Identifying distinct prognostic and predictive contributions of tumor epithelium versus tumor microenvironment in colorectal cancer

Accumulating evidence has suggested that cancer progression and therapeutic response depend on both tumor epithelium (EPI) and tumor microenvironment (TME). However, the dependency of clinical outcomes on the tumor EPI vs. the TME has neither been clearly defined nor quantified.

We classified 2373 colorectal cancer (CRC) tumors into the consensus molecular subtypes (CMS1-4) and generated the 10-gene TME^S and the 10-gene EPI^S signatures as the serendipitous derivatives of the most (positively vs. negatively) correlated genes of a highly-prognostic, ~ 500-gene signature we previously identified. Distinct TME vs. EPI cellular features of the signature genes were identified by CIBERSORT deconvolution and validated by scRNASEQ in an independent public dataset.

The TME^S signature was strongly associated with the immune/stromal TME-rich CMS1/CMS4 subtypes that portended worse survival, whereas the EPI^S signature was predominantly related to the TME-poor, epithelial CMS2/CMS3 classes that portended better survival. Multivariable Cox regression analysis against 29 TME-related signatures revealed that the TME^S signature was the most strikingly impacted by the "Cancer-associated fibroblasts" signature (HR: 10.87 vs. 0.13, both P < 0.0001). Moreover, the TME^S score was strongly correlated with EMT, SRC activation and MEK inhibitor resistance in 2373 CRC tumors (Spearman r = 0.727, 0.802, 0.824, respectively), which was validated in two independent CRC datasets (n = 626 and n = 566). By contrast, the EPI^S score was the dominant force in associating with longer progression free survival in cetuximab-treated metastatic CRC patients derived from two independent clinical trials (Logrank trend P = 0.0005/n = 80; P = 0.0013/n = 44). This finding was further validated in a large real-world clinico-genomics dataset with EGFR inhibitor therapy, which demonstrated that higher EPI^S scores were associated with increased overall survival (EGFRi, Logrank trend P < 0.0001/n = 2343) and time on treatment (cetuximab, P = 0.003/n = 953; panitumumab, P < 0.0001/n = 1307).

Here we identified a pair of new, distinct 10-gene signatures (the EPI^S vs. the TME^S) capable of distinguishing the cellular contribution of the tumor EPI vs. the TME in determining CRC prognosis and therapeutic outcomes. With targeted approaches emerging to address both tumor epithelial cells and the TME, the EPI^S vs. TME^S signature scores may have a novel biomarker role to permit optimization of CRC therapy by identifying sensitive vs. resistant subpopulations.