In Study III, we employed a custom-made Perlegen genotyping assay covering 5,257 SNPs targeting 155 ion transport genes. These genes were selected from publicly available databases based on function as ion channels, following on the channelopathy hypothesis (see Figure 20). An initial scan of 841 unrelated migraine samples and 884 unrelated controls failed to detect any significant associations. However, given the relatively low power to detect common variants with this sample size, we selected all SNPs with an allelic p-value < 0.005 (n=66) for follow-up study. No SNPs showed significant association to migraine, and thus we conclude that common variants of moderate effect size in ion transport genes do not play a major role in susceptibility to common migraine within these European populations. Similarly, the three known FHM genes showed no association in the screen, which is in line with previous research by our group (Kaunisto et al., 2005).
3.a. Target selection
The “common variant – common disease” hypothesis proposes that common diseases rise from the genetic effects of variants of relatively high frequency in the general population, which explains why certain diseases are common. From an evolutionary point of view, these common variants would possess only minimal effect on reproductive capability, involving largely diseases appearing after fertility period (neurodegenerative diseases, metabolic syndrome etc.), and thus escape evolutionary pressures.
Before it was possible to use GWA data, which at the time was prohibitively expensive, educated guesses had to be made in the selection of target genes in order to study the role of common variants in ion channel genes. Based on the fact that FHM studies successfully identified causative genes that are all ion channels, we targeted all available ion channel genes (including the three FHM genes) based on the data-mining of the available biological databases (such as the Kyoto Encyclopedia of
Figure 20. The research plan for Study III. SNP – single nucleotide polymorphism.
Genes and Genomes [KEGG] at genome.jp/kegg) for known ion channel genes. An initial list of 768 partially overlapping genes from the various databases was eventually reduced to 155 based on the reliability of their detection and the quality of available data (see Table 11). Regions covering the genes with an additional 20 kb at the 5’ end and 10 kb at the 3’ end of each gene were included the study. TagSNPs were then selected to cover all common variants with MAF ≥ 10% in the study regions with an r2 ≥ 0.8. The resulting total was 5,269 SNPs. These markers were first studied in a Finnish sample of 841 unrelated migraine cases and 884 population-based controls with SNPs showing p-values <0.005 selected for replication in multinational samples from Germany, the Netherlands and Australia.
3.b. No association to common variants either with diagnosis or trait component analysis
In the initial Finnish study sample, no significant association to migraine was observed. The SNP with the lowest p-value was rs13276133 (p-value 0.00041);
however, after multiple testing correction through permutation, the empirical p-value was only 0.7713. A total of 66 SNPs in nine genes (ATP2C2, CACNA1E, CACNB2, KCNE2, KCNK12, KCNK2, KCNS3, SCN5A and SCN9A) were selected for replication.
In the replication analysis, none of the 66 SNPs chosen for replication showed consistent evidence of association between cohorts. Given that the estimated power to replicate an association at a p<0.05 level from the initial study was over 70% for each of the top SNPs, we can be relatively confident that the original associations were spurious in nature, especially as even the lowest replication p-value was relatively high (rs400922 on chromosome 16; empirical p-value 0.15). The use of TCA did not considerably improve results (highest TCA signal was seen with marker rs12996816 on chromosome 2, which had an allelic p-value of 0.001 using the MA diagnosis, and 6.32 x 10-5 using the pulsation trait; Bonferroni-corrected significance limit for the study was 1.9 x 10-5). For the trait component analysis part of the study, we used the traits photo- and phonophobia in addition to pulsation, as they were considered to be the best traits based on the 2006 study (Anttila et al., 2006). The top results for the other traits were with markers rs13276133 (chr 8q21.11; p-value 0.00021, photophobia) and rs12054449 (chr 3p21.1; p-value 0.00031; phonophobia).
Category Genes on array
Voltage-gated calcium channels 26
Voltage-gated potassium channels 74
Voltage-gated sodium channels 14
Chloride channels 21
ATPase ion transporters 20
Total 155
Table 11. Categories of ion channels and pumps represented in the array.
V. Anttila - Identification of genetic susceptibility loci for migraine
74
3.c. Possible signs of epistasis between ion channel genes
As many of the ion channels are hetero- or multidimers and thus act together in cells, a SNP x SNP epistasis analysis was conducted among the studied SNPs using the epistasis analysis option in PLINK (Purcell et al., 2007). Given the large number of SNPs available, we followed the recommendation by Blangero et al. (Blangero et al., 2000) to restrict the analysis to those SNPs that have been chosen for replication for showing nominal association (p < 0.005) to migraine. In this analysis, uncorrelated (r2
= 0.0009) SNPs in genes KCNB2 (rs1431656) and CACNB2 (rs7076100) were found to have an interaction p-value (pointwise p-value 1.99 x 10-5, 0.022 after correction) that remained significant after Bonferroni correction adjusted for independence of tests using SNPSpD (Nyholt, 2004). The interaction estimated from the SNP data would confer an OR of 1.64 for migraine. These genes are located on two different chromosomes (KCNB2 on 8q13 and CACNB2 on 10p12). However, it should be noted that the interaction was only found in the Finnish cohort. It almost replicated in the Australian cohort (p=0.057), but no interaction was present in the Leiden cohort (both SNPs were successfully genotyped in these three populations only). The CACNB2 SNP is unfortunately not included in the GWA platform used in Study IV, and so we could not replicate the interaction in the larger study. No previously reported connection was found between the two by a PubMed search or in any of the relevant databases (such as the Kyoto Encyclopedia of Genes and Genomes).
3.d. Conclusions
The results of this study were fairly straightforwardly negative, both with the diagnosis-based approach and the TCA approach. Given that the most likely reason for this, not tagging enough of the variation surrounding these genes, was mostly excluded as a cause in Study IV, and therefore we concluded that common variants of moderate to high effect sizes in these genes do not play a role in migraine
susceptibility. Untagged rare variants may of course still play a role in these genes, and given the known FHM pathology, this may even be likely. However, the answer to the role of rare variants in ion channel genes can only be sufficiently answered through resequencing.