Cross-cohort analysis identifies shared gut microbial signatures and validates microbial risk scores for colorectal cancer

Abstract

Background: Microbiome-wide association studies showed links between colorectal cancer (CRC) and gut microbiota. However, the clinical application of gut microbiota in CRC prevention has been hindered by the diversity of study populations and technical variations. We aimed to determine CRC-related gut microbial signatures based on cross-regional, cross-population, and cross-cohort metagenomic datasets, and elucidate its application value in CRC risk assessment.

Methods: We used the MMUPHin tool to perform a meta-analysis of our own cohort and seven publicly available metagenomics datasets to identify gut microbial species associated with CRC across different cohorts, comprising of 570 CRC cases and 557 controls. Based on differential species sets, we constructed the microbial risk score (MRS) using α-diversity of the sub-community (MRS_α), weighted/unweighted summation methods and machine learning algorithms. Cohort-to-cohort training and validation were performed to demonstrate the transferability.

Results: We found that MRS_α of core species was better validated and more interpretable than those constructed with summation methods or machine learning algorithms. Six species, including Parvimonas micra, Clostridium symbiosum, Peptostreptococcus stomatis, Bacteroides fragilis, Gemella morbillorum, and Fusobacterium nucleatum, were included in MRS_α constructed by half or more of the cohorts. The AUC of MRS_α, calculated based on the sub-community of six species, varied between 0.619 and 0.824 across the eight cohorts.

Conclusion: We identified six CRC-related species across regions, populations, and cohorts. The constructed MRS_α could contribute to the risk prediction of CRC in different populations.

Keywords: Colorectal cancer; Cross-cohort; Disease prediction; Gut microbiota; Microbial risk score.

Fig. 1

The workflow of the microbial risk score (MRS) establishment

Fig. 2

Geographic distribution of participants in 8 colorectal cancer-related studies

Fig. 3

Differential species across cohorts identified via MMUPHin analysis of 8 colorectal cancer study populations. Bar plots of significant differential species (adjusted P < 0.05, Benjamini–Hochberg method) were presented with red for enriched species in colorectal cancer group and blue for enriched species in control group

Fig. 4

Cohort-to-cohort validation of microbial risk score (MRS) constructed with α-diversity of the sub-community. The heatmap shows the AUC values of MRS built with α-diversity of the sub-community of differential species in each cohort. The average value refers to the AUC values in the other validation cohorts

Fig. 5

Cohort-to-cohort validation of microbial risk score (MRS) constructed with LightGBMXT. The heatmap shows the AUC values of MRS built with LightGBMXT in each cohort. The average value refers to the AUC values in the other validation cohorts. The gray squares indicate an AUC less than 0.5