Abstract
Background and objectives Dystonia is a genetically complex disease with both monogenic and polygenic causes. For the latter, numerous genetic associations studies have been performed with largely inconsistent results. The aim of this study was to perform a field synopsis including systematic meta-analyses of genetic association studies in isolated dystonia
Methods For the field synopsis we systematically screened and scrutinized the published literature using NCBI’s PubMed database. For genetic variants with sufficient information in at least two independent datasets, random-effects meta-analyses were performed, including meta-analyses stratified by ethnic descent and dystonia subtypes.
Results A total of 3,575 articles were identified and scrutinized resulting in the inclusion of 42 independent publications allowing 134 meta-analyses on 45 variants across 17 genes. While our meta-analyses pinpointed several significant association signals with variants in TOR1A, DRD1, and ARSG, no single variant displayed compelling association with dystonia in the available data.
Conclusions Our study provides an up-to-date summary of the status of dystonia genetic association studies. Additional large-scale studies are needed to better understand the genetic causes of isolated dystonia.
Introduction
Genetically, dystonia is a complex disease with the parallel occurrence of forms with and without causative genetic factors. Over the past decades more than two dozen loci have been proposed as monogenic causes of dystonia[1–3] and 32 confirmed genes (28 with “DYT” labels)[4,5] have been included in the new list of isolated, combined, and complex hereditary dystonia proposed by the International Parkinson and Movement Disorder Society’s (MDS) Task Force on Genetic Nomenclature in Movement Disorders[5]. Substantially less is known about the genetic underpinnings of sporadic (i.e. non-monogenic) dystonia. Along these lines, only two GWAS on dystonia subtypes, i.e. musician’s dystonia[6] and cervical dystonia[7], have been published to date. The remainder of the existing genetics literature on sporadic dystonia is comprised of candidate gene association studies. These typically focused on genes known to cause familial dystonia as well as functionally founded candidates, such as genes involved in dopamine metabolism or the brain-derived neurotrophic factor (BDNF)[8]. To date, no systematic review covering all genetic polymorphisms investigated in dystonia has been published and only very few meta-analyses utilizing genetic association data exist[9–12]. As a result, it is becoming increasingly difficult to evaluate and interpret the genetics literature pertaining to sporadic dystonia. The aim of this work was to overcome this limitation and perform the first systematic synopsis - including meta-analyses of all available genetic association data - in the dystonia field. To this end, we carefully screened more than 3,500 articles and performed a total of 134 meta-analyses across 52 case-control datasets from 42 independent publications, thereby vastly increasing the number of meta-analyses currently available for dystonia[9–12]. The accrued literature database and meta-analysis results are made available in the supplement of this manuscript.
Methods
1. Literature searches
1.1. Systematic literature searches
Overall, our study followed the approach developed earlier by our group for systematic field synopses in Alzheimer’s disease[13] and Parkinson’s disease[14]. At the outset, this entailed a systematic literature search using NCBI’s “PubMed” database for papers published until January 23, 2018. In a first stage, we searched for keywords “(dystoni* OR (writer* AND cramp) OR graphospasm OR blepharospasm OR torticollis OR meige OR spasmodic OR dysphonia OR retrocollis OR antecollis OR laterocollis OR (musician* AND cramp) OR (occupation* AND cramp) OR (golf* AND cramp) OR yips) AND associat* AND gene*”. In the second stage, we separately searched for each gene identified in stage 1 and for 28 loci with a “DYT” label (Figure 1; DYT designations taken from refs. 5 and 15; see Supplementary Tables S1 and S2 for more details). Of note, prior to the revision of the dystonia classification and nomenclature in 2013, isolated dystonia (i.e. pure dystonia with or without dystonic tremor) was referred to as “primary” dystonia, a term mixing phenomenological and etiological features, whose use is no longer recommended[16]. As the clinical part of the definition of “primary” dystonia was identical to that of “isolated” dystonia, we exclusively use the new and unambiguous term “isolated” dystonia for all studies referred to in this article.
Flowchart of literature search, data extraction, and analysis strategies applied for the dystonia field synopsis. † Search for query terms “(dystoni* OR (writer* AND cramp) OR graphospasm OR blepharospasm OR torticollis OR meige OR spasmodic OR dysphonia OR retrocollis OR antecollis OR laterocollis OR (musician* AND cramp) OR (occupation* AND cramp) OR (golf* AND cramp) OR yips) AND associat* AND gene*”. ‡ Search for query terms outlined above and “AND ([gene name] OR [gene alias(es)])” using HGNC (HUGO Gene Nomenclature Committee at the European Bioinformatics Institute) nomenclature27. ¶ Only 45 out of 233 polymorphisms were investigated in at least two independent data sets. Furthermore, two studies were excluded due to sample overlap: 1. Djarmati et al, 2009 (ref. 19) fully excluded due to overlap with Lohman et al, 2012 (ref. 21), 2. Newman et al, 2012 (ref. 18) exclusion of data for polymorphisms overlapping with Newman et al, 2014 (ref. 20; see Table S2 for more details).
1.2. Inclusion / exclusion criteria
We only considered publications representing association studies between biallelic genetic polymorphisms (minor allele frequency in included data sets or in the general population ≥0.01 based on ref. 17) and isolated dystonia phenotypes. Additional eligibility criteria were: publication in a peer-reviewed journal, publication in English, and analysis in at least 10 dystonia cases and 10 control subjects. We excluded studies assessing dystonia with a known etiology (previously known as “secondary” dystonia), studies using family-based designs, and studies on mitochondrial DNA. Furthermore, we excluded studies that contrasted their case groups against “database controls” and those with cases carrying a known dystonia mutation.
2. Database
From each eligible article, we extracted a range of informative variables into a Microsoft Excel database (full database in Supplementary Table S2). The initial data entry was performed by O.O. followed by double checking of all variant IDs and genotype distributions by V.D.
3. Statistical analyses
For all eligible datasets, we calculated odds ratios (OR) and corresponding 95% confidence intervals (CIs), assuming an additive genetic model. For datasets with allele frequency information only, we calculated the corresponding genotype distributions assuming Hardy-Weinberg equilibrium. For four variants, the same or largely overlapping data sets were published in separate articles[18–21]. In these instances, we only included the genotype distributions for the larger of the two datasets, i.e. refs. 20 and 21. For variants with independent genetic data in at least two non-overlapping datasets we calculated random effects summary ORs using PLINK 1.9[22,23]. Whenever sufficient data were available, meta-analyses were stratified by ethnic descent group in addition to calculating summary ORs across all ethnic groups combined. In addition to calculating meta-analyses across all forms of isolated dystonia, we divided our analyses into focal dystonia (i.e. cervical dystonia, blepharospasm, musician’s dystonia, writer’s cramp, and spasmodic dysphonia), segmental dystonia, and generalized dystonia. Whenever possible, subgroups were further stratified into groups of differing ethnic descent. Overall, this led to 134 meta-analyses of 45 variants across 17 genes. Pairwise linkage disequilibrium estimates across SNPs within the same ±1Mb interval in the 1000GP (phase 3) data[17] were calculated using PLINK 1.9 (Supplementary Table S3). This identified 32 independent (r2<0.3) variants; accordingly, the study-wide significance threshold was set to α = 0.00156 (=0.05/32). Heterogeneity across studies was assessed by calculating the I2 metric (I2 >75% indicates excess heterogeneity[14]).
Results
1. Literature search
The systematic literature search (up-to-date until January 23rd, 2018) identified 3,575 potentially eligible articles (Figure 1). After systematic review of titles, abstracts and full text (as needed) versions of these papers we identified 42 eligible publications (describing association data in 52 independent datasets. Analyzed datasets originated from a total of 17 different countries spread across four continents. In total, 233 polymorphisms across 33 loci were investigated in these 42 publications. Most (n=40) used a candidate gene approach, while two papers were GWAS. The entire database of this field synopsis is provided in Supplementary Table 2.
2. Meta-analysis results
For the meta-analyses, we only considered variants with at least two independent non-overlapping datasets with available genotype or allele data. This included screening 557,621 SNPs from a musician’s dystonia GWAS dataset[6] for an overlap with the 233 polymorphisms identified in the literature screen. Overall, this procedure resulted in sufficient data for a total of 45 variants across 17 genetic loci (median sample size per data set: 458, range 88-5,385). Of these, 42 variants could be analysed after stratification in for datasets of Caucasian descent, and 6 after stratification for Asian descent. Only five variants in TOR1A, DRD1 and ARSG showed nominally significant evidence for association: i.e. rs4532 (OR [95 % CI]: 1.37 [1.131.67], P=0.00153), rs35153737 (OR [95% CI]: 1.46 [1.14-1.88], P=0.00315), rs13283584 (OR [95% CI]: 1.32 [1.08-1.61], P=0.00584), rs7342975 (OR [95 % CI]: 2.00 [1.15-3.47], P=0.00572), rs9972951 (OR [95 % CI]: 2.16 [1.06-4.40], P=0.0342), but only variant rs4532 in DRD1 survived correction for multiple testing. Meta-analyses in subgroups stratified by ethnic descent did not reveal any additional associations (Table 1).
Results of random-effects meta-analyses based on an additive model in all types of isolated (a.k.a. “primary”) dystonia.
Meta-analyses divided by diagnostic subgroup are summarized in Table 2 and yielded five nominally significant results, two of which were not observed in the analyses without diagnostic stratification, i.e. rs1801968 in TOR1A (OR [95% C.I.]: 3.10 [1.25-7.68], P=0.0142), associated with writer’s cramp, and rs11655081 in ARSG (OR [95% C.I.]: 4.42 [2.72-7.19], P=2.11e-09, respectively) associated with musician’s dystonia (Table 2). While the latter finding surpassed the threshold of study- and genome-wide significance, we note that this meta-analysis is only based on two independent datasets, both of which originate from the same (and only) paper on the topic[6]. The same variant was also assessed in independent datasets with other focal (i.e. cervical, blepharospasm; Table 2B, D) and segmental (Table 2G) forms of dystonia but did not show any evidence for association with these dystonia subtypes[6].
Results of random-effects meta-analyses based on an additive model in diagnostic subgroups of isolated dystonia: a) all types of focal dystonia, b) musician’s dystonia, c) blepharospasm, d) cervical dystonia, e) spasmodic dystonia, f) writer’s dystonia, g) segmental dystonia, h) generalized dystonia.
Discussion
This work represents the first systematic field synopsis of genetic association studies in dystonia. Overall, we present the results of 134 meta-analyses on 45 variants across 17 genes. While nominal associations with several loci were observed, only variants in DRD1 and ARSG survived multiple testing correction. However, even these two most significant results should be interpreted with caution owing to comparatively small sample size.
Prior to this study meta-analyses on genetic association data in dystonia were only available for two genes, i.e. TOR1A[9,10] and BDNF[11,12], both with conflicting results, i.e. one study interpreted their results in favor of a role in dystonia susceptibility, while the other study reached an opposite conclusion. Although our results provide some degree of support for an involvement of TOR1A SNP rs35153737 in Asians and rs13283584 in Caucasians (Table 1), no association evidence was observed with the only available (but widely tested) SNP (rs6265) in BDNF.
One notable observation and important limitation of our study is that most meta-analyses are based on relatively small sample sizes (median combined sample size of across all metaanalyses was only 2,882 individuals, compared to ~4,500 for Alzheimer’s[13] and ~7,500 for Parkinson’s disease[14]). Provided that the genetic liability of dystonia is mostly governed by small effect variants with ORs <1.2, our power to detect such associations was small, and only efforts with substantially increased sample sizes for both patients and controls will have sufficient power to detect such effect sizes[14]. Other limitations of our study relate to the search strategy, which might not have identified some eligible papers, and an erroneous data extraction and representation procedure. However, while such errors are unavoidable, our ample experience in the collection, annotation and curation of genetic association data in the context of systematic field synopses[13,14,24–27] suggest that these errors will likely be infrequent and will not affect the major conclusion(s) of our paper.
Notwithstanding these and possibly other unrecognized limitations, our study represents the first systematic synopsis of genetic association studies in dystonia and the largest collection of meta-analyses in the field. Despite its comprehensiveness, and in contrast to genetic studies on monogenic forms of the disease, it currently provides only little support for an existence of strong genetic risk factors acting in sporadic dystonia. This situation will likely change upon completion of additional large-scale GWAS of sufficient size, which will hopefully provide new insights into the pathogenetic forces underlying the onset and progression of this debilitating disease.
Acknowledgements
This study was supported by by the German Research Foundation (DFG) (FOR2488, main support by subproject “P6” BE2287/6-1 to LB, additional support by subprojects “P7” LI2654/2-1 to CML, “Z1” KL1134/17-1 to CK, and “P4” LO1555/9-1 to KL). CK is the recipient of a career development award from the Hermann and Lilly Schilling Foundation. The authors report no conflict of interest.
Footnotes
Funding: This study was supported by the German Research Foundation (DFG grant FOR2488: Main support by subproject “P6” BE2287/6-1 to LB; additional support by subprojects “P7” LI2654/2-1 to CML, “Z1” KL1134/17-1 to CK, and “P4” LO1555/9-1 to KL).
References
