Differential immune gene signatures to inform clinical course of patients with metastatic colorectal cancer.

e15592 Background: Patients with proficient mismatch repair (pMMR) colorectal cancer liver metastases (CRLM) constitute a distinct subgroup with poor survival and attenuated responses to immunotherapy. We hypothesized immune gene expression signatures may predict clinical outcomes. Using data from our single institution experience and from a national database we sought to define relationships between overall survival (OS) and gene expression profiles among patients with metastatic colorectal cancer (CRC). Methods: Using our institutional cohort of CRLM patients, we compared gene expression between patients in the lowest OS quartile (n = 9) and those alive at follow-up (n = 21). 15,605 metastatic pMMR CRC tumors underwent Next-Gen sequencing of DNA and RNA at Caris Life Sciences. CRLM (n = 10,600) were stratified into high- and low-risk subgroups based on OS quartiles to validate findings in the Emory data. Additional metastatic sites were lung (n = 2995) and peritoneal (n = 2010). Deconvolution of bulk tumor RNA expression was used to profile the tumor microenvironment (TME) using quanTIseq. Significance was calculated using X 2 /Fisher’s exact, or Mann-Whitney U with p-values adjusted for multiple comparisons. OS from time of tissue collection to last contact was estimated from insurance claims data using Cox proportional hazards models to calculate hazard ratio (HR) and log-rank tests to calculate p-values. Results: Among our institutional cohort, tumors from low-risk patients demonstrated signatures associated with increased immune enrichment and myeloid predominance. Among pMMR CRLM within the Caris cohort, 3,113 (29%) were considered low-risk and 4973 (47%) were considered high-risk. Of the top 40 differentially expressed genes identified between Emory high- and low-risk cohorts, 21 (52.5%) genes were validated by the Caris cohort. Among pMMR CRLM patients within the low-risk cohort, there were signatures denoting increased immune infiltration with a similar myeloid predominance. Patients with pMMR CRLM tumors had worse OS compared to lung mets (21 vs 35 m; HR 1.7, 95% CI: 1.6-1.8 p < .001) but better OS compared to peritoneal mets (21 vs 18 m; HR 0.8, 95% CI: 0.8-0.9, p < 0.001). Compared to CRLM TMEs, peritoneal mets had increased Tregs, B cells, M1-like, and M2-like macrophages (FC: 1.3-2.3, p < .001) but decreased neutrophils and NK cells (FC: 0.65-0.89, p < .001), while lung mets had increased Tregs, B cells and M2-like macrophages (FC: 1.2-2.5, p < .001). Conclusions: Results from a single institution cohort and a national dataset identify key differences in immune signatures between CRLM that may influence outcomes. These trends persisted based on CRC metastatic site of disease: CRLM are devoid of T cells while peritoneal mets harbor tumors with suppressive TME features. These distinct immune profiles may contribute to worse OS observed among peritoneal and CRLM as compared to lung mets.