Gut Microbiome Classifier Improves Colorectal Cancer Risk Stratification

Clinicians therefore need tools that improve risk stratification while fitting into stool-based testing workflows. Although the gut microbiome has been implicated in tumor biology, prior findings have often varied across cohorts and sequencing methods, limiting its clinical use. Building on this need, a large cross-cohort analysis has now defined a robust microbial signature and introduced a classifier that could inform future diagnostics and dietary interventions.

Researchers from the European Molecular Biology Laboratory (EMBL) and Leiden University Medical Center, working within the EU-funded CartoHostBug project, developed a machine-learning classifier trained on gut microbiome profiles. The model integrates data generated by different sequencing approaches and produces a quantitative “cancer-like microbiome” score for individual stool samples. The study was published in Cell Host & Microbe on July 8, 2026.

The analysis re-examined nearly 6,800 gut microbiome profiles drawn from 27 independent studies conducted in Germany, the Netherlands, and Switzerland. By harmonizing heterogeneous datasets, the team identified a generalizable microbial signature that distinguished colorectal cancer from non-cancer across populations, geographies, age groups, and both early- and late-onset disease. The approach addresses prior limitations of small, inconsistent cohorts that hindered translation.

To link noninvasive readouts with tumor biology, investigators analyzed 906 intestinal tissue samples. Microbes enriched in tumor tissue mirrored organisms driving the stool-based signature, and cancer-associated taxa appeared even in early-stage tumors. However, stool-based detection of precancerous adenomas remained challenging, reflecting weaker microbial signals that underscore the need for more sensitive strategies or multimodal combinations.

Diet emerged as a key modifier of the microbiome score. Lower dietary fiber intake correlated with more cancer-like patterns, while dietary intervention studies suggested that increasing fiber could reduce these scores. Species-level analysis refined understanding of Fusobacterium, showing that Fusobacterium nucleatum subsp. animalis was consistently more prevalent in cancer samples worldwide, whereas other subspecies displayed region-specific patterns. While the classifier is not a diagnostic test, it establishes a reference framework for future clinical tools and interventional research under CartoHostBug.

Related Links
EMBL
Leiden University Medical Center