Evidence of at least two type 1 diabetes susceptibility genes in the HLA complex distinct from HLA-DQB1, -DQA1 and DRB1

(2003) 4, 46 53 & 2003 Nature Publishing Group All rights reserved 1466-4879/03 $25.00 www.nature.com/gene Evidence of at least two type 1 diabetes susceptibility genes in the HLA complex distinct from HLA-DQB1, -DQA1 and DRB1 S Johansson 1, 6, BA Lie 1,6, JA Todd 2, F Pociot 3, J Nerup 3, A Cambon-Thomsen 4, I Kockum 5, HE Akselsen 1, E. Thorsby 1 and DE Undlien 1 1 Institute of Immunology, Rikshospitalet University Hospital, Norway; 2 JDRF/WT Diabetes and Inflammation Laboratory, Cambridge Institute for Medical Research, University of Cambridge, UK; 3 Steno Diabetes Center, Gentofte, Denmark; 4 Inserm U 558, Toulouse, France; 5 Department of Molecular Medicine, Karolinska Institutet, Stockholm, Sweden Susceptibility to, and protection against development of type 1 diabetes (T1D) are primarily associated with the highly polymorphic exon 2 sequences of the HLA class II genes: DQB1, DQA1 and DRB1. However, several studies have also suggested that additional genes in the HLA complex influence T1D risk, albeit to a lesser degree than the class II genes. We have previously shown that allele 3 of microsatellite marker D6S2223, 4.9 Mb telomeric of DQ in the extended class I region, is associated with a reduction in risk conferred by the DQ2-DR3 haplotype. Here we replicate this finding in two populations from Sweden and France. We also show that markers in the HLA class II, III and centromeric class I regions contribute to the DQ2- DR3 associated risk of T1D, independently of linkage disequilibrium (LD) with both the DQ/DR genes and the D6S2223 associated gene. The associated marker alleles are carried on the DQ2-DR3-B18 haplotype in a region of strong LD. By haplotype mapping, we have located the most likely location for this second DQ2-DR3 haplotype-modifying locus to the 2.35 Mb region between HLA-DOB and marker D6S2702, located 970 kb telomeric of HLA-B. (2003) 4, 46 53. doi:10.1038/sj.gene.6363917 Keywords: autoimmunity; genetics of type 1 diabetes; haplotype analysis; HLA complex Introduction Type 1 diabetes (T1D) is a T-lymphocyte-mediated complex disease, in which both genetic and environmental factors are involved in the pathogenesis. Many chromosomal regions have been suggested to harbour disease susceptibility genes. 1,2 The strongest genetic contribution in humans has been mapped to the HLA complex on the short arm of chromosome 6 (IDDM1). The HLA class II region genes DQB1, DQA1 and DRB1 are of primary importance for disease susceptibility and resistance. 3,4 The primary aetiological variants are their highly polymorphic exon 2 sequences that encode the peptide-binding sites of the corresponding HLA molecules. The strongest susceptibility haplotypes among Caucasoid populations are DQB1*0201-DQA1*0501- DRB1*03 (DQ2-DR3) and DQB1*0302-DQA1*03-DRB1*04 (DQ8-DR4), in particular when they occur together. However, not all DQ2-DR3 and DQ8-DR4 haplotypes are equally predisposing, even though they carry the same DQ/DR alleles. 5 10 The number of non-dq/dr genes involved and their identities are unknown. Correspondence: Dr S Johansson, Institute of Immunology, Rikshospitalet University Hospital, 0027 Oslo, Norway. E-mail: stefan.johansson@labmed.uio.no 6 These authors contributed equally. Localisation of these genes is particularly difficult owing to strong linkage disequilibrium (LD) in the HLA complex. 11,12 Recently, we reported an association with T1D for the polymorphic microsatellite D6S2223 marker in the extended HLA class I region, independent of LD with DQ/DR. Allele 3 at D6S2223 is associated with a reduction in the DQ2-DR3 haplotype-associated risk of T1D. 8 Indeed, in the same study 8 we also found some evidence for an association to the microsatellite locus D6S273. Allele 2 at this locus, located in the class III region, is positively associated with disease, again on the DQ2-DR3 haplotype, but independently of DQ2-DR3 itself. Similarly, in a Finnish study 9 an island of association with T1D, associated with DQ8-DR4 haplotypes but independent of the DQ8 or DR4 alleles themselves, was reported. The authors suggested that a disease locus could be mapped to a 240 kb region encompassing HLA-B. Thirdly, in a study of Sardinian T1D, it was suggested that there could be three non-dq/ DR loci in the class II and class III regions affecting the DQ2-DR3 haplotype susceptibility. 10 Nevertheless, LD can leap-frog across regions, and an isolated island of disease-associated markers may not be a reliable indicator of disease locus position, 11,13 nor of independence between these genetic modifiers. Thus, the Sardinian data 10 can also be explained by one susceptibility gene on an extended haplotype. In the absence of complete genotyping of all genes in the relevant regions and data

from functional studies of candidate polymorphisms in these islands, haplotype mapping may help locate additional aetiological variants. 6,14 16 This method relies on identifying recombination break points and blocks of LD and then measuring the disease association of the blocks, or ancestral haplotype segments. In the present study, we first tested if our previous findings for disease associations with markers D6S2223 and D6S273 were replicated in two independent populations from Sweden and France. Secondly, we genotyped a more dense set of microsatellites surrounding D6S273 in the class III and flanking region and studied if these markers were associated with T1D independently of DQB1 and DRB1 in a combined set of families from Norway, Denmark, United Kingdom, Sweden and France. Next, we used haplotype mapping to define the boundaries for the associated region, and test the dependence between this association and the previously found D6S2223 association. We present evidence that two DQ2-DR3 haplotype risk modifying loci exist within the HLA complex: one in LD with D6S2223*3 and a second provisionally mapped to the 2.35 Mb region between HLA-DOB and D6S2702. The latter susceptibility gene is in strong LD with the ancestral DQ2-DR3- B18 haplotype which we demonstrate confers higher risk than other DQ2-DR3 haplotypes. Materials and methods Populations The families used in this study were selected from a large number of T1D families from Norway (NO), Denmark (DK) and the United Kingdom (UK) as described in previous studies, 8 and from Sweden (SW) and Southern France (FR) (from the Midi-Pyrenees region or Atlantic Pyrenees county). 17,18 Only families with at least one parent homozygous for HLA-DQB1-DQA1-DRB1 haplotypes were included in the study. This resulted in a material consisting of 273 families comprising 147 and 73 parents homozygous for the high-risk haplotypes DQB1*0201-DQA1*0501-DRB1*03 and DQB1*0302- DQA1*03-DRB1*0401, respectively (see Table 1). The case control material consisted of 181 Norwegian T1D patients and 354 healthy controls selected from the Norwegian Bone Marrow Donor Registry. The subjects were either DQ2-DR3 homozygotes or DQ2-DR3/DQ8- DR4(0401) or DQ2-DR3/DQ8-DR4(0404) heterozygotes. Both cases and controls have been used and described in previous studies. 19 Genotyping The subjects have been previously genotyped for HLA- DQB1, DQA1 and DRB1 alleles mainly as described elsewhere. 20 22 In the first step of the study, 12 microsatellites in the HLA complex, between RING3 and D6S510, covering a region of approximately 2.9 Mb were genotyped in all families with at least one parent homozygous for DQ2-DR3 (see Figure 1 for a list of markers and relative distances). All families were also genotyped for microsatellite marker D6S2223 in the extended class I region and for DMB and selected alleles at DOB. Primer sequences for microsatellites were obtained from the Genome Database (GDB, http:// gdbwww.gdb.org/). Genotyping of microsatellites was performed as described previously. 8 Each microsatellite allele was given a numerical value, not always in accordance with GDB terminology, starting at the allele with the lowest number of repeats observed. However, the MICA alleles were termed A4, A5, A5.1, A6 and A9 according to the number of alanine repeats. DMB and DOB genotyping were performed using sequence-specific oligonucleotide (SSO) probes. In brief, for DMB alleles, PCR amplification was performed using primers described elsewhere 23 and five SSO probes (at positions 144 and 179 of DMB) were used to distinguish between the six known alleles. PCR primers for amplification of DOB exon 4 are described in Zavattari et al 24 and four SSO probes (at positions 208 and 218) were used to distinguish DOB*0103 and DOB*0104 from other DOB alleles (0101 and 0102). In the second step of the study, selected families were genotyped for HLA-A and -B and DPB1. HLA-A and -B genotyping were performed using Dynal RELI TM SSO HLA-A and HLA-B Typing Kit (Dynal Biotech, Oslo, Norway). DPB1 genotyping was performed either by PCR-SSOP or sequencing as described elsewhere. 25 All cases and controls were genotyped for D6S273, MIB, HLA-B, DMB and DOB. Genotyping was performed as described for the family material except for the DQ2- DR3/DQ8-DR4(0401) and DQ2-DR3/DQ8-DR4(0404) heterozygous patients who were typed for the existence of HLA-B8 and -B18 using an allele specific PCR assay. The controls were serologically typed for HLA-B. Statistical analysis Transmission of single microsatellite alleles and haplotypes from parents heterozygous for the marker (or haplotype) to diabetic probands (first diagnosed child) were analysed using the transmission/disequilibrium test (TDT). 26 Only transmissions from parents 47 Table 1 Populations used in the study Population No. of families DQB1*0302-DRB1*0401 homozygous parents DQB1*0201-DRB1*03 homozygous parents Markers tested previously 8 Norway 65 18 42 D6S273, D6S2223 Denmark 50 24 21 D6S273, D6S2223 United Kingdom 64 8 41 D6S273, D6S2223 Sweden 76 23 23 None France a 18 0 20 None Total 273 73 147 F a The French families were from the Midi-Pyrenees region or Atlantic Pyrenees county.

48 The extended HLA complex Class II Class III Class I Extended class I DM DQ DP DR B DOB TNF C A 0 1 2 3 4 5 (Mb) RING3CA TAP1CA D6S273 TNFd TNFc TNFa C12A MICA MIB D6S2700 D6S2702 D6S510 D6S2223 Figure 1 Map of the extended HLA complex showing the microsatellite markers used in the study. Map distances are relative to DPB1 and are based on the consensus sequence of the HLA region available at the Sanger Centre and from the working draft sequence of the human genome of chromosome 6 (January 2002). homozygous for the established high-risk HLA-haplotypes DQ2-DR3 or DQ8-DR4(0401) were analysed. This approach is described as the homozygous parent TDT (HPTDT) and eliminates effects due to LD with the mentioned high-risk DQB1, DQA1 and DRB1 alleles. 8 Transmissions were analysed separately from DQ2-DR3 and DQ8-DR4(0401) homozygous parents, and marker alleles with a frequency less than 5% were not included. LD between adjacent markers among all DQ2-DR3 bearing parents (396 haplotypes) was calculated as normalised pairwise disequilibrium (D ) 27 using the Arlequin software. 28 For haplotype analyses, the criterion for assignment of a DQ2-DR3-B18 haplotype was that no mismatches were allowed for the eight relevant alleles (D6S273*2, TNFd*5, TNFc*2, TNFa*1, C12A*1, MICA*A4, MIB*2, B18) and at least five of them had to be unambiguously scored. However, three out of the 54 assigned DQ2-DR3-B18 haplotypes carried TNFd*3 instead of the expected TNFd*5 but were still assigned as DQ2-DR3-B18 haplotypes based on identity at all other relevant loci both telomeric and centromeric of TNFd. In the case control data set, the number of DQ2-DR3- B18 haplotypes was determined by manual inspection and haplotype frequencies were estimated using the Arlequin software. 28 Odds ratios were calculated using Woolf s formula 29 and chi-square or Fisher s exact test when appropriate was used to test for statistical significance. Unless otherwise stated the P-values presented are not corrected for number of comparisons. Results T1D associations on DQ2-DR3 haplotypes independently of DQ and DR gene contributions The Swedish and French DQ2-DR3 homozygous families provided an opportunity to independently test whether our previous finding of association of D6S273*2 and D6S2223*3 with T1D could be replicated (Tables 1 and 2). The results from the DQ2-DR3 HPTDT show that both D6S273*2 and D6S2223*3 are associated with T1D also in these families, further supporting an association of both alleles, independently of the DQ2-DR3 contribution. The results of the genotyping scan of the HLA complex are shown in Table 3 (see Figure 1 for marker positions). Only the most strongly associated allele for each marker from the HPTDT is presented. Furthermore, only transmissions from the 147 DQ2-DR3 homozygous parents are shown, since no statistically significant deviations from 50% transmission were evident from DQ8-DR4(0401) homozygous parents (data not shown). One allele for each marker between HLA-DOB and Table 2 DQ2-DR3 homozygous parent TDT for D6S2223*3 and D6S273*2 from previous publication compared to results from the present study Populations No. of DQB1*0201- DRB1*03 homozygous parents Marker alleles %T (T/NT) P Norway, Denmark and the 117 D6S2223*3 20 (9/37) 0.00004 United Kingdom (from Lie et al 8 ) D6S273*2 81 (21/5) 0.002 Sweden and France 43 D6S2223*3 18 (2/9) 0.03 (not previously published) D6S273*2 83 (10/2) 0.02 Total 160 D6S2223*3 19 (11/46) 3.6E 06 D6S273*2 82 (31/7) 0.0001 % T=percent transmission, T=number of transmitted alleles, NT=number of not transmitted alleles.

Table 3 (n ¼ 147) Transmission of alleles from DQ2-DR3 homozygous parents to diabetic probands Marker alleles % T (T/NT) CI P RING3CA*2 65 (30/16) 51 79 0.04 DMB*0101 58 (14/10) 39 78 Ns TAP1CA*1 58 (18/13) 41 75 Ns DOB*0103 94 (16/1) 83 100 0.0003 D6S273*2 81 (29/7) 68 93 0.0002. TNFd*5 74 (28/10) 60 88 0.004 TNFc*2 68 (21/10) 51 84 0.05 TNFa*1 79 (23/6) 65 94 0.002 C12A*1 68 (21/10) 51 84 0.05 MICA*A4 77 (24/8) 60 90 0.005 MIB*2 80 (24/6) 66 94 0.001 D6S2700*5 85 (17/3) 69 100 0.002 D6S2702*12 54 (14/12) 35 73 Ns D6S510*6 75 (15/5) 56 94 0.03 D6S2223*1 100 (10/0) F 0.002 % T = percent transmission, T = number of transmitted alleles, NT = number of not transmitted alleles, CI = 95% confidence interval, n.s. = not significant. D6S2700 was transmitted more frequently than the expected 50% under the assumption of no association, and the P-values for DOB*0103 (P c ¼ 0.01), D6S273*2 (P c ¼ 0.01) and MIB*2 (P c ¼ 0.04) remain significant after conservative correction for numbers of alleles tested (n ¼ 40). LD between all of the positively associated alleles is very strong on the DQ2-DR3 haplotype (Table 4). Haplotype analyses revealed that the associated alleles are predominantly parts of DQ2-DR3 haplotypes that all share a common haplotype segment between DQB1 and MIB (located B30 kb centromeric of HLA-B). It appears that virtually all the presented single marker associations (Table 3) are caused by a single risk haplotype marked by microsatellite marker alleles D6S273*2-TNFd*5-TNFc*2- TNFa*1-C12A*1-MICA*A4-MIB*2 (DR3-(2-5-2-1-1-A4-2)- haplotype). All family trios with a DQ2-DR3 homozygous parent carrying this high-risk haplotype were genotyped for HLA-DPB1, HLA-B and HLA-A for further haplotype characterisation. Of the 54 extended DR3-(2-5- 2-1-1-A4-2) haplotypes, all carried the B18 allele. LD was less, but still considerable, between the DR3-(2-5-2-1-1- A4-2) haplotype and DPB1 and HLA-A (DPB1*0202 and HLA-A*30 being the most frequent alleles). The DQ2-DR3-(2-5-2-1-1-A4-2)-B18 haplotype (hereafter called DQ2-DR3-B18) is the most prevalent DQ2- DR3 haplotype in the French population studied, constituting 65% of DQ2-DR3 haplotypes. In contrast, among the other populations included, the DQ2-DR3- B18 haplotype is uncommon and makes up only 10% (range 6 14%) of all parental DQ2-DR3 haplotypes. The transmission deviation of the DQ2-DR3-B18 haplotype using the HPTDT was similar in all five populations: 80% T, P ¼ 0.001 (24 T, 6 NT) in all populations combined. To confirm the finding in our family data set of a positive association for the DQ2-DR3-B18 haplotype, we next genotyped an independent Norwegian case control data set with individuals stratified for known highrisk genotypes (i.e. DQ2-DR3 homozygotes and DQ2DR3/DQ8-DR4(0401) or DQ2-DR3/DQ8-DR4(0404) Table 4 Pairwise standardised disequilibrium values between selected alleles present on DQ2-DR3-B18 haplotypes among all DQ2-DR3 haplotypes in the study (n=396) RING3CA*2 DMB*0101 TAP1CA*1 DOB*0103 D6S273*2 TNFd*5 TNFc*2 TNFa*1 C12A*1 MICA*A4 MIB*2 D6S2700*5 D6S2702*12 D6S510*6 D6S2223*1 RING3CA*2 DMB*0101 1 TAP1CA*1 0.61 0.05 DOB*0103 0.93 1 0.97 D6S273*2 0.66 0.1 0.59 1 TNFd*5 0.6 0.1 0.48 1 0.93 TNFc*2 0.42 0.14 0.51 1 0.93 0.83 TNFa*1 0.66 0.08 0.65 0.94 1 0.94 1 C12A*1 0.56 0.21 0.48 0.96 0.9 0.85 0.78 1 MICA*A4 0.57 0.1 0.61 0.91 0.89 0.86 0.89 0.98 0.88 MIB*2 0.62 0.08 0.62 0.97 0.96 0.92 0.98 1 0.94 0.92 D6S2700*5 0.77 0.58 0.77 0.87 0.97 0.87 1 1 0.96 1 0.97 D6S2702*12 0.59 0.49 0.59 0.85 0.64 0.55 0.57 0.63 0.61 0.64 0.62 0.86 D6S510*6 0.54 0.03 0.59 0.68 0.81 0.79 0.8 0.83 0.83 0.78 0.75 0.76 0.68 D6S2223*1 0.65 1 0.6 0.68 1 0.8 1 1 1 1 1 1 1 1 All P-values 50.01, except for DMB*0101 which did not show evidence for LD with any other allele. 49

50 heterozygotes). All 181 diabetic patients and 354 healthy controls were genotyped for D6S273, MIB and for the presence of HLA-B8 or -B18. Due to the strong LD described above, samples positive for D6S273*2, MIB*2 and B18 could confidently be assumed to carry the full conserved DQ2-DR3-(1-2-5-2-1-1-A4-2)-B18 haplotype (also confirmed by haplotype estimates using the Arlequin software). The results are shown in Table 5 (only DQ2-DR3 haplotypes are included in the analysis). In agreement with the results from the HPTDT, a higher proportion of DQ2-DR3 haplotypes are DQ2-DR3-B18 among DQ2-DR3 homozygous and DQ2-DR3/DQ8- DR4(0401/4) heterozygous diabetic patients compared to healthy controls (OR ¼ 3.4, P ¼ 0.002). This result supports the fact that this conserved DQ2-DR3-B18 haplotype carries a susceptibility gene(s) contributing to T1D independently of LD with DR and DQ genes. Analyses of extended haplotypes to define the borders for the associated region Association mapping based on allelic association alone is not reliable to map the candidate region. We therefore investigated extended haplotypes and used haplotype mapping to better predict the location of the involved non-dq-dr genes. The 54 extended DQ2-DR3-B18 haplotypes from the DQ2-DR3 HPTDT family set allowed for extensive characterisation of these haplotypes between DPB1 and HLA-A (3.1 Mb). This analysis showed that both telomeric and centromeric of the DQB1 HLA-B interval, one common haplotype was more frequent, probably representing the ancestral DQ2-DR3-B18 haplotype, while the remaining haplotypes were diverse. We then compared the transmission of haplotypes carrying the telomeric and centromeric segments of the ancestral DQ2-DR3-B18 haplotype (grey bars in Table 6) with other DQ2-DR3-B18 haplotypes, to define which segments are common for the transmitted chromosomes and hence most likely to carry the susceptibility gene. It can be seen that all DQ2-DR3-B18 haplotypes are positively transmitted (HPTDT) irrespective of which segments are conserved telomeric or centromeric of the DQB1 HLA-B region. This suggests that the causal polymorphism most likely is located in the common conserved region between HLA-DOB and D6S2700. However, for some of the comparisons in the stratification analysis, the number of haplotypes is small. Thus, on the telomeric side, it cannot be formally excluded that DQ2-DR3-B18-D6S2700*5 confers greater risk than DR3- DQ2-B18-D6S2700*non-5. Therefore, it is possible that Table 5 Relative distribution of DQ2-DR3-B18 haplotypes compared to all DQ2-DR3-haplotypes among Norwegian T1D patients and healthy controls stratified for known high-risk DR-DQ genotypes DQ2-DR3-D6S273*2-MIB*2-B18 DQ-DR groups Patients n (%) Controls n (%) OR (CI) P DQ2-DR3/DQ2-DR3 4(6.5) 4(1.4) 4.8 (0.9-26.5) 0.04 DQ2-DR3 /DQ8-DR4(0401/4) 11(7.3) 7(3.3) 2.3 (0.8-6.8) 0.08 Combined 15(7.1) 11(2.2) 3.4 (1.4-8.0) 0.002 n=number of haplotypes, %=proportion of DQ2-DR3-B18 compared to all DQ2-DR3 haplotypes, OR=odds ratio, CI=confidence interval (95%). Table 6 Comparison of the transmission of different extended DQ2-DR3-B18 haplotypes using the HPTDT DPB1*0202 DOB*0103 TAP1CA*1 DMB*0101 RING3CA*2 DQ2-DR3-B18 D6S2700*5 D6S2702*12 HLA-A30 D6S510*6 %T(T/NT) + 80 (24/6) 0.001 + + 86 (18/3) 0.001 + Other 64 (7/4) 0.4 + + + 81 (13/3) 0.01 + Other 75 (12/4) 0.05 + + + ++ 82 (14/3) 0.008 + Other 68 (13/6) 0.1 ++++ + 88 (15/2) 0.002 Other + 65 (11/6) 0.2 + ++++ + 87 (13/2) 0.005 Other + 68 (13/6) 0.1 + ++++ + + + ++ 77 (10/3) 0.05 Other + Other 79 (15/4) 0.01 P % T=percent transmission, T=number of transmitted haplotypes, NT=number of not transmitted haplotypes. Darkgrey regions represent the conserved segment of the DQ2-DR3-B18 haplotype with the alleles indicated above. Nonshaded regions represent all other diverse DQ2-DR3-B18 haplotypes.

the aetiological polymorphism is located as far telomeric as D6S2702 (970 kb telomeric of HLA-B). Furthermore, we cannot rule out that the highly conserved DPB1*0202- RING3CA*2-DMB*0101-TAP1CA*1-DOB*0103-DQ2-DR3- B18 haplotype (which constitutes 89% of the DQ2-DR3- B18 haplotypes in the French population and 31% (range 29 33%) in the other populations tested) is more susceptible than other DQ2-DR3-B18 haplotypes. Hence, while the data suggest that the putative susceptibility gene is located telomeric of DOB, it could theoretically be located more centromeric, and, of course, there could be more than one causal locus. Since our haplotype analysis could not formally exclude an involvement of the candidate genes DMB and DOB, we genotyped these loci in the DQ-DR stratified case control set. No association was evident for DMB while DOB*0103 was again positively associated with T1D and in strong LD with DQ2-DR3-B18. More DQ2-DR3 homozygous patients compared to controls carried the DOB*0103 allele (27% vs 3.1%, OR ¼ 11.6, P ¼ 0.0003) and in the combined set of DR-DQ stratified individuals the OR was 2.4 (10% vs 4.5%, P ¼ 0.02). The D6S2223-associated locus Our haplotype analyses (Table 6) suggest that the likely telomeric border for the putative DQ2-DR3-B18 linked susceptibility gene is D6S2702. This marker is located 2.7 Mb centromeric of D6S2223 which we have shown is associated with T1D both in the present and a previous study. 8 Together, these data may suggest two independent susceptibility loci telomeric of DQ and DR. However, our data also demonstrate very strong LD on DQ2-DR3-B18 haplotypes: 28/54 (53%) are conserved to marker D6S510, covering a region of 2.6 Mb, and as many as 10/54 (19%) of the haplotypes are conserved all the way from DQ2 to D6S2223 (5.3 Mb; the full haplotype supplemented with additional genotyping 8 being DQ2-DR3-B18-D6S2700*5-D6S2702*12-D6S265*3-D6S510* 6-HLA*A30-D6S2222*3-D6S306*8-D6S1001*3-D6S464*11- D6S2223*1-D6S2225*4-D6S105*11-D6S2219*3-D6S2218*4- D6S1260*8-D6S1558*3). Therefore, we wanted to further study the possibility of only one gene effect explaining both the D6S2223*3 and the DQ2-DR3-B18 associations. Table 7 shows that the DQ2-DR3-B18 haplotype confers susceptibility even in the presence of the most strongly associated and protective allele D6S2223*3 (compare rows 1 and 3). Similarly, D6S2223*3 is associated with T1D independently of the DQ2-DR3-B18 haplotype effect since D6S2223*3 is significantly negatively transmitted (24% T, P ¼ 0.005) in the homozygous parent TDT also after removal of all parents carrying DQ2-DR3-B18. Table 7 Comparison of T1D association conferred by extended DQ2-DR3-B18-D6S2223 haplotypes, using the DQ2-DR3 homozygous parent TDT DRB1 HLA-B D6S2223 T NT % T P 3 18 3 11 6 65 0.2 3 18 Non-3 15 2 88 0.002 3 Non-18 3 12 43 22 3E-05 3 Non-18 Non-3 23 10 70 0.02 % T=percent transmission, T=number of transmitted haplotypes, NT=number of not transmitted haplotypes. Discussion We have presented the first independent confirmation of an association between D6S2223 and T1D, which cannot be explained by LD to DR-DQ alleles. 8 Using the DQ2- DR3 homozygous parent TDT, we found reduced transmission of D6S2223*3 in French and Swedish T1D families (Table 2). We also show, by a combination of single-point association mapping using the HPTDT (Table 3), stratified case control analysis (Table 5), and extended haplotype mapping (Table 6), that several marker alleles in the central region of the HLA complex are associated with T1D independent of LD with DR and DQ genes. Haplotype mapping suggests that these associations are most likely caused by a susceptibility gene located in the 2.35 Mb DOB D6S2702 interval. The most strongly associated marker alleles in this region are predominantly carried en bloc on the DQ2-DR3-B18 haplotype. The conserved susceptibility haplotype we found in this study carries B18 and TNF-microsatellite alleles that are all markers for the known ancestral DQ2-DR3-B18 (18.2) haplotype which has its highest frequencies in Southern European populations, especially among Basques and Sardinians. 30,31 The DQ2-DR3-B18 haplotype has often been referred to as diabetogenic, 32 34 but from previous studies it is not possible to completely rule out the effect of DR and DQ when comparing the relative susceptibility of DQ2-DR3-B18 haplotypes with other DQ2-DR3 haplotypes. The homozygous parent TDT eliminates this problem and our data strongly imply that the DQ2-DR3- B18 haplotype indeed confers significantly increased risk compared to other haplotypes with the same DQB1, DQA1 and DRB1 alleles, including the DQ2-DR3-B8 haplotype which differs from the former haplotype telomeric of DRB1. The frequency of the DQ2-DR3-B18 is low in most Caucasian populations and hence its overall contribution to T1D is limited. We observed very strong LD between alleles on the DQ2-DR3-B18 haplotype, and this remarkable haplotype conservation poses considerable difficulties in further fine mapping. Interestingly, extensive LD is present centromeric of DQ with one common DQ2-DR3-B18 haplotype marked by DOB*0103 that was strongly conserved, even through the recombination hotspots located in the region, 12,35 to the DPB1 locus. We cannot exclude that this conserved haplotype confers higher risk than other DQ2-DR3-B18 haplotypes lacking this segment. However, comparisons between distantly related populations such as Sardinian and Northern European populations can provide some information about the additional disease locus. In a recent study of Sardinian families, 40% of DQ2-DR3 haplotypes were found to be non-predisposing and the predisposing haplotypes all had the conserved haplotype DMB*0101-DOB*1- TNFc*169. In the populations studied here DMB cannot explain the observed association with the DQ2-DR3-B18 haplotype, nor was it found associated in a DQ2-DR3 matched case control analysis. Concerning DOB*0103, surprisingly we find a positive association with T1D (93% T, P ¼ 0.0003 in families, OR ¼ 2.4, P ¼ 0.02 in the case control study). This is in stark contrast to the study by Zavattari et al, 10 where they found DOB1*0103 (designated DOB*2 in their study) to be significantly decreased on DQ2-DR3 haplotypes among patients 51

52 compared to controls. Moreover, whereas we find strong positive LD between DOB*0103 and DQ2-DR3-B18 in our family material (D 0 ¼ 0.83, 67% of DQ2-DR3-B18 haplotypes carry DOB*0103), the DOB*0103 allele is exceedingly rare on DQ2-DR3 haplotypes in Sardinia (2.2% on transmitted DQ2-DR3 haplotypes and 6.6% on non-transmitted DQ2-DR3-haplotypes) despite the fact that the DQ2-DR3-B18 haplotype is the most frequent DQ2-DR3 haplotype in Sardinia (73% of Sardinian DQ2- DR3 haplotypes 36 ). This comparison gives us valuable mapping information. The differences in allele composition at DMB and DOB between the DQ2-DR3-B18 haplotypes in the currently studied populations and Sardinia suggest that these genes may not influence the risk conferred by the conserved susceptibility haplotype defined here. This is consistent with several studies on the neighbouring TAP and LMP genes which have been found not to be primarily associated with T1D in several studies (for references see Undlien et al 20 ). However, we cannot exclude that DPB1*0202 or another gene in the class II region on the conserved DOB*0103-DQ2-DR3-B18 haplotype may be primarily involved in susceptibility to T1D. DPB1 has previously been implicated in T1D susceptibility 37 40 with a suggested predisposing role for DPB1*0301 and DPB1*0202. Here, we show that DPB1*0202 is in strong LD, not only with DQ2-DR3, but throughout the HLA complex on the strongly predisposing DOB*0103-DQ2-DR3-B18 haplotype. Again, it is difficult to determine which effect is primary and secondary to LD, and the data illustrate the importance of extensive haplotype information for reliable mapping of disease genes even in a region interrupted by recombination hotspots. We have previously determined that DQ2-DR3-associated T1D susceptibility was affected by a locus in LD with the D6S2223 marker, in the extended class I region. 8 The most common allele, D6S2223*3, was associated with a reduction in risk of DQ2-DR3 haplotypes. We have shown here that this effect is independent of the DQ2- DR3-B18-associated effect. Taken together our study provides evidence for two non-dq/dr disease genesfone in the telomeric class I/ extended class I region marked by D6S2223, and another centromeric of the class I marker D6S2702 (440 kb centromeric of HLA-A). Whether these loci also may influence the susceptibility conferred by non-dq2-dr3 haplotypes cannot be determined in this study. We do not find any significant transmission distortion in the class III and class I regions on DQB1*0302-DQA1*03- DRB1*0401-haplotypes, which may suggest little, or no, effect on this particular DQ-DR background or, alternatively, lack of power to detect the possible additional effect. Zavattari et al 10 proposed the existence of additional modifying gene effects other than DPB1 centromeric of DQB1. However, their data could be interpreted as the effect of a single as yet untyped locus located in the class II centromeric class I region, in accordance with our data and with a recent regression-based analysis. 41 In addition, it is possible that the same single locus in the class II centromeric class I region also explains the heterogeneity in risk reported for DQ8-DRB1*0404 haplotypes in Finland. 7,9 Zavattari et al 10 suggested that three markers, DMB, DOB and TNFc, were independently associated with T1D, but these data could be explained by a single extended susceptibility haplotype. Nonetheless, Sardinian haplotypes differ significantly centromeric of DQB1, and a third non-dq/dr variant reducing the DQ2-DR3 haplotype susceptibility in this population is a possibility. The suggestion of three non- DQ/DR/DP loci 10 rather than one, not including the D6S2223-associated region, can only be verified by further genotyping and haplotype mapping. The suggested centromeric boundary (HLA-DOB) for the putative susceptibility gene in LD with the DQ2-DR3- B18 haplotype should be confirmed in other studies since haplotype analysis cannot completely rule out a more centromeric location. In fact, it cannot formally be excluded that the increased transmission of alleles covering the 2.35 Mb region between DOB and D6S2702 may be explained by DPB1*0202 although comparison with other studies argues against this explanation. This further illustrates the daunting task of mapping disease genes in the HLA complex. However, we show here the potential of using haplotype analyses for fine mapping of candidate regions. Even in a region of such strong LD as the HLA complex, by comparing different haplotypes across ethnically distinct populations, we can begin to tease out the possible location of the non-dq/dr T1D susceptibility genes, one of which appears to contain an allele in strong LD with the DR3-DQ2-B18 haplotype. Direct and complete analysis of the candidate genes in the 2.6 Mb region between DPB1 and D6S2702 may provide a distribution of allelic variants for each candidate gene that either explains the haplotype associations observed, or not. Such studies will need to be coupled to functional studies, to help identify the variants that can have a substantial influence on the action of the DQ/DR genes in T1D. Acknowledgements We thank Anne Brit Thoresen and Siri Flåm for excellent technical assistance. The study was financially supported by The Norwegian Research Council, Juvenile Diabetes Research Foundation International (grants 1-2000-514, 3-2000-515 and 4-2000-948), the Welcome Trust, The Novo Nordisk Foundation, The Norwegian Diabetes Association, the Swedish Diabetes Association, the Swedish Childhood Diabetes Fund and the Swedish Medical Research Council (K98-19X-12674-01A). We thank the Norwegian Bone Marrow Donor Registry, the Norwegian Study Group for Childhood Diabetes, the Danish Study Group of Diabetes in Childhood, the Danish IDDM Epidemiology and Genetics Group, Diabetes UK, Swedish Childhood Diabetes Study, the Diabetes Incidence Study in Sweden, the French Network Inserm for IDDM and MS in Atlantic Pyrenees and the Regional Observatory of Health in Aquitaine region (Dr J Doutreix) for the collection of samples for this study. References 1 Concannon P, Gogolin-Ewens KJ, Hinds DA et al. A secondgeneration screen of the human genome for susceptibility to insulin-dependent diabetes mellitus. Nat Genet 1998; 19: 292 296. 2 Mein CA, Esposito L, Dunn MG et al. A search for type 1 diabetes susceptibility genes in families from the United Kingdom. Nat Genet 1998; 19: 297 300.

3 She JX. Susceptibility to type 1 diabetes: HLA-DQ and DR revisited. Immunol Today 1996; 17: 323 329. 4 Thorsby E. HLA associated diseases. Anniversary review article Hum Immunol 1997; 53: 1 11. 5 Robinson WP, Barbosa J, Rich SS, Thomson G. Homozygous parent affected sib pair method for detecting disease predisposing variants: application to insulin dependent diabetes mellitus. Genet Epidemiol 1993; 10: 273 288. 6 Degli-Esposti MA, Abraham LJ, McCann V, Spies T, Christiansen FT, Dawkins RL. Ancestral haplotypes reveal the role of the central MHC in the immunogenetics of IDDM. Immunogenetics 1992; 36: 345 356. 7 Reijonen H, Nejentsev S, Tuokko J et al. HLA-DR4 subtype and -B alleles in DQB1*0302-positive haplotypes associated with IDDM. The Childhood Diabetes in Finland Study Group. Eur J Immunogenet 1997; 24: 357 363. 8 Lie BA, Todd JA, Pociot F et al. The predisposition to type 1 diabetes linked to the human leukocyte antigen complex includes at least one non-class II gene. Am J Hum Genet 1999; 64: 793 800. 9 Nejentsev S, Gombos Z, Laine AP et al. Non-class II HLA gene associated with type 1 diabetes maps to the 240-kb region near HLA-B. Diabetes 2000; 49: 2217 2221. 10 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. 11 Undlien DE, Lie BA, Thorsby E. HLA complex genes in type 1 diabetes and other autoimmune diseases. Which genes are involved? Trends Genet 2001; 17: 93 100. 12 Carrington M. Recombination within the human MHC. Immunol Rev 1999; 167: 245 256. 13 Clerget-Darpoux F. The HLA component of type I diabetes. Am J Hum Genet 2000; 66: 1468. 14 Degli-Esposti MA, Andreas A, Christiansen FT, Schalke B, Albert E, Dawkins RL. An approach to the localization of the susceptibility genes for generalized myasthenia gravis by mapping recombinant ancestral haplotypes. Immunogenetics 1992; 35: 355 364. 15 Todd JA, Mijovic C, Fletcher J, Jenkins D, Bradwell AR, Barnett AH. Identification of susceptibility loci for insulindependent diabetes mellitus by trans-racial gene mapping. Nature 1989; 338: 587 589. 16 Goldstein DB. Islands of linkage disequilibrium. Nat Genet 2001; 29: 109 111. 17 Bennett ST, Wilson AJ, Esposito L et al. Insulin VNTR allelespecific effect in type 1 diabetes depends on identity of untransmitted paternal allele. The IMDIAB Group. Nat Genet 1997; 17: 350 352. 18 Jongeneel CV, Briant L, Udalova IA, Sevin A, Nedospasov SA, Cambon-Thomsen A. Extensive genetic polymorphism in the human tumor necrosis factor region and relation to extended HLA haplotypes. Proc Natl Acad Sci USA 1991; 88: 9717 9721. 19 Lie BA, Sollid LM, Ascher H et al. A gene telomeric of the HLA class I region is involved in predisposition to both type 1 diabetes and coeliac disease. Tissue Antigens 1999; 54: 162 168. 20 Undlien DE, Akselsen HE, Joner G et al. No independent association of LMP2 and LMP7 polymorphisms with susceptibility to IDDM. Diabetes 1997; 46: 307 312. 21 Kockum I, Sanjeevi CB, Eastman S, Landin-Olsson M, Dahlquist G, Lernmark A. Complex interaction between HLA DR and DQ in conferring risk for childhood type 1 diabetes. Eur J Immunogenet 1999; 26: 361 372. 22 Kimura A, Sasasuki T. Eleventh International Histocompatibility Workshop reference protocol for the HLA DNA-typing technique. In: Tsuji K, Sasazuki MAT (eds). HLA 1991, Vol. I. Oxford University Press: Oxford, 1992, pp 397 419. 23 Siegmund T, Donner H, Braun J, Usadel KH, Badenhoop K. HLA-DMA and HLA-DMB alleles in German patients with type 1 diabetes mellitus. Tissue Antigens 1999; 54: 291 294. 24 Zavattari P, Lampis R, Mulargia A et al. Confirmation of the DRB1-DQB1 loci as the major component of IDDM1 in the isolated founder population of Sardinia. Hum Mol Genet 2000; 9: 2967 2972. 25 Lie BA, Akselsen HE, Joner G et al. HLA associations in insulin-dependent diabetes mellitus: No independent association to DP genes. Hum Immunol 1997; 55: 170 175. 26 Spielman RS, McGinnis RE, Ewens WJ. Transmission test for linkage disequilibrium: the insulin gene region and insulindependent diabetes mellitus (IDDM). Am J Hum Genet 1993; 52: 506 516. 27 Klitz W, Stephens JC, Grote M, Carrington M. Discordant patterns of linkage disequilibrium of the peptide-transporter loci within the HLA class II region. Am J Hum Genet 1995; 57: 1436 1444. 28 Schneider S, Kueffer J, Roessli D, Excoffier L. Arlequin ver 1.1: a Software for Population Genetic Data Analysis. Genetics and Biometry Laboratory, University of Geneva: Geneva, Switzerland, 1997. 29 Tiwari JL, Terasaki P. The data and statistical analysis. In: Tiwari JL, Terasaki P (eds). HLA and Disease Associations. Springer-Verlag: New York Heidelberg Berlin Tokyo, 1985, 18 27. 30 Crouau-Roy B, Bouzekri N, Carcassi C, Clayton J, Contu L, Cambon-Thomsen A. Strong association between microsatellites and an HLA-B, DR haplotype (B18-DR3): implication for microsatellite evolution. Immunogenetics 1996; 43: 255 260. 31 Ramsbottom D, Carcassi C, de Salages H et al. MHC haplotypes including eight microsatellites in French Basques and in Sardinians. In: Charron D (ed). Proceedings of the Twelfth International Histocompatibility Conference. Vol. II, EDK Medical and Scientific International Publisher: Paris, 1996; pp 178 180. 32 Cambon-de Mouzon A, Ohayon E, Hauptmann G et al. HLA- A, B, C, DR antigens Bf, C4 and glyoxalase I (GLO) polymorphisms in French Basques with insulin-dependent diabetes mellitus (IDDM). Tissue Antigens 1982; 19: 366 379. 33 Contu L, Deschamps I, Lestradet H et al. HLA haplotype study of 53 juvenile insulin-dependent diabetic (I.D.D.) families. Tissue Antigens 1982; 20: 123 140. 34 Deschamps I, Marcelli-Barge A, Poirier JC et al. Two distinct HLA-DR3 haplotypes are associated with age related heterogeneity in type 1 (insulin-dependent) diabetes. Diabetologia 1988; 31: 896 901. 35 Jeffreys AJ, Kauppi L, Neumann R. Intensely punctate meiotic recombination in the class II region of the major histocompatibility complex. Nat Genet 2001; 29: 217 222. 36 La Nasa G, Carcassi C, Cirillo R et al. Serological and molecular studies of HLA in insulin-dependent diabetes mellitus in Sardinia. Dis Markers 1990; 8: 333 340. 37 Erlich HA, Rotter JI, Chang JD et al. Association of HLA- DPB1*0301 with IDDM in Mexican-Americans. Diabetes 1996; 45: 610 614. 38 Valdes AM, Thomson G, Erlich HA, Noble JA. Association between type 1 diabetes age of onset and HLA among sibling pairs. Diabetes 1999; 48: 1658 1661. 39 Noble JA. The HLA class II locus DPB1 can influence susceptibility to type 1 diabetes. Diabetes 2000; 49: 121 125. 40 Cucca F, Dudbridge F, Loddo M et al. The HLA-DPB1 associated component of the IDDM1 and its relationship to the major loci HLA-DQB1, -DQA1, and -DRB1. Diabetes 2001; 50: 1200 1205. 41 Cordell HJ, Clayton DG. A Unified Stepwise Regression Procedure for Evaluating the Relative Effects of Polymorphisms within a Gene Using Case/Control or Family Data: Application to HLA in Type 1 Diabetes. Am J Hum Genet 2001; 70: 1. 53