) pathway-level interpretation of targeted gene panel.
- The 30-month truncation justification is statistically questionable. Long survivors are not outliers in survival analysis, but they are legitimate right-tail observations. Therefore, removing them biases hazard estimation. Also, survival analysis methods (Kaplan-Meier, Cox models) are specifically designed to handle long survivors. The epidemiological argument (high mortality in PDAC) raised by the authors does not justify removing data. Indeed, the correct approach is (i) keep all data, (ii) censor naturally or (iii) run sensitivity analysis. Hence, despite the small number reported (15–17 patients) probably does not change the result materially, the lack of sensitivity analysis represents an issue of methodological clarity. While the number of such cases appears small, it would be helpful for the authors to explicitly discuss this limitation in the Methods/Discussion and clarify whether the primary conclusions remain unchanged when all survival data are retained.
- The continuous analysis performed contradicts the authors’ main claim, since they found that PROS1 continuous expression is not associated with survival. This implies that the authors’ dichotomized result may reflect threshold effects or statistical artifact. Indeed, the authors were correct by reporting it, but the manuscript must clearly state that the prognostic value is not robust across modeling approaches. Please clarify that continuous modeling does not support a linear prognostic effect and discuss this limitation explicitly in the Discussion.
- The manuscript still interprets the attenuation of several EMT-, hypoxia-, and invasion-associated genes as indicative of broader biological programs. However, these conclusions are derived from a curated panel of genes rather than an unbiased transcriptome-wide analysis. While the authors have clarified the exploratory nature of the study, the manuscript should ensure that statements referring to “coordinated suppression of oncogenic pathways” or related pathway-level interpretations remain clearly framed as associations derived from selected markers rather than evidence of pathway-wide regulation. A brief clarification in the Discussion would help avoid potential overinterpretation.
Overall Recommendation: Minor revisions
The authors have made meaningful revisions that address the majority of the reviewer’s concerns. Provided that the above points are clarified, the manuscript would represent a clearer and more appropriately framed exploratory analysis of PROS1-associated transcriptomic patterns in PDAC. I therefore consider the manuscript suitable for publication pending these minor clarifications.
Author Response
Comment 1:
The 30-month truncation justification is statistically questionable. Long survivors are not outliers in survival analysis, but they are legitimate right-tail observations. Therefore, removing these biases from hazard estimation. Also, survival analysis methods (Kaplan-Meier, Cox models) are specifically designed to handle long survivors. The epidemiological argument (high mortality in PDAC) raised by the authors does not justify removing data. Indeed, the correct approach is (i) keep all data, (ii) censor naturally or (iii) run sensitivity analysis. Hence, despite the small number reported (15–17 patients), which probably does not change the result materially, the lack of sensitivity analysis represents an issue of methodological clarity. While the number of such cases appears small, it would be helpful for the authors to explicitly discuss this limitation in the Methods/Discussion and clarify whether the primary conclusions remain unchanged when all survival data are retained.
Response 1:
Thank you for your valuable insight. We have added the following lines to the Limitations section to add further clarification to the methodology.
“Furthermore, the survival analysis incorporated a 30-month truncation, which may exclude a small number of long-term survivors who represent legitimate right-tail observations in survival distributions. Although the number of long-term survivors was small and unlikely to influence the primary findings, the absence of a formal sensitivity analysis including all survival times represents a limitation and should be considered when interpreting results.”
Comment 2:
The continuous analysis contradicts the authors’ main claim, since they found that PROS1 continuous expression is not associated with survival. This implies that the authors’ dichotomized result may reflect threshold effects or statistical artifacts. Indeed, the authors were correct by reporting it, but the manuscript must clearly state that the prognostic value is not robust across modeling approaches. Please clarify that continuous modeling does not support a linear prognostic effect and discuss this limitation explicitly in the Discussion.
Response 2:
Thank you for indicating the importance of clarifying this difference in the Discussion. We have added the following lines to the Discussion accordingly:
“However, this was not observed when PROS1 was modeled as a continuous variable, as discussed earlier, suggesting that the observed prognostic signal may depend on expression stratification rather than a consistent linear effect. Accordingly, the prognostic relevance of PROS1 should be interpreted in the context of model specification.”
Comment 3:
The manuscript still interprets the attenuation of several EMT-, hypoxia-, and invasion-associated genes as indicative of broader biological programs. However, these conclusions are derived from a curated panel of genes rather than an unbiased transcriptome-wide analysis. While the authors have clarified the exploratory nature of the study, the manuscript should ensure that statements referring to “coordinated suppression of oncogenic pathways” or related pathway-level interpretations remain clearly framed as associations derived from selected markers rather than evidence of pathway-wide regulation. A brief clarification in the Discussion would help avoid potential overinterpretation.
Response 3:
Thank you for noting the importance of clarification in the Discussion to avoid potential overinterpretation. We have added the following lines accordingly:
“Notably, the observed magnitude Δρ of expression of several EMT-, hypoxia-, and invasion-associated genes should be interpreted as associations derived from a curated set of candidate markers rather than evidence of transcriptome-wide pathway suppression. Accordingly, these findings suggest potential relationships between PROS1 expression and selected biological processes but do not establish coordinated regulation of entire oncogenic pathways.”