Lack of Association of PAX4 Gene With Type 1 Diabetes in the Finnish and Hungarian Populations

All patients with type 1 diabetes were diagnosed at <15 years of age according to the World Health Organization criteria.

Children with type 1 diabetes were derived mainly from the Departments of Pediatrics at the universities in Turku, Helsinki (southern Finland), and Oulu (northern Finland) (n = 309, 50.5% males; mean age at diagnosis 8.4 ± 4.2 years). The control group comprised consecutive healthy infants born in the University Hospitals of Turku and Oulu who were randomly selected for this study (n = 535, 47.5% males).

Patients were consecutively diagnosed in Baranya County (southern Hungary) over a 17-year period from 1980 to 1996 and recruited through the Hungarian Childhood Diabetes Registry (n = 189, 49.7% males; mean age at diagnosis 8.5 ± 4.8 years). Healthy schoolchildren born in the same geographical region and matched for sex and age were used as control subjects (n = 290, 50.7% males; mean age at sampling 10.6 ± 2.7 years).

Four single nucleotide polymorphisms (SNPs) in the PAX4 gene (OMIM 167413, 7q32) were analyzed using DNA sequencing (MegaBace1000 DNA Sequencer; GE Healthcare). Three nonsynonymous SNPs were located in exon 9 [rs712701 (P321H), rs2233583 (V330A), and rs2233584 (Ter333W)], whereas the rs2233578 SNP (R133W) was in exon 3. HLA typing was performed in all Finnish subjects using methods described earlier (12,13). The INS −23 HphI SNP was typed with a similar oligonucleotide hybridization format (T probe 5′-Europium-CTGTCTCCCAGA-3′; A probe 5′-Terbium-CTGTCACCCAGA-3′). The CTLA4 CT60 SNP was genotyped using minisequencing (SnuPe kit; GE Healthcare). Primer sequences are available on request.

For comparisons of genotype frequencies between the groups, χ² statistics with Yates’ correction were used. P values <0.05 were considered significant. The effect of PAX4 and the other studied genes was also examined by logistic regression analysis (SPSS for Windows, version 11.0.1; SPSS, Chicago, IL). Power calculations were done using standard methods.

The R133W variant in exon 3 (rs2233578) was analyzed in 46 healthy newborn infants and in 124 patients. One Finnish patient carried the arginine/tryptophan heterozygous genotype; all other subjects were arginine/arginine homozygous. At the V330A polymorphism in exon 9 (rs2233583), one child with type 1 diabetes from Finland carried the heterozygote genotype, whereas all other individuals (n = 1,322) were alanine/alanine homozygous. In addition, no variation was detected at the rs2233584 SNP (Ter333W) among the 1,323 study subjects.

The C/A variant at coding sequence position +1,168 is a nonsynonymous SNP (P321H; rs712701) that was found polymorphic. No differences in the genotype frequencies between cases and control subjects were observed in the Finnish and Hungarian series (Table 1). Accordingly, the comparison on the combined datasets showed no difference between patients and control subjects. The genotype distributions were in Hardy-Weinberg equilibrium both among cases and control subjects in the two series (P = 0.82 and 0.69, respectively, for the Finnish set; P = 0.50 and 0.96, respectively, for the Hungarian set). We stratified the Finnish patients according to age at diagnosis and sex and compared the PAX4 genotype frequencies with control subjects, but no differences were seen (Table 1). Moreover, to elucidate gene-gene interactions, we analyzed the distribution of the PAX4 genotypes in patients stratified according to the HLA DRB1-DQA1-DQB1, insulin gene (−23) HphI, and CTLA4 CT60 genotypes and observed no differences (Table 1). The lack of association of PAX4 with type 1 diabetes was confirmed also by logistic regression analysis.

We compared the genotype and allele frequencies (rs712701 SNP) in the Finnish newborn series between northern and southern Finland, and no differences were detected (data not shown). Similarly, no regional differences were either observed in the genotype frequencies among the Finnish patients (data not shown, P = 0.69, df = 4). In contrast, the +1,168 genotype distributions showed a difference between the Finnish and Hungarian control series (P = 0.002) since the A allele was more frequent in the Finnish population (27.0 vs. 19.5% in Hungarians, P = 0.0008). As expected, genotype distributions were also different between the two populations among the patients affected by type 1 diabetes (P = 0.007).

The identification of novel type 1 diabetes genes is challenging since the disease is polygenic and the current mapping methods, such as linkage analysis, seem to have very low sensitivity, and association studies are extremely sensitive to population stratification biases. The situation is further complicated by linkage disequilibrium heterogeneity that exists between different ethnic groups (14). In addition, it is likely that there are several alternative disease pathways, and the effect of certain genetic variants is likely to be restricted to a particular molecular mechanism. Our previous studies suggest that the INS locus is associated with the emergence of insulin autoantibodies and could play a role in the development of insulin autoimmunity (15,16). In contrast, INS was not associated with multiple autoantibody positivity in children without insulin autoantibodies (16). As a consequence, new strategies for mapping of genes predisposing to type 1 diabetes are required using such cohorts, where the patients are well characterized for molecular markers such as β-cell–specific autoantibodies.

In this study, we attempted to reproduce the findings of Biason-Lauber et al. (10) on the association of the gene encoding the PAX4 transcription factor and type 1 diabetes. However, we were unable to confirm a role for the +1,168 polymorphism of PAX4 in disease pathogenesis in the two populations studied with profoundly different socioeconomic conditions, climate, and environment. Similarly, no effect of PAX4 could be seen in any patient subgroups analyzed according to age at diagnosis, sex, HLA, and CTLA4 CT60 genotypes. Since PAX4 is a pancreatic transcriptional regulator, we hypothesized that its effect could be restricted to individuals with specific INS genotypes. We found, however, no evidence for that.

The Finnish population has the highest disease incidence in the world (55.3 · 100,000⁻¹ · year⁻¹ in 2004; A. Reunanen, personal communication), whereas the Hungarian population is among the low incidence countries in Europe (8.9 · 100,000⁻¹ · year⁻¹) (17). In addition to this difference, the two populations also show divergence in genetic background, socioeconomic characteristics, and environment. Although the allele frequencies were somewhat different in the two populations, the disease association was consistently lacking in both series. It is unlikely that the findings in our study are explained by interpopulation differences. The statistical power of our study to detect a difference of the magnitude reported by Biason-Lauber et al. (10) at α = 0.05 was more than 99%, and it was more than 80% to detect a significant OR of 1.2. To explore the striking difference between our data and those reported by Biason-Lauber et al. (10), we applied the classical χ² goodness-of-fit test for Hardy-Weinberg equilibrium (HWE) on the PAX4+1,168 genotype data from the Swiss and German population. We found that the +1,168 genotype frequencies showed severe departure from the expected values, both among the cases and control subjects (χ² = 20.2, P = 0.00004; χ² = 73.9, P < 0.000001, respectively). Deviation from HWE might be seen in multiple SNPs in a strong linkage disequilibrium block, and it may also be caused by factors such as recent population migration or genetic drift. However, the most common reason for this phenomenon is genotyping error, as shown by the study of Xu et al. (18), who found that as much as 12% of the genotype data on 133 SNPs published in 75 studies were inconsistent with HWE in control subjects. Our comparison of the Swiss and German data on the +1,168 PAX4 genotype frequencies also showed that particularly the distribution of the C/A heterozygous genotype was affected by this bias, being 10.8% lower in cases and 16.2% higher among control subjects compared with the expected values. In large SNP genotyping efforts, indicators of assay quality other than HWE such as genotype call rate or Mendelian error rate may also help to reduce false-positive findings.

Holm et al. (10) observed in the 7q32 region an increase in the logarithm of odds from 2.2 to 5.3 in a pathway restricted linkage analysis using 408 Scandinavian multiplex families. They analyzed three intronic SNPs in the PAX4 gene and observed marginally significant P values in the haplotype transmission analysis. It has to be noted that these do not remain significant after correction for multiple comparisons. The T-C-C PAX4 haplotype, however, showed a P value of 0.001 in the transmission distortion test, but it was extremely rare.

In summary, we suggest that the PAX4+1,168 C/A variant does not play any essential role in genetic susceptibility to type 1 diabetes. The strong coherence between our datasets on two ethnic groups with conspicuously different disease incidences, socioeconomic characteristics, and profoundly diverse environment strengthens this finding. The striking observations made on the German and Swiss population cohorts require replication and a more detailed analysis of this gene to assess whether the difference between the two studies could be explained by heterogeneity in linkage disequilibrium patterns between ethnic groups or might reflect profound variation in the effect of PAX4 on type 1 diabetes susceptibility in particular patient groups.

This study was supported by the Juvenile Diabetes Research Foundation (JDRF grants 4-1998-274 and 4-1999-731); JDRF/European Foundation for the Study of Diabetes/Novo Nordisk Focused Research Grant for Type 1 Diabetes; the Novo Nordisk Foundation; Special Federal Grants to Turku, Oulu, and Tampere University Hospitals; the Academy of Finland; and the Hungarian Scientific Research Fund (OTKA) grant T046923.

We thank Eija Nirhamo, Piia Nurmi, Mia Karlson, Ritva Suominen, and Terhi Laakso for their skillful technical assistance.