Background

The four susceptibility loci identified so far in T1D, the HLA class II gene complex [1], INS [2], CTLA4 [3] and PTPN22 [4, 5] indicate that the regulation of T cell development, activation, expansion and homeostasis is a central component of disease development. A fifth locus, the IL2R2/CD25 region [6] awaits independent replication and fine mapping of the aetiological variant. With the exception of INS [7], these genes contain polymorphisms that are associated with susceptibility to other immune-mediated diseases.

Therefore, we hypothesised that further susceptibility variants for T1D may reside in genes previously associated with other immune-mediated diseases, as prior evidence suggests the presence of shared disease susceptibility genes. For example, in families with T1D, other immune-mediated diseases, such as RA and autoimmune thyroid disease (AITD), occur more frequently than expected by chance, indicative of a partly shared genetic basis [7]. This model has gained significant support recently with the association of the Arg620Trp non-synonymous SNP in the PTPN22 gene, encoding the lymphoid specific phosphatase, LYP; not only with T1D [4] but also with GD, RA and SLE [5, 8, 9]. Likewise CTLA4 has been associated with T1D, AITD, RA and SLE [3, 10].

Furthermore, as susceptibility to T1D and other autoimmune diseases is probably directly related to the homeostatic, regulatory state of the immune system, it is possible that variants in immune response genes influence susceptibility to T1D via alteration of networks of immune regulation. For example, the CARD15 gene product, NOD2, influences the development of the adaptive immune response [11, 12] and functional variants of the gene predispose to the inflammatory bowel disease (IBD), Crohn disease (Table 1). Both the Th1 and mucosal immune system are thought to be important in T1D aetiology [13].

Table 1 Previously associated polymorphisms with other immune-mediated diseases and references.
Full size table

The aim of this study was to determine whether previously associated polymorphisms of immune-mediated disease also predispose to T1D. We genotyped a total of 41 polymorphisms, including three microsatellite markers, from 16 genes, shown in Table 1, in large T1D collections. For two genes C-reactive protein (CRP), a marker for inflammation and associated with susceptibility to SLE [14, 15], and FCER1B, high-affinity receptor for immunoglobulin E (IgE) (MS4A2), located in the putative T1D locus IDDM4, and associated with atopic illness [16, 17], we also carried out a re-sequencing effort, to gain a more comprehensive profile of allelic variation of the genes and their potential association with T1D. Our sample sizes had good statistical power for ORs greater than 1.5 (Additional File 1) [18].

Methods

Subjects

T1D families were white European or of white European descent, with two parents and at least one affected child comprising DNA samples from up to 476 multiplex Diabetes UK Warren 1families [19], 278 multiplex HBDI families [20], 250 simplex Northern Ireland families [21], 260 simplex Norwegian families and 233 simplex Romanian families with inclusion criteria as reported in Vella et al. [22]. The T1D cases [23]and the 1958 BBC controls [24] have been described previously [5]. All DNA samples were collected after approval from the relevant research ethics committees and written informed consent was obtained from the participants. This project has run over a number of years during which samples of cases were still being collected. Consequently, owing to availability of DNA, polymorphisms were genotyped either in "set 1" families (n = 754 UK and USA multiplex families), "set 2" families (n = 743 Norwegian, Romanian and Northern Irish simplex families) and/or a British case-control collection consisting of between approximately 1,500 and 4,400 cases and 1,500 and 4,600 controls.

SNP identification and genotyping

CRP (EMBL Accession number AL445528) and FCER1B (AP001181) were annotated locally, importing Ensembl information into a temporary ACeDB database as described previously [25]. After confirmation of gene structures by BLAST analysis, these were re-extracted in GFF format and submitted to T1Dbase [26].

Direct sequencing of nested PCR products from 32 T1D individuals was carried out for all exons of CRP and 3 kb 5' and 3' of the gene, using Applied Biosystems 3700 capillary sequencer. Polymorphisms were identified using the Staden Package and loaded into T1Dbase. FCER1B was also sequenced, as for CRP, in 96 T1D individuals for 2 kb 5' and 3 kb 3' of the gene and all exons, except exon 6, were successfully sequenced.

SNPs were genotyped using either TaqMan MGB chemistry (Applied Biosystems), Invader Biplex assay (Third Wave Technologies, Madison) [3] or PCR RFLP. The microsatellites were genotyped on an ABI3700 using fluorescent primers. All genotyping data were double scored to minimize error.

Family studies of variants with a MAF less than 5% may be compromised by apparent under transmission of alleles resulting from undetected genotyping errors [27]. To evaluate potential genotyping errors we genotyped a large selection of samples from three SNPs: SNP8 (rs2066844), SNP12 (rs2066845), SNP13 (ss28514842), twice, using two different methods, either TaqMan, Invader or PCR RFLP. The concordance rate between the methods was >99.2%.

Statistical analyses

All statistical analyses were performed within STATA [28] making specific use of the Genassoc and htSNP2 packages for association and tag SNP selection, available from [29]. All genotyping data of unaffected parents and controls were assessed for, and found to be in Hardy-Weinberg equilibrium (P > 0.05). Associations of the SNPs and microsatellites were tested by the Transmission/disequilibrium test in the families. Allelic and genotypic relative risks were calculated using pseudo-controls and cases with conditional logistic regression [30]. In order to minimise any confounding due to variation in allele frequencies across Great Britain [31], case-control data were stratified by broad geographical region within the logistic regression model used to calculate ORs and test associations. Both genotypic and allelic effects were allowed for by modelling loci as three-level, 2 degrees of freedom (2 df), categorical variables corresponding to a model assuming no particular mode of inheritance and continuous (1 df) variables equating to a multiplicative model. Where a difference was found, by a likelihood ratio test, between the multiplicative model and a model assuming no particular mode of inheritance, the 2 df P-value is reported. The tag SNPs were analysed by use of a multivariate test statistic [6, 32].

Results

For CRP and FCER1B, we resequenced and followed a tag selection approach, selecting tags that capture the variation of the remaining common SNPs (MAF ≥ 0.05) with a minimum R 2 of 0.8 [32]. Nineteen SNPs were identified in CRP, five of which were novel (Additional File 2) [26] and seven tag SNPs were selected and genotyped in set 1 families (multilocus P = 0.49). We also genotyped the tag SNPs in the case-control collection (1,607 cases and 1,636 controls) and found no support for association (multilocus P = 0.42), indicating that common variants of CRP are unlikely to influence T1D susceptibility in a major way. The CRP intronic microsatellite (GT)n polymorphism (ss28514831) (Table 1 and Additional file 2) was also genotyped in set 1 and 2 families but showed no association with T1D (P = 0.90).

On resequencing FCER1B, which is located in the putative diabetes-susceptibility region IDDM4 [33], we identified 34 SNPs, 17 of which were novel, and selected five tag SNPs (Additional File 3) [26]. We were unable to detect the non-synonymous SNP in exon 7, Gly237Glu, in 96 DNA samples, that has previously been associated with atopic asthma [16]. We initially genotyped the five tag SNPs in set 1 families (multilocus P = 0.085), and then followed this result up in set 2 families giving a combined multilocus P = 0.070 (adjusted for two-stage design). In 1,600 cases and 1,636 controls we obtained P = 0.24, and the combination of the multilocus family and case-control results indicated that variants of FCER1B are unlikely to play a major role in T1D susceptibility (P = 0.23). The FCER1B intronic microsatellite polymorphism (ss28514807) also showed no evidence of association with T1D in set 1 and 2 families (P = 0.65).

Five SNPs from FCRL3, CFH, SLC9A3R1, PADI4, RUNX1 and six SNPs from SPINK5 were genotyped in set 1 families and a minimum of 1,500 cases and 1,500 controls (Table 2). All except FCRL3 (TDT P = 0.04) and one locus in SPINK5 (P = 0.03), showed a P value of > 0.05. However, we did not obtain any further evidence of association of FCRL3 in 1,896 cases and 2,020 controls (Table 2). IL1RN, IL1RA, IGR2198 and the CARD15 SNPs 8, 12 and 13 were genotyped in sets 1 and 2 families, and all showed P > 0.10 (Table 3). The two SLC22A4 SNPs and the single LAG3 SNP were genotyped in a minimum of 3,290 cases and 3,549 controls and showed no evidence of association with T1D (Table 3).

Table 2 Association analyses in type 1 diabetes families and case-control sample sets for immune-mediated disease associated polymorphisms.
Full size table
Table 3 Association analyses in type 1 diabetes families and case-control sample sets for rheumatoid arthritis, inflammatory bowl disease and multiple sclerosis associated polymorphisms.
Full size table

As a potential atopy/asthma susceptibility locus, ADAM33 was considered as a candidate gene for T1D because atopic illness and T1D have been inversely associated [34]. No functional candidate SNPs in ADAM33 have been identified, but it is plausible that the SNPs in introns or the 3' region are located in regulatory sequences and thus may affect transcriptional efficiency or transcript stability [35]. We genotyped six SNPs that showed a P ≤ 0.03 in asthma case-control data [35] in set 1 families (Tables 1 and 4). One SNP, ADAM33 V4 (rs2787094), showed P = 0.0004 (RR = 0.78, 95% CI = 0.67–0.89) initially in 754 families, and P = 4.4 × 10-6 (RR = 0.77, 95% CI = 0.69–0.86) when genotyped in the additional set 2 families. However, we did not obtain additional support for association in 4,326 cases and 4,610 controls (P = 0.57) (Table 4).

Table 4 Association analyses in type 1 diabetes families and case-control sample sets for ADAM33 SNPs
Full size table

Finally, we genotyped the NFKB1 dinucleotide repeat (CA) microsatellite polymorphism, which has been associated with T1D [36], albeit inconsistently [37] (Table 1). In our set 1 families, we failed to find any evidence for an association with T1D (P = 0.68).

Discussion

Co-localization and overlapping of genetic loci in autoimmune diseases suggests that in some cases, common biological pathways may be involved in the aetiology of T1D and other clinically distinct immune-mediated diseases. In this study we examined 16 genes implicated in autoimmune and other immune-mediated diseases and report that none of the variants tested, consistently showed P values of less than 0.05 in association tests with T1D. Our data indicate that, although common immune-mediated disease loci are present in the genome, there are disease genes that are distinct to certain diseases. Indeed, the known T1D susceptibility loci follow this observation: while both CTLA4 and PTPN22 loci are associated with several autoimmune diseases [3, 38], the insulin VNTR locus is likely to be T1D-specific rather than a general autoimmune locus [7].

Conclusion

The possibility remains that some of the investigated genes and variants are associated with T1D, albeit with weak genetic effects, such as with ORs of less than 1.3, for which the sample size employed in our study does not provide sufficient statistical power. Even a case-control sample of 8,000 cases and 8,000 controls would only have 48% statistical power at a type 1 error rate α of 0.001 for a disease variant with a MAF of 0.10 and OR of 1.15. As genotyping costs decrease, it will be necessary to test the variants in larger sample sets than reported here, because the identification of genetic effects with ORs in the 1.15–1.25 range can be instrumental in the understanding of the disease process of T1D [3]. For example, we note that an observed genetic effect of OR 1.15, such as that exerted by the T1D susceptibility locus CTLA4, does not reflect the importance of the biological effect contributed by the locus through the protein(s) it encodes and the pathways it regulates [3].

Lack of association with variants tested here may also be partly due to false positive results obtained in the reported primary disease association studies. As shown in Table 1, these generally employed small sample sets, which is a common explanation in the reporting of P values that cannot be replicated in independent studies using larger sample sizes [39–42]. Nevertheless, for ORs >1.5 our present study had good statistical power even for alleles at 0.01 frequency.