Common variants in the regulative regions of GRIA1 and GRIA3 receptor genes are associated with migraine susceptibility

Two hundred fifty outpatients consecutively observed from January 2006 to December 2008 at the Service of Diagnosis and Therapy of Headache of the Second University, Naples and two hundred sixty healthy volunteers consecutively enrolled during the same time interval were included in the study. Cases and controls were matched for sex, came from the same small geographical area and were Caucasian. All gave informed consent to the research procedures including molecular genetic analyses. The study protocol included history, direct clinical and neurological examination, MRI with both arterial and venous angiography sequences; venous blood collection for routine blood analysis as well as DNA extraction from lymphocytes according to standard salting out procedure [9]. Migraineurs were diagnosed as having either migraine with (MA) or migraine without (MO) aura, according to the 2004 International Headache Society [10] diagnostic criteria. The study population was comprised of 244 migraineurs (M/F 56/188; age 14-64 years; MA/MO 135/109); 7 patients who resulted affected with Familial Hemiplegic Migraine were excluded from the study.

For the statistical analysis three experimental groups were considered: all migraineurs together (MA+MO= M) and MA and MO separately. Healthy control individuals (204 females and 56 males aged 21-64 years) of Caucasian origin matching for age and sex the migraineurs served as control for all three experimental groups.

SNP selection was performed inspecting the HapMap database PhaseIII/Rel 2, Feb09 on NCBI B36 assembly, dbSNP b126 http://​www.​hapmap.​org/​index.​html.​en. Default parameters in Haploview were used to create haplotype blocks (95% confidence intervals) as described by Gabriel et al (2002) [11].

TagSNPs were selected at an r ² threshold of 0.80 from all SNPs with minor allele frequency (MAF) > 0.15 allowing us to capture more than 80% of the haplotypic diversity of these genes.

To detect association between the tested markers and migraine, we performed chi-square (χ²) analysis to test for significant differences in allele and genotype frequencies in case versus control results. χ² provides the likelihood of a deviation in the distribution of the same attributes in different classes (e.g. allelic frequencies in controls versus affected subjects). If the probability (p-value) of an equal distribution between the two groups is below a determined significance level α (in percent), the statistical output will show enough significance to assume LD and therefore association. The p value of this test is asymptotically equal to the p value obtained in Armitage's trend test (ATT). ATT takes into account genotypes rather than alleles, avoiding a possible bias caused by doubling the sample size. It assumes an additive (or co-dominant) disease model where all disease alleles have equal and independent contribution to disease risk. The ATT is valid and powerful under a broad range of disease mechanisms. Second, allele and genotype frequencies, dominant and recessive genetic models as well as odds ratios were calculated to characterise the distribution of a distinct genotypes in different phenotypic subgroups of the population (using FINETTI, http://​ihg.​gsf.​de/​cgi-bin/​hw/​hwa1.​pl). Hardy-Weinberg Equilibrium (HWE) was also calculated using FINETTI. The analysis of the statistical power was performed using the Genetic Power Calculator software http://​pngu.​mgh.​harvard.​edu/​~purcell/​gpc/​cc2.​html[14] assuming a significance level (α) of 0.05, a lifetime risk of 2, and the MAF of each SNP as calculated in our control group. In the population-based association analysis, the nominal significance threshold was set at P < 0.05 and lowered to P < 0.0017 after the multiple comparison correction of Bonferroni considering 22 SNPs, the three clinical groups analyzed, genotypes-alleles and females-males. Confidence intervals at 95% are provided for all odds-ratio (OR) values. Haplotype frequencies were calculated using the Estimation Haplotype (EH) program (Jurg Ott, Rockefeller University). A P level of less than 0.05 was considered statistically significant.

To determine whether the c.-1952T>C SNP in GRIA3 promoter region alters nuclear protein-DNA interactions, EMSAs was performed using radio-labelled oligonucleotide probes containing either -1952T or -1952C. Nuclear extracts were prepared from HEK 293 and SH5Y5Y cell lines as described by Granelli-Piperno et al. [15]. Binding reactions mixture contained 10 ng 32P-labeled oligonucleotide probe (10⁵ cpm), 15 μg nuclear extract, 2 μg poly (dI-dC), 100 mM KCl, 10 mM MgCl2, 20 mM HEPES (pH 7.9), 0.2 mM EDTA, 0.5 mM dithiothreitol, 20% glycerol and protease inhibitor, in a total vol of 20 μl. To perform super shift experiments, 2 μg of the anti-HSF1 antibody (Celbio) was added to the binding reaction mixture. In competition experiments, 100-fold molar excess of cold oligonucleotides was added to the binding reaction mixture. Reaction mixtures were incubated for 30 min at room temperature, resolved on non-denaturing 5% polyacrylamide gel, dried, and exposed to autoradiography. The same assay was performed for the c.-2012C>T SNP rs548294 in GRIA1 promoter region and for c.+561G>A SNP rs2195450 in intron 1.

To test the functional activity of the T>C polymorphism in the promoter region of the GRIA3 gene, we first identified the putative start site at -823 to the ATG, using the Promoter scan software vs2. A fragment of 1854 bp of the GRIA3 promoter encompassing base pairs -1411 to +325 (referred to the predicted start site) and including the T>C polymorphism was amplified from genomic DNA of migraine patients with genotype CC and controls with genotype TT. The primers used for generating this fragment were sense: 5'-GCGCGCCTCGAGTTTGAAGATGAGAGAACTGG-3' and antisense: 5'-GCGCGCAAGCTTCAAAAGAAAGAGAACGAAAG-3' which contain 5'-XhoI and 3'-HindIII restriction sites (highlighted bold sequences), respectively.

PCR products were double digested using XhoI and HindIII and then legated into a linearized pGL3-Basic vector (Promega, Madison USA) containing a Firefly Luciferase reporter gene. The inserts of the constructs were verified by sequencing before transient transfection.

Transfections were carried out using HEK293T cells. Cells were co-transfected with 1 μg of reporter plasmid and 10 ng of pRL-TK control vector (Promega, Madison USA) containing Renilla luciferase, which was used to correct for transfection efficiency. Transfection cells were harvested after 24 h and cell lysates were assayed sequentially for Firefly and Renilla luciferase activity using a dual-reporter assay system (Promega, Madison USA). Transfections were performed in quadruplicate for each plasmid construct. Changes in gene transcription were calculated relative to the corrected luciferase activity of the empty pGL3-Basic vector (control).

This research was reviewed and approved by the University of Naples Human Research Ethics Committee and all subjects participating in the study gave informed consent.

We selected SNPs for each gene (GRIA1-GRIA4) considering their distribution in LD blocks (figure 1). All variants were genotyped in a well-characterized panel of 244 cases and 260 controls (see methods). In controls, all SNPs analyzed are in Hardy-Weinberg equilibrium (HWE) (with the exception of rs2195450) and showed allele frequencies comparable to those observed in the Caucasian population and reported in the HapMap or SNP browser databases.

No significant association with the disease for any of the four SNPs in GRIA2 and GRIA4 genes tested was observed.

Significant association was observed for two SNPs located in the first block of LD of GRIA1 gene including the regulative region and the first two exons. Allelic and genotype frequencies distribution of the promoter -2012C/T (rs548294) marker in cases and controls showed significant association with migraine (p = 0.00009 and p = 0.0005, respectively). Stratified analyses of migraine subtypes were also undertaken indicating association primarily attributed to MO subtype group (P = 0.0003) and particularly in females (P = 0.0007) (Table 2). These differences remained statistically significant when Bonferroni correction for multiple comparisons was applied (Migraine all: alleles p = 0.002; genotypes p = 0.01; MO alleles: p = 0.008; genotypes p = 0.03; Females all: alleles p = 0.02; genotypes p = 0.05) (Table 2).

The analysis of the +561 G>A SNP (rs2195450) also showed significant association in both allelic and genotype distributions (p = 0.0002 and p = 0.0007, respectively) (p = 0.005 and p = 0.02 after Bonferroni corrections). Stratified analyses of migraine subtypes showed a specific association with MA subtype group for both allelic and genotypic frequencies (p = 0.00002 and p = 0.00008) (p = 0.0005 and p = 0.002 after Bonferroni corrections) (Table 2). However, the SNP rs2195450, was the only SNP found in Hardy-Weinberg disequilibrium (DHW) both in controls and in patients (p < 0.05).

Regarding GRIA3 gene, we obtained statistically significant evidence for an association to migraine phenotype for the -1952T/C variant (rs3761555) located in the regulative region of the gene (allele C frequencies: controls-migraineurs = 22% vs. 34%) (Table 2). In consideration of the X chromosomal localization of GRIA3 gene, we analyzed the data by gender, showing significant associations in females for both allelic and genotype distributions (p = 0.0006 and p = 0.0009, respectively). Stratified analyses of migraine subtypes showed a specific association with the female MA subtype group for either allelic or genotypic frequencies (p = 0.0001 and p = 0.0002, respectively) (Table 2). These differences remained statistically significant after Bonferroni correction (alleles p = 0.003; genotypes p = 0.005) (Table 2). The male group was not analyzed independently due to its limited size (n = 56). The statistical power for the three associated SNPs in GRIA1 and GRIA3 genes was 97% for rs548294, 97% for rs2195450 and 93% for rs3761555. When we calculated the statistical power for subgroups (MA, MO, females, males) it remained high in MA and females, but it decreased for MO and males subgroup due to their limited size. We analyzed the distribution of the haplotypes formed by the associated SNPs in GRIA1 gene in cases and controls. Haplotype frequencies (calculated using the EH program) were significantly different in cases and controls (chi-square = 12.72, df = 3, p-value 0.005). Haplotype-specific testing showed that the putative risk haplotype T_A (rs548294- rs2195450) was more frequent in cases than in controls, resulting in p-value of 0.002 and an OR of 2.1. Conversely, the putative protective haplotype C_G (rs548294- rs2195450) was more frequent in controls than in cases resulting in a p-value of 0.00001 and an OR of 0.4 (Table 3).

Possible additive interactions of the two most associated SNPs in GRIA1 (rs548294) and GRIA3 (rs3761555) genes were evaluated in those subjects who had both SNPs characterization. GRIA1 SNP rs2195450 was not analyzed because it was in HWD. In Migraine patients, the p-value for the sole risk allele of the GRIA1 gene (TT/TT, CT/TT genotypes) was 0.03, OR 1.68 (CI = 1.02-2.77); the p-value for the sole risk allele of the GRIA3 gene (CC/TC, CC/CC genotypes) was 0.005, 2.28 (CI = 1.27-4.11) and the p-value for the combined two risk alleles (TT/TC, TT/CC, CT/TC, CT/CC genotypes) was 0.00001, OR 3.35 (CI = 1.96-5.74) (Table 4).

Bioinformatics predictions by TRANSFAC [16] showed that the -1952T>C associated variant could affect the putative transcription factor binding sites HSF (Heat Shock Factor) and CdxA. To determine whether the c.-1952T>C SNP alters nuclear protein-DNA interactions, EMSAs were performed using radio-labelled oligonucleotide probes containing either -1952T or -1952C (see materials and methods). Two distinct complexes were revealed on incubation of the probes with nuclear extract from unstimulated HEK293T cells (figure 2A) and the -1952T probe showed stronger DNA protein-binding activity compared with the -1952C probe. EMSA assay was also performed using nuclear extract from neuroblastoma cell line SH5Y5Y and obtained the same results (data not shown). To further assess the binding specificity and the differences in binding affinity between the T and C allele, competition assay was performed with unlabeled -1952T or -1952C oligonucleotide. The results revealed that unlabeled -1952T oligonucleotide but not -1952C with 100- fold molar excess fully blocked the binding of the radiolabeled -1952T probe or -1952C probe with nuclear protein(s). Moreover, to determine if HSF1 is one of the transcription factors bound to the complexes, anti-HSF1 antibody was used in super shift assays. In these experimental conditions none of the complexes was specific for HSF1 (figure 2A). These data suggest that the -1952T>C polymorphism can affect the binding affinity of the GRIA3 promoter with several transcription factor(s) and the variant T allele has a stronger binding strength compared with the C allele. To test the functional effect of the -1952 T>C variant we performed the luciferase assay using the promoter region of GRIA3 gene.

When HEK293T cells were transfected with the -1411 to +325 region, the luciferase activity generated by the C construct was similar to that generated by the T construct but interestingly, under heat shock condition (at 40°C for 1 h) the luciferase activity generated by the C construct was 1.3-fold greater than that generated by the T construct (figure 3).

Bioinformatics predictions did not show any clear alteration of transcription factor binding sites both for SNP -2012C>T and +561G>A. To determine whether these variations could alter nuclear protein-DNA interactions we performed EMSAs assays using radiolabelled oligonucleotide probes containing either -2012C or -2012T and +561G or +561A (see materials and methods) (figure 2B). No binding affinity was revealed for both +561G and +561A alleles. In contrast, two distinct complexes were revealed on incubation of the probes -2012T and -2012C with nuclear extract from unstimulated HEK293T cells and in particular, the -2012T probe showed stronger DNA protein-binding activity compared with the -2012C probe (figure 2B). The binding specificity and the differences in binding affinity between the T and C allele was also demonstrated in competition assay using unlabeled -2012T or -2012C oligonucleotide. The results revealed that unlabeled -2012T oligonucleotide but not -2012C with 100- fold molar excess fully blocked the binding of the radiolabeled -2012T probe or -2012C probe with nuclear protein(s) (figure 2B).

The identification of associated variants in the promoter regions of both GRIA1 and GRIA3 genes and their evidence for functional roles, suggested us to investigate about the evolutive conservation of this class of genes. We selected from databases the full-length sequences of the transcripts of GRIA1-GRIA4 genes and the regulative regions at the 5' of each gene. To postulate a coordinate regulation of these genes, all sequences were compared using the vista software and the results suggest that GRIA1 seems be the ancestral gene and is more similar to GRIA3 gene as for other genes (GRIA2 and GRIA4) (figure 4).