Strategies for Mining Metagenomic Markers of the Gestational Diabetes Mellitus Microbiome

Abstract

Whole metagenome sequencing of pregnant mothers’ gut flora captures a snapshot of the microbial and metabolic landscape. Using case and control cohorts of pregnant Chinese women (n=60, 30‐30), metagenomic sequencing and clinical health data were mined using a simple correlation‐based coverage binning strategy to investigate microbiome features of gestational diabetes mellitus (GDM). Our approach sought to uncover novel associations between the hosts’ clinical variables of interest and specific genome regions of the resident gut bacteria. Candidate species were selected using two sample non‐parametric tests to determine differentially abundant species from sequencing. Next, available genomic reference sequences for the selected species were subdivided into 500 bp (base pairs) length sequence bins. Each bin is represented by the number of mapped sequences that overlap the location. After normalizing for depth of sequencing, each 500 bp region was assessed for its Spearman correlation with clinical variables across all samples. This method identified regions of interest that highly correlated with the results of the oral glucose tolerance tests after one hour (OGTT‐1). Using genome feature annotations of the reference sequence to identify nearby genes, our method identified the Starch‐binding protein SusD as an associated bacterial gene. Later, validation by PCR confirmed its differential abundance in GDM mothers compared to control pregnancies. These findings are to be published in an upcoming article; however, the mining strategy itself has not been fully investigates. In this work we compare the simple binning‐correlation strategy with more sophisticated de novo assembly followed by gene catalog construction and mapping. We find that the binning strategy is highly effective offering a simple but powerful data mining tool that compares favorably to more sophisticated and computationally costly methods.