Background
Migraine is a highly prevalent and disabling neurological disorder, which is comorbid with a variety of neuropsychiatric disorders, including an intriguing sensorimotor disease-restless legs syndrome (RLS) [
1,
2]. Restless legs syndrome (RLS) is an intriguing sensorimotor disorder characterized by an urge to move legs, which occurs mostly at night and disturbs sleep, being exacerbated by lying down with unpleasant sensations in legs, and can be temporarily relieved by voluntary leg movements [
3]. Evidence has suggested complex associations between migraine and RLS. The prevalence of RLS in patients with migraine [
1] could be up to seven times higher than that in the general population [
4]. The severity of RLS in patients with migraine is worse than that of non-migraineurs [
5], and the occurrence of RLS is more frequent in chronic compared with episodic migraineurs [
6]. Moreover, RLS and migraine were found to have bidirectional trigger effects [
7]. Yet, detailed mechanisms underlying comorbid RLS in migraineurs are unclear.
Both migraine and RLS are known to have high heritability, and genome-wide association studies (GWASs) have made substantial progress in identifying susceptibility genes for both diseases [
8‐
17]. Dysfunctional dopaminergic neurotransmission and iron homeostasis have been proposed to be common mechanisms shared by RLS [
18‐
20] and migraine [
21,
22]; however, the genetic constituents contributing to RLS in migraineurs remain to be explored. We previously identified that a single-nucleotide polymorphism (SNP) rs2300478 at
MEIS1, the gene responsible for iron homeostasis [
23], increased the risk of RLS by 1.42-fold in migraine subjects via a candidate gene approach [
24]. A recent small-scaled GWAS also suggested additional genes may contribute to RLS in migraineurs [
25]. To further decipher the role of genetic variants in RLS in patients with migraine, we implemented a two-stage GWAS followed by in vivo functional analyses with zebrafish [
26‐
28].
Methods
Study participants and data collection
A two-stage case–control GWAS was implemented to identify susceptible genes for RLS in migraineurs by comparing the cases (i.e., migraineurs with RLS) with controls (i.e., migraineurs without RLS). The significant findings of the discovery cohort were validated in the replication cohort, and a combined analysis of both cohorts was employed to examine the significance of the validated SNPs. In addition, we also examined the significant SNPs in an independent normal control cohort unaffected with restless legs syndrome or migraine. Consecutive patients with migraine were enrolled in the headache clinic of Taipei Veterans General Hospital (TVGH). They filled out a structured questionnaire with questions regarding personal information, medical history, and headache history. Participants were interviewed and their questionnaires and medical records were reviewed simultaneously by board-certified neurologists specialized in headache diagnosis. Migraine was diagnosed according to the criteria proposed in the International Classification of Headache Disorders, 3
rd edition [
29]. Subjects with secondary headache disorders except for medication overuse headache were excluded. RLS was diagnosed based on the criteria proposed by the International RLS Study Group [
30]. Subjects with ferritin < 50 ng/ml, anaemia, creatinine > 1.5 mg/dL or pregnancy were eliminated to exclude secondary RLS. Subjects with any RLS symptom proposed in the criteria or periodic limb movements in sleep based on self-reported nocturnal leg jerks during sleep were excluded from the control groups.
Genotyping in the discovery cohort
We genotyped 642,832 SNPs using the Affymetrix Axiom Genome-Wide CHB 1 Array Plate, which has high coverage of genome-wide common variants for Han Chinese. SNP genotypes were called using the Axiom GT1 algorithm. Quality control (QC) criteria were applied to exclude SNPs if they (a) were monomorphic in both cases and controls, (b) had a total call rate of less than 95%, (c) had a minor allele frequency of less than 5% and a total call rate of less than 99%, or (d) showed significant (P < 1 × 10–8) deviation from Hardy–Weinberg equilibrium in controls. For sample filtering, arrays with generated genotypes for < 95% of the loci were excluded.
Heterozygosity of SNPs on the X-chromosome was used to verify the sex of the samples. PLINK version 1.09 [
31] was used to identify samples with genetic relatedness, indicating that they were from the same individual (or monozygotic twins) or from first-, second- or third-degree relatives. These determinations were made based on evidence for cryptic relatedness from identity-by-descent status (pi-hat cut-off of 0.125).
Genotyping in the replication cohorts
We selected SNPs that were within 200 kb of a gene which contains at least two adjacent SNPs with a P value of < 1 × 10–4. Single SNPs with a trend P value < 1 × 10–4 but not within 200 kb of a gene were not chosen for replication because we aimed to explore known protein coding genes. Genotyping was performed in replication cohorts using the Sequenom MassARRAY iPLEX platform (Sequenom Inc., San Diego, CA, USA). Genotyping in both cohorts are services provided by the National Center for Genome Medicine (NCGM).
Imputation for the discovery case–control GWAS
We conducted a genotype imputation analysis in the discovery cohort using the 1000 Genomes Phase 3 reference data by implementing IMPUTE2 [
32]. Well-imputed SNPs (info score > 0.4) were retained followed by systematic QC as described above.
Morpholino translational knockdown
Morpholino oligonucleotide can block translation by targeting the 5’ untranslated region (UTR) of mRNA or inhibit RNA splicing by targeting exon/intron junctions. We designed six 25-base morpholinos (Gene Tools, Philomath, OR) that target the 5’UTR or splicing junction of
ccdc141 and
vstm2l (Additional file
1).
CRISPR interference
CRISPR gRNAs were designed with Benchling and the cloning sequences are shown in Additional file
2. Oligonucleotides were annealed in a thermoblock at 95 °C for 5 min and cooled to room temperature. Annealed oligonucleotides were cloned into pT7-gRNA plasmid at BsmBI site and verified by sequencing. To make dCas9 mRNA, dead Cas9 plasmid [
33] was linearized by XbaI enzyme and purified by Gel extraction kit (Qiagen, Hilden, Germany). mRNA was synthesized by mMESSAGE mMACHINE T3 kit (Life Technologies, Carlsbad, CA) and purified by RNeasy mini kit (Qiagen). To make gRNA mRNA, pT7-gRNA plasmid was linearized by BamHI enzyme and purified by Gel extraction kit. RNA probe was synthesized by in vitro transcription using a MEGAscript® T7 Transcription kit (Thermo Fisher Scientific, Waltham, MA) and purified by ethanol precipitation.
Transient and stable CRISPR/Cas9 knockout (KO)
The CRISPR/Cas9 KO is carried out by a non-for-profit service offered by the Taiwan Zebrafish Technology and Resource Center (TZTRC) according to previous reports. Briefly, together with the common tracrRNA and Cas9 protein, 4 gene-specific crRNAs (Additional file
3; Horizon, Waterbeach, UK), 2 for each gene, were injected into one-cell stage embryos separately [
34]. Transient CRISPR/Cas9-injected embryos (crispants) have been demonstrated to largely phenocopy mutants [
35]. The CRISPR/Cas9 activity detection and mutation screening were performed by high resolution melting analysis [
36]. The stable KOs were confirmed by Sanger sequencing and maintained according to the standard operating protocol [
37].
Tyrosine hydroxylase RNA in situ hybridization
Tyrosine hydroxylase (TH) is an enzyme responsible for the biosynthesis of dopamine precursors. The 3–5 dpf wild-type and injected embryos were used for in situ hybridization following previously established protocol [
38]. The embryos were fixed in 4% fresh-made paraformaldehyde at 4 °C overnight and then treated with 3% H
2O
2 and 5% KOH for depigmentation. Embryos were washed and transferred into 100% methanol at -20 °C overnight. Digoxigenin-labelled antisense RNA probes were used for labelling to detect the distribution of dopaminergic cells, and then the embryos were mounted in glycerol for observation and photography.
Fin movement observation
We utilized a video system under normal laboratory lighting to observe pectoral fin movement and evaluate whether the injected embryos had hyperkinetic movements mimicking the “restlessness” and “urge to move the limbs” in patients with RLS. The 5 dpf embryos were used because pectoral fins and body organs are relatively well-developed. Embryos were mounted on glass slides covered with 1% low melting agar and put under a dissecting microscope to observe fin movements. One-to-three-minute videos were filmed by DFK 23UP031 USB Camera (The Imaging Source Asia Co., Taipei, Taiwan). Video Analysis Tools, After Effects and Tracker (Adobe, San Jose, CA), were used. The average flapping frequency (times/second) was acquired by catching the fin movement in x and time in y coordinates.
Quantitative RT-PCR (qRT-PCR)
Dechorionated 2 dpf embryos were collected and total RNA was extracted by RNAzol® RT reagent (Molecular Research Center, Inc.). cDNA was synthesized by SuperScript™ III Reverse Transcriptase kit (Thermo Fisher Scientific). The experiment was conducted by LightCycler® 480 Instrument II with SensiFAST™ SYBR® Hi-ROX kit (Bioline). Actin was used as an internal control in all triplicated experiments. The qPCR data was analysed by LightCycler® 480 software version 1.5.0.39.
Statistics
Association analyses were carried out by comparing allele/genotype frequencies between cases and controls using a single-point method: Cochran–Armitage trend test. The distribution of expected
P values under the null hypothesis and genomic inflation value (λ) were calculated. The Manhattan and quantile–quantile (Q-Q) plots were created using the R package [
39]. Genetic analyses were conducted using PLINK (version 1.09) [
31]. Detection of possible population stratification was carried out by using principal component analysis (PCA) implemented in EIGENSTRAT to infer continuous axes of genetic variation. We adjusted for potential genetic heterogeneity by incorporating the first 10 PCs in the logistic regression tests of association with RLS. Joint analysis was conducted by combining data from the discovery and replication samples. In addition, we also examined the association of significant variants with migraine in an independent migraine case–control cohort. For studies involving zebrafish, data are reported as the mean ± SD or median and interquartile range. Student’s t test was used for comparison of continuous variables; Mann–Whitney U test was used for comparisons of unpaired nonparametric variables. All calculated
P-values were two-tailed, and statistical significance was defined as
P-value less than 0.05. These analyses were performed using Graphpad Prism, version 7.00 (GraphPad Software, La Jolla, CA).
Discussion
By using a two-stage GWAS, we identified two novel susceptibility genes, VSTM2L and CCDC141, accountable for an increased risk of RLS in patients with migraine. These two genes were highly expressed in the central nervous system (CNS) among species. Inhibiting expression of these two genes at the transcriptional or translational level resulted in morphological changes involving fin development, decreased number of dopaminergic neurons, and hyperkinetic movements of pectoral fins in zebrafish, compatible with the clinical symptoms and putative pathogenic pathways of RLS. Gene rescue reversed the phenotypes of the morphants, which further supports that these findings are not due to non-specific toxic effects from morpholino and augmented the functional roles of these two genes in RLS pathogenesis.
Our data confirmed the crucial role of
VSTM2L and
CCDC141 in RLS in patients with migraine; however, pre-existing information regarding these two genes is scarce.
VSTM2L, short for V-set and transmembrane domain containing 2 like, was previously known as C20orf102. The protein encoded by
VSTM2L has an exquisitely CNS-specific expression and is known to be a secreted antagonist of a neuroprotective mitochondrial peptide Humanin [
40].
CCDC141 (short for coiled-coil domain containing 141), also named CAMDI after coiled-coil protein associated with myosin II and DISC1 (disrupted in schizophrenia 1), is known to affect neuronal development by impairing radial migration through DISC1 and myosin II-mediated centrosome positioning [
41]. How these known functions of
VSTM2L and
CCDC141 contribute to RLS is unclear, but our data indicate that it might be mediated through affecting the development and distribution of dopaminergic neurons. The A11 dopaminergic nucleus of the dorsal-posterior hypothalamus has been considered to be important in the pathogenesis of RLS [
19] and migraine [
21] in rodent models. In zebrafish, we also demonstrated that inhibition of the expression of
vstm2l and
ccdc141 could affect the distribution of dopaminergic cells in the CNS. Though the
th expression level of DC2,4–6 (A-11 type, the rodent A11 equivalent) [
43] did not change, that of DC7, which is considered as caudal hypothalamus, did decline (Fig.
3A). Of note, the distribution of DC2,4–6 neurons seem dispersed in morphants. Nevertheless, we could not exclude the possibility that it was due to morphological changes. Interestingly, the
th expression of A11-type dopaminergic neurons, LC and MeO neurons with far-ranging projections is not affected, while that of DC7 neurons and retinal amacrine cells projecting exclusively locally or to adjacent brain regions is decreased [
42]. Evolutionarily, there is no direct zebrafish counterpart of mammalian substantia nigra/ventral tegmental area dopaminergic neurons. A trans-species comparison of the A11-type and other dopaminergic systems, which are also less well studied in mammals [
43], and behavioral phenotypes need to be examined.
Previous GWASs have identified six RLS risk loci (
MEIS1, BTBD9, MAP2K5, PTPRD, TOX3, and an intergenic region on chromosome 2p14) [
14‐
17]; however, only
MEIS1 has been found to be associated with RLS in patients with migraine via candidate gene approach [
24]. Hence, susceptibility genes for RLS in migraineurs might not be completely the same as those for RLS in general population. None of the above genes were identified associated with risks of RLS in migraineurs in this study. Whether
CCDC141 and
VSTM2L also contribute to the risk of RLS in general population remains to be explored.
We have used translational knockdowns (morphants), transcriptional knockdowns, and transient knockouts (crispants) in the zebrafish system to examine the functional relationship of
CCDC141 and
VSTM2L to the symptoms of RLS and migraine and obtained relatively consistent results. The stable
ccdc141 and
vstm2l KO lines did not show a decrease in
th-expressing cells or a hyperkinetic movement in pectoral fin and basically behaved like wildtype embryos. Though unexpectedly, some similar cases have been reported in zebrafish, such as
egfl7 and
slc25a46 [
33,
48]. The mechanism of genetic compensation for
egfl7 has been shown to be transcriptional adaptation that is triggered by degradation of the mutated mRNA through nonsense-mediated mRNA decay (NMD) to upregulate sequence-similar genes that thereby enable functional compensation [
49,
50]. However, the mechanism for
slc25a46 is currently unknown [
48]. The expression of
ccdc141 and
vstm2l in corresponding KO mutants is decreased (Additional file
12), suggesting a transcriptional adaptation caused by NMD [
48,
50]. To overcome the genetic compensation and examine the phenotypes in adult animals, different animal models may help. For example, various mouse
Slc25a46 mutants exhibit a spectrum of disorders similar to those in patients with recessive loss of
SLC25A46 function [
51‐
53].
Our study has several implications. First, although the true biological significance of the genes identified from GWAS for complex disorders is often questioned, our findings provide evidence to support the functional roles of the identified genes which is consistent with the prevailing theories of RLS pathogenesis. Of note, the function of
CCDC141 and VSTM2L has not been fully elucidated. Further studies for these two genes might provide novel mechanisms of RLS, particularly in patients with migraine. Second, only one previous study had employed zebrafish to evaluate the function of
Meis1 gene; however, the study investigated only hindbrain development [
54], without phenotypic studies to simulate RLS. Our study further demonstrated the utility of zebrafish to model the behavioural phenotypes of RLS in humans. Spreading depression (or depolarization) (SD) could be used as a preclinical model for migraine study, particularly migraine with aura [
55]. A recent paper has established the method to measure SD in the adult zebrafish tectum [
56]; therefore, it can be used to examine the “migraine-like” phenotype in the corresponding adult zebrafish mutants in the future. With accurate diagnoses and strict criteria for the patient recruitment, we obtained significant signals with a limited sample size. However, only common variants were included from the GWAS results in this study. Further investigations are required to look at rare variants with fine mappings. Moreover, we focused on SNPs located in or near a gene in the replication analysis for reasons stated in Methods. The possibility that SNPs not mapped to a gene have roles in pathogenesis remains to be examined. Finally, our findings provide biological insights on the ample clinical evidence supporting the RLS-migraine comorbidity, which may support the implement of a detailed questionnaire about sleep disorder and restless legs symptoms in patients with frequent migraine in clinical practice. For those with symptoms with RLS, testing for iron, ferritin or other secondary causes of RLS may be mandatory. Moreover, it may be appropriate to treat RLS with dopaminergic D2 agonist in patients with migraine, which may be beneficial for both RLS symptoms and migraine in these patients [
57].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.