Original Article

Genes and Immunity (2009) 10, 56–67; doi:10.1038/gene.2008.74; published online 2 October 2008

Conditional analyses on the T1DGC MHC dataset: novel associations with type 1 diabetes around HLA-G and confirmation of HLA-B

M C Eike1,2, T Becker3,4, K Humphreys4, M Olsson4,5 and B A Lie1

  1. 1Institute of Immunology, Rikshospitalet University Hospital, Oslo, Norway
  2. 2Institute of Immunology, Faculty Division Rikshospitalet, University of Oslo, Oslo, Norway
  3. 3Institute for Medical Biometry, Informatics and Epidemiology, University of Bonn, Bonn, Germany
  4. 4Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden
  5. 5Mathematical Statistics, Chalmers University of Technology, Gothenburg, Sweden

Correspondence: MC Eike, Institute of Immunology, Rikshospitalet University Hospital, Sognsvannsveien 20, Oslo N-0027, Norway. E-mail: Morten.Eike@rr-research.no

Received 16 June 2008; Revised 18 August 2008; Accepted 27 August 2008; Published online 2 October 2008.

Top

Abstract

The major histocompatibility complex (MHC) is known to harbour genetic risk factors for type 1 diabetes (T1D) additional to the class II determinants HLA-DRB1, -DQA1 and -DQB1, but strong linkage disequilibrium (LD) has made efforts to establish their location difficult. This study utilizes a dataset generated by the T1D genetics consortium (T1DGC), with genotypes for 2965 markers across the MHC in 2321 T1D families of multiple (mostly Caucasian) ethnicities. Using a comprehensive approach consisting of complementary conditional methods and LD analyses, we identified three regions with T1D association, independent both of the known class II determinants and of each other. A subset of polymorphisms that could explain most of the association in each region included single nucleotide polymorphisms (SNPs) in the vicinity of HLA-G, particular HLA-B and HLA-DPB1 alleles, and SNPs close to the COL11A2 and RING1 genes. Apart from HLA-B and HLA-DPB1, all of these represent novel associations, and subpopulation analyses did not indicate large population-specific differences among Caucasians for our findings. On account of the unusual genetic complexity of the MHC, further fine mapping is demanded, with the possible exception of HLA-B. However, our results mean that these efforts can be focused on narrow, defined regions of the MHC.

Keywords:

major histocompatibility complex, type 1 diabetes, conditional analysis, T1DGC

Top

Introduction

Type 1 diabetes (T1D) is characterized by an autoimmune destruction of insulin-producing β-cells of the pancreas. The major heritable factor in the development of this disease is known to be conferred by certain haplotypes of the major histocompatibility complex (MHC) class II genes HLA-DRB1, -DQA1 and -DQB1 (for example see, Thomson et al., Todd et al., Sheehy et al., Noble et al.,1, 2, 3, 4). However, the MHC contains an unusually high density of genes involved in immune functions,5 many of which are good candidates for involvement in T1D. Not surprisingly, therefore, numerous reports have strongly suggested the existence of additional T1D risk factors within this region (for example see, Aly et al., Lie et al., Nejentsev et al., Zavattari et al., Johansson et al.6, 7, 8, 9, 10, 11, 12). The identification of these factors has, however, proven difficult, mainly because of the strong and extensive linkage disequilibrium (LD) in this region, with ancestral haplotypes conserved as far as 9Mb.6 In combination with the sheer strength of the DRB1-DQA1-DQB1 association, which outweighs that of any other genetic T1D risk factor found to date, this serves as a severe confounding factor that compromises conventional association strategies. Furthermore, controlling for these factors is challenged by the complexity of the association and high polymorphism of the DRB1-DQA1-DQB1 loci, resulting in a vast number of parameters.

The identification of additional susceptibility loci within the MHC has been the focus of the T1D genetics consortium (T1DGC13) MHC fine-mapping project, under which 2321 multiplex families with T1D have been genotyped for 2957 single nucleotide polymorphisms (SNPs), 66 microsatellites and eight HLA-loci spread across the entire classical MHC. We have used a set of complementary approaches with focus on controlling not only for LD effects of the DRB1-DQA1-DQB1 loci, but also for that of neighbouring markers, on the SNP and HLA-data from this dataset. This way, we identified four independent regions and narrowed these further down to a few markers. Although we cannot claim to have identified all of the additional T1D risk factors in this region, this represents a significant step towards this goal, and will guide further efforts to a small subset of genes instead of the entire MHC region.

Top

Results

Conventional transmission disequilibrium tests of the SNPs and HLA-loci in the T1DGC MHC dataset yielded hundreds of markers with highly significant associations across the entire MHC region, with the maximum association peaking around the HLA-DRB1, -DQA1 and -DQB1 genes (data not shown). Thus, the pivotal role of these genes in T1D and the extensive LD in the MHC were confirmed. To examine additional effect in this region, we applied a multi-step approach, summarized as follows: (1) All markers were tested, one at a time, for an association independent of DRB1-DQA1-DQB1. This was done by (individually) adding markers in a regression model which already included genotypes for DRB1-DQA1-DQB1; (2) A smaller subset of markers that could explain the results from the previous step was identified by forward and backward selection on top of DRB1-DQA1-DQB1, treating SNPs and HLA-loci separately; (3) The latter procedure was repeated, but this time also adjusting for the best markers of the other type identified in step 2: that is, SNPs were modeled adjusting for DRB1-DQA1-DQB1 and the best HLA-loci from step 2, and vice versa; (4) All of the significant results from steps 2 and 3 were analysed in more detail by conditional allelic tests based on haplotype estimates, as well as the analyses of LD patterns. Importantly, before all analyses we recoded the HLA-DRB1, -DQA1 and -DQB1 genotypes into phased haplotypes spanning all three loci, and treated these haplotypes as alleles of a single locus to reduce complexity (see Supplementary Methods).

First step: single point associations independent of DRB1-DQA1-DQB1

Single point main effect tests of the SNPs, adjusted for the DRB1-DQA1-DQB1 genotypes, were performed in the all-affecteds dataset (including all-affected children, see Materials and methods). After Bonferroni correction for multiple testing (threshold=0.05/2296 SNPs=2.2 × 10–5), 76 SNPs remained with significant T1D association independent of DRB1-DQA1-DQB1 (Supplementary Table 1). These SNPs were concentrated in four regions (Figure 1), broadly marked by proximity to either MHC class I loci (HLA-G and HLA-C or -B, respectively) or MHC class II loci (DRB1-DQA1-DQB1 and HLA-DPA1 or -DPB1, respectively). LD plots for these SNPs showed that the four regions appeared distinct, especially in terms of r2 (average r2 within each of the regions: 0.17–0.30; average r2 between regions <0.03; Figure 2a). Regions 1 and especially 4 also appeared distinct from the other regions in terms of D', whereas the border between regions 2 and 3 was less well-defined (Figure 2b).

Figure 1.
Figure 1 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact help@nature.com or the author

Main effects test of SNPs adjusted for DRB1-DQA1-DQB1 genotypes. Analyses were performed in the all-affecteds dataset. Only SNPs with significant results after Bonferroni-correction (P<2.2 × 10−5) are shown. SNPs that withstood regression modeling (Tables 2 and 4) and the four regions discussed in the text are indicated. Note that the results for rs2076522, rs660895 and rs6457617 in region 3 were later shown to be artefacts. Positions of genotyped HLA-loci (black) and other genes discussed in the text (grey) are included for reference. Positional values are along chromosome 6, genome build 36.

Full figure and legend (113K)

Figure 2.
Figure 2 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact help@nature.com or the author

LD plots of significant SNPs from main effects test adjusted for DRB1-DQA1-DQB1 genotypes. Analyses were performed in the all-affecteds dataset. (a): r2 plot; (b); D′ plot. The four regions discussed in the text are indicated by numbers, and genotyped HLA-loci with arrows pointing at the closest SNP.

Full figure and legend (115K)

Single point main effects tests were also applied to HLA-loci, but using the proband dataset, which included only one affected child per family (more suitable for testing of multiallelic markers; see Materials and methods). In these tests, all HLA-loci except HLA-DPA1 showed effects on T1D risk independent of the DRB1-DQA1-DQB1 genotypes, with the most significant association for HLA-B and -DPB1 (Table 1).


Second step: separate model selection for SNPs and HLA-loci

To explore dependency between the SNP associations, we performed conditional logistic regression modeling on the 76 significant SNPs from the single point tests, using the all-affecteds dataset. We used the SNP with the lowest P-value in each of the four regions as starting points, and then performed forward and backward selection until no more SNPs could be included (α=0.01). The final model contained nine SNPs spread across the four regions (Table 2).


The HLA-loci were tested by a similar strategy, but using the proband dataset. Because of the smaller number of markers, we included all HLA-loci from the start. The results showed that the best model was provided by inclusion of HLA-C, -B and -DPB1 together with the DRB1-DQA1-DQB1 haplotypes (Table 3). As in the previous step, HLA-DPA1 was non-significant and HLA-B the most significant. In contrast to the single-point tests, HLA-A only appeared to have a marginal effect in this step. This appeared to be because of the presence of HLA-B in the model: when HLA-B was omitted from the model, HLA-A was significantly associated (P=0.0060), but if HLA-B was present and instead HLA-DPB1 was removed, HLA-A obtained a P-value of 0.087.


Third step: model selection combining SNPs and HLA-loci

To investigate the relationship between associated SNPs and HLA-loci, we repeated the regression modeling procedures described above, but with the additional inclusion of the best markers of the other type identified in the previous step. For the SNPs, this meant modeling on the 76 significant SNPs from the first step, with HLA-C, -B and -DPB1 (best model step 2; Table 3) in the regression model together with DRB1-DQA1-DQB1. With this approach, the final model contained six SNPs, two in each of the first three regions, but none in region 4 (Table 4). One of the SNPs, rs6926530 in region 2, was also present in the final SNP model from step 2.


Similarly, we repeated the regression modeling of the HLA-loci with the additional inclusion of the best SNPs from step 2 (nine SNPs; Table 2) in the regression model. This procedure left only HLA-C and HLA-B with significant P-values (Table 5). The significance levels of these markers were also markedly changed compared with previous results (cf. Tables 1 and 3), with higher significance for HLA-C and lower significance for HLA-B. Note, however, that the inclusion of biallelic markers in the model is likely to reduce the power of the multiallelic markers because the phase in many cases becomes unresolvable (resulting in only one pseudo-control per case compared with the three pseudo-controls achieved with phase resolved). Therefore, these P-values are not directly comparable to those of the analyses performed in steps 1 and 2.


Fourth step: validation using LD patterns and haplotype estimates

To evaluate the significance, validity and relationship between the best model markers identified in the regression analyses, we examined LD patterns and performed conditional allelic tests based on haplotype estimates. In addition to using the complete all-affecteds dataset, these analyses were also performed separately in each of the three European subpopulations because of the possible influence of population stratification. To minimize the number of analyses and issues of multiple testing, we only included markers with significant P-values in steps 2 and 3 (Tables 2, 3, 4 and 5).

The overall relationship between the best model markers in terms of global LD (D′) patterns in the complete dataset is given in Figure 3a. As for the LD plot of the 76 significant SNPs in the single point tests (Figures 2a and b), these results supported the idea of independent regions, although some LD were observed between the regions (average D′ within each of the four regions: 0.48–0.86; average D′ between regions <0.31). The general LD patterns were similar in the subpopulations, but with some notable exceptions for rs6926530 (Figures 3b–d).

Figure 3.
Figure 3 - Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, please contact help@nature.com or the author

LD plot (D′) of SNPs and HLA-loci from best model selections. Analyses were performed in the all-affecteds dataset. (a) All samples; (b) Northern Europeans; (c) Eastern Europeans; (d) Southern Europeans. The four regions discussed in the text are indicated in each plot, and the D′ value scale is given beneath the figure. Values are for non-transmitted haplotypes. HLA-DR-DQ: phased haplotypes of the HLA-DRB1, -DQB1 and -DQA1 loci.

Full figure and legend (263K)

The distribution of alleles of the best model markers on each separate DRB1-DQA1-DQB1 haplotype was mapped using family-based haplotype estimates. Significant associations from analyses of one SNP at a time conditioned on the different DRB1-DQA1-DQB1 haplotypes are given in Supplementary Table 2, and for HLA-loci in Supplementary Table 3. For the HLA-loci, we additionally calculated approximate associations for individual alleles across all DRB1-DQA1-DQB1 haplotypes by reorganising the results according to the allele, regardless of the significance level (Supplementary Table 4). Note that sibship dependencies in the all-affecteds dataset also introduce some linkage (see Supplementary Methods): the results of these analyses should therefore be regarded as descriptive, and only be used for evaluation of the relative contribution of each polymorphism or allele.

After mapping each polymorphism separately, we additionally investigated relationships between associations mapping to the same DRB1-DQA1-DQB1 haplotypes, within and across regions, by mapping two and two polymorphisms together. This often resulted in small numbers and non-significant associations (data not shown). However, the underlying haplotype patterns between the associated alleles were revealed, which to some extent made it possible to determine whether one of the polymorphisms could explain the other, or if they marked independent effects.

Region 1: rs4122198, rs1619379, rs1611133 and rs2394186

In region 1, the four best SNPs identified by conditional regression modeling (Tables 2 and 4) showed high LD in terms of D′ for all combinations (D′=0.72–0. 99 in the complete dataset, Figure 3a), although the individual associations partly mapped to different DRB1-DQA1-DQB1 haplotypes (Supplementary Table 2). The haplotype patterns observed when mapping combinations of two SNPs together on the different DRB1-DQA1-DQB1 haplotypes also indicated that rs1619379 and rs1611133 marked the same effects, whereas other SNP combinations appeared at least partly independent of each other (data not shown). Relatively high LD was also observed between rs1611133 and polymorphisms in region 2 (up to D′=0.65 in the complete dataset; Figure 3a). In concordance with this, mapping two-locus combinations from region 1 and region 2 together on the different DRB1-DQA1-DQB1 haplotypes revealed that the associations on DRB1*03-DQA1*0501-DQB1*0201 for both rs1611133 and rs1619379 (but not rs4122198) could be explained by HLA-B, most notably by the B*18 and B*08 alleles (data not shown). Moreover, the associations for rs4122198 and rs2394186 on the DRB1*08-DQA1*0401-DQB1*0402 haplotype could be explained by B*39. However, no further connections were apparent with polymorphisms outside region 1 (data not shown). Consequently, three SNPs, that is, rs4122198 and rs2394186 together with rs1619379 or rs1611133, were necessary to explain the association in region 1, which although picking up some of the association in region 2, appeared to mark effects that were truly independent of other regions.

Region 2: rs6926530, rs3130695, rs4713468, rs2246626, HLA-C and -B

In region 2, four SNPs, HLA-C and -B appeared in the best regression models (Tables 2, 3, 4 and 5). Allelic association tests conditioned on DRB1-DQA1-DQB1 haplotypes showed that HLA-C alleles C*05, C*06, C*07 and C*12 and HLA-B alleles B*13, B*18 and B*39 generally appeared to be predisposing, whereas C*01, C*02, C*04, B*08, B*35 and B*44 appeared to be protective (Supplementary Table 3). Notably, however, C*05, C*06, C*07, C*12, B*08 and B*44 also showed non-significant tendencies in the opposite direction on certain DRB1-DQA1-DQB1 haplotypes (Puncorrected<0.05 in the complete dataset, data not shown), which may indicate that the associations for these alleles were secondary. In contrast, B*18 and in particular B*39 showed a strikingly consistent and strong predisposing association pattern, both on individual DRB1-DQA1-DQB1 haplotypes (Supplementary Table 3; also when regarding non-significant tendencies: data not shown) and across all haplotypes (Supplementary Table 4). Moreover, mapping of HLA-C and -B together on the different DRB1-DQA1-DQB1 haplotypes revealed that B*39, B*18 or B*13 were able to explain most of the positive association of C*05, C*06, C*07 and C*12 (data not shown). This was consistent with the high LD observed between these two markers (D=0.84 in the complete dataset; Figure 3a) and the above suggested secondary association for these particular HLA-C alleles. Therefore, although C*01, C*02 and C*04 showed signs of independent association, HLA-B appeared to be largely primary to HLA-C.

As for region 1, some of the SNPs within this region were in relatively high LD, especially rs2246626 and rs4713468 (D′=0.99 and r2=0.63 in the complete dataset; Figure 3a). Rs6926530, which was the only SNP that showed up in both regression models (Tables 2 and 4), appeared distinct from the other SNPs in this region, but displayed a high global LD with DRB1-DQA1-DQB1 (D=0.94) or rs660895 (D=0.81) in region 3 in the Northern or Eastern European datasets, respectively (Figures 3b and c). Both of these observations could indicate that this SNP association was an artefact (see next section). Moreover, rs6926530*G is a rare allele (frequency 1.9% in founders), with only 10 informative transmissions behind the single significant association observed (Supplementary Table 2). Therefore, this SNP is unlikely to mark an important contribution to the risk of developing T1D, if at all.

When including the best SNPs in the regression model together with DRB1-DQA1-DQB1, the HLA-B association was heavily reduced (best HLA-model step 3; Table 5). Although this may partly be attributed to the loss of power when including biallelic SNPs (see Third step, p. 4), it is noteworthy that one of these SNPs, rs3130695, was in high global LD with HLA-B (D=0.81 in the complete dataset, Figure 3a). Moreover, strong allelic LD (D′ but not r2) was observed between most of the associated HLA-B alleles and alleles at rs3130695, rs2246626 or rs4713468, although sometimes between alleles showing opposite risk tendencies (data not shown). It may therefore be speculated that the reduction in significance for HLA-B partly was attributable to a redundancy between HLA-B and the SNPs. Accordingly, exploring combinations of HLA-B and either of these SNPs revealed that all of the significant individual allelic associations where HLA-B and an SNP mapped to the same DRB1-DQA1-DQB1 haplotype (cf. Supplementary Tables 2 and 3) could be explained by HLA-B alleles (data not shown). An exception was the association of rs3130695 on the DRB1*1301-DQA1*0103-DQB1*0603 haplotype, which appeared to be independent both of HLA-B and any other marker, inside or outside region 2 (data not shown).

In summary, HLA-B, and in particular alleles B*18 and B*39, appeared to best capture the T1D association in region 2, although the inclusion of HLA-C (in particular alleles C*01, C*02 and C*04) and rs3130695 may be necessary to explain all of the observed association in this region.

Region 3: rs3132959, rs2076522, rs660895 and rs6457617

In region 3, associations for the SNPs rs2076522, rs660895 and rs6457617 predominantly mapped to the group of rare DRB1-DQA1-DQB1 haplotypes (P<1.0 × 10–8; Supplementary Table 2). This is a heterogeneous group, and these associations should therefore be regarded as artefacts. All of these SNPs were in strong LD with DRB1-DQA1-DQB1 (D=0.88–0.96 in the complete dataset; Figure 3a), which was reflected in near invariance on the majority of individual DRB1-DQA1-DQB1 haplotypes, particularly for rs660895 and rs6457617. A significant association was seen for rs6457617 on the DRB1*0901-DQA1*0301-DQB1*0303 haplotype (P=0.0013), but the association on the grouped rare haplotype (P=5.6 × 10–9) appeared to be the main contributor to the overall association. The remaining SNP in this region, rs3132959, showed a different association pattern and lower LD with DRB1-DQA1-DQB1 (D=0.63 in the complete dataset; Figure 3a). However, the only significant association, on DRB1*07-DQA1*0201-DQB1*0201 (Supplementary Table 2), could apparently be explained by the HLA-B associations mapping to the same haplotype (B*13 and B*44; data not shown). Therefore, none of the SNPs in region 3 appeared to represent truly independent associations.

Region 4: rs9368757, rs439121 and HLA-DPB1

In region 4, two SNPs and the HLA-DPB1 locus emerged from the regression modeling procedures (Tables 2 and 3). The haplotype analyses showed that DPB1*0301 and DPB1*0401 appeared predisposing, whereas DPB1*0402 appeared protective (Supplementary Table 3). Non-significant association tendencies (Puncorrected<0.05 in the complete dataset; data not shown) for DPB1*0301 and in particular DPB1*0402 were consistent across several additional haplotypes, whereas DPB1*0401 showed a contrasting protective tendency on certain haplotypes, which, similar to some of the marker alleles in region 2, may indicate secondary association.

The regression modeling procedures left no significant SNPs or HLA-loci in this region when adjusting for markers of the other type (best model step 3, Tables 4 and 5). Haplotype-based analyses combining these polymorphisms with those in other regions did not indicate any strong connections (data not shown), in agreement with the low global LD observed across the border of region 4 (Figures 1 and 3a–d). Therefore, a possible explanation for the regression results is that the associations seen with the two marker types within this region were closely connected, partly cancelling each other out. The global LD was also markedly higher between HLA-DPB1 and the SNPs than between the SNPs themselves (D′=0.62–0.76 compared with D=0.06 in the complete dataset; Figure 3a). Furthermore, mapping of HLA-DPB1 together with either SNP on the different DRB1-DQA1-DQB1 haplotypes showed that DPB1*0402 and DPB1*0301 were almost always located together with the protective and predisposing SNP alleles, respectively (data not shown). This was reflected in generally high LD values between these specific alleles, especially between rs9368757*A and DPB1*0402 (D=0.90, r2=0.53). Moreover, the significant individual allelic associations of HLA-DPB1 and rs9368757 broadly mapped to the same DRB1-DQA1-DQB1 haplotypes (cf. Supplementary Tables 2 and 3), and neither appeared better at explaining the overall association than the other (data not shown). HLA-DPB1 and rs9368757 therefore appeared to mark the same effect. In contrast, the relatively high LD between certain alleles was not reflected in the individual association patterns of HLA-DPB1 and rs439121, with only one haplotype in common (cf. Supplementary Tables 2 and 3), and mapping of these two markers together indicated that these associations were largely independent (data not shown).

Taken together, in region 4, rs439121 and either rs9368757 or the HLA-DPB1 locus (principally involving the DPB1*0402 and 0301 alleles) seemed necessary to explain the association with T1D.

Extended LD analyses

The SNPs genotyped in this screen represent only a limited number of the total number of SNPs in the MHC, and so might tag other non-genotyped causal SNPs. Therefore, we performed extended analyses of LD to look for SNPs in particular high LD (r2>0.8) with our best model SNPs, as these are expected to convey much of the same information. For these analyses, we used HapMap data (CEPH population), as well as all of the SNPs in the T1DGC all-affecteds dataset within the defined regions. Boundaries for these regions were demarcated by the 76 best SNPs (Figure 1), but excluding rs6926530, rs4713468 and rs2246626 in region 2 and all of the SNPs in region 3 (because of results of the haplotype-based analyses, pp. 5–6). For region 1, we additionally included flanking regions extending to the UBD gene and HLA-A gene, respectively, because of recently reported T1D associations with these genes.6, 10 All of the SNPs that were in high LD with our prime candidate SNPs were located close to the best model SNPs, with the greatest distance observed in region 4 (23kb; Supplementary Table 5). Moreover, the genes most often implicated by these LD SNPs were the same as those marked by the best model SNPs (HLA-G, COL11A2 and RING1, see Discussion, pp. 7–8).

Top

Discussion

In this paper, we have used a combination of stepwise conditional regression and conditional allelic tests with family-based haplotype estimates to narrow down candidate regions for T1D risk factors within the classical MHC. By these means, we identified four separate regions with highly significant T1D association independent of the known risk factors HLA-DRB1, -DQA1 and -DQB1, and demonstrated independence for three of these regions also of each other. The associations in region 3 were shown to be dependent on HLA-B in the neighbouring region 2 or on artefacts of rare DRB1-DQA1-DQB1 haplotype grouping, aspects that were only revealed by haplotype-based analyses. Thus, large physical distance (cf. Miretti et al.14), regression modeling procedures and LD plots were not sufficient. This level of complexity is not surprising in the MHC, which is known to contain frequent ancestral haplotypes with variable conservation between distant markers,15, 16, 17, 18 and LD patterns that show great variation depending on the particular DRB1-DQA1-DQB1 haplotype.19, 20, 21 Therefore, it seems clear that only analyses involving haplotypes will give the necessary depth in the results to establish true independence.

The regions and markers identified in this study partly confirm associations reported previously, but also represent novel findings. Although none of our prime candidate SNPs have been reported associated with T1D earlier, potential connections with previous findings were evident.

Novel associated SNPs in region 1 are proximal to HLA-G

The best SNPs we identified in region 1—rs4122198, rs2394186, rs1619379 and rs1611133—appeared partly redundant, especially the two latter SNPs. We have previously shown that a microsatellite located in the middle and within 46kb of these SNPs, D6S2773, is also associated independently of DRB1-DQA1-DQB1 in the T1DGC MHC dataset (Eike et al., submitted). High global LD was observed between this marker and rs4122198 or rs1611133 (D=0.76 and 1, respectively), and between D6S2773*227 (predisposing) and the predisposing alleles of all the SNPs (D=0.76–1; r2=0.02–0.52). Together with the strong LD between the SNPs themselves and the relatively small distances, this suggests that all of these markers, at least in part, mark the same locus. The only NCBI Reference Sequence (RefSeq22) gene within this region is the nonclassical MHC class I gene HLA-G (D6S2773 being the closest marker, 3.5kb 5′ of the coding region; see also Figure 1), which has been implicated in asthma.23, 24 More importantly, a recent report demonstrated constitutive expression of HLA-G in pancreatic islets and suggested an immunoregulatory role with possible implications for autoimmunity.25 Unlike the classical MHC class I genes, the non-classical genes exhibit low polymorphy in the coding regions. Therefore, these results suggest that fine mapping of the HLA-G gene, with the surrounding region to include putative regulatory sites, is warranted in T1D.

In a recent study by Nejentsev et al.,10 using large British T1D case–control datasets, HLA-A was reported to be associated in regression analyses adjusting for HLA-DRB1 and -DQB1, and additionally when adjusting for HLA-B. Although this was not observed in a family dataset in the same study (850 British and US T1D families, most of which are included in the T1DGC MHC dataset; however, a different panel of SNPs were genotyped), this was mainly attributed to weaker statistical power but is in concordance with the negative finding in our study. However, an alternative explanation is that HLA-A acted as a proxy for another locus in the vicinity. HLA-A showed high global LD (D=0.70–0.94) with all of the best four associated SNPs we identified in region 1. More importantly, protective or predisposing alleles of the SNPs in many cases showed LD values close or equal to D=1 with the corresponding HLA-A alleles reported to be associated in the study by Nejentsev et al. (A*01, A*11 and A*31 and A*03 and A*24, respectively). Moreover, a recent study by Aly et al.6 utilizing parts of the T1DGC MHC dataset (1240 families) and the DAISY cohort, were also not able to confirm the association with HLA-A, but reported two SNPs, rs389419 and rs1233478 in the UBD/MAS1 L gene region, which is located ~300kb telomeric of region 1. Again, rs389419 was in high global LD with the four best SNPs we identified in region 1 (D=0.88-0.96), as well as with the HLA-A gene (D=0.95). This SNP may therefore potentially reflect the same effect as reported in both our and Nejentsev's study. In contrast, rs1233478 was in low or moderate LD with the same markers (D=0.10–0.57). However, neither of the SNPs reported by Aly et al. showed a significant effect independent of DRB1-DQA1-DQB1 in our analyses. Moreover, extended LD analyses between our associated SNPs and SNPs from HLA-A to the UBD/MAS1 L region showed the strongest LD with SNPs within or very close to region 1, both for T1DGC families and for the HapMap CEPH population.

Consequently, there is separate evidence of association between T1D and several loci in the telomeric part of the classical MHC. Even though the final conclusions in each report seem contradictory, the apparent LD between most of the identified markers may be taken as a strong evidence that there is an additional T1D susceptibility locus in this part of the MHC, although the exact location remains to be verified.

Strong evidence for a primary role of HLA-B in region 2

The strongest evidence in any of our conditional analyses was obtained for HLA-B, which appeared to account for most of the association in region 2, independently of all other investigated markers. Several earlier reports have pointed to this gene as a T1D risk factor independent of HLA-DRB1 and -DQB1, including convincing replications in independent case–control populations in the report by Nejentsev and co-workers, with the most consistent evidence, as in our study, for the B*39 and B*18 alleles.6, 9, 10, 26, 27 We also found a protective effect of B*44, consistent with earlier findings.27 However, the association tendencies for a predisposing effect, which we observed on some of the DRB1-DQA1-DQB1 haplotypes, could indicate that this was not a primary association.

HLA-C alleles C*01, C*02 and C*04 and the SNP rs3130695 25kb telomeric of this gene (the closest gene according to RefSeq) also showed indications of truly independent associations in our study. HLA-C has previously been reported associated with T1D independently of the HLA-DRB1 and -DQB1 genes,27, 28, 29 but there is little overlap between the specific alleles reported. Moreover, Nejentsev et al.10 could not identify an independent association of this gene in their families, although this could be because of lower statistical power (850 vs 2301 families; HLA-C was not investigated in their case–control materials). Therefore, HLA-C as a risk factor in T1D remains to be verified.

It is interesting to note that Nejentsev et al.10 reported an association with rs4151651 in the CFB gene (between regions 2 and 3), independently of HLA-B, -DRB1 and -DQB1, but this SNP did not show significant independent association in our analyses (data not shown).

To conclude, the accumulated evidence strongly suggests a primary role in T1D for the HLA-B gene, although additional effects conferred by HLA-C or another locus in high LD with HLA-C might also exist in this region. HLA-B is the most polymorphic of the classical MHC class I genes, which also include HLA-A and HLA-C. The peptides encoded by these genes have central roles in both adaptive and innate immunity (in antigen presentation and as KIR ligands, respectively). Moreover, with the direct role in CD8+ T-cell-mediated destruction of β-cells in the non-obese diabetic mouse model of T1D,30, 31, 32, 33, 34, 35 these genes are excellent candidates for T1D involvement.

The associations in region 4 cannot be explained by HLA-DPB1 alone

In region 4, associated alleles of HLA-DPB1 (primarily DPB1*0402 and 0301) and rs9368757 (2.1kb. downstream of the COL11A2 gene) showed a high degree of redundancy, but with additional effects marked by rs439121 (12.4kb downstream of the RING1 gene; see also Figure 1). This redundancy, in addition to the lower power expected particularly for HLA-DPB1 (because of inclusion of bi-allelic markers in the model), could explain why no markers in region 4 were included in the best models of step 3. Of note, rs2229634 in the ITPR3 gene, 445kb centromeric of rs439121 (and thus outside region 4), was recently reported associated with T1D in a Swedish study,36 but did not show significant independent association in our analyses (data not shown), which is in agreement with other replication attempts.6, 10, 37 HLA-DPB1 has been implicated in T1D previously, predominantly involving the DPB1*0202, 0301 and 0402 alleles,4, 28, 29, 38, 39, 40, 41, 42, 43 but conflicting evidence has led some authors to question the status of this locus as an independent risk factor.44, 45 Some of this controversy could be explained by our findings, as HLA-DPB1 alone was not enough to explain all of the association in this region, and the redundancy with rs9368757 points to the possibility that another locus could account for the apparent effect of HLA-DPB1. Thus, our results suggest that there could either be more than one additional risk factor in this region, or that the causal variant is a yet unidentified locus in high LD with all of the markers we identified. The extended LD analyses did not indicate that the observed SNP effects extended far beyond the RING1 and COL11A2 genes, which suggests further fine mapping concentrated around these genes, in addition to HLA-DPB1.

Associations limited to a subset of DRB1-DQA1-DQB1 haplotypes

None of the investigated markers showed association on all individual DRB1-DQA1-DQB1 haplotypes (Supplementary Tables 2 and 3). To some degree this is expected, as the power of the haplotype-based tests relies both on allele frequencies of each of the tested marker and of the DRB1-DQA1-DQB1 loci. However, this cannot explain the whole picture, as several haplotypes were relatively common. The most striking observation was that very few of the markers seemed to influence the high-risk DRB1*0401-DQA1*0301-DQB1*0302 haplotype (only rs1619379 and rs1611133 in region 1), in contrast to the other major risk haplotype DRB1*03-DQA1*0501-DQB1*0201, despite the fact that both were present at high frequencies. In theory, such apparent haplotype-specificity could indicate specific epistatic effects between marked aetiological loci and the DRB1-DQA1-DQB1 loci. However, the underlying causal variant may also form complex haplotype structures, such that the marking by the investigated polymorphisms was detectable only on certain DRB1-DQA1-DQB1 haplotypes. It is well-known that both the LD and allelic content vary on different HLA haplotypes.19, 20, 21, 46

Population stratification and results in European subpopulations

Although the T1DGC MHC dataset consists mostly of samples of Caucasian ancestry (only approximately 2% of the samples have known non-Caucasian ancestry), population stratification is still a possible confounding factor. However, there are several reasons why this is not likely to have had a large impact on our results: Firstly, the logistic regression approach used in the first step to identify candidate markers for further analyses is, unlike the haplotype estimations, not sensitive to population stratification.47 Secondly, global LD plots of the involved markers did not show large differences between the different subpopulations and the complete dataset (Figures 3a–d). Thirdly, results in the subpopulations generally showed the same tendencies as in the complete dataset, even if results in many cases were not significant. This lack of significance was most likely because of a loss of statistical power, for example, the Southern and Eastern European datasets comprised only 161 and 228 families, respectively. Also, small population sizes become critical when dealing with rare effects, which was obvious for several of the associations reported here (for example, HLA-B*39 had a frequency of only 2.9 % in the complete dataset). On top of this, the stratification by DRB1-DQA1-DQB1 haplotypes, which in several cases had low frequencies, meant that the cell counts quickly became too small to perform meaningful tests.

Top

Conclusion

In conclusion, our results present strong evidence for the existence of at least three T1D susceptibility loci in the MHC additional to the well-established DRB1-DQA1-DQB1 loci. Furthermore, by inclusion of all of the best markers as measured by independence of DRB1-DQA1-DQB1 in conditional regression modeling, use of a complementary method based on haplotype estimates and LD analyses, we have demonstrated the independence of these loci also of each other. In particular, novel T1D risk variants have been identified close to or within the HLA-G gene, and within a region demarcated by the HLA-DPB1/COL11A2/RING1 genes. In addition, our findings provide strong support for a primary role of HLA-B and suggest involvement of HLA-C or a locus in high LD with this gene. However, the complexity of the identified associations indicate that the true aetiological polymorphisms remain largely unidentified, or in the case of HLA-B, possibly incomplete. Therefore, replication in other T1D populations and further fine mapping in or near the suggested regions is warranted.

Top

Materials and methods

The T1DGC MHC dataset (February 2007 release) consists of 2321 families (11279 individuals) from Europe, USA and the Asia Pacific, predominantly of Caucasian ancestry (see ‘Ethnic groups’ section). Each family included two affected siblings or more (59 families with only one affected child) and both parents (1718 families). SNP genotyping was performed by the Wellcome Trust Sanger Institute using the two dense standard Illumina mapping and exon-centric MHC panels (1536 SNPs in each panel with 115 overlapping SNPs). Genotyping of HLA-A, -C, -B, -DRB1, -DQA1, -DQB1, -DPA1 and -DPB1 involved four genotyping laboratories using line strip technology, with standardized methods, common reagents and a central computing system and software processing package. Some samples (mainly from the British Diabetes Association cohort; 423 families) have been genotyped for these loci previously using other methods. Further details are available on the T1DGC website (http://www.t1dgc.org).

Sample selection and handling of genotypes

Genotyping data and pedigree information were handled using Progeny Lab v6 (Progeny Software LLC, South Bend, IN, USA). The dataset was edited to contain only two generations per family (affecting 34 extended pedigrees), and children without any genotype data were removed. An unusually high number of recombinations within a family points to a potentially wrong pedigree structure. Therefore, we used FAMHAP48 to compute haplotypes for all windows of two neighbouring markers including all marker types (after application of quality controls, see ‘Marker quality checks’ section, p. 10) and counted for each family the number of windows with a recombination. Among families with at least one recombination, on average 38 (out of 2381, s.d.=60) windows per family showed a recombination. Families with more than this average +3s.d. (that is, 218 recombinations) were removed (n=21). In total, the remaining number of samples was 10618 in 2301 families (referred to as the ‘all-affecteds dataset’).

The statistical tests of multiallelic loci showed unreliable results when using the sandwich variance estimator for adjustment of sibling dependencies (see ‘Statistical analyses’ section, p. 10). We therefore made an alternative dataset containing one proband per family. The proband was chosen as the child with the youngest age of onset, if available, otherwise as the affected child with the highest analytic ID. Unaffected siblings were retained to improve phase assignment, whereas siblings diagnosed with T1D were removed. This resulted in a total number of 8190 samples in 2301 families (referred to as the ‘proband dataset’).

For extended LD analyses, we used data from the HapMap project,49 and limited our analyses to the samples most likely to be similar to the T1DGC families, the CEPH population (30 trio families resident in Utah with ancestry from Northern and Western Europe).

Ethnic groups

Ethnic groups were specified on the basis of self-reported ethnicity. The following groups were used in analyses: Caucasian (n=10436/2262 families): all families (including those where no ethnicity information was available) except those of Asian, African or indigenous non-Caucasian ancestry; Northern Europe (n=3530/781 families): families from the United Kingdom and Ireland, Scandinavia, Germany and the Netherlands; Eastern Europe (n=1029/228 families): families from Poland, the Czech Republic, Slovenia, Hungary, Serbia, Romania, Bosnia and Slovakia; Southern Europe (n=768/161 families): families from Italy, Spain, Portugal, France and Malta.

Recoding and grouping of alleles at HLA-loci

Genotypes for HLA-DRB1, -DQA1 and -DQB1 were replaced by a haplotype code spanning all three loci, with phase inferred based on common haplotypes. Mendelian consistency (verified with Pedcheck50) was evaluated for quality control. Details of this procedure are described in Supplementary Methods. For HLA-A, -B and -C, all genotypes were downcoded to 2-digit resolution to avoid artefacts of varying degrees of typing resolution. To limit the number of input variables, alleles with frequencies below 1% (calculated in parents of the all-affecteds dataset) were grouped for each HLA-locus. This included the DRB1-DQA1-DQB1 haplotype codes, with the exception of the DRB1*0403-DQA1*0301-DQB1*0302 haplotype (frequency: 0.48%) because of the demonstrated protective effect of the DRB1*0403 allele (for example see, Thomson et al., Undlien et al.1, 51).

Marker quality checks

SNP positions were defined against genome build 36 (NCBI dbSNP build 126). In cases where the rs-ID or reference sequence could not be found or showed double positional hits in dbSNP build 126, genotypes were removed. In addition, we removed SNPs with reports of more than two alleles in dbSNP build 127, SNPs that were monoallelic or with minor allele frequencies below 1% (calculated in parents of the all-affecteds dataset) and SNPs with genotype success rates below 80% (calculated among individuals in the all-affecteds dataset with at least one successful result for any SNP). Hardy–Weinberg equilibrium was calculated in the above described ethnic groups using the exact test implemented in the program Pedstats.52 SNPs showing disequilibrium with P<10−5 in any one of the populations, or <10−3 in at least two separate populations, were removed. All genotypes were investigated for Mendelian errors using Pedcheck, and inconsistent genotype data was removed for the affected polymorphism in the entire involved family. The above procedures resulted in a final dataset containing 2296 SNPs and all 5 HLA-loci, in addition to the DRB1-DQA1-DQB1 haplotype code.

For HapMap data, SNPs with minor allele frequency <0.001, that were not in Hardy–Weinberg equilibrium (P<0.05) or with Mendelian errors were excluded.

Statistical analyses

Transmission disequilibrium test for the SNPs were performed in the all-affecteds dataset using PLINK v1.053 and for HLA-loci in the proband dataset using UNPHASED v2.4.54 Conditional logistic regression, adjusted for DRB1-DQA1-DQB1 genotypes, was used for evaluation of the importance of SNPs and allelic variants at the HLA-A, -C, -B, -DPA1 and -DPB1 loci on T1D susceptibility. We used the conditioning strategy 4 described in Cordell and Clayton,55 implemented in STATA (STATA corp., TX, USA). Allele models were used for the SNPs, and genotype models for the multiallelic HLA-loci. Each SNP was tested in the all-affecteds dataset using the Wald test with the sandwich variance estimator, to account for dependence between affected children in the same family. For multiallelic loci this approach gave unreliable results, most likely due to having too many low-frequent genotypes in the model. Therefore, we restricted the analyses of the HLA-loci to the proband dataset, using the likelihood ratio test.

LD measures (D′ and r2) between SNPs were calculated using Haploview v4.0,56 whereas global LD (D′) between SNPs and multiallelic loci were analyzed using FAMHAP, based on the generalized D′ measure. Raw data plots from the FAMHAP output were generated with GOLD57 and edited using Adobe Illustrator. Calculations of LD between specific alleles (D′ and r2) were performed using UNPHASED with expectation maximization (EM) estimation and values for non-transmitted haplotypes.

Details of the haplotype-based methods are available in the Supplementary Methodssection. Briefly, maximum-likelihood haplotype frequency estimates combining markers with DRB1-DQA1-DQB1 haplotypes were computed in the all-affecteds dataset (both in the total material and separately in each of the subpopulations) using FAMHAP, excluding families who were recombinant for the respective markers. Based on these estimates, haplotype transmission/non-transmission (T/NT) tables were constructed as described elsewhere (counting T/NT from heterozygous parents48) and haplotypes were organized in separate groups for each DRB1-DQA1-DQB1 haplotype. Statistically significant deviations of a marker within each of these DRB1-DQA1-DQB1 groups were tested in 2 × 2 contingency tables, using Pearson's χ2 test, or Fisher's exact test (two-tailed; using EpiCalc 2000 v1.02; http://www.brixtonhealth.com/epicalc.html) when any of the cell counts were below 5, and odds ratios were calculated. To exclude possible artefacts of rare genotyping errors, results for T+NT<10 were excluded. For calculations of approximate association levels of individual HLA-alleles across all DRB1-DQA1-DQB1 haplotypes (excluding the grouped rare haplotypes, which provides no meaningful information), we reorganized the above results according to HLA-alleles and added together χ2-values for all individual tests for each allele where T and NT were greater than 4.

Top

References

  1. Thomson G, Valdes AM, Noble JA, Kockum I, Grote MN, Najman J et al. Relative predispositional effects of HLA class II DRB1-DQB1 haplotypes and genotypes on type 1 diabetes: a meta-analysis. Tissue Antigens 2007; 70: 110–127. | Article | PubMed | ChemPort |
  2. Todd JA, Bell JI, McDevitt HO. HLA-DQ beta gene contributes to susceptibility and resistance to insulin-dependent diabetes mellitus. Nature 1987; 329: 599–604. | Article | PubMed | ISI | ChemPort |
  3. Sheehy MJ, Scharf SJ, Rowe JR, Neme de Gimenez MH, Meske LM, Erlich HA et al. A diabetes-susceptible HLA haplotype is best defined by a combination of HLA-DR and -DQ alleles. J Clin Invest 1989; 83: 830–835. | Article | PubMed | ISI | ChemPort |
  4. 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. | PubMed | ISI | ChemPort |
  5. Horton R, Wilming L, Rand V, Lovering RC, Bruford EA, Khodiyar VK et al. Gene map of the extended human MHC. Nat Rev Genet 2004; 5: 889–899. | Article | PubMed | ISI | ChemPort |
  6. Aly TA, Baschal EE, Jahromi MM, Fernando MS, Babu SR, Fingerlin TE et al. Analysis of SNPs Identifies Major Type 1A Diabetes Locus Telomeric of the MHC. Diabetes 2008; 57: 770–776. | Article | PubMed | ChemPort |
  7. Aly TA, Ide A, Jahromi MM, Barker JM, Fernando MS, Babu SR et al. Extreme genetic risk for type 1A diabetes. Proc Natl Acad Sci USA 2006; 103: 14074–14079. | Article | PubMed | ChemPort |
  8. Lie BA, Todd JA, Pociot F, Nerup J, Akselsen HE, Joner G 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. | Article | PubMed | ISI | ChemPort |
  9. Nejentsev S, Gombos Z, Laine AP, Veijola R, Knip M, Simell O 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. | Article | PubMed | ChemPort |
  10. Nejentsev S, Howson JM, Walker NM, Szeszko J, Field SF, Stevens HE et al. Localization of type 1 diabetes susceptibility to the MHC class I genes HLA-B and HLA-A. Nature 2007; 450: 887–892. | Article | PubMed | ChemPort |
  11. Zavattari P, Lampis R, Motzo C, Loddo M, Mulargia A, Whalen 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-DQB1, -DRB1 disease loci. Hum Mol Genet 2001; 10: 881–889. | Article | PubMed | ChemPort |
  12. Johansson S, Lie BA, Todd JA, Pociot F, Nerup J, Cambon-Thomsen A 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. | Article | PubMed | ISI | ChemPort |
  13. Rich SS, Concannon P, Erlich H, Julier C, Morahan G, Nerup J et al. The Type 1 Diabetes Genetics Consortium. Ann NY Acad Sci 2006; 1079: 1–8. | Article | PubMed | ChemPort |
  14. Miretti MM, Walsh EC, Ke X, Delgado M, Griffiths M, Hunt S et al. A high-resolution linkage-disequilibrium map of the human major histocompatibility complex and first generation of tag single-nucleotide polymorphisms. Am J Hum Genet 2005; 76: 634–646. | Article | PubMed | ISI | ChemPort |
  15. Yunis EJ, Larsen CE, Fernandez-Vina M, Awdeh ZL, Romero T, Hansen JA et al. Inheritable variable sizes of DNA stretches in the human MHC: conserved extended haplotypes and their fragments or blocks. Tissue Antigens 2003; 62: 1–20. | Article | PubMed | ISI | ChemPort |
  16. Degli-Esposti MA, Leaver AL, Christiansen FT, Witt CS, Abraham LJ, Dawkins RL. Ancestral haplotypes: conserved population MHC haplotypes. Hum Immunol 1992; 34: 242–252. | Article | PubMed | ChemPort |
  17. Alper CA, Awdeh Z, Yunis EJ. Conserved, extended MHC haplotypes. Exp Clin Immunogenet 1992; 9: 58–71. | PubMed | ISI | ChemPort |
  18. Aly TA, Eller E, Ide A, Gowan K, Babu SR, Erlich HA et al. Multi-SNP analysis of MHC region: remarkable conservation of HLA-A1-B8-DR3 haplotype. Diabetes 2006; 55: 1265–1269. | Article | PubMed | ISI | ChemPort |
  19. Blomhoff A, Olsson M, Johansson S, Akselsen HE, Pociot F, Nerup J et al. Linkage disequilibrium and haplotype blocks in the MHC vary in an HLA haplotype specific manner assessed mainly by DRB1*03 and DRB1*04 haplotypes. Genes Immun 2006; 7: 130–140. | Article | PubMed | ChemPort |
  20. Ahmad T, Neville M, Marshall SE, Armuzzi A, Mulcahy-Hawes K, Crawshaw J et al. Haplotype-specific linkage disequilibrium patterns define the genetic topography of the human MHC. Hum Mol Genet 2003; 12: 647–656. | Article | PubMed | ISI | ChemPort |
  21. Alper CA, Larsen CE, Dubey DP, Awdeh ZL, Fici DA, Yunis EJ. The haplotype structure of the human major histocompatibility complex. Hum Immunol 2006; 67: 73–84. | Article | PubMed | ChemPort |
  22. Pruitt KD, Tatusova T, Maglott DR. NCBI reference sequences (RefSeq): a curated non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids Res 2007; 35 (database issue): D61–D65. | Article | PubMed | ChemPort |
  23. Nicolae D, Cox NJ, Lester LA, Schneider D, Tan Z, Billstrand C et al. Fine mapping and positional candidate studies identify HLA-G as an asthma susceptibility gene on chromosome 6p21. Am J Hum Genet 2005; 76: 349–357. | Article | PubMed | ISI | ChemPort |
  24. Tan Z, Randall G, Fan J, Camoretti-Mercado B, Brockman-Schneider R, Pan L et al. Allele-specific targeting of microRNAs to HLA-G and risk of asthma. Am J Hum Genet 2007; 81: 829–834. | Article | PubMed | ChemPort |
  25. Cirulli V, Zalatan J, McMaster M, Prinsen R, Salomon DR, Ricordi C et al. The class I HLA repertoire of pancreatic islets comprises the nonclassical class Ib antigen HLA-G. Diabetes 2006; 55: 1214–1222. | Article | PubMed | ChemPort |
  26. Nejentsev S, Reijonen H, Adojaan B, Kovalchuk L, Sochnevs A, Schwartz EI et al. The effect of HLA-B allele on the IDDM risk defined by DRB1*04 subtypes and DQB 1*0302. Diabetes 1997; 46: 1888–1892. | Article | PubMed | ChemPort |
  27. Valdes AM, Erlich HA, Noble JA. Human leukocyte antigen class I B and C loci contribute to Type 1 Diabetes (T1D) susceptibility and age at T1D onset. Hum Immunol 2005; 66: 301–313. | Article | PubMed | ChemPort |
  28. Bugawan TL, Klitz W, Alejandrino M, Ching J, Panelo A, Solfelix CM et al. The association of specific HLA class I and II alleles with type 1 diabetes among Filipinos. Tissue Antigens 2002; 59: 452–469. | Article | PubMed | ISI | ChemPort |
  29. Noble JA, Martin A, Valdes AM, Lane JA, Galgani A, Petrone A et al. Type 1 diabetes risk for human leukocyte antigen (HLA)-DR3 haplotypes depends on genotypic context: Association of DPB1 and HLA class I loci among DR3- and DR4-matched Italian patients and controls. Hum Immunol 2008; 69: 291–300. | Article | PubMed | ChemPort |
  30. de Jersey J, Snelgrove SL, Palmer SE, Teteris SA, Mullbacher A, Miller JF et al. Beta cells cannot directly prime diabetogenic CD8 T cells in nonobese diabetic mice. Proc Natl Acad Sci USA 2007; 104: 1295–1300. | Article | PubMed | ChemPort |
  31. Kay TW, Parker JL, Stephens LA, Thomas HE, Allison J. RIP-beta 2-microglobulin transgene expression restores insulitis, but not diabetes, in beta 2-microglobulin null nonobese diabetic mice. J Immunol 1996; 157: 3688–3693. | PubMed | ISI |
  32. Utsugi T, Yoon JW, Park BJ, Imamura M, Averill N, Kawazu S et al. Major histocompatibility complex class I-restricted infiltration and destruction of pancreatic islets by NOD mouse-derived beta-cell cytotoxic CD8+ T-cell clones in vivo. Diabetes 1996; 45: 1121–1131. | Article | PubMed | ChemPort |
  33. Sumida T, Furukawa M, Sakamoto A, Namekawa T, Maeda T, Zijlstra M et al. Prevention of insulitis and diabetes in beta 2-microglobulin-deficient non-obese diabetic mice. Int Immunol 1994; 6: 1445–1449. | Article | PubMed | ChemPort |
  34. Serreze DV, Leiter EH, Christianson GJ, Greiner D, Roopenian DC. Major histocompatibility complex class I-deficient NOD-B2mnull mice are diabetes and insulitis resistant. Diabetes 1994; 43: 505–509. | Article | PubMed | ISI | ChemPort |
  35. Wicker LS, Leiter EH, Todd JA, Renjilian RJ, Peterson E, Fischer PA et al. Beta 2-microglobulin-deficient NOD mice do not develop insulitis or diabetes. Diabetes 1994; 43: 500–504. | Article | PubMed | ISI | ChemPort |
  36. Roach JC, Deutsch K, Li S, Siegel AF, Bekris LM, Einhaus DC et al. Genetic mapping at 3-kilobase resolution reveals inositol 1,4,5-triphosphate receptor 3 as a risk factor for type 1 diabetes in sweden. Am J Hum Genet 2006; 79: 614–627. | Article | PubMed | ISI | ChemPort |
  37. Qu HQ, Marchand L, Szymborski A, Grabs R, Polychronakos C. The association between type 1 diabetes and the ITPR3 gene polymorphism due to linkage disequilibrium with HLA class II. Genes Immun 2008; 9: 264–266. | Article | PubMed | ChemPort |
  38. Erlich HA, Rotter JI, Chang JD, Shaw SJ, Raffel LJ, Klitz W et al. Association of HLA-DPB1*0301 with IDDM in Mexican-Americans. Diabetes 1996; 45: 610–614. | Article | PubMed | ChemPort |
  39. Cruz TD, Valdes AM, Santiago A, Frazer de Llado T, Raffel LJ, Zeidler A et al. DPB1 alleles are associated with type 1 diabetes susceptibility in multiple ethnic groups. Diabetes 2004; 53: 2158–2163. | Article | PubMed | ChemPort |
  40. Noble JA, Valdes AM, Thomson G, Erlich HA. The HLA class II locus DPB1 can influence susceptibility to type 1 diabetes. Diabetes 2000; 49: 121–125. | Article | PubMed | ChemPort |
  41. Valdes AM, Noble JA, Genin E, Clerget-Darpoux F, Erlich HA, Thomson G. Modeling of HLA class II susceptibility to Type I diabetes reveals an effect associated with DPB1. Genet Epidemiol 2001; 21: 212–223. | Article | PubMed | ChemPort |
  42. Cucca F, Dudbridge F, Loddo M, Mulargia AP, Lampis R, Angius E 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. | Article | PubMed | ChemPort |
  43. Baschal EE, Aly TA, Babu SR, Fernando MS, Yu L, Miao D et al. HLA-DPB1*0402 protects against type 1A diabetes autoimmunity in the highest risk DR3-DQB1*0201/DR4-DQB1*0302 DAISY population. Diabetes 2007; 56: 2405–2409. | Article | PubMed | ChemPort |
  44. Lie BA, Akselsen HE, Joner G, Dahl-Jorgensen K, Ronningen KS, Thorsby E et al. HLA associations in insulin-dependent diabetes mellitus: no independent association to particular DP genes. Hum Immunol 1997; 55: 170–175. | Article | PubMed | ChemPort |
  45. Johansson S, Lie BA, Pociot F, Nerup J, Cambon-Thomsen A, Kockum I et al. HLA associations in type 1 diabetes: DPB1 alleles may act as markers of other HLA-complex susceptibility genes. Tissue Antigens 2003; 61: 344–351. | Article | PubMed | ChemPort |
  46. Horton R, Gibson R, Coggill P, Miretti M, Allcock RJ, Almeida J et al. Variation analysis and gene annotation of eight MHC haplotypes: The MHC Haplotype Project. Immunogenetics 2008; 60: 1–18. | Article | PubMed | ChemPort |
  47. Cordell HJ, Barratt BJ, Clayton DG. Case/pseudocontrol analysis in genetic association studies: A unified framework for detection of genotype and haplotype associations, gene-gene and gene-environment interactions, and parent-of-origin effects. Genet Epidemiol 2004; 26: 167–185. | Article | PubMed |
  48. Becker T, Knapp M. Maximum-likelihood estimation of haplotype frequencies in nuclear families. Genet Epidemiol 2004; 27: 21–32. | Article | PubMed | ISI |
  49. The International HapMap Consortium. The International HapMap Project. Nature 2003; 426: 789–796. | Article | PubMed | ISI | ChemPort |
  50. O'Connell JR, Weeks DE. PedCheck: a program for identification of genotype incompatibilities in linkage analysis. Am J Hum Genet 1998; 63: 259–266. | Article | PubMed | ISI | ChemPort |
  51. Undlien DE, Friede T, Rammensee HG, Joner G, Dahl-Jorgensen K, Sovik O et al. HLA-encoded genetic predisposition in IDDM: DR4 subtypes may be associated with different degrees of protection. Diabetes 1997; 46: 143–149. | Article | PubMed | ISI | ChemPort |
  52. Wigginton JE, Abecasis GR. PEDSTATS: descriptive statistics, graphics and quality assessment for gene mapping data. Bioinformatics 2005; 21: 3445–3447. | Article | PubMed | ISI | ChemPort |
  53. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet 2007; 81: 559–575. | Article | PubMed | ISI | ChemPort |
  54. Dudbridge F. Pedigree disequilibrium tests for multilocus haplotypes. Genet Epidemiol 2003; 25: 115–121. | Article | PubMed | ISI |
  55. 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 2002; 70: 124–141. | Article | PubMed | ISI | ChemPort |
  56. Barrett JC, Fry B, Maller J, Daly MJ. Haploview: analysis and visualization of LD and haplotype maps. Bioinformatics 2005; 21: 263–265. | Article | PubMed | ISI | ChemPort |
  57. Abecasis GR, Cookson WO. GOLD—graphical overview of linkage disequilibrium. Bioinformatics 2000; 16: 182–183. | Article | PubMed | ISI | ChemPort |
Top

Acknowledgements

This research utilizes resources provided by the Type 1 Diabetes Genetics Consortium, a collaborative clinical study sponsored by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), National Institute of Allergy and Infectious Diseases (NIAID), National Human Genome Research Institute (NHGRI), National Institute of Child Health and Human Development (NICHD) and Juvenile Diabetes Research Foundation International (JDRF) and supported by U01 DK062418. MC Eike and BA Lie were supported by JDRF Grant 1-2004-793, by the Novo Nordisk Foundation and the Norwegian Diabetes Association. K Humphreys acknowledges support from the Swedish Research Council. M Olsson was supported by SSF Grant A3 02:129.

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

Top

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated

NEWS AND VIEWS

International histocompatibility testing workshop 1977

Nature News and Views (22 Dec 1977)