Circulating bacterial DNA as a tool towards noninvasive biomarkers for colorectal adenocarcinoma and adenoma.

Abstract

3045 Background: The gut microbiota is closely associated with the progression of colorectal neoplasia. While most metagenomics studies utilized fecal samples, circulating bacteria DNA in colorectal adenocarcinoma (ADC) or adenoma (ADM) patients remain unexplored. This study aimed to characterize microbiota DNA in plasma samples and build a machine-learning model for ADC and ADM early detection. Methods: In this proof-of-concept study, we performed whole genome sequencing (~30X) of plasma samples from 25 ADC patients, 10 ADM patients, and 22 healthy controls (HC). Significant biomarkers were identified in the discovery cohort (12 ADC and 11 HC) and built into a random-forest model which was tested in the validation cohort (13 ADC and 11 HC). These biomarkers were further examined in ADM and tested for abundance difference with ADC and HP. Results: In the discovery cohort, 111 species had increased relative abundance in ADC compared to HC and 165 species had decreased relative abundance. Alteration in several species such as Flavobacterium and Ruminococcus torques were consistent with previously published results in faecal and gut microbiome samples. The random forest-recursive feature elimination model selected 28 significant species from the discovery cohort (mean AUC = 0.98, repeated 2-fold cross-validation) and yielded an AUC of 1 in the validation cohort. Interestingly, most biomarker species in ADM patients, with abundance intermediate between ADC and HC, were distinguished from HC. To further test the clinical utility of this model, sequencing data were randomly down-sampled to 1X and the model performance remained robust (AUC = 1) at distinguishing ADC and ADM from HC. Conclusions: This study is the first effort to characterize circulating bacteria DNA in patients with ADC and ADM. Our findings revealed significant difference in relative abundance of several bacterial species between ADC, ADM and HC. A predictive model constructed with selected microbial features accurately distinguished ADC and ADM from HC. Circulating bacteria biomarkers represent potential non-invasive tools for early diagnosis of colorectal neoplasia.