MSI colon adenocarcinoma transcriptomic subtypes are biomarkers of response to immunotherapy

Background: Microsatellite instability-high (MSI) is a predictive biomarker for response to immune checkpoint inhibitors (ICI) in metastatic colon adenocarcinoma (COAD), but not all patients respond to ICI. Thus, there is a clinical need to develop novel biomarkers to more accurately predict ICI response in metastatic MSI-COAD. Methods: The transcriptomic subtyping cohort consisted of 793 primary tumor MSI-COAD samples with DNA & RNA-seq (Tempus xT/xR) from Tempus' real-world database. The ICI cohort consisted of 59 metastatic MSI-COAD patients who had a primary tumor sample profiled via Tempus xT/xR. We performed transcriptomic subtyping via non-negative matrix factorization (NMF). Survival analyses were performed via risk set adjusted Kaplan-Meier and CoxPH models with right censor cutoff of 48 months post ICI. Consensus molecular subtype (CMS) calls were made using Tempus CMS classifier, immune deconvolution was performed using xCell, and GSEA using fgsea. A FDR < .05 was considered significant. Results: Clustering of primary MSI-COAD tumor transcriptomes revealed two subtypes that we named “Immune/Stromal” (C1, n=390) and “Goblet/Enterocyte” (C2, n=403). C1 tumors displayed significantly higher proportion of CD8+ T cells, cytolytic scores, and enrichment of EMT, TNFɑ, and IFNɑ/gamma response gene sets. C2 tumors displayed significantly higher expression of goblet cell and enterocyte marker genes and enrichment of metabolic gene sets. LOF mutations in epigenetic regulators ARID1A, ARID1B, and EP300 were enriched in C1 tumors. In the ICI cohort of metastatic MSI-COAD patients, C2 was associated with significantly shorter PFS (HR = 6.6, 95% CI: 2.2-19.6, p < 0.0001) and OS (HR = 4.4, 95% CI: 1.4-13.3, p < 0.001). In C2 patients, the median PFS and OS following ICI were 8 and 34 months, respectively; in C1, neither median was reached. C2's association with shorter PFS but not OS following ICI was independent of ICI regimen, line of therapy, collection procedure type, age, sex, and TMB in a CoxPH model (HR = 7.0, 95% CI: 1.8-26.6, p = 0.005). In the subset of our ICI cohort treated with Nivolumab + Ipililumab (n=10), we observed that 4/4 C2 patients progressed within 6 months of treatment while 0/6 C1 patients progressed (p = 0.049, HR = Inf). Finally, we demonstrate that NMF subtypes presented here have a stronger association with PFS in CoxPH models following ICI compared to CMS subtypes (CMS3 vs CMS1 HR=3.5, 95% CI: 1.1-11.1, p = 0.04, CMS4 vs CMS1 HR = 4.2, 95% CI: 0.7-24.6, p = 0.11). Conclusions: We identified novel MSI-COAD transcriptomic subtypes that stratify patient response to ICI, including Nivolumab + Ipililumab. In addition to personalizing patient care, these subtypes could inform clinical trial design and target discovery by identifying patients with an unmet clinical need. Acknowledgements: We acknowledge FightCRC for funding this study and colleagues at Tempus AI and FightCRC for helpful discussion.