Background & Purpose: The incidence of colorectal cancer (CRC) has risen sharply in recent decades, with a disproportionate increase observed among younger individuals in Japan and other countries, representing an emerging public health challenge. The etiological contribution of the gut microbiota to CRC pathogenesis is increasingly recognized, yet the mechanisms involved remain to be fully clarified. We integrated whole-genome sequencing (WGS) and transcriptome profiling of CRC with whole-genome metagenomic sequencing (WGMS) of fecal samples to interrogate host–microbiome interactions at high resolution.
Methods: We integrated a newly established cohort (Cohort-1) with our previously reported dataset (Cohort-2), affording a total of 600 participants (454 patients with CRC and 146 healthy controls (HCs)). Whole-genome and whole-transcriptome sequencing were performed on Japanese CRC cases from Cohort-1 (n = 200) to delineate their genomic landscape comprehensively. In parallel, whole-genome metagenomic sequencing (WGMS) was conducted on fecal samples from 395 individuals to characterize gut microbial profiles. An interpretable artificial intelligence (AI) framework stratified CRC cases into distinct subgroups according to microbial composition.
Results : Application of interpretable artificial intelligence enabled the stratification of CRC into four distinct microbiome-informed subtypes. WGS analysis identified mutational signatures SBS88 and ID18, linked to colibactin exposure, as early clonal events detected in 44.8% of non-hypermutated patients (82/183). Notably, these signatures were significantly more frequent among patients born after the 1960s, suggesting the involvement of temporal and environmental factors. Colibactin-induced mutations were identified in common driver genes in 15.3% of patients (28 of 183), including an APC splice-site hotspot mutation causing a frameshift deletion, and TP53 mutations. Microbiome-based subclassification revealed subtype-specific clinical and molecular features. Subtype 3 was characterized by a minimal contribution from Fusobacterium nucleatum, a marked temporal increase in incidence after 1980, and a younger age at onset, accompanied by an increased prevalence of the colibactin-associated mutational signature. 
Conclusions: This study represents the first integrative analysis combining stool-derived metagenomic profiling with WGS of CRC tissues. Through this approach, patients were stratified into four microbiome-based subtypes. Subtype 3, characterized by a low contribution of F. nucleatum in stool as determined by WGMS, was paradoxically associated with younger age and a higher prevalence of colibactin-related mutational signatures. The colibactin-associated mutational signature was observed in ~45% of non-hypermutated CRCs in Japan, with a particularly high prevalence among younger patients. These findings suggest that preventive strategies aimed at limiting colonization by pks? bacteria and suppressing colibactin production may represent viable approaches to reducing the global burden of CRC.