Spatial Heterogeneity, Stromal Phenotypes, and Therapeutic Vulnerabilities in Colorectal Cancer Peritoneal Metastasis

Abstract

Purpose: Peritoneal metastases (PM) in colorectal cancer portend a poor prognosis. We sought to elucidate molecular features differentiating primary tumors (PT) from PMs and actionable targets facilitating transcoelomic dissemination and progression.

Experimental design: We performed multiomic profiling of 227 samples from 136 patients, including 56 PTs and 120 synchronous PMs comprising 34 matched PT-PM pairs. Whole-exome and bulk RNA sequencing analyses were conducted to identify underlying genomic aberrations and transcriptomic differences between primary and peritoneal lesions. We spatially characterized the microenvironment of tumor-stroma compartments and studied the roles of stromal phenotypes in promulgating tumorigenesis.

Results: Whole-exome sequencing found that genomic alterations and clonality patterns between PTs and PMs remain broadly similar. Transcriptomic profiles, however, suggest a transition as tumors reach the peritoneum toward a more mesenchymal tumor profile and fibrotic tumor microenvironment. Applying spatial profiling, we identify a fibro-collagenous and immune-infiltrated stromal phenotype [stromal cluster (SC) 2] characterized by increased cancer-associated fibroblasts, memory B cells, M2 macrophages, and T-cell exhaustion. These findings were orthogonally validated by multiplex IHC. Patients with SC2 stroma had poorer survival and were characterized by high SERPINE-1 (PAI-1) expression. PMs in patients with SC2 stroma were associated with enriched oncogenic pathways such as TGF-β. PAI-1 inhibition of colorectal cancer PM cell lines with a novel biologic demonstrated reduced IL2-STAT5 and TGF-β pathways and cell death.

Conclusions: Our findings unveil distinctive and actionable molecular signatures, offering deeper insights into the intricate cross-talk between tumor cells and stromal microenvironments enabling PM in colorectal cancer.

Figure 1.

Cohort overview and genomic evolution between primary and peritoneal tumors. A, Graphical summary of samples retrieved in the cohort and assays utilized. B, Sample location and assay type overview of cohort. C, Oncoprint of genomic aberrations identified. Genes shown here were previously reported by Househam and colleagues (22). D, Co-bar plot comparison between PT and PM in colorectal cancer. All comparisons were not statistically significant. E, PT–PM comparisons of TMB, fraction of altered genome, median MAF of dominant clone, whole genome duplication and clonality. MAF, minor allele frequency. [Created in BioRender. Sundar, R. (2025), https://BioRender.com/y53n114.]

Figure 2.

Transcriptomic and microenvironmental evolution in transcoelomic metastasis. A, UMAP of gene expression profiles of primary and peritoneal tumor samples in colorectal cancer (PT, n = 28; PM, n = 44). B, Gene set enrichment analysis pathway changes between primary and peritoneal tumor samples (PT, n = 28; PM, n = 44). Pathways were shown if |NES| was greater than 1.00. Pathways were highlighted if adjusted P values were less than 0.05. C, Immune deconvoluted cell subtypes by xCell and CIBERSORT (PT, n = 28; PM, n = 44). Comparisons were undertaken with an unpaired t test. Immune cells were shown if |t-statistic| was greater than 1.75. Immune cells were highlighted if P values retrieved from an unpaired t test were less than 0.05. D, Differences in EMT score (24), tumor purity [ESTIMATE (25) algorithm], CMS (26, 27) subtype, and TME subtype (PT, n = 28; PM, n = 44; ref. 28). E, Ranked analysis of putative targets in paired PT–PM comparisons. Genes were highlighted if P < 0.05 on a paired t test. F, Violin plots of tumor and (G) immune-related putative therapeutic targets in colorectal cancer PM (unpaired: PT n = 28 and PM n = 44; paired: PT n = 27 and PM n = 36). H, Overview of prognostic significance (overall survival) of identified putative targets. When patients had more than one sample, the mean gene expression was utilized. P values were retrieved from the log-rank test. The cohort (high vs. low) was dichotomized by median gene expression. I, Kaplan–Meier plots of SERPINE-1 expression vs. overall survival in primary and peritoneal tumor samples. J, Scatter plot of EMT score and SERPINE-1 gene expression. The relationship was assessed using Pearson correlation. CMS, consensus molecular subtype; F, fibrotic; FPKM, fragment per kilobase of transcript per million; IE, immuno-enriched; NES, normalized enrichment score; UMAP, Uniform Manifold Approximation and Projection for Dimension Reduction.

Figure 3.

Microenvironmental convergence between PT–PM. A, Illustration of DSP ROI retrieval with NanoString GeoMx. B, UMAP of spatially resolved ROIs in CRC PM. C, Heatmap of SpatialDecon (31) enumerated immune cell types. D, Scatter plot of immune cell type comparisons between tumor–stromal compartments in PT and PM. Immune cell types were highlighted if P < 0.05 from the Dirichlet regression model in either comparison. E, Illustration of COMET Lunaphore mIHC. F, Stacked barchart of cell type proportions stratified by ROI location (stroma vs. TSI vs. tumor) and site (PT vs. PM). G, Immune cell type density curves against distance from TSI stratified by PT vs. PM. H, Immune cell type density curves against distance from TSI stratified by immune cell types. Tumor compartments were taken to be at negative distance whereas stromal compartments were taken to be at a positive distance from the TSI. Smoothed conditional means density curves were retrieved with the geom_smooth() function in ggplot2. I, Average immune cell type–specific nearest neighbor distance per ROI stratified by immune cell types across PT vs. PM. Only ROIs with >100 cells per immune cell type were included. mDC, myeloid dendritic cell; PanCK, pan-cytokeratin; pDC, plasmacytoid dendritic cells; Treg, regulatory T cell; TSI, tumor stromal interface; UMAP, Uniform Manifold Approximation and Projection for Dimension Reduction.

Figure 4.

Spatially revolved stromal phenotypes in CRC PM. A, Consensus clustering of stromal ROIs identifies three distinct stromal clusters. B, Proportion of stromal ROIs per stromal cluster stratified by the site of sample. C, Venn diagram depicting stromal cluster overlaps. D, Kaplan–Meier plot of overall survival stratified by stromal clusters (site agnostic). E, Stromal ROI H&E and DSP slides. F, Heatmap of pathway comparisons across compartments. G, SpatialDecon (31) immune cell type comparisons across compartments. H,SERPINE-1 gene expression across compartments. I, Dotplot of SERPINE-1 gene expression against EMT GSVA enrichment scores. J, Dotplot of PROGENy pathway comparisons between tumor compartments from patients with and without SC2. K, Heatmap of immune cell density from retrieved from mIHC (COMET, Lunaphore) across ROIs identifies two distinct stromal clusters consistent with DSP findings. L, Average immune cell type–specific nearest neighbor distance per stromal ROI stratified by immune cell types across PT vs. PM and stromal cluster. Only ROIs with >100 cells per immune cell type were included. M, Example reconstructed ROIs of SC1-like stroma (CRC_S17-4025_3-2_A11) and SC2-like stroma (CRC_S19-008963_F-5_A09) illustrating divergent spatial distributions of T cells and fibroblasts between stromal phenotypes. CRC, colorectal cancer; GSVA, gene set variation analysis.

Figure 5.

Spatial distribution of SC2 cell types and development of an SC2 signature. A, Overview of the retrieval of SC2 signature score. B, Internal validation of the SC2 score on DSP data. C, Heatmap of SC2 signature genes across compartments. D, Overall survival Kaplan–Meier curves of other colorectal cohorts (Lenos and colleagues, TCGA, and Gallois and colleagues) and SERPINE-1 gene expression stratified by SC2 signature (Q1 vs. Q4). For patients with multiple samples, the mean SC2 signature score was taken. GSVA, gene set variation analysis. [Created in BioRender. Sundar, R. (2025), https://BioRender.com/a03r463.]

Figure 6.

Development of A5, a novel PAI-1–neutralizing antibody. A, Graphical illustration of phage screen and identification of PAI-1–neutralizing antibody. B, Recognition of different conformations of recombinant PAI-1 by A5 in Western blotting. C, A5 neutralized PAI-1 function in inhibiting tPA and uPA. D, Competitive binding of A5 with fluorine-containing tiplaxtinin on stable active PAI-1 assessed by 19F-NMR. E, Inhibition of proliferation of PM cells treated with PAI-1–positive ascites in vitro by 150 µg/mL A5 in biological triplicates. Human IgG served as a negative control. F, STAT3 signaling activation upon the exposure to PM ascites was suppressed by A5 as compared with IgG control. G, GSEA comparisons of HALLMARK pathway changes between CRC cells treated with A5 (n = 2) vs. IgG (n = 2). Pathways with an FDR adjusted P value of less than 0.05 are opacified. CRC, colorectal cancer; NES, normalized enrichment score.