# Prediction of immunotherapeutic responses by a classifier model based on inflammation-associated tumor microenvironment signatures in colorectal cancer

## TL;DR

This study identifies inflammation-related cell subgroups in colorectal cancer that predict resistance to immunotherapy and develops a model to forecast treatment responses.

## Contribution

A novel classifier model based on inflammation-associated tumor microenvironment signatures to predict immunotherapeutic responses.

## Key findings

- Three inflammation-related cell subgroups (CEMIP+ Monocytes, CCL4+ Neutrophils, and MMP3+ Fibroblasts) are linked to immune suppression and poor immunotherapy response.
- The classifier model based on inflammatory signatures accurately predicts immunotherapeutic responses across cancer types.
- The study reveals a correlation between pro-inflammatory factors and inhibited anti-tumor immunity in the tumor microenvironment.

## Abstract

In recent years, the application of immunotherapy has greatly improved the prognosis of cancer patients. However, a proportion of patients will acquire resistance to immunotherapy, leading to a lower response rate and poorer clinical outcome. The underlying mechanisms contributing to the therapeutic resistance and accurate biomarkers to predict immunotherapy responses remain unclear.

We comprehensively analyzed a single cell RNA-sequencing dataset of microsatellite instability-high colorectal cancer patients received anti-PD1 immunotherapy. We dissected the heterogeneity of the immunosuppressive tumor microenvironment contributing to the therapeutic resistance and highlighted on a correlation between pro-inflammatory factors and inhibited immune responses. We established a classifier model using Random Forest algorithm based on the common marker genes of inflammation-associated subpopulations. The validation of the model and further analysis between potential responders and non-responders was also performed in bulk RNA-seq cohorts.

Three inflammation-related cell subgroups, including CEMIP+ Monocytes, CCL4 + Neutrophils and MMP3 + Fibroblasts were identified to be associated with immune-suppressed signatures and unfavorable responses to immunotherapy. The classifier model based on inflammatory signatures exhibited acceptable accuracy and robustness to predict immunotherapeutic responses across cancer types.

Our study dissected the heterogeneity of the immunosuppressive tumor microenvironment and highlighted a correlation between pro-inflammation signatures and inhibited anti-tumor immunity. We also developed a novel classifier model based on inflammation-related signatures to predict patients’ responses to immunotherapy.

The online version contains supplementary material available at 10.1007/s12672-026-04548-6.