Introduction

Genetic studies in familial hemiplegic migraine (FHM), a monogenic subtype of migraine with aura, have yielded several migraine genes. Studies in the common forms of migraine have had limited success. This review will address how molecular insight coming from studies in FHM can help understand the pathophysiology of migraine. In addition, we will provide a detailed overview of the most relevant linkage and association studies in common migraine. Emphasis will be given to clinical and genetic heterogeneity in common migraine and alternative strategies how these problems can be tackled in future studies. We envisage that these novel strategies will yield gene variants for the common forms of migraine and will increase our insight into the molecular pathways and mechanisms involved in common migraine.

Migraine is an episodic neurovascular disorder that is diagnosed according to the criteria of the International Classification of Headache Disorders (ICHD-II) from the International Headache Society (IHS) (HCC 2004). Migraine attacks are characterized by severe, often unilateral, pulsatile headache that can be accompanied by nausea, vomiting, photo- and/or phonophobia, and which last several hours to days (migraine without aura, MO) (Table 1). One-third of migraine patients experience transient (e.g., lasting between 20 and 60 min) neurological aura symptoms before the headache phase (migraine with aura, MA). The aura symptoms almost always include visual symptoms, but sensory- or speech-related symptoms can also be involved (Russell and Olesen 1996). Migraine is a very prevalent disorder that affects about 15% of the adult general population and is about three times more prevalent in females than in males (Stewart et al. 1994; Launer et al. 1999; Stovner et al. 2006). Migraine can be very disabling with 10% of migraine patients having weekly attacks. Consequently, the World Health Organization (WHO) rated severe migraine among the most disabling chronic disorders (Menken et al. 2000). Current acute and prophylactic treatments are not optimal as they are effective only in about half of the patients (Ramadan et al. 1997). A better understanding of disease pathophysiology is needed to identify novel targets for better intervention. The identification of migraine genes will help pinpointing such targets and molecular pathways in migraine.

Table 1 International Headache Society (IHS) criteria for migraine without aura (MO) and migraine with aura (MA) (Headache Classification Subcommittee of the International Headache Society (HCC) 2004)
Full size table

Migraine pathophysiological mechanisms

Although it was previously thought that migraine either had a vascular or a neurogenic origin, the current view is that migraine has a neurovascular origin (for review see Goadsby 2007). Headache is not merely the consequence of painful vasodilatation, but is due to the activation of the trigeminovascular system (TGVS) that consists of meningeal and superficial cortical blood vessels that are innervated by the trigeminal nerve. The TGVS projects to the trigeminal nucleus caudalis in the brainstem, which in turn, projects into higher-order pain centers giving rise to the headache. It is well accepted that the migraine aura is not due to reactive vasoconstriction, but is neurally driven and most likely caused by the human equivalent of the cortical spreading depression (CSD) of Leao (Leao 1944; Lauritzen 1994). In experimental animals, CSD is characterized by a short-lasting, intense wave of neuronal and glial cell depolarization that spreads slowly over the cortex at a rate of approximately 2–5 mm/min and that is accompanied by massive fluxes of ions (Ca2+, Na+, and K+) followed by a long-lasting inhibition of spontaneous and evoked neuronal activity (for review see Somjen 2002). The electrophysiological changes are associated with changes in cerebral blood flow (CBF). Initially there is a small, brief reduction in CBF followed by a profound increase of CBF lasting minutes, after which blood flow is reduced again for up to an hour. There is a considerable body of clinical evidence that CSD is the likely basis of the migraine aura. Visual aura symptoms in humans (Lashley 1941; Milner 1958; Russell et al. 1994) typically spread from the center of the visual field to the periphery with a propagation rate comparable to CSD evoked in experimental animals. Positive (e.g., scintillations, paresthesias) and negative (e.g., scotomata, paresis) phenomena of the migraine aura can be explained by the initial transient hyperexcitation front of CSD followed by neuronal depression. Most importantly, however, functional neuroimaging studies in humans using blood-oxygen level dependent (BOLD) signals have convincingly demonstrated that CBF changes that occur during a migraine aura are very similar to those observed in experimental animals during CSD (Hadjikhani et al. 2001). Additional support for the importance of CSD in migraine pathophysiology comes from a study by Ayata and coworkers who demonstrated that chronic, but not acute, treatment of rats with migraine prophylactic drugs of different pharmacological classes (e.g., β-adrenergic receptor blockers, tricyclic antidepressants, anticonvulsants and serotonergic drugs) that are effective in reducing the frequency and/or intensity of migraine headaches in patients, dose-dependently suppressed CSD frequency by 40–80% and increased the threshold for inducing CSD (Ayata et al. 2006).

Animal studies have shown that CSD can activate the TGVS, and thus might trigger headache mechanisms (Bolay et al. 2002). However, the connection between CSD and headache in patients remains an open question (Blau 1992; Goadsby 2001). For instance, it would not explain how the headache phase is triggered in the majority of migraine patients that never experience an aura. Also, the fact that ketamine treatment can reduce aura symptoms but fails to prevent the headache (Kaube et al. 2000) would argue against a key role of CSD in triggering the headache. Although one can hypothesize that spreading depression may occur in migraine without aura patients in clinically silent subcortical areas of the brain without propagating to the visual cortex (Goadsby et al. 2002; Haerter et al. 2005), this has never been demonstrated.

Migraine is a genetic disorder

Migraine has a strong genetic component. Many patients have first-degree relatives who also suffer from migraine (Russell and Olesen 1993). Population-based family studies showed that the familial risk of migraine is increased (Russell and Olesen 1995; Stewart et al. 1997). First-degree relatives of probands with MO had an almost twofold increased risk to suffer from this disorder, but had only 1.4 times the risk of MA compared with the general population. Instead, first-degree relatives of probands with MA had a nearly fourfold increased risk for MA, but no increased risk for MO (Russell and Olesen 1995). Studies of mono- and di-zygotic twin pairs are the classical method to investigate the relative importance of genetic and environmental factors. Migraine concordance rates are between 1.5 and 2 times higher in monozygotic twins than in dizygotic twins for both MO and MA (Ulrich et al. 1999; Gervil et al. 1999), indicating that genetic factors are important in migraine susceptibility. A large population-based twin study comprising of some 30,000 twin pairs revealed that genetic and environmental factors had an almost equally large contribution (Mulder et al. 2003). Shared environmental factors seemed to play a minor role as shown by studies comparing twins that were raised together or apart (Ziegler et al. 1998; Svensson et al. 2003).

For genetic study designs, it is important whether MO and MA should be seen as different disease entities. Evidence from Danish studies that show no increased co-occurrence of MO and MA compared to the product of the prevalence of MO and MA in the general population (Russell and Olesen 1995; Russell et al. 1996) or in twins (Russell et al. 2002) supports this view. Others belief that MO and MA are different expressions of the same disease with pure MO to pure MA at both ends of the clinical spectrum (Kallela et al. 2001; Nyholt et al. 2004; Ligthart et al. 2006). Clinical observations support this view as both migraine types share identical headache symptoms and frequently co-occur in an individual. Future studies will have to show whether at least some migraine susceptibility genes are shared by both migraine types suggesting that there is a migraine continuum.

Hemiplegic migraine: a monogenic form of migraine with aura

Genetic studies in FHM: genes encode ion transporters

The most straightforward approach to identify genes and unravel pathways for complex genetic disorders is the study of monogenetic subtypes of the disease. An example of a monogenic subtype of migraine is familial hemiplegic migraine (FHM), a rare form of migraine with aura. FHM can be considered a model for the common forms of migraine because the headache and aura features, apart from the hemiparesis, are identical (Thomsen et al. 2002) and two-thirds of the FHM patients have, in addition to attacks of FHM, also attacks of common non-hemiplegic migraine (Ferrari et al. 2007).

Three genes have been identified studying families with FHM. The first FHM gene that was identified is CACNA1A (FHM1), which is located on chromosome 19p13 (Ophoff et al. 1996). CACNA1A encodes the α1 subunit of neuronal CaV2.1 (P/Q-type) voltage-gated calcium channels that are widely expressed throughout the central nervous system (CNS) (Westenbroek et al. 1995). All the 21 known FHM1 mutations (Fig. 1a) are missense mutations. FHM1 mutations are associated with a broad spectrum of clinical features besides hemiplegic migraine (Ducros et al. 2001). Cerebellar ataxia (Ducros et al. 1999; Battistini et al. 1999; Kors et al. 2003; Alonso et al. 2004; Stam et al. 2008a) and epilepsy, both during severe FHM attacks (Vahedi et al. 2000) or independent of FHM attacks (Kors et al. 2004; Beauvais et al. 2004), are not uncommon. The phenotype in for instance FHM1 S218L mutation carriers can be very severe, even lethal, after mild head trauma (Kors et al. 2001; Curtain et al. 2006; Chan et al. 2008). The second FHM gene, ATP1A2 (FHM2), is located on chromosome 1q23 (De Fusco et al. 2003). It encodes the α2 subunit of sodium–potassium pumps. There are now over 30 FHM2 mutations (Fig. 1b) that, with a few exceptions, are all found in single families. Almost all FHM2 mutations are amino acid changes, but there are also small deletions (Riant et al. 2005) and a mutation affecting the stop codon causing an extension of the ATP1A2 protein by 27 amino acid residues (Jurkat-Rott et al. 2004). Most of the ATP1A2 mutations are associated with pure FHM without additional clinical symptoms (De Fusco et al. 2003; Riant et al. 2005; Jurkat-Rott et al. 2004; Kaunisto et al. 2004; Pierelli et al. 2006). However, over the years, a number of FHM2 mutations (Fig. 1b) have been reported that are associated with FHM and cerebellar problems (Spadaro et al. 2004), childhood convulsions (BFIC) (Vanmolkot et al. 2003), epilepsy (Jurkat-Rott et al. 2004; Deprez et al. 2008), permanent mental retardation (Jurkat-Rott et al. 2004; Vanmolkot et al. 2006). Interestingly, certain ATP1A2 mutations were shown to be associated with non-hemiplegic migraine phenotypes, such as basilar migraine (Ambrosini et al. 2005) and even common migraine (Todt et al. 2005), although causality has not been established for all mutations by functional testing. The most recently identified FHM gene is the SCN1A (FHM3) gene, which is located on chromosome 2q24 (Dichgans et al. 2005). SCN1A encodes the α1 subunit of neuronal NaV1.1 voltage-gated sodium channels and is a well-known epilepsy gene with over 100 truncating and missense mutations that are associated with childhood epilepsy (i.e., severe myoclonic epilepsy of infancy (SMEI) or generalized epilepsy with febrile seizures (GEFS+)) (Meisler and Kearney 2005; Mulley et al. 2005). Only five FHM3 mutations (Fig. 1c) have been identified that all change amino acid residues (Dichgans et al. 2005; Vanmolkot et al. 2007a; Castro et al. 2009; Vahedi et al. 2009). Missense mutations Q1489K and L1649Q were identified in large families and are associated with pure FHM (Dichgans et al. 2005; Vanmolkot et al. 2007a). Notably, three out of five FHM carriers of the L263V mutation had generalized tonic–clonic epileptic attacks, occurring independently from their hemiplegic migraine attacks (Castro et al. 2009). Recently two novel FHM3 SCN1A mutations (i.e., Q1489H and F1499L) were reported (Vahedi et al. 2009). Four of five carriers of the FHM3 Q1489H mutation, in addition to having hemiplegic migraine, also suffered from ‘elicited repetitive transient daily blindness’ (ERDB) that was not associated with headache or other neurological symptoms (Le Fort et al. 2004). Moreover, both the proband and his daughter also suffered from childhood epilepsy. In the FHM3 family with the F1499L mutation, there was also a link with ERBD, but only in the proband.

Fig. 1
figure 1

Schematic representation of SHM and FHM mutations in FHM gene-encoded proteins. a Mutations in the α1A subunit of the voltage-gated CaV2.1 Ca2+ channel encoded by the FHM1 CACNA1A gene (Genbank Ac. nr. X99897). The protein is located in the plasma membrane and contains four repeated domains, each encompassing six transmembrane segments. b Mutations in the α2 subunit of the Na+,K+ ATPase encoded by the FHM2 ATP1A2 gene (Genbank Ac. nr. NM_000702). The protein is located in the plasma membrane and contains ten transmembrane segments. c Mutations in the α1 subunit of the voltage-gated NaV1.1 Na+ channels encoded by the FHM3 SCN1A gene (Genbank Ac. nr. NM_006920). The protein is located in the plasma membrane and contains four repeated domains. Symbols: Circle with solid line FHM, circle with dotted line SHM, circle with horizontal striped pattern basilar-type migraine, circle with vertical striped pattern common migraine. Asterisk Mutation for which also SHM was reported, black circles mutation was tested for functional consequences, white circle mutation was not tested for functional consequences

Full size image

The fact that not all FHM families are linked to one of the three known FHM loci implies that there are additional FHM genes. Recently, a novel FHM locus on chromosome 14q32 was proposed in a Spanish migraine family (Cuenca-León et al. 2009). However, genetic information of six FHM patients and seven patients with either MO or MA was combined to reach significant LOD scores. Therefore, future studies will have to show whether this locus contains a gene that can cause both FHM and common migraine.

Not all hemiplegic migraine patients are part of FHM families. The so-called sporadic hemiplegic migraine (SHM) patients do exist, and exhibit clinical symptoms that are similar to those of hemiplegic migraine patients (Thomsen et al. 2003). For instance, also SHM—like FHM—patients can have attacks of common migraine that are not associated with hemiparesis. Also the prevalence of familial and sporadic hemiplegic migraine in the population is similar; both are rare with a prevalence of approximately 0.01% (Thomsen and Olesen 2003). Several studies have tried to identify mutations in FHM genes in SHM patients (Terwindt et al. 2002; de Vries et al. 2007; Thomsen et al. 2008. Whereas about 15% of SHM patients in a Dutch clinic-based study had gene mutations (predominantly in the ATP1A2 gene) (de Vries et al. 2007), hardly any mutations were identified in 100 patients from a Danish population-based study (Thomsen et al. 2008). SHM mutations are depicted in Fig. 1. Although mutations in known FHM genes have been identified in a proportion of SHM patients, and SHM thus, in part, belongs to the genetic migraine spectrum, other genetic factors likely play a role. Sporadic occurrence of disease can be explained by several genetic mechanisms; a de novo mutation, reduced penetrance of a hemiplegic migraine gene mutation, or an unfavorable combination of low-risk factors of common migraine gene variants that are present in a single patient.

There are no obvious clinical differences between mutation carriers of the three FHM genes, although patients with FHM1 mutations, more often exhibit cerebellar ataxia or trauma-triggered attacks (van den Maagdenberg et al. 2007; Ducros et al. 2001). Strikingly, for all three FHM genes, there are mutation carriers that have epilepsy (Haan et al. 2008). This is perhaps not so surprising given the epidemiological evidence that there is a bidirectional comorbidity of migraine and epilepsy (Ottman and Lipton 1994), suggesting that both disorders have, at least in part, a shared pathophysiology. The identification of gene mutations that can cause both FHM and epilepsy provides a unique opportunity to study these mechanisms.

Functional studies of FHM mutations

Functional consequences of FHM gene mutations have been studied in cellular and animal models. Not less than 12 FHM1 mutations have been tested in heterologous expression systems expressing recombinant CaV2.1 channels for their consequences on calcium channel functioning using whole cell or single channel electrophysiology (for review see Pietrobon 2007). At the single channel level, all eight investigated FHM1 mutations were shown to open at more negative voltages and have an enhanced channel open probability, compared to normal channels (Hans et al. 1999; Tottene et al. 2002, 2005). This gain-of-function effect results in increased Ca2+ influx, which would predict increased neurotransmission. At the whole cell level, however, neurons from CaV2.1-α1 knockout mice (Jun et al. 1999; Fletcher et al. 2001) transfected with either wildtype or mutant CaV2.1-α1 cDNA constructs, all seem to show a loss-of-function effect (Hans et al. 1999; Tottene et al. 2002, 2005; Cao et al. 2004). Hippocampal neurons from CaV2.1-α1 knockout mice, that were transfected with wildtype or mutant CaV2.1-α1 cDNA constructs showed a reduced neurotransmitter release and a decreased contribution of P/Q-type channels in controlling neurotransmitter release (Cao et al. 2004). However, when mutant CaV2.1 channels are expressed at their normal level in their natural environment in knock-in mice that harbor the human FHM1 R192Q mutation that is associated with FHM in patients (Ophoff et al. 1996), electrophysiological analysis of cerebellar granule cells revealed a gain-of-function effect at the whole cell level. This is in line with gain-of-function effects observed at the single channel level in transfected neurons. The paradoxical loss-of-function effect on whole cell level in transfected cells is thought to be an artifact of overexpression (van den Maagdenberg et al. 2004), although others propose that a species differences (mice vs. humans) is the reason (Cao et al. 2004). Evoked and spontaneous neurotransmission was shown to be increased at the neuromuscular junction; a synapse in the peripheral nervous system where release of transmitter is predominantly determined by CaV2.1 calcium channels. Most relevant for migraine pathophysiology, in FHM1 R192Q mutant mice, the threshold for CSD was lowered and the propagation velocity for CSD was increased. These observations indicate that the FHM1 mutant mice are useful models to study the pathophysiology of migraine in vivo. Future studies have to reveal exactly how a lower activation threshold of mutated CaV2.1 Ca2+ channels can lead to an episodic disease. One may think that only when the stimulus is strong (for instance with repetitive firing) and the threshold is temporarily lowered (for instance by hormonal changes), hyperexcitability of neurons in a susceptible brain will lead to an attack. Decreased inhibition by G-protein modulation as exhibited by several FHM1 mutations (R192Q, S218L, Y1245C) (Melliti et al. 2003; Weiss et al. 2008; Serra et al. 2009) may contribute to this neuronal hyperactivity by promoting deinhibition. The complexity of CaV2.1 Ca2+ channel modulation is further demonstrated by the fact that the functionality of channels expressing certain FHM1 mutations (K1336E, W1684R, V1696I) can vary when combined with different auxillary β-subunits (Müllner et al. 2004). At this moment, it is still unclear to what extent such modulations play a role in disease pathophysiology.

The functional consequences of a large number of ATP1A2 mutations causing either FHM or SHM have been investigated in various in vitro assays. Many were shown to be dysfunctional as they—unlike wildtype—were not (or only partially) able to rescue cell survival in assays in which endogenous sodium potassium pumps were inactivated by the drug ouabain (Koenderink et al. 2005). In these assays, wildtype or mutant α2 Na+,K+-ATPase cDNAs were insensitive to the ouabain challenge. Other studies investigated the consequences of ATP1A2 mutations in more detail and revealed a wide variety of functional changes. FHM2 mutations G301R, T376M, L764P, W887R, led to non-functional proteins (Capendeguy and Horisberger 2004; Koenderink et al. 2005; Tavraz et al. 2008, 2009). Other FHM2 mutations resulted in sodium potassium pumps with partial activity with decreased (in the case of T345A and A606T) or increased (in the case of R689Q, M731T, R763H, and X1021R) affinity for potassium (Segall et al. 2004, 2005; Tavraz et al. 2008). For five FHM2 mutations (i.e., R383H, R689Q M731T, R763H, and R834Q) a reduced turn-over rate was identified. Since FHM2 mutations compromised pump function, Atp1a2 knockout mice may serve as a good model for FHM. However, Atp1a2 knockout mice that lack the α2-subunit have a very severe phenotype and die immediately after birth because of their inability to start breathing (Ikeda et al. 2003; James et al. 1999). Heterozygous mice are viable and exhibit enhanced fear and anxiety following conditioned fear stimuli (Ikeda et al. 2003). These mice have not been used as potential migraine mouse models.

Five FHM3 mutations have been identified so far (Dichgans et al. 2005; Vanmolkot et al. 2007a; Castro et al. 2009; Vahedi et al. 2009) and for three of them the functional consequences have been investigated. Early functional studies of the FHM3 Q1489K and L1649Q mutations used cardiac NaV1.5 cDNA as the backbone for making mutant cDNAs and revealed various gain-of-function effects (Dichgans et al. 2005; Vanmolkot et al. 2007a). When expressed in the more appropriate NaV1.1 protein, FHM3 mutations Q1489K and L1649Q revealed clear loss-of-function effects (Kahlig et al. 2008). The third FHM3 mutation L263V that in patients causes FHM and in the majority of carriers also generalized tonic–clonic epilepsy, essentially had gain-of-function effects (Kahlig et al. 2008). It was hypothesized that loss of sodium channel activity primarily disturbs the functioning of inhibitory neurons, where the NaV1.1 are expressed normally (Ogiwara et al. 2007; Yu et al. 2006), whereas gain-of-activity has a predominant effect on excitatory neurons. From a recent in vitro study expressing FHM3 mutation Q1489K in cultured neurons, it is clear that the functional consequences of FHM3 mutations can be very complex. Depending on the test paradigm, the mutation had functional consequences fitting either hyperexcitability or hypoexcitability (self-limiting hyperexcitability capacity) (Cestele et al. 2008), but this has not been tested in for instance knock-in mice.

How do FHM mutations cause disease?

Can the molecular genetic findings of the three FHM genes (CACNA1A, ATP1A2 and SCN1A) be integrated into one common pathway? More specifically, can we link the functional consequences of the three genes to increased propensity for CSD (Moskowitz et al. 2004)? Mutant CaV2.1 calcium channels from FHM1 R192Q knock-in mice predict increased glutamate release in the cerebellar cortex and thereby can easier induce, maintain, and propagate CSD; this is in line with the observed decreased threshold for CSD in knock-in mice (van den Maagdenberg et al. 2004). FHM2 mutations in the sodium–potassium pump predict in vivo reduced glial uptake of K+ and glutamate from the synaptic cleft. FHM3 mutations in the NaV1.1 sodium channel predict in vivo hyperexcitability of excitatory neurons. Therefore, the predicted consequence for FHM1, FHM2, and FHM3 is that mutations seem to cause increased levels of glutamate and K+ in the synaptic cleft and because both are facilitators of CSD, an increased propensity for CSD. The increased propensity for CSD could well explain the aura. As discussed earlier, it remains to be established whether this also would result in a more readily activated TGVS and thereby the headache.

Is it possible to translate genetic results from FHM to common migraine?

As the main clinical symptoms of headache and aura are similar in FHM and common migraine, it is thought that they may share a common pathophysiology (Ferrari et al. 2007). Several studies have investigated the role of FHM1 and FHM2 loci in the common forms of migraine. These studies led to conflicting results with some evidence in favor of the hypothesis (May et al. 1995; Nyholt et al. 1998; Terwindt et al. 2001; Todt et al. 2005), while others find no evidence for their involvement in common migraine (Hovatta et al. 1994; Jones et al. 2001; Jen et al. 2004). Some of the studies hypothesized that mutations found in FHM may cause common migraine, while it is more likely that ‘milder’, less penetrant, DNA variants are involved. A recent comprehensive study, including some 2,800 migraine patients from various European countries, tested whether common DNA variants in ion transport genes are involved in common migraine (Nyholt et al. 2008). Over 5,000 SNPs in 155 ion transport genes (including the 3 FHM genes) were studied, but none of the original significant SNPs (66 SNPs in 12 genes) was significant across all four replication cohorts. From this study it seems that common variants in ion transport genes do not play a major role in susceptibility for common migraine. Rarer variants or variants with a smaller effect size would not have been detected with this study design.

Other strategies for the genetic dissection of migraine?

In addition to investigating patients with rare monogenic FHM, what other options do we have to find gene variants for migraine? Several strategies have been tried. First of all, migraine can be part of other monogenic disorders. Identification of genes for these disorders possibly shed light on why migraine co-occurs in mutation carriers. Other strategies in common forms of migraine are candidate gene association studies and linkage approaches (Fig. 2). The various strategies and the results that came from them are discussed below.

Fig. 2
figure 2

Successful strategies for identification of genes in monogenic and complex forms of migraine. The linkage approach yielded genes in FHM that have a high effect size and a low population allele frequency, whereas candidate gene association studies tested potential migraine gene variants with a low effect size and relatively high allele frequencies in the population

Full size image

Monogenetic syndromes associated with migraine

Migraine can be part of the clinical spectrum of certain monogenetic disorders. Perhaps the clearest example of migraine associated with a monogenic disease is cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) (Stam et al. 2008b). CADASIL is caused by mutations in the NOTCH3 gene, which encodes the Notch3 receptor that plays a key role vascular smooth muscle cell function in small arteries and arterioles of the brain (Joutel et al. 1996). Up to one-third of CADASIL patients suffer from migraine with aura, where migraine often is the presenting clinical symptom (Dichgans et al. 1998). The link with MA and not the more frequent MO, suggests that increased susceptibility for CSD is somehow caused by NOTCH3 mutations, but this has not been investigated yet. A second example is retinal vasculopathy with cerebral leukodystrophy (RVCL), which is caused by mutations in the TREX1 gene that encodes the major 3′–5′ DNA exonuclease TREX1 (Richards et al. 2007). RVCL is primarily characterized by progressive blindness due to vascular retinopathy and can be associated with a wide range of systemic and cerebral symptoms, including cerebral infarcts and white matter hyperintensities, vascular dementia, liver and kidney dysfunction, Raynaud’s phenomenon, and migraine. Particularly in a Dutch RVCL family, migraine and Raynaud’s phenomenon is very prominent (Terwindt et al. 1998). Comorbidity of migraine with CADASIL and RVCL indicates that cerebral or meningeal vasculopathy and vascular dysfunction may play a role in migraine (Vanmolkot et al. 2008). The vascular role in migraine was also shown by altered functional arterial properties and abnormal endothelial progenitor cells in migraine patients (Vanmolkot et al. 2007b; Lee et al. 2008). Future research is needed to reveal the possible role of NOTCH3 and TREX1 in migraine.

Linkage studies in common migraine

Another major genetic strategy to identify migraine susceptibility genes is classical linkage analysis, which aims to find chromosomal loci using a family-based approach. Over the years, a number of chromosomal loci (Table 2) have been identified using either migraine without or with aura patients diagnosed according criteria of the International Classification of Headache Disorders (ICHD-II) (HCC 2004). Except for a few loci, replication successes have been scarce. A promising migraine susceptibility locus resides on chromosome 4. A linkage study of 50 Finnish MA families revealed a locus on chromosome 4q24 (Wessman et al. 2002). Subsequently, an Icelandic study identified a locus on chromosome 4q21, but using MO patients (Bjornsson et al. 2003). The Finnish and Icelandic migraine loci overlap and it is unclear whether they harbor different genes for MA and MO, or one gene that causes both migraine types. For none of the reported chromosomal loci, the causative gene variant has been identified. There are several possible explanations for this rather disappointing outcome. Probably, due to the high prevalence of migraine, it has been difficult to ascertain “clean” pedigrees for linkage where migraine genes from spouses do not interfere. However, it is well conceivable that a major reason for the lack of success is the way migraine patients are diagnosed. Diagnosis of migraine is mainly based on questionnaires and (sometimes) interviewing the patients. A more objective method of diagnosing patients, such as a biochemical test in blood is not available. It may well be that some of the published migraine susceptibility loci in fact are false positive findings.

Table 2 Summary of relevant linkage results performed for migraine using the International Headache Classification (IHS) classification guidelines
Full size table

Diagnostic issues; from endpoint to trait classifications

The ICHD criteria resulting in an end diagnosis of migraine with or without aura are well suited for clinical practice (that is to diagnose the attacks that patients have), but may not be very useful for genetic research. Why is that? The IHS diagnosis consists of a number of traits (e.g., the presence of pulsating quality of the headache, nausea, vomiting) that can occur in patients in different combinations and it is the combination of these traits that is used to diagnose a patient with migraine with or without aura (although patients can also have both types of attacks). Consequently, an end diagnosis inevitably carries a lot of underlying heterogeneity, thereby complicating genetic research. However, knowledge of the traits in individual patients provides possibilities for new avenues of intermediary or endophenotyping, which can increase the power of the genetic analyses. Interestingly, for several other complex disorders such as attention-deficit/hyperactivity disorder (ADHD) and schizophrenia, it was shown that the heritability of the individual traits is higher than of the syndrome as a whole (i.e., combination of traits) (Rommelse et al. 2008; Tuulio-Henriksson et al. 2002). Recently, researchers investigating migraine genetics have adopted two alternative approaches that are based on the individual traits: “trait component analysis (TCA)” and “latent class analysis (LCA)”. Whereas TCA is a rather straightforward analysis of the individual traits, LCA is a complex statistical empirical clustering approach based on factor analysis that combines the information of several migraine symptoms. In principle, TCA and to a certain degree LCA, reflects the underlying processes in migraine pathophysiology as they utilize the questionnaire-based information in a more optimal manner, compared to the dichotomous IHS end diagnosis (Anttila et al. 2008). It can be expected that by using TCA, the clinical heterogeneity will be reduced, since traits better reflect the biological pathways that are influenced by specific genetic variations.

Recent linkage studies in migraine using alternative phenotyping methods

Nyholt and colleagues were the first to use LCA, to identify subgroups of migraine patients (Nyholt et al. 2004, 2005). In their studies, patient subgrouping was based on the presence and clustering of individual migraine symptoms. This resulted in four classes of (1) asymptomatic individuals (CL0), (2) patients with a mild form of recurrent non-migrainous headache (CL1), (3) patients with a moderately severe form of migraine, often without visual aura (CL2), and (4) patients with a severe form of migraine, often with aura (CL3). This classification reflects disease severity and does not specifically separate MO from MA. The LCA strategy seems to provide support for migraine as a continuum caused by multiple factors contributing to migraine severity and susceptibility, and not so much separate migraine with and without aura. Although more individuals are labeled affected using the LCA approach, compared to traditional IHS classification, all individuals considered affected by the IHS criteria are also considered affected in the LCA classification.

The LCA approach yielded a novel migraine susceptibility locus on chromosome 5q21 (Table 3) (Nyholt et al. 2005). Subsequent trait analyses revealed that the 5q21 locus is associated predominantly with pulsating headache. Some evidence for linkage was reported for other chromosomes with certain migraine symptoms, but none passed the threshold for multiple testing. Another Australian study recently reported linkage for the LCA severe phenotype (CL3) to a locus on chromosome 18p11. The strength of the LCA approach, according to this study, was shown by the fact that using traditional IHS criteria, no linkage was found (Lea et al. 2005). A similar conclusion about LCAs potential to detect loci, came from a Dutch study that reported suggestive linkage to a locus on chromosome 10 (Lighthart et al. 2008).

Table 3 Summary of linkage results performed for migraine grouped for phenotyping methods latent class analysis (LCA) and trait components analysis (TCA)
Full size table

Trait components analysis in migraine was introduced by Palotie and colleagues (Anttila et al. 2006), who later compared TCA with LCA and classical clinical end diagnoses (Anttila et al. 2008). The most prominent result of their studies was a highly significant LOD score to a locus on chromosome 10q22-q23 with the TCA unilaterality phenotype (Table 3). The same locus was identified with five other TCA phenotypes, the IHS MA end diagnosis, and the LCA migrainous headache class. This locus showed robust replication in an independent Australian sample (but only for the TCA pulsation trait) and in the Dutch study that was mentioned earlier (Lighthart et al. 2008). Although TCA and LCA yielded a novel migraine susceptibility locus, these new phenotyping methods still need to show whether they will lead to identification of gene variants. Although, the field will certainly welcome the identification of variants for TCA traits, it remains to be determined how a gene variant that is associated with a specific trait such as nausea will increase our insight into migraine pathophysiology and/or patient’s diagnosis.

Association studies: candidate genes and genome-wide approaches

Another commonly used strategy to identify gene variants in complex genetic disorders are case–control association studies. These studies test for significant differences in allele frequencies between cases and controls. Many candidate gene association studies have been performed; mainly of genes involved in the serotonin and dopamine pathways—but also in other genes with an already suspected function in migraine pathophysiology—have been studied (Table 4). Unfortunately, the majority of the associations could not be replicated, suggesting that many of the original findings may represent false positive findings. There are several reasons for the inconsistencies among which, small sample sizes, not taking into account LD blocks when choosing polymorphisms, inadequate correction for multiple testing, and phenotyping issues. Clearly more systematic approaches are needed to assess whether candidate genes are indeed involved in migraine. To give an example, it is surprising that despite the strong evidence from neuroanatomical, pharmacological, clinical studies that the dopamine system is involved in the etiology of migraine (for review see Akerman and Goadsby 2007), most of the association studies investigating genes in this pathway were negative and positive results could often not be replicated. A more optimal design was used in a recent German study systematically investigating genes of the dopamine system (Todt et al. 2009). Using a two-step design, testing 50 polymorphisms in that were selected using a haplotype-based strategy in an adequately sized sample identified several associated polymorphisms in the dopamine β-hydroxylase, dopamine D2 receptor and dopamine transporter genes.

Table 4 Summary of association studies performed for migraine that contained at least 275 cases and controls
Full size table

The best example of a genetic association with migraine is the 5′,10′-methylenetetrahydrofolate reductase (MTHFR) gene. MTHFR is a key enzyme in folate and homocysteine metabolism (Goyette et al. 1994). Most studies found an association of the T-allele of the MTHFR C677T polymorphism with migraine (Kowa et al. 2000; Kara et al. 2003; Lea et al. 2004; Oterino et al. 2004, 2005; Scher et al. 2006), although negative findings have been reported as well (Todt et al. 2006; Kaunisto et al. 2006). A recent meta-analysis of over 2,000 migraine patients revealed that the T-allele that results in moderately increased levels of homocysteine, was not associated with migraine overall, but only with MA (Rubino et al. 2007). Increased homocysteine levels may cause migraine through a vascular endothelium dysfunction effect, but evidence for this hypothesis is still lacking.

Genome-wide association

Only very recently, genotyping techniques and platforms have greatly improved making massive genotyping in large patient cohorts feasible. For many complex disorders, genome-wide association studies (GWAS) have led to successes, where linkage and candidate gene-based associations fell short (Aulchenko et al. 2008; van Es et al. 2008). The unbiased way of testing for the involvement of common variants throughout the genome in GWAS turned out more powerful than testing variants in candidate gene association studies. Large study populations (ideally well over a thousand cases and controls are used, although recent studies use even much larger samples) are needed to have sufficient power to detect genetic effects after correction for multiple testing. A critical factor for GWAS is the quality of phenotyping the cases and, depending on the design, the controls. No GWAS results have been reported for migraine, but several studies are currently being performed. Important lessons from recent GWAS are that, with few exceptions, identified variants have a low relative risk (RR) (that is well below 2) and contribute little to the overall genetic load. One may ask whether it is worthwhile to invest in the identification of risk factors with a RR of around 1.1, as this require huge sample sizes and will in all likelihood provide only a marginal contribution to the population attributable risk. Other factors such as epistasis and copy number variation have to be included in forthcoming study designs. Furthermore, it cannot be excluded that much of the genetic load is carried by a large number of allelic variants that may have a high RR and a (very) low allele frequency. This genetic variation is usually not captured in GWAS, but requires large-scale deep sequencing approaches. Fortunately, the appropriate technology is now emerging. Although the clinical predictive value of low-risk factors will be limited, it is important to obtain novel insights into the pathology of disease. The next few years will have to show what this new technology can bring for migraine.

Some thoughts on epigenetic and trigger mechanisms in genetic research

One possibility that has not gained much attention is that epigenetic mechanisms may play a role in migraine pathophysiology. Epigenetics is defined by (meiotic and mitotic) modifications in gene expression, due to methylation of cytosines in specific CpG dinucleotides and histone modifications that are heritable but are not encoded in the DNA sequence. These modifications decrease (or increase) gene expression and thus influence molecular pathways (Jirtle and Skinner 2007). Epigenetic effects occur in early development but may change over lifetime, by lifestyle and diet, and it has been suggested that changes for instance in development of the individual or triggers such has stress determine the ‘epigenetic outcome’ and thus may affect migraine susceptibility (Montagna 2008). It would be interesting to subgroup migraine patients for genetic and/or epigenetic studies according to their self-reported triggers. A complicating factor, however, may be that often-mentioned triggers such as stress, hormonal changes, sleeping and eating habits, are not very reliable and not well understood (Lambert and Zagami 2008). For instance, although stress-sensitive patients reported an increase in perceived stress in the days before an attack, this was not accompanied by objective signs indicating a biological stress response (such as morning or evening cortisol levels) (Schoonman et al. 2007).

Opportunities for biomarkers in genetic research

As mentioned earlier, an important drawback for genetic migraine research is the lack of biomarkers. No objective, reliable, quantitative measurements are available for diagnosis. This should trigger research initiatives to identify such markers in migraine patients using cerebrospinal fluid, blood, urine, and/or imaging (Loder and Rizzoli 2006). Biomarkers will help defining more homogeneous groups of patients for genetic studies. The identification of biomarkers may also be of direct use in patient care, as they may help monitoring drug response, disease prognosis, and/or disease progression.

Conclusion/summary

The identification of genetic factors conferring migraine susceptibility will be important for our understanding of pathophysiological disease mechanisms. Studies in FHM, a rare Mendelian form of migraine with aura, yielded three genes, which all play a role in controlling ion and neurotransmitter levels in the brain. Functional studies of gene mutations in various cellular and animal models point toward neuronal hyperexcitability as an important disease mechanism. No convincing evidence has been obtained that the same genes play a major role in the common forms of migraine, although there is convincing evidence that disturbed ion balances and neurotransmitter pathways play a role. Until now, genetic studies investigating the common forms of migraine were largely unsuccessful. However, the use of alternative phenotyping methods (i.e., LCA and TCA) and large-scale GWAS approaches hold all the promises that migraine susceptibility genes will be identified soon.