The association between candidate migraine susceptibility loci and severe migraine phenotype in a clinical sample

Abstract

Introduction

The objective of the study was to follow up and to test whether 12 previously identified migraine-associated single nucleotide polymorphisms were associated as risk factors and/or modifying factors for severe migraine traits in a Danish clinic-based population.

Methods

Semi-structured migraine interviews, blood sampling and genotyping were performed on 1806 unrelated migraineurs recruited from the Danish Headache Center. Genotyping was also performed on a control group of 6415 people with no history of migraine. Association analyses were carried out using logistic regression and odds ratios were calculated assuming an additive model for risk. The proxies for severe migraine traits (early onset of migraine; many lifetime attacks, prolonged migraine and tendency to chronification of migraine) were tested against the 12 single nucleotide polymorphisms and a combined genetic score in both a case-control and case-only logistic regression model.

Results

We successfully replicated five out of the 12 previously reported loci and confirmed the same direction of effects for all the 12 single nucleotide polymorphisms. In line with the recently published genome-wide association meta-analysis, the associations were significant for all migraine and migraine without aura but not for migraine with typical aura. Two single nucleotide polymorphisms (rs2274316 and rs11172113) conferred risk of many lifetime attacks inthe case-control analysis. In the case-only analysis, only three single nucleotide polymorphisms showed nominal association with many lifetime attacks and prolonged migraine attacks.

Conclusion

Our study supports previously reported findings on the association of several single nucleotide polymorphisms with migraine. It also suggests that the migraine susceptibility loci may be risk factors for severe migraine traits.

Introduction

Migraine with typical aura (MTA) and migraine without aura (MO) are prevalent and burdensome neurological disorders with high socioeconomic cost (1,2). The hereditary component of migraine has previously been estimated to be as high as 57% and is likely to be caused by multiple genetic factors, thus characterizing migraine as a common complex disorder (3–7).

The prevailing method of genetically investigating complex disorders is the genome-wide association study (GWAS), in which the frequency of common variants is compared between the disease group and healthy controls. Several studies have attempted to elucidate genetic factors for migraine and the past years have witnessed promising advances. So far, three GWAS (8–10) and a genome-wide meta-analysis spearheaded by the International Headache Genetics Consortium have identified 12 genetic loci as conferring risk of migraine (11). This recent effort, a large-scale meta-analysis comprising 23,285 migraineurs and 95,425 controls, pooled mostly new results from 29 different GWAS and resulted in the identification of five new migraine-loci as well as confirmation of previously reported loci (11). Furthermore, the study demonstrated that the subgroup composed of clinic-based samples had larger effect sizes for all 12 genome-wide significant loci compared to migraine cases identified in population-based sampling; suggesting that the clinic-based cases may be genetically enriched compared to cases sampled from the general population (11).

We aimed to evaluate if the 12 candidate single nucleotide polymorphisms (SNPs) could be replicated in an independent clinic-based migraine cohort and to test whether these migraine susceptibility loci were specifically associated with severe migraine characteristics inherent in clinically ascertained migraineurs recruited from a tertiary referral centre.

Materials and methods

Study population

The migraine sample comprised 2463 participants from both a recent and a previously collected migraine cohort from the Danish Headache Center (DHC), Glostrup Hospital. DHC is a national tertiary referral centre that receives headache patients from general practitioners, neurology departments and neurology practices. The study protocol and detailed information about the migraine cohort have been described previously (12,13). A semi-structured interview was conducted using a previously validated questionnaire. The interviews were performed one-to-one at the DHC or over the telephone by trained physicians or senior medical students. The survey included all the pertinent clinical data for correct migraine diagnosing according to the ICHD-2 criteria (14). Standard blood sampling was performed for the purpose of DNA-extraction and genotyping.

In an effort to minimize both a clinical and a potential genetic heterogeneity, cases diagnosed with ‘probable’ migraine with regard to headache characteristics and accompanying symptoms were excluded. We did, however, choose to include the cases with tendency of prolonged migraine attacks since this was a feature of migraine severity which we intended to study. In addition, we also filtered for relatedness retaining only the proband. Cases with missing genotypes were excluded. The remaining sample comprised 1806 individuals, out of which 796 had MTA and 1010 had MO. The skewed proportion of MO and MTA in our cohort compared to the population prevalence was deliberate as we meant to recruit an equal number of cases with MTA and MO for our genetic studies. The mean age was 44.7 (±12.4) years and the male to female ratio was 1:4.6 (322:1484).

The control samples included 6415 healthy and unrelated individuals of Danish ethnicity enrolled as controls for epidemiological and genetic studies in Denmark. These controls were pooled from three sources: 458 were migraine-free controls from the DHC, recruited using the same semi-structured questionnaire mentioned above; 1326 blood donors, who reported the absence of any pre-existing medical condition or taking any medication were recruited under the auspices of the Danish Blood Donor Study (M-20090237); 4631 persons with no self-reported history of severe headaches from the population-based Inter99 study sampled at the Research Centre for Prevention and Health, Glostrup Hospital, Denmark. The Inter99 study (ClinicalTrials.gov: NCT00289237) is a randomized, non-pharmacological intervention study for the prevention of ischaemic heart disease (15). The present study protocol was in accordance with the Helsinki Declaration and approved by the Danish Ethical Standards Committee and Danish Data Protection Agency (Protocol number: H-2–2010–122). Written informed consent was obtained from all participants.

SNP selection and genotyping

For genotyping we selected migraine-associated SNPs that exceeded the threshold for genome-wide significant association (P < 5 × 10⁻⁸) in the abovementioned migraine meta-analysis (11). The 12 markers that were available at the time of this study were included (rs2651899/PRDM16; rs10915437/near AJAP1; rs12134493/near TSPAN2; rs2274316/MEF2D; rs7577262 -surrogate marker (r²= 1.00) for rs6741751/TRPM8; rs6790925/near TGFBR2; rs9349379/PHACTR1; rs13208321 -surrogate marker (r²= 0.91) for rs11759769/FHL5; rs4379368/c7orf10; rs10504861/near MMP16; rs6478241/ASTN2; rs11172113/LRP1).

All samples were typed using Centaurus assays at deCODE genetics, Reykjavik, Iceland. For each SNP, genotype call rate was at least 95% in both cases and controls and a test for deviation from the Hardy–Weinberg equilibrium in controls was non-significant. For rs6790925, plates containing 347 samples could not be included due to poor genotyping quality.

Statistical analysis

Follow-up case-control analysis of the 12 SNPs

The follow-up association analysis for the 12 SNPs mentioned above was carried out using logistic regression with 1806 migraineurs and 6415 healthy controls. The analysis was performed for MTA and MO both separately and combined. Odds ratios (ORs) were calculated assuming an additive model for risk. The level of significance was set at P < 0.004 (0.05/12).

Association analysis of the 12 SNPs and severe migraine characteristics in a case-control and a case-only logistic regression model

Migraine is sub-classified as MTA and MO (14), but as the primary aim of this analysis was to investigate clinical migraine as a subgroup, with a focus on severe migraine characteristics, we have chosen not to discriminate between the two subtypes.

In a logistic regression analysis assuming an additive genetic effect, we investigated the single locus association and the association of a cumulative risk score, with severe migraine traits. This was done, first, by comparing severe migraine cases with non-migraineurs, i.e. healthy controls (case-control analysis) and, second, the same analysis was carried out again comparing cases with severe migraine traits with migraine cases without these severe sub-phenotypes (case-only analysis). The rationale for using these two distinct logistic regression models is that they represent two fundamentally different assumptions. The first model represents a conventional way to re-assess the migraine risk alleles that were identified in a case-control association study and investigates if the SNPs confer risk of severe migraine. In the latter model we perform a conditional analysis to test the influence of the variants on specific traits within a severely affected migraine population. Thus, in the case-only analysis we aim to test whether the risk alleles have a modifying effect on specific severe migraine traits.

To our best knowledge, no previous studies have described severe migraine characteristics or traits: hence, these traits were defined by consensus among headache specialists at the DHC as traits that are frequently reported by migraineurs at a tertiary referral centre for headache (Olesen, internal communication). Severe migraine characteristics were defined as the following: early onset of migraine (migraine onset =/<10 years of age); many lifetime attacks (=/>100 migraine attacks during the entire lifetime); tendency to have prolonged migraine attacks (migraine attacks lasting >72 hours); chronification of migraine (headache =/>15 days per month and having at least eight migraine attacks monthly over a period of 3 months. This characteristic was not adjusted for prior medication overuse).

The effect allele was set as the unfavourable allele associated with migraine as determined by the previous report (11). In an additive model based on an expected allelic dosage model for SNP markers, we calculated a multi-locus cumulative genetic score for each individual by adding up the number of risk alleles from the original study (11). We also calculated a cumulative genetic risk score using only the seven SNPs that were significant for ‘all migraine’ in the original study (11), since these variants were significantly associated with migraine in a larger sample size.

The level of significance was set at P < 0.001 (0.05/52).

Data on the binary end-points were extracted from the semi-structured interview. Results were presented as ORs, 95% confidence intervals (CI) and P-values. All of the analyses were adjusted for sex and age. Statistical analysis was performed using statistical software SAS version 9.3 (Copyright (c) 2002–2008 by SAS Institute Inc., Cary, NC, USA) for Microsoft Windows 10.

Results

All the 12 candidate migraine SNPs showed the same direction of effects as in the original study (11) and seven out of the 12 SNPs were replicated in our clinical migraine cohort (Table 1). However, the two SNPs rs6478241 (ASTN2) and rs6790925 (near TGFBR2) only reached nominal significance resulting in five out the 12 SNPs succeeding the significance threshold for association with migraine after correction for multiple testing (Table 1; P < 0.004). A two-tailed binomial test showed that the observed consistency of effects with the previous study (11) is significantly different than what would be expected by chance for the 12 SNPs (P = 2 × 10⁻⁴). In accordance with the GWAS meta-analysis data (11), our data show that the SNPs show slightly larger effect sizes in MO than in MTA and also associate more significantly with MO than with MTA (Table 1). The lowest P-value in MTA was seen for rs2274316 (MEF2D) (P = 8.1 × 10⁻³). The most significant P-values for combined migraine were seen for rs12134493 (near TSPAN2), rs2274316 (MEF2D), rs9349379 (PHACTR1), rs11759769 (FHL5) and rs11172113 (LPR1). Although population sub-stratification among the controls was considered unlikely, a sensitivity analysis was performed in two subsets (the Danish Blood Donor Study and the DHC controls compared with the Inter99 study cohort) and showed no significant differences in association with the 12 SNPs (P > 0.05 for all candidate SNPs).

Table 1.

Association results of 12 meta-genome wide significant single nucleotide polymorphisms (SNPs) (11) with migraine with typical aura (MTA), migraine without aura (MO) and all migraine in a clinic-based sample. N_cases = 796 MTA and 1010 MO. N_controls = 6415.

SNP	Chr: Position (bp)a	Gene/Nearby gene	Alleleb	MTA OR (95% CI)	MTA P-value*	MO OR (95% CI)	MO P-value*	All migraine OR (95% CI)	All migraine P-value*	Direction of effect previous/ current study
rs2651899	1:3073572	PRDM16	C	0.95 (0.86–1.06)	3.7 × 10−1	1.04 (0.95–1.15)	4.0 × 10−1	1.00 (0.93–1.08)	9.6 × 10−1	+/+
rs10915437	1:4082866	Near AJAP1	G	0.94 (0.84–1.05)	2.5 × 10−1	1.02 (0.93–1.13)	6.5 × 10−1	0.99 (0.91–1.06)	6.9 × 10−1	−/−
rs12134493	1:115479469	Near TSPAN2	A	1.08 (0.92–1.27)	3.4 × 10−1	1.25 (1.09–1.44)	1.5 × 10−3	1.18 (1.05–1.32)	4.6 × 10−4	+/+
rs2274316	1:154712866	MEF2D	C	1.16 (1.04–1.29)	8.1 × 10−3	1.27 (1.16–1.40)	<1.0 × 10−4	1.22 (1.13–1.32)	<1.0 × 10−4	+/+
rs6741751c	2:234483608	TRPM8	A	0.87 (0.72–1.04)	1.1 × 10−1	0.94 (0.81–1.10)	4.3 × 10−1	0.91 (0.80–1.03)	1.2 × 10−1	−/−
rs6790925	3:30455089	Near TGFBR2	T	1.08 (0.98–1.21)	1.3 × 10−1	1.09 (0.99–1.19)	8.6 × 10−2	1.09 (1.01–1.17)	3.0 × 10−2	+/+
rs9349379	6:13011943	PHACTR1	G	0.87 (0.78–0.97)	1.1 × 10−2	0.85 (0.77–0.94)	1.0 × 10−3	0.86 (0.80–0.93)	1.0 × 10−4	−/−
rs11759769d	6:96967075	FHL5	A	1.11 (0.99–1.26)	7.3 × 10−2	1.18 (1.06–1.32)	2.2 × 10 − 3	1.16 (1.06–1.26)	1.0 × 10−3	+/+
rs4379368	7:40432725	c7orf10	T	1.05 (0.89–1.24)	5.4 × 10−1	1.09 (0.94–1.26)	2.4 × 10−1	1.08 (0.96–1.21)	2.2 × 10−1	+/+
rs10504861	8:89617948	Near MMP16	T	0.92 (0.80–1.06)	2.4 × 10−1	0.99 (0.89–1.14)	9.5 × 10−1	0.97 (0.88–1.07)	5.0 × 10−1	−/−
rs6478241	9:118292450	ASTN2	A	1.09 (0.98–1.22)	1.3 × 10−1	1.14 (1.03–1.25)	9.0 × 10−3	1.12 (1.03–1.21)	5.7 × 10−3	+/+
rs11172113	12:55813550	LRP1	C	0.98 (0.88–1.09)	7.1 × 10−1	0.79 (0.72–0.87)	<1.0 × 10−4	0.87 (0.81–0.94)	3.0 × 10−4	−/−

*

Significance level P < 4.2 × 10⁻³ (0.05/12). Significant values are shown in bold.

a

NCBI build 36.

b

Minor allele.

c

The surrogate SNP rs7577262 was used (r²= 1.00).

d

The surrogate SNP rs13208321 was used (r²= 0.91).

OR: odds ratio; CI, confidence interval.

In the following analysis, testing severe migraine traits in migraineurs with non-migraineurs, the migraine candidate SNPs rs2274316 (MEF2D) and rs11172113 (LRP1) showed statistical significant association with frequent migraine attacks (P < 0.0001) and rs9349379 (PHACTR1) showed a near significant association (P = 0.0008) of migraine with frequent attacks (Table 2). There was an observed increased risk of severe traits with increasing numbers of risk alleles for all four severe traits, but the association was only statistically significant for frequent attacks (P <0.0001) and nominally significant for prolonged migraine attacks (P = 0.03) and chronification of migraine (P = 0.05) (Table 2).

Table 2.

Association results of 12 meta-genome wide significant single nucleotide polymorphisms (SNPs) (11) with severe migraine phenotypes in a case vs. controls analysis. Logistic regression assuming additive genetic effect.

SNP	Risk allele/ other	Early onset of migraine (<10 years) N with/without trait=257/7438 OR (95% CI)	P-value	Many lifetime attacks (>100 attacks) Nwith/without trait=1258/6422 OR (95% CI)	P-value	Prolonged migraine attacks (>72 hours) Nwith/without trait=93/7612 OR (95% CI)	P-value	Chronification of migraine Nwith/without trait=153/7142 OR (95% CI)	P-value
rs2651899	G/A	1.08 (0.91–1.29)	0.39	1.07 (0.98–1.17)	0.15	1.14 (0.85–1.53)	0.37	0.96 (0.76–1.21)	0.74
rs10915437	A/G	0.97 (0.81–1.16)	0.73	0.99 (0.91–1.09)	0.89	0.94 (0.70–1.26)	0.68	0.87 (0.69–1.09)	0.25
rs12134493	A/C	1.09 (0.84–1.41)	0.53	1.15 (1.01–1.32)	0.03	1.40 (0.94–2.07)	0.10	1.18 (0.81–1.59)	0.47
rs2274316	C/A	1.22 (1.02–1.46)	0.03	1.26 (1.15–1.38)	<0.0001	0.99 (0.73–1.34)	0.93	1.37 (1.09–1.73)	0.008
rs6741751	G/A	1.13 (0.83–1.53)	0.45	1.11 (0.96–1.29)	0.17	0.81 (0.51-.1.27)	0.35	1.36 (0.89–2.08)	0.16
rs6790925	T/C	1.12 (0.94–1.33)	0.22	1.08 (0.99–1.18)	0.08	1.41 (1.06–1.88)	0.02	1.23 (0.96–1.51)	0.11
rs9349379	A/G	0.99 (0.83–1.19)	0.89	1.17 (1.07–1.28)	0.0008	1.47 (1.07–2.02)	0.02	1.33 (1.05–1.70)	0.02
rs11759769	A/G	1.04 (0.84–1.28)	0.75	1.16 (1.04–1.28)	0.006	1.19 (0.85–1.65)	0.32	1.06 (0.81–1.39)	0.66
rs4379368	T/C	1.09 (0.84–1.43)	0.51	1.07 (0.93–1.22)	0.36	0.57 (0.32–1.01)	0.05	0.81 (0.55–1.19)	0.29
rs1050486	G/A	0.78 (0.62–0.97)	0.02	0.93 (0.83–1.05)	0.23	0.97 (0.66–1.42)	0.86	0.91 (0.68–1.22)	0.51
rs6478241	A/G	1.00 (0.82–1.19)	0.92	1.14 (1.04–1.25)	0.007	1.24 (0.92–1.68)	0.17	1.28 (1.02–1.61)	0.04
rs11172113	T/C	1.13 (0.94–1.35)	0.20	1.22 (1.12–1.33)	<0.0001	1.22 (0.90–1.65)	0.19	1.01 (0.80–1.27)	0.96
Cumulative genetic risk scorea	1→24 Risk-alleles	1.04 (0.97–1.10)	0.27	1.11 (1.08–1.15)	<0.0001	1.12 (1.01–1.24)	0.03	1.08 (1.00–1.17)	0.05

a

The cumulative genetic score was defined as the total number of risk alleles carried by the individual.

Age and sex-adjusted logistic regression model comparing the risk allele association with severe migraine subtypes in migraine cases (both migraine with typical aura (MTA) and migraine without aura (MO)) and controls.

Significance level P < 0.0001. Significant values are shown in bold.

OR: odds ratio; CI, confidence interval.

For the case-only analysis only three migraine risk SNPs were found to associate nominally with traits reflecting severe migraine: ‘many lifetime attacks and rs2651899 (PRDM16), and rs11172113 (LPR1); ‘prolonged migraine attacks and rs6790925 (near TGFBR2) (Table 3).

Table 3.

Association results of 12 meta-genome wide significant single nucleotide polymorphisms (SNPs) (11) with severe migraine phenotypes in a case vs. case analysis. Logistic regression assuming additive genetic effect.

SNP	Risk allele/ other	Early onset of migraine (<10 years) Nwith/without trait=257/1524 OR (95% CI)	P-value	Many lifetime attacks (>100 attacks) Nwith/without trait=1258/511 OR (95% CI)	P-value	Prolonged migraine attacks (>72 hours) Nwith/without trait=93/1664 OR (95% CI)	P-value	Chronification of migraine Nwith/without trait=153/1207 OR (95% CI)	P-value
rs2651899	G/A	1.09 (0.90–1.32)	0.36	1.17 (1.00–1.36)	0.04	1.15 (0.85–1.56)	0.36	0.94 (0.74–1.21)	0.65
rs10915437	A/G	0.95 (0.78–1.14)	0.32	0.92 (0.79–1.07)	0.26	0.92 (0.68–1.24)	0.57	0.85 (0.67–1.08)	0.19
rs12134493	A/C	0.99 (0.75–1.31)	0.95	1.10 (0.88–1.38)	0.40	1.28 (0.86–1.92)	0.22	1.04 (0.74–1.48)	0.81
rs2274316	C/A	1.05 (0.87–1.27)	0.62	1.11 (0.95–1.29)	0.18	0.84 (0.62–1.16)	0.29	1.17 (0.92–1.50)	0.20
rs6741751	G/A	1.08 (0.79–1.49)	0.62	1.08 (0.85–1.38)	0.52	0.77 (0.49-.1.22)	0.26	1.36 (0.87–2.10)	0.18
rs6790925	T/C	1.07 (0.89–1.29)	0.47	1.08 (0.93–1.26)	0.31	1.39 (1.04–1.87)	0.03	1.16 (0.91–1.47)	0.23
rs9349379	A/G	0.87 (0.72–1.06)	0.16	1.09 (0.93–1.26)	0.29	1.38 (0.97–1.87)	0.07	1.22 (0.95–1.57)	0.13
rs11759769	A/G	1.06 (0.75–1.49)	0.44	1.05 (0.88–1.25)	0.62	1.08 (0.76–1.53)	0.66	0.94 (0.71–1.25)	0.67
rs4379368	T/C	1.06 (0.80–1.41)	0.69	1.01 (0.80–1.27)	0.92	0.53 (0.30–0.95)	0.03	0.75 (0.50–1.12)	0.16
rs1050486	G/A	0.75 (0.59–0.95)	0.02	0.78 (0.63–0.96)	0.02	0.96 (0.64–1.42)	0.83	0.88 (0.64–1.20)	0.41
rs6478241	A/G	0.99 (0.74–1.10)	0.30	1.04 (0.89–1.22)	0.60	1.15 (0.84–1.56)	0.38	1.19 (0.93–1.52)	0.16
rs11172113	T/C	1.01 (0.83–1.22)	0.95	1.19 (1.02–1.38)	0.03	1.09 (0.80–1.48)	0.59	0.89 (0.70–1.13)	0.33
Cumulative genetic risk scorea	1→24 Risk-alleles	0.99 (0.67–1.45)	0.94	1.26 (0.92–1.73)	0.15	1.09 (0.61–1.98)	0.77	0.79 (0.47–1.31)	0.36

a

The cumulative genetic score was defined as the total number of risk alleles carried by the individual.

Age and sex-adjusted logistic regression model comparing the risk allele association with severe migraine subtypes amongst migraine cases (both migraine with typical aura (MTA) and migraine without aura (MO)).

Significance level P < 0.0001.

OR: odds ratio; CI, confidence interval.

The multi-locus analysis, combining the risk-alleles of the 12 identified SNPs into a cumulative genetic score, showed a non-significant trend towards a larger risk of severe migraine traits with increasing genetic load for the traits ‘many lifetime attacks’ and ‘prolonged migraine attacks’ while there was no clear trend for early debut of migraine and for tendency of chronification (Table 2). The genetic risk score that included the seven ‘all migraine’ SNPs further strengthened the trend observed for many lifetime attacks and prolonged migraine attack showing at least a nominally significant association with ‘many lifetime attacks’ (P = 0.008; data not shown).

Discussion

This study re-evaluated 12 previously established migraine candidate SNPs (11) in an independent clinic-based migraine sample. The rationale for performing this study was to follow up on the interesting findings that came out of the large GWAS meta-analysis (11); namely, that the SNPs were associated with MO only and that their effect sizes were larger in migraineurs ascertained from a clinical setting. It is likely that the migraineurs ascertained from clinical settings show greater effect sizes due to genetic enrichment, but another equally likely explanation may be ascertainment bias in that these cases may show greater penetrance due to co-segregation of environmental and other risk factors.

We replicated five out the 12 previously reported migraine associated loci and confirmed that the direction of the effects of all 12 were the same as in the original study (11). Although the direction of the effect was concurrent with the previous study, none of the four loci showing the strongest association with the clinical subgroup in the previous report (Supplementary Table 6 from original report; 11) was found to be significant in our clinical sample, nor could the association with the variants PRDM16 and TRPM8 be replicated in our clinical migraine sample.

These findings merit several comments. First, the candidate migraine loci that were replicated in our relatively small and highly selected sample might be the SNPs that are associated widely with migraine across the severity spectrum. The migraine-associated markers in PRDM16 and TRPM8 may be more prevalent in population-sampled migraineurs than in more severely affected clinical cases and it is likely that other genetic variants may be associated with the severest types of migraine. Some of the discrepancy in our association results compared to the original study may be attributed to our relatively smaller sample size (the clinical subgroup in the original study was 2.5 times larger than our clinical cohort (11)) and perhaps also our stringent correction for multiple comparisons. We did, however, choose to include all the 12 SNPs in the analysis of severe traits despite the fact that only five of these were successfully replicated in our study. We assumed that these loci are robustly associated with migraine despite our lack of replication and because we were interested to test whether any of the 12 SNPs show a preferential association with a specific subset of severely affected migraineurs in our sample.

Second, in accordance with recent reports (11,16), we also found that the 12 loci were associated more significantly with MO than MTA. It should be noted that our MTA sample was smaller than the MO sample (796 vs.1010; Table 1), although this is likely not the only reason for this surprising lack of association. It may indicate that MTA is mediated through multiple rare variants. This hypothesis is supported by earlier studies suggesting that MTA is more hereditary than MO (3,4). Consequently, the genetic architecture of MTA may involve more penetrant genetic factors occurring at lower population frequencies. However, in a recently published study by Nyholt et al., this issue is debated (17). The authors investigated the genetic overlap between migraine subgroups in the large GWAS meta-analysis and found a significant concordance of the migraine associated SNPs across MTA and MO. Furthermore, the authors reasoned that the lack of association between the 12 SNPs and MTA may be due to a higher degree of heterogeneity among MTA cases in which genetically differing MTA subgroups may to varying degrees share the same genetic risk factors with MO (17). We believe that, in fact, both hypotheses are valid and that MTA may be caused by both rare and common variants. Furthermore, a greater clinical heterogeneity in MTA than in MO may, in particular, be a problem in studies using self-reported migraine samples rather than in carefully diagnosed migraine cases.

In our secondary analyses, we aimed to elaborate further on the observation that the 12 candidate migraine SNPs had stronger effect sizes in the clinic-based migraine samples than the population based migraine sample (11). Our approach to this matter was to investigate the interaction of the migraine susceptibility loci with severe migraine traits often found at our headache centre. Two SNPs, rs2274316 and rs11759769, appeared to confer risk of having frequent migraine attacks and the risk of having three out of four severe migraine traits (‘many lifetime attacks’, ‘prolonged migraine attacks’ and tendency of ‘chronification of migraine’) increased with a higher load of risk alleles. The results for the case-only analysis with severe migraine showed associations of nominal significance for many lifetime attacks and prolonged migraine but none of these associations withstood correction for multiple testing. Albeit a fairly weak association, this may indicate that these genetic factors could play a modifying role in the development of severe migraine in migraineurs in which other risk factors (both genetic and environmental) carry the brunt of the disease phenotype. These nominally associated results should, nevertheless, be interpreted with caution and, ideally, our findings would have to be validated by replication in another clinic-based migraine sample. However, at the time of the study these specific migraine traits were not accessible from other cohorts.

The strength of the present study is its contribution to the characterization of GWAS findings in migraine and in bridging the genetic findings to clinical interpretation. This has not been done extensively before. The weakness of the study is of course the effect of stringent multiple testing, which is crucial, especially in genetic studies but also when testing several outcomes. As the impact of the reported migraine susceptibility genes on migraine pathophysiology is unknown, we explored the association of the SNPs with several clinical end-points. The downside of exploratory studies is penalization by stringent multiple testing correction in which potentially true signals may be missed, especially in data with potentially correlated outcomes. Furthermore, in the case-only study, we tested subtypes among migraine cases only, which likely results in diminished statistical power to detect effects with appropriate statistical significance, leading to a risk of committing a type-2 error.

Lastly, our clinical end-points were chosen as proxies for the most affected and severe cases of the migraine spectrum, as no established parameters for severe migraine exist. Utilizing these migraine severity measures made sense for the purpose of this study but, in reality, other parameters such as decreased quality of life might depict severe migraine better. We believe that this is an important yet unknown characterization of migraine, which future studies may focus on.

In conclusion, this is the first study to follow up on the leads from a recently reported migraine meta- GWAS analysis in an independent clinic-based sample. We have investigated whether the 12 migraine susceptibility loci play roles as risk factors or modifiers in patients severely affected by migraine. Clinical characterization of susceptibility loci in non-Mendelian migraine may provide valuable insights that can assist in future prioritizing of susceptibility genes for i.e. functional studies.