Genome-wide association study of susceptibility to hospitalised respiratory infections

Abstract

Background: Globally, respiratory infections contribute to significant morbidity and mortality. However, genetic determinants of respiratory infections are understudied and remain poorly understood. Methods: We conducted a genome-wide association study in 19,459 hospitalised respiratory infection cases and 101,438 controls from UK Biobank (Stage 1). We followed-up well-imputed top signals from our Stage 1 analysis in 50,912 respiratory infection cases and 150,442 controls from 11 cohorts (Stage 2). We aggregated effect estimates across studies using inverse variance-weighted meta-analyses. Additionally, we investigated the function of the top signals in order to gain understanding of the underlying biological mechanisms. Results: From our Stage 1 analysis, we report 56 signals at P<5×10 ^-6, one of which was genome-wide significant ( P<5×10 ^-8). The genome-wide significant signal was in an intron of PBX3, a gene that encodes pre-B-cell leukaemia transcription factor 3, a homeodomain-containing transcription factor. Further, the genome-wide significant signal was found to colocalise with gene-specific expression quantitative trait loci (eQTLs) affecting expression of PBX3 in lung tissue, where the respiratory infection risk alleles were associated with decreased PBX3 expression in lung tissue, highlighting a possible biological mechanism. Of the 56 signals, 40 were well-imputed in UK Biobank and were investigated in Stage 2. None of the 40 signals replicated, with effect estimates attenuated. Conclusions: Our Stage 1 analysis implicated PBX3 as a candidate causal gene and suggests a possible role of transcription factor binding activity in respiratory infection susceptibility. However, the PBX3 signal, and the other well-imputed signals, did not replicate in the meta-analysis of Stages 1 and 2. Significant phenotypic heterogeneity and differences in study ascertainment may have contributed to this lack of statistical replication. Overall, our study highlighted putative associations and possible biological mechanisms that may provide insight into respiratory infection susceptibility.

Keywords: GWAS; Respiratory infections; UK Biobank; electronic medical records.

Figure 1.. Overview of study design.

Figure 2.. Frequency of individual ICD-10 codes used to define the 19hospitalised respiratory infection phenotype.

Frequency (log ₁₀ scale) of individual ICD-10 codes used to define the hospitalised respiratory infection phenotype. To improve visualisation, only codes that occurred in 10 or more individuals are shown. Individuals may contribute to the overall count of more than one ICD-10 code. A description of each ICD-10 code, as well as the ICD-10 code itself, is shown.

Figure 3.. Forest plot for the sentinel variant in the genome-wide significant signal from the Stage 1 analysis following meta-analysis of Stages 1 and 2.

Forest plot for the sentinel variant, rs10564495, in the genome-wide significant signal identified in the Stage 1 following inverse variance-weighted fixed effects meta-analysis of results from Stages 1 and 2. The A allele for this variant was taken to be the coded allele. Where a proxy variant was used, which was consistently the rs10819083 variant, the T allele was taken to be the allele that corresponds to the A allele of the rs10564495 variant, as reported by the LDpair tool in the LDlink suite of online applications.

Figure 4.. Hospitalised respiratory infection GWAS versus eQTL for PBX3 in lung tissue (GTEx v7): probability of colocalisation = 87%.

Each point corresponds to a genetic variant, with genomic position (GRCh37) on the x-axis and –log ₁₀( p-value) on the y-axis. The top plot shows regional association results for the genome-wide significant signal (sentinel variant: rs10564495) from the hospitalised respiratory infection GWAS. The bottom plot shows regional association results for the genome-wide significant signal from the eQTL analysis. The plotting window extends 1Mb either side of the sentinel variant in the region. The sentinel variant is represented by a blue triangle, with all other genetic variants in the region coloured according to the extent of pairwise linkage disequilibrium with the sentinel variant: red points reflect genetic variants that have r ² >0.8 with the sentinel variant, orange points reflect genetic variants that have 0.5< r ² ≤0.8 with the sentinel variant, yellow points reflect genetic variants that have 0.2< r ² ≤0.5 with the sentinel variant, and grey points reflect genetic variants that have r ² ≤0.2 with the sentinel variant. The area shaded in light pink represents the gene implicated by the eQTL analysis. The red dashed line represents a p-value threshold of 5×10 ^-8.