Pathology-derived clinical micro-architectural diagnostics of tumour-microbiome interactions in colorectal cancer.

BACKGROUND: Classical tumour pathology reports contain a largely untapped layer of information that may indicate tumour-microbial interactions. However, routine colorectal cancer pathology staging does not take into account microbiome-associated tumour micro-architecture signatures, thus limiting insights into intratumoral microbial ecology, prognostic stratification and treatment-relevant microbial information. In this study, we analysed scanned USA pathology reports to quantify likely intratumoral microbiome-associated micro-architectural signatures.

METHODS: We studied 1,978 TCGA colorectal cancer pathology reports from 1,249 colon adenocarcinomas, 559 rectal adenocarcinomas and 170 reports without a definitive anatomic site using rule-based natural language processing to extract microbiome-linked micro-architectural features. Barrier-disruption and invasion-access signatures were identified from the reports as microbiome-associated pathology micro-architecture signatures that occur with microbial-related necrosis, hypoxia, toxins, colonisation, persistence, metabolic activity and/or tumour interaction. We developed a z-scored composite index called Report-based Microbial Ecology Likelihood Score (RMELS) and used Kaplan-Meier log-rank analyses, multivariable Cox regression, Kruskal-Wallis tests and receiver operation characteristic curves with bootstrap confidence intervals. Proportional hazards assumptions were tested for statistical significance at two-sided p < 0.05.

RESULTS: Microbiome-associated pathology micro-architectural signatures were highly prevalent in the pathology reports. Barrier-disruption features, including ulceration (41.1%) and mucin alteration (16.7%), were common and increased with tumour stage (Kruskal-Wallis p < 0.0001). Prominent invasion-access features included infiltrative growth (59.4%, 95% CI 57.2-61.5), lymphovascular invasion (18.6%, 95% CI 17.0-20.4) and perineural invasion (22.9%, 95% CI 21.1-24.8). All showed heterogeneous, non-monotonic distributions across pathologic stages, indicating activation of microbial injury and invasion programmes. Integration of these features into our signature score, ordered tumours along a continuous microbiome-permissiveness gradient independent of pathological stage. With limited information, our signature score discriminated early (T1) from advanced (T4) disease more effectively than barrier or invasion features alone (AUC = 0.66, 95% CI 0.58-0.74, p < 0.0001). Right-sided colonic tumours exhibited significantly higher scores than left-sided colonic and rectal tumours (FDR q < 0.001), aligning with known microbial biogeography. In multivariable Cox models adjusted for pathological stage, our signature score RMELS showed modest but directionally consistent association with overall and progression-free survival, capturing microbiology-relevant risk not resolved by pathological staging.

CONCLUSIONS: Routine classical colorectal cancer pathology reports contain intratumoral microbiome-associated pathology micro-architectural signatures. Quantifying these exploratory tumour-microbial signatures using digital pathology will enable scalable, microbiology-informed risk stratification and prognostic modelling to complement the current pathological staging.