Intratumoral microbiota-host gene associations characterize stage-associated microecological patterns in colorectal cancer

Abstract

Intratumoral microbiota are emerging as important ecological components of the colorectal cancer (CRC) microenvironment. However, their stage–associated patterns and coordination with host gene expression remain poorly understood. This study aimed to characterize stage–associated associations between intratumoral microbiota and the host transcriptome in CRC. We profiled 73 regional tumor specimens from 30 independent, treatment-naïve CRC patients using full-length 16S rRNA gene sequencing and mRNA sequencing. Microbial diversity, predicted functional profiles, co-occurrence networks, differential abundance, and microbiota–host transcriptome associations were analyzed. Regularized canonical correlation analysis and machine-learning modeling with SHAP-based feature interpretation were performed to identify stage-discriminative features. Candidate associations were evaluated in two independent public microbiome datasets. Mendelian randomization and immune-infiltration analyses were further conducted to explore the potential biological relevance of representative microbiota–gene associations. Disease stage was a major factor shaping intratumoral microbial composition. Advanced-stage CRC exhibited increased microbial α-diversity, distinct predicted functional profiles, and altered microbial co-occurrence patterns. Integrated microbiota–transcriptome analysis identified ten genus-level and six species-level candidate microbiota–gene association pairs. Core genera showed reproducible stage-discriminative signals in two independent public microbiome datasets. SHAP analysis identified the principal microbial and host transcriptomic features contributing to stage classification. Integrating microbial and host gene signatures improved classification performance, with the combined model achieving an AUROC greater than 0.90. The Alloprevotella –MDH1 pair was selected as a representative association axis. Alloprevotella abundance was negatively correlated with MDH1 expression, while low MDH1 expression was associated with poorer survival and distinct metabolic–immune states in CRC. Exploratory Mendelian randomization analysis provided preliminary genetic epidemiological support for an association between Alloprevotella abundance and CRC risk. Intratumoral microbiota and host transcriptional profiles exhibit coordinated, stage-associated alterations in CRC. The identified microbiota–gene associations, particularly the Alloprevotella –MDH1 axis, provide potential biological links between tumor microbial ecology, host metabolism, immune states, and CRC progression. These findings support the value of integrated microbiome–transcriptome analysis for characterizing the intratumoral microenvironment and identifying candidate microbial–host regulatory relationships.