Gut Metabolite Genes Could Predict Colorectal Cancer Survival and Immunotherapy Response

Colorectal cancer remains one of the most common and lethal malignancies worldwide, and clinicians have long struggled with a deceptively simple question: which patients will do well, and which will not? Standard staging systems, built on tumor size, depth of invasion, and lymph node spread, capture only part of the answer. Two patients with seemingly identical disease can follow dramatically different courses, and the molecular machinery driving those divergences has proved stubbornly difficult to pin down. Now, a team of researchers at Ganzhou People’s Hospital in Jiangxi, China, has turned to an unexpected corner of tumor biology for clues: the metabolic handling of propionate, a short-chain fatty acid produced in abundance by the gut microbiome. In a study published in Medical Oncology, Shanmin Yuan, Sha Li, and Cong Chen describe a four-gene signature rooted in propionate metabolism that robustly stratifies colorectal cancer patients into distinct risk groups and appears to encode information about the immune landscape of their tumors.

The rationale for looking at propionate is grounded in a decade of work on the microbiome–cancer axis. Short-chain fatty acids such as acetate, butyrate, and propionate are generated when gut bacteria ferment dietary fiber, and they act on the colonic lining through multiple channels. They serve as histone deacetylase inhibitors, reshaping gene expression programs in epithelial and immune cells; they influence the differentiation of regulatory T cells, which maintain mucosal tolerance; and they modulate macrophage polarization, a process with direct consequences for tumor immune evasion. Recent studies have also implicated propionate metabolism in antitumor T-cell function through enzymes such as propionyl-CoA carboxylase, and in immune escape through pathways involving PD-L1. Yet despite this growing appreciation of propionate as a metabolic axis shaping colorectal cancer biology, no one had systematically built a molecular stratification tool around the genes that govern it.

To fill that gap, the researchers assembled a training cohort by integrating transcriptomic profiles from The Cancer Genome Atlas colon adenocarcinoma and rectum adenocarcinoma collections, and reserved the GSE38832 dataset as an external validation set. They curated a list of propionate metabolism–related genes from the GeneCards database and intersected it with genes differentially expressed between tumor and normal tissue, producing a focused set of candidates. Univariate Cox regression then filtered this list down to genes with genuine prognostic weight. Applying consensus clustering to those prognostic genes revealed two molecular classes of colorectal cancer, distinguishable not only by their survival patterns but also by the composition of their tumor microenvironments. The team characterized those microenvironments using complementary computational tools: ESTIMATE to estimate immune and stromal content, single-sample Gene Set Enrichment Analysis, and CIBERSORT to deconvolve the fractions of infiltrating immune cell types.

From this foundation, the authors distilled a four-gene prognostic signature comprising NLGN1, CILP2, HEPACAM2, and HOXC6. Each of these genes carries its own emerging biography in cancer research. NLGN1, better known as a neuronal adhesion protein, has been shown to promote colorectal cancer progression through the APC/β-catenin pathway and to regulate inflammatory signaling in tumor-derived exosomes. CILP2 has been proposed as a biomarker for peritoneal metastases in colorectal cancer. HEPACAM2 has recently been analyzed as a potential prognosis biomarker and immunotherapy target in the disease. HOXC6, a homeobox transcription factor, promotes metastasis through Wnt/β-catenin signaling and has been linked to effector T-cell exhaustion and interactions with M2 macrophages in microsatellite-instable tumors. The signature scores each patient based on the expression of these four genes, assigning a risk level that splits the cohort into low- and high-risk subsets with clearly separated survival curves.

The technical performance of the signature is a central claim of the study. In the training cohort, the risk score separated patients with stable time-dependent areas under the curve, a measure of how well the model discriminates survivors from non-survivors across follow-up windows. Critically, the same predictive pattern was reproduced in GSE38832, an independent external dataset, which guards against the overfitting that plagues many gene-signature studies. After adjustment for clinicopathological characteristics in Cox proportional hazards modeling, the signature retained independent prognostic relevance, meaning it added information beyond what age, stage, and other conventional variables provide. To translate the model into something usable at the bedside, the researchers constructed a nomogram combining the significant variables, and evaluated it with calibration analysis and decision curve analysis, both of which assess whether a predictive tool would actually improve clinical decisions rather than merely fit retrospective data.

Perhaps the most provocative findings concern immunity and immunotherapy. The two risk groups differed substantially in immune-cell infiltration, checkpoint-gene expression, and immunotherapy-associated signatures. High-risk tumors showed enrichment of immune and stromal components, higher Tumor Immune Dysfunction and Exclusion scores, and lower Immunophenoscores, a pattern generally interpreted as an immunologically hostile microenvironment less likely to respond to checkpoint blockade. In an exploratory analysis of the IMvigor210 urothelial carcinoma cohort, which serves as a benchmark dataset for predicting anti–PD-L1 response, the low-risk group showed a more favorable predicted response. The authors are careful about the limits of that cross-cancer extrapolation, but the direction of the association is consistent with the broader hypothesis that metabolic programs in tumor cells help determine whether immunotherapy can succeed.

The study also mapped the mutational landscape of the risk subgroups using the maftools package, finding that high-risk tumors carried a higher tumor mutational burden alongside their differential immune features, a combination that could reflect distinct evolutionary trajectories or microenvironmental pressures. On the therapeutic side, the researchers used pRRophetic prediction and CellMiner-based drug response analyses to infer potential therapeutic susceptibilities, identifying differential drug sensitivity between the risk groups. This kind of pharmacogenomic inference does not prescribe treatments, but it generates testable hypotheses about which existing compounds might be repurposed for molecularly defined subsets of colorectal cancer patients, an approach that aligns with the broader push toward individualized oncology.

It is worth situating this work within the larger context of colorectal cancer classification. The field already uses the Consensus Molecular Subtypes framework, which divides colorectal cancer into four biologically coherent groups, and numerous metabolic signatures have been proposed for other nutrient pathways, including lactate metabolism. What distinguishes the propionate axis is its direct connection to the microbiome, the organ-adjacent ecosystem that is increasingly recognized as a modulator of immunotherapy response. Studies have shown that short-chain fatty acids can either enhance or undermine checkpoint inhibitor efficacy depending on context, and microbiota-directed interventions, from dietary fiber to specific bacterial strains, are being explored as adjuvants to immunotherapy. A gene signature that quantifies the tumor-side footprint of propionate metabolism offers a way to measure that axis in standard tumor biopsies, without requiring microbiome sequencing.

The caveats are the usual ones for computational oncology. The signature was derived and validated in retrospective public datasets, and it will need prospective testing in clinical cohorts before it can inform treatment decisions. The immunotherapy findings are exploratory and partly rely on a cohort from a different cancer type. The four genes themselves are correlates, not necessarily causal drivers, and the mechanistic question of how propionate metabolism shapes the expression of NLGN1, CILP2, HEPACAM2, and HOXC6 remains open. Still, the study delivers a coherent framework: two propionate metabolism–related molecular subtypes with distinct survival and immune profiles, a validated four-gene risk score, a nomogram for individualized estimation, and a set of hypotheses linking microbial metabolism to checkpoint response and drug sensitivity. As the burden of colorectal cancer continues to grow globally, tools that convert metabolic biology into actionable risk stratification will be watched closely, and this propionate-centered signature has earned a place in that conversation.

Subject of Research: A propionate metabolism–related gene signature for prognosis and immune characterization in colorectal cancer

Article Title: A propionate metabolism–related gene signature predicts prognosis and immunological features in colorectal cancer

Article References: Yuan, S., Li, S., & Chen, C. (2026). A propionate metabolism–related gene signature predicts prognosis and immunological features in colorectal cancer. Medical Oncology, 43(11), Article 311. https://doi.org/10.1007/s12032-026-03431-2

Image Credits: AI Generated

DOI: 10.1007/s12032-026-03431-2

Keywords: colorectal cancer, propionate metabolism, gene signature, prognosis, tumor microenvironment, immunotherapy, short-chain fatty acids, biomarker, TCGA, CIBERSORT, PD-L1, nomogram