Computational analysis of shotgun metagenomic data from human gut microbiota

Abstract

Shotgun metagenomic sequencing is rapidly becoming the most commonly used approach to identify and quantify complex microbial samples, especially for the study of human gut microbiota. A variety of bioinformatics methods have been developed for shotgun metagenomics data, yet there is still a lot of unresolved issues that limit this technology’s full potential. In this thesis, I examine several important issues that arise from the analysis of shotgun metagenomics data and their application to clinical studies. These issues include: 1. Identifying factors associated with metagenomic species abundance estimation: through benchmarking several key metagenomic sequencing steps such as sample storage, sequencing platform and abundance estimation methods, I identified experimental and bioinformatic factors that influenced the estimation of species abundance estimation. I further proposed a new “gradient framework” to perform quality assessment of individual species abundance estimations. 2. Application of shotgun metagenomic to uncover the association between human gut microbiome and hypertension. Using a cohort of 241 Hong Kong adults, we used various statistical methods with careful covariate adjustment to identify a strong association of gut microbiome and hypertension in females. Specifically, the species Ruminococcus gnavus, Clostridium bolteae, and Bacteroides ovatus were found enriched in hypertensive females. Circulating blood short-chain fatty acid, propionic acid, also associated with gut microbiome and hypertension in females which supports a possible mediating role. 3. Discovery of a novel phenomenon of uneven read coverage (‘bias’) at transcription start sites of bacterial genes. I found that the extent of this bias was correlated with gene expression which currently does not have an obvious biological explanation. Comprehensive investigation of possible technical artefacts such as read mappability or sequencing platform bias was excluded, however GC bias demonstrated strong association with read coverage, although independent of its association with gene expression. We demonstrated this phenomenon arises due to strain diversity in the intergenic region upstream of TSSs where mutations from the reference genome lead to an underrepresentation of read coverage. This collection of work contributes to new knowledge, computer algorithms, and biological insights that enable the potential of metagenomics shotgun sequencing to unlock novel clinical associations as well as interesting biological phenomenon.