# Finding the True Responders: Stratifying dMMR/MSI-H Tumors for ICI Response

## TL;DR

This study identifies a 20-gene signature that helps predict which patients with dMMR/MSI-H tumors will respond best to immunotherapy.

## Contribution

A novel 20-gene signature was developed to accurately identify true responders to immune checkpoint inhibitors in dMMR/MSI-H tumors.

## Key findings

- The 20-gene signature showed high reproducibility with a mean AUC of 0.95 and accuracy of 89%.
- The signature identified tumors with higher PD-L1 blockade response in the IMvigor210 cohort.
- It captures immune heterogeneity and improves survival trends in the TMB-high subset.

## Abstract

Immune checkpoint inhibitors (ICIs) have transformed cancer therapy, but not all patients with mismatch repair-deficient (dMMR) or microsatellite instability-high (MSI-H) tumors respond effectively. To explore this variability, we analyzed immune profiles of dMMR/MSI-H tumors and classified them into four immune subgroups. A distinct “HotHigh” group showed strong immune activation, from which we developed a 20-gene signature that accurately identifies patients most likely to benefit from ICIs. This signature was validated across multiple datasets, supporting its potential as a reliable biomarker to guide precision immunotherapy in clinical practice.

Background/Objectives: Immune checkpoint inhibitors (ICIs) show durable efficacy in tumors with deficient mismatch repair (dMMR) or high microsatellite instability (MSI-H), yet clinical responses remain heterogeneous. This study aimed to define immune subgroups within dMMR/MSI-H tumors and develop a reproducible transcriptomic signature predictive of ICI response. Methods: Four MSI-H-enriched cancer types (UCEC, COAD, READ, STAD) from The Cancer Genome Atlas were analyzed. Tumors were stratified by immune cell infiltration (MCP-counter immune composite score) and T-cell-inflamed gene expression profiles (GEP score). Integrating these two axes defined four immune subgroups. Differential expression, random forest feature selection, and pathway enrichment were performed to identify immune programs. A 20-gene immune signature representing the most immune-active subgroup was developed and validated across TCGA, GEO (GSE39582), and IMvigor210 cohorts. Results: Among the four subgroups, the most immune-active group showed strong activation of interferon signaling, antigen presentation, and T-cell-mediated pathways. The 20-gene signature—including CD74, STAT1, TAP1, and HLA-class genes—achieved high reproducibility (mean AUC = 0.95 ± 0.02; accuracy ≈ 89%). In the IMvigor210 cohort, this signature identified tumors with higher PD-L1 blockade response (55.6% vs. 32.8%, p = 0.034) and improved survival trends in the TMB-high subset. Conclusions: The proposed 20-gene signature quantitatively captures immune heterogeneity in dMMR/MSI-H tumors and serves as a practical, interpretable biomarker to identify true ICI responders and guide precision immunotherapy.