Evidence for linkage of migraine in Rolandic epilepsy to known 1q23 FHM2 and novel 17q22 genetic loci

Authors


Abstract

Migraine headaches are a common comorbidity in Rolandic epilepsy (RE) and familial aggregation of migraine in RE families suggests a genetic basis not mediated by seizures. We performed a genome-wide linkage analysis of the migraine phenotype in 38 families with RE to localize potential genetic contribution, with a follow-up in an additional 21 families at linked loci. We used two-point and multipoint LOD (logarithm of the odds) score methods for linkage, maximized over genetic models. We found evidence of linkage to migraine at chromosome 17q12-22 [multipoint HLOD (heterogeneity LOD) 4.40, recessive, 99% penetrance], replicated in the second dataset (HLOD 2.61), and suggestive evidence at 1q23.1-23.2, centering over the FHM2 locus (two-point LOD 3.00 and MP HLOD 2.52). Sanger sequencing in 14 migraine-affected individuals found no coding mutations in the FHM2 gene ATP1A2. There was no evidence of pleiotropy for migraine and either reading or speech disorder, or the electroencephalographic endophenotype of RE when the affected definition was redefined as those with migraine or the comorbid phenotype, and pedigrees were reanalyzed for linkage. In summary, we report a novel migraine susceptibility locus at 17q12-22, and a second locus that may contribute to migraine in the general population at 1q23.1-23.2. Comorbid migraine in RE appears genetically influenced, but we did not obtain evidence that the identified susceptibility loci are consistent with pleiotropic effects on other comorbidities in RE. Loci identified here should be fine-mapped in individuals from RE families with migraine, and prioritized for analysis in other types of epilepsy-associated migraine.

Both migraine and epilepsy are common, chronic and paroxysmal disorders (Forsgren et al. 2005; Stovner & Andree 2010) with complex genetic etiologies (Anttila et al. 2010; Chasman et al. 2011; Pal et al. 2010b), and their comorbidity in both rare and common disorders is well established (Haut et al. 2006; Ottman & Lipton 1994). Indeed, a recent study indicated a shared genetic susceptibility to migraine with aura in many types of non-acquired focal and generalized epilepsies (Winawer et al. 2013). Such comorbidity has a great impact on quality of life and treatment success (Baca et al. 2011; Salpekar & Dunn 2007).

A recent meta-analysis of 29 genome-wide association studies (GWAS) identified 12 loci associated with migraine susceptibility, 6 of which are found in migraine without aura only and the others in combined sample groups (Anttila et al. 2013). Seven confirmed previously reported associations (PRDM16, MEF2D, TRPM8, TGFBR2, PHACTR1, ASTN2 and LRP1) and five were new (AJAP1, TSPAN2, FHL5, C7orf10 and MMP16). These genes have wide-ranging functions including roles in glutamate homeostasis, pain pathways, neuronal differentiation and synaptic morphology (reviewed in Silberstein & Dodick 2013). The study of pedigrees with atypical migraine and Mendelian inheritance, such as familial hemiplegic migraine (FHM) and migraine in genetic vasculopathies such as cerebral autosomal dominant arteriopathy with subcortical infarcts and leukencephalopathy (CADASIL), has also revealed the importance of ion channel and metabolic pathway genes, e.g. CACNA1A, ATP1A2, SCN1A and NOTCH3 in migraine genetics (De Fusco et al. 2003; Dichgans et al. 2005; Joutel et al. 1996; Ophoff et al. 1996). Some of these genes are also implicated in rare forms of epilepsy (Bianchin et al. 2010). Recent discoveries have shown that mutations in the transmembrane protein PRRT2 cause several episodic conditions including paroxysmal dyskinesia, infantile convulsions and migraine (Meneret et al. 2013). Despite these recent studies, genetic influences on the comorbidity between migraine and epilepsy are still not well understood. The determinants could be the same as those that cause common migraine or epilepsy; the same as those for more rare or monogenic disorders; or distinct and specific to the comorbidity. In this article, we set out to identify the genetic determinants of migraine susceptibility in the commonly occurring and complex genetic disorder of Rolandic epilepsy (RE) (Clarke et al. 2009).

We have previously shown that migraine is comorbid in RE [hazard ratio (HR) 2.46, 95% confidence interval (CI): 1.06–5.70], and clusters independently to the seizures along with an increased risk in siblings (HR 2.86, 95% CI: 1.10–7.43) (Clarke et al. 2009). This suggests a shared susceptibility to migraine and RE in these families that is not mediated directly by seizures. Here, we perform a genome-wide linkage analysis to determine whether migraine in RE is: (1) linked to known monogenic migraine or epilepsy loci; (2) linked to loci for common migraine; (3) linked to loci for either the electroencephalographic (EEG) endophenotype of RE, centrotemporal spikes (CTS); or (4) linked to genetically influenced comorbidities of RE such as speech sound disorder (SSD) (Pal et al. 2010a; Strug et al. 2009) or reading disability (RD) (Strug et al. 2012).

Methods

Subjects

Rolandic epilepsy probands and their families were recruited for genetic linkage analysis from centers in the USA by single ascertainment (Strug et al. 2009). Proband cases were enrolled if they met stringent eligibility criteria for RE, consisting of typical orofacial seizures, age of onset between 3 and 12 years, no previous epilepsy type, normal global developmental milestones, normal neurological examination, EEG with centrotemporal sharp waves and normal background and neuroimaging that excluded an alternative structural, inflammatory or metabolic cause for the seizures. Board-certified experts in epileptology, neurophysiology and neuroimaging centrally reviewed all the probands' charts, EEGs and neuroimaging for eligibility prior to recruitment, providing a consistent and homogeneous epilepsy phenotype. Furthermore, epilepsy cases were not required to be comorbid with any neuropsychiatric disorder, and comorbidities were unknown at the time of ascertainment.

We conducted linkage analysis in two stages. The first stage was a genome-wide linkage screen that included 38 two- or three-generation families. In the follow-up stage, we analyzed 21 additional families typed at a subset of microsatellites – see Genotyping section for details. The number of genotyped individuals with RE, or other comorbidities, is listed for each family in Table S1, Supporting Information. The phenotype of migraine was that of classic migraine without aura, with the age of onset in childhood (months) (Headache Classification Subcommittee of the International Headache Society 2004). No individual reported migraine with aura. There were no isolated periictal migraines; migraine equivalents were not assessed. We do not see phenotypic heterogeneity of migraine across individuals in our study, but the nature of the head pain, severity, frequency and associated sensory symptoms, triggers and relieving factors are exactly as in common migraine. Out of the families genotyped, there were 21 migraine multiplex pedigrees in the stage 1 dataset and 18 migraine multiplex pedigrees in the stage 2 analysis.

Ethics statement

The IRB of New York State Psychiatric Institute and collaborating centers consisting of Yale School of Medicine IRB, Drexel School of Medicine IRB, The Children's Hospital of Philadelphia IRB, Lifespan Office of Research Administration IRB, Continuum Health Partners IRB and Robert Wood Johnson Medical School IRB approved the study. All subjects and/or their guardians provided written informed consent.

Phenotyping

A pediatric-trained physician (TC or DKP) interviewed the case families. Both parents were interviewed, either together or separately, and the proband and siblings were also interviewed when age appropriate. The investigator administered a 125-item questionnaire covering perinatal, developmental, medical, educational, family history and detailed seizure semiology and treatment history (Clarke et al. 2007). For family members where no EEG was performed, the presence of CTS was coded as unknown. The questionnaire included diagnostic criteria from the International Classification of Headache Disorders (ICHD-2) (Headache Classification Subcommittee of the International Headache Society 2004), which have improved diagnostic sensitivity for migraine in pediatric patients. For migraine affectedness, children under age 12 were coded as unknown as they were still at risk of developing the condition. Other family members were coded as unknown if they had not answered the questions from the ICHD-2 criteria completely. Those above age 12 who completed the questions and did not meet the diagnostic criteria were labeled as unaffected as recall for migraine history is generally good. The questionnaire also included 13 items addressing speech articulation disorder F80.0. Nine items addressed the ICD-10 definitions of reading disorder F81.0. Reading disorder was thus identified by significant impairment in the development of reading skills not solely accounted for by mental age, sensory problems, mother tongue or inadequate schooling. Further details on the SSD and RD criteria as well as the battery of tests to assess general intelligence can be found in our earlier work (Clarke et al. 2007; Pal et al. 2010a). Siblings who were below the age range for diagnosis of SSD or RD were classified as unknown, and siblings and parents who did not meet the diagnostic requirements were classified as unaffected.

Genotyping

Blood or saliva samples (Oragene, DNA Genotek, ON, Canada) were collected from probands and all potentially informative available family members. DNA was extracted using standard protocols. Individuals from the first 38 families were genotyped at deCODE Genetics, Iceland using the deCODE 1000 marker single tandem repeat (STR) set, which has an average genome-wide resolution of 4 cM. In the second stage, individuals from 21 unrelated families were typed for markers in the same deCODE STR set on chromosomes 1–3, 5–7, 11, 12 and 15–17. Twenty additional markers not previously genotyped as part of the deCODE 1000 marker set were also typed in the replication sample in these chromosomes, marking other loci known to be linked to RE comorbidities. Amplified fragments were typed using ABI 3700 and ABI 3730 DNA analyzers (Applied Biosystems, Grand Island, NY, USA) with CEPH family DNA used as standards. Alleles were called automatically and checked for consistency with Mendelian inheritance and Hardy–Weinberg equilibrium.

Statistical analysis

We conducted a genome-wide linkage analysis of migraine with both two-point LOD (logarithm of the odds) and multipoint heterogeneity LOD (HLOD) score calculations. The first group of 38 families was used for genome-wide linkage analysis, and the second group of 21 families for linkage analysis over selected chromosomal regions of interest. We assessed linkage under parametric (i.e. specified) dominant and recessive models of inheritance because, apart from the inheritance model at the locus being studied, lodscore analysis is robust to misspecification of other parameters, and is often a more powerful approach than non-parametric methods even when the genetic model is unknown as in the case of comorbid migraine (Abreu et al. 1999; Clerget-Darpoux et al. 1986; Durner et al. 1999; Elston 1989; Goldin 1994; Greenberg & Hodge 1989; Pal et al. 2001; Vieland et al. 1992, 1993). Using the analysis approach where LOD scores are maximized over the unknown trait model (referred to as MMLS) (Hodge et al. 1997), we calculated two-point LOD scores under both dominant and recessive modes of inheritance (Abreu et al. 2002; Durner & Greenberg 1992) with a dominant gene frequency of 0.01 and a recessive gene frequency of 0.14, a sporadic rate of 0.0002 and a penetrance of 50%. In regions of linkage, we maximized the LOD scores over a grid of penetrance values from 0.1 to 0.99 (Clerget-Darpoux et al. 1986; Pal et al. 2001). A modified version of LIPED (Ott 1974) was used to calculate the two-point LODs using this MMLS procedure.

Migraine is common in the general population; therefore, parametric analysis allows one to explicitly account for the possibility of heterogeneity (i.e. there being more than one causal locus or gene) through the calculation of HLOD scores (Durner & Greenberg 1992). We followed up two-point results with multipoint analysis using the GENEHUNTER program (Kruglyak et al. 1996), with a sex-averaged genetic map, again using the MMLS procedure followed by penetrance maximization and computation of HLOD scores, which we refer to here as multipoint HLODs. The consequent increase in type I error owing to the maximization procedure is equivalent to half a degree of freedom under a χ2 distribution (Hodge et al. 1997), and can be conservatively compensated for by increasing the two-point LOD score critical value for significance by 0.3 LOD score units, i.e. to 3.3. Critical values for multipoint HLODs are not fixed, but rather are a function of pedigree number and structure (Mattheisen et al. 2008). Priority for locus follow-up was either genome-wide significance or loci with suggesting linkage evidence in previously reported regions for either epilepsy or migraine.

Evidence of common etiology

To assess the evidence that a single locus contributes to both migraine and other components of the RE phenotype, we redefined the ‘affected’ phenotype as those with migraine or the comorbid phenotype of interest, i.e. CTS, RDG or SSD. Those with unknown affectedness for one or more phenotype remained as unknown unless they were affected by a second phenotype. Pedigrees were then reanalyzed for linkage using the same methodology described above. An increase in the maximum LOD score is interpreted as evidence consistent with the hypothesis of pleiotropy. This approach has been shown to provide qualitatively similar conclusions when compared with a formal test of pleiotropy using a less-powerful affecteds-only design (Pal et al. 2010a). We did not implement this affecteds-only test for pleiotropy owing to insufficient numbers of families with the required traits.

Candidate gene sequencing

We used bidirectional Sanger sequencing and standard protocols at the Institute of Psychiatry, King's College London, for mutation screening of the ATP1A2 gene in 14 migraine-affected family members from seven families linked to the chromosome 1 locus. Single nucleotide polymorphisms (SNPs) of potential interest were sequenced using the same methods in an additional 14 family members in five more linked families. We designed primers to amplify all the exons and alternatively spliced exons, as well as the splice sites, promoter and 5′ and 3′ untranslated regions (UTRs) of ATP1A2. We sequenced DNA amplified by polymerase chain reaction (PCR) using the ABI 3130 DNA Analyzer and interpreted sequence using the Invitrogen Vector NTI Advance Suite (Life Technologies, Grand Island, NY, USA). See Table S2 for PCR primer sequences.

Results

Genome-wide linkage analysis of migraine

In two-point linkage analysis of the migraine phenotype in families with RE, we observed suggestive linkage on chromosome 17 with a maximum LOD of 2.41 at D17S788 in the first group of 38 families, maximizing under a recessive model of inheritance, at 99% penetrance, and a recombination fraction of 0.01. Multipoint linkage analysis of the migraine phenotype in these families resulted in a maximum HLOD of 4.40 on chromosome 17 (inline image = 0.92) between markers D17S788 and D17S787, i.e. at the same locus as in the two-point analysis and with the same mode of inheritance. Twenty of the 38 analyzed families provided a positive LOD score at these markers. A second more proximal peak also maximized with this genetic model on chromosome 17, with an HLOD of 3.0 at D17S2194-D17S1842 (inline image = 0.78). Analysis using sex-specific recombination maps did not provide any evidence of imprinting at these chromosome 17 regions (Fig. 1).

Figure 1.

Linkage plot of migraine in RE families on chromosome 17. Linkage analysis in original dataset of 38 families (black line) and second dataset of 21 families (gray line) with RE. Maximum multipoint HLOD is 4.38 between D17S788 and D17S787 for the first group of families under a recessive model of inheritance and 99% penetrance. The highest two-point LOD for these families was 2.42 at D17S788 (green dot). Linkage for the second group of families maximized under the proximal peak with HLOD of 2.62 at D17S1842 and a dominant model of inheritance and 90% penetrance. Blue line shows information content. Arrow indicates site of 1.5-Mb genomic inversion.

A critical value for replication of linkage results is reported to be between LOD scores of 1.0 and 2.0 (Morton 1998). Linkage to the more proximal chromosome 17 peak was replicated in the second group of 21 families, with maximum HLOD of 2.61 (inline image = 1) at D17S1842 under a dominant model with 90% penetrance. There is a known 1.5-Mb genomic inversion (DGV 37189) and common 0.7-Mb deletion/duplication (DGV 4034) at the point where the linkage signal drops between the two peaks at markers D17S1867-D17S1818. The explanation for this pattern of two peaks is uncertain: they may represent either one large or two separate susceptibility loci. We were unable to define a shorter segregating haplotype that may have helped resolve this issue. Most families have positive HLOD scores at both linkage peaks, but there is some heterogeneity in the sample as would be expected for individual families with small HLODs. Specifically, two families provided negative LOD scores at the proximal peak, four yielded negative LOD scores at the distal peak and three were not linked to either chromosome 17 marker. We did not see any differences between families with positive and negative LOD scores by ethnicity or phenotype that might have explained the heterogeneity in the proximal/distal peaks or linked/unlinked group of families.

Loci on chromosomes 1q, 2q and 12q provided genome-wide suggestive evidence for linkage with two-point or multipoint analyses generating LOD scores above 2.5 (Table 1 and Fig. S1). The 2q and 12q loci were not chosen for follow-up at this stage as priority was given to regions that were either genome-wide significant or that provided suggesting linkage evidence to previously reported loci for either epilepsy or migraine.

Table 1. Linkage to migraine in RE families at chromosomes 1, 2 and 17
Microsatellite (cytoband)Two-point linkage LOD (model, penetrance)Multipoint linkage HLOD (model, penetrance)
  1. Maximum LOD scores for both two-point and multipoint analysis for the original group of 38 families.

  2. R, recessive; D, dominant.

D1S1653 M48 (1q23.1)3.0 (R, 0.99)2.52 (R, 0.9)
D1S2707 M49 (1q23.2)1.37 (R, 0.99)2.51 (R, 0.9)
D2S428 M33 (2p12)2.46 (D, 0.99)1.25 (D, 0.9)
D2S293 M37 (2q12.2)1.35 (D, 0.99)2.78 (D, 0.9)
D2S1896 M38 (2q13)1.26 (D, 0.99)2.78 (D, 0.9)
D12S369 (12q24.21)1.29 (R, 0.99)2.52 (R, 0.85)
D12S366 (12q24.23)1.36 (R, 0.99)2.46 (R, 0.85)
D17S788 M25 (17q22)2.42 (R, 0.99)4.38 (R, 0.99)
D17S787 M26 (17q22)1.74 (R, 0.99)4.38 (R, 0.99)

Two-point LOD scores for chromosome 1 maximized at 3.00 at marker D1S1653 under a recessive mode of inheritance with 99% penetrance and a recombination fraction of 0.01. Multipoint linkage analysis provided a maximum HLOD score of 2.52 (inline image = 1) at 1q23.1-23.2 again with a recessive mode of inheritance and 90% penetrance at D1S1653-D1S2707 (Fig. 2). D1S2707 is 12.9 kb away from the gene ATP1A2, the closest gene to this marker. The suggestive linkage peak on chromosome 1 is notable because mutations in ATP1A2 cause FHM type 2 (FHM2, OMIM 602481) and therefore this gene represented the best candidate to sequence in a 1-LOD linkage interval.

Figure 2.

Linkage plot of migraine in RE families on chromosome 1. Linkage in the original dataset of 38 families with RE. Maximum two-point linkage occurs at D1S1653 with a maximum LOD of 3.0 under a recessive model of inheritance and 99% penetrance (black line). Maximum multipoint linkage occurs between D1S1653 and D1S2707 with a HLOD of 2.52 again under a recessive inheritance model and 90% penetrance (orange and green dots). Blue line shows information content.

Candidate gene sequencing of ATP1A2

We sequenced 14 individuals, from families linked to the FHM2 locus, across the coding and regulatory regions of ATP1A2 (Table 2). We identified 17 SNPs: two were synonymous coding SNPs, one novel and three other SNPs were located within splice junctions with a predicted functional effect on splicing. Allele frequencies are listed in Table 2. One insertion was identified just before exon 21 and a 4-bp deletion before exon 2. Typing in 14 further family members from five of the seven linked families showed that none of the polymorphisms were inherited solely with affection status for migraine, but cannot be conclusively ruled out.

Table 2. SNPs, insertion and deletion present from ATP1A2 sequencing of 14 individuals from seven families linked to 1q23 with RE and migraine
SNPPosition hg19 (bp)Frequency of alternate allelePosition in gene and proteinFunctional effect on protein
rs2854246160 090 674All homozygousc.13-22T>CNon-coding
rs67915591/CCTT Del160 090 682-54 homozygous, 5 heterozygousc.13-14_17delCCTTNon-coding
rs2820581160 093 222All homozygousc381+16C>TNon-coding
rs41265765160 093 6923 heterozygousc.382-41G>TPredicted splicing
rs2295623160 097 3152 heterozygousc.749-27C>ANon-coding
rs61734524160 098 5437 heterozygous, 1 homozygousc.1119G>A p.S373SCoding synonymous
rs17846713160 100 2012 heterozygousc.1462-11C>GNon-coding
rs17846715160 105 3671 heterozygousc.2259C>T p.A753ACoding synonymous
rs12083034160 106 5315 heterozygous, 1 homozygousc.2706+20G>ANon-coding
rs111331208160 106 8573 heterozygousc.2840+36G>ANon-coding
rs56077630160 109 4092 heterozygousc.2841-21insCNon-coding
rs17846717160 109 6362 heterozygousc.2943-47G>CNon-coding
rs17846718160 109 6562 heterozygousc.2943-27C>GNon-coding
rs41288127160 109 7882 heterozygousc.3034+14C>TPredicted splicing
G>A NOVEL160 109 8442 heterozygousc.3034+70G>ANon-coding novel
rs2070702160 111 7015 heterozygous, 3 homozygousc.3063+589G>CPredicted splicing
rs55843060160 111 7982 heterozygousc.3063+686G>ANon-coding
rs62620182160 111 8585 heterozygous, 3 homozygousc.3063+746C>TNon-coding
rs2070703160 111 8605 heterozygous, 3 homozygousc.3063+748C>TNon-coding

Multitrait analysis of migraine loci

The LOD scores at the chromosome 1, 2, 12 and 17 loci all decreased dramatically for both two-point and multipoint analyses when the affectedness definition was extended to migraine or CTS, indicating that pleiotropy is unlikely at these loci for the EEG phenotype of RE. The LOD scores also decreased when either RD or SSD was included in the definition of affectedness status (Table 3 and Fig. S2). There were no other peaks approaching, or above genome-wide significance for these linkage analyses.

Table 3. Maximum multipoint genome-wide HLOD scores above 2.5 for the first group of 38 RE families
ChromosomeMarker (flanking)Migraine HLOD (flanking)MIG or CTS or RE HLOD (flanking)MIG or RDG HLOD (flanking)MIG or SSD HLOD (flanking)
  1. HLOD scores for affectedness definition of migraine alone, or assessing pleiotropy using migraine (MIG) or CTS or RE, migraine or reading disability (RDG) and migraine or SSD. HLOD maximized under a dominant ‘D’ or recessive ‘R’ model of inheritance, with penetrance indicated.

1q23

Model and penetrance

D1S1653 (D1S498, D1S2707)

2.52 (1.51, 2.49)

R, 0.9

0.01 (0.53, 0.0)

R, 0.99

0.0 (0.0, 0.0)

R, 0.85

0.33 (0.56, 0.00)

R, 0.99

2q12-13

Model and penetrance

D2S293-D2S1896 (D2S2264, D2S363)

2.78 (1.38, 1.40)

D, 0.9

0.05 (0.15, 0.01)

R, 0.4

0.0 (0.0, 0.0)

D, 0.95

0.04 (0.01, 0.00)

D, 0.99

17p21.33-q22

Model and penetrance

D17S788-D17S787 (D17S1795, D17S957)

4.38 (3.59, 3.79)

R, 0.99

0.0 (0.0, 0.0)

R, 0.3

0.52 (0.13, 0.35)

D, 0.99

0.51 (0.44, 0.15)

D, 0.8

Discussion

Migraine is a common disorder of complex inheritance where the problem of genetic and phenotypic heterogeneity has limited the ability to identify genetic contributors. Many of the genetic breakthroughs in migraine have emerged from studying rare, multiplex, families with atypical forms like FHM or CADASIL and Mendelian inheritance (Dichgans et al. 2005; Joutel et al. 1996). However, migraine is not only exceedingly common in the general population but is also frequently comorbid with other neurological conditions, notably epilepsy. Studying migraine in a specific and common form of childhood epilepsy reduces phenotypic and genetic heterogeneity. In this study, both the epilepsy and migraine phenotypes were very homogeneous. However, we still observed that multiple susceptibility loci may be at play, and that these loci may be both related to genes implicated in Mendelian or common forms (e.g. at 1q23.1-23.3), or entirely novel and possibly specific to migraine in RE (17q12-22). Results from this study could be complemented by genetic studies of comorbid migraine in other types of idiopathic epilepsy. This would be especially interesting with regard to migraine with vs. migraine without aura, where shared genetic susceptibility seems to depend on the type of epilepsy studied, as well as the age of the subjects recruited (Winawer et al. 2013).

We have discovered a potential novel locus for migraine without aura on chromosome 17q12-22 both in a discovery and replication set of families ascertained through RE probands. This locus has not previously been reported in the migraine genetics literature, suggesting that genetic influences on comorbid migraine in epilepsy may be different to those for common migraine, although the current list of susceptibility genes for common migraine is not exhaustive. This locus is also not known to influence other common epilepsies. However, there is suggestive evidence for linkage of multiple seizure types at 17q22 (multipoint LOD 2.7) in a rare Finnish pedigree (Siren et al. 2010). The maximum LOD in the Finnish pedigree was obtained at D17S1606, a marker we did not type but which is only 130 kb away from D17S957, a linked marker in our study. These converging pieces of evidence suggest the possibility of shared genetic contributions that merit further investigation.

Confirmation of a susceptibility gene at the novel 17q12-22 migraine locus presented here will require further investigation to resolve several complexities. The locus is broad (the 2-LOD interval covers one third of the chromosome) and the linkage evidence (Fig. 1) appears as two peaks. It is not clear if there is one, or two separate, underlying susceptibility genes at this locus. We have excluded loss of marker information content across the region as an alternative explanation for the findings (Fig. 1), and in-depth analysis of recombination did not reveal a shorter haplotype in one peak alone that could be associated with the disorder. Most, but not all, families provided positive LOD scores at both peaks. A 1.5-Mb genomic inversion (DGV 37189) at markers D17S1867-D17S1818, the point where the linkage signal drops, further complicates the interpretation of the linkage evidence as it may be present in our samples. This inversion, also reported in control samples, encompasses 19 genes. Another common (71% of controls) 0.7-Mb deletion/duplication (DGV 4034) can also be present at this location (Redon et al. 2006) and may also interfere with linkage signals. The DECIPHER database [Database of Chromosomal Imbalance and Phenotype in Humans using Ensembl Resources (Firth et al. 2009)] also lists two individuals with deletions characterized by very similar breakpoints, and three cases with duplications, all with mental retardation, and two comorbid with speech delay. Therefore, analysis now underway of structural genomic variants in RE families might provide clarification of the linkage findings if these variants are present. Next-generation sequencing of a targeted capture from the linkage region would enable all the variation at this locus to be uncovered in a larger sample. The segregation of potential causal variants might then lead to the discovery of a susceptibility gene.

We also found suggestive linkage for migraine at 1q23.1-2. Two genome-wide screens for common migraine also found suggestive linkage to the same region: Ligthart et al. at D1S1653 (1q23.1), the same marker as our maximum multipoint HLOD (Ligthart et al. 2008), and Nyholt et al. at D1S1679, 1q23.3 (Nyholt et al. 2005). This region also contains the gene ATP1A2, which is mutated in cases of FHM type 2, FHM2 (De Fusco et al. 2003). More than 50 different mutations have been found in the gene, clustering within the functional domains in families with this autosomal dominant disorder (Jurkat-Rott et al. 2004). ATP1A2 forms the catalytic subunit of a Na+,K+-ATPase, whose activity is partially or completely compromised by these mutations. In about 20% of cases, patients with FHM2 also have seizures (Deprez et al. 2008) and some experience benign infantile convulsions (Vanmolkot et al. 2003). Mutations and common variation in ATP1A2 are not a cause of common migraine (e.g. Nyholt et al. 2008). However, studies have not investigated rare variants at ATP1A2, intermediate between mutations and common variation. Therefore, it is a possibility that ‘milder’, less penetrant genomic variants could be involved in a common pathophysiology. ATP1A2 was thus an obvious candidate gene for mutation screening in our families. Variants identified within ATP1A2 in the families sequenced here are all common (MAF > 0.023) and found in controls (Riant et al. 2005). Thus, if ATP1A2 is involved in migraine in RE, these variants do not seem likely to be contributory. The novel SNP located 70 bases into the 3′-UTR is also unlikely to be functional as it is far from the end of the gene, and not conserved in other ATPase α-subunits or other species. The absence of variants that segregate with migraine in our families suggests that the susceptibility variant(s) at this locus may lie elsewhere than in ATP1A2-coding regions. However, incomplete penetrance and phenocopies are known to occur in FHM2 families (Riant et al. 2005) and so we cannot entirely rule out these variants, nor those which are functional but intronic. Use of the ENCODE project (Encyclopedia of DNA elements: http://genome.ucsc.edu/ENCODE/) to reveal conserved regulatory elements in the region could prove useful for screening of potential SNP function.

The discovery of these two susceptibility loci for migraine in RE advances our understanding of the genetic model of this common form of childhood epilepsy, as well as for migraine itself. It now seems clear that the majority of neurological comorbidities and EEG endophenotypes in RE families are likely to be genetically influenced (Clarke et al. 2007, 2009; Strug et al. 2009): the 11p13 locus is pleiotropic for both the EEG CTS pattern and SSD in RE (Pal et al. 2010a) but has not been reported in other SSD linkage studies (Smith et al. 2005; Stein et al. 2006); while the RD experienced by approximately half of RE cases (Clarke et al. 2007) may be influenced by separate loci on chromosomes 1q42 and 7q21 (Strug et al. 2012) that are not involved in ‘pure’ dyslexia. The 1q23 and 17q12-22 migraine susceptibility loci discovered here do not overlap (or show evidence consistent with pleiotropy) loci associated with other RE comorbidities or endophenotypes (Pal et al. 2010a; Strug et al. 2009, 2012), suggesting further independent genetic effects. The genetic basis of seizures in RE has not been reported, and therefore we do not know whether migraine and seizures are linked to the same loci.

We have shown that migraine susceptibility in epilepsy may have a unique genetic component on chromosome 17, and this would not otherwise have been uncovered through the study of migraine alone. We have also provided further evidence of linkage to the migraine locus at 1q23 in our families with migraine in RE. Future studies will be able to ascertain if variants at the two loci described here are common to other epilepsy-associated migraine or contribute to migraine susceptibility in the general population. Identifying and understanding this genetic comorbidity in other types of epilepsy will also help to further our knowledge of both conditions and potentially lead to new treatments.

Acknowledgments

Our thanks to members of the IRELAND Study consortium who also referred patients to the study: Linda Leary, MD, Frances Rhoads and Maria Younes, MD. Thanks also to the late Lewis Kull, to Cigdem Akman, MD, Jeff Rogg, MD and Jerry Boxerman, MD, PhD for reviewing subjects' charts and investigations for eligibility. The authors declare that they have no conflict of interest.

Ancillary