Tumor-restrictive type III collagen in the breast cancer microenvironment: prognostic and therapeutic implications

Collagen plays a critical role in regulating breast cancer progression and therapeutic resistance. An improved understanding of both the features and drivers of tumor-permissive and -restrictive collagen matrices are critical to improve prognostication and develop more effective therapeutic strategies. In this study, using a combination of in vitro, in vivo and in silico experiments, we show that type III collagen (Col3) plays a tumor-restrictive role in human breast cancer. We demonstrate that Col3-deficient, human fibroblasts produce tumor-permissive collagen matrices that drive cell proliferation and suppress apoptosis in noninvasive and invasive breast cancer cell lines. In human TNBC biopsy samples, we demonstrate elevated deposition of Col3 relative to type I collagen (Col1) in noninvasive compared to invasive regions. Similarly, in silico analyses of over 1000 breast cancer patient biopsies from The Cancer Genome Atlas BRCA cohort revealed that patients with higher Col3:Col1 bulk tumor expression had improved overall, disease-free and progression-free survival relative to those with higher Col1:Col3 expression. Using an established 3D culture model, we show that Col3 increases spheroid formation and induces formation of lumen-like structures that resemble non-neoplastic mammary acini. Finally, our in vivo study shows co-injection of murine breast cancer cells (4T1) with rhCol3-supplemented hydrogels limits tumor growth and decreases pulmonary metastatic burden compared to controls. Taken together, these data collectively support a tumor-suppressive role for Col3 in human breast cancer and suggest that strategies that increase Col3 may provide a safe and effective modality to limit recurrence in breast cancer patients.


Figure S2: High Col3 staining is associated with non-aligned collagen fibers.
Human biopsy images were stained for Col3 and SHG images were simultaneously obtained. Col3 staining (integrated density) was calculated and plotted with collagen fiber alignment (SHG FFT aspect ratio) for each image obtained (N=23 tumors; 7-10 images per tumor). Linear regression and Pearson coefficient was used to test for correlation.

Col3 expression in human breast cancer
When using our bioinformatics approach to mine for associations between Col3 expression and patient clinical outcome, we aimed to address the issue of varying stromal content between patient tumors -a potential source of bias when analyzing  Fig. S3B). Furthermore, when COL1A1, COL1A2, and COL3A1 gene expression were stratified by PAM50 subtype, they were found to significantly vary (p<0.0001, Kruskal-Wallis H test) in a pattern inverse to that of tumor purity (Supp. Fig. S3C). The collection of these findings illustrates that Col1 and Col3 expression varies between samples by corresponding tumor heterogeneity, suggesting that inconsistent stromal cell content across the BRCA cohort confounds traditional expression analysis of collagen genes.
We hypothesized that mining for significant genomic correlations with Col1 and Col3 gene expression in the BRCA cohort should largely yield associations with stromal cell genes and extracellular matrix components subject to the same purity bias.
In order to validate this, Pearson correlations and corresponding p-values were calculated independently for COL1A1, COL1A2, and COL3A1 versus the other 20,530 unique gene expression estimates from the RNA-seq platform for the TCGA BRCA cohort. Q-values were calculated by adjusting the p-values for false discovery rate using the Benjamini-Hochberg procedure. The resulting q-value matrices for each of COL1A1, COL1A2, and COL3A1 gene expression were ranked to obtain the top 200 gene hits, which were then input to g:Profiler for gene set and pathway enrichment analysis. The resulting number of significantly enriched gene sets in the top correlates for each of COL1A1, COL1A2, and COL3A1 were similar at 336, 286, and 297, with 243 of the gene sets overlapping between all 3 genes. Moreover, the most significantly enriched biological pathways within the top correlates for each of the Col1 and Col3 genes were identical extracellular matrix-based pathways ("extracellular matrix", GO:0031012; "extracellular matrix organization", GO:0030198; "extracellular structure organization", GO:0043062; and "collagen-containing extracellular matrix", GO:0062023), demonstrating that expression analysis of Col1 and Col3 in tumors incidentally selects for similar associations that are subject to tumor purity biases.

Figure S3: Sample heterogeneity skews expression analysis of bulk tumor tissue data. (A) Gene expression of COL1A1, COL1A2, and COL3A1 inversely correlate with the estimated proportion of tumor cells within samples of the TCGA BRCA cohort (p<0.0001, Pearson correlation coefficient t-test), demonstrating that Col1 and Col3 expression within samples vary based on corresponding normal cell and tumor cell fractions. (B)
Tumor purity values were found to significantly vary between PAM50 subtype groupings within the TCGA BRCA cohort (p<0.0001, Kruskal-Wallis H test). (C) COL1A1 and COL1A2 (not shown) and COL3A1 gene expression were found to significantly vary between PAM50 subtype groupings (p<0.0001, Kruskal-Wallis H test) in a pattern inverse to that of purity.
To address the issue of varying stromal content between patient tumors and thereby allow us to evaluate the clinical implications of relative Col3 levels in patient tumors, we sought to analyze a ratio of Col1:Col3 gene expression. After calculating correlations between tumor purity and the ratios of COL1A1:COL3A1 and COL1A2:COL3A1 gene expression, we observed diminished relationships with corresponding p-values that, while still significant, were 15 to 22 orders of magnitude larger than those calculated between tumor purity and individual COL1A1, COL1A2, and COL3A1 gene expression (p<0.0001, Pearson correlation coefficient t-test) (Supp.   Fig. S4B).