Connecting the dots : integrative analysis of genomic, metabolomic and phenotypic data from a population cohort

Abstract

Lower back pain (LBP) is one of the most prevalent global health issues and a main cause of disability. Lumbar disc degeneration (LDD) is one of the major reasons for LBP, which could be evaluated by radiographic observations through magnetic resonance imaging (MRI). Nevertheless, these MRI observations, as diagnostics for LBP/LDD, are prone to human error and may be insufficient in detecting real-time variations of complex biological systems. This thesis aims to identify novel LDD biomarkers related to altered metabolism, which could aid personalized diagnosis and treatment of LBP.

This study is based on a population cohort of 3584 southern Chinese. Over 1000 individuals were followed with MRI scans, which were read by experienced physicians specialized in LDD. Statistical analyses showed that the severity of LDD is significantly greater in the lower lumbar region, the lower disc levels forming a cluster more related to age. Accordingly, a systematic way to quantify the degree of LDD from raw MRI reads is proposed in this thesis. Apart from phenotyping, 2482 samples in the cohort were genotyped, and the serum samples of 757 individuals were acquired for proton nuclear magnetic resonance spectroscopy, resulting in 130 metabolomic measurements over three molecular windows.

In order to discover genetic variants associated with different metabolomic measurements, genome-wide association studies (GWAS) were conducted on 571 individuals for each of the 130 metabolomic measurements. In total, 123 unique single nucleotide polymorphisms were found to be significantly associated with one or more metabolomic measurements; among them, intergenic variants were underrepresented, whereas exonic, intronic and UTR3 variants were enriched. My results suggest significant associations between 42 different metabolomic traits and a number of genetic loci. Polyunsaturated fatty acids were found to be significantly associated with the FADS1/FADS2 loci, and CTTNBP2 was identified as a potential risk locus for a cluster of lipid / fatty acid related metabolites.

The human metabolome was next estimated based on the summary statistics from previous GWAS and genomic data via meta-analysis and polygenic scoring. The associations between (estimated) metabolomic data and various phenotypes (anthropometric, behavioral, clinical, and LDD-related) were tested using different regression methods, ranging from simple linear models to Lasso. Potential metabolomic biomarkers for LDD were identified, including blood lipid levels, the mean diameter for very low density lipoprotein particles, sphingomyelins and tyrosine.

Through GWAS, polygenic scoring and association analysis, this study pinpoints metabolomic biomarkers for LDD with a purely data-driven approach. It also proposes a new way to analyze genomic, metabolomic and phenotypic data in an integrative manner, utilizing metabolome prediction models. This process of the integration of big omics data could help us discover known and novel metabolomic biomarkers associated with complex traits and gain a better understanding of the mechanisms of these associations.