Genetic Susceptibility to Tuberculosis Associated with Cathepsin Z Haplotype in a Ugandan Household Contact Study

Abstract

Tuberculosis (TB), caused by Mycobacterium tuberculosis (Mtb), causes 9 million new cases worldwide and 2 million deaths annually. Genetic linkage and association analyses have suggested several chromosomal regions and candidate genes involved in TB susceptibility. This study examines the association of TB disease susceptibility with a selection of biologically relevant genes on regions on chromosomes 7 (IL6 and CARD11) and 20 (CTSZ and MC3R), and fine mapping of the chromosome 7p22-p21 region, identified through our genome scan. We analyzed 565 individuals from Kampala, Uganda who were previously included in our genome-wide linkage scan. Association analyses were conducted for 1417 single-nucleotide polymorphisms (SNPs) that passed quality control. None of the candidate gene or fine mapping SNPs were found significantly associated with TB susceptibility (P > 0.10). When we restricted the analysis to HIV-negative individuals, two SNPs on chromosome 7 were significantly associated with TB susceptibility (P < 0.05). Haplotype analyses identified a significant risk haplotype in Cathepsin X (CTSZ) (p=0.0281, OR = 1.5493, 95% CI [1.039, 2.320]).

1. Introduction

Tuberculosis (TB) is an infectious disease caused by the bacterium Mycobacterium tuberculosis (Mtb). The World Health Organization (WHO) reports that over 9 million new cases of TB are estimated each year, resulting in approximately 2 million deaths worldwide [1]. Only 5–10% of individuals infected with Mtb actually advance to disease [2]. The pathogenesis of TB follows a two-stage process: a productive infection of Mtb whereby symptoms do not develop, followed by Mtb replication and the expression of disease symptoms [3]. TB disease is characterized by growth of Mtb on culture, presence of cavities on chest x-ray, and symptoms such as cough and fever.

Several arguments have been made for a genetic risk factor in development of TB, based on the observation that only a small percentage of individuals infected with Mtb progress to disease development. Although results remain somewhat inconsistent, animal models [4,5], twin studies [6–8], segregation analysis [9], candidate gene studies (reviewed by [10]), and linkage analysis [11–15], have all found evidence in support of a genetic component in the risk for developing TB. Stein et al. [16] conducted a genome-wide linkage analysis of Ugandan individuals, including both HIV-negative and HIV-positive individuals. Suggestive linkage to TB disease was found on a 34-cM long segment on chromosome 7 (P = 0.0002), in addition to a 25-cM long region on chromosome 20 (P = 0.002)[16]. Chromosome 7p22-7p21 contains the IL6 and CARD11 genes. IL6 is an immunoregulatory cytokine that inhibits production of TNF-α and IL-1β, and thus may have a role in the response to mycobacterial infections [17]. IL6-deficient mice succumb to TB infection while wild-type mice do not [18]. However, Oral et al. [19] did not find significant differences in the distribution of the IL6 gene polymorphisms or differences in IL6 allele frequencies between human TB cases and controls. Such contradictory evidence merits further pursuit of the IL6 gene’s involvement in TB susceptibility. Further upstream in this chromosome 7 region is the gene CARD11 (caspase recruitment domain family, member 11), which is part of the NOD-like receptor (NLR) pathway. This gene is of interest because NLRs have non-redundant roles in Mtb recognition [20]. Though both IL6 and CARD11 are promising candidates, it is possible that a novel locus exists under the linkage peak reported by Stein et al. [16].

The same region on chromosome 20q13 observed by Stein et al. [16] was found to be a major susceptibility locus for TB in a study of South African and Malawian sibling pairs, HIV-negative and HIV-positive cases included [21]. Two genes in this chromosome 20 region, melanocortin 3 receptor (MC3R) and cathepsin Z (CTSZ), were mapped in South African and Malawian populations; these presented new candidate genes for TB [21]. Both chromosome 20 genes are biologically relevant to TB susceptibility, as MC3R plays a suggested role in the regulation of energy homeostasis, while CTSZ is expressed in cancer cell lines with possible involvement in host defense and tumorigenesis [22].

The twin goals of this study were (1) to examine the association of TB susceptibility and the biologically relevant genes IL6, CARD11, MC3R and CTSZ; and (2) to fine map the 1-LOD significance region on chromosome 7 identified through our previous genome scan.

2. Material and Methods

2.1 Data description

Study participants were enrolled in the Household Contact Study (Phase I [23]), between 1995 and 1999, and the Kawempe Community Health Study, enrolling between 2002–2004 (Phase II [24]), in Kampala, Uganda. The individuals included in this analysis are those that were analyzed in our genome scan [16] and who had sufficient DNA remaining for additional genotyping. Of the individuals reported in that study, 18 have since developed culture-confirmed TB and 1 has since developed HIV infection; their trait data was recoded accordingly for this analysis.

A complete summary of study design and enrollment procedures can be found in our previous work [16,23,24]. Briefly, households were identified through an index case (proband) with culture-confirmed TB. All individuals within the household who provided informed consent were included in the study; individuals were not excluded based on HIV infection or relationship to the index case. All participants received a full clinical examination, including HIV testing, tuberculin skin test (TST), and sputum collection for TB suspects. Individuals with confirmed TB disease were treated appropriately [25]; in addition, individuals with latent Mtb infection were also offered preventative therapy [26]. Households were followed actively for two years after enrollment; after that two year period, individuals that developed symptoms of TB returned to the clinic for further evaluation and treatment. One of the strengths of this prospective design is the ability to observe all clinically relevant events, including both primary progressive and reactivation disease, as well as tuberculin skin test conversion. This long duration of follow-up significantly reduces the chances of misclassifying cases as controls [27]. This family-based design also has several advantages for genetic association studies [28]; in addition to addressing genetic epidemiological concerns such as population stratification, the family-based design enables the characterization of exposure to an infectious TB case [29].

2.2 Genotyping

Genotyping was performed using a custom Illumina GoldenGate SNP panel comprising 1,536 SNPs with 1,528 from chromosome 7 (1479 in the fine mapping region) and 8 from chromosome 20 (Supplemental Table 1). We first selected SNPs in our four candidate genes of interest: IL6, CARD11, CTSZ, and MC3R. Haplotype tagging SNPs were selected for IL6, CTSZ, and MC3R using all three African HapMap reference populations [30,31] with an r² threshold of 0.8; these were selected using the tagger application of the Genome Variation Server (SeattleSNPs Program for Genomic Applications PGA, 2009; http://gvs.gs.washington.edu/GVS/index.jsp). Because of the large size of CARD11, rather than choose tag SNPs, we genotyped SNPs approximately 8 kb apart. Second, we identified SNPs for fine mapping the chromosome 7 region. A 1-LOD drop region [32] of the chromosome 7 linkage peak reported by Stein et al. [16] was covered by genotyping SNPs approximately every 11 kb. In order for a SNP to be included on the custom array, it had to meet both a minor allele frequency (MAF) threshold of 5% and SNP score quality criterion (Illumina SNP quality score > 0.6). SNP quality scores were determined by Illumina and are based on the probability of success of the assay and validation of the SNP in at least two populations.

Prior to analysis, Mendelian inconsistencies were removed using MARKERINFO (S.A.G.E. v6.0). Call rates by plate were considered as an additional measure of quality control (QC) and any SNPs that did not meet the call rate of 90% were excluded. Also, signal intensities were verified as falling into three distinct genotype groups, and thus no SNPs were lost due to inadequate signaling. Based on these QC measures, a total of 119 SNPs were excluded from analyses (Supplementary Table 1). In addition, individual samples with call rates below 90% were excluded from the analysis.

2.3 Statistical analysis

Demographic factors included age, sex, HIV status, and presence of BCG scar. Significant differences between individuals with and without TB were examined using the Pearson Chi-square for categorical variables, and Mann-Whitney test for age since age was non-normally distributed. Prior to conducting association analysis, we examined the SNP genotypes for departure from Hardy-Weinberg proportions using Haploview [33]. Genetic association analyses were conducted using a generalized linear mixed model (GLMM), in SAS’s PROC GLIMMIX procedure (Cary software, North Carolina SAS Institute Inc. 2004), with each individual included in the analysis model. Since TB is a binary trait, we used a logit link function. To account for familial relationships, we used a generalized estimating equation (GEE) framework and applied an exchangeable correlation structure within each pedigree, assuming that all individuals had a correlation of ρ = 1 with themselves, and a correlation ρ with any other member in their family, i.e. the G covariance matrix, defined as

$$G=\left(\begin{array}{ccc}1& & \rho \\ & \ddots & \\ \rho & & 1\end{array}\right).$$

Each SNP was entered into the GLMM individually, along with the three covariates, HIV status, sex, and dichotomized age (< five years or ≥ to five years) based on results by Lewinsohn et al. [34]. SNP genotypes were coded in an additive fashion to represent the number of minor (risk) alleles and were analyzed one at a time in the analysis, along with the aforementioned covariates. This GEE model approach to examining genetic association to TB has been used in other family-based studies [35–37]. Also, this method allows for the inclusion of all individuals, both related and unrelated, into the analysis, thereby maximizing power [28],

In addition, we examined haplotypes in IL6, CTSZ, and MC3R; because of the large size and low SNP density in CARD11, we did not examine haplotypes for this gene. Based on an r² of 0.8, haplotypes were constructed in DECIPHER (S.A.G.E. v6.0). Four haplotype blocks were identified in the analysis: 1 in CTSZ, 1 in MC3R, and 2 blocks in the IL6 gene. The most likely haplotype phase was used for each individual in the analysis. Haplotypes were then treated similar to SNP genotypes in the GLMM analysis, coded in an additive fashion. Rare haplotypes with frequency < 10% in the sample were pooled for analysis.

Finally, because some of our associated SNPs were identified in regions known to contain copy number variants (CNVs) [38], we investigated the Illumina intensity data for CNVs using PennCNV [39] and the default parameter settings. PennCNV infers DNA copy number from SNP intensity data via a hidden Markov model, in which overall intensity and the individual allele intensities reflect an underlying, unknown copy number “state”.

3. Results and Discussion

After removal of samples with genotyping call rates less than 90%, 565 individuals with complete genotype data were included from both Phase I and Phase II. The sample comprised 318 females (56.3%) and 247 males (43.7%); 429 (82.5%) individuals were HIV-negative while 91 (17.5%) individuals were HIV-positive, and the other individuals’ HIV serostatuses were unknown (Table 1) because HIV testing was only conducted in adult and children who had HIV-positive mothers. A total of 135 (23.9%) individuals had culture-confirmed TB. Of the individuals without TB, 271 individuals had a positive tuberculin skin test (TST) without TB disease either at baseline or converted to TST positive sometime during study follow-up (48% of total sample). The sample comprised 243 pedigrees, including 73 singletons, 230 parent-offspring pairs, and 32 sibling pairs, with a mean family size of 5.08 individuals, and a standard deviation of 5.87. The median age was 16 years. TB patients were significantly older (median age = 28) than unaffected individuals (median age = 12) (Mann Whitney P < 0.0001).

Table 1.

Descriptive Statistics for Analysis Sample (% within category)

Category	Total (% of total sample)	TB	No TB	p-valuea
Male	247 (44%)	72 (29%)	175 (71%)
Female	318 (56%)	63 (20%)	255 (80%)	< 0.0001
HIV−	429 (83%)	77 (18%)	352 (82%)
HIV+	91 (17%)	58 (64%)	33 (36%)	0.0098
BCG Scar	326 (67%)	62 (19%)	264 (81%)
No BCG Scarb	140 (30%)	45 (32%)	95 (68%)	0.0020
Total	565 (100%)	135 (24%)	430 (76%)
Age Range		<1 – 53	<1 – 65	< 0.0001c

^aPearson chi-squared test for difference in proportion of TB patients within category

^bBCG scar unclear for some individuals, so numbers do not sum to 100%

^cDifference in age distributions in TB patients vs. healthy individuals using Mann-Whitney test

Departure from Hardy-Weinberg equilibrium (HWE) in the control individuals (without active TB) was found in 23 SNPs (P < 0.0001, data not shown). None of the candidate gene SNPs were significantly associated with TB susceptibility (P > 0.10). For the fine mapping analysis, a total of 1,479 SNPs were genotyped across the 17.84-Mb region on chromosome 7, and of these 1,367 met the 90% call rate threshold. No significant associations with TB susceptibility were found with these SNPs (P > 0.10) (Table 2). Both the candidate gene analysis and fine mapping analysis were repeated using only the 429 HIV-negative individuals. Two SNPs, rs10233991 and rs12700594, were associated with TB susceptibility at the α = 0.05 level, with nominal P = 0.0308 and P = 0.0390, respectively; rs12700594 is approximately 107 kb away from CARD11. Three additional SNPs had P-values between 0.05 and 0.10. It is likely that more significant associations are seen in the HIV negative subset because these individuals do not have a compromised immune system simply due to HIV; in other words, HIV infection may confound the effects of genotype at certain loci. We could not examine the HIV-positive subset of individuals due to a reduced sample of only 91 such individuals, limiting our analysis.

Table 2.

Single-SNP association analysis results on chromosome 7 with P < 0.10 in entire sample and HIV-negative subsample

			Entire Sample		HIV- Subset
SNP	SNP location	Genea	Odds ratio	P-value	Odds ratio	P-value
rs10233991	170805		0.9983	0.36	0.9945	0.031
rs12700594	3123227		0.9988	0.51	0.9949	0.039
rs13437998	13838749		0.9991	0.61	0.9954	0.073
rs7811444	1562722	TMEM184A	0.9994	0.75	0.9960	0.083
rs7783310	1712658		0.9991	0.63	0.9957	0.093

^aGenes containing SNPs within 500kb were identified using SNPDoc (Guy et al, Submitted)

We then conducted haplotype analyses in both the full sample as well as the HIV-negative subset. CTSZ haplotypes were constructed using the following SNPs: rs10369, rs9760, rs163790, rs163800, and rs163801. Haplotype analyses identified a significant association within the CTSZ gene (Table 3). The frequency of haplotype “GAGGG” was significantly greater in cases than in controls, adjusting for HIV status, age, and sex (P = 0.0281); this haplotype had a frequency of 21.0% in the full sample. Also, the number of copies of the rare haplotypes (frequencies less than 10%) were included in one indicator variable, and this variable was also significant, P = 0.0354. This suggests that at least one rare haplotype is associated with increased risk for TB, but because of our limited sample size, we did not have the statistical power to dissect which haplotype(s) was/were the risk haplotype(s); such examination of rare variants is a contemporary challenge in genetic epidemiology [40]. When only HIV-negative individuals were analyzed, the frequency of rare haplotypes was significantly greater in cases than controls (P = 0.0128) though the GAGGG haplotype was no longer significantly associated (P = 0.2436). Haplotype analyses of MC3R and IL6 did not yield statistically significant associations at the α = 0.05 level (data not shown).

Table 3.

Results of CTSZ haplotype association analysis in entire sample and HIV-subsample

Haplotype (% of entire sample with haplotype)	Entire Sample		HIV- Subset
	Odds ratio	P-value	Odds ratio	P-value
“GAGGG” (20.7%)	1.5493	0.0281	1.3644	0.2436
All rare (pooled) (11.4%)	1.6809	0.0354	2.1514	0.0128

CTSZ haplotype includes rs10369, rs9760, rs163790, rs163800, and rs163801.

From these analyses, none of the candidate gene SNPs or fine mapping SNPs were significantly associated with TB susceptibility at a significance threshold accounting for multiple comparisons (for candidate genes P < 0.0125, and for fine mapping P < 4.13 × 10⁻⁵). It is possible that significant SNP associations were not identified underlying the linkage peaks reported by Stein et al. [16] because the exact familial correlations could not be integrated into the generalized linear model, thus correlations between full siblings were most likely underestimated. Also, linkage analysis is more powerful for detecting rare variants with strong effect sizes, while association analysis is more powerful for detecting common variants [41], therefore, it is possible that our genome scan points to a rare variant (or multiple rare variants) on chromosome 7. It is also conceivable that the chromosome 7 linkage identified in our whole genome scan was a false-positive finding, particularly because the trait status in 18 individuals changed [27].

Replication of a single-SNP between CTSZ and TB, as originally reported by Cooke et al. [21] was not successful. However, this significant association with CTSZ was based upon the analysis of a single SNP, rs34069356. This SNP did not meet design criteria for the GoldenGate assay and thus was not included in the panel of markers tested here. Thus, direct replication using published guidelines [42] of the Cooke et al. analysis of this SNP was not possible, although we did identify nominally significant associations between CTSZ haplotypes and TB. It is possible that the risk allele reported by Cooke et al. [21] resides within these haplotypes. We could not evaluate this hypothesis via imputation, though, because rs34069356 is not included in the HapMap III set of SNPs. In general, our results support a role for CTSZ as a TB susceptibility gene, and further research in this area is warranted. Recently, a study in an independent Cape Town, South Africa study population both replicated the single SNP association of Cooke et al. and also found CTSZ haplotypes associated with TB (Adams et al. [49]). Future analyses will examine all of these CTSZ SNPs, including the SNP reported by Cooke et al., in a larger, independent dataset in the future.

The significant associations of the MC3R SNPs, rs3746619 and rs3827103, with TB susceptibility as reported by Cooke et al. [21] were also not replicated. Another potential reason for the non-replication between the present study and Cooke et al. could be differences in population genetics. The original association with CTSZ and MC3R were found in a South African population of fully independent sibling pairs, including “Coloureds”, South Africans of mixed ancestry, but a case-control analysis in a West African sample within the same report failed to replicate the association. Based on our previous linkage disequilibrium (LD) and haplotype analyses, we found that the Ugandan population carries genetic distinctions from other African populations, including novel polymorphisms and LD structure in a different South African sample [43]. Therefore, it is not unexpected that the Cooke et al. [21] results were not replicated. In addition, we did not find associations with IL6 as reported by Oral et al. [19]. IL6 plays a role in TB immunopathogenesis, so the role of host genetics for this gene is not clear.

Our fine mapping results do not point to any clear candidates for causal variants. As seen in Table 2, many of these SNPs are not within 500 kb of a gene coding region. It is possible that these SNPs lie in regulatory regions, similar to those identified through eQTL studies [44]. Many trait-associated SNPs (> 40%) are intergenic or intronic [45], therefore it is not surprising that we identified TB-associated SNPs in a potential gene desert. Interestingly, rs13437998 is approximately 1 megabase away from rs7787531, which was an imputed SNP that attained P = 8 × 10⁻⁶, reported by a recent genome-wide association analysis of TB in populations from Ghana and the Gambia [46]. Additionally, two SNPs, rs10233991 and rs13437998, reside in regions containing copy number variants (CNVs) [38], thus we examined the Illumina intensity data for CNVs.. We found that rs13437998 is in a CNV region with 3 copies (compared to the 2 expected when there are no CNVs) (data not shown). Structural variants may be relevant for TB, since other studies have shown an association between CCL3L1 copy number and HIV infection and progression [47]. CNVs may be in LD with SNPs, as a result an observed SNP association may actually be due to a CNV [48].

This study followed the previously reported linkage signals and their significant linkage to TB susceptibility in a genome-wide scan [16]. Therefore, levels of significance in these association tests should be interpreted with caution. P-values may not be the best determinant of significant association to TB susceptibility; with a purpose to locate and identify the most likely point in a region already identified by linkage, a nominal P-value may be inappropriate given prior probability of association in that region. A major limitation to consider is the finite sample size of only 564 individuals in families, which was underpowered to detect small effect sizes. Another limitation is the spacing of the fine mapping SNPs; 11kb spacing may not be dense enough to detect untyped loci of small effect in LD with our SNPs.

Abbreviations used

CARD11

caspase recruirtment domain family, member 11

CNV

copy number variant

CTSZ

cathepsin Z

GLMM

generalized linear mixed model

HIV

human immunodeficiency virus

HWE

Hardy-Weinberg Equilibrium

IL6

Interleukin-6

LD

linkage disequilibrium

MC3R

melanocortin 3 receptor

Mtb

Mycobacterium tuberculosis

QC

quality control

SNP

single nucleotide polymorphism

TB

Tuberculosis

TNF-α

tumor necrosis factor-α

WHO

World Health Organization