Common variants at 11p13 are associated with susceptibility to tuberculosis

Abstract

After imputation of data of the 1000 Genomes Project into a genome-wide data set of Ghanaian tuberculosis cases and controls, we identified a resistance locus on chromosome 11p13, downstream of the Wilms' tumour 1 gene. The strongest signal was obtained at SNP rs2057178 (P = 2.63 × 10⁻⁹). Replication in Gambian, Indonesian and Russian TB case-control study groups increased the significance level to P = 2.57 × 10⁻¹¹.

The influence of host genetic factors on susceptibility to tuberculosis (TB) is well established by twin, linkage and candidate gene analyses^1-3. Recently, in a combined genome-wide association study (GWAS) of African TB case-control groups from Ghana, The Gambia and Malawi we have identified a susceptibility locus on chromosome 18q11.2⁴. Here, we present a novel association of a genetic locus on chromosome 11p13 with resistance to TB, obtained only after imputation of data provided by the 2010-08 release of the 1000 Genomes Project into the genome-wide Ghanaian data.

After genotyping of 1329 TB cases and 1847 controls (Affymetrix SNP Array 6.0) and quality control of SNPs, 793,964 variants were available for imputation analyses (Supplementary Note). Population stratification was low as indicated by a lambda factor (λ) of 1.03 (Supplementary Figure 1). Genotypes from the 1000 Genomes Project data set now offer African data from 78 Yoruban individuals (Nigeria), 67 Luhya in Webuye (Kenya), 24 individuals of African ancestry from Southwest USA and 5 Puerto Ricans. These genotypes were imputed into the Ghanaian dataset using the minimac software.

For association testing, the allelic dosages, which represent the expected number of copies of a distinct allele rather than the best-guess imputed genotypes of each SNP, were analysed in a logistic regression framework in order to account for imputation uncertainty. Adjustment for the population structure was performed with the mach2dat software by including the first three principal components derived from an Eigenstrat analysis of genotype data as covariates (Supplementary Note).

We chose imputed SNPs with minor allele frequencies (MAF) of > 1% and MACH RSQ values of at least 0.3 for further analyses. A MACH RSQ value is a post-imputation quality score not directly related to pairwise linkage disequilibrium measures between SNPs and indicates the correlation between true and estimated allele counts of imputed SNPs⁵. Applying these criteria, 10,921,004 genetic variants were successfully imputed. We selected 46 of these SNPs representing independent signals and yielding after imputation P values of < 1 × 10⁻⁵ and genotyped them using LightTyper assays in the same GWAS dataset.

Two of the 46 variants were mono-allelic in the Ghanaian study population. Eleven variants only provided after genotyping association signals of P < 1 × 10⁻⁵. The low degree of concordance between imputation with genotyping results may have several reasons. First, the data available for imputation consists of preformed and calculated haplotypes from the 1000 Genomes Project African subset under inclusion of individuals of defined ancestry. Optimal would be if the dataset would include the population currently studied. This was not the case here. Second, imputation has previously been shown to be less accurate in African populations⁶. Third, the imputation quality essentially depends on the size of the reference panel which is in our case, with 348 haplotypes of the African subset, suboptimal. Last, the lower the frequency of the occurrence of distinct SNPs, the lower the imputation precision.

The eleven variants with P values < 1 × 10⁻⁵ were tested in a replication sample of 817 TB cases and 3805 controls, constituting a total of 7798 Ghanaian individuals (Supplementary Table 1). Variant rs2057178 on chromosome 11p13 yielded the strongest genome-wide significant association result (P value 2.63 × 10⁻⁹, odds ratio (OR) 0.77, 95% confidence interval [CI] 0.71-0.84) (Table 1). Further genotyping of variants at this locus with imputation P values of < 10⁻⁵ revealed two additional variants, rs11031728 and rs11031731, with genome-wide significant results as well (P = 5.25 × 10⁻⁹, OR = 0.77, 95% CI 0.71-0.84; P = 7.01 × 10⁻⁹, OR = 0.78, 95% CI 0.71-0.85) (Supplementary Table 2). These three variants are in strong linkage disequilibrium in the different ethnic groups represented in the HapMap Project (LD; all pairs r² = 0.98) with each other (Supplementary Figure 2), thus making it virtually impossible to distinguish them with regard to their impact on the infection phenotype. It may be assumed that strong LD applies in other populations as well. Notably, variant rs11031728 is part of the conserved transcription factor binding site V$TCF11MAFG_01 (UCSC, HMR Conserved Transcription Factor Binding Site track, GRCh37 genome assembly). Variant rs2057178, which provided the strongest association signal (Figure 1) was investigated further after confirmation of ethnic homogeneity (Mantel-Haenszel statistic; Supplementary Table 3).

Table 1.

Meta-analysis of SNP rs2057178 in four TB case-control groups

rs2057178 – A allele	Controls		Cases		OR (95% CI)	P value
	N	Freq	N	Freq
Ghana	5636	0.32	2127	0.27	0.77 (0.71-0.84)	2.63E-09
The Gambia	1349	0.31	1207	0.27	0.80 (0.70-0.91)	4.87E-04
Indonesia	983	0.11	1025	0.09	0.84 (0.68-1.03)	0.099
Russia	5874	0.13	4441	0.12	0.91 (0.82-0.99)	0.02
Total	13859		8821			2.57E-11

Figure 1.

Association plot of the chromosome 11 hit region with markers identified by imputation in the Ghanaian data set (red, orange and grey diamonds) and of the meta-analysis (blue diamond). Genes in the long range vicinity are given.

We genotyped SNP rs2057178 in additional TB case-control groups originating from The Gambia (1207 cases vs. 1349 controls; P = 4.9 × 10⁻⁴, OR 0.80, 95% CI 0.70-0.91), Indonesia (1025 cases vs. 983 controls; P = 9.9 × 10⁻², OR 0.84, 95% CI 0.68-1.03) and Russia (4441 cases vs. 5874 controls; P = 2.0 × 10⁻², OR 0.91, 95% CI 0.82-0.99). Results from the Ghanaian study group were corroborated in a meta-analysis including the results of SNP rs2057178 of the four study groups, with a combined P value of 2.57 × 10⁻¹¹ (Table 1, Supplementary Figure 3, Supplementary Note). The consistent effect of rs2057178 in the study populations from West-Africa, Indonesia and Russia, that may have undergone different regional adaption and selection processes, suggests functional relevance of rs2057178 or strong linkage of rs2057178 to a causal variant yet to be identified.

Applying a fixed effect model in the Cochran Q test, the between-study heterogeneity was negligible when comparing the two African study groups (P = 0.67). A result close to significance (P = 0.062) was obtained when testing all four study groups, indicating a certain degree of inter-study heterogeneity between the two African and the two non-African groups.

SNP rs2057178, which showed the strongest association, and the other two SNPs in strong LD with it, rs11031728 and rs11031731, are located in an intergenic region 45 kb downstream of the Wilms' tumour 1 gene (WT1; MIM ID *607102). Whether the associated locus on chromosome 11 affects or even regulates WT1 expression is not clear at present. WT1 is a zinc-finger transcription factor and involved in the development of the urogenital system. Genetic variants of WT1 have been shown to be associated with the occurrence of the Wilms' tumour, but also with acute myeloid leukaemia and the Denys-Drash, Frazier and other syndromes⁷. WT1 also plays a role in the activation of the vitamin D receptor (VDR)⁸ and was, in a mouse model, found to suppress interleukin (IL)10 expression⁹. Both VDR and IL-10 have been claimed to be important in the pathophysiology of TB¹⁰ and genetic variation of the VDR and IL10 genes has been reported to be associated with TB susceptibility¹¹.

Further genes telomeric of the association peak include RCN1, PAX6, ELP4 and IMMP1L and, centromerically from WT1 the genes WIT1, EIF3M and CCDC73. There is no evidence so far of an involvement of any of these genes in the phenotype arising after a M. tuberculosis infection.

More than 10 million common African variants (genotype frequencies of > 1%) have been reported in the 1000 Genomes Project and it is clear that the number of SNPs to be looked at in African groups has increased markedly. Enlargement of the number of SNPs by imputation analyses has proven useful in this study and led to the identification of a new TB locus. With the advent of affordable genome-wide sequencing technologies, more common, but also rare variants will be identified and hopefully unfold new strategies to tackle TB.