CIITA variation in the presence of HLA-DRB1*1501 increases risk for multiple sclerosis

Abstract

The MHC class II transactivator gene (CIITA) is an important transcription factor regulating gene required for HLA class II MHC-restricted antigen presentation. Association with HLA class II variation, particularly HLA-DRB1*1501, has been well-established for multiple sclerosis (MS). In addition, the −168A/G CIITA promoter variant (rs3087456) has been reported to be associated with MS. Thus, a multi-stage investigation of variation within CIITA, DRB1*1501 and MS was undertaken in 6108 individuals. In stage 1, 24 SNPs within CIITA were genotyped in 1320 cases and 1363 controls (n = 2683). Rs4774 (missense +1614G/C; G500A) was associated with MS (P = 4.9 × 10⁻³), particularly in DRB1*1501 +individuals (P = 1 × 10⁻⁴). No association was observed for the −168A/G promoter variant. In stage 2, rs4774 was genotyped in 973 extended families; rs4774*C was also associated with increased risk for MS in DRB1*1501+ families (P = 2.3 × 10⁻²). In a third analysis, rs4774 was tested in cases and controls (stage 1) combined with one case per family (stage 2) for increased power. Rs4774*C was associated with MS (P = 1 × 10⁻³), particularly in DRB1*1501+ cases and controls (P = 1 × 10⁻⁴). Results obtained from logistic regression analysis showed evidence for interaction between rs4774*C and DRB1*1501 associated with risk for MS (ratio of ORs = 1.72, 95% CI 1.28–2.32, P = 3 × 10⁻⁴). Furthermore, rs4774*C was associated with DRB1*1501+ MS when conditioned on the presence (OR = 1.67, 95% CI = 1.19–2.37, P = 1.9 × 10⁻³) and absence (OR = 1.49, 95% CI = 1.15–1.95, P = 2.3 × 10⁻³) of CLEC16A rs6498169*G, a putative MS risk allele adjacent to CIITA. Our results provide strong evidence supporting a role for CIITA variation in MS risk, which appears to depend on the presence of DRB1*1501.

INTRODUCTION

Multiple sclerosis (MS) is a chronic inflammatory disorder of the central nervous system characterized by demyelination, astrogliosis, varying degrees of axonal pathology and a relapsing or progressive course (1). A strong but complex genetic component in MS pathogenesis is indicated by both an increased relative risk in non-twin siblings compared with the general population, and by an increased concordance rate in monozygotic compared with dizygotic twins (25 versus 5%) (2–4). The strongest and most consistent evidence for a susceptibility gene in MS is within the major histocompatibility complex (MHC) on chromosome 6p21.3. Associations with the human leukocyte antigen (HLA)-DR15 haplotype (DRB1*1501-DQA1*0102-DQB1*0602) have been repeatedly demonstrated in multiple populations, primarily in those of Northern European descent (5,6). Haplotype analysis of HLA class II genes in admixed African Americans has demonstrated HLA-DRB1*15 is the primary susceptibility allele for MS (7). However, a complex pattern of allelic heterogeneity at the DRB1 locus in MS highlights the intricate nature of this genetic association (8). Transgenic animal models of autoimmune demyelination confirm the critical role of DRB1 gene products in initiating and maintaining a damaging anti-myelin immune response and suggest DQB1*0602 is associated with anti-myelin autoimmunity (9,10).

The identification of all non-MHC genetic risk factors in MS, while progressing steadily over the past couple of years, is far from complete. Several whole genome linkage screens in MS previously identified a large number of regions of modest contributions with little overlap, underscoring a complex polygenic pattern of inheritance contributing to disease susceptibility (11–14). Recent genome-wide association (GWA) and replication studies have identified several genetic risk loci for MS, including IL2RA, IL7RA, CLEC16A, CD58, TNFRSFA1 and IRF8 (15–20), though each contributes very modestly to the overall genetic risk for disease. Therefore, a substantial component of the genetic susceptibility to MS remains unknown. While GWA studies are attractive for many reasons, including that in principle they are ‘hypothesis free’, it is clear that experiments using current technology will be limited in their ability to identify the entire genetic contribution for most complex diseases, including MS (21). Candidate gene studies have historically failed to identify susceptibility loci with conclusive evidence. However, revisiting candidate gene studies with well-powered data sets and strong hypotheses based on prior research remains an important strategy for disease gene identification.

The MHC class II transactivator gene (CIITA, also called MHC2TA) encodes the CIITA protein, a transcription factor essential for the expression of HLA class II molecules and involved in the expression of HLA class I molecules (16,18,22–28). CIITA spans 48 kb on chromosome 16p13 and has four alternate first exons in a 12 kb promoter region (I–IV) (29). The gene is adjacent to a recently identified MS risk locus (C-type lectin domain family 16, member A gene, CLEC16A, previously called KIAA0350) (16,18). The CIITA protein contains a highly conserved and relatively rare domain also encoded by the neuronal apoptosis inhibitory protein gene (NAIP, also called BIRC1), associated with spinal muscular atrophy, and the nucleotide-binding oligomerization domain protein 2 gene (NOD2), associated with Crohn's disease (30–33). Mutations in CIITA cause a rare and severe immunodeficiency characterized by HLA class II deficiency (bare lymphocyte syndrome) (34). Thus, CIITA is an attractive candidate for genetic studies of autoimmune diseases and other inflammatory conditions for which HLA associations have been well established. In this multi-stage study, we investigated whether genetic variation in CIITA is associated with MS risk and assessed effect modification by DRB1*1501.

RESULTS

In stage 1, 24 CIITA SNPs were genotyped and analyzed in a European data set of 1320 MS cases and 1363 independent healthy controls (total n = 2683) (Table 1 and Supplementary Material, Table S1). In the second stage, we genotyped and analyzed rs4774 in an independent data set of 973 MS families of self-identified European ancestry (total n = 978 offspring MS cases, 2917 individuals) (Table 1) (16). DRB1*1501 was strongly associated with MS in case–control [odds ratio (OR) = 2.71, 95% confidence interval (95% CI) = 2.36–3.11, P = 1 × 10⁻⁶] and family (P < 1 × 10⁻⁶) data sets, as expected (data not shown; Materials and Methods).

Table 1.

Characteristics of MS cases and controls analyzed in this study (total n = 3656)

Clinical characteristic	MS, % (n = 1320)	Controls, % (n = 1363)	MS family probands, % (n = 973)
Site
Brigham and Women's Hospital, Boston	22.5 (294)	0.6 (8)	5.4 (53)
School of Clinical Medicine, Cambridge	50.2 (664)		46.6 (453)
University of California, San Francisco	27.3 (362)	44 (599)	48.0 (467)
Wellcome Trust Case Control Consortium		55.4 (756)
Sex
Male	37.7 (502)	35 (473)	48.4 (471)
Female	62.3 (818)	65 (890)	51.6 (502)
Age, years
Range	25–91	25–86	21–92
Mean ± SD	52.0 ± 10.2	49.9 ± 12.3	50.7 ± 10.8
Age-at-onset, years
Range	4–64		11–60
Mean ± SD	33.3 ± 9.9		30.3 ± 8.7
Disease course
Relapsing-remitting	52.5 (693)		66.9 (651)
Secondary-progressive	23.2 (306)		20.3 (198)
Primary-progressive	12.0 (159)		4.7 (46)
Clinically isolated syndrome	3.4 (45)		3.3 (32)
Progressive-relapsing	1.5 (20)		1.2 (12)
Unknown	7.4 (97)		3.5 (34)
Number of HLA-DRB1*1501 allelesa
0	47.3 (625)	74.7 (1018)	44.6 (434)
1	45.4 (599)	23.3 (318)	46.5 (452)
2	7.3 (96)	2.0 (27)	8.9 (87)

HLA, human leukocyte antigen; MS, multiple sclerosis; SD, standard deviation.

^aThe rs3135388 (A/G) SNP was genotyped to characterize DRB1*1501 status, due to very strong correlation between the presence of rs3135388*A and DRB1*1501 as previously described (76).

Of the 24 CIITA SNPs tested, evidence for association with MS was observed only for rs4774 (χ² = 8.14, P = 4.9 × 10⁻³) after application of a conservative correction for number of statistical tests (Materials and Methods, Table 2). Furthermore, rs4774 and MS were more strongly associated in individuals carrying DRB1*1501 (χ² = 18.8, P = 1 × 10⁻⁴) (Table 2). In both comparisons (overall and stratified by presence of DRB1*1501), the minor rs4774*C allele frequency was increased in MS cases versus controls (overall: OR = 1.19, 95% CI = 1.06–1.34, DRB1*1501+: OR = 1.6, 95% CI = 1.3–1.97) (Supplementary Material, Table S2). This result exceeded our threshold for statistical significance. No evidence for association was observed for the previously reported −168A/G CIITA promoter polymorphism (rs3087456). CIITA SNP haplotypes were assigned based on block structure (see Materials and Methods) and also compared between MS cases and controls. A total of six blocks were observed (Fig. 1). Results did not indicate the presence of any other CIITA SNP associations stronger than rs4774. Interestingly, rs4774 did not fall within surrounding SNP blocks, and LD with neighboring SNP variant rs3087456 was not present (r² = 0.01). No evidence for any global haplotype associations between CIITA and MS were observed (see Supplementary Material, Table S3). The family-based analysis of rs4774*C in 539 DRB1*1501+ MS extended families also showed evidence of association with increased MS risk, albeit weak (P = 2.3 × 10⁻²) (Table 3). MS families were stratified based on DRB1*1501 status in the proband.

Table 2.

P-values^a from the Cochran–Armitage test of association in 1320 MS cases and 1363 controls stratified by the presence of the HLA-DRB1*1501^b risk allele (total n = 2683)

SNP	Marker	MS MAF	Controls MAF	Overall	DRB1*1501+	DRB1*1501−
1	rs4436808	0.01	0.01	1.00	0.24	0.37
2	rs6498114	0.24	0.25	0.19	0.64	0.22
3	rs9302456	0.34	0.34	0.81	0.42	0.38
4	rs4781010	0.01	0.01	0.79	0.20	0.37
5	rs3087456	0.25	0.27	0.16	0.87	0.07
6	rs12928665	0.24	0.26	0.22	0.90	0.10
7	rs12932187	0.06	0.07	0.70	0.38	0.59
8	rs12925158	0.02	0.02	0.84	0.71	0.63
9	rs8043545	0.27	0.28	0.24	1.00	0.13
10	rs6498119	0.07	0.06	0.22	0.06	0.78
11	rs4781015	0.20	0.22	0.06	0.51	0.07
12	rs7189406	0.07	0.07	0.86	0.91	0.43
13	rs6498124	0.44	0.42	0.07	1.8 × 10−3	0.79
14	rs4781016	0.29	0.27	0.15	8.5 × 10−3	0.82
15	rs4774	0.30	0.27	4.9 × 10−3	1 × 10−4	0.81
16	rs13330686	0.09	0.07	0.04	0.05	0.39
17	rs13336804	0.09	0.07	0.03	0.05	0.37
18	rs4781019	0.46	0.44	0.15	0.09	0.54
19	rs6498126	0.20	0.21	0.66	0.20	0.40
20	rs4781020	0.30	0.30	0.92	0.65	0.53
21	rs6498132	0.12	0.11	0.30	1.3 × 10−3	0.12
22	rs2229322	0.11	0.10	0.57	0.25	0.81
23	rs4781024	0.41	0.41	0.76	0.07	0.15
24	rs1139564	0.20	0.20	0.42	0.40	0.86

HLA, human leukocyte antigen; MAF, minor allele frequency; SNP, single nucleotide polymorphism.

^aCharacterized by rs3135388 genotyping.

^bP-values based on 10 000 permutations.

Bold value indicates statistically significant.

Figure 1.

R² plot illustrating the LD structure of CIITA SNP variants in healthy controls; darker gray indicates higher r² between pairs of SNPs. LD, linkage disequilbrium; SNP, single nucleotide polymorphism.

Table 3.

Results for PDT of rs4774 in 973 extended MS families stratified by the presence of the HLA-DRB1*1501 risk allele (total n = 2917)

Result	Overall (n = 973)	DRB1*1501+ (n = 539)	DRB1*1501− (n = 434)
Number of informative families	892	497	395
Number of informative DSPsa	489	288	201
Siblings affected: not affected, % (n): % (n)
rs4774*G	68.8 (530): 72.1 (705)	68.5 (304): 73.4 (423)	69.3 (226): 70.1 (282)
rs4774*C	31.2 (240): 27.9 (273)	31.5 (140): 26.6 (153)	30.7 (100): 29.9 (120)
Number of informative trios	674	357	317
Alleles transmitted: not transmitted, % (n): % (n)
rs4774*G	70.3 (948): 69.0 (930)	69.3 (495): 70.5 (504)	71.5 (453): 67.2 (426)
rs4774*C	29.7 (400): 31.0 (418)	30.7 (219): 29.5 (211)	28.5 (181): 32.8 (208)
P-valueb	0.25	2.3 × 10−2	0.10

DSP, discordant sibling pair; HLA, human leukocyte antigen; MS, multiple sclerosis; PDT, pedigree disequilibrium test.

^aThe informative DSPs contained 385, 222 and 163 affected offspring in the overall, DRB1*1501+ and DRB1*1501− samples, respectively.

^bOne-sided, asymptotic P-values.

Bold value indicates statistically significant.

Rs4774 frequencies did not differ between MS cases from the case–control [minor allele frequency (MAF) 0.304] and family-based (MAF 0.306) data sets utilized in this study (P = 0.83). Therefore, a pooled analysis, whereby both MS case groups were combined (from stages 1 and 2) and compared to controls (from stage 1), was performed for increased statistical power (stage 3). One MS case per family was selected. A total of 3656 individuals (n = 2293 cases and 1363 controls) were included in the analysis. Rs4774*C was associated with increased MS risk (χ² = 10.9, P = 1 × 10⁻³), and this association was stronger in DRB1*1501+ cases and controls (χ² = 18.9, P = 1 × 10⁻⁴) (Table 4). We also conducted a combined analysis that maintained the robustness of the family-based component against potential confounding due to population stratification and controlled for potential differences between study populations in stages 1 and 2. Specifically, we used the sibling TDT to analyze the extended families and the Cochran–Mantel–Haenszel test for stratified tables to incorporate the independent cases and controls. Results supported those obtained from the pooled analyses. Rs4774*C was associated with increased risk of MS in the overall sample [χ² = 5.4 (1249 expected, 1298 observed), P = 3.2 × 10⁻²], and this association was stronger in DRB1*1501+ individuals [χ² = 17.4 (687.8 expected, 746 observed), P = 3 × 10⁻⁴]. No association was observed in DRB1*1501− individuals [χ² = 1.1 (568.2 expected, 552 observed), P = 0.33].

Table 4.

Frequencies for the rs4774 variant and HLA-DRB1*1501^a in MS cases (n = 1320), controls (n = 1363) and one MS case per family (n = 973)^b

Number of rs4774*C alleles	Presence of DRB1*1501a	MS, % (n)	Controls, % (n)
0	−	23.7 (543)	39.0 (531)
1	−	18.8 (430)	29.7 (405)
2	−	3.8 (86)	6.0 (82)
0	+	24.9 (570)	15.3 (209)
1	+	23.2 (532)	8.1 (110)
2	+	5.8 (132)	1.9 (26)

HLA, human leukocyte antigen; MS, multiple sclerosis.

^aDRB1 or rs3135388 genotype data were available for all cases and controls in this study (8,16,55).

^bTotal n = 3656 individuals (2293 affected).

Results obtained from a logistic regression analysis of MS cases and controls demonstrated evidence of interaction between DRB1*1501 and rs4774*C [ratio of ORs (ROR_GxG) = 1.72, 95% CI = 1.28 to 2.32, P = 3 × 10⁻⁴] (Table 4). Furthermore, the rs4774 genotype was associated with the presence of DRB1*1501 in a case-only analysis (OR = 1.19, 95% CI = 1.05–1.36, P = 6.8 × 10⁻³), adding further evidence for interaction. Also, the presence of both DRB1*1501 and rs4774*C, when compared with the presence of DRB1*1501 alone, was associated with MS (OR = 1.79, 95% CI = 1.39–2.30, P = 2.6 × 10⁻⁶) (Table 4).

To determine whether the CIITA rs4774 association observed in the current study was independent from the previously reported MS association with nearby CLEC16A rs6498169 (16,18), conditional haplotype analysis using the larger stage 3 data set and genotypes for both loci was performed. Here, rs4774*C was associated with DRB1*1501+ MS when conditioned on both presence (OR = 1.67, 95% CI = 1.19–2.37, P = 1.9 × 10⁻³) and absence (OR = 1.49, 95% CI = 1.15–1.95, P = 2.3 × 10⁻³) of the CLEC16A rs6498169*G MS risk allele (see Table 5 for haplotype frequencies). CLEC16A rs6498169*G was associated with MS when conditioned on the presence (OR = 1.32, 95% CI = 1.09–1.61, P = 4.9 × 10⁻³) and absence (OR = 1.19, 95% CI = 1.05–1.35, P = 6.6 × 10⁻³) of rs4774*C, and this association trended toward significance in the DRB1*1501 stratified subsets (DRB1*1501+/rs4774*C+, OR = 1.38, 95% CI = 0.96–2.01, P = 0.08; DRB1*1501+/rs4774*C−, OR = 1.23, 95% CI = 0.90–1.54, P = 0.06; DRB1*1501−/rs4774*C+, OR = 1.33, 95% CI = 1.03–1.72, P = 2.6 × 10⁻²; DRB1*1501−/rs4774*C−, OR = 1.16, 95% CI = 0.98–1.37, P = 0.07). These results indicate that the association between CIITA and MS is independent of CLEC16A. Furthermore, in a logistic regression, rs4774*C demonstrated association with MS (OR = 1.24, 95% CI 1.07–1.44, P = 6 × 10⁻³) even after adjusting for the presence of DRB1*1501 (OR = 3.33, 95% CI 2.85–3.89, P < 2 × 10⁻⁶) and CLEC16A rs6498169*G (OR = 1.24, 95% CI 1.06–1.44, P = 4.9 × 10⁻³) (data not shown).

Table 5.

Rs4774-rs6498169 haplotype frequencies in the combined HLA-DRB1-1501+^a MS cases (n = 918)^b and controls (n = 345) (total n = 1263)

Presence of rs4774*Cc	MS, % (n)	Controls, % (n)	OR (95% CI)	P-value
CG	14.9 (274)	8.8 (61)	1.67 (1.19–2.37)	1.9 × 10−3
GG	24.4 (448)	24.2 (167)
CA	17.9 (329)	14.7 (101)	1.49 (1.15–1.95)	2.3 × 10−3
GA	42.8 (786)	52.3 (361)

HLA, human leukocyte antigen; MS, multiple sclerosis.

^aDRB1 or rs3135388 genotype data were available for all cases and controls in this study (8,16,55).

^bGenotyping data for the rs6498169 CLEC16A variant was available for all 1320 cases from the case–control sample (stage 1) (DRB1*1501+, n = 695) and 518 cases from the family sample (stage 2) (DRB1*1501+, n = 223).

^cAssociation testing was performed for rs4774 (C/G in bold) conditioned on the presence or absence of the CLEC16A rs6498169*G risk allele.

DISCUSSION

Due to the strong association between HLA-DRB1*1501 and MS, and the influence of CIITA on the expression of HLA class II genes, the CIITA locus has long been considered a strong MS candidate gene. Almost a decade ago, Rasmussen et al. (35) screened for variants in 111 MS cases and 105 controls from the UK, and through sequencing, identified the −168A/G variant in the type III CIITA promoter region (rs3087456), as well as five variants in the 3′ untranslated region (UTR). Association between MS and CIITA in the overall or DR15-stratified sample was not detected; however, an association between the −168A/G variant and primary progressive MS (P < 0.04) was reported. Shortly thereafter, Patarroyo et al. (36) discovered four additional CIITA SNPs, including the rs4774 (+1614G/C) missense mutation in exon 11, through bidirectional sequencing of lymphocyte cDNAs from 50 healthy individuals of Northern European ancestry. A more recent study reported association between the −168A/G variant and increased susceptibility to both MS and rheumatoid arthritis (RA), as well as lower expression of CIITA after stimulation of leukocytes with interferon γ (37). In addition, ex vivo stimulation with interferon-γ of peripheral blood cells from RA cases with the −168G/G genotype exhibited decreased expression of CIITA and the HLA class II alleles DQA1 and DRA, compared with RA patients with the −168A/G or −168A/A genotype, and the difference was greater with increased stimulation (37). However, a meta-analysis of 10 studies revealed no evidence for association between the −168A/G variant and RA (38).

Thus far, studies of the CIITA −168A/G variant and MS have yielded conflicting results (37,39–41). No evidence for association between the −168A/G variant and MS was observed in the current study, despite achieving 80% power to detect an allelic OR of 1.25 (MAF 0.27) under a two-sided α = 4.9 × 10⁻³, or even an allelic OR of 1.18 under relaxed significance criteria (α = 0.05). As part of the current investigation, we also performed a meta-analysis of allelic association between the −168A/G polymorphism and MS in 3322 cases and 4260 controls, which were obtained from stage 3 of this study plus four additional published case–control studies (37,39–41). Between-study heterogeneity and publication bias were not present (data not shown). There was no evidence for association (summary OR = 1.06, 95% CI = 0.91–1.24, P = 0.47). Our results collectively and definitively exclude any major effect of the −168A/G variant on risk for MS (data not shown).

This study is the first to report evidence for interaction between the rs4774 (+1614G/C) missense mutation and DRB1*1501 associated with MS. This variant, located in exon 12, causes an amino acid substitution from glycine to alanine. Based on sequence homology and physical properties of amino acids, this amino acid substitution is predicted to have a tolerable effect on protein function (42). However, the exact functional consequences that result are not known. We have replicated this finding in an extended family-based sample; though the effect was in the same direction, the significance of the observed association was modest. A combined analysis of MS cases, controls and one MS case from each family also supports a role for rs4774 in MS susceptibility, particularly in the presence of DRB1*1501. Three previous studies have examined association between rs4774 and MS and reported negative findings (37,40,41). We performed a meta-analysis of rs4774 and MS in 2669 cases and 3773 controls, obtained from stage 3 of this study plus three additional published case–control studies (37,40,41). While some evidence for between-study heterogeneity was detected (P < 0.02), publication bias was not present (data not shown). The meta-analysis revealed no evidence for allelic association when all MS cases and controls were considered (summary OR = 1, 95% CI = 0.81–1.23, P = 0.99). Unfortunately, DRB1*1501 data were not available for published studies to perform stratified analyses. The association seen in the current study between rs4774 and MS appears to depend on the presence of DRB1*1501.

Rs4774*C has been reported to be over-represented in MS cases with active replication of human herpes virus 6A (HHV-6A), compared with MS cases and controls without active replication of HHV-6A (43,44), although this association needs to be replicated in an independent sample. Actively replicating HHV-6A was defined as at least one positive serum sample for HHV-6A among the five serum samples collected in a 2-year period. The importance of this finding is not yet known. Larger studies of MS and CIITA that include environmental exposure data are warranted.

Variation within the MHC class II transactivator gene in animal studies (rat Ciita, 10q11) has been reported to affect the quantity of MHC class II expression in the brain and on immune cells, as well as risk and severity of experimental allergic encephalomyelitis (EAE) (45). The rat strain most susceptible to neurodegeneration and central nervous system inflammation from experimental nerve injury [dark Agouti (DA) strain] was bred with the rat strain most resistant to experimental nerve injury [the Piebald Virol Glaxo (PVG) strain]. Compared with DA rats carrying the DA Ciita locus, DA rats with the PGV Ciita locus exhibited decreased MHC class II expression in the brain upon stimulation with IFN-γ, decreased MHC class II expression on B cells and dendritic cells, and reduced risk and severity of EAE. In MS, the relationship between variation at the CIITA locus and gene expression for both CIITA and MHC class II loci, as well as the resulting biological implications for the immune response and MS pathogenesis, are poorly understood. Large and comprehensive studies, particularly ones that can also fully explore clinical MS phenotypes, are needed.

We carefully considered the potential impact of population stratification on the current study (46). In stage 1, European ancestry was estimated in MS cases and controls using genetic markers and only individuals with ≥90% European ancestry were included in further analyses of CIITA variation. In the second stage, the family-based analysis used to replicate our initial finding was not subject to population stratification. Finally, one MS case per family was selected and combined with other cases for a larger case–control analysis. Because ancestry informative marker information was not available for all familial MS cases in this final stage, it is possible that ancestral differences in frequencies could have contributed to a spurious association in stage 3. However, the majority of families utilized here were subjected to rigorous testing for population outliers, as previously described (16). Despite these efforts, it is possible that within European population stratification may still be present. A conservative correction for multiple testing was also employed to help guide interpretation of testing; however, further replication studies will be required for confirmation.

GWA studies have not identified CIITA as a susceptibility locus for MS (16,47,48). Further, results for CIITA analysis in the current study would not meet criteria for genome-wide significance. Because the recently confirmed CLEC16A MS locus is adjacent to CIITA on chromosome 16, we examined LD patterns between 14 CLEC16A SNPs and 24 CIITA SNPs in the controls from stage 1, and 274 CLEC16A SNPs and 40 CIITA SNPs in HapMap samples of northern and western European origin (CEU, release 24) (49). There was no evidence of LD between CIITA and CLEC16A in the controls (r² ≤ 0.10). The rs1139564 CIITA 3′-UTR variant exhibited weak LD with the intronic rs8055876 CLEC16A variant in CEU (r² = 0.46). The rs8055876 CLEC16A variant was neither genotyped nor tagged in our controls, but the rs1139564 CIITA 3′-UTR variant was not in LD with rs4774 in either our controls (r² = 0.001) or CEU (r² = 0.001). Also, rs8055876 and the rs6498169 CLEC16A MS risk variant were not in LD in CEU (r² = 0.09). Thus, based on patterns of LD derived from two independent samples and results from our comprehensive analyses, including logistic regression modeling, it does not appear that association observed between CIITA and MS, specifically the effect on disease risk conferred by rs4774, is due to CLEC16A.

Approximately 80% of the common genetic variation in CIITA was captured in the current study, based on r² > 0.8 in CEU. An additional 19 CIITA HapMap variants were captured (see Supplementary Material, Table S4). Though this study failed to capture eight common CIITA variants, all of these were intronic, and perhaps less likely to play a role in MS susceptibility (see Supplementary Material, Table S5). In addition, rare variants in CIITA were not directly investigated in the current study, and must be considered in future studies.

In conclusion, this is the first large study of CIITA in MS to fully characterize common genetic variation in CIITA, including the assessment of haplotypes and gene x gene interaction with DRB1*1501. Our results confirm that the previously reported −168A/G promoter variant (rs3087456) is not associated with MS, and provide strong evidence for association between MS and the CIITA non-synonymous coding variant (rs4774; missense G/C; G500A) in the presence of DRB1*1501. Given the functional relevance of CIITA, and the relationship between CIITA and the class II DRB1 locus, our results will help further the understanding of biological mechanisms contributing to MS pathogenesis.

MATERIALS AND METHODS

Study subjects

MS cases and controls (first stage) were collected by the Brigham and Women's Hospital in Boston (BWH), School of Clinical Medicine in Cambridge (CSU), University of California in San Francisco (UCSF) and Wellcome Trust Case-Control Consortium (Table 1). The second stage included 973 extended MS families of self-identified European ancestry collected by BWH, CSU and UCSF (total n = 978 offspring MS cases, 32 parental MS cases, 2917 individuals) (Table 1). Five hundred and eighteen trios from a GWA study by the International Multiple Sclerosis Genetics Consortium (IMSGC) were part of this family sample (16). There was no overlap between MS cases or other individuals in the two data sets. All MS cases in this study met the revised McDonald MS criteria (50). A total of 5600 genotyped individuals were studied. Based on the available genetic ancestry data for all cases and controls, and to apply the most stringent criteria possible for genetic analysis of CIITA, only MS cases and controls with ≥90% European ancestry were analyzed (stage 1). These steps reduced the potential impact of population stratification on our investigation.

Genotyping

Initially, 29 CIITA single nucleotide polymorphisms (SNPs) and the CLEC16A rs6498169 MS risk variant were genotyped in the case–control sample with a custom Illumina iSelect 48K chip (San Diego, CA, USA). Four monomorphic missense variants (rs34648899, rs35451230, rs4781022 and rs8046121) and one missense variant (rs7197779) with a low MAF (<0.001) were omitted from further analyses. Hardy–Weinberg equilibrium (HWE) for all marker genotypes was examined in cases and controls separately with the exact test (PLINK v1.07) (51,52). There was no evidence of deviation from HWE in the cases or controls (P < 1 × 10⁻⁴). Ultimately, 24 CIITA variants were tested in the first stage (see Supplementary Material, Table S1). Rs4774 was also genotyped in an independent extended family sample (n = 455 families) using an Applied Biosystems TaqMan assay (Foster City, CA, USA). There were no Mendelian errors (PedCheck v1.1) and the exact test showed no deviation from HWE in pedigree founders (PEDSTATS v0.6.8) (53,54). In addition, genotyping data for rs4774 and CLEC16A rs6498169 were available for 518 trio families (DRB1*1501+ n = 223) from a GWA study that used the Affymetrix GeneChip Human Mapping 500K array (16). DRB1 or rs3135388 genotype data were available for all cases and controls in this study. The DRB1 locus was characterized as previously described (8,16,55).

Statistical analysis

European ancestry was estimated from 1000 markers included on the Illumina iSelect 48K chip using a Bayesian clustering algorithm based on two populations under the admixture model using a burn-in length of 10 000 for 10 000 repetitions (STRUCTURE v2.3.1) (56). European ancestry estimates were tested for association with MS in the combined cases and controls (all individuals from stage 1) and subgroups stratified by DRB1*1501 status with logistic regression (R v2.9; http://www.r-project.org). European ancestry estimates were statistically indistinguishable when compared between stage 1 MS cases [mean = 0.99, standard deviation (SD) = 0.018] and controls (mean = 0.99, SD = 0.017) (asymptotic P = 0.20), even in the DRB1*1501+ (P = 0.71) and DRB1*1501− subsets (P = 0.27).

Association was tested in the case–control sample with a Cochran–Armitage test for trend [degrees of freedom (df) = 1] (PLINK) (57,58). Trend tests for CIITA SNPs were stratified by the presence or absence of DRB1*1501 in the cases and controls. To address multiple testing concerns for CIITA, we determined significance in stage 1 (P ≤ 4.9 × 10⁻³) by controlling the false discovery rate (FDR) using the Benjamini–Hochberg method at a level of 10% (59). Allelic ORs and 95% CIs are also reported (see Supplementary Material, Table S2). DRB1*1501 was also tested with the Cochran–Armitage test. Haplotypes were estimated in the controls with the expectation-maximization algorithm and haplotype blocks were determined with the confidence bound algorithm (HAPLOVIEW v4.1) (Fig. 1) (60). We computed maximum likelihood estimates of haplotype probabilities for the MS cases and controls with the expectation-maximization algorithm and conducted global haplotype tests of six haplotype blocks encompassing 16 SNPs using score statistics under the additive genetic model (HAPLOSTATS v1.4.3, R) (61). Haplotypes with inferred frequencies <5% were excluded. Analyses were stratified by DRB1*1501 status. ORs and 95% CIs are reported.

Association between rs4774 and MS was tested in extended families with the pedigree disequilibrium test (PDT v6.2.4), an extension of the transmission disequilibrium test (TDT) (62,63). We chose the PDT to take advantage of the data available for families with more than one affected offspring and families with one or more unaffected siblings. Eight hundred and ninety-two families [674 affected offspring trios and 489 discordant sibling pairs (DSPs)] were informative for the PDT, and a one-tailed asymptotic P-value was calculated (Table 3). Four hundred and ninety-seven DRB1*1501+ families (358 affected offspring trios and 288 DSPs) were informative for the PDT. DRB1*1501 was also tested with the PDT and the informative sample consisted of 848 families with 630 affected offspring trios and 488 DSPs (data not shown).

A combined analysis (stage 3) of cases and controls (from stage 1) and one proband per family (from stage 2) was conducted using a 1-df Cochran–Armitage test for trend (PLINK). Rs4774 allele frequencies between the cases from stages 1 and 2, as well as between the controls from stage 1 and the affected family-based controls (AFBACs, or non-transmitted parental alleles) from stage 2, were calculated with AFBAC v1.13 (64). Frequencies were tested with chi-square tests of heterogeneity (df = 1) in R and asymptotic P-values were calculated. Rs4774 frequencies in controls from stage 1 (MAF 0.268) and the AFBACs from stage 2 (MAF 0.311) were different (P = 2.5 × 10⁻⁴). Therefore, we chose not to utilize the AFBACs as controls in the combined analysis (stage 3). The AFBACs were utilized only in family-based analyses robust against effects of population stratification.

We also tested rs4774 for association in the combined data sets from stages 1 and 2 with a TDT that incorporates the sibling TDT and parental phenotypes and tests unrelated cases and controls as sibships with the Cochran–Mantel–Haenszel test for stratified tables (PLINK) (57,62,65–67). We considered conducting a conditional logistic regression analysis of one proband per family and three ‘pseudo-controls’ based on non-transmitted parental alleles, along with the cases and controls (68–70). However, this method would have substantially reduced power because a third of the families (n = 301) did not have rs4774 genotypes available for both of the proband's parents.

Interaction between the presence of rs4774*C and DRB1*1501 (gene x gene) in the combined cases and controls (stage 3) was tested using logistic regression (R) (Table 4); we report these results as the ratio of ORs (ROR_GxG) and include 95% CIs. A case-only interaction test was also conducted by using logistic regression (R) to test for association between the rs4774 genotype and the presence of DRB1*1501 (71,72). The conditional haplotype method (73) (HAPLOVIEW, R) was used to test for association between rs4774 and MS, conditional on the presence or absence of the CLEC16A rs6498169*G MS risk allele; asymptotic P-values were calculated. The presence of rs4774*C was tested for association with MS in a logistic regression (R), adjusted for presence/absence of DRB1*1501 and the presence of the CLEC16A rs6498169*G MS risk allele.

All reported P-values are empirically based on ≥10 000 permutations and are two-tailed, unless otherwise noted in the methods. Power was estimated (PGA v2.0) assuming a two-sided type I error of α = 4.9 × 10⁻³, to account for number of statistical tests (74). Stage one of the current study had 80% power to detect an OR ranging from 1.22 to 1.38, under varying MAF (0.1–0.5).

A meta-analysis of the −168A/G variant and MS was performed for 3322 cases and 4260 controls in stage 3 plus four additional case–control studies (37,39–41). ORs and 95% CIs were calculated to test whether the G allele [AA compared against GA + GG carriers (dominant model)] or the GG genotype [AA + AG compared against GG carriers (recessive model)] increased risk for MS. A meta-analysis of the rs4774 variant and MS was also performed for 2669 cases and 3773 controls in stage 3 plus three additional case–control studies (37,40,41). ORs and 95% CIs were calculated to test whether the C allele (GG compared against GC + CC carriers) or the CC genotype (GG + CG compared against CC carriers) increased risk for MS. Results did not differ under the recessive models (data not shown). Between-study heterogeneity was assessed with a χ² test of heterogeneity and publication bias was evaluated with a funnel plot. We calculated summary ORs and 95% CIs using a random effects model and asymptotic P-values, as previously described (38,75).

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

FUNDING

NIH/NINDS R01 NS049477 to the International MS Genetics Consortium; NMSS RG4201-A-1 to J.L.H., RG2901 to J.R.O. and RG4201 to J.L.M.; NIH/NIAID F31 AI075609 to P.G.B. and U19 AI067152 to S.L.H. This work was also supported by the Medical Research Council (G0700061) and the Cambridge NIHR Biomedical Research Centre.