HLA class II sequence variants influence tuberculosis risk in populations of European ancestry

Abstract

Mycobacterium tuberculosis (M. tuberculosis) infections cause 9.0 million new tuberculosis (TB) cases and 1.5 million deaths annually¹. To search for sequence variants that confer risk of TB we tested 28.3 million variants identified through whole-genome sequencing of 2,636 Icelanders for association with TB (8,162 cases and 277,643 controls), pulmonary TB (PTB), and M. tuberculosis infection. We found association of three sequence variants in the HLA class II region: rs557011[T] (MAF=40.2%) with M. tuberculosis infection (OR =1.14, P=3.1×10^-13) and PTB (OR=1.25, P=5.8×10^-12) and rs9271378[G] (MAF=32.5%) with PTB (OR=0.78, P=2.5×10^-12), both located between HLA-DQA1 and HLA-DRB1. Finally, a missense variant p.Ala210Thr in HLA-DQA1, (MAF=19.1%, rs9272785) shows association with M. tuberculosis infection (P=9.3×10^-9, OR=1.14). The association of these variants with PTB was replicated in large samples of European ancestry from Russia and Croatia (P< 5.9×10^-4). These findings demonstrate that the HLA class II region contributes to the complex genetic risk of tuberculosis, possibly through reduced presentation of protective M. tuberculosis antigens to T cells.

Mycobacterium tuberculosis (M. tuberculosis) causes 9.0 million new tuberculosis (TB) cases and 1.5 million deaths annually, and M. tuberculosis carriers who are not infected with HIV have a 10% lifetime risk of developing active TB disease¹. The majority of infected individuals control the pathogen by mounting a successful, long-lived immune response, leading to clinically latent infection. Immunological impairment caused by malnutrition, diabetes, HIV/AIDS, aging and smoking, plays a major role in the epidemiology of TB and heritability studies have implicated genetic susceptibility^2,3. It is believed that TB was rare in Iceland until the 19th century when it spread rapidly and reached its peak in the beginning of the nineteen thirties. In the year 1935 approximately 20.9% of 8 year olds and 34.2% of 13 years olds had positive tuberculin skin test (TST⁺), indicating that a substantial part of the population had been exposed to M. tuberculosis^4,5. Since then the incidence of TB in Iceland has decreased to the lowest in Europe⁶.

We imputed 28.3 million single nucleotide polymorphisms (SNPs) and insertions/deletions identified through whole-genome sequencing of 2,636 Icelanders into 104,220 chip-typed Icelanders and their first and second degree relatives⁷. The Icelandic Tuberculosis Data Registry contains TB diagnosis from 1900 to 2010, including confirmed TB diagnoses of 8,162 individuals with genotype information, whereof 3,686 had a confirmed pulmonary TB (PTB) (Table 1). 14,724 individuals with genotype information had contracted M. tuberculosis infection, since in addition to the TB cases, 6,562 individuals who did not develop TB, were recorded TST⁺ and had not been Bacillus Calmette-Guerin (BCG) vaccinated (Supplementary Table 1). Since TST positivity may also be caused by vaccination with BCG, we excluded those who had been BCG vaccinated and considered the remaining TST⁺ subjects as M. tuberculosis-infected. We tested the imputed sequence variants for association with PTB, TB, and M. tuberculosis infection (TB and/or TST⁺) using population controls (N>277,643). Sequence variants were weighed according to their prior probability of affecting gene function by applying thresholds for genome-wide significance that depend on the variant class. We allocated the type I error rate of 0.05 equally between three classes of variants. We tested 5,955 loss of function variants, 157,106 missense variants and 28,113,695 other variants yielding class specific Holm-Bonferroni genome-wide significance thresholds of 2.8 × 10^-6 for loss of function, 1.1 × 10^-7 for missense and 5.9 × 10^-10 for other variants.

Table 1. Characteristics of the Icelandic TB cases.

For both chip typed individuals and family imputed individuals the average year of birth and standard deviation is given as well as percent of individuals alive and percent of males.

	Pulmonary TB N=3,686		Tuberculosis N=8,162		M. tuberculosis infection (including TB) N=14,723
	Chip typed	1st,2nd degree relatives	Chip typed	1st,2nd degree relatives	Chip typed	1st,2nd degree relatives
N cases	1,188	2,498	2,765	5,397	6,105	8,618
Mean year of birth (SD)	1931(12)	1914(16)	1933(13)	1914(17)	1931(12)	1919(17)
Percent alive	53.0%	8.4%	57.4%	10.0%	54.7%	16.2%
Males %	40.7%	48.0%	41.1%	47.5%	42.3%	48.9%

Using these criteria, several sequence variants in the human leukocyte antigen (HLA) region on chromosome 6p21 showed genome-wide significant association (Table 2, Supplementary Tables 2 and 3, Figure 1, Supplementary Figure 1). The strongest association was between rs557011[T] (MAF=40.2%), located between HLA-DQA1 and HLA-DRB1, and M. tuberculosis infection (P=3.1×10^-13, OR=1.14) and PTB (P=5.8×10^-12, OR=1.25). rs9271378[G] (MAF=32.5%), also located between HLA-DQA1 and HLA-DRB1 (Figure 1), associates with reduced risk of PTB (P=2.5×10^-12, OR=0.78). Finally, a missense variant p.Ala210Thr, (MAF=19.1%, rs9272785) in exon 4 of HLA-DQA1 associates with M. tuberculosis infection (P=9.3×10^-9, OR=1.14). p.Ala210Thr corresponds to the classical HLA-DQA1*03 superallele. Three missense variantsp.Thr49Ser (rs1048023), p.Gly79Arg (rs12722072 and rs12722074) and p.Met99Val (rs1064944), that along with p.Ala210Thr define HLA-DQA1*03, did not align to the reference sequence. These variants were called by separately aligning each exon to one of the haplotypes found in the GATK/hla caller database⁸ and imputing for HLA typing (see methods). The variants correlate perfectly (r²>=0.99) with rs9272785/p.Ala210Thr and show identical association (Table 2). The HLA-DQA1*03 superallele (p.Ala210Thr/p.Thr49Ser/p.Gly79Arg/p.Met99Val) and the two non-coding variants are moderately correlated (pairwise r² between 0.11 and 0.35, Supplementary Table 4). The association of each of the two non-coding variants remains significant when conditioned on the other, while p.Ala210Thr remained significant when conditioned on rs9271378 but not rs557011 (Supplementary Table 5). No sequence variant associates significantly with TB disease as a whole, but rs557011 (P=3.7×10^-9, OR=1.15) and rs9272785 (P=2.3×10^-9, OR=0.86) still represent the top signals with consistent direction of effect.

Table 2. Sequence variants showing genome-wide significant association with Tuberculosis in Iceland.

For each sequence variant the reference SNP ID number (rs#), chromosome (Chr), hg18 position, its effect on a gene (Coding effect), minor allele frequency (MAF), gene name and modeled allele are provided in addition to the odds ratio (OR) for tuberculosis and the corresponding P values.

								Pulmonary TB			Tuberculosis			M. tuberculosis infection (including TB)
								N=3,686			N=8,162			N=14,724
rs#	chr	pos	MAF (%)	Info	coding	gene	Effect allele	P	OR	95% CI	P	OR	95% CI	P	OR	95% CI
rs557011	chr6	32694991	40.2	0.997	-	-	T	5.8×10-12	1.25	[1.17-1.33]	3.7×10-9	1.15	[1.10-1.20]	3.1×10-13	1.14	[1.10-1.18]
rs9271378	chr6	32695278	32.5	0.998	-	-	G	2.5×10-12	0.78	[0.73-0.84]	1.2×10-9	0.86	[0.82-0.90]	1.3×10-7	0.90	[0.87-0.94]
rs9272785	chr6	32718379	19.1	0.996	p.Ala210Thr*	HLA-DQA1	A	3.5×10-7	1.22	[1.13-1.32]	7.3×10-4	1.10	[1.04-1.16]	9.3×10-9	1.14	[1.09-1.19]

^*p.Thr49Ser (rs1048023) , p.Gly79Ser (rs12722072) and p.Met99Val (rs1064944) correlate (r²=0.99) with rs9272785/p.Ala210Thr and show identical association.

Figure 1.

Regional association plot of the HLA 6p21 loci for pulmonary tuberculosis (N=3,686), all tuberculosis (N=8,162) and M. tuberculosis infected w/wo TB (N=14,723). The -log₁₀ P values (y axis) of each SNP are presented on the basis of their chromosomal positions (x axis). Only P values below 1×10^-3 are plotted.The horizontal dotted lines represend significance thresholds for missense and non-coding variants.

We followed the three sequence variants identified in Iceland up in PTB samples from Russia (5,530 cases and 5,607 controls) and Croatia (438 cases and 1,009 controls). HLA-DQA1*03 (represented by p.Ala210Thr) and the two non-coding variants associate with PTB in the Russian sample (P<5.4×10^-4) and rs557011 associates with PTB in the smaller Croatian sample (P=0.0074). All three variants show a consistent direction of effects (Table 3). The ORs for rs557011 and rs9272785 are lower in the Russian sample than in Iceland (P<0.011).

Table 3. Associations of the sequence variants from follow up in samples from Russia and Croatia.

Results from combined analysis of Iceland, Russia and Croatia are also provided. For each sequence variant the reference SNP ID number (rs#) and modeled allele are given in addition to the allele frequency in each cohort, the odds ratio (OR) for pulmonary tuberculosis, the corresponding P values and 95% confidence intervals. Significant heterogeneity was observed between the non-Icelandic and Icelandic samples for rs557011 (P=0.016) and rs9271378 (P=0.0024) but not for DQA1*03 (P=0.21). The heterogeneity is driven by the difference between the Icelandic and the Russian samples.

			rs557011[T]				rs9271378[G]				p.Ala210Thr
	#Cases	#Controls	P	OR	95% CI	AF%	P	OR	95% CI	AF%	P	OR	95% CI	AF%
PTB Iceland	3,686	287,427	5.8×10-12	1.25	[1.17-1.33]	40.2	2.5×10-12	0.78	[0.73-0.84]	32.5	3.5×10-7	1.22	[1.13-1.32]	19.1
PTB Russia	5,530	5,607	8.5×10-5	1.12	[1.06-1.19]	34.4	1.9×10-5	0.89	[0.84-0.94]	46.6	5.4×10-4	1.15	[1.06-1.24]	13.9
PTB Croatia	438	1,009	0.0074	1.26	[1.06-1.49]	36.1	0.052*	0.85*	[0.72-1.00]	57.5*	0.66	1.06	[0.82-1.37]	10.9
PTB Russia+Croatia combined	5,968	6,616	4.7×10-6	1.13	[1.07-1.20]	-	3.0×10-6	0.89	[0.84-0.93]	-	5.9×10-4	1.14	[1.06-1.23]	-
PTB all combined	9,654	294,043	2.0×10-15	1.18	[1.13-1.23]	-	3.2×10-15	0.85	[0.82-0.89]	-	1.9×10-9	1.18	[1.12-1.25]	-

^*Pvalue, OR and AF are given for rs113013369 which correlates with rs9271378 (r² = 0.8)

To better understand the effect of the variants we compared PTB and TB cases to M. tuberculosis-infected individuals without TB, as well as non-pulmonary TB cases and M. tuberculosis-infected individuals to population controls (Table 4). HLA-DQA1*03 and rs557011 both associate with M. tuberculosis infection in the absence of TB disease (P=1.1×10^-7 and 2.5×10^-6, respectively), but not with TB disease among the M. tuberculosis-infected (P>0.13). rs557011 associates with PTB (P=0.0035) among the M. tuberculosis-infected while HLA-DQA1*03 does not (P=0.063). This pattern is inversed for rs9271378 which showed no association for M. tuberculosis infection in the absence of tuberculosis (P=0.20), but a significant association with TB (P=0.0035) and PTB (P=3.7×10^-6) among M. tuberculosis-infected. For all of the variants, the association with non-pulmonary TB was weaker than for PTB, possibly because greater phenotypic heterogeneity among non-pulmonary TB cases. Microbiologically confirmed PTB cases and TB cases were additionally tested for association. All three variants associate more strongly with microbiologically confirmed cases (Supplementary Table 6). We use TST-positivity as a surrogate for M. tuberculosis infection, recognizing that it is not a perfect surrogate, since environmental mycobacteria may also cause TST positivity. It was reported that TST reactivity is controlled by the TST1 locus at 11p14 and the TST2 locus at 5p15 in a hyperendemic area in South-Africa ⁹ and TST negativity in a low endemic area in France was recently shown to be controlled by a chromosomal region at 11p14-15 close to TST1 and overlapping with the TNF1 locus¹⁰. We therefore tested the effects of the TB-associated variants reported here with TST negativity (Supplementary Table 1) and found no association (P>0.05).

Table 4. Association of the sequence variants with TB sub-phenotypes.

For each sequence variant the reference SNP ID number (rs#), chromosome (Chr), hg18 position, its effect on a gene (Coding effect), minor allele frequency (MAF), gene name and modeled allele are provided in addition to the odds ratio (OR) for TB and the corresponding P values.

							PTB (N = 3,686) vs M. tuberculosis-infected wo TB (TST+, N = 6,562)			TB (N=8,162) vs M. tuberculosis-infected wo TB (TST+, N = 6,562)			Non-pulmonary TB (N = 4,476)			M. tuberculosis-infected wo TB (TST+, N = 6,562)
rs#	chr	pos	MAF (%)	coding	gene	Effect allele	P	OR	95% CI	P	OR	95% CI	P	OR	95% CI	P	OR	95% CI
rs557011	chr6	32694991	40.2	-	-	T	0.0035	1.12	[1.04-1.21]	0.6	1.02	[0.95-1.10]	0.05	1.06	[1.00-1.12]	2.5×10-6	1.11	[1.06-1.16]
rs9271378	chr6	32695278	32.5	-	-	G	3.7×10-6	0.83	[0.77-0.90]	0.0035	0.91	[0.85-0.97]	0.028	0.94	[0.89-0.99]	0.2	0.97	[0.93-1.02]
rs9272785	chr6	32718379	19.1	p.Ala210Thr	HLA-DQA1	A	0.061	1.10	[1.00-1.22]	0.13	0.94	[0.87-1.02]	0.91	1.00	[1.00-1.00]	1.1×10-7	1.16	[1.10-1.23]

We tested imputed HLA alleles of six of the classical HLA genes; HLA-A, HLA-B, HLA-C, HLA-DQA1, HLA-DQB1 and HLA-DRB1 for association with PTB, TB and M. tuberculosis infection (Supplementary Table 7). Of these, the HLA-DQA1*03:01 allele showed the strongest association. However, the associations of rs9271378 and rs557011 cannot be fully explained by the available HLA alleles (Supplementary Table 8). This suggests either that the association is with an HLA allele that was not well imputed or that the association is not driven by the classical HLA alleles, with the possible exception of HLA-DQA1*03:01. Given that the strongest PTB association is with rs9271378, the possible involvement of other genes in that region cannot be excluded. The TB associated variants correlate with increased or decreased expression of HLA-DQA1, HLA-DQB1 and/or HLA-DRB1 in white blood cells, although other variants in the region are more strongly correlated with the level of expression of these genes, but had weaker association with TB in our study (Supplementary Table 9)¹¹.

SNPs in the ASAP1 gene on chromosome 8q24 have been associated with PTB in the Russian population ¹². However, in Icelanders these SNPs do not associate with PTB (rs10956514[G] P=0.66, OR=0.99, Supplementary Table 10). Previously, a genome-wide association scan in African samples from Ghana and The Gambia yielded common variants associating with PTB: rs4331426[G] (OR=1.19, MAF=44%) at 18q11 and rs2057178[G] located downstream of the WT1 gene on chromosome 11p13 (OR=0.77, MAF=32%) ^13,14. We tested these variants for association in the Icelandic data. Both variants are rarer in Iceland than in the African populations and have smaller effects (Supplementary Table 10). These results resonate with findings of previous TB GWAS in the South African¹⁵ and Russian¹² populations that neither showed TB association with rs4331426 on chromosome 18q11, but detected association of rs2057178 on chromosome 11p13. No SNPs in the HLA region were reported to have a genome-wide significant association with TB in the African samples, the most suggestive associations in the region were found with rs9469220 in HLA-DQA1 (P=0.002, OR=1.15) and rs9272346 (P=0.022, OR=1.10). In the Icelandic data, rs9469220[A] (MAF=38.8%) shows association with PTB (P=1.1×10^-5, OR=0.86,) and correlates weakly with our top PTB signal rs9271378 (r²=0.23). On the other hand, rs9272346[A] does not associate with TB in our data (P>0.21). Lack of replication of previously reported variants maybe due to factors such as phenotypic heterogeneity, BCG vaccination rates, exposure to environmental mycobacteria and the predominant circulating M.tuberculosis strains, that may cause effects of variants to vary between populations, affecting statistical power to detect their association, different LD to causal variants, in addition to the possibility of a spurious initial finding. We show genome-wide significant association of rs557011[T] and rs9271378[G] located between HLA-DQA1 and HLA-DRB1, and rs9272785/p.Ala210Thr in HLA-DQA1 with TB phenotypes and replication in follow up studies. However, it should be kept in mind when attempting to transfer results to other populations that effect sizes are likely to vary based on the case composition.

Our study suggests that allele DQA1*03 may contribute to TB. DQA1*03 associates with several autoimmune diseases. DQA1*03 is a well known risk factor for the gluten sensitive enteropathy coeliac disease (CD) as part of the DQA1*03-DQB1*03:02 haplotype, and trans-DQA1*03:01/DQB1*02:01, encoding the DQ8 and DQ2.3 molecules, respectively¹⁶. DQ8 and DQ2.3 bind gliadin peptide T cells epitopes, providing a functional evidence for the contribution of DQA1*03 to the pathogenesis of the disease^17,18. DQA1*03 also increases the susceptibility to Type 1 Diabetes (T1D)¹⁹ and other autoimmune diseases ^20,21. Many M. tuberculosis -derived epitopes are recognized by HLA-restricted CD4+ and CD8+ T cells in humans infected with M. tuberculosis²². In latently TB-infected individuals the CD4+ T cells recognizing M. tuberculosis epitopes are confined to the CXCR3+CCR6+ Th1 subset²³ and high reactivity is associated with recognition of a few discrete dominant antigenic regions²⁴. The ability of DQA1*03:01-containing HLA molecules to present dominant epitopes of critical protective M. tuberculosis antigens is unknown. Cell-surface expression of HLA-DQ alleles varies extensively, indicating an allelic hierarchy in the intrinsic stability of HLA-DQ molecules²⁵, and DQ8, containing DQA1*03, is among the most unstable DQ molecules. Alterations in disease-associated amino acids located outside of the peptide-binding groove regulate the stability of DQ molecules²⁵. The Ala210Thr mutation is in exon four, encoding the transmembrane region of the DQA1 chain, whereas the p.Thr49Ser, p.Gly79Arg and p.Met99Val missense mutations are in exon 2, in the vicinity of α−β chain dimerization interfaces at each end of the peptide binding groove^26,27, and may affect peptide binding. Furthermore, the DQ8 molecule containing DQA1*03 interacts poorly with HLA-DM (DM), that plays a critical role in peptide loading during antigen presentation, resulting in reduced loading and DM-editing of antigenic peptides^26,28. Surface expression HLA class II peptide complexes on antigen presenting cells is also regulated by ubiquitination; their assembly, endocytosis, recycling and turnover²⁹. The non-coding variants rs557011 and rs9271378 do not overlap with known biologically relevant regions (Supplementary Notes). However, a correlated marker rs1846190 (r²=0.81 with rs557011) is located in an enhancer site in CD4+ CD25- IL17+ T cells³⁰ and CTCF binding site which seems to regulate HLA-DRB1 expression in lymphoblastoid cell lines³¹. It is conceivable that the contribution of DQA1*03 and the missense mutations to the risk of M. tuberculosis infection and TB disease is through reduced stability of DQA1*03 containing molecules, and poor presentation of critical M. tuberculosis antigens, resulting in poor activation of protective T cells. The effects of the TB-associated variants on mRNA of DQA1*03 and its surface expression in antigen presenting cells have not been studied.

The HLA region plays a key role in immune responses and associates with infections³² and autoimmune^33,34 diseases and has been the focus of many tuberculosis candidate gene studies yielding conflicting results³⁵. We found sequence variants, located in HLA-DQA1 and between HLA-DQA1 and HLA-DRB1 that associate with TB. rs557011 associates with PTB and seems to confer susceptibility to both M. tuberculosis infection and risk of development of TB disease. On the contrary, rs9271378 does not associate with M. tuberculosis infection per se, but protects against development of TB disease in M. tuberculosis-infected individuals. For all variants the effects on PTB were the strongest.

Materials and Methods

The Icelandic discovery population

Individuals who had been diagnosed with TB or been infected with M. tuberculosis (TST positive) without developing TB during the 20^th century, according to the Icelandic Tuberculosis Database (ITBDB) were invited to participate in the study. The ITBDB contains information on TB diagnosis, major and minor sites of infection based on M. tuberculosis culture and microscopic analysis, histology and roentgen result. It has also records on number and duration of TB episodes and hospitalizations, symptoms and signs, treatment and outcome and family history, as well as nationwide results of TST testing and BCG (see Supplementary Notes). In this study we used genotype data for 3,686 patients with confirmed pulmonary TB, 8,162 with any TB and 14,723 with M. tuberculosis infection with or without developing TB, using TST positivity as a surrogate for having been M. tuberculosis infected (excluding BCG vaccinated individuals) and ≥ 277,643 controls (remaining chip-typed Icelanders and their relatives).

The study was approved by the Icelandic Data Protection Authority (ref. 2004120649) and the National Bioethics Committee (ref. VSN 04-172 VSNb2004120008-03-1). All participating subjects who donated blood signed informed consent. Personal identities of the participants and biological samples were encrypted by a third party system approved and monitored by the Icelandic Data Protection Authority.

The Russian TB sample set

The study in the Russian cohort has been done as described previously¹². Briefly, TB patients have been diagnosed using information about TB contact, medical history and clinical symptoms, presence of acid fast bacilli in sputum smear and symptoms characteristic of pulmonary TB on chest X rays. For all patients diagnosis has been confirmed by culture of M. tuberculosis from sputum. Patients with extra-pulmonary TB and all HIV-positive subjects were excluded. Controls were healthy adult blood bank donors with no history of TB. M. tuberculosis infection status of these controls was unknown.

The Croatian TB sample set

DNA was isolated from blood samples of PTB patients (N=244) and contact controls (n=85) treated at the Section of Pulmology, Department of Internal Medicine, Clinical Hospital Centre Rijeka, Rijeka and Hospital for Lung Diseases “Jordanovac”, University Hospital Center “Zagreb”, and healthy blood donors (n=924) collected at Departments of Transfusion Medicine in Rijeka and in Zagreb, Croatia, as previously described^36,37. DNA samples from the blood of additional 194 PTB patients, treated in Hospital for Lung Disease “Jordanovac”, University Hospital Center “Zagreb”, Croatia, were isolated at deCODE, Reykjavik, Iceland. The age and gender frequency of the latter group of patients was not significantly different from the former cohort of cases. All study subjects provided oral and written informed consent. The Medical Research Council ethics committees at Zagreb and Rijeka approved the research.

Illumina chip genotyping

Icelandic chip-typed samples were assayed using the Illumina HumanHap300, HumanCNV370, HumanHap610, HumanHap1M, HumanHap660, Omni-1, Omni 2.5 or Omni Express bead chips at deCODE genetics. SNPs were excluded if they had (i) yield less than 95%, (ii) minor allele frequency less than 1% in the population or (iii) significant deviation from Hardy-Weinberg equilibrium in the controls (P <0.001), (iv) if they produced an excessive inheritance error rate (over 0.001), or (v) if there was substantial difference in allele frequency between chip types (from just a single chip if that resolved all differences, but from all chips otherwise). All samples with a call rate below 97% were excluded from the analysis. For the HumanHap series of chips, 304,937 SNPs were used for long range phasing, whereas for the Omni series of chips 564,196 SNPs were included. The final set of SNPs used for long-range phasing was composed of 707,525 SNPs.

Genotyping of the Russian PTB cases and controls has been done using Affymetrix Genome-Wide Human SNP Array 6.0 and SNPs across the genome were imputed as described previously¹². Association analysis of the HLA SNPs rs557011, rs9271378 and rs9272785 has been done using PLINK³⁸.

Whole genome sequencing

Whole genome sequencing was performed for 2,636 Icelanders, selected for various conditions. All of the individuals were sequenced at a depth of at least 10X (average sequencing depth = 22X).

Template DNA fragments were hybridized to the surface of flow cells (GA PE cluster kit (v2) or HiSeq PE cluster kits (v2.5 or v3)) and amplified to form clusters using the Illumina cBot. In brief, DNA (2.512 pM) was denatured, followed by hybridization to grafted adaptors on the flow cell. Isothermal bridge amplification using Phusion polymerase was then followed by linearization of the bridged DNA, denaturation, blocking of 3’ ends and hybridization of the sequencing primer. Sequencing-by-synthesis (SBS) was performed on Illumina GAIIx and/or HiSeq 2000 instruments. Paired-end libraries were sequenced at 2 x 101 (HiSeq) or 2 x 120 (GAIIx) cycles of incorporation and imaging using the appropriate TruSeq™ SBS kits. Each library or sample was initially run on a single GAIIx lane for QC validation followed by further sequencing on either GAIIx (³ 4 lanes) or HiSeq (³ 1 lane) with targeted raw cluster densities of 500800 k/mm2, depending on the version of the data imaging and analysis packages (SCS2.6-2-9/RTA1.6-1.9, HCS1.3.8-1.4.8/RTA1.10.36-1.12.4.2). Real-time analysis involved conversion of image data to base-calling in real-time.

Generation of whole-genome genotype data

SNP and INDEL calling from the whole-genome sequence data of the 2,230 Icelanders and generation of imputed genotypes has been described previously³⁹ (Supplementary Notes). Briefly, the genotypes identified were imputed into chip genotyped and long range phased Icelanders. Probabilities of genotypes were furthermore predicted for relatives of chip-typed individuals.

HLA typing

For each of the six MHC genes, the most common alleles present in the Icelandic population were selected from the allelefrequencies database, using the most ethnically related populations: Norway and Ireland. The exonic sequence of the alleles were downloaded from Broad’s HLA reference.

We genotyped “in silico” a set of 2615 whole genome sequenced (WGS) individuals. For each sequenced individual we selected reads that either: 1) mapped to the public reference sequence using BWA or 2) were unmapped by BWA and could be aligned to one of the haplotypes found in the GATK/hla caller database.

For every gene genotyped we align each read in this set to each exon separately. We consider a read to belong to a given haplotype if: 1) The read can be aligned to the exon from GATK with no mismatches or indels, allowing for the possibility that the read only partially overlaps the exon as long as the overlap is at least 40 basepairs and the overlap does not introduce a mismatch or an indel. 2) The mate of this read can be aligned to some sequences occurring within +/- 1000 basepairs from the exon using BWA default parameters.

If a read r aligns to an exon of allele A we arbitrarily say that P(r|A) = 1-PE and if r does not align to A we set P(r|A) = PE. We arbitrarily choose PE = 0.001 and have P(r|A) = 0.9999 for reads r that align to the superallele A and P(r|A) = 0.001 for reads r that do not align to the superallele A. If we let R be the set of reads and assume independence of reads and the two alleles carried by an individual we can then compute $P(R|A1,A2)={\prod }_{r\in R}(\frac{1}{2}P(r|A1)+\frac{1}{2}P(r|A2)).$ The HLA alleles were imputed into the Icelandic sample set as previously described for genotypes³⁹.

To check the accuracy of the imputation at least three individuals carrying each haplotype were HLA typed for the 6 genes using All-Set TM Gold SSP (Life Technologies, DQA1, DQB1, DRB1, HLA-A high resolution typing; HLA-B and HLA-C low resolution typing). Accuracy between imputation and wet-lab genotyping was 90%-99% and frequency weighted correlation was 95-99.6%. (Supplementary Table 12, Accuracy was 0.989 for DQA1)

Genotyping of single variants

Single SNP genotyping of rs9271378 and rs9272785 in the Croatian sample set was carried out by deCODE Genetics in Reykjavik, Iceland, applying the Centaurus (Nanogen) platform⁴⁰ . Sanger sequencing of rs557011 in the Croatian sample set was performed by deCODE Genetics.

Association testing

Logistic regression was used to test for association between sequence variants and disease (tuberculosis), treating disease status as the response and genotype counts as covariates. Other available individual characteristics that correlate with disease status were also included in the model as nuisance variables. These characteristics were: Sex, county of birth, current age or age at death (first and second order terms included), blood sample availability for the individual and an indicator function for the overlap of the lifetime of the individual with the timespan of phenotype collection (Supplementary Notes). Conditional analysis was performed by including the sequence variant being conditioned on as a covariate in the model under the null and the alternative in the generalized linear regression.

Principal component analysis

We calculated principal components using the EIGENSTRAT software⁴¹ (Supplementary Figure 2). In the Russian population, the first four principal components were included in the association analysis (Supplementary Table 11). The first 20 principal components only explain 0.6% of the variance of the Icelandic genotype data and the first two components correlate strongly with the county of birth, already included in the association test⁴². Including the first five principal components in the association testing of the chip typed Icelandic samples had a minimal effect (Supplementary Table 3) and principal components were not included in the final Icelandic association testing which also included relatives of the chip typed individuals.

Genotype imputation information

The informativeness of genotype imputation was estimated by the ratio of the variance of imputed expected allele counts and the variance of the actual allele counts:

$$\frac{Var(E(\theta |chipdata))}{Var(\theta )},$$

where θ is the allele count. Var(E(θ|chip data)) was estimated by the observed variance of the imputed expected counts and Var(θ) was estimated by p(1 − p), where p is the allele frequency. Sequence variants with imputation information below 0.8 were excluded from the analysis.

Gene and variant annotation

For the annotation of the exome data coordinates of variants were converted between hg18 and hg19 using the liftOver tool from UCSC⁴³. Variants in hg19 coordinates were annotated with information from Ensembl release 70 using Variant Effect Predictor (VEP) version 2.8 ⁴⁴. Only protein coding transcripts from RefSeq Release 56⁴⁵ were considered.

Correction for relatedness of the Icelandic subjects and genomic control

Individuals in both the Icelandic case and control groups are related, causing the χ² test statistic to have a mean > 1 and median > 0.675. We estimated the inflation factor λ_g based on a subset of about 300,000 common variants and the P values adjusted by dividing the corresponding χ² values by this factor to adjust for both relatedness and potential population stratification⁴⁶.

Thresholds for genome-wide significance

We weighed sequence variants according to their prior probability of affecting gene function by applying thresholds for genome-wide significance that depend on the variant class. The Bonferroni correction for multiple testing can be adjusted to account for prior importance of sequence variants. We performed a weighted Holm-Bonferroni correction based on giving equal weight to the classes of LoF, MSSNS, and other variants ⁴⁷. For example sequence variants in the LoF class get weight $\frac{1}{3\cdot 6,476}.$ The sum of these weights over all the variants in the genome is 1 and the Bonferroni threshold for significance within a class containing m sequence variants will be $\frac{0.05}{3m}.$

Expression analysis

Samples of RNA from human peripheral blood were hybridized to Agilent Technologies Human 25K microarrays as described previously¹¹. We quantified expression changes between two samples as the mean logarithm (log₁₀) expression ratio (MLR) compared with a reference pool RNA sample. In comparing expression levels between groups of individuals with different genotypes, we denoted the expression level for each genotype as 10^{(average MLR)}, where the MLR is averaged over individuals with the particular genotype. We determined s.e.m. and significance by regressing the MLR values against the number of risk alleles carried. We took into account the effects of age, sex and differential cell type count in blood as explanatory variables in the regression. P-values were adjusted for familial relatedness of the individuals by simulation.

Assessment for potential overlap with regulatory regions

To identify TB associated variants that might have regulatory effects. We took the strongest non-coding signals (rs557011 and rs9271378) and identified all SNPs in LD at r2 > 0.8 (excluding SNPs with low imputation information values). This added rs1846190 and rs508318 (both r2=0.81 with rs557011). For each of the 4 variants identified, we searched for overlaps with known regulatory regions as follows: 1) First we used ENSEMBL to determine whether the variant had been assigned a regulatory region ENSR number. Then we examined the ENCODE data and looked for any evidence of ChIP-Seq transcription factor binding and DNaseI hypersensitivity sites⁴⁸. 2) We also looked for enhancer and promoter chromatin segmentation states using the 25 state HMM from the Roadmap consortium³⁰. 3) Then we looked for correlations between DNaseI hypersensitive sites and local gene expression using results described by Sheffield et al³¹. 4) We examined SiPhy and GERP conservation scores^49,50 and 5) we viewed the Factorbook and HaploReg_v3 data^51,52 to search for potential changes in transcription factor binding motifs with ChIP-Seq evidence of the cognate transcription factor.