Gut fungal signatures in colorectal cancer and their potential for supporting diagnosis: a multi-cohort metagenomic analysis.

BACKGROUND: Colorectal cancer (CRC) is influenced by host factors and environmental exposures that shape gut microbial ecosystems. Although bacterial and viral alterations in CRC have been widely investigated, the role of gut fungi remains underexplored, partly because of their low biomass and the limited availability of well-curated fungal reference genomes. METHODS: We conducted a large-scale metagenomic analysis across 9 publicly available cohorts comprising 1,433 fecal samples to characterize CRC-associated fungal alterations and fungal-bacterial co-abundance patterns. The predictive value of microbial signatures was assessed using LASSO and random forest models, with external validation performed in 6 independent cohorts comprising 272 samples. RESULTS: Multi-cohort analysis revealed CRC-associated alterations in gut fungal community structure and selected diversity measures. Differential abundance analysis identified 15 fungal species with recurrent changes across cohorts. Among them, Saccharomyces cerevisiae c86 and Trichophyton rubrum c61 showed predominant enrichment in healthy controls, whereas Barnettozyma c122 and Pseudopithomyces c302 showed predominant enrichment in CRC. Fungal-only models exhibited limited standalone predictive capacity. However, integrating fungal features with bacterial biomarkers modestly improved CRC prediction performance compared with bacterial-only models. In external validation, the random forest-based fungal-bacterial model increased the mean AUC from 0.722 to 0.762, with improved AUCs in 5 of the 6 validation cohorts. CONCLUSIONS: This study suggests that CRC is associated with gut fungal dysbiosis and supports the exploratory value of gut fungal signatures as adjunctive features in microbiome-based CRC prediction models. These findings highlight the importance of incorporating fungal communities into CRC microbiome research while emphasizing the need for prospective and mechanistic validation.