From clusters to clinic: An 8-gene signature combined with mucinous component stratifies benefit of anti-CTLA-4 addition to anti-PD-1 in dMMR/MSI-H metastatic colorectal cancer

Background Around 15–50% of patients with dMMR/MSI-H metastatic colorectal cancer (mCRC) experience primary and secondary resistance to immune checkpoint inhibitors (ICI). We previously described three transcriptomic clusters (A, B, C) linked to progression-free survival (PFS), with cluster A (“StromalHIGH – ProliferationLOW”) associated with the shortest PFS. Here we combine gene signature and mucinous phenotype to predict the benefit from anti-CTLA-4 plus anti-PD-1. Methods We used 3’RNA-sequencing data from pre-treatment samples of two independent cohorts of patients with dMMR/MSI-H mCRC treated with anti-PD-1 ±anti-LAG-3 (“Mono”) or anti-PD-1 +anti-CTLA-4 (“Combo”). A rank-based k-Top Scoring Pairs (k-TSP) classifier was trained in Cohort 1 (103 patients) to predict assignment to Cluster A versus B/C and validated in Cohort 2 (87 patients). Patients receiving an anti-PD-L1 and with ECOG ≥ 2 were excluded. We defined a composite stratum crossing Clusterk-TSP (A vs B/C) with mucinous component (present vs absent). Results Cluster A was enriched for mucinous tumors and had shorter PFS in both cohorts. The k-TSP selected four gene pairs predictive of Cluster A (YPEL2 >H2AZ1, H6PD>COX7A2, STAT5B>SNRPE, ZHX3 >TRAPPC4) and was validated in Cohort 2 with 68.1% balanced accuracy. In the pooled population (163 patients), the composite stratum × treatment interaction improved model fit (likelihood-ratio p ≈ 0.04). Patients with Clusterk-TSP A + mucinous (25/163, 15.3%) derived advantage from Combo versus Mono: adjusted 24-month PFS 72.2% vs 13.8% (HR 0.10, 95% CI 0.02–0.39; p < 0.001). Conclusions An 8-Gene signature combined with mucinous component identifies a subgroup that preferentially benefits from anti-PD-1 +CTLA-4, warranting prospective validation to guide regimen selection in dMMR/MSI-H mCRC.