Review Reports

Comments 1: Typographical errors (e.g., “FAP” instead of “FAPα”)

Response 1: Thank you for pointing this out. We agree with this comment. Therefore, we have corrected the nomenclature of fibroblast activation protein by explicitly defining it at its first mention as fibroblast activation protein alpha (FAP-α). This correction has been implemented in the Introduction (page 2, paragraph 4, lines 73–74). In addition, the same terminology has been applied consistently in the subsequent reference to immunohistochemical markers (Introduction, paragraph 6, line 89).

Updated text in the manuscript:

Reactive stroma is histologically defined as a fibrotic and proinflammatory micro-environment that emerges during early stages of carcinogenesis. It is characterized by the activation of fibroblasts into a myofibroblast-like phenotype, with increased synthesis of ECM components such as type I collagen, tenascin-C and fibronectin, as well as the ex-pression of mesenchymal markers like vimentin and fibroblast activation protein alpha (FAP-α) [20,21].

However, its clinical quantification has been limited by the lack of objective and standardized methodologies capable of distinguishing it from the remaining stromal content. While some studies have employed immunohistochemical markers such as FAP-α or α-SMA to detect activated [26,28]

Comments 2: Low quality of figures

Response 2: We appreciate this important observation. The figures included in the revised manuscript have been updated and are embedded in the Word file for review purposes. Final high-resolution versions of all figures will be resubmitted separately as individual files, in full compliance with the journal’s publication requirements.

Comments 3: Limitation regarding the use of bulk transcriptomic material instead of reactive stroma–specific gene expression methods.

Response 3: We thank the reviewer for this valuable methodological consideration. We agree that spatially resolved approaches, such as RNA-FISH or related techniques, would allow a more direct assessment of gene expression specifically within the reactive stromal compartment. In the present study, transcriptomic analyses were performed on bulk FFPE tissue, in line with material availability and with an emphasis on clinical applicability, reflecting routine diagnostic conditions. The transcriptomic results were interpreted in the context of spatially quantified reactive stroma and are supported by previously described gene signatures associated with extracellular matrix remodeling and tumor microenvironment modulation. Future studies will incorporate spatially resolved methodologies, including spatial transcriptomics or RNA-FISH–based validation, to further characterize gene expression specifically within reactive stromal regions and to strengthen the mechanistic validation of these findings.

Comments 1: In this study, H&E staining was used to quantify total stroma and Masson’s trichrome staining to quantify reactive stroma, combined with QuPath analysis. How does this design ensure correlation and accuracy between both staining methods? Were cross-validation or control experiments implemented to exclude staining-related bias?

Response 1: We thank the reviewer for this methodological question. Our approach was designed to ensure robustness and comparability through (i) supervised automated segmentation and (ii) proportion-based normalization.

First, quantification was performed in QuPath (v0.5.0) using a supervised pixel-classification (Random Trees) workflow trained from manual annotations to discriminate relevant tissue compartments (tumor, stroma/reactive stroma, and other structures). This pipeline enables objective and reproducible segmentation, reducing observer-related variability.

Second, and critically, reactive stroma was expressed as a percentage of the total stroma present in the same biopsy. Specifically, we quantified total stromal area in H&E and then quantified the collagen-rich ECM fraction (reactive stroma) in Masson’s trichrome, reporting it relative to the total stromal area. This normalization was implemented to ensure that reactive stroma quantification is not dependent on the amount or type of tissue obtained in the biopsy, but rather standardized to the stromal content available within each sample.

Finally, reproducibility of the automated measurements was assessed through expert pathologist validation in a subset of cases (H&E n=17; Masson n=14), showing high concordance for total stroma (r = 0.902; p < 0.001) and significant concordance for reactive stroma (r = 0.577; p < 0.05), supporting the accuracy of the method.

Comments 2: Total stroma was associated with overall survival (OS) and metastasis-free survival (MFS) in univariate analysis but did not retain independent prognostic value in multivariate analysis, whereas reactive stroma showed significant independent prognostic value. How can this biological difference be explained from the perspective of functional heterogeneity in the tumor microenvironment?

Response 2: We appreciate this insightful question, as it helps clarify the conceptual contribution of our work. Total stroma, quantified on H&E, primarily captures an architectural metric (the proportion of non-tumoral tissue) but aggregates biologically heterogeneous components (fibroblasts across activation states, vasculature, immune infiltrates, and potentially acellular/mucoid stromal areas). Consequently, its prognostic signal may be confounded after adjustment for clinicopathologic covariates (age, pT, pN, ER/PR/HER2 subtype, grade). This is consistent with our findings: total stroma showed strong univariate associations (OS HR = 3.49; p < 0.001; MFS HR = 4.11; p < 0.001) but lost independent significance for OS (p = 0.209) and remained only a trend for MFS (p = 0.073) in multivariate models.

By contrast, reactive stroma, quantified on Masson’s trichrome, operationalizes a functional stromal phenotype, focusing on the collagen-rich, densely remodeled ECM fraction, typically linked to activated CAF programs and pro-metastatic biology (stiffness, mechanotransduction, adhesion/migration, angiogenesis, and immune modulation). Accordingly, high reactive stroma retained independent prognostic value for metastasis (MFS: HR = 3.76; 95% CI 1.91–7.39; p < 0.001) after adjustment for age, ER, and PR (while OS showed a non-significant trend, p = 0.069).

Importantly, transcriptomic integration supports this functional distinction: reactive stroma–high tumors showed enrichment of ECM remodeling pathways (Reactome/KEGG; GSEA ECM organization/disassembly; NES ~1.7) and TGF-β signaling (NES ~1.5–1.7), together with negative enrichment of T-cell activation–related programs, providing a mechanistic framework consistent with increased metastatic risk.

Comments 3: Based on the association of reactive stroma with TGF-β signaling and extracellular matrix remodeling, what impact would be expected from inhibiting TGF-β signaling or targeting ECM-related genes (such as FN1 and OLR1) on the metastatic potential of breast cancer models? Is there preliminary evidence supporting this expectation?

Response 3: We appreciate this highly relevant translational question. There is direct preclinical evidence supporting the expectation that inhibition of TGF-β signaling can significantly reduce invasion and metastasis in breast cancer. Notably, Fang et al. reported in Journal of the National Cancer Institute in 2013 that YR-290, a selective inhibitor of the TGF-β type I receptor (TGFβR1), inhibits TGF-β–induced migration and invasion in vitro and almost completely block metastasis in murine models, while also prolonging animal survival. This study establishes a direct functional link between TGF-β signaling and metastatic potential.

These findings are consistent with the broader biological framework describing the role of TGF-β in advanced tumor progression. In a comprehensive review published in Journal of Mammary Gland Biology and Neoplasia in 2011, Drabsch and ten Dijke described how TGF-β signaling can switch from an early tumor-suppressive role to a pro-metastatic role in advanced breast cancer, promoting invasion and metastasis to lung and bone, and discussed the therapeutic rationale for targeting this pathway in preclinical settings.

Regarding extracellular matrix remodeling, evidence indicates that fibrillar networks composed of type I collagen and fibronectin (FN1) are associated with key metastatic behaviors. Soikkeli et al. demonstrated in American Journal of Pathology in 2010 that metastatic outgrowth is accompanied by coordinated upregulation of type I collagen and FN1 and their assembly into fibrillar networks that regulate cell adhesion, migration, and growth, providing mechanistic support for the role of ECM-rich stromal states in tumor dissemination. Consistently, Graf et al. reported in FASEB Journal in 2021 that ECM proteins, including type I collagen and fibronectin, can stimulate cancer cell migration, reinforcing the functional relevance of these ECM components in invasive phenotypes.

With respect to OLR1 (LOX-1), available evidence in breast cancer supports its involvement in metastasis-relevant processes, although it does not directly establish a CAF-specific role in this disease. Liang et al. reported in Breast Cancer Research in 2007 that endothelial LOX-1 induction promotes adhesion and trans-endothelial migration of breast cancer cells, a critical step in metastatic dissemination. In addition, Pucci et al. showed in Cell Death Discovery in 2019 that LOX-1 is overexpressed in a substantial fraction of human breast cancer samples and is associated with features of tumor aggressiveness. Taken together, these published data provide preliminary experimental evidence supporting the hypothesis that targeting TGF-β signaling and ECM remodeling axes, including FN1- and collagen-rich fibrillar networks, may reduce metastatic potential in breast cancer models characterized by high reactive stroma, providing a strong rationale for future intervention studies.

Comments 4: RNA sequencing was performed in only 12 samples. Could this small sample size affect the reliability of differential gene and pathway enrichment results? What experimental designs could further validate the generalizability of these transcriptomic features?

Response 4: We agree that the RNA-seq sample size (n = 12) is a limitation for broad transcriptomic generalization. In this study, RNA-Seq was designed as an exploratory, hypothesis-generating component, using histologically characterized tumors from the same FFPE series to provide mechanistic context complementing the robust clinical evidence (n = 182; median follow-up 60 months) and survival analyses.

Despite the limited n, we observed consistent separation between reactive stroma–high vs –low tumors (heatmap), identification of DEGs under stringent criteria (adj p ≤ 0.05; |log2FC| > 1), and convergence across multiple analytical layers (DEG + Reactome + KEGG + GSEA), highlighting ECM/TGF-β programs together with negative enrichment of T-cell activation signatures. Nevertheless, we do not present these data as a definitive molecular classifier.

Future validation will include: (i) increasing transcriptomic sample size and validating in independent cohorts; (ii) implementing spatially resolved approaches (spatial transcriptomics or RNA-FISH) to capture compartment-specific signals within reactive stroma; and (iii) exploring single-cell RNA-seq when feasible to resolve stromal cellular heterogeneity.

Comments 5: The study proposes that reactive stroma quantification could be integrated into routine pathology workflows. Considering inter-hospital variability in slide preparation (e.g., section thickness and staining duration), how can stability and standardization of this method be ensured? Is large-scale multicenter validation necessary to define a unified cutoff value (53.2%)?

Response 5: We thank the reviewer for this clinically relevant implementation point. Our approach relies on algorithm-driven, standardized segmentation (QuPath) and relative area-based quantification, which reduces sensitivity to moderate variations in staining intensity. The workflow also includes supervised classifier training and tissue-class definitions, enabling local recalibration if pre-analytical conditions differ across laboratories.

We fully agree that broad clinical adoption requires multicenter validation to assess inter-laboratory robustness (slide preparation, staining protocols, scanners) and to determine whether a unified cutoff such as 53.2% remains stable or should be refined using harmonization strategies (e.g., reference sets, digital QC, or batch normalization). We consider this validation a critical next step prior to widespread clinical implementation.