Families were included in the study if at least two siblings had type 1 diabetes. Both affected and unaffected siblings were invited to participate, as well as their parents. The study was approved by the centers’ ethics committees and participants signed a written, informed consent form. Clinical information was collected by means of standardised questionnaires delivered at each of the participating centres.9 Blood samples were obtained and HLA was genotyped centrally in Malmo¿, Sweden, using a PCR-based, sequence-specific oligonucleotide probe system and alleles were read with specific software (SCORE).10

The HLA haplotypes were constructed using the R package alleHap, a deterministic algorithm for imputing missing genetic data and reconstructing haplotypes from pedigree databases.11 Risk and protective class II HLA haplotypes were defined, based on previous T1DGC results in Caucasian populations.10 Briefly, haplotypes DRB1*0405-DQA1*0301-DQB1*0302, DRB1*0401-DQA1*0301-DQB1*0302, DRB1*0301-DQA1*0501-DQB1*0201, DRB1*0402-DQA1*0301-DQB1*0302 and DRB1*0404-DQA1*0301-DQB1*0302 were considered high-risk, whereas DRB1*0701-DQA1*0201-DQB1*0303, DRB1*1401-DQA1*0101-DQB1*0503, DRB1*1501-DQA1*0102-DQB1*0602, DRB1*1101-DQA1*0501-DQB1*0301, DRB1*1104-DQA1*0501-DQB1*0301, DRB1*1303-DQA1*0501-DQB1*0301, DRB1*1301-DQA1*0103-DQB1*0603 and DRB1*0403-DQA1*0301-DQB1*0302 were considered protective. All haplotypes not included in the mentioned categories were defined as “other”.

In the case of incomplete haplotype reconstruction, if only one option of complete haplotype was possible according to the rest of the database, then the appropriate, complete haplotype would be inferred and classified. For example, in the case of the incomplete DRB1*0301-DQA1*?-DQB1*0201, the high-risk DRB1*0301-DQA1*0501-DQB1*0201 haplotype would be inferred. If several alternatives were possible, albeit in the same category, then the haplotype was also classified. For example, DRB1*1303-DQA1*?-DQB1*0402 could be translated into different possibilities, although none of them classified as high-risk or protective. Thus, it would be classified as “other”. Finally, if several alternatives were possible, in different categories, the haplotype would be considered “lost” and would not be included in the analysis. For example, DRB1*?-DQA1*0301-DQB1*0302, could be both high-risk or protective and would thus be declared missing.

The distribution of protective, high-risk and other haplotypes was compared in the first two affected siblings per family and in the unaffected parents in the Canarian and non-Canarian Spanish participants, using chi-squared. Since the majority of the non-Canarian families were recruited in Catalonia, a separate analysis was also performed, comparing the Canarian and Catalonian families.

Fully or partially complete (one allele missing), unambiguous DRB1-DQA1-DQB1 HLA haplotypes were obtained in 144 Canarian (74 siblings with type 1 diabetes and 70 non-diabetic parents) and 301 non-Canarian subjects (162 siblings with type 1 diabetes and 139 non-diabetic parents). This led to 456 complete unambiguous and 16 partially complete haplotypes among the affected siblings and 411 and 7, respectively, among the unaffected parents, that could be classified according to their risk. There were no significant differences between regions in the distribution of the haplotypes into their risk categories for the siblings or for the non-affected parents (see Table 1). Table 2 displays the frequency of selected, specific, complete, unambiguous haplotypes in the affected siblings and in the parents. The comparison between the Canarian and Catalonian families did not show any significant differences, either (data not shown).

According to this family-based study, the high incidence of childhood-onset type 1 diabetes in the Canarian population does not seem to be explained by higher-risk class II HLA haplotypes.

The strengths of this study include its family-based nature and the use of standardised methods for data and sample collection and processing, which allowed for haplotype imputation and direct comparison.

Previous studies performed in Spain show a high frequency of high-risk HLA haplotypes in patients with type 1 diabetes, comparable to other populations. A higher frequency and higher risk associated with DR3-DQ2 (compared with DR4-DQ8) has been found in mainland Spain in previous reports.12–14 Furthermore, populations-specific, extended DR3 HLA haplotypes have been identified.15 In the present study, DR3 was the most frequent class II HLA in peninsular Spain, but not in the Canary Islands. In a recent study performed in Gran Canaria in children with type 1 diabetes, more than 95% of the participants showed at least one DQ2 or DQ8 allele (which are often assumed to be part of high-risk DR3-DQ2 and DR4-DQ8 haplotypes, respectively).3 In that study 40% of alleles were DQ2 and 36% were DQ8,3 which does not fully agree with the presently described results. Indeed, this family-based population, with at least two affected siblings per family, may not necessarily be representative of the more frequent, sporadic form of type 1 diabetes. Furthermore, the families included in the T1DGC need not be strictly representative of all families with type 1 diabetes. Finally, a type II error due to small sample size cannot be ruled out. In the present study, differences between groups ranged between 1.2 and 1.5% in the case of the affected siblings, and between 3.5 and 8% in the case of the non-affected parents. For the study to have an 80% power to detect a 1.5% difference, the sample should have included 8000 participants per group (8000 Canarian and 8000 non-Canarian) and, to detect a 3.5% difference, we would still need about 1600 individuals per group.

Although the Canarian population is mostly of Iberian-European descent,16 North African genetic markers are also present.17 In fact, studies performed in the North of Africa also identify DR3 and DR4 as the main high-risk alleles associated with type 1 diabetes.18,19 More extensive haplotyping might allow for the identification of population-specific haplotypes, as in other Spanish regions. On the other hand, different genotyping methods, with varying resolutions, limit the comparisons that can be made among studies. Furthermore, haplotype imputation is often not possible, especially in case-control studies, where parents might not be available for genotyping.

HLA accounts for most, albeit not all the genetic risk of type 1 diabetes. About 40 additional genetic loci have been associated with moderate disease susceptibility, although only variants in INS, PTPN22, CTLA4, and IL2RA are associated with OR above 1.1.6 Although they have not been assessed in the current analysis, they should be available for future studies.

Several environmental factors have also been associated with the risk of type 1 diabetes, including virus infections, early dietary patterns and vitamin D.6 These and/or other environmental or genetic factors should be sought that may account for the high incidence of childhood-onset type 1 diabetes in the Canary Islands.

In conclusion, despite a higher incidence of childhood-onset type 1 diabetes in the Canary Islands, our family-based results do not show an increased prevalence of high-risk HLA haplotypes when compared with mainland Spain.

This research uses resources provided by the Type 1 Diabetes Genetics Consortium (T1DGC), the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK, U01 DK062418), the National Institute of Allergy and Infectious Diseases (NIAID), the National Human Genome Research Institute (NHGRI), the National Institute of Child Health and Human Development (NICHD), the Juvenile Diabetes Research Foundation International (JDRF). In addition, the authors were supported by the European Foundation for the Study of Diabetes (Albert Renold Travel Grant- N Medina) and the Instituto de Salud Carlos III (PI11/02441).

The authors are not aware of any conflicts of interest regarding the contents of this manuscript.