Advances in genetics of migraine

Migraine is thought to be a complex brain network disorder that occurs when the brain loses control of its homeostasis, leading to the activation of the trigeminovascular system and a cascade of events [2]. Signals from activated nociceptors innervating the cranial blood vessels are transmitted to the trigeminal bipolar neurons, and further relayed to thalamic and cortical areas [3, 4]. The signal from the perivascular neurons is transmitted by endogenous mediators, including the vasoactive neuropeptides calcitonin gene-related peptide (CGRP), substance P, neurokinin A, and pituitary adenylate cyclase-activating peptide (PACAP), as well as release of vasoactive inflammatory mediators such as nitric oxide, coincident with inflammation in the meninges [2, 5]. Sensitization of pain relevant brainstem regions, including peripheral trigeminovascular neurons to dural stimuli, is thought to produce the characteristic sensation of throbbing pain in migraine [6, 7].

During migraine, distinct areas of the brain are activated, each contributing to aspects of migraine pathophysiology, whether this is triggering the attack, generating the pain, or playing roles in some of the associated neurological symptoms that occur during an attack [2]. Migraine is characterised by multiple phases; trigeminal activation occurs in the headache phase, but these may be preceded by a premonitory phase, in which symptoms including fatigue, mood changes, food cravings, yawning, muscle tenderness, and photophobia may be experienced up to 3 days before the headache [8]. Some individuals also experience an aura phase, which may feature visual, sensory, speech/language, and motor disturbances, as well as disruption of higher cortical function, immediately preceding or concurrent with the headache [8]. Cortical spreading depression (CSD) is a slowly propagating wave of depolarization in neuronal and glial cell membranes accompanied by massive ion fluxes, which spreads across the brain cortex, followed by a suppression of activity [9]. It coincides with the initiation and progression of aura symptoms, but whether CSD is causally linked to the initiation of headache is still debated [10]. Evidence from experimental animals supports a pivotal role of CSD in aura, headache initiation and activation of trigeminal nociception [11‐13]; CSD-associated opening of neuronal Panx1 mega-channels releases molecules that trigger an inflammatory cascade, which activates neighboring astrocytes and leads to sustained release of inflammatory mediators [13]. Most migraineurs, however, do not experience aura, and it is unlikely that CSD is involved in initiating the complete syndrome of migraine. Alternative triggers for trigeminovascular activation, such as cortical hyperexcitability and brain stem or hypothalamic dysfunction, may also be important [14].

A variety of imaging techniques have revealed both structural and functional brain alterations in individuals that suffer migraine [14]. Furthermore, clinical and neurophysiological studies have found chronic hypersensitivity to sensory stimuli and or abnormal processing of sensory information in migraineurs [15‐17], as well as cortical excitability which may make them more susceptible to CSD [17, 18]. While some of these changes may be the result of repetitive exposure to pain or stress, the brain biology of migraine sufferers appears to differ from healthy controls [2]. Migraine may be triggered by a range of external factors, including chemicals, lack of sleep, stress, and skipping meals. However, these triggers only lead to migraine in migraineurs. Some aspects of the altered brain biology are likely to be genetically predetermined.

The prevalence of HM has been found to be up to 0.01% in European populations, with both familial and sporadic forms [23, 25, 26]. FHM is diagnosed when there is at least one 1st or 2nd degree relative in the family who also suffers HM attacks. FHM usually shows an autosomal dominant pattern of inheritance (with 70–90% penetrance) and is considered to be monogenic, but genetically heterogeneous. To date three main causative genes – CACNA1A, ATP1A2 and SCN1A – have been identified through linkage studies and mutational screening in FHM family pedigrees. FHM can be classified as FHM1 (MIM #141500), FHM2 (MIM #602481) and FHM3 (MIM #609634) according to whether patients have mutations in CACNA1A, ATP1A2 or SCN1A, respectively. Clinically these FHM sub-types are indistinguishable, as symptoms overlap, but there is wide variation in phenotypes, including between individuals with mutations in the same gene, or even family members with the same mutation [27‐29]. This suggests that other genes or environmental factors can modify phenotype. It should be noted that the majority of cases (< 25%) do not appear to have mutations in the CACNA1A, ATP1A2 or SCN1A genes [30] and our results (under review). Nevertheless, identifying and studying the known FHM genes and mutations has greatly improved diagnostics as well as understanding of the underlying biology of HM. The three main HM genes encode ion channel or ion transport proteins, leading to the supposition that HM is a channelopathy [31].

Sporadic Hemiplegic Migraine (SHM) is diagnosed when there is no family history of HM, and estimates suggest in the general population approximately one-third of cases are sporadic [25]. SHM can be caused by pathogenic variants in the known FHM genes, including those that have arisen de novo, which may then become familial cases [41, 74, 112] Variants in ATP1A2 have been the most commonly found in SHM cases, possibly reflecting greater genetic heterogeneity, or more variable penetrance, in this gene [62]. SHM may result from less penetrant variants in the known FHM genes, mosaicism in the transmitting parent, pathogenic variants in other genes, and/or other modes of inheritance, e.g. compound recessive mutations and gene/environment interactions [23, 93]. Some SHM cases may also represent a phenotypic extreme of common migraine due to a combination of lower-risk genetic variants. For example, Pelzer et al. (2018) found that individuals with HM, but without mutations in CACNA1A, ATP1A2 or SCN1A, generally have a milder phenotype than those with mutations in those genes [41].

Although rare, pathogenic variants in other genes, including PRRT2, PNKD, SLC2A1, SLC1A3, SLC4A4, have been reported in HM. Mutations in PRRT2 and PNKD are more commonly associated with paroxysmal conditions, in particular movement disorders [113]. PNKD is the main causal gene for paroxysmal non-kinesigenic dyskinesia (PNKD; MIM #118800) [114, 115], while PRRT2 mutations can cause paroxysmal kinesigenic dyskinesia (PKD; MIM #128200) [116, 117], paroxysmal non-kinesigenic dyskinesia (PNKD) [118], paroxysmal exercise-induced dyskinesia (PED), and childhood epilepsy/seizure disorders [119, 120]. Some patients presenting with HM have been found to have mutations in PRRT2 [118, 121‐124], leading to the suggestion that it is a fourth HM gene [121]. However, the relationship is complicated due to the clinical heterogeneity and pleiotropy of phenotypes, and it may mainly act in a modifying role [125]. PRRT2 encodes Proline Rich Transmembrane Protein 2 (PRRT2), a presynaptic transmembrane protein which interacts with members of the SNAP Receptor (SNARE) complex [126]. It is involved in synaptic vesicle fusion and regulation of voltage-gated calcium channels in glutamatergic neurons, and is important in the final steps of neurotransmitter release [127‐129]. Heterozygous PRRT2 c.649dupC (p.Arg217Profs*8) or c.649delC (p.Arg217Glufs*12) loss-of-function truncating mutations are the most common in PRRT2-related conditions, including HM, and are likely to result in impaired interaction with the SNAP25/SNARE complex and increased presynaptic vesicle release, leading to a state of hyperexcitability [118].

Mutations in both PNKD, the main causal gene for PNKD, and SLC2A1, the glucose transporter protein type 1 (GLUT1 or EAAT2) gene implicated in PED and GLUT1 deficiency syndrome (MIM #606777), have also been found in HM patients [118, 130, 131]. They likely act via disruption of neurotransmitter regulation and impaired synaptic vesicle release [118]. Mutations in SLC1A3, the gene for the glial glutamate transporter EAAT1, can cause episodic ataxia, type 6, (EA6; MIM #612656), but have also been associated with HM [132, 133]. Similarly, mutations in SLC4A4, the gene for the sodium bicarbonate cotransporter NBCe1, which is usually involved in renal tubular acidosis syndromes (MIM #604278) are also found in some HM cases [134]. Analysis of whole exome sequencing (WES) data of HM patients without CACNA1A, ATP1A2 and SCN1A mutations suggests that mutations in all these genes are rare [41] and our results (under review), but should be considered in the molecular diagnosis of patients without mutations in the main HM genes.

A monogenic form of typical MA in a large multigenerational pedigree identified a frameshift mutation (F139Wfsx24) in the TWIK-related spinal cord potassium channel (TRESK, encoded by KCNK18), segregating with migraine [135]. TRESK is a member of the two pore domain potassium channel (K2P) family, which regulate the excitability of a variety of neurons involved in transducing pain stimuli, including the somatosensory neurons of the dorsal root ganglia (DRG) and trigeminal ganglia [136, 137]. KO mouse models suggest TRESK functions to modify certain forms of nociceptive afferentation [138, 139]. Functional analysis suggested a dominant negative effect of the TRESK F139Wfsx24 mutation on whole-cell TRESK currents resulting in hyperexcitability of trigeminal ganglion neurons [140]. However, another dominant negative TRESK mutation, C110R, which is not associated with migraine [141], does not trigger sensory neuron hyperexcitability, even though it reduces TRESK currents in sensory neurons [142]. A recent study by Royal et al. (2019) sheds light on this apparent contradiction and has revealed a novel mechanism by which frameshift mutations can alter the function of a gene [143]. Firstly, they found that TRESK can heterodimerise with two other K2P channels, TREK1 and TREK2, which when knocked out together in mice results in a migraine-like allodynia phenotype. The TRESK-C110R protein inhibits TRESK activity on dimerization, but does not affect TREK1 and TREK2, while TRESK-F139Wfsx24 inhibits activity of all three channels. Interestingly, the 2 bp frameshift puts an alternative start codon in frame, which results in translation of a second TRESK fragment. It is this that specifically downregulates TREK1 and TREK2 function, which appears to contribute to migraine induction. Furthermore, Royal et al. (2019) identified another TRESK frameshift mutation (Y121LfsX44) in a human exome sequence database, and which is associated with migraine in ClinVar, that appears to work via the same mechanism which they have termed frameshift mutation-induced alternative translation initiation [143]. Finally, this work suggests that TREK-related genes may also be involved in migraine.

Casein kinase 1 delta (CKIδ) is a central component of the circadian clock. Mutations in the CKIδ gene, CSNK1D, were found to cause familial advanced sleep phase syndrome (FASPS) in two large independent pedigrees [144, 145]. FASPS patients show severe disruption of the sleep-wake-cycle and other circadian rhythms, but interestingly, the phenotype also co-segregated with MA in these pedigrees. Mice carrying a transgene with the human CKIδ-T44A mutation displayed sensitization to pain after triggering migraine with nitroglycerin, and a reduced threshold for CSD; cultured astrocytes showed increased spontaneous and induced calcium signalling [144, 145]. Further details of its role in migraine are to be elucidated, but CKIδ is a ubiquitous serine-threonine kinase that phosphorylates the circadian clock protein PER2, as well as other proteins involved in brain signalling [146]. CSNK1D is a notable exception to the ion channel and glutamatergic-related genes implicated in the majority of monogenic migraine, and the connection between migraine and FASPs is consistent with a likely role of the hypothalamus in regulating physiological stresses and migraine susceptibility [147‐149].

ROSAH is a recently described distinct autosomal dominant ocular systemic disorder, which features migraine headache as one of the key clinical features. Exome and genome sequencing identified a heterozygous missense pathogenic variant in the ALPK1 gene (c.710C > T, p.[Thr237Met]) in five independent families [150]. ALPK1 encodes Alpha Kinase 1, which may play roles in inflammation and intracellular trafficking, although its function is poorly defined, and it is not yet understood the how mutations in the protein would contribute to migraine.

There are a number of primarily vascular disorders caused by mutations in single genes, in which migraine is a common symptom. Cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL), is a cerebral small-vessel disease (SVD) characterised by vascular degeneration, recurrent subcortical ischaemic strokes, cognitive decline, dementia, and premature death [54]. It is the most common heritable cause of stroke and vascular dementia in adults, caused by toxic gain-mutations in NOTCH3, which are usually autosomal dominant. Migraine, in particular the MA subtype, is a common symptom accompanying CADASIL (in up to 75% cases) [151‐154], often presenting decades before the onset of other symptoms [54, 155]. For example, a study of 300 symptomatic CADASIL patients found that three quarters had migraine (90% of which was MA), and in two-thirds of the patients it was the presenting symptom [153].

Other SVDs that commonly feature migraine include syndromes such as retinal vasculopathy with cerebral leukodystrophy (RVCL; MIM #192315) caused by mutations in TREX1 [156, 157], and COL4A1 and COL4A2-related disorders [158‐160]. The exact mechanism through which vascular disorders lead to an increased prevalence of migraine is unknown [154], but they indicate that some genes with roles in vascular function are also implicated in migraine, something which has also become apparent in polygenic migraine from both epidemiological studies and GWAS [161, 162].

Until relatively recently HM genetic testing involved Sanger sequencing of selected exons in one, two or all three main HM causative genes (CACNA1A, ATP1A2 and SCN1A). This form of iterative testing was limited and could be costly and time consuming. The development of next-generation sequencing (NGS) technologies, in which millions of small fragments of DNA are sequenced in parallel, have revolutionised genomic research, allowing specific regions of interest to the entire genome to be sequenced concurrently. NGS applications include targeted gene panels, WES (in which all the coding regions of the genome are sequenced), and Whole Genome Sequencing (WGS), which also captures introns, regulatory regions and all other non-coding DNA. NGS has been applied clinically in genetic diagnostics, including for HM and overlapping disorders, facilitating the discovery of novel HM mutations [163‐165]. Using a five gene panel designed for HM and overlapping disorders (EA2 and CADASIL), our laboratory has found that diagnostic success rates have increased considerably (~ 21%) when compared to that of previous Sanger sequencing testing methods (~ 9%), and have identified a number of novel causative variants for HM and related disorders [166, 167]. Clinicians also appreciate the option to test for overlapping neurological disorders when presented with complex cases with HM-related symptoms.

Importantly, recent application of NGS sequencing techniques to screen HM patients have shown that the majority do not have exonic mutations in the main HM genes [30]. We find that > 75% patients sent for testing do not have likely pathogenic exonic variants in CACNA1A, ATP1A2 or SCN1A (under review). Furthermore, analysis of data from NGS panels or WES has revealed that likely pathogenic variants in other known familial migraine and migraine-related genes are also rare [41], [our results (under review)]. This low level of diagnostic success may largely be due to other causative genes or genetic factors, although no other major HM loci have been found so far [41]. In addition to the three main genes, HM may be highly genetically heterogeneous. From what is already known of the biology, other genes likely to be involved in HM may include ion channel and solute transporter genes, as well as genes involved in aspects of glutamatergic neurotransmission and vascular biology. Assigning causality for variants that are less dominant or penetrant than those in the known HM genes will be challenging. This is exemplified in a study by Klassen et al. (2011) comparing ion channel variant profiles of unaffected individuals to those with sporadic idiopathic epilepsy from targeted exome sequencing; rare missense variants were prevalent in both groups at similar complexity, demonstrating that even deleterious ion channel variants confer uncertain risk to an individual depending on the other variants with which they are combined [168]. In fact Hiekkala et al. have hypothesized that HM may not be a true monogenetic disease, but that it may reflect an extreme phenotype in the MA spectrum where rare and/or multiple common variants contribute to the disease outcome [30].

Determining the biological effect of variants on protein function is a major limitation in medical genetics. As NGS techniques reveal many more variants, particularly if HM is highly genetically heterogeneous, it will be necessary to improve functional testing pipelines to filter those likely to be pathogenic. Public databases which provide variant frequency (e.g. dbSNP, Genome Aggregation Database [169]) and previously reported pathogenicity information (e.g. ClinVar [170], Leiden Open Variation Databases), and in silico bioinformatics tools which predict functional consequences (e.g. SIFT [171], Polyphen2 [172], and MutationTaster) are useful in prioritising lists of candidate variants by providing first assessments of pathogenicity [173, 174, 175]. In silico methods to predict the impact of regulatory variants are also being developed [176, 177]. In addition to in silico analysis, functional assays are necessary to provide further evidence for pathogenicity, or otherwise, for prioritised variants, and to explore molecular mechanisms. Testing exogenous DNA constructs with engineered variants in cell and animal models can be complimented with genome-editing technologies, particularly the clustered regularly interspaced short palindromic repeats (CRISPR)-Cas9 system, which allows more refined and faster generation of knock-out or knock-in lines [178]. Coupled with induced pluripotent stem cells (iPSCs), which are able to be differentiated into various neuronal cell types [179, 180], as well as brain organoids [181], variants can be functionally tested in more relevant cell models, or generated from patients so they can be studied in the context of their genomic background. A range of approaches to scale up such assays are being developed [182], e.g. deep mutational scanning, which combines large scale generation of variants with deep sequencing, is a technique allowing the effect of a combination of variants to be tested at once [183], and high throughput electrophysiology platforms are available for testing ion channel variants [184].

A molecular diagnosis is likely to improve management and treatment efficiencies for neurological disorders, even if symptoms may overlap, as the specific pathway or mechanism can be targeted. E.g. Glut1 deficiency caused by SLC2A1 mutations can be treated using a ketogenic diet and HM symptoms, if present, have been found to improve on a modified Atkins diet [131]. In HM cases with PRRT2 mutations, some benefit has been observed with carbamazepine, the most frequently used drug in treating PKD and PKD/IC patients [185]. A range of acute and prophylactic drugs are used for HM, and some may be more effective than others depending on the nature of the causative genetic mutation [22].

Analyses of the genes in the vicinity of GWAS loci has suggested the types of gene function and pathways that may be involved in migraine, however, it is important to remember that for the majority of loci, the gene that is actually influenced by the SNP remains unknown. SNPs affect the diversity of human traits/diseases via various mechanisms: changing encoded amino acids of a protein (non-synonymous) may affect its function or localisation; and SNPs that are either silent (synonymous), or more commonly, in noncoding regions, may affect gene expression levels via messenger RNA (mRNA) conformation and stability, subcellular localization, or its promoter/enhancer activity. Making the leap from associated SNPs to causal genes, and then to functional mechanisms, still presents a formidable task in the interpretation of GWAS.

Methods have been developed to fine-map GWAS loci, combining statistical and functional evidence [214, 215]. Firstly, association-test statistics can be combined with LD information to prioritise a credible set of SNPs likely to contain the causal disease-associated SNP. As susceptibility SNPs often lie in introns or intergenic regions, the next hurdle is to identify which gene is affected (not necessarily the nearest), by connecting the variants with genes by a range of methods and resources, complementing functional annotation with information from projects such as ENCyclopedia of DNA Elements (ENCODE), NIH Roadmap Epigenomics, and FANTOM5, which have characterized regulatory regions and expression quantitative trait loci (eQTL) [162, 214]. Once putative variants and genes have been pinpointed via in silico analysis, further functional experiments are required to confirm and understand molecular mechanisms. This process is illustrated by investigations into rs9349379 in intron 3 of the PHACTR1 gene, which has been identified as a causal susceptibility SNP in a range of vascular disorders including migraine [216]. From epigenomic data from human tissues, Gupta et al. (2017) identified an enhancer signature over rs9349379 in aorta suggesting a vascular regulatory function; then using CRISPR-edited stem cell-derived endothelial cells they demonstrated that the SNP actually regulates expression of the endothelin 1 gene (EDN1), located 600 kb upstream of PHACTR1 [216]. EDN1 encodes a 21 amino acid peptide that, along with its receptor, promotes vasoconstriction, vascular smooth muscle cell proliferation, extracellular matrix production, and fibrosis; these factors would contribute to the increased risk of coronary artery disease and decreased risk of cervical artery dissection, fibromuscular dysplasia and migraine, conferred by the SNP [216]. This work underlines the importance of functional assays in cellular and animal models in further characterisation of migraine GWAS signals.

In another effort to refine GWAS loci, Hannon et al. applied summary-data-based Mendelian randomization (SMR) to large DNA methylation quantitative trait locus (mQTL) datasets generated from blood and fetal brain to prioritize genes for > 40 complex traits with well-powered GWAS data, including migraine [217]. Using this approach they showed that, with respect to the HEY2-NOCA7 GWAS signal identified by Gormley et al. [53], whole blood and fetal brain have a mQTL profile highly comparable to that of the migraine GWAS, which implicated HEY2 in migraine. These results are consistent with genetic signals influencing DNA methylation in both tissues and migraine, and shows utility of this approach in prioritizing specific genes within genomic regions identified by GWAS [217]. The expansion of resources with gene expression and epigenetic data in tissues relevant to migraine-related pathophysiology will be critical to advancing these types of studies. Recent studies have used gene expression datasets (including single cell analysis) to begin to link genetic loci to their expression in migraine-relevant brain tissues and cell types [218‐220].

There has been some discussion about whether MO and MA are different entities or part of a disease spectrum [221‐223]. Subtype analysis in high-powered GWAS with large samples sizes may reveal whether particular genes may contribute to phenotypic consequences. Most of the migraine loci identified by Gormley et al., (2016) were implicated in both MO and MA, although seven genomic loci (near TSPAN2, TRPM8, PHACTR1, FHL5, ASTN2, near FGF6 and LRP1) were significantly associated with the MO subtype [53]. None were significant for MA, likely reflecting the smaller sample size. Some genetic loci may be selectively associated with particular features (e.g. pain character, duration, frequency, nausea, photophobia and triggers) of the migraine attack [224, 225]. Menstrual migraine affects a subset of female MO sufferers; replication of migraine GWAS loci in a menstrual migraine case-control cohort suggested a particular role for NRP1 in this subgroup [226]. However, the small sample sizes often make it difficult to obtain robust associations for such specific phenotypes. Nevertheless, it will be interesting to identify genes that might be involved in specific aspects of migraine.

A wider view is also informative and can be used to explore the etiology of related and comorbid traits. A GWAS of broadly defined headache using the UK Biobank data found significant associations at 28 loci, of which 14 overlapped with migraine, including the rs11172113 in the LRP1 as the top SNP [227]. Some migraine-associated genes and SNPs have more systemic effects and are involved in a wide range of disorders. A large analysis of shared heritability between common brain disorders found that while most psychiatric and neurologic disorders share relatively little common genetic risk, suggesting largely independent etiological pathways, migraine appears to share some genetic architecture with psychiatric disorders, including attention deficit hyperactivity disorder (ADHD), Tourette’s syndrome, and major depressive disorder [228]. This, together with genetic correlations with other neurological (epilepsy) and vascular disorders (stroke, coronary artery disease), is consistent with comorbidities that have been documented for migraine and suggests they are underpinned by shared genetic factors [228‐233]. Similarly, the monogenic migraine disorders show comorbidity with epilepsy, depression, vascular and sleep disorders [54, 145, 234, 235]. Understanding these relationships can impact the management and treatment of conditions with overlapping etiologies [235, 236].

As the large migraine GWAS have been performed in predominantly Caucasian populations of European heritage, questions remain as to whether the genes and SNPs identified are relevant to other ethnicities, and if there are population-specific genes and polymorphisms. One way to address the former is to test whether there is replication of association of the GWAS SNPs in a particular population. A number of studies have taken this approach, both in specific European cohorts, as well as North Indian and Han Chinese. For example, association of the minor C allele for the PRDM16 polymorphism rs2651899 was replicated in Swedish [237], Spanish [238] and Han Chinese cohorts [239, 240], while rs2651899 and LRP1 rs11172113 showed a protective effect on migraine susceptibility in a North Indian population [241]. Polymorphisms rs4379368 (Succinyl-CoA:Glutarate-CoA Transferase gene locus, C7orf10) and rs13208321 (FHL5) showed some replication in a cohort of the Chinese She people [242]. However, GWAS conducted in specific ethnic populations will determine whether the genetic contributions to migraine vary, and identify migraine susceptibility loci which may be particular to different groups. While still limited, and with relatively small sample sizes, GWAS have been performed in Norfolk Islander, Taiwanese Han Chinese and African American pediatric cohorts [243‐245]. The Norfolk Island genetic isolate is a unique admixed Polynesian-Caucasian population with a high prevalence of migraine (25%). A GWAS for migraine revealed a number of loci of suggestive significance near neurotransmitter-related genes [245]. A GWAS in Taiwanese Han Chinese identified two novel migraine susceptibility SNPs: rs655484 in DLG2, a gene involved in glutamatergic neurotransmission; and rs3781545 in GFRA1, which encodes a receptor for glial cell line-derived neurotrophic factor (GDNF) in trigeminal neurons [243]. The GWAS in American African children found association of migraine with SNPs, including rs72793414, which were strongly correlated with the mRNA expression levels of NMUR2, encoding the G protein-coupled receptor of the CNS neuropeptide neuromedin-U [244].