Bmc Medical Genetics Two-stage Case-control Association Study of Dopamine-related Genes and Migraine

Background: We previously reported risk haplotypes for two genes related with serotonin and dopamine metabolism: MAOA in migraine without aura and DDC in migraine with aura. Herein we investigate the contribution to migraine susceptibility of eight additional genes involved in dopamine neurotransmission.


Background
Migraine is a highly prevalent neurological disorder involving multiple susceptibility genes and environmental factors [1,2]. The current clinical classification follows the International Criteria for Headache Disorders (ICHD-II), with the two main categories of migraine without aura (MO) and migraine with aura (MA) [3]. The pathophysiology of migraine is not entirely understood, but a role for dopamine (DA) was already suggested thirty years ago [4]. The DA hypothesis relies on the observed signs of central DA hypersensitivity in migraine patients and the known capacity of DA receptors to regulate nociception, vascular tone and autonomic responses [5]. Studies in animal models revealed that DA receptors are present in the trigeminovascular pathway and showed that DA can act as an inhibitor of nociceptive trigeminovascular transmission in the rat brain [6]. Along this line, DA antagonists have proved useful in aborting migraine headache or associated symptoms [5]. However, DA antagonists are not always selective and may act through DA receptorindependent mechanisms [7]. Also, a review of pharmacological and therapeutic studies in migraine could not provide convincing evidence of a direct role of DA in migraine pathogenesis [8].
Several association studies in different populations have focused on genes encoding proteins of the dopaminergic neurotransmission system, including DA receptors, the DA transporter, and enzymes involved in the synthesis and catabolism of DA. These studies provided conflicting results [7], although a recent, most comprehensive analysis of 10 dopamine-related genes in MA suggested that DBH and SLC6A3, at least, might be involved in migraine pathogenesis [9].
In a previous study that evaluated the contribution of 19 serotonin-related genes to migraine susceptibility in our cohort of Spanish migraineurs, we reported risk haplotypes in MAOA for migraine without aura and in DDC for migraine with aura [10], both genes being key players in the serotonin and dopamine metabolic pathways. In order to further elucidate the involvement of the dopaminergic system in migraine liability, nine dopamine-related genes were selected for a two-stage casecontrol association study in the Spanish population. showing monogenic inheritance, were excluded. The control samples consisted of Caucasian Spanish unrelated adult subjects (blood donors, individuals that underwent surgery unrelated to migraine or unaffected partners of migraine patients) that were matched for sex with patients and recruited in the same geographic areas (HUVH and FPGMX). Migraine and positive family history of migraine or any type of severe or recurrent headache in first -degree relatives were excluded in all control subjects through personal interview. Genomic DNA was isolated from peripheral leukocytes or saliva. Research was approved by the local Ethics Committees and all the adult participants, and the children or their parents gave their informed consent prior to participate in the study, according to the Helsinki Declaration.

SNP selection, SNPlex design, genotyping and quality control
We selected SNPs in nine candidate genes involved in dopaminergic neurotransmission. These genes encode five DA receptors (DRD1, DRD2, DRD3, DRD4 and DRD5), three enzymes involved in DA degradation (COMT and DBH) or synthesis (TH), and the DA transporter (SLC6A3 or DAT1; [see Additional File 1]).
In order to ensure a full genetic coverage of these genes and to minimize redundancy, we used the LD-select software [11] and the HapMap database (http://www.hap map.org; release 20) [12] to evaluate the LD pattern of the region spanning each candidate gene plus three to five kb flanking regions. TagSNPs were selected at an r 2 threshold of 0.85 and minimal allele frequency (MAF)>0.15 for genes with less than 15 tagSNPs and MAF>0.25 for those genes with 15 or more tagSNPs (COMT, DBH and SLC6A3). A total of 69 tagSNPs (26 in multi-loci bins and 43 singletons) were chosen [see Additional File 1]. Of these, five did not pass through the SNPlex design pipeline. After genotyping, MAF were determined in our control population 1 [see Additional File 2]. To ensure that no population stratification was present in the sample, 45 anonymous unlinked SNPs located at least 100 kb distant from known genes were also analyzed [13] by means of STRUCTURE [14], FSTAT [15] and the method by Pritchard and Rosenberg [16], as previously described [17].
Genotyping was performed at the Barcelona node of the National Genotyping Center http://www.cegen.org using the SNPlex technology [18]. Two CEPH DNA samples were included in the different genotyping assays, and a concordance rate of 100% with HapMap data was obtained. The allelic variants of the SNPs under study were named on the coding strand of each gene.

Statistical analyses
The minimal statistical power, calculated post hoc in population 1 using the Genetic Power Calculator software [19], assuming a disease prevalence of 0.12, an odds ratio (OR) of 1.7, a significance level (α) of 0.05 and a MAF of 0.123, the lowest in control population, was 85% and decreased to 74% for MO and 68% for MA. Given these estimates, we decided to begin our study by performing a joint analysis of the MO and MA groups, and to proceed to separate analyses of clinical subgroups only if a positive association was obtained in the whole sample.
Individuals with <40% successful genotypes were excluded from the analysis; SNPs with >10% missing genotypes were considered as failed; SNPs at r 2 > 0.85 from any other studied SNP or showing deviation from Hardy-Weinberg equilibrium (HWE; threshold set at 0.01) as calculated in our control population 1 were also excluded. The SNPs analyzed in the follow-up population were also in HWE.

Single-marker analysis
The analysis of HWE as well as case-control comparisons of both allele and genotype frequencies under a codominant model were performed with the SNPassoc R library [20] initially in population 1, adjusting by sex. When a nominal association was identified (P < 0.05), dominant and recessive models were also analyzed. The significance threshold under the Bonferroni correction for multiple testing was set at P < 5e-04 upon consideration of 50 SNPs analyzed, genotype and allele comparisons and a single clinical group. Under a False Discovery Rate (FDR) of 10% the threshold was set at P < 0.0035, using the qvalue R library [21].

Multiple-marker analysis
Risk haplotypes were assessed in the whole migraine group (MO + MA) with the UNPHASED software [22], only for genes showing association in the single-marker analysis after the Bonferroni or FDR corrections. The best up to five-marker haplotype was selected as previously described [17]. Significance was estimated by a 10,000 permutation procedure with UNPHASED [22]. The specific assignment of haplotypes to individuals was performed independently in cases and controls with the PHASE 2.0 software [23]. The comparisons of risk haplotype carriers vs non carriers were performed using the SNPassoc R library [20] adjusting by sex. Subsequently, risk haplotypes originally identified in the all-migraine group and SNPs trespassing the FDR threshold were tested in the MO and MA subgroups.

Follow-up replication study
Risk haplotypes identified in population 1 were tested in the replication population 2. For SNPs showing nominal association with migraine in population 1 a comparison of genotype and allele frequencies was undertaken in population 2.

Results
Initially, 64 SNPs from nine candidate genes encoding proteins related with DA neurotransmission were genotyped [see Additional File 1]. Fourteen SNPs were excluded from statistical analysis after data depuration in the first population: ten had genotype call rates <90%, one was monomorphic and three were in strong LD with other SNPs (r 2 > 0.85, [see Additional File 1]). The 50 remaining SNPs had MAF>0.12 and were in HWE in control population 1 (P > 0.01; [see Additional File 2]). After the exclusion of individuals with low genotyping rate, population 1 consisted of 263 patients and 274 controls, and population 2 was composed of 259 patients and 287 controls. No evidence of population stratification was found in any of the two populations studied by applying the STRUCTURE software [see Additional File 3], the Fst coefficient (theta = 0, 99%CI = 0.000-0.001 for population 1 and theta = 0, 99%CI = 0.000-0.002 for population 2) and the method by Pritchard and Rosenberg (P = 0.57 for population 1 and P = 0.05 for population 2).
In the single-marker analysis, genotype and allele frequencies were compared between patients and controls in population 1 [see Additional File 2]. Six SNPs within five genes (DRD1, DRD2, DRD3, DBH and TH) displayed Pvalues < 0.05 (table 1). Two of them, rs2283265 in DRD2 and rs2070762 in TH, remained significant after applying a FDR of 10% (Table 1A) and were further considered for the multiple-marker analysis. No SNP withstood the restrictive Bonferroni correction for multiple testing. For these two SNPs we sought to detect a specific association with either one of the clinical subtypes, MO or MA. We found that in population 1, DRD2 rs2283265 was associated wit both MO and MA, while TH rs2070762 was only associated with MO (Table 1B).
Multiple-marker analysis of the two SNPs (rs6356 and rs2070762) in TH showed a different overall distribution between cases and controls (best adjusted P-value = 0.015, Table 2), due to an over-representation the A-C allelic combination in cases (P = 0.037, OR = 1.34, 95%CI = 0.94-1.90), and G-T in controls (P = 0.00573, OR = 1.47, 95%IC = 1.11-1.94; Table 3). However, individual haplotype assignation did not identify differences in the frequency of risk haplotype carriers between cases and controls. Moreover, the analysis of rs6356-rs2070762 haplotype distributions in cases and controls of population 2 found no evidence of association with migraine (table 3).
Finally, we aimed to determine whether those variants nominally associated with the disease phenotype in population 1 after the single-marker analysis, could be replicated in population 2, especially rs2283265 in the DRD2 gene and rs2070762 in the TH gene, which maintained significance after FDR correction in the initial analysis. The comparison of genotype and allele frequencies between cases -either the whole group of migraineurs, or MO or MA subgroups-and controls did not reveal significant differences for rs2283265 (codominant genotypes P = 0.62 and alleles P = 0.36), rs2070762 (codominant genotypes P = 0.44 and alleles P = 0.22) nor for any other SNP (table 1).

Discussion
We performed a two-stage case-control association study of eight dopamine-related genes in the Spanish population. In order to capture the common haplotype variation of these genes in the European population, we selected haplotype-tagging SNPs which covered each gene and its flanking regions. In population 1, a five-marker risk haplotype in the DRD2 gene and a single variant in the TH gene were found to be associated with migraine, and both remained significant after applying correction for multiple comparisons. In the initial single-marker analysis, pointing at five genes including the two above, no SNP withstood the Bonferroni correction. However, it is well known that this correction is often over-conservative as it assumes independence of all the tests performed, whereas many SNPs within the genes studied, although not in strong LD, are not independent. When markers found associated in population 1 were analyzed in the follow-up population, the results could not be replicated. As special attention was paid to rule out the existence of stratification and both populations were comparable in terms of size, gender distribution, ethnicity (Caucasians), geographical origin (Spain) and diagnostic criteria, failure to replicate the results suggests that the associations identified in population 1 may be spurious and that the genes analyzed here would not be involved in migraine susceptibility. However, these findings should be taken with caution, as the genetic coverage of some of the studied genes is not optimal for several reasons: First, SNPs with low frequencies, which would require very large sample sizes to produce significant results, were not selected. Second, some SNPs within the studied genes, for which no LD data were available in the HapMap database, were not included. And third, SNPlex design constraints and low genotyping call rates of some specific SNPs forced additional exclusions that left the DRD4 gene out of the study. Of note, the same Spanish cohort analyzed in the present work was previously scrutinized by us to detect association of MA or MO with genes related with serotonin neurotransmission [10]. Among the three genes that displayed significant association, two belong to the dopamine metabolic pathway: MAOA, found to be associated with MO, and DDC, which was associated with MA. However, these findings still await replication.
A number of association studies have focused on dopamine-related genes. The first susceptibility polymorphism identified in this system was the NcoI variant in the DRD2 gene (rs6275), with an over-representation of the C allele in MA [24]. Subsequent studies failed to replicate this association [25][26][27][28] or that with other DRD2 polymorphisms [27,29]. It is worth mentioning, however, that a (TG)n repeat variant in DRD2 was found associated with yawning and nausea in a small subgroup of migraine patients [30]. We analyzed DRD2 rs2242592, in strong LD with rs6275, that belonged to a risk haplotype identified in population 1 but not confirmed in population 2. Subsequent studies found association between migraine phenotypes and polymorphisms in DRD4 [31,32] and DBH [9,25,33,34], although negative associations have also been described [9,30,31,33]. No associations have been identified in any of the polymorphisms analyzed in genes DRD1, DRD3, DRD5 or COMT [9,27,30,[35][36][37][38]. The genetic marker set selected in the present analysis is in many respects not comparable with the polymorphisms analyzed in the previous studies. However, our results agree with previous negative findings in DRD1, DRD3 and one polymorphism in DBH.
A recent study carried in two German populations [9], analyzed the contribution of the nine dopamine-related genes we have examined, plus DDC, to MA susceptibility. In that study, MA was associated with three SNPs, SLC6A3 rs40184, DRD2 rs7131056 and DBH rs2097629. Overall, they analyzed 43 SNPs belonging to the nine genes stud-    with the German study, while rs2097629 in DBH was not included in our study because of design constraints. In addition to differences in the respective SNP sets, our samples were composed of both MO and MA patients, and therefore a comparison of our results with those of Todt et al. is not altogether straightforward. Also, our analytical design set that the two population samples could only be grouped for analysis in case nominal associations were found in both populations 1 and 2, while in the German study their two samples were analyzed as a single group for all SNPs within the three genes that showed nominal association in only one population. This strategy produced significant associations despite lack of replication in their follow-up sample. Future studies combining both marker sets might help to reconcile these apparently discordant findings.
Much evidence points to dopamine hypersensitivity in migraineurs, particularly those displaying the premonitory symptoms of yawning or nausea. In our study, such specific symptoms could not be analyzed, since they were not available in the whole sample. To our knowledge, no well-powered association study has addressed the relationship between endophenotypes based on dopaminergic symptoms and genetic susceptibility using a pathwaybased approach. Alternatively, latent class analysis of migraine symptoms, as used to enhance clinical homogeneity in genetic linkage analysis [39,40], might define migraine phenotypes, not necessarily related to ICHD-II migraine subtype diagnoses, and thus uncover specific genetic susceptibility factors.

Conclusion
In summary, our results do not support the involvement of a set of dopamine-related genes in the genetic vulnerability to migraine in the Spanish population, albeit a previous association study in the same cohort identified DDC and MAOA as potential susceptibility genes [10]. Further studies in larger samples or family-based sets may help to clarify the contribution of dopamine-related genes to migraine genetic background.