A multi-phenotype genome-wide association study of clades causing tuberculosis in a Ghanaian- and South African cohort

Abstract

Despite decades of research and advancements in diagnostics and treatment, tuberculosis remains a major public health concern, particularly in low- and middle-income countries. New bioinformatics and computational methods are needed to interrogate the intersection of host- and bacterial genomes and identify novel targets for anti-tuberculosis drugs. Host genotype datum and paired infecting bacterial isolate information were analysed for associations using a multinomial logistic regression framework implemented in SNPTest. Two geographically distinct cohorts were evaluated: a cohort of 947 participants self-identifying as belonging to a five-way admixed South African population and a Ghanaian cohort consisting of 3 311 participants. We report potential associations between host genetic variants and multiple members of the Mycobacterium tuberculosis complex (MTBC). Although none of the variants analyzed in the South African cohort passed the GWAS cut-off for significance, 32 single nucleotide polymorphisms were identified in the Ghanaian cohort as being statistically significantly associated with risk for infection with strains of different members of the MTBC. Further analysis revealed that two of these SNPs were directly genotyped, and the rest were imputed using the 1000 Genomes Phase 3 reference panel. The availability of paired host-pathogen data is imperative for investigating strain-specific interactions between MTBC and its host. As demonstrated by this study, the implementation of a multinomial logistic regression using paired host-pathogen data may prove valuable for further research investigating the complex relationships driving infectious disease.