Quantification of the Total Stroma Percentage in BC. (A) Flowchart illustrating the methodology used to quantify the total stroma percentage in FFPE breast tumor tissue samples. The samples were stained with H&E and analyzed via QuPath software to identify the stroma and tumor areas, with validation performed by a pathologist. (B) Representative images of H&E staining (top) and corresponding QuPath predictions (bottom) in samples with low, medium, and high total stroma percentages. (C) Correlation between total stroma quantification by QuPath and pathologist validation (n = 17).

Maximally selected rank statistical analysis identified an optimal cutoff of 71.8% total stromal content for overall survival (OS) stratification (Fig. 2A). In the study cohort, the total stromal content ranged from 23.9–96.7%, with a mean of 67.8% and a standard deviation of 16.5%. On the basis of this threshold, 102 patients were classified as having low total stroma, and 80 patients were classified as having high total stroma, corresponding to 56.0% and 44.0% of the cohort, respectively.

Associations of total stromal content with clinical outcomes and molecular subtypes in patients with breast cancer. (A–B) Maximally selected rank statistics and Kaplan–Meier survival analysis for overall survival (OS) in patients with breast cancer stratified by total stromal content (n = 182). (A) Determination of the optimal cutoff value (71.8%) via maximally selected rank statistics. (B) Kaplan–Meier curve showing significantly shorter OS in patients with low stromal content (HR 3.49; 95% CI: 1.84–6.62; p < 0.001). (C–D) Analysis of metastasis-free survival (MFS) in a subset of patients (n = 134). (C) The optimal cutoff value for MFS was 80.8%. (D) Kaplan–Meier curve indicating significantly worse MFS in the Stroma-Low group (HR 4.11; 95% CI: 1.62–10.4; p < 0.001). (E) Representative H&E-stained images and corresponding QuPath-based segmentation of stromal and tumor areas in tumors classified as ER+/PR+, HER2+, or triple-negative breast cancer (TNBC). Stromal areas are highlighted in green and tumor regions in red. (F) Percentage of total stromal content across molecular subtypes (ER+/PR+, HER2+, and TNBC) quantified from H&E-stained sections via QuPath. No statistically significant differences were observed among the subtypes (ns, Kruskal–Wallis test).

Given the lack of significant associations in multivariate models and the observation that total stromal content does not vary across molecular subtypes, we investigated the biological nature of the stromal compartment itself. While total stroma provides a quantitative measure of tissue microarchitecture, it may not fully capture the functional activity of stromal cells involved in tumor progression. To address this, we evaluated the expression of genes commonly expressed by CAF and other stromal components. Specifically, we assessed the expression of VIM, PDGFRA, PDGFRB, and ACTA2, which are genes predominantly expressed by stromal cells, particularly CAF, and are involved in mesenchymal activation and extracellular matrix remodeling, via RNA sequencing of a representative subset of the same FFPE samples used for stromal quantification.

Among these genes, PDGFRA and PDGFRB were significantly upregulated in tumors classified as total stroma-high (p < 0.05) (Fig. 3A–D). Although VIM and ACTA2 were more highly expressed in the total stroma-high group, these differences were not statistically significant. Moreover, correlation analysis revealed that gene expression levels were strongly positively correlated with the stromal percentage in the total stroma-low group (Fig. 3E–H), suggesting a complex relationship between stromal abundance and transcriptional activity. In addition, MCP-counter analysis revealed higher fibroblast scores in tumors with high total stromal content (Fig. 3I), reflecting increased infiltration of stromal cell populations in these samples.

Although low total stromal content was significantly associated with poorer OS and MFS, these results indicate that the total stromal percentage does not necessarily reflect the abundance or activation state of CAF. Given that CAF play a pivotal role in tumor progression, we hypothesized that the total stromal content may fail to capture the functionally active stromal compartment. Therefore, we quantified the reactive stroma, defined as the fraction of stroma characterized by a densely remodeled ECM enriched in activated CAF with a myofibroblastic phenotype and high secretory activity of protumorigenic factors, including collagen types I and II. This component represents the biologically active portion of the tumor stroma and is directly involved in invasion, angiogenesis, immunosuppression, and therapeutic resistance. Its quantification enabled a more specific assessment of its prognostic relevance.

Expression of CAF-Associated Genes and the Abundance of Tumor Microenvironment Cell Populations According to the Total Stromal Content in Breast Cancer. (A–D) Box plots showing the variance-stabilized expression levels of the CAF markers VIM (A), PDGFRA (B), PDGFRB (C), and ACTA2 (D) in tumors classified as Stroma-High and Stroma-Low based on the total stromal percentage. (E–H) Scatter plots illustrating the correlation between VST-normalized gene expression and total stromal area across samples. Pearson’s r and p-values are shown for each stromal group. (I) MCP-counter analysis showing the estimated abundance of tumor microenvironment cell populations in patients with high and low total stromal content.

Reactive Stroma and Its Impact on Breast Cancer Survival

To quantify the reactive stroma, we analyzed 146 Masson’s trichrome–stained samples, measuring the area corresponding to collagen I and III relative to the total stromal area previously identified in H&E images. This quantification was performed via QuPath with supervised pixel classification, allowing the identification of reactive stroma, tumor, and other tissue structures (Fig. 4A-B). The method was validated in a subset of 14 cases by comparison with manual annotations from expert pathologists, which revealed a significant correlation (r = 0.577; p < 0.05) (Fig. 4C).

Quantification of Reactive Stroma in BC. (A) Flowchart illustrating the methodology used to quantify the percentage of reactive stroma relative to total stroma in FFPE breast tumor tissue samples. The samples were stained with Masson’s trichrome and analyzed via QuPath software to identify the tumor and reactive stroma areas, followed by clinical validation by a pathologist. (B) Representative images of Masson’s trichrome staining (top) and corresponding QuPath predictions (bottom) in samples with low, medium, and high percentages of reactive stroma. (C) Correlation between reactive stroma quantification by QuPath and validation by pathologists (n = 14).

Maximally selected rank statistical analysis identified optimal cutoffs of 42.5% and 53.2% reactive stromal content for OS and MFS, respectively (Figs. 5A and 5C). In the OS cohort (n = 146), the percentage of reactive stromal content ranged from 3.7–100%, with a mean of 55.2% and a standard deviation of 21.9%. Based on the 42.5% cutoff, 42 patients were classified as reactive stroma-low, and 104 patients as reactive stroma-high. For the MFS analysis, a subset of 85 patients from the same cohort for whom time to metastatic progression was available was included. In this subgroup, the reactive stromal content ranged from 14.5–100%, with a mean of 48.7% and a standard deviation of 21.0%. According to the 53.2% threshold, 58 patients were assigned to the Reactive Stroma-Low group, and 27 were assigned to the Reactive Stroma-High group. This stratification enabled a more specific evaluation of the prognostic impact of the biologically active stromal compartment, offering additional insights beyond total stromal quantification.

Associations of the Reactive Stromal Content with Clinical Outcomes, Molecular Subtypes, and Multivariate Prognostic Value in Breast Cancer. (A–B) Maximally selected rank statistics and Kaplan–Meier analysis for overall survival (OS) based on the percentage of reactive stroma (n = 146). (A) Optimal cutoff point identified at 42.5% reactive stroma. (B) Kaplan–Meier curve showing a trend toward worse OS in patients with highly reactive stroma (HR = 2.51; 95% CI: 0.97–6.49; p = 0.058). (C–D) Analysis of metastasis-free survival (MFS) in a subset of patients (n = 85). (C) Optimal cutoff point determined at 53.2% reactive stroma. (D) Kaplan–Meier curve demonstrating a significantly lower MFS in the reactive stroma-high group (HR = 3.75; 95% CI: 1.98–7.09; p < 0.01). (E) Representative Masson’s trichrome-stained images and corresponding QuPath segmentation of reactive stroma in ER+/PR+, HER2+, and triple-negative breast cancer (TNBC) tumors. (F) Quantitative comparison of the reactive stromal content across molecular subtypes revealed no significant differences, suggesting that the reactive stroma is a subtype-independent feature. (G–H) Forest plots of multivariate Cox regression analyses for OS (G) and MFS (H).

Transcriptomic Profile Associated with the Reactive Stromal Content

Given the strong prognostic value of histologically defined reactive stroma, we aimed to investigate whether this phenotype is associated with specific biological programs. To this end, RNA sequencing (RNA-Seq) was performed on 62 tumor samples with available FFPE material to assess their transcriptomic profile. From these, a representative subset of 12 tumors—previously analyzed histologically and selected on the basis of high or low reactive stromal content—was used for differential gene expression (DEG) analysis, followed by gene set enrichment analysis (GSEA) and functional annotation via the KEGG and Reactome databases, with the aim of identifying signaling pathways differentially regulated between tumors with high and low reactive stromal content.

To visualize overall transcriptomic differences, a heatmap was generated using representative differentially expressed genes between tumors stratified by reactive stromal content (Fig. 6A). A clear separation was observed between the high and low groups, indicating that distinct transcriptomic profiles are associated with the stromal phenotype. The volcano plot highlights differentially expressed genes between the two groups, with several showing high magnitude changes and statistical significance (adjusted p-value < 0.05; log2-fold change > |1|) (Fig. 6B). Among the genes upregulated in tumors with high reactive stromal content were FN1, OLR1, EDN2, and MSR1, which are involved in key processes such as extracellular matrix (ECM) remodeling, inflammatory signaling, and modulation of the tumor microenvironment—all mechanisms associated with tumor progression and metastasis.

Pathway enrichment analysis via the Reactome database revealed an overrepresentation of ECM-related and immune-associated processes, including collagen degradation, ECM activation, and ECM organization, in tumors with high reactive stromal content (Fig. 6C). Consistently, functional analysis via the Kyoto Encyclopedia of Genes and Genomes (KEGG) database confirmed these findings, revealing significant enrichment of ECM-related and immune signaling pathways, such as ECM-receptor interactions, cell adhesion molecules, NF-kappa B signaling, and B cell receptor signaling, in the highly reactive stroma group (Fig. 6D).

To validate these findings at the gene set level, we performed GSEA focused on ECM-related processes. Three key gene sets—extracellular matrix disassembly (normalized enrichment score, NES = 1.68), extracellular structure organization (NES = 1.74), and extracellular matrix organization (NES = 1.75)—were significantly positively enriched in tumors with high reactive stromal content (Fig. 6E–G). Additionally, positive enrichment was observed in three gene sets associated with TGF-β signaling, including the regulation of the TGF-β signaling pathway (NES = 1.66), the cellular response to the TGF-β stimulus (NES = 1.60), and the transforming growth factor beta receptor superfamily signaling pathway (NES = 1.49) (Figs. 6H–J), supporting the activation of profibrotic and tissue remodeling programs in these patients.

Gene expression and pathway enrichment analyses associated with reactive stromal content. (A) Heatmap of DEGs between tumors with high and low reactive stromal content. The samples are clustered according to the stromal phenotype. (B) Volcano plot from DEG analysis. Genes with statistically significant differences (adjusted p-value < 0.05) and | log2(fold change) | > 1 are highlighted. (C) Reactome pathway enrichment analysis revealed that tumors with high reactive stromal content were enriched in pathways related to Fc gamma receptor-mediated phagocytosis, collagen degradation, platelet activation, and extracellular matrix organization. (D) KEGG pathway enrichment analysis showing immune-related and extracellular matrix-associated pathways enriched in tumors with high reactive stromal content. (E–G) Gene set enrichment analysis (GSEA) plots for the terms extracellular matrix disassembly (E), extracellular matrix organization (F), and extracellular structure organization (G), indicating significant upregulation of these processes in tumors with highly reactive stroma. (H–J) GSEA plots for the terms regulation of the transforming growth factor beta receptor signaling pathway (H), cellular response to transforming growth factor beta stimulus (I), and the transforming growth factor beta receptor superfamily signaling pathway (J), suggesting transcriptional upregulation of TGF-β–associated programs in tumors with high reactive stromal content.

DISCUSSION

The TME has emerged as a key driver of cancer progression [32, 33]. In this context, our study demonstrated that the proportion of reactive stroma represents an independent prognostic variable in patients with breast cancer. Tumors enriched in reactive stroma exhibited significantly shorter metastasis-free survival (MFS) than those enriched with lower stromal content, a finding that remained robust after adjustment for age and hormone receptor status. These results indicate that the reactive stroma is a biologically relevant feature of the TME, independent of conventional clinicopathological factors.

Reactive stroma, originally described in prostate cancer [20], is characterized by the activation of fibroblasts and the deposition of ECM proteins such as type I collagen, tenascin, and fibronectin. Indeed, abundant ECM accumulation not only provides structural support but also actively promotes tumor progression. Excessive ECM accumulation has been associated with poor prognosis across multiple solid tumors [34–36], emphasizing the functional relevance of stromal remodeling in cancer biology.

However, desmoplasia, or reactive stroma, is often acellular or composed predominantly of the extracellular matrix [37, 38]. This characteristic may limit the detection of specific cellular transcripts in bulk RNA-Seq analyses. Therefore, transcriptional profiles derived from tumors with extensive acellular stroma should be interpreted cautiously, particularly when assessing cell-specific gene expression.

In our study, samples were objectively quantified via Masson's trichrome staining, which enabling the precise identification of collagen-rich tissue components. By applying digital image analysis, we achieved accurate segmentation of the fibrous ECM fraction, allowing standardized assessment of the reactive stroma content across tumor samples. This approach provides a robust histological basis for linking stromal content to clinical outcomes and molecular alterations.

Several studies have shown that collagen-enriched ECM promotes tumor progression and metastatic dissemination [39, 40]. Although previous studies have suggested that luminal tumors (ER+/HER2−) present greater stromal content than triple-negative breast cancer (TNBC) [41], our analysis did not reveal significant differences in total or reactive stroma among molecular subtypes. This lack of association underscores the possibility that reactive stroma operates as a transversal feature across breast cancer subtypes, reinforcing its prognostic significance.

Further examination of the reactive stromal compartment revealed significant associations with MFS. These findings suggest that the reactive stroma is not merely a subtype-specific phenomenon but represents a functionally active component of the TME broadly implicated in metastatic risk.

The biological basis for the prognostic role of the reactive stroma is supported by the functional properties of its major components, particularly type I collagen. In addition to providing mechanical support, accumulated collagen promotes cell migration through the activation of integrins and receptors, such as DDR1, leading to cytoskeletal reorganization and acquisition of an invasive phenotype [42–45]. Increased collagen density induces tissue stiffness, which modulates cellular behavior through mechanotransduction pathways, notably via focal adhesion kinase (FAK) activation [46, 47]. This process triggers downstream cascades, including ERK activation, Rho-GTPase signaling, and integrin clustering, promoting cellular contractility, invasion, and morphological plasticity [46, 48]. These dynamic interactions between tumor cells and the ECM contribute to tissue remodeling, altered cell–cell and cell–ECM adhesion, and ultimately drive tumor cells toward an invasive phenotype capable of penetrating the matrix and initiating the metastatic cascade [46–49]. Our findings reinforce the concept that, in addition to its architectural role, collagen content constitutes a biologically active component of the tumor stroma that is associated with poor prognosis.

Consistent with observations from other solid tumors, our results highlight the clinical relevance of stromal features. In prostate cancer, Ruder et al. developed a quantitative reactive stroma (qRS) biomarker independently associated with increased disease-specific mortality and biochemical recurrence [50]. In breast cancer, Sharma et al. validated Stratipath Breast, which has strong prognostic performance in ER+/HER2 − patients (HR = 2.76; p < 0.001) [51]. While stratification focuses on predicting general disease progression, our study specifically addresses metastasis, demonstrating that reactive stroma ≥ 53.2% was associated with a significantly shorter MFS (HR = 3.75; p < 0.01), and reactive stroma ≥ 42.5% was associated with a trend toward worse OS (HR = 2.51; p = 0.058).

Importantly, assessing reactive stroma with routine histological methods, such as Masson's trichrome staining combined with digital image analysis, provides a simple, reproducible, and clinically feasible strategy. Unlike complex molecular assays, incorporating reactive stroma evaluation into standard pathology workflows entails minimal modification while offering valuable prognostic information in breast cancer.

A key strength of our study is the integration of digital pathology with transcriptomic profiling of FFPE tumor tissue. Although transcriptomic data validated the activation of fibroblast-related and ECM remodeling pathways, they remain complementary to histological quantification.

By applying RNA-Seq to a subset of histologically characterized tumors, we identified a transcriptional signature enriched in biological programs associated with fibroblast activation and ECM remodeling. Notably, tumors with high reactive stroma content exhibited significant overexpression of genes such as FN1, OLR1, EDN2, and MSR1. FN1, which encodes fibronectin, is a glycoprotein expressed in the ECM as a dimer or polymer [52, 53]. This protein plays a pivotal role in physiological processes, such as wound healing, and pathological processes, such as tumor progression, facilitating cell adhesion, migration, and invasion [54, 55]. In line with our histological findings, fibronectin is deposited as fibrillar complexes that reorganize the ECM, generating a dense and stiff matrix enriched in type I collagen. This structural reorganization promotes mechanotransduction, activates pro-oncogenic signaling pathways such as FAK and YAP, and enhances cell migration, immune evasion, and therapy resistance [56, 57].

The correlation between FN1 overexpression and high reactive stroma content supports the concept that the ECM is not merely a static scaffold but also an active participant in tumor progression. Specifically, we demonstrated that tumors with abundant type I and III collagen synthesized by CAF presented a greater risk of metastasis, underscoring the functional importance of the stromal compartment in disease dissemination.

Consistently, we also observed the overexpression of OLR1, a receptor for oxidized lipoproteins (oxLDLs), which has previously been implicated in promoting a protumoral CAF phenotype. OLR1 activation stimulates fibroblast activation, collagen synthesis, and ECM remodeling, creating a dense and mechanically active microenvironment that facilitates tumor cell migration and immune evasion [58]. Additionally, OLR1 expression has been associated with increased infiltration of M2 macrophages and upregulation of immune checkpoint molecules such as PD-L1 [59], further reinforcing its role in establishing an immunosuppressive TME.

Similarly, the overexpression of EDN2, a member of the endothelin family and an adverse prognostic marker in breast cancer, was observed. EDN2 promotes tumor microenvironment remodeling through proliferation, migration, and activation of pro-oncogenic pathways such as the STAT3 pathway [60].

In addition to individual genes, functional pathway enrichment analyses provided broader insights into stromal activity. Reactome and KEGG analyses revealed significant enrichment of ECM-related pathways—including those related to collagen degradation, ECM organization, and ECM-receptor interactions—emphasizing the functional involvement of the stroma in poor-prognosis tumors. Notably, the TGF-β signaling pathway was also enriched in tumors with highly reactive stroma. TGF-β is a well-established driver of fibroblast-to-myofibroblast transition and ECM deposition, and its persistent activation within the TME has been linked to epithelial‒mesenchymal transition (EMT), immunosuppression, and increased metastatic potential [61–65].

Conversely, we observed negative enrichment of immune-related gene sets, including those related to T-cell receptor signaling, T-cell proliferation, and leukocyte adhesion [66, 67]. These findings suggest that CAF-mediated ECM remodeling not only promotes invasion but also restricts immune cell infiltration, fostering an immunosuppressive and immune-excluded microenvironment that facilitates metastatic dissemination.

Future studies could also leverage digital pathology and machine learning algorithms to incorporate additional stromal parameters such as fiber density, length, linearity, and thickness. The quantitative assessment of these features could refine risk stratification and enhance the prognostic utility of the reactive stroma.

In summary, our transcriptomic results reinforce the concept that the reactive stroma is a biologically active compartment fostering an invasive and immune-suppressive tumor phenotype. The integration of molecular findings with histological quantification provides a mechanistic rationale for its prognostic value and highlights its potential as a clinical biomarker in breast cancer.

While our study benefits from a robust cohort (n = 182) and a median follow-up of 60 months, future validation in independent, multi-institutional series will further strengthen its generalizability. Our use of Masson’s trichrome staining with digital segmentation offers precise, reproducible measurements of the ECM and could be complemented by multiplex immunohistochemistry or spatial transcriptomics to refine the cellular resolution within the stroma. Finally, as treatment paradigms have evolved over the 2006–2020 period, ongoing investigations incorporating contemporary therapeutic regimens will ensure continued relevance to current clinical practice.

Taken together, our findings provide a comprehensive characterization of the reactive stromal compartment in breast cancer, highlighting its dual role as a structural and signaling component that promotes metastasis. By combining histological and transcriptomic evidence, we establish the reactive stroma as an integral feature of the tumor microenvironment with both prognostic and biological significance. These insights offer a compelling rationale for incorporating stromal assessment into routine diagnostic workflows and for the development of stromal-targeted therapies. As the field moves toward more personalized oncology, evaluating stromal dynamics may refine risk stratification and guide treatment decisions beyond conventional tumor cell–centric models.

CONCLUSIONS

This study advances our understanding of tumor–stroma interactions in breast cancer by establishing the reactive stroma as a clinically relevant prognostic marker of metastasis, independent of molecular subtype. By integrating pathology and transcriptomics from FFPE tissue, we propose a clinically feasible strategy to assess stromal activity, complementing existing molecular classifiers. Future research should aim to validate these findings prospectively and explore therapeutic opportunities targeting the reactive stromal compartment.

Declarations

Ethics approval and consent to participate: The study was approved by the following ethics committees: the Research Ethics Committee of Fundación Arturo López Pérez (FALP; breast cancer protocol ID: 2022-0232-RES-CRC-MUL), Instituto Nacional del Cáncer (INC; project number CRI20050, under the framework of UC project 201016011, as the INC does not have its own ethics committee), and Red Salud UC-CHRISTUS (project ID: 201016011). A waiver of informed consent was granted by all committees, as the study met the criteria for minimal risk, impracticability, and protection of participants’ rights and welfare. All procedures complied with relevant institutional guidelines and regulations.

Consent for publication: Not applicable.

Availability of data and materials:

The scanned histological images (H&E and Masson's trichrome) are available from the corresponding author upon reasonable request due to file size limitations and institutional storage restrictions. The RNA-Seq data will be made publicly available in the Gene Expression Omnibus (GEO) repository upon acceptance of the manuscript. The accession numbers and corresponding links are included in the final published version.