FULL PAPER Genotype effects and epistasis in type 1 diabetes and HLA-DQ trans dimer associations with disease

(2004) 5, 381 388 & 2004 Nature Publishing Group All rights reserved 1466-4879/04 $30.00 www.nature.com/gene FULL PAPER Genotype effects and epistasis in type 1 diabetes and HLA-DQ trans dimer associations with disease BPC Koeleman 1,2, BA Lie 3, DE Undlien 4, F Dudbridge 5, E Thorsby 3, RRP de Vries 1, F Cucca 6, BO Roep 1, MJ Giphart 1 and JA Todd 7 1 Department of ImmunoHaematology and Blood Transfusion, Leiden University Medical Center, Leiden, The Netherlands; 2 Department of Medical Genetics, University Medical Center, Utrecht, The Netherlands; 3 Institute of Immunology, Rikshospitalet University Hospital, Oslo, Norway; 4 Institute of Medical Genetics, Ulleval University Hospital, University of Oslo, Norway; 5 MRC Human Genome Mapping Project Resource Centre, Cambridge, UK; 6 Dipartimento di Scienze Biomediche e Biotecnologie, Universita di Cagliari, Italy; 7 Juvenile Diabetes Research Foundation/Wellcome Trust Diabetes and Inflammation Laboratory, Cambridge Institute for Medical Research, University of Cambridge, UK Alleles of HLA class II genes DQB1, DQA1, and DRB1 in the MHC region are major determinants of genetic predisposition to type 1 diabetes (T1D). Several alleles of each of these three loci are associated with susceptibility or protection from disease. In addition, relative risks for some DR-DQ genotypes are not simply the sum or product of the single haplotype relative risks. For example, the risk of the DRB1*03-DQB1*02/DRB1*0401-DQB1*0302 genotype is often found to be higher than for the individual DRB1*03-DQB1*02 and DRB1*0401-DQB1*0302 homozygous genotypes. It has been hypothesized that this synergy or epistasis occurs through formation of highly susceptible trans-encoded HLA-DQ(a1, b1) heterodimers. Here, we evaluated this hypothesis by estimating the disease associations of the range of DR-DQ genotypes and their inferred dimers in a large collection of nuclear families. We determined whether the risk of haplotypes in DRB1*0401-DQB1*0302-positive genotypes relative to the DRB1*03-DQB1*02-positive genotypes is different from that of DRB1*01-DQB1*0501, which we used as a baseline reference. Several haplotypes showed a different risk compared to DRB1*01-DQB1*0501, which correlated with their ability to form certain trans-encoded DQ dimers. This result provides new evidence for the potential importance of transencoded HLA DQ molecules in the determination of HLA-associated risk in T1D. (2004) 5, 381 388. doi:10.1038/sj.gene.6364106 Published online 27 May 2004 Keywords: HLA; type 1 diabetes; genotype risk Introduction The association between genes in the class II region of the HLA complex on chromosome 6p (IDDM1) and type 1 diabetes (T1D), first described in 1974, is one of the most extensively studied genetic susceptibility factors for a complex disease. 1 4 A large number of publications have established that particular combinations of alleles at HLA-DQB1, -DQA1, and -DRB1 show strong association with T1D. 5 12 The complex nature of this association is partly caused by the presence of multiple susceptible and protective HLA class II haplotypes, and may be modified by the presence of multiple other loci in the MHC region. 13,14 The haplotypes DRB1*0401-DQA1*0301-DQB1*0302 and DRB1*0301-DQA1*0501-DQB1*0201 have been found to be the major susceptible haplotypes in most populations, whereas DRB1*1501-DQA1*0102-DQB1* Correspondence: Dr BPC Koeleman, Department of Medical Genetics, PO Box 85090, NL-3508 AB, Utrecht, The Netherlands. E-mail: b.p.c.koeleman@med.uu.nl Received 11 February 2004; revised 08 April 2004; accepted 15 April 2004; published online 27 May 2004 0602 has been described to be protective in a dominant manner. 2 Many previous studies have measured the disease risk of DR-DQ haplo- and genotypes by odds ratios (ORs) and have categorized the haplo- or genotypes in terms of susceptibility, neutrality, and protection. 2,6,7,10,11 Inferences have also been made concerning the strength of susceptibility or protection, for example, DRB1*04 subtypes have been shown to display different ORs on an identical DQA1*0301-DQB1*0302 haplotype, with ranking 04054040140404404034 0406 8. 2,4,15 However, the presence of multiple diseaseassociated haplotypes complicates the analysis of the DR-DQ loci. Most of the studies cited above use ORs to evaluate the risk of haplotypes or genotypes. Several authors have recognized that the OR of one haplotype can be influenced by the risk associated with the other haplotype present in an individual, for example, ORs of neutral factors can be significantly decreased due to the increased frequency of a strong susceptibility factor. 2 To address this problem, several other measures have been proposed, like attributable risk and relative predispositional effect. 16,17 However, these measures only partly solve the problem. Here, relative risks are estimated

382 relative to a baseline reference haplo- or genotype, which alleviates the above problem. Furthermore, we estimate the risk conferred by one DR-DQ haplotype or genotype by family-based logistic regression conditional on parental genotypes, which circumvents assumption of Hardy Weinberg (HW) equilibrium and random mating in the general population. 18,19 Finally, this family-based approach and conditioning on parental genotypes avoids inaccurate risk estimation due to population admixture effects or ascertainment on multiple affected siblings, which can affect case control and family-based AFBAC designs. 20 It is evident that the risks of HLA genotypes are not simply the sum or product of the single haplotype relative risks. Apart from their incomplete disease penetrance, the risk of the DRB1*03-DQB1*02/DRB1* 0401-DQB1*0302 heterozygote genotype was found to be increased compared to that of the individual DRB1*03- DQB1*02 and DRB1*0401-DQB1*0302 homozygous genotypes, indicating an epistatic or synergistic effect. 4,21 Other genotype effects have been found for DRB1*08- DQB1*0402 (increased risk when together with DRB1*0401-DQB1*0302) and DRB1*09-DQB1*0303 (increased risk mainly with DRB1*03-DQB1*02), 6,22,23 occurring fairly consistently across several ethnic groups. 4 It has been postulated that the unexpected increase in risk of these genotypes is caused by highly susceptible transencoded DQ(a1, b1) heterodimers as formed in the DRB1*03-DQB1*02/DRB1*0401-DQB1*0302 genotype. 4,24 The DQ molecule is an important antigen-presenting molecule for CD4-positive T cells, and consists of a DQ(a1, b1) heterodimer for which the DQb1 and DQa1 chains are encoded by the DQB1 and DQA1 genes, respectively. Both cis- and trans-heterodimer molecules are expressed in class II heterozygous individuals, for example, in the high-risk DRB1*03-DQB1*02/DRB1* 0401-DQB1*0302 heterozygotes. 25 To explain the genotype effect found for DRB1*03-DQB1*02/DRB1*0401- DQB1*0302, the trans-encoded dimers, namely DQ(a1* 0501, b1*0302) and DQ(a1*0301, b1*0201), have been proposed to confer higher risk than the cis dimers, DQ(a1*0301, b1*0302) and DQ(a1*0501, b1*0201). It has been postulated that similar molecules are formed in trans in DRB1*0401-DQB1*0302/DRB1*08-DQB1*0402 (DQ(a1*0401, b1*0302)-positive individuals, with a1*0401 being structurally similar to a1*0501, and for DRB1*03-DQB1*02/DRB1*09-DQB1*0303 (DQ(a1* 0301, b1*0201)) genotypes, 4,24 supporting the hypothesis that these trans-encoded molecules confer high risk. The aim of this study was to estimate the relative risk (RR) for disease for HLA class II haplo- and genotypes and determine whether genotype effects indicative of trans effects are present for genotypes other than those previously described. The results indicate that DQ trans dimer formation could be an important factor in the association of HLA with T1D. Results Haplotype risks in separate populations Populations were first analysed separately and DRB1- DQB1 haplotype ORs were found comparable in different populations (Table 1). Only haplotypes with counts larger than 10 were analysed. No genotype analysis in separate populations was performed, since the result was unreliable because of limited number of observations. Differences between populations in magnitude and rank of haplotype risk were observed, although the main features of the HLA class II associations were maintained: DRB1*04-DQB1*0302 haplotypes were universally the most potent source of susceptibility except for those with the DRB1*0403 allele, and equally significant disease protection compared to DRB1*01- DQB1*05 haplotypes by the common DRB1*11-DQB1* 0301 and DRB1*15-DQB1*06 haplotypes. Table 2 shows the DRB1-DQB1 haplotype RRs for the populations combined. Overall, this analysis confirms susceptible and protective haplotypes reported in many previous studies, with major high-risk haplotypes DRB1* 03-DQB1*02, DRB1*0401-DQB1*0302, and protective haplotypes DRB1*11-DQB1*0301, DRB1*14-DQB1*05, DRB1*07-DQB1*09, DRB1*0407-DQB1*0301, and DRB1 *15-DQB1*06 (Table 2). The DRB1*04-DQB1*0302 subtypes were ranked from most to least susceptible: *0401Z*04054*0404B*04024*0403. Genotype risks The number of observations for each individual data set was too small to estimate genotype risks. We, therefore, used conditional extended transmission disequilibrium test (CETDT) to avoid the problem of inaccurate risk estimation due to population admixture or ascertainment bias and combined the data from the different populations to calculate genotype OR estimates (relative transmission probability (RTP) and OR-r), relative to the reference DRB1*01-DQB1*05/DRB1*01-DQB1*05 genotype (Table 3). This reference genotype was chosen as it represents the most frequent and relatively neutral genotype. No major differences were observed when a different reference was chosen (not shown). As expected, the DRB1*0401-DQB1*0302/DRB1*03-DQB1*02 genotype shows the highest RTP (RTP ¼ 33), which was significantly different from the DRB1*0401-DQB1*0302 homozygous (RTP ¼ 26.6) and the DRB1*03-DQB1*02/DRB1* 03-DQB1*02 genotype (RTP ¼ 8.1) (P ¼ 1 10 5 and 1.4 10 5, respectively) (Table 3). This observation confirms the epistatic synergy between DRB1*0401- DQB1*0302 and DRB1*03-DQB1*02 haplotypes. Significant susceptibility was also found for DRB1*0401- DQB1*0302/DRB1*08-DQB1*04 (RTP ¼ 17.4), DRB1* 0401-DQB1*0302/DRB1*01-DQB1*05 (RTP ¼ 8.8), DRB1* 0401-DQB1*0302/DRB1*13-DQB1*06 (RTP ¼ 8.1), DRB1* 0401-DQB1*0302/DRB1*07-DQB1*02 (RTP ¼ 6.1), and DRB1* 0401-DQB1*0302/DRB1*0401-DQB1*0301 (RTP ¼ 5.2). The main question investigated here was whether synergy between haplotypes in terms of the ability to form disease susceptible trans-encoded DQ dimers can be observed in genotypes other than DRB1*0301- DQB1*0201/DRB1*0401-DQB1*0302. The DRB1*01- DQB1*0501 haplotype-encoded DQ b1 and a1 chains cannot form trans dimers with the DQ b1 and a1 chains from the DRB1*0401-DQB1*0302 or DRB1*03-DQB1*02 haplotypes 22 and, therefore, the difference in risk of the DRB1*0401-DQB1*0302/DRB1*01-DQB1*0501 genotype (RTP ¼ 8.8) compared to the DRB1*03-DQB1*02/ DRB1*01-DQB1*0501 genotype (RTP ¼ 2.5) can be taken as the neutral reference for the difference in risk between DRB1*0401-DQB1*0302 and DRB1*03-DQB1*02

Table 1 Haplotype risk estimation for separate populations DRB1 DQA1 DQB1 NL NOR UK US SAR RTP T NT OR-r (CI) RTP T NT OR-r (CI) RTP T NT OR-r (CI) RTP T NT OR-r (CI) RTP T NT OR-r (CI) 0401 0301 0302 6.4**** 100 18 6.0 (4.3 8.4) 13.7**** 259 52 12.1 (9.5 15.4) 8.0**** 157 33 7.1 (5.5 9.2) 5.0** 68 20 4.5 (3.1 6.5) 0404 0301 0302 4.0* 21 9 2.5 (1.6 3.9) 5.8**** 71 37 4.7 (3.5 6.1) 2.5* 36 24 2.3 (1.7 3.1) 6.2** 31 10 4.0 (2.6 6.2) 0405 0301 0302 2.5 11 2 4.5 (2.3 9.0) 3.4 11 3 7.4 (4.0 13.6) 3.7 11 4 3.7 (2.1 6.4) 13.2**** 78 20 9.3 (6.5 13.4) 0801 0401 0401 2.5 17 11 1.7 (1.1 2.6) 1.7 33 34 2.4 (1.8 3.2) 0.9 7 18 0.6 (0.4 1.0) 2.8 7 5 1.8 (1.0 3.4) 0301 0501 0201 2.1 115 66 2.0 (1.5 2.6) 6.0**** 258 103 6.1 (4.9 7.7) 3.1*** 206 97 3.2 (2.6 4.0) 3.7* 108 39 3.7 (2.7 5.2) 6.9* 290 114 6.3 (4.7 8.4) 0402 0301 0302 1.8 8 3 3.6 (1.9 6.8) 5.8* 15 8 4.4 (2.7 7.2) 0405 0301 0201 3.6 6 8 1.9 (1.1 3.4) 0401 0301 0301 1.8 7 9 0.9 (0.5 1.5) 0.8 13 29 1.1 (0.8 1.7) 0.9 22 36 1.0 (0.7 1.3) 0.1 2 11 0.3 (0.2 0.7) 1302 0102 0604 1.1 15 21 0.8 (0.6 1.2) 2.2 29 31 2.3 (1.7 3.2) 09 0301 0303 1.9 13 17 1.9 (1.3 2.9) 0.7 6 10 1.0 (0.6 1.6) 0701 0201 0202 1.0 13 20 0.8 (0.5 1.1) 0.8 13 51 0.7 (0.5 0.9) 0.5 18 54 0.5 (0.4 0.7) 0.4 6 28 0.3 (0.2 0.5) 0.4 5 21 0.7 (0.4 1.2) 01 01 0501 1.0 36 41 1.0 1.0 38 95 1.0 1.0 50 77 1.0 1.0 23 32 1.0 1.0 18 46 1.0 0403 0301 0302 1.0 4 15 0.8 (0.4 1.4) 0405 0301 0301 0.9 4 16 0.7 (0.4 1.3) 16 01 05 0.3 2 9 0.3 (0.2 0.7) 1.8 62 104 1.5 (1.1 2.0) 13 01 06 0.4 16 48 0.5 (0.4 0.7) 0.6 13 28 0.7 (0.4 1.0) 1301 0103 0603 0.3 7 18 0.5 (0.3 0.8) 0.5 7 61 0.3 (0.2 0.5) 12 0501 0301 0.2 1 10 0.5 (0.2 1.1) 11 0501 0301 0.1** 4 33 0.2 (0.1 0.3) 0.2* 2 34 0.2 (0.1 0.4) 0.1* 2 25 0.2 (0.1 0.3) 0.2* 3 28 0.2 (0.1 0.3) 0.2*** 5 87 0.2 (0.1 0.3) 07 0201 0303 0.1* 2 13 0.2 (0.1 0.5) 0.0 0 13 0.2 (0.1 0.5) 0.0** 0 13 0.1 (0.0 0.3) 1001 01 05 0.5 1 9 0.5 (0.2 1.1) 1.2 2 13 0.5 (0.3 1.1) 14 01 0503 0.1 1 10 0.2 (0.1 0.5) 0.0 0 10 0.2 (0.1 0.6) 0.0* 0 11 0.2 (0.1 0.6) 15 0102 0602 0.0**** 1 57 0.0 (0.0 0.1) 0.0**** 0 150 0.0 (0.0 0.0) 0.1**** 3 53 0.1 (0.1 0.2) 0.0*** 1 37 0.1 (0.0 0.2) DRB1, DQA1, and DQB1 haplotypes are given. For some haplotypes (partial), subtyping of the alleles is present, otherwise serological typing is given. RTP denotes relative transmission probability (see methods) estimates with corresponding P-values given as ****o1e 05, ***o1e 04, **o1e 03, *o1e 02, relative to the DQB1*5-DRB1*01 reference haplotype, which is in bold and italic. Other columns give T: count of transmitted haplotypes, NT: count of nontransmitted haplotypes, OR-r: ORs (confidence interval in parentheses) calculated relative to the DQB1*5-DRB1*01 reference haplotype. Only haplotypes with total counts of transmitted and nontransmitted haplotypes higher than 10 are shown. 383

384 Table 2 Haplotype risk estimation for populations combined DRB1 DQA1 DQB1 RTP P-value T NT OR-r 0401 0301 0302 8.6 1.7E 42 588 124 8.3 (7.2 9.6) 0405 0301 0302 6.3 1.2E 12 119 31 6.6 (5.3 8.3) 0404 0301 0302 4.5 6.8E 14 160 80 3.5 (3.0 4.1) 0301 0501 0201 3.8 7.3E 23 978 419 4.1 (3.7 4.6) 0402 0301 0302 3.2 3.0E 3 38 16 4.0 (3.0 5.5) 0405 0301 02 2.0 7 9 1.4 (0.9 2.3) 0408 0301 0301 1.8 4 7 1.1 (0.6 2.0) 08 0401 0402 1.6 65 71 1.6 (1.3 2.0) 09 0301 0303 1.0 26 36 1.3 (1.0 1.7) 16 01 05 1.0 74 123 1.1 (0.9 1.3) 0401 0301 0301 1.0 44 86 0.9 (0.7 1.1) 01 01 05 1.0 165 292 1.0 13 01 06 0.7 92 215 0.8 (0.6 0.9) 07 0201 0202 0.7 55 174 0.6 (0.5 0.7) 0403 0301 0302 0.6 10 27 0.7 (0.5 1.0) 12 0501 0301 0.6 8 21 0.7 (0.5 1.1) 0405 0301 0301 0.5 4 17 0.5 (0.3 0.8) 10 01 05 0.4 4 33 0.3 (0.2 0.4) 13 0501 0301 0.3 5 21 0.5 (0.3 0.8) 11 0501 0301 0.1 7.2E 13 16 208 0.1 (0.1 0.2) 14 01 05 0.1 5.2E 06 2 48 0.1 (0.1 0.2) 07 0201 0303 0.1 1.4E 05 2 46 0.1 (0.1 0.2) 0407 0301 0301 0.0 3.0E 4 1 17 0.2 (0.1 0.4) 15 01 06 0.0 1.1E 27 7 307 0.0 (0.0 0.1) DRB1, DQA1, and DQB1 haplotypes are given. For some haplotypes (partial), subtyping of the alleles is present, otherwise serological typing is given. RTP denotes relative transmission probability (see methods) estimates relative to the DQB1*5-DRB1*01 reference haplotype, which is in bold and italic. Other columns give, P-value, T: count of transmitted haplotypes, NT: count of nontransmitted haplotypes, OR-r: ORs (confidence interval in parentheses) calculated relative to the DQB1*5-DRB1*01 reference haplotype. Only haplotypes with total counts of transmitted and nontransmitted haplotypes higher than 10 are shown. haplotypes. Our observation that an approximate threefold difference in risk exists between the DRB1*0401- DQB1*0302 and DRB1*03-DQB1*02 homozygous genotypes (RTP ¼ 26.6 vs RTP ¼ 8.1, respectively, see Table 3), and a similar three-fold difference in risk exists between the DRB1*0401-DQB1*0302/DRB1*01-DQB1*0501 and DRB1*03-DQB1*02/DRB1*01-DQB1*0501 genotypes (RTP ¼ 26.6 vs RTP ¼ 8.1, respectively, see Table 3) is consistent with this assumption. We, therefore, compared the difference in risk of haplotypes in a DRB1*0401-DQB1* 0302-positive genotype relative to the DRB1*03-DQB1* 02-positive genotype with the difference in risk of DRB1* 01-DQB1*05/DRB1*0401-DQB1*0302 vs the DRB1*01- DQB1*05/DRB1*03-DQB1*02 genotype (Table 4). The haplotypes DRB1*11-DQA1*0501-DQB1*0301, DRB1*07-DQA1*0201-DQB1*0202, and DRB1*08- DQA1*0401-DQB1*0402 showed higher increased risk in DRB1*0401-DQB1*0302-positive genotypes than in DRB1*03-DQB1*02-positive genotypes compared to the reference DRB1*01-DQB1*05 haplotype. These haplotypes can potentially form putative high-risk transencoded DQ dimers in a DRB1*0401-DQB1*0302-positive genotype, DQ(a1*0501, b1*0302), DQ(a1*0301, b1*02), and DQ(a1*0401, b1*0302), respectively. Note that the DQ(a1*0401) molecule, encoded by the DRB1*08- DQA1*0401-DQB1*0402 haplotype, is structurally very similar to the DQ(a1*0501) molecule. In contrast, DRB1*0401-DQB1*0301 and DRB1*09-DQB1*0303 haplotypes can form the DQ(a1*0301, b1*02) trans-encoded dimer in DRB1*03-DQB1*02-positive genotypes, but not in DRB1*0401-DQB1*0302-positive genotypes (Table 4). This can explain why the increase in risk of DRB1*0401- DQB1*0301 and DRB1*09-DQB1*0303 haplotypes in a DRB1*0401-DQB1*0302- compared to DRB1*03-DQB1* 02-positive genotype is even lower than that of DRB1*01- DQB1*0501, suggesting interaction of these haplotypes with DRB1*03-DQB1*02. Discussion In this large data set analysed using a statistical approach resistant to population admixture effects, ascertainment bias and the complexity due to the multiplicity of both predisposing and protective alleles at the HLA class II loci, we confirm that the DRB1*03-DQB1*02/DRB1*0401- DQB1*0302 genotype is significantly more associated with T1D than DRB1*0401-DQB1*0302/DRB1*0401- DQB1*0302 and DRB1*03-DQB1*02/DRB1*03-DQB1*02 genotypes. Furthermore, evidence is provided for synergy between either DRB1*0401-DQB1*0302 or DRB1*03 -DQB1*02 and two previously implicated haplotypes (DRB1*08-DQA1*0401-DQB1*04 and DRB1*09-DQA1* 0301-DQB1*0303), and for three new haplotypes (DRB1* 11-DQA1*0501-DQB1*0301, DRB1*07-DQA1*0201-DQB1* 02, and DRB1*0401-DQA1*0301-DQB1*0301). The haplotypes DRB1*11-DQA1*0501-DQB1*0301, DRB1*08-DQA1* 0401-DQB1*04, and DRB1*07-DQA1*0201-DQB1*02 show interaction with DRB1*0401-DQB1*0302 haplotype. Assuming that DQA1*0401 is structurally similar to DQA1*0501, in a DRB1*0401-DQB1*0302-positive genotype, the first two haplotypes can form the DQ(a1*0501, b1*0302) trans-encoded heterodimer, whereas the latter can form DQ(a1*0301, b1*0201). The haplotypes DRB1*

Table 3 HLA class II genotype risks relative to the DRB1*01-DQB1*05 homozygous genotype for populations combined 385 DRB1 DQA1 DQB1 DRB1 DQA1 DQB1 RTP P T NT OR-r 0401 0301 0302 0301 0501 0201 33.0 4.3E 10 282 12 47.2 (26.7 83.3) 0401 0301 0302 0401 0301 0302 26.6 2.4E 07 32 5 11.9 (6.1 23.1) 0405 0301 0302 0301 0501 0201 21.8 1.5E 06 68 6 21.4 (11.3 40.2) 0404 0301 0302 0301 0501 0201 21.6 3.0E 07 85 10 16.9 (9.4 30.5) 0401 0301 0302 08 0401 0402 17.4 1.9E 05 31 5 11.6 (5.9 22.5) 0401 0301 0302 01 01 05 8.8 3.0E 04 44 14 6.5 (3.7 11.6) 0402 0301 0302 0301 0501 0201 8.7 3.0E 03 20 2 15.2 (6.9 33.4) 0301 0501 0201 0301 0501 0201 8.1 6.1E 04 100 20 10.4 (6.0 18.0) 0401 0301 0302 13 01 06 8.1 1.2E 03 38 11 7.0 (3.9 12.8) 0401 0301 0302 0404 0301 0302 8.1 3.4E 03 17 5 6.5 (3.3 12.9) 0401 0301 0302 07 0201 0202 6.1 8.4E 03 18 4 8.2 (4.1 16.7) 0401 0301 0301 0401 0301 0301 5.2 9 1 10.8 (4.3 27.1) 0301 0501 0201 09 0301 0303 3.2 16 6 5.3 (2.7 10.3) 0301 0501 0201 16 01 05 3.0 36 25 3.1 (1.8 5.4) 0403 0301 0302 0301 0501 0201 2.8 7 6 2.5 (1.2 5.1) 0404 0301 0302 13 01 06 2.6 7 8 1.9 (1.0 3.8) 0301 0501 0201 01 01 05 2.5 44 37 2.6 (1.5 4.4) 01 01 05 16 01 05 2.5 6 9 1.5 (0.8 3.0) 0401 0301 0302 11 0501 0301 2.1 5 5 2.2 (1.0 4.6) 0401 0301 0301 01 01 05 2.1 11 15 1.6 (0.9 3.0) 0301 0501 0201 08 0401 0402 2.1 13 18 1.6 (0.9 2.9) 0401 0301 0301 0301 0501 0201 2.0 8 6 2.8 (1.4 5.6) 07 0201 0202 13 01 06 1.8 3 10 0.8 (0.4 1.7) 0405 0301 0302 0405 0301 0302 1.4 5 5 2.2 (1.0 4.6) 0404 0301 0302 01 01 05 1.3 8 17 1.1 (0.6 2.1) 0301 0501 0201 07 0201 0202 1.1 9 23 0.9 (0.5 1.7) 0301 0501 0201 13 01 06 1.1 20 34 1.3 (0.7 2.3) 01 01 05 08 0401 0402 1.1 3 12 0.7 (0.3 1.4) 08 0401 0402 13 01 06 1.0 2 8 0.7 (0.3 1.7) 01 01 05 01 01 05 1.0 5 12 1.0 (0.5 2.0) 0401 0301 0301 07 0201 0202 0.8 2 10 0.6 (0.3 1.3) 01 01 05 07 0201 0202 0.5 6 22 0.7 (0.3 1.3) 0401 0301 0301 13 01 06 0.5 2 12 0.5 (0.2 1.1) 0404 0301 0302 15 01 06 0.3 2 9 0.7 (0.3 1.5) 01 01 05 13 01 06 0.3 3 21 0.4 (0.2 0.8) 0401 0301 0302 15 01 06 0.3 2 20 0.3 (0.1 0.7) 0301 0501 0201 11 0501 0301 0.2 3 47 0.2 (0.1 0.4) 01 01 05 11 0501 0301 0.1 1 25 0.2 (0.1 0.4) 0301 0501 0201 15 01 06 0.1 0.01 2 44 0.1 (0.1 0.3) 01 01 05 15 01 06 0.0 7.2E 04 0 35 0.1 (0.0 0.2) 13 01 06 15 01 06 0.0 4.2E 03 0 31 0.1 (0.0 0.2) 07 0201 0202 15 01 06 0.0 0.01 0 22 0.1 (0.0 0.3) 11 0501 0301 16 01 05 0.0 0 12 0.2 (0.1 0.5) 11 0501 0301 15 01 06 0.0 0.01 0 19 0.1 (0.0 0.3) 15 01 06 15 01 06 0.0 0 20 0.1 (0.0 0.3) 0401 0301 0301 15 01 06 0.0 0 13 0.2 (0.0 0.5) 13 01 06 11 0501 0301 0.0 0 13 0.2 (0.0 0.5) 11 0501 0301 11 0501 0301 0.0 0 12 0.2 (0.1 0.5) Subtyping of the alleles is given if available, otherwise serological type is given. Second column gives RTP (see methods) estimates with corresponding P-values in the third column, relative to the DRB1*01-DQB1*5 homozygous reference genotype. Following columns give, T: count of transmitted genotypes, NT: count of nontransmitted genotypes, OR-r: ORs (confidence interval in parentheses) calculated relative to the DRB1*01-DQB1*5 homozygous reference genotype (ref). Only genotypes with total counts of transmitted and nontransmitted genotypes higher than 10 are shown. 09-DQA1*0301-DQB1*0303 and DRB1*0401-DQA1*0301- DQB1*0301 show interaction with DRB1* 03-DQB1*02 haplotype and can form in individuals with a DRB1*03- DQB1*02-positive genotype the DQ(a1*0301, b1*0201) trans heterodimer. This correlation between haplotype interaction and possible formation of the trans-encoded heterodimers similar to those formed in the high-risk DRB1*03-DQB1*02/DRB1*0401-DQB1*0302 genotype provides new evidence for the possibility of transencoded HLA DQ dimer disease susceptibility. One potential mechanism is that the trans-encoded dimers bind certain peptides differently from the cis heterodimers. Although one study did not find support for a relevance of an increase in the peptide-binding repertoire, this issue requires further investigations. 26 Further studies will have to be carried out with potential autoantigenic epitopes to test this hypothesis. Finally, we have shown new evidence for synergy of three haplotypes (DRB1*11-DQA1*0501-DQB1*0301, DRB1*07- DQA1*0201-DQB1*02, and DRB1*0401-DQA1* 0301- DQB1*0301), but it is difficult from these data to conclude which of the two trans-encoded DQ hetero-

386 Table 4 Cis- and trans-encoded DQA1-DQB1 dimers DRB1 DQA1 DQB1 DRB1 DQA1 DQB1 Cis Cis Trans Trans RTP P-value 11 0501 0301 0401 0301 0302 0501 0301 0301 0302 0301 0301 0501 0302 11.9 3.0E 03 11 0501 0301 0301 0501 0201 0501 0301 0501 0201 0501 0301 0501 0201 1.0 Ref 08 0401 0402 0401 0301 0302 0401 0402 0301 0302 0301 0402 0401 0302 8.8 9.0E 06 08 0401 0402 0301 0501 0201 0401 0402 0501 0201 0501 0402 0401 0201 1.0 Ref 07 0201 0202 0401 0301 0302 0201 0202 0301 0302 0301 0202 0201 0302 5.6 4.0E 04 07 0201 0202 0301 0501 0201 0201 0202 0501 0201 0501 0202 0201 0202 1.0 Ref 01 01 05 0401 0301 0302 0101 0501 0301 0302 None None 3.7 7.6E 05 01 01 05 0301 0501 0201 0101 0501 0501 0201 None None 1.0 Ref 0401 0301 0301 0401 0301 0302 0301 0301 0301 0302 0301 0301 0301 0302 2.6 NS 0401 0301 0301 0301 0501 0201 0301 0301 0501 0201 0501 0301 0301 0201 1.0 Ref 09 0301 0303 0401 0301 0302 0301 0303 0301 0302 0301 0303 0301 0302 1.6 NS 09 0301 0303 0301 0501 0201 0301 0303 0501 0201 0501 0303 0301 0201 1.0 Ref Genotypes are given as DRB1-DQA1-DQB1 haplotypes in serological typing. Heterodimers are shown as the corresponding DQA1-DQB1 allele combination in subtypings (deduced from linkage disequilibrium data). RTP (see methods) is calculated using CETDT, comparing the DRB1*0401-DQB1*0302 genotypes with their equivalent DRB1*03-DQB1*02 genotype, with corresponding P-values. Susceptible heterodimers are in bold, with those that occur in cis in the high-risk DRB1*03-DQB1*02/DRB1*0401-DQB1*0302 genotype bold, and those that occur in trans bold and underlined. dimers in DRB1*03-DQB1*02/DRB1*0401-DQB1*0302 may be associated with the strongest susceptibility, that is, DQ(a1*0301,b1*0201) or DQ(a1*0501, b1*0302), although some suggestions have been made previously. 4,24 Obviously, it cannot be ruled out that the interaction observed here is at another locus in the HLA region, or through an alternative mechanism. 27 Furthermore, it follows from Table 3 that the absolute risk of a genotype depends more on the sum of susceptible and protective elements, such as the presence of protective DQ heterodimers, DR molecules, and non-hla factors. It has been shown that efficiency of dimer formation differs between different DQa1 and DQb1 allele products. 28 The DQ molecules from the DRB1*0101- DQB1*0501 haplotype are not able to form trans-encoded dimers with the DQ b1 and a1 chain products from either of the DRB1*0401-DQB1*0302 or DRB1*03-DQB1*02 haplotypes. This suggests that only cis-encoded dimers are expressed by DRB1*0401-DQB1*0302/DRB1*01- DQB1*05 and DRB1*03-DQB1*02/DRB1*01-DQB1*05 genotypes. The risk of DRB1*0401-DQB1*0302/ DRB1*01-DQB1*05 is approximately three-fold that of DRB1*03-DQB1*02/DRB1*01-DQB1*05 (Table 3). Interestingly, the risk of the DRB1*0401-DQB1*0302 homozygous genotype is also three-fold that of the DRB1*03- DQB1*02 homozygous genotype (Table 3). Since it is unlikely that non-multiplicative epistatic effects are present in these genotypes, the similarity between the increase in risk suggests that these genotypes can be considered as pure measures of the susceptibility provided by the DRB1*03-DQB1*02 and DRB1*0401- DQB1*0302 haplotypes. However, the risk of the homozygous genotypes is higher than the heterozygous DRB1*01-DQB1*05 genotypes, indicating that the amount of expressed highly susceptible heterodimers is important. 29 An alternative explanation is that the DRB1*01-DQB1*05 haplotype confers dominant protection, which seems unlikely when comparing the risk of DRB1*01-DQB1*05 to that of established protective haplotypes such as DRB1*14- DQB1*05 and DRB1*15-DQB1*06, and from transmission rates of DRB1*01-DQB1*05 to affected siblings that do not carry either DRB1*0401-DQB1*0302 or DRB1*03- DQB1*02. 30 Taken together with the observed associations of trans-encoded dimers, these observations suggest that part of the variation in risk observed is influenced by the number of expressed high-risk dimers, which is determined by the number of combinations of DQb and DQa chains and their affinity to form heterodimers. 31 Evidence for such a gene dosage effect of HLA-DQ2 in correlation to T-cell response to gluten peptide epitopes has been found recently in coeliac disease, in which the T-cell response was correlated to the number of DQ2 molecules. 32 In coeliac disease, the susceptible dimer is DQA1*0501-DQB1*02, which is also formed in trans when the DRB1*07-DQB1*02 haplotype is found in a DRB1*03-DQB1*02-positive genotype. It has been well documented that the increased susceptibility of the DRB1*07-DQB1*02 haplotype is only present when combined with another haplotype that carries DQA1*0501. In narcolepsy, the high-risk haplotype is DRB1*15-DQB1*06, which in homozygous form has the highest genotype risk for disease, suggestive of a dosage effect. Remarkably, DQB1*0501 and DQA1*01 (non- *0102) alleles are protective in narcolepsy. 33 Following our hypothesis, the protection of DQB1*0501 and non- DQA1*0102 can be explained by the effect of the DRB1*01-DQA1*0101-DQB1*0501 haplotype through reduction in the number of susceptible heterodimers, since it is likely that DQB1*05 can form a heterodimer with DQA1*0102 of the DRB1*15-DQB1*06 haplotype. Apart from the trans dimer effect described here, several other factors are likely to play a role in the susceptibility of HLA class II. First, it has been described that the response of a DQ2.1 restricted HSV-2 specific T- cell clone responding to HSV peptides was not correlated with gene dosage, but rather with the peptide MHC avidity. 34 Second, the importance of DRB1 is obvious by the difference in susceptibility of DR4 subtypes. Most clear is the lower risk of DRB1*0403 and DRB1*0406 as previously described. 35 37 DRB1*0403 and DRB1*0406 share the E at position 74, which differentiates the two from the high-risk DRB1*0401 allele. Position 74 is part of the P4 pocket of the DRB1 molecule, which has been

consistently implicated in several models for disease prediction in rheumatoid arthritis. 38,39 The predictive value of the P4 pocket has also been demonstrated by others, 2,40,41 and has a strong pocket structural similarity between DRB1*0403 and DQB1*0301, both protective for T1D. 2 Third, several reports suggest the presence of an additional susceptibility locus in the class II, III, and I regions in several autoimmune diseases, although the necessity of fully compensating for the class II effect in these studies is a complicating factor. 14,27 Finally, variations within the regulatory sequences of HLA-DQ have also been found that could influence susceptibility. Together, these susceptibility factors probably have a large influence on the risk as measured for HLA class II and could contribute to the synergy observed in DRB1*03-DQB1*02/DRB1*0401-DQB1*0302 genotypes. Methods Estimation of the risk of genotypes in case control design Several authors have recognized that standard risk calculation by OR can be biased when more than one risk allele is present. A common correction for this bias is to standardize the OR by setting one haplo- or genotype to unity as a reference. Here we show ORs calculated with DRB1*01-DQB1*0501 haplo- or homozygous genotype as the reference (OR-r). Calculation with a different reference did not lead to significant differences in ranking or relative magnitude of the haplo- or genotypes. Estimation of the risk of haplotype and genotype in nuclear families Relative risks can be estimated using simplex families. Likelihood methods for these case parent triads can be used, conditional on the parental genotype. 42 45 These methods alleviate bias caused by population admixture, nonrandom mating, and deviations from HW equilibrium, and conditioning on parental genotypes avoids bias by ascertainment on multiple affected siblings. 18,20,42 45 With the likelihood method, haplotype relative risks (RR i ) or genotype relative risks (RR ij ) can be estimated, for haplotype i or genotype ij. We have previously written a program (CETDT) that outputs the ETDT estimate RR i or, when using the full genotype model, RR ij. 19 For clarity, we refer to this ETDT measure as RTP. Furthermore, CETDT can be used to test directly whether two different haplotypes have different risks. Here, we have used this program to analyse HLA class II haplotypes and genotypes in T1D families from several populations. For the multiplex families, one random affected sibling was chosen to reduce these families to a simplex form. Material Nuclear diabetic families with HLA class II DQB1, DQA1, and DRB1 typing were available from Dutch (219 families), Norwegian (386 families), UK (379 families), US (176 families), and Sardinian populations (261 families), in total comprising 1421 families. Full subtyping was available for US and Norwegian families. DQB1*06 subtyping was available for the Dutch families, and DRB1*04 subtyping was available for all populations. Abbreviations OR, Odds ratio; CETDT, Conditional Extended Transmission Disequilibrium Test; RTP, Relative Transmission Probability; RR, Relative Risk for disease. Acknowledgements We thank the Juvenile Diabetes Research Foundation (JDRF), the Wellcome Trust, Dutch Diabetes Research Foundation (97.137,2001.10.004), The Netherlands Organisation for Health Research and Development (ZonMW) and The Juvenile Diabetes Research Foundation International (JDRF) (2001.10.004), and the Italian Telethon for financial support. We gratefully acknowledge the participation of all patients and family members, including provision of samples from T1D families from the Human Biological Data Interchange and Diabetes UK repositories, and sample collections by The Norwegian Study Group for Childhood Diabetes, and Italy. We thank Sarah Nutland and Helen Rance for DNA preparation and HLA typing. References 1 Cucca F, Todd JA. HLA susceptibility to type 1 diabetes: methods and mechanisms. In: Browning MJ, McMichael AJ (eds). HLA and MHC: Genes, Molecules and Function. BIOS Scientific Publishers Ltd: Oxford, 1996, pp 23 60. 2 Cucca F, Lampis R, Congia M et al. A correlation between the relative predisposition of MHC class II alleles to type 1 diabetes and the structure of their proteins. Hum Mol Genet 2001; 10: 2025 2037. 3 Thorsby E, Undlien D. The HLA associated predisposition to type 1 diabetes and other autoimmune diseases. J Pediatr Endocrinol Metab 1996; 9 (Suppl 1): 75 88. 4 She JX. Susceptibility to type I diabetes: HLA-DQ and DR revisited [see comments]. Immunol Today 1996; 17: 323 329. 5 Cucca F, Muntoni F, Lampis R et al. Combinations of specific DRB1, DQA1, DQB1 haplotypes are associated with insulindependent diabetes mellitus in Sardinia. Hum Immunol 1993; 37: 85 94. 6 Noble JA, Valdes AM, Cook M, Klitz W, Thomson G, Erlich HA. The role of HLA class II genes in insulin-dependent diabetes mellitus: molecular analysis of 180 Caucasian, multiplex families. Am J Hum Genet 1996; 59: 1134 1148. 7 Ronningen KS, Spurkland A, Iwe T, Vartdal F, Thorsby E. Distribution of HLA-DRB1, -DQA1 and -DQB1 alleles and DQA1-DQB1 genotypes among Norwegian patients with insulin-dependent diabetes mellitus. Tissue Antigens 1991; 37: 105 111. 8 Sheehy MJ, Scharf SJ, Rowe JR et al. A diabetes-susceptible HLA haplotype is best defined by a combination of HLA-DR and -DQ alleles. J Clin Invest 1989; 83: 830 883. 9 Schipper RF, Koeleman BP, Bruining GJ et al. HLA class II associations with type 1 diabetes mellitus: a multivariate approach. Tissue Antigens 2001; 57: 144 150. 10 Erlich HA, Zeidler A, Chang J et al. HLA class II alleles and susceptibility and resistance to insulin dependent diabetes mellitus in Mexican-American families. Nat Genet 1993; 3: 358 364. 11 Park YS, Wang CY, Ko KW et al. Combinations of HLA DR and DQ molecules determine the susceptibility to insulin-depen- 387

388 dent diabetes mellitus in Koreans. Hum Immunol 1998; 59: 794 801. 12 Mijovic CH, Jenkins D, Jacobs KH, Penny MA, Fletcher JA, Barnett AH. HLA-DQA1 and -DQB1 alleles associated with genetic susceptibility to IDDM in a black population. Diabetes 1991; 40: 748 753. 13 Johansson S, Lie BA, Todd JA et al. Evidence of at least two type 1 diabetes susceptibility genes in the HLA complex distinct from HLA-DQB1, -DQA1 and -DRB1. Genes Immun 2003; 4: 46 53. 14 Zavattari P, Lampis R, Motzo C 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-DQB1, -DRB1 disease loci. Hum Mol Genet 2001; 10: 881 889. 15 Undlien DE, Friede T, Rammensee HG et al. HLA-encoded genetic predisposition in IDDM: DR4 subtypes may be associated with different degrees of protection. Diabetes 1997; 46: 143 149. 16 Thomson G, Motro U, Selvin S. Statistical aspects of measuring the strength of associations between HLA antigens and diseases. Tissue Antigens 1983; 21: 320 332. 17 Payami H, Joe S, Farid NR et al. Relative predispositional effects (RPEs) of marker alleles with disease: HLA-DR alleles and Graves disease. Am J Hum Genet 1989; 45: 541 546. 18 Cordell H, Clayton DG. A unified stepwise regression procedure for evaluating the relative effects of polymorphisms within a gene using case/control of family data: application to HLA in type 1 diabetes. Am J Hum Genet 2002; 70: 124 141. 19 Koeleman BP, Dudbridge F, Cordell HJ, Todd JA. Adaptation of the extended transmission/disequilibrium test to distinguish disease associations of multiple loci: the Conditional Extended Transmission/Disequilibrium Test. Ann Hum Genet 2000; 64: 207 213. 20 Cordell HJ. Effect of recombination, ascertainment, and multiple affected offspring on case/pseudo-control analysis. Genet Epidemiol 2003; 25: 244. 21 Thomson G, Robinson WP, Kuhner MK et al. Genetic heterogeneity, modes of inheritance, and risk estimates for a joint study of Caucasians with insulin-dependent diabetes mellitus. Am J Hum Genet 1988; 43: 799 816. 22 Chuang LM, Wu HP, Tsai WY, Lin BJ, Tai TY. Transcomplementation of HLA DQA1-DQB1 in DR3/DR4 and DR3/DR9 heterozygotes and IDDM in Taiwanese families. Diabetes Care 1995; 18: 1483 1486. 23 Ronningen KS, Markussen G, Iwe T, Thorsby E. An increased risk of insulin-dependent diabetes mellitus (IDDM) among HLA-DR4,DQw8/DRw8,DQw4 heterozygotes. Hum Immunol 1989; 24: 165 173. 24 Thorsby E. Invited anniversary review: HLA associated diseases. Hum Immunol 1997; 53: 1 11. 25 Nepom BS, Schwarz D, Palmer JP, Nepom GT. Transcomplementation of HLA genes in IDDM. HLA-DQ alpha- and betachains produce hybrid molecules in DR3/4 heterozygotes. Diabetes 1987; 36: 114 117. 26 Reichstetter S, Kwok WW, Nepom GT. Impaired binding of a DQ2 and DQ8-binding HSV VP16 peptide to a DQA1*0501/ DQB1* 0302 trans class II heterodimer. Tissue Antigens 1999; 53: 101 105. 27 Thorsby E, Caillat-Zucman S, Dorman J et al. Additional disease predisposing genes in the HLA complex. A summary of the 13.IHWS disease component studies. HLA 2004. Immunobiology of the Human MHC. 28 Kwok WW, Kovats S, Thurtle P, Nepom GT. HLA-DQ allelic polymorphisms constrain patterns of class II heterodimer formation. J Immunol 1993; 150: 2263 2272. 29 Nepom GT, Kwok WW. Molecular basis for HLA-DQ associations with IDDM. Diabetes 1998; 47: 1177 1184. 30 Undlien DE, Kockum I, Ronningen KS et al. HLA association in type 1 diabetes among patients not carrying high-risk DR3-DQ2 or DR4-DQ8 haplotypes. Tissue Antigens 1999; 54: 543 551. 31 Sant AJ, Germain RN. Intracellular competition for component chains determines class II MHC cell surface phenotype. Cell 1989; 57: 797 805. 32 Vader W, Stepniak D, Kooy Y et al. The HLA-DQ2 gene dose effect in celiac disease is directly related to the magnitude and breadth of gluten-specific T cell responses. Proc Natl Acad Sci USA 2003; 100: 12390 12395. 33 Mignot E, Lin L, Rogers W et al. Complex HLA-DR and-dq interactions confer risk of narcolepsy-cataplex in three ethnic groups. Am J Hum Genet 2001; 68: 686 699. 34 Reichstetter S, Kwok WW, Kochik S, Koelle DM, Beaty JS, Nepom GT. MHC-peptide ligand interactions establish a functional threshold for antigen-specific T cell recognition. Hum Immunol 1999; 60: 608 618. 35 Awata T, Kuzuya T, Matsuda A, Iwamoto Y, Kanazawa Y. Genetic analysis of HLA class II alleles and susceptibility to type 1 (insulin-dependent) diabetes mellitus in Japanese subjects. Diabetologia 1992; 35: 419 424. 36 Cucca F, Muntoni F, Lampis R et al. Combinations of specific DRB1, DQA1, DQB1 haplotypes are associated with insulindependent diabetes mellitus in Sardinia. Hum Immunol 1993; 37: 85 94. 37 Cucca F, Lampis R, Frau F et al. The distribution of DR4 haplotypes in Sardinia suggest a primary association of type 1 diabetes with DRB1 and DQB1 loci. Hum Immunol 1995; 43: 301 308. 38 Zanelli E, Gonzalez-Gay MA, David CS. Could HLA-DRB1 be the protective locus in rheumatoid arthritis? Immunol Today 2001; 16: 274 278. 39 Auger I, Roudier J. HLA-DR and the development of rheumatoid arthritis. Autoimmunity 1997; 26: 123 128. 40 Zamani M, Cassiman JJ. Reevaluation of the importance of polymorphic HLA class II alleles and amino acids in the susceptibility of individuals of different populations to type I diabetes. Am J Med Genet 1998; 76: 183 194. 41 Djoulah S, Busson M, Sasazuki T et al. A new predictive model for insulin-dependent diabetes mellitus susceptibility based on combinations of molecular HLA- DRB1 and HLA-DQB1 pockets. Tissue Antigens 1999; 54: 341 348. 42 Self SG, Longton G, Kopecky KJ, Liang KY. On estimating HLA/disease association with application to a study of aplastic anemia. Biometrics 1991; 47: 53 56. 43 Schaid DJ, Sommer SS. Genotype relative risks: methods for design and analysis of candidate-gene association studies. Am J Hum Genet 1993; 53: 1114 1126. 44 Schaid DJ. Likelihoods and TDT for the case parents design. Genet Epidemiol 1999; 16: 250 260. 45 Sham PC, Curtis D. An extended transmission/disequilibrium test (TDT) for multi-allele marker loci. Ann Hum Genet 1995; 59: 323 336.