Abstract
Increasing evidence has supported dysbiosis in the faecal microbiome along control-adenoma-carcinoma sequence. In contrast, the data is lacking for in situ tumor bacterial community over colorectal cancer (CRC) progression, resulting in the uncertainties of identifying CRC-associated taxa and diagnosing the sequential CRC stages. Through comprehensive collection of benign polyps (BP, N = 45) and the tumors (N = 50) over the four CRC stages, we explored the dynamics of bacterial communities over CRC progression using amplicons sequencing. Canceration was the primarily factor governing the bacterial community, followed by the CRC stages. Besides confirming known CRC-associated taxa using differential abundance, we identified new CRC driver species based on their keystone features in NetShift, including Porphyromonas endodontalis, Ruminococcus torques and Odoribacter splanchnicus. Tumor environments were less selective for stable core community, resulting in heterogeneity in bacterial communities over CRC progression, as supported by higher average variation degree, lower occupancy and specificity compared with BP. Intriguingly, tumors could recruit beneficial taxa antagonizing CRC-associated pathogens at CRC initiation, a pattern known as "cry-for-help". By distinguishing age- from CRC stage-associated taxa, the top 15 CRC stage-discriminatory taxa contributed an overall 87.4% accuracy in diagnosing BP and each CRC stage, in which no CRC patients were falsely diagnosed as BP. The accuracy of diagnosis model was unbiased by human age and gender. Collectively, our findings provide new CRC-associated taxa and updated interpretations for CRC carcinogenesis from an ecological perspective. Moving beyond stratifying case-control, the CRC-stage discriminatory taxa could add the diagnosis of BP and the four CRC stages, especially the patients with poor pathological feature and un-reproducibility between two observers.