Meta-Analysis and Validation of a Colorectal Cancer Risk Prediction Model Using Deep Sequenced Fecal Metagenomes

Area of Science:

• Microbiome Research
• Oncology
• Genomics

Background:

• The gut microbiome represents a modifiable risk factor for colorectal cancer (CRC).
• Previous studies have explored the association between gut bacteria and CRC, necessitating meta-analyses and validation.
• Identifying reliable microbiome-based biomarkers is crucial for early CRC detection and prevention.

Purpose of the Study:

• To identify a gut microbiome signature for predicting colorectal cancer (CRC) risk and precancerous lesions.
• To validate predictive models using independent datasets, including the Colorectal Cancer Screening (COLSCREEN) study.
• To investigate the role of specific bacterial species in CRC development and progression.

Main Methods:

• Re-analysis of eight published stool sequencing datasets and a meta-analysis of microbial-wide associations (MWAS).
• Development of cross-validated LASSO predictive models to identify a CRC microbiome signature.
• Validation of the predictive models in the COLSCREEN dataset (n=156) within a CRC screening context.

Main Results:

• The MWAS meta-analysis identified 95 bacterial species significantly associated with CRC (FDR < 0.05).
• The LASSO model achieved an area under the receiver operating characteristic curve (aROC) of 0.81 for CRC prediction, with validation at 0.75.
• The CRC-trained model showed limited predictive accuracy (aROC=0.52) for precancerous lesions, indicating microbial changes might be tumor-consequential.

Conclusions:

• A 32-bacterial species signature accurately predicts CRC but not precancerous lesions.
• The identified microbial signature's inability to predict early lesions suggests it may be a consequence of CRC rather than a cause.
• Future research should focus on developing microbiome signatures for both CRC and precancerous lesions for enhanced screening programs.