Genes and Immunity (2009) 10, 141–150 & 2009 Macmillan Publishers Limited All rights reserved 1466-4879/09 $32.00 www.nature.com/gene

ORIGINAL ARTICLE Genetic variants of the HLA-A, HLA-B and AIF1 loci show independent associations with type 1 diabetes in Norwegian families

MC Eike1,2, M Olsson3, DE Undlien4,5, K Dahl-Jørgensen6,7, G Joner6,8, KS Rønningen9, E Thorsby1,2 and BA Lie1 1Institute of Immunology, Rikshospitalet University Hospital, Oslo, Norway; 2Institute of Immunology, Faculty Division Rikshospitalet, University of Oslo, Oslo, Norway; 3Mathematical Statistics, Chalmers University of Technology, Gothenburg, Sweden; 4Institute of Medical Genetics, Faculty Division Ulleva˚l University Hospital, University of Oslo, Oslo, Norway; 5Department of Medical Genetics, Ulleva˚l University Hospital, Oslo, Norway; 6Department of Paediatrics, Ulleva˚l University Hospital, Oslo, Norway; 7Faculty of Medicine, University of Oslo, Oslo, Norway; 8Institute of Health Management and Health Economics, University of Oslo, Oslo, Norway and 9Division of Epidemiology, Norwegian Institute of Public Health, Oslo, Norway

The main genetic predisposition to type 1 diabetes (T1D) is known to be conferred by the HLA-DRB1, -DQA1 and -DQB1 genes in the major histocompatibility complex (MHC). Other genetic factors within this complex are known to contribute, but their identity has often been controversial. This picture is shared with several other autoimmune diseases (AIDs). Moreover, as common genetic factors are known to exist between AIDs, associations reported with other AIDs may also be involved in T1D. In this study, we have used these observations in a candidate gene approach to look for additional MHC risk factors in T1D. Using complementary conditional methods (involving conditional logistic regression and family-based haplotype tests) and analyses of linkage disequilibrium (LD) patterns, we confirmed association for alleles of the HLA-A and HLA-B genes and found preliminary evidence for a novel association of a single-nucleotide polymorphism (rs2259571) in the AIF1 gene, independent of the DRB1-DQA1-DQB1 genes and of each other. However, no evidence of independent associations for a number of previously suggested candidate polymorphisms was detected. Our results illustrate the importance of a comprehensive adjustment for LD effects when performing association studies in this complex. Genes and Immunity (2009) 10, 141–150; doi:10.1038/gene.2008.88; published online 6 November 2008

Keywords: type 1 diabetes; autoimmune disease; major histocompatibility complex; human leukocyte antigen; genetic predisposition; conditional analysis

Introduction AIDs, including ankylosing spondylitis, celiac disease, Graves’ disease, juvenile idiopathic arthritis, multiple Type 1 diabetes (T1D) is an autoimmune disease (AID) sclerosis, rheumatoid arthritis, sarcoidosis, systemic characterized by the destruction of insulin-producing lupus erythematosus and ulcerative colitis (cf. Supple- b-cells of the pancreas, causing irreversible insulin mentary Table 1). This commonality between AIDs is deficiency and a number of serious clinical complica- also reflected in genome-wide linkage scans14 and/or tions. The most important susceptibility factors reside association studies revealing shared genetic factors within the major histocompatibility complex (MHC) elsewhere in the genome, most notably variants of the where particular human leukocyte antigen (HLA) class CTLA4 and PTPN22 genes.15 This has led to the belief II DRB1-DQA1-DQB1 haplotypes confer high disease that, despite a wide variety of clinical manifestations, risk.1,2 However, numerous studies have strongly AIDs share certain biological features that are common to implicated additional T1D risk loci within the MHC.3–12 the autoimmune process. Genetic mapping of the MHC This complex contains an unusually high density of in a particular AID, such as T1D, therefore represents genes involved in immunological responses,13 of which potential value not only for the disease under study, but many are good candidates for involvement in auto- also for mapping of this region in other AIDs. However, immune processes. The observation of multiple suscept- association studies in the MHC are subject to severe ibility loci in the MHC is shared with a number of other confounding due to strong, variable and extensive linkage disequilibrium (LD), which has substantially complicated such efforts. Consequently, other than in Correspondence: MC Eike, Institute of Immunology, Rikshospitalet cases of strong primary associations (frequently for University Hospital, Sognsvannsveien 20, N-0027 Oslo, Norway. E-mail: [email protected] antigen-presenting HLA class II molecules, such as in Received 31 July 2008; revised 2 September 2008; accepted 8 October T1D), evidence for disease associations has often been 2008; published online 6 November 2008 conflicting. In an effort to resolve some of these conflicts, Conditional MHC screen of T1D candidate loci MC Eike et al 142 we used a candidate gene approach, examining pre- LD with the known class II risk variants (using single- viously suggested susceptibility loci from various AIDs, point main effects tests adjusted for DRB1-DQA1-DQB1 to identify additional risk variants for T1D within the genotypes), only five markers remained with significant MHC. results (Po0.05), that is, HLA-A, HLA-B and the single- nucleotide polymorphisms (SNPs) rs1800750 (tumor necrosis factor (TNF) À376), rs3132451 and rs2259571 Results (in the allograft inflammatory factor 1 (AIF1) gene), respectively (Figure 1b and Table 1; cf. Supplementary Polymorphisms were selected on the basis of earlier Table 2 for further details). reports of involvement in T1D and/or other AIDs, either directly (association reported for the marker) or indir- ectly (association or functional evidence reported for the gene in which the marker is located). In an effort to break Tests of independence between significant markers up the strong LD pattern, we also included additional To test whether these five T1D-associated genetic polymorphisms to maximize the number of haplotypes variants were independent of LD also with each other, picked up by the sum of markers, based on the work of, we repeated the main effects tests, still adjusting for among others, Allcock et al. (2004).16 Supplementary DRB1-DQA1-DQB1 genotypes, but with the additional Table 1 summarizes the most important earlier references inclusion of each of the other significant markers in the (reviews where appropriate), with direct affirmative model, one at a time (Table 1). None of the markers evidence reported for the polymorphisms investigated remained significant across all tests, although HLA-B here (Supplementary Table 2), and at least some form of only had one insignificant, but borderline, P-value. Note, adjustment for LD effects with known associated however, that the number of pseudo-cases/controls variants of other MHC loci. varies between the tests due to slight differences in genotyping success rates and inclusion of more markers Association tests in the model, and that some of the variation in Unconditional transmission-disequilibrium tests of all significance therefore may be attributed to variable loss markers showed the lowest P-value for the DRB1-DQA1- of statistical power. Nonetheless, all markers appeared to DQB1 haplotypes (P ¼ 5.8 Â 10À101), as expected, and influence the results of some of the other markers to significant P-values (at a ¼ 0.05) for 39 of the 53 some extent. The greatest impact appeared to be investigated markers (Figure 1a and Supplementary conferred by HLA-B, which reduced the P-values of both Table 2). Twenty seven of these were still significant HLA-A and the TNF SNP rs1800750 to nonsignificant after Bonferroni correction for multiple markers levels, but HLA-A also appeared to have some effect on (Po9.4 Â 10À4). However, after adjusting for effects of HLA-B. Also, the associations of the AIF1 SNPs rs3132451

Unconditional TDT 1E-27

1E-21

1E-15 -value

P 1E-09

1E-03 0.05 corr

Conditioned on DRDQ 1E-03

HLA-B rs3132451 1E-02 HLA-A rs1800750 rs2259571

-value 0.05

P 1E-01

pos (kb) 30020 3056731500 31750 32000 32250 32500 32750

IIIIIIregions genes TNF HLA-A HLA-E HLA-B MICA MICB BAT1 NFKBIL1 LTA AIF1 HSPA1L HSPA1B CFB AGER BTNL2 DRDQ C2

Figure 1 Results of association tests. (a) Unconditional transmission-disequilibrium tests. (b) Main effects tests adjusted for DRB1-DQA1- DQB1 (DRDQ) genotypes. P-values for DRDQ are not shown. Significance levels at a ¼ 0.05, Bonferroni-corrected for multiple markers À4 (0.05corr ¼ 9.4 Â 10 ) and uncorrected, respectively, are marked with dashed lines. Positions of the classical MHC regions and genes with genotyped markers are included for reference. Positions are along chromosome 6, genome build 36. Complete association results and further details are available in Supplementary Table 2.

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 143 Table 1 Regression modelling of associated markers to reveal associations independent of each other in Norwegian families

Test marker Markers adjusted for (included in model)

DRDQ DRDQ and HLA-A DRDQ and HLA-B DRDQ and rs1800750 DRDQ and rs3132451 DRDQ and rs2259571

nP-value nP-value nP-value nP-value nP-value nP-value

HLA-A 1414 0.012 — — 1226 0.12 1318 0.077 1036 0.031 1206 0.0073 HLA-B 1284 0.0052 1226 0.055 — — 1196 0.030 942 0.0015 1090 0.011 rs1800750 1460 0.023 1318 0.094 1196 0.69 — — 1042 0.071 1214 0.068 rs3132451 1102 0.0046 1036 0.0068 942 0.016 1042 0.0030 — — 988 0.086 rs2259571 1306 0.034 1206 0.027 1090 0.0058 1214 0.058 988 0.33 — —

Tests were adjusted for DRB1-DQA1-DQB1 (DRDQ) genotypes and additionally for genotypes of other markers with significant independent effects (Po0.05) of DRDQ (cf. Figure 1b). n: number of pseudo-cases and -controls. P-values are for a likelihood ratio test (uncorrected). Degrees of freedom were 1 for tests of SNPs, 23 for HLA-B, 14 for HLA-A adjusted for DRDQ genotypes and 13 for HLA-A adjusted for DRDQ genotypes and additional markers. and rs2259571 appeared connected, but were seemingly DRB1*04XX-DQA1*0301-DQB1*0302 and DRB1*XX- unaffected by HLA-A or -B. DQA1*0501-DQB1*0301) are internally heterogeneous (possibly with differential risk conferred by various Analyses of alleles, haplotypes and LD patterns DRB1*04XX or DRB1*XX alleles), which may have Reduction in statistical power is inherent in any caused an artificial association in the regression analyses. conditional approach, especially when considering more This, and the small numbers of informative transmis- than one conditional locus. Therefore, the employment of sions (Table 2), questions the status of this SNP as an complementary conditional approaches, which may important risk factor in T1D. restrain power in a different way, is important to validate To explore the association dependencies suggested by results. Moreover, as noted by Thomson et al. (2008), no the regression results, we performed analyses of the single statistical package has yet been developed that is implicated two-marker combinations (HLA-B with HLA-A capable of handling the complexity of MHC associations. or rs1800750, and rs2259571 with rs3132451; see above) In this situation, visual inspection and treatment of data by mapping them together on the DRB1-DQA1-DQB1 is necessary to reveal the underlying genetic causes of an haplotypes (combined mapping). As a further guide, we association.17 In this study, we mapped the allelic examined overall LD patterns, ascertaining that depen- distribution of each of the significant markers from the dencies between allelic associations were best captured first regression step (Figure 1b) on each DRB1-DQA1- by looking at either transmitted or nontransmitted DQB1 haplotype and tested whether these distributions haplotypes, depending on the risk direction (predispos- deviated significantly from that expected by LD alone. In ing or protective, respectively). These patterns, including addition to providing information on risk directions and all markers or alleles reported in Table 2, are given in the particular alleles involved, this approach provides Figure 2 (allelic D0; Figures 2c and d, respectively; valuable information about haplotype patterns, which corresponding global D0-values are given in Figures 2a may reveal LD dependencies not only with particular and b). Note, however, that these overall LD patterns DRB1-DQA1-DQB1 haplotypes, but also between alleles may not always reflect the LD patterns on particular at these loci. However, it should be noted that resulting DRB1-DQA1-DQB1 haplotypes, which may be stronger associations in these tests may be haplotype specific, and or weaker, depending on the underlying ancestral not necessarily descriptive of the true effect size of tested haplotypes. markers. In light of the limited power and to evaluate also nonsignificant trends, markers with frequency Association dependencies between HLA-A and HLA-B alleles deviations between transmitted and nontransmitted The global LD between HLA-A and -B was quite low haplotypes (Po0.10) from analyses of one marker at a (Figures 2a and b; D0o0.53), and among the HLA-A and time conditioned on the different DRB1-DQA1-DQB1 -B alleles with overlapping risk tendencies and DRB1- haplotypes are given in Table 2. In general, the HLA DQA1-DQB1 haplotypes (A*24, B*18 and B*39, cf. alleles A*24 and B*39 were significantly (Po0.05) Table 2), only A*24 and B*39 showed an appreciable increased, whereas A*01 and A*31 were significantly level of LD (Figure 2c, transmitted alleles; D0 ¼ 0.53 vs reduced among transmitted haplotypes. In addition, the D0 ¼À0.09 for A*24-B*18). In line with this, combined HLA-B allele B*18 showed a marginally significant mapping of A*24 and B*39 showed that these alleles predisposing effect on the DRB1*03-DQA1*0501- were often transmitted together on the DRB1*08- DQB1*0201 haplotype. For the SNPs, only rs2259571 DQA1*0401-DQB1*0402 haplotype (15 transmissions of (AIF1) yielded significant results in these stratified A*24-B*39 vs 8 transmissions of other haplotypes with analyses, with the minor allele C appearing to be either allele). B*39 seemed slightly better at explaining protective. When also looking at nonsignificant tenden- the association on this haplotype, as the association was cies (0.054Po0.10), rs1800750*A (TNF) and rs3132451*G somewhat stronger, although numbers were too small to (AIF1) (minor alleles) appeared predisposing. However, make a clear distinction (data not shown). Nonetheless, both of the DRB1-DQA1-DQB1 haplotypes where A*24 also showed significant associations on other rs3132451 showed frequency deviations (that is, DRB1-DQA1-DQB1 haplotypes (Table 2), which almost

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 144 Table 2 Allelic tests of markers conditional on DRB1-DQA1-DQB1 haplotypes in Norwegian families

Conditional haplotype Marker*Allele T NT Tfreq NTfreq OR P-value (exact)

DRB1 DQA1 DQB1

03 0501 0201 HLA-A*01 102 51 0.43 0.60 0.50 7.4 Â 10À3 HLA-A*24 25 3 0.11 0.04 3.20 0.070 HLA-B*18 18 1 0.09 0.01 6.56 0.051 rs1800750*A 15 1 0.07 0.02 4.55 0.13a rs2259571*C 6 6 0.03 0.10 0.28 0.034 04 0301 0301 rs2259571*C 6 31 0.60 0.86 0.24 0.087 0401 0301 0302 HLA-A*24 38 2 0.16 0.04 4.35 0.038 HLA-A*31 4 4 0.02 0.08 0.19 0.029 0404 0301 0302 HLA-A*31 7 9 0.10 0.26 0.33 0.049 04XXb 0301 0302 rs3132451*G 7 1 0.47 0.10 7.88 0.088 08 0401 0402 HLA-A*24 19 5 0.61 0.15 9.15 1.2 Â 10À4 HLA-B*39 19 4 0.66 0.13 13.30 2.2 Â 10À5 rs2259571*C 4 13 0.12 0.42 0.20 0.021 1302 0102 0604 HLA-A*02 16 11 0.62 0.34 3.05 0.063 HLA-A*03 1 7 0.04 0.22 0.14 0.063 HLA-A*24 5 0 0.19 — — 0.014 15 0102 0602 HLA-B*40 1 4 0.50 0.03 29.50 0.079 XXc 0501 0301 HLA-A*01 2 3 0.50 0.07 14.33 0.045 rs3132451*G 2 4 0.67 0.12 14.54 0.066

Only tests where Po0.10 and number of informative haplotypes 44 are shown, except footnote a. aThe only haplotype with sufficient numbers of the rare rs1800750*A allele. Marker*Allele: tested marker with allele given after the *. T and

NT: number of transmitted and nontransmitted haplotypes, respectively. Tfreq and NTfreq: frequency of transmitted and nontransmitted (informative) haplotypes on the given conditional haplotype. OR: odds ratio. P-value (exact) is for Fisher’s exact test (two-tailed). bDRB1*04XX denotes all DRB1*04 alleles except 0401 and 0404. cDRB1*XX includes alleles 0103, 11, 12 and 1303.

never carried B*39 (only one transmission of A*24-B*39 found exclusively together with B*18, whereas B*18 also on DRB1*1302-DQA1*0102-DQB1*0604). Therefore, at had one transmission (and no nontransmissions) with least some of the association with A*24 appeared rs1800750*G. Although these numbers were too small to independent of B*39. make a distinction, this indicated that the association In contrast to A*24 and B*39, but still in line with the with rs1800750*A could be explained by LD with B*18 LD plots, combined mapping of the predisposing alleles and not the other way around, which was supported by A*24 and B*18 did not suggest any dependency, as only the overall results of the conditional regression analyses four transmitted haplotypes carrying both alleles were (Table 1). observed across all DRB1-DQA1-DQB1 haplotypes (n ¼ 3 Neither the conditional regression results nor the LD on the DRB1*03-DQA1*0501-DQB1*0201 haplotype, patterns suggested dependence between the associations where both showed association; Table 2). As all of the of rs2259571 (AIF1) and HLA-A or -B (Table 1 and Figures implicated HLA-B alleles showed predisposing tenden- 2a–d; D0o0.57 for global LD or between alleles with the cies, these were unlikely to be responsible for the same risk directions). In contrast, strong overall LD was association of the (largely) protective A*01 allele. Also, observed between rs2259571 and the other AIF1 SNP none of the HLA-B alleles were implicated on the DRB1- rs3132451 (especially on transmitted haplotypes; D0 ¼ 1, DQA1-DQB1 haplotypes, where A*02, A*03 or A*31 r2 ¼ 0.23; Figure 2c), in line with the dependency showed transmission distortions, and vice versa for B*40 suggested by the conditional regression results (Table 1). (Table 2). Overall, therefore, with a few exceptions, the However, the individual associations all mapped to associations of HLA-A and -B appeared to be largely different DRB1-DQA1-DQB1 haplotypes (Table 2), and independent. therefore these associations nonetheless appeared to be independent. Association dependencies between SNPs and HLA loci To summarize, the association with rs1800750 (TNF) The predisposing rs1800750*A allele (TNF), which is rare appeared to be secondary to LD with B*18, whereas (frequency in founders: 0.017; Supplementary Table 2), the association with rs3132451 (AIF1) appeared indepen- was mostly transmitted on the DRB1*03-DQA1*0501- dent, but instead may have represented an artifact. DQB1*0201 haplotype (15 transmissions vs only two on Conversely, HLA-A, HLA-B and rs2259571 (AIF1) other haplotypes). The global LD between this SNP and seemed to represent largely independent associations, HLA-B was high, especially on transmitted haplotypes with particularly strong evidence for the A*01 and (Figure 2a; D0 ¼ 0.99), and the allelic LD patterns B*39 alleles. indicated that this was connected with the predisposing B*18 allele (Figure 2c; D0 ¼ 0.86 and r2 ¼ 0.42). Moreover, Replication in T1DGC MHC data set combined mapping revealed that on the DRB1*03- The three SNPs rs1800750, rs3132451 and rs2259571 were DQA1*0501-DQB1*0201 haplotype, rs1800750*A was additionally investigated in an independent data set

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 145 rs1800750 rs2259571 rs3132451 rs2259571 rs3132451 rs1800750 HLA-A HLA-B HLA-B HLA-A DRDQ DRDQ

Global rs3132451*G rs3132451*G rs2259571*C rs2259571*C rs1800750*A rs1800750*A HLA-A*01 HLA-A*24 HLA-A*31 HLA-B*18 HLA-B*39 HLA-B*40 HLA-A*01 HLA-A*24 HLA-A*31 HLA-B*18 HLA-B*39 HLA-B*40 + + + + + + + + + + + +

0.23 0.12

0.08 0.42 0.07

Allelic

0.21 0.16 0.06

Transmitted Non-transmitted

1.00 -1.00

Figure 2 Linkage disequilibrium (LD) between markers and alleles with significant association in the main effects tests. Cf. Figure 1b and Table 1 for selection of markers and alleles; minor alleles are given for SNPs. (a and b) Global LD: D0 values, including DRB1-DQA1-DQB1 haplotype codes (DRDQ). (c and d) Allelic LD: D0 and r2 (only r240.05 is given, as text inside the grid squares). (a) and (c) are for transmitted haplotypes, (b) and (d) for nontransmitted haplotypes. Scale is from 0 to 1 for global D0 and from À1 to 1 for allelic D0, illustrated at the bottom of the figure. Dashed boxes contain no LD values. ‘ þ ’ and ‘À’ denote predisposing and protective effect of the given allele. Note that where a D0-value between a human leukocyte antigen-allele and a SNP allele is negative, D0 given the other single-nucleotide polymorphism allele will be the opposite, positive value and vice versa.

Table 3 Associations of SNPs adjusted for DRB1-DQA1-DQB1 genotypes in the T1DGC MHC data set

Marker All Northern Europe Eastern Europe Southern Europe

nP-value nP-value nP-value nP-value rs1800750a 13 288 9.9 Â 10À6 4616 0.019 1400 0.063 728 0.21 rs3132451 11 872 0.095 4086 0.78 1258 0.71 680 0.080 rs2259571 11 340 0.30 3844 0.21 1188 0.29 680 0.92

All: all samples; Northern, Eastern and Southern Europe: subpopulations. n: number of transmitted and nontransmitted haplotypes. aNot in Hardy–Weinberg equilibrium (P ¼ 3.0 Â 10À6). provided by the T1D Genetics Consortium (T1DGC), conditioned on the DRB1-DQA1-DQB1 haplotypes consisting of genotype data from 2321 multiplex T1D showed similar patterns to those observed in the families. Investigations of HLA-A and -B in this data set Norwegian families, with the only significant association have been described earlier.18 The main effects tests on the DRB1*03-DQA1*0501-DQB1*0201 haplotype adjusted for DRB1-DQA1-DQB1 genotypes left only (Table 4) and apparent dependency on the HLA-B the TNF SNP rs1800750 with a significant result associations on this haplotype (B*18 in particular; data À4 (Pexact ¼ 9.0 Â 10 ; Table 3). However, this SNP was not not shown). Haplotype-based analyses of the two other in Hardy–Weinberg equilibrium in the T1DGC data set SNPs showed that rs2259571, but not rs3132451 (data not À6 (Pexact ¼ 3.0 Â 10 for unrelated patients), which ques- shown), was significantly associated on the same DRB1- tions the validity of this result. Moreover, haplotype- DQA1-DQB1 haplotypes as in the Norwegian families based analyses of rs1800750 with or without HLA-B (Tables 4 and 2).

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 146 Table 4 Allelic tests of rs1800750 and rs2259571 conditional on DRB1-DQA1-DQB1 haplotypes in the T1DGC MHC data set

Conditional haplotype Marker*Allele T NT Tfreq NTfreq OR P-value (exact)

DRB1 DQA1 DQB1

03 0501 0201 rs1800750*A 410 119 0.20 0.15 1.45 9.0 Â 10À4 03 0501 0201 rs2259571*C 72 49 0.04 0.07 0.54 1.1 Â10À3 08 0401 0402 rs2259571*C 34 53 0.24 0.37 0.55 0.028

Only significant results (Po0.05) for DRB1-DQA1-DQB1 haplotypes implicated in the Norwegian families (cf. Table 2) are shown. Cf. Table 2 for definitions.

Discussion with disease status in our study and those of others overlap with those identified with relevance for these In this study, we confirmed T1D associations indepen- specific aspects (that is, A*24, B*18 and B*39 associated dent of the DRB1-DQA1-DQB1 genes for alleles of the both with disease status, accelerated progression and/or HLA-A and HLA-B genes in the MHC class I region and younger age of onset, whereas reversed associations found novel evidence for DRB1-DQA1-DQB1-indepen- have been reported for A*01). Moreover, a direct role for dent T1D association for three SNPs (rs1800750, class I genes has been demonstrated in CD8 þ T-cell rs3132451 and rs2259571) in the TNF cluster of the mediated destruction of b-cells in the nonobese diabetic MHC class III region. In addition, we demonstrated that, mouse model of T1D.28–33 These reports indicate that with the exception of rs1800750, all of these associations HLA-A and/or -B, or other loci in high LD with the showed independence also of each other. associated alleles, may be particularly important in the processes between initiation of autoimmunity and Confirmation of MHC class I associations progression to overt disease. Furthermore, HLA class A DRB1-DQA1-DQB1-independent association between II-independent associations with HLA-A and -B have also T1D and certain HLA-B alleles has been demonstrated in been reported with other AIDs. In particular, HLA-A has several reports,3–6,10,18 but most convincingly by a British been implicated in multiple sclerosis34–36 and juvenile combined case–control/family study3 and in a large idiopathic arthritis,37,38 whereas HLA-B is a well-known family material provided by the T1DGC.18 In both earlier risk factor in ankylosing spondylitis39 and has studies and this study, the most consistent evidence was been implicated in systemic lupus erythematosus.40 This observed for the B*39 and B*18 alleles, with demon- raises the possibility that these genes represent suscept- strated independence of a high number of neighboring ibility loci that are involved in common, autoimmune markers. The combined evidence for HLA-B strongly processes. suggests that this is a primary risk locus in T1D, implying that future T1D studies in the MHC should Novel AIF1 association adjust for LD effects from alleles of this locus in addition In addition to confirm earlier reports for HLA-A and -B, to the DRB1-DQA1-DQB1 loci. we found preliminary, novel evidence for risk effects DRB1-DQA1-DQB1-independent association between both independently of DRB1-DQA1-DQB1, HLA-A and HLA-A and T1D has also been reported earlier,3,19,20 most -B, for rs2259571 in the 50-untranslated region/intron 3 convincingly in the above-mentioned British study, (depending on isoform) of the AIF1 gene. The AIF1 although an independent association could only be protein is involved in inflammatory responses, allograft demonstrated in the case–control materials, and not in rejection and macrophage regulation,41 and has been families.3 The alleles showing individual significant implicated in T1D by functional studies, showing associations in our study (A*01, A*24 and A*31) also regulatory effects on insulin secretion in mice and match with this last report, and partly (A*01 and/or accumulation in the pancreas of prediabetic Bio-Breeding A*24) with earlier reports.19,20 Hence, the present results (BB) rats.42 Also, a number of studies have implicated show that the DRB1-DQA1-DQB1-independent T1D this protein in other AIDs,41 including a recent report association to alleles at the HLA-A locus is consistent showing that AIF1 may be crucial in the pathogenesis of across different populations and provides further evi- rheumatoid arthritis.43 In addition to rs2259571, we dence for an association also in families. Importantly, our observed an independent association of rs3132451 in results provide confirmatory evidence that the HLA-A the promoter of the same gene, but this result was most associations are largely independent of HLA-B. Even so, likely an artifact, both judging from the heterogeneity recent studies have indicated that the HLA-A associa- within each of the two individual DRB1-DQA1-DQB1 tions may be secondary to LD with another, yet haplotypes the associations mapped to and the negative unconfirmed, locus telomeric of this gene,4,18 which result in the T1DGC families. The replication attempt in cannot be excluded with our results. the T1DGC data set was also negative for rs2259571 for Alleles of the HLA-A and -B loci have been associated, the regression analyses, raising the possibility of a type I independently of the DRB1-DQA1-DQB1 genes, with error. However, when mapping the allele distribution on accelerated b-cell destruction and progression to clinical the different DRB1-DQA1-DQB1 haplotypes in the disease in T1D patients (most notably, A*24),21–24 as well T1DGC families, rs2259571 showed significant associa- as with variable age of onset.3,6,25–27 Although our study tions on the same haplotypes as in the Norwegian did not allow for appropriate testing of these effects (see families. Therefore, this SNP could still mark the same Materials and methods), most of the alleles associated effect in both data sets, although the SNP itself is not

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 147 most likely to be a primary locus. A possible explanation Similar to T1D, many AIDs show indications of for the lack of association in the regression analyses multiple susceptibility loci within the MHC, and thus could be that the underlying LD pattern with the share many of the challenges in efforts to fine-map this primary locus was population specific, and therefore region. Each of these loci are most likely to act as could be weaker in a relatively heterogeneous material, confounding factors, illustrated by the growing body of such as the T1DGC families, which, although mostly evidence for T1D associations attributable to LD with Caucasoid, contains samples from all over Europe, North HLA-B alleles, in addition to the known primary America and Australia. However, this question can only susceptibility loci DRB1-DQA1-DQB1. Moreover, defini- be answered by replication studies and further fine- tions of risk alleles at identified primary loci may vary mapping in other, preferably homogenous populations. (for example, controversies about the ‘shared epitope’ hypothesis in rheumatoid arthritis45), possibly leading to Associations limited to a subset of DRB1-DQA1-DQB1 insufficient adjustment for confounding factors even if haplotypes conditional approaches are used. Hence, conflicting None of the associated alleles of HLA-A, HLA-B or the association reports in this region are not surprising. AIF1 SNP rs2259571 showed signs of frequency devia- Our approach to these problems has been to select tions across all DRB1-DQA1-DQB1 haplotypes, which candidate polymorphisms on the basis of reports with at could indicate specific epistatic effects with the class II least some form of adjustment for known MHC risk loci. loci. However, other possible explanations include This strategy allowed us to examine the results of these limited power resulting from stratification according to reports with respect to each other and increased the DRB1-DQA1-DQB1 haplotypes or, in the event that the a priori likelihood (and thus reduced the expected type I markers do not constitute real, etiological loci, complex error rate) compared with a hypothesis-free screening underlying LD patterns that make the associations approach. Moreover, our strategy employed two com- visible only on certain DRB1-DQA1-DQB1 haplotypes. plementary conditional approaches, increasing the va- Answers to these questions can only be given after more lidity of our results. Both of these considered all definite evidence is obtained for these markers, in DRB1-DQA1-DQB1 haplotype variants, which is justified particular rs2259571 and HLA-A, and also for certain by the observed risk-continuum from predisposing to HLA-B alleles, and/or the real etiological loci have been protective haplotypes of these loci in T1D.1,2 identified.

No independent T1D association with other previously Conclusion reported AID-associated variants In conclusion, our findings support the association No evidence for DRB1-DQA1-DQB1-independent asso- between alleles at the HLA-B locus (in particular, B*18 ciation was found for polymorphisms in the other and B*39) and T1D and provides further evidence for an candidate genes, including HLA-E, MICA, MICB, BAT1, independent association with HLA-A alleles (in particu- NFKBIL1, LTA, HSPA1L/B, C2, CFB, AGER and BTNL2. lar, A*01, A*24 and A*31), or alleles at an unidentified This was despite the fact that a large number of these locus in high LD with these alleles. In addition, a novel 0 polymorphisms showed highly significant association independent association of rs2259571 in intron 3/5 - before adjusting for LD with the DRB1-DQA1-DQB1 loci, untranslated region of the AIF1 gene was observed, demonstrating the necessity of a conditional approach warranting follow-up studies in other populations. throughout the MHC. Several of the genotyped loci, including variants in the BAT1, NFKBIL1, LTA, TNF and AIF1 genes, are located within the TNF cluster in the Materials and methods MHC class III region. This unusually dense cluster of genes has been implicated in predisposition for T1D and Family materials other AIDs in numerous reports (for example, Supple- A total of 434 Norwegian T1D families were recruited as mentary Table 1), with particular focus on the TNF gene described elsewhere46 and patients diagnosed before the itself, partly because of the demonstrated clinical benefits age of 15 years according to the EURODIAB criteria.47 of TNF blockers in several AIDs. A significant DRB1- The inclusion only of patients with a relatively young age DQA1-DQB1-independent association was indeed ob- of onset is most likely to result in a clinically homo- served for rs1800750 (also known as TNF À376) in this geneous patient population, presumably excluding gene, both in the Norwegian and the T1DGC families. atypical adult manifestations of T1D (including latent However, the observed association with rs1800750 could autoimmune diabetes in adults48). Even though associa- apparently be explained by LD with the B*18 allele in tions between MHC loci and age of onset have been both data sets. Moreover, none of the other polymorphi- suggested,3,6,25–27 such analyses were not justified in this sms earlier reported to be independently associated, study, as our analyses already included a large number most notably rs1800629 (TNF -308), showed evidence of of parameters in a relatively small population, with independent association in our study. This is in agree- resulting loss of statistical power. Moreover, the range of ment with a recent report, showing that apparent age of onset in our study was small and would therefore association of rs909253 (LTA NcoI RFLP), rs1800629 have allowed only for limited tests of this effect. The (TNF À308) and rs361525 (TNF À238), all of which were families mainly consisted of trios with one affected child also investigated in this study, could be explained by LD and both parents, but 20 families had two or more with either DRB1-DQB1 haplotypes or the B*18 allele.44 siblings included, for a total of 1316 individuals. For These observations further illustrate our earlier point of replication studies of rs1800750, rs3132451 and adjusting also for HLA-B for T1D association studies in rs2259571, we used data from the T1DGC MHC data the MHC. set (2007.02.MHC). The editing and recoding of this data

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 148 set has been described elsewhere,18 resulting in a data set could indicate a sample switch or erroneous family with a total of 2301 multiplex T1D families. member assignment), the affected sample or (in cases where the responsible sample was ambiguous) the entire Genotyping family was removed from analysis (38 samples in 15 Genotyping data and pedigree information were families, 11 of these families with all members removed). handled using Progeny Lab v6 (Progeny Software LLC, Genotypes for two SNPs in the MICA gene were South Bend, IN, USA). Genotyping was performed on removed because of high Mendelian error rates in the whole genome-amplified DNA (GenomiPhi DNA Am- remaining families. Remaining errors were removed for plification Kit, GE Healthcare, Piscataway, MD, USA). the affected polymorphism in the entire involved family. A total of 58 SNPs were genotyped using SNPlex Hardy–Weinberg equilibrium was calculated in unre- (Applied Biosystems, Foster City, CA, USA), TaqMan lated subjects using the exact test implemented in the (Applied Biosystems) or MassARRAY technology (Se- program Pedstats.50 One SNP showed significant devia- quenom, San Diego, CA, USA), and four insertions– tion from expected genotype frequencies (Po0.01) and deletions (indels) using fragment length analysis with was removed. Details of the removed markers are fluorescence-labelled, tailed primer pairs on an ABI3730 available in Supplementary Table 2. automated sequencer (Applied Biosystems). Genotyping The above-mentioned procedures resulted in a data set with SNPlex and MassARRAY technology was per- with 1278 samples in 423 families, with genotypes for 46 formed at CIGENE (A˚ s, Norway). Detailed reaction SNPs and 4 indels, in addition to HLA-A, HLA-B and the setup and primer sequences are available in Supplemen- DRB1-DQA1-DQB1 haplotype codes. tary Text 1. Genotyping of HLA-A and HLA-B was performed using sequence-based typing protocols and primers amplifying exons 2 and 3, using the Platinum Statistical analysis Taq PCRx DNA polymerase kit (Invitrogen, Carlsbad, Transmission-disequilibrium test was performed using CA, USA). Primers and detailed PCR conditions are the extended transmission-disequilibrium test imple- available from the authors upon request. Sequencing was mented in UNPHASED v2.403.51 Conditional logistic performed using BigDye Terminator v3.1 chemistry regression, adjusted for DRB1-DQA1-DQB1 genotypes, (Applied Biosystems) on an ABI3730 automated sequen- was used for evaluation of the importance of SNPs, cer. Sequence alignment and allele assignment were indels and allelic variants at the HLA-A, and -B loci on performed using Assign v3.2.7 software (Conexio Geno- T1D susceptibility. We used the conditioning strategy 4 mics, Applecross, WA, Australia). In addition, some described by Cordell and Clayton,52 implemented in samples were genotyped using RELI SSO HLA-A and -B STATA (STATA Corp., College Station, TX, USA). This Typing Kits (Invitrogen) to validate results. Genotyping method involves ‘main effects’ tests, where comparisons of HLA-DQA1, -DQB1 and -DRB1 has been performed are made between regression models including the earlier using several protocols, as described elsewhere.46 effects of a primary locus and models including the In addition, a few remaining samples with DRB1*04 effects of the primary locus and one or more additional alleles at two-digit resolution were genotyped in the loci. A slightly modified version of the main effects test context of this study using sequence-based typing, as by Cordell and Clayton, implemented in UNPHASED, described above, to increase resolution to four digits. was used for evaluation of the rs1800750, rs3132451 and rs2259571 SNPs in the T1DGC data set. DRB1-DQA1-DQB1 haplotype codes Global and allelic LD measures (D0 and D0 and r2, Genotypes for HLA-DRB1, -DQA1 and -DQB1 were respectively) were calculated using UNPHASED with replaced by a haplotype code spanning all three loci, expectation-maximization estimation. Raw data plots with phase inferred on the basis of common haplotypes. were generated in GOLD53 and edited using Adobe Mendelian consistency (verified with Pedcheck49) was Illustrator. evaluated as a quality control. A total of 99.9% of the Details of the haplotype-based methods have been samples had genotypes at the HLA-DQA1 and -DQB1 described elsewhere.18 Briefly, maximum-likelihood hap- loci (four-digit resolution), but only 64.7% had genotypes lotype frequency estimates combining markers with at the HLA-DRB1 locus (mixed resolution, 2–6 digits), DRB1-DQA1-DQB1 haplotypes were computed using and of these, 40.6% were conclusively typed for only one FAMHAP.54 On the basis of these estimates, tables with allele. However, due to strong LD, the remaining haplotype transmission/nontransmissions from hetero- genotypes at this locus could be inferred from haplo- zygous parents were constructed as described else- types of the two other loci (the exception was DRB1*04 where,54 and haplotypes were organized in separate subtypes, which was always determined by direct groups for each DRB1-DQA1-DQB1 haplotype. Statisti- genotyping), resulting in a total ‘genotyping’ rate of the cally significant deviations of one marker allele at a time DRB1-DQA1-DQB1 haplotype code equal to that of the or haplotypes of two markers (combined mapping) HLA-DQA1 and -DQB1 loci. The rationale behind this within each of these DRB1-DQA1-DQB1 groups were method has been discussed earlier.18 tested in 2 Â 2 contingency tables with Fisher’s exact test (two-tailed; using EpiCalc 2000; http://www.brixton- Quality control and pruning health.com/epicalc.html) and odds ratios were calcu- Markers with genotyping call rates below 79% (n ¼ 1), lated. For combined mapping, the resulting small bad allele clustering (n ¼ 2) or that were monomorphic numbers among individual haplotypes did not allow (n ¼ 6) were removed from further analyses. Mendelian for conditional tests adjusting for more than the DRB1- inheritance was verified using Pedcheck. In cases where DQA1-DQB1 loci. However, colocalization of predispos- an unusually high amount of Mendelian errors across all ing or protective alleles at two different loci on markers were observed within a single family (which transmitted or nontransmitted haplotypes, respectively,

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 149 can still reveal whether these alleles were most likely to 10 Nejentsev S, Gombos Z, Laine AP, Veijola R, Knip M, Simell O represent the same association. et al. Non-class II HLA gene associated with type 1 diabetes Bonferroni-correction for multiple testing (n ¼ 53) was maps to the 240-kb region near HLA-B. Diabetes 2000; 49: performed only for the unconditional transmission- 2217–2221. disequilibrium test analyses due to limited statistical 11 Johansson S, Lie BA, Todd JA, Pociot F, Nerup J, Cambon- power in the conditional analyses. This is also justified Thomsen A et al. Evidence of at least two type 1 diabetes by our approach, investigating candidate polymorphi- susceptibility genes in the HLA complex distinct from HLA- sms and/or genes, which results in an elevated a priori DQB1, -DQA1 and -DRB1. Genes Immun 2003; 4: 46–53. 12 Zavattari P, Lampis R, Motzo C, Loddo M, Mulargia A, Whalen likelihood of association. M et al. Conditional linkage disequilibrium analysis of a complex disease superlocus, IDDM1 in the HLA region, reveals the presence of independent modifying gene effects influencing the type 1 diabetes risk encoded by the major HLA- Acknowledgements DQB1, -DRB1 disease loci. Hum Mol Genet 2001; 10: 881–889. 13 Horton R, Wilming L, Rand V, Lovering RC, Bruford EA, This study was supported by JDRF Grant 1-2004-793, by Khodiyar VK et al. Gene map of the extended human MHC. the Novo Nordisk Foundation and the Norwegian Nat Rev Genet 2004; 5: 889–899. Diabetes Association. M Olsson was supported by SSF 14 Becker KG, Simon RM, Bailey-Wilson JE, Freidlin B, Biddison Grant A3 02:129. We thank CIGENE, A˚ s, Norway, WE, McFarland HF et al. Clustering of non-major histocom- for performing the genotyping with the SNPlex and patibility complex susceptibility candidate loci in human MassARRAY assays and Linda Haugse for assistance in autoimmune diseases. Proc Natl Acad Sci USA 1998; 95: HLA genotyping. This research utilizes resources pro- 9979–9984. vided by the Type 1 Diabetes Genetics Consortium, a 15 Pearce SH, Merriman TR. Genetic progress towards the molecular basis of autoimmunity. Trends Mol Med 2006; 12: collaborative clinical study sponsored by the National 90–98. Institute of Diabetes and Digestive and Kidney Diseases 16 Allcock RJ, Windsor L, Gut IG, Kucharzak R, Sobre L, Lechner (NIDDK), National Institute of Allergy and Infectious D et al. High-density SNP genotyping defines 17 distinct Diseases (NIAID), National Human Genome Research haplotypes of the TNF block in the Caucasian population: Institute (NHGRI), National Institute of Child Health implications for haplotype tagging. Hum Mutat 2004; 24: and Human Development (NICHD) and Juvenile Dia- 517–525. betes Research Foundation International (JDRF) and 17 Thomson G, Barcellos LF, Valdes AM. Searching for additional supported by U01 DK062418. disease loci in a genomic region. Adv Genet 2008; 60: 253–292. 18 Eike MC, Becker T, Humphreys K, Olsson M, Lie BA. Conditional analyses on the T1DGC MHC dataset: novel associations with type 1 diabetes around HLA-G and References confirmation of HLA-B. Genes Immun (advance online pub- lication, 2 October 2008; doi:10.1038/gene.2008.74). 1 Erlich H, Valdes AM, Noble J, Carlson JA, Varney M, 19 Noble JA, Martin A, Valdes AM, Lane JA, Galgani A, Petrone Concannon P et al. HLA DR-DQ haplotypes and genotypes A et al. Type 1 diabetes risk for human leukocyte antigen and type 1 diabetes risk: analysis of the type 1 diabetes (HLA)-DR3 haplotypes depends on genotypic context: asso- genetics consortium families. Diabetes 2008; 57: 1084–1092. ciation of DPB1 and HLA class I loci among DR3- and DR4- 2 Thomson G, Valdes AM, Noble JA, Kockum I, Grote MN, matched Italian patients and controls. Hum Immunol 2008; 69: Najman J et al. Relative predispositional effects of HLA class II 291–300. DRB1-DQB1 haplotypes and genotypes on type 1 diabetes: a 20 Noble JA, Valdes AM, Bugawan TL, Apple RJ, Thomson G, meta-analysis. Tissue Antigens 2007; 70: 110–127. Erlich HA. The HLA class I A locus affects susceptibility to 3 Nejentsev S, Howson JM, Walker NM, Szeszko J, Field SF, type 1 diabetes. Hum Immunol 2002; 63: 657–664. Stevens HE et al. Localization of type 1 diabetes susceptibility 21 Tait BD, Colman PG, Morahan G, Marchinovska L, Dore E, to the MHC class I genes HLA-B and HLA-A. Nature 2007; Gellert S et al. HLA genes associated with autoimmunity and 450: 887–892. progression to disease in type 1 diabetes. Tissue Antigens 2003; 4 Aly TA, Baschal EE, Jahromi MM, Fernando MS, Babu SR, 61: 146–153. Fingerlin TE et al. Analysis of SNPs identifies major type 1A 22 Nakanishi K, Kobayashi T, Murase T, Naruse T, Nose Y, Inoko diabetes locus telomeric of the MHC. Diabetes 2008; 57: H. Human leukocyte antigen-A24 and -DQA1*0301 in 770–776. Japanese insulin-dependent diabetes mellitus: independent 5 Nejentsev S, Reijonen H, Adojaan B, Kovalchuk L, Sochnevs contributions to susceptibility to the disease and additive A, Schwartz EI et al. The effect of HLA-B allele on the IDDM contributions to acceleration of beta-cell destruction. J Clin risk defined by DRB1*04 subtypes and DQB1*0302. Diabetes Endocrinol Metab 1999; 84: 3721–3725. 1997; 46: 1888–1892. 23 Nakanishi K, Kobayashi T, Murase T, Nakatsuji T, Inoko H, 6 Valdes AM, Erlich HA, Noble JA. Human leukocyte antigen Tsuji K et al. Association of HLA-A24 with complete beta-cell class I B and C loci contribute to type 1 diabetes (T1D) destruction in IDDM. Diabetes 1993; 42: 1086–1093. susceptibility and age at T1D onset. Hum Immunol 2005; 66: 24 Nakanishi K, Inoko H. Combination of HLA-A24, -DQA1*03, 301–313. and -DR9 contributes to acute-onset and early complete beta- 7 Aly TA, Ide A, Jahromi MM, Barker JM, Fernando MS, Babu cell destruction in type 1 diabetes: longitudinal study of SR et al. Extreme genetic risk for type 1A diabetes. Proc Natl residual beta-cell function. Diabetes 2006; 55: 1862–1868. Acad Sci USA 2006; 103: 14074–14079. 25 Tait BD, Harrison LC, Drummond BP, Stewart V, Varney MD, 8 Cheong KY, Allcock RJ, Eerligh P, Witt CS, Christiansen FT, Honeyman MC. HLA antigens and age at diagnosis of McCann V et al. Localization of central MHC genes influencing insulin-dependent diabetes mellitus. Hum Immunol 1995; 42: type I diabetes. Hum Immunol 2001; 62: 1363–1370. 116–122. 9 Lie BA, Todd JA, Pociot F, Nerup J, Akselsen HE, Joner G et al. 26 Fujisawa T, Ikegami H, Kawaguchi Y, Yamato E, Takekawa K, The predisposition to type 1 diabetes linked to the human Nakagawa Y et al. Class I HLA is associated with age-at-onset leukocyte antigen complex includes at least one non-class II of IDDM, while class II HLA confers susceptibility to IDDM. gene. Am J Hum Genet 1999; 64: 793–800. Diabetologia 1995; 38: 1493–1495.

Genes and Immunity Conditional MHC screen of T1D candidate loci MC Eike et al 150 27 Valdes AM, Thomson G, Erlich HA, Noble JA. Association several predisposing loci. Rheumatology (Oxford) 2005; 44: between type 1 diabetes age of onset and HLA among sibling 1368–1373. pairs. Diabetes 1999; 48: 1658–1661. 41 Liu G, Ma H, Jiang L, Zhao Y. Allograft inflammatory 28 de Jersey J, Snelgrove SL, Palmer SE, Teteris SA, Mullbacher factor-1 and its immune regulation. Autoimmunity 2007; 40: A, Miller JF et al. Beta cells cannot directly prime diabetogenic 95–102. CD8 T cells in nonobese diabetic mice. Proc Natl Acad Sci USA 42 Chen ZW, Ahren B, Ostenson CG, Cintra A, Bergman T, 2007; 104: 1295–1300. Moller C et al. Identification, isolation, and characterization of 29 Kay TW, Parker JL, Stephens LA, Thomas HE, Allison J. RIP- daintain (allograft inflammatory factor 1), a macrophage beta 2-microglobulin transgene expression restores insulitis, polypeptide with effects on insulin secretion and abundantly but not diabetes, in beta 2-microglobulin null nonobese present in the pancreas of prediabetic BB rats. Proc Natl Acad diabetic mice. J Immunol 1996; 157: 3688–3693. Sci USA 1997; 94: 13879–13884. 30 Utsugi T, Yoon JW, Park BJ, Imamura M, Averill N, Kawazu S 43 Kimura M, Kawahito Y, Obayashi H, Ohta M, Hara H, Adachi et al. Major histocompatibility complex class I-restricted T et al. A critical role for allograft inflammatory factor-1 in the infiltration and destruction of pancreatic islets by NOD pathogenesis of rheumatoid arthritis. J Immunol 2007; 178: mouse-derived beta-cell cytotoxic CD8+ T-cell clones in vivo. 3316–3322. Diabetes 1996; 45: 1121–1131. 44 Noble JA, Valdes AM, Lane JA, Green AE, Erlich HA. Linkage 31 Sumida T, Furukawa M, Sakamoto A, Namekawa T, Maeda T, disequilibrium with predisposing DR3 haplotypes accounts Zijlstra M et al. Prevention of insulitis and diabetes in beta 2- for apparent effects of tumor necrosis factor and lymphotoxin- microglobulin-deficient non-obese diabetic mice. Int Immunol alpha polymorphisms on type 1 diabetes susceptibility. Hum 1994; 6: 1445–1449. Immunol 2006; 67: 999–1004. 32 Serreze DV, Leiter EH, Christianson GJ, Greiner D, Roopenian 45 Morgan AW, Haroon-Rashid L, Martin SG, Gooi HC, DC. Major histocompatibility complex class I-deficient NOD- Worthington J, Thomson W et al. The shared epitope B2mnull mice are diabetes and insulitis resistant. Diabetes hypothesis in rheumatoid arthritis: evaluation of alternative 1994; 43: 505–509. classification criteria in a large UK Caucasian cohort. Arthritis 33 Wicker LS, Leiter EH, Todd JA, Renjilian RJ, Peterson E, Rheum 2008; 58: 1275–1283. Fischer PA et al. Beta 2-microglobulin-deficient NOD mice do 46 Bjørnvold M, Undlien DE, Joner G, Dahl-Jørgensen K, not develop insulitis or diabetes. Diabetes 1994; 43: 500–504. Njølstad PR, Akselsen HE et al. Joint effects of HLA, INS, 34 Brynedal B, Duvefelt K, Jonasdottir G, Roos IM, Akesson E, PTPN22 and CTLA4 genes on the risk of type 1 diabetes. Palmgren J et al. HLA-A confers an HLA-DRB1 independent Diabetologia 2008; 51: 589–596. influence on the risk of multiple sclerosis. PLoS ONE 2007; 2: 47 EURODIAB ACE Study Group. Variation and trends in e664. incidence of childhood diabetes in Europe. EURODIAB ACE 35 Fogdell-Hahn A, Ligers A, Gronning M, Hillert J, Olerup O. Study Group. Lancet 2000; 355: 873–876. Multiple sclerosis: a modifying influence of HLA class I genes 48 Cervin C, Lyssenko V, Bakhtadze E, Lindholm E, Nilsson P, in an HLA class II associated autoimmune disease. Tissue Tuomi T et al. Genetic similarities between latent autoimmune Antigens 2000; 55: 140–148. diabetes in adults, type 1 diabetes, and type 2 diabetes. 36 Harbo HF, Lie BA, Sawcer S, Celius EG, Dai KZ, Oturai A et al. Diabetes 2008; 57: 1433–1437. Genes in the HLA class I region may contribute to the HLA 49 O’Connell JR, Weeks DE. PedCheck: a program for identifica- class II-associated genetic susceptibility to multiple sclerosis. tion of genotype incompatibilities in linkage analysis. Am J Tissue Antigens 2004; 63: 237–247. Hum Genet 1998; 63: 259–266. 37 Zeggini E, Donn RP, Ollier WE, Thomson W. Evidence for 50 Wigginton JE, Abecasis GR. PEDSTATS: descriptive statistics, linkage of HLA loci in juvenile idiopathic oligoarthritis: graphics and quality assessment for gene mapping data. independent effects of HLA-A and HLA-DRB1. Arthritis Bioinformatics 2005; 21: 3445–3447. Rheum 2002; 46: 2716–2720. 51 Dudbridge F. Pedigree disequilibrium tests for multilocus 38 Hall PJ, Burman SJ, Laurent MR, Briggs DC, Venning HE, haplotypes. Genet Epidemiol 2003; 25: 115–121. Leak AM et al. Genetic susceptibility to early onset pauciarti- 52 Cordell HJ, Clayton DG. A unified stepwise regression cular juvenile chronic arthritis: a study of HLA and comple- procedure for evaluating the relative effects of polymorphisms ment markers in 158 British patients. Ann Rheum Dis 1986; 45: within a gene using case/control or family data: application to 464–474. HLA in type 1 diabetes. Am J Hum Genet 2002; 70: 124–141. 39 Rahman P. Genetics of ankylosing spondylitis: an update. Curr 53 Abecasis GR, Cookson WO. GOLD—graphical overview of Rheumatol Rep 2007; 9: 383–389. linkage disequilibrium. Bioinformatics 2000; 16: 182–183. 40 Smerdel-Ramoya A, Finholt C, Lilleby V, Gilboe IM, Harbo 54 Becker T, Knapp M. Maximum-likelihood estimation of HF, Maslinski S et al. Systemic lupus erythematosus and the haplotype frequencies in nuclear families. Genet Epidemiol extended major histocompatibility complex—evidence for 2004; 27: 21–32.

Supplementary Information accompanies the paper on Genes and Immunity website (http://www.nature.com/gene)

Genes and Immunity