Genetic Diversity and Expanded Phenotypes in Dystonia: Insights from Large-Scale Exome Sequencing

Study population

A total of 1,967 samples were included in the exome sequencing study, comprising 1,950 patients with dystonia and 17 unaffected family members (parents). One sample failed sequencing, and 25 samples were removed after relationship analysis (Supplementary Methods) revealed that those sample IDs referred to identical patients recruited twice from different sites. Altogether, this yielded a total number of 1,924 patients with dystonia without a previous genetic diagnosis who were included in the final cohort.

The vast majority of the samples were recruited in the framework of two large dystonia registries, DysTract (https://www.isms.uni-luebeck.de/en/research/dystract/) and the Dystonia Coalition (https://www.dystoniacoalition.org/). These samples had undergone genetic prescreening. Prescreening comprised: 1) Hot spot screening for known pathogenic variants in dystonia genes in about 80% of the samples using the Global Screening Array (GSA V01, Illumina; with custom content) and 2) Gene panel sequencing in about a third of the samples, comprising the genes ANO3, GCH1, GNAL, KMT2B, PRKRA, SGCE, THAP1, and TOR1A (results published elsewhere^17,18). Patients with negative prescreening, early age at onset, positive family history, or multisite involvement were prioritized when selecting samples from the dystonia registries. Patient samples were collected in movement disorder clinics across Europe, North America, and Australia. Additionally, 72 samples from patients of Asian ethnicity (primarily from Malaysia) were included. The cohort consisted of diverse ethnic backgrounds, including White (n=1,787), Asian (n=72), African American (n=32), Latino (n=18), Ashkenazi Jewish (n=7), American Indian (n=5), and Mixed (n=3) individuals. Patients were phenotyped by movement disorder specialists and secondary causes, including acquired dystonia, were excluded. Written informed consent was obtained from all participants prior to the genetic testing, and the study was approved by the ethics committee at the University of Lübeck (04–180). The patient IDs used in this study cannot be linked to individual patients by anyone outside the research group.

Sequencing and data processing

Genomic DNA was extracted from peripheral blood samples, and exome sequencing was performed at the Competence Centre for Genomic Analysis in Kiel, Germany, using Illumina NovaSeq. Sequencing reads were preprocessed and aligned to the Hg38 reference genome. Variant calling and annotation were conducted using DeepVariant, GLnexus, and VEP, with further details provided in the Supplementary Methods.

Variant filtering

A comprehensive list of dystonia-linked genes was generated by systematically querying the PubMed database (search term: dystoni* [TITLE/ABSTRACT] AND (gene* OR genetic* OR mutation* OR mutated OR varia*) AND “english”[Language]) until April 2024, resulting in 405 distinct genes that have been linked to dystonic symptoms in the literature (Supplementary Table 1). In this study, we utilized this phenotype-driven candidate gene list to identify genetic causes in our dystonia sample.

We filtered out variants with the following criteria: (a) low variant allele frequency (<20%), (b) low sequencing depth (<10 reads), (c) synonymous variants, (d) minor allele frequency >0.05% in the gnomAD population database, (e) variants classified as (likely) benign in ClinVar, and (f) single heterozygous variants in recessive disease genes. For the established dystonia genes that have been assigned a DYT prefix by the MDS Nomenclature Task Force,⁷ the remaining variants underwent detailed individual evaluation, and their pathogenicity was assessed using ACMG standards and guidelines.¹⁵ For genes not classified as established dystonia genes by the Task Force, additional filtering was applied to further narrow down the large number of rare variants: filtering out single heterozygous loss-of-function (LoF) variants in genes with good tolerance for such variants (gnomAD pLI-score = 0), missense variants classified as likely benign by the alpha missense score,¹⁹ and variants with a low CADD Score (<10),²⁰ before individual evaluation by ACMG standards.

Only likely pathogenic or pathogenic variants according to ACMG criteria were considered disease-causing and underwent validation through Sanger sequencing. Candidate variants were tested for segregation whenever the DNA of family members was available.

The relevant literature for each gene was carefully reviewed to evaluate the phenotypic and genetic information associated with each gene. We categorized the phenotype-genotype relationships for all presumably pathogenic variants into four categories: consistent, partially consistent, quite inconsistent, and unclear/unknown. The unclear/unknown category was used when detailed clinical information was unavailable or for dystonia candidate genes without an established gene-phenotype relationship.

Statistical analysis

Statistical analysis and visualization were performed in R version 4.3.2. The analysis utilized base R functions for statistical tests and data manipulation, as well as the pROC package for generating and analysing ROC curves and the ggplot2 package for creating visualizations. For calculating diagnostic yield, patients were grouped based on clinical criteria, including age at onset (AAO: <30 years, 30–50 years, and >50 years), dystonia subtypes (focal, segmental/multifocal, generalized), family history (positive or negative), and the presence of additional features versus isolated dystonia. These groupings, except for AAO, were also used for ROC analysis and statistical comparisons using Fisher’s exact test. AAO was treated as a continuous variable for both the Mann-Whitney U test comparing resolved and unresolved patients and in the ROC curve analysis. This approach was chosen to capture clinically relevant differences. Missing data were handled by excluding cases with NA values during analysis. All statistical tests were two-sided, and a p-value < 0.05 was considered statistically significant.

Episignature analysis for KMT2B variants

For patients with rare variants in KMT2B and a sufficient amount of DNA available (n=45), the functional effect was assessed by analysing the diesease-specific methylation pattern (“episignature”) in peripheral blood, using the Illumina MethylationEPIC BeadChip. The mean of the normalized methylation levels (mean(z)) and the coefficient of variation (CV = SD/ |mean|) were used as quantifiers (Supplementary Methods).

Patient characteristics and diagnostic yield

After quality control and excluding duplicated samples, the investigated cohort comprised a total of 1,924 dystonia patients without a previous genetic diagnosis, including 1,895 index cases and 29 affected family members. Patients had a median age at examination (AAE) of 54 years (interquartile range (IQR), 43–65), a median age at onset (AAO) of 33 years (IQR, 22.5–43.5), and included 1,148 (59.7%) females and 776 (40.3%) males (Table 1). Of the patients, 1,045 (54.3%) had focal, 549 (28.5%) segmental or multifocal, and 236 (12.3%) generalized dystonia (Table 1). Among the focal dystonia patients, cervical dystonia was the most common subtype, observed in 550 (28.6%) patients. About 93% of all patients had isolated dystonia, while in 130 patients (6.8%), additional neurological (other than tremor) or systemic features were reported, such as bradykinesia (n=41), ataxia (n=35), myoclonus (n=35), or neurodevelopmental features (n=12). A positive family history was reported in 561 index patients (29.6%, 101 unknown).

Exome sequencing revealed likely pathogenic or pathogenic variants in dystonia-linked genes in 163 patients, including 153 out of 1,895 index patients (8.1% diagnostic yield), with an additional 321 index patients (16.9%) having variants of uncertain significance (VUS) in dystonia-linked genes (Figure 1A, 2, and 3, Supplementary Table 2 and 3). Patients with likely pathogenic or pathogenic variants had a median AAO of 19 years (IQR, 5.25–32.75), and included 93 females and 70 males (Table 1); 73 patients had generalized (73/163, 44.8%), 36 multifocal or segmental (22.1%), and 53 focal dystonia (32.5%). For 51 patients (31.3%), additional features were reported, including myoclonus (n=20), neurodevelopmental features (n=12), bradykinesia (n=6), ataxia (n=5), and chorea (n=4). Positive family history was reported in 51/153 index patients (33.3%, 9 unknown). Regarding ethnicity, 140 patients were White (85.9%), 2 African American, 3 American Indian, 10 Asian, and for 8 patients, ethnicity was unknown.

Compared to the unsolved patients and patients with VUS combined, patients with a genetic diagnosis had a significantly earlier AAO (Mann-Whitney U Test, p<0.0001), higher prevalence of generalized dystonia (Fisher’s exact test, p<0.0001), positive family history (Fisher’s exact test, p=0.043), and additional clinical features (Fisher’s exact test, p<0.0001). Notably, there were no significant differences between the patients with VUS and the unsolved patients regarding these characteristics.

The diagnostic yield varied substantially across index patient groups, with the highest yield in patients with additional features (36.3%), followed by those with generalized dystonia (29.3%) and those with an AAO < 30 years (20.4%) (Figure 4a). ROC curves were generated to evaluate the predictive performance of four variables—AAO, positive family history, generalized dystonia, and additional features—along with a combined model incorporating all four variables (Figure 4b). The area under the curve (AUC), which measures predictive accuracy, was 0.68 for AAO, 0.67 for generalized dystonia, 0.62 for additional features, and 0.53 for family history. The combined model, incorporating all four predictors, achieved an AUC of 0.76, indicating the best overall predictive performance, with AAO and generalized dystonia being the strongest individual predictors.

Genetic findings

Among the dystonia index patients with a genetic diagnosis (n=153, 8.1%), 137 distinct variants were identified in 51 genes previously linked to dystonic syndromes (Supplementary Table 2, Figure 2). These variants included 67 missense changes, 24 stop-gain variants, 23 frameshift variants, 17 splice site variants, and six in-frame deletions or insertions (Figure 1C and 2). The majority (132/137, 96.4%) of variants were found in the heterozygous state in genes known to be linked to dominant inheritance. One variant (in MECP2) was found in the hemizygous state in a male patient, associated with X-linked inheritance, and four variants (two in GCH1 and one each in SETX and SPR) were biallelic and linked to autosomal recessive disorders. At least six variants occurred de novo (in ADCY5, GNB1, IR2BPL, KCNN2, KMT2B, and VPS16) (Supplementary Figure 1). Over half of the identified presumably pathogenic variants (77/137; 56.2%) were not previously reported in ClinVar or MDSGene.²¹. Among these, four recurred in at least two unrelated index patients (in EIF2AK2, KCNMA1, SLC2A1, and VPS16; see below), supporting pathogenicity. Details about the variants, pathogenicity scoring, and clinical information can be found in Supplementary Table 2.

A wide spectrum of genetic causes was identified, encompassing established dystonia genes as defined by the MDS Nomenclature Task Force in 103 out of 163 diagnoses, movement disorder or developmental delay genes that frequently include dystonia as a phenotype (9/163 and 4/163 diagnoses, respectively), genes typically linked to other neurological disorders in which dystonia has been rarely reported (41/163 diagnoses), and dystonia candidate genes (7/163 diagnoses) (Supplementary Table 2). Genotype-phenotype relationships were consistent with the literature in most patients (111/163; 68.1%), partially consistent in 14 (8.6%), quite inconsistent in 25 (15.3%), and unclear/unknown in 14 (8.6%) (Supplementary Table 2). Notably, all variants in established dystonia genes demonstrated at least partial genotype-phenotype consistency. Meanwhile, cases classified as “quite inconsistent” were mostly focal (n=15) or segmental/multifocal (n=8) dystonia and involved variants in genes typically associated with other disorders where dystonia is rarely reported, such as C19orf12, GRN, KIF1A, and PRKCG.

Genotype-phenotype relationships and uncommon findings

The genetic causes identified in this study are diverse, as evidenced by the detection of presumably pathogenic variants in 51 genes, highlighting the significant genetic heterogeneity observed in dystonia. Findings encompassed variants in established dystonia genes as defined by the MDS Nomenclature Task Force (62.6% of diagnoses), movement disorder genes or neurodevelopmental delay genes that frequently include dystonia as a phenotype (5.5% and 2.5%), genes typically linked to other neurological disorders where dystonia has been rarely reported (25.2%), and dystonia candidate genes (4.3%).

For the majority of variants found in isolated dystonia genes, the patients’ phenotypes generally aligned with the phenotypic spectrum described in the literature for the respective genes, thus, the genotype-phenotype relationships were classified as consistent (Supplementary Table 2). However, some unusual findings were also observed.

One patient with DYT-EIF2AK2 featured bradykinesia and rigidity, symptoms not previously reported in this form.^12,21 A patient with DYT-ANO3 presented with spasticity and parkinsonism, features described only twice in the literature,^11,21 suggesting they may be recurrent characteristics. Bradykinesia and resting tremor were observed in a DYT-VPS16 patient; while resting tremor has been previously reported in one patient,²⁵ presence of bradykinesia is novel. Additionally, neurodevelopmental delay was noted in another DYT-VPS16 patient, which had not been previously described. Furthermore, chorea was observed in a DYT-VPS16 patient, a feature recently reported in a single patient,²⁶ highlighting that additional features are not uncommon in this genetic form despite isolated dystonia cases predominating. We also identified two DYT-VPS16 patients with unusually late ages at onset (in their 60s or 70s). The reported onset of DYT-VPS16 ranges from 3 to 50 years.²¹ The occurrence of both very early-onset and these exceptionally late-onset cases, both caused by truncating mutations, suggests that additional genetic or environmental factors contribute to the variability in severity observed in DYT-VPS16. Of note, reduced penetrance, which can be regarded as extremely late age at onset, has also been reported for DYT-VPS16.²⁷

Findings in combined dystonia genes encompassed ATP1A3, GCH1, GNAO1, and SGCE, generally aligning with the phenotypic spectrum described in the literature for each gene. Unusual presentations included a DYT/PARK-GCH1 patient with macrocephaly in addition to dopa-responsive dystonia. However, the possibility that the unusual additional phenotype is caused by another genetic variant (dual molecular diagnosis) cannot currently be excluded. Among the 18 DYT/MYC-SGCE patients, 13 exhibited typical myoclonus-dystonia phenotypes, whereas in the remaining five, myoclonus was either absent or not clearly documented. SGCE variants can also manifest as isolated dystonia, albeit rarely.²¹ GNAO1 variants are typically linked to severe combined chorea-dystonia, but recent studies have identified splice site variants and haploinsufficiency linked to milder phenotypes and isolated dystonia.^28,29 Interestingly, in our study, a patient with a novel splice site variant presented solely with adolescence-onset cervical dystonia, consistent with these recent observations.

Variants in ATP1A3 show a broad phenotypic spectrum, complicating genotype-phenotype evaluations. Two of our five patients experienced a rapid onset of dystonic symptoms in adolescence, a known presentation for ATP1A3, although typically accompanied by parkinsonism.¹³ One of these patients also had deafness, often associated with ATP1A3 in the context of CAPOS syndrome (Cerebellar ataxia, areflexia, pes cavus, optic atrophy, and sensorineural hearing loss).³⁰ Notably, this patient carried the same variant (p.Glu818Lys) previously described in CAPOS. Four of the five patients had isolated dystonia. This includes a patient with hemidystonia and the known pathogenic variant p.Thr613Met, which has only been associated with rapid-onset dystonia-parkinsonism until now.³¹ These findings suggest that isolated dystonia can be part of the broad phenotypic spectrum of ATP1A3.

Findings in genes linked to dystonia with additional neurological or systemic features included ACTB, IRF2BPL, SPR, and TUBB4A, with phenotypes generally aligning with previous reports. One case involved generalized dystonia and deafness due to a known ACTB variant (p.Arg183Trp), while another patient with a de novo frameshift variant in IRF2BPL (p.Gln167*) exhibited generalized dystonia, learning disability, spasticity, and gait dysfunction. Additionally, a known homozygous missense variant in SPR (p.Arg150Gly) was linked to infancy-onset, dopa-responsive upper limb dystonia, along with depression and an anxiety-related disorder. Although SPR is typically associated with dystonia, parkinsonism, and developmental delay, isolated dystonia has been observed in 3% of cases,³² predominantly affecting the upper limbs, consistent with our patient’s presentation. TUBB4A variants are associated with a broad phenotypic spectrum, including isolated dystonia, typically laryngeal, cervical, or upper limb dystonia, ³³ as seen in three patients in this study.

As new discoveries unfold, understanding the correlations between clinical manifestations and genetic variants has become increasingly complex. Mechanisms such as variable expressivity, incomplete penetrance, and genetic pleiotropy contribute to this complexity. As anticipated, our study revealed instances where the relationship between genotype and phenotype was quite inconsistent (25 out of 163 diagnoses, 15.3%). This included heterozygous variants in the genes AFG3L2, ATL1, C19orf12, CHMP2B, COL4A1, GRN, KCNMA1, KCTD7, KIF1A, PDE10A, PRKCG, PSEN1, SCN8A, SLC2A1, and SLC6A1. For some of these genes, further investigations are warranted to evaluate their true pathogenicity, as they are not backed by enough evidence to postulate a phenotypic expansion. However, some findings are particularly interesting:

Three patients carried variants in the frontotemporal dementia gene GRN (a recurrent missense and a splice site variant) that have all been previously described in frontotemporal dementia patients. In contrast to previous reports, all our patients had adult-onset focal dystonia (cervical or upper limb) without additional features to date (AAE: 25–76 years). Remarkably, previous reports describe phenotypic differences even within the same family,³⁵ and the splice site variant found in our patient was previously reported in a patient with corticobasal syndrome in combination with focal dystonia.³⁴ This suggests that dystonia may occur without or prior to cognitive and behavioral symptoms due to GRN variants.

We found three patients with novel frameshift variants in KCNMA1, including a recurrent variant in two unrelated individuals. KCNMA1 is typically associated with paroxysmal non-kinesigenic dyskinesia, developmental delay, and seizures, with LoF being a known disease mechanism.³⁶ The three patients presented with childhood- or adult-onset focal persistent dystonia (cervical or upper limb) without additional features, except for tremor and left eye ophthalmoplegia in one patient. Although movement disorders are a consistent feature of KCNMA1-related disease, particularly in LoF variants,³⁶ this is the first report of isolated dystonia. Three high-confident LoF variants, including a recurrent one, were identified, suggesting that KCNMA1 may present with isolated dystonia.

We identified four likely pathogenic variants in KIF1A, including two truncating variants, one splice site variant, and one missense variant affecting the kinesin motor domain, where the majority of pathogenic variants are found.³⁷ Although KIF1A is typically associated with spastic paraplegia, the patients in this study presented with isolated upper body dystonia (cranial, cervical, or upper limbs), with two patients exhibiting tremor and one showing proximal muscle atrophy at the shoulders. While dystonia is not uncommon in KIF1A-related diseases, this is the first report of patients without spasticity, potentially expanding the known phenotypic spectrum.

Another interesting gene was CACNA1A, which has been linked to several diseases, including episodic ataxia, cerebellar ataxia, and developmental and epileptic encephalopathy. Recent reports show that dystonia can also be primary and/or generalized in cases with CACNA1A-variants.^38–40 In our study, we identified three patients with likely pathogenic variants in CACNA1A, all of whom had isolated focal or segmental dystonia affecting the cranial and/or cervical muscles, with disease onset in their 40s. This confirms that variants in CACNA1A are also linked to isolated dystonia.

Dystonia candidate genes

We also screened our data for proposed dystonia candidate genes, identifying presumably pathogenic variants in CHD6, KCNN2, KLC1, NR4A2, and ZMYND11, but not in others, e.g., ATP5F1B, ACBD6, CIZ1, SHQ1 and SPTBN1.

Variants in KCNN2 were initially associated with neurodevelopmental disorders, sometimes accompanied by movement disorders.⁴¹ Meanwhile, they were linked to tremulous myoclonus-dystonia,⁴² currently awaiting replication as a myoclonus-dystonia gene.⁷ We here identified two patients with likely pathogenic missense variants. The first patient had infancy-onset generalized dystonia, myoclonus, ataxia, and bradykinesia – symptoms previously reported in KCNN2-related disease⁴¹ - and carried a de novo variant, supporting its pathogenicity and confirming the role of KCNN2 in myoclonus-dystonia.

The second missense variant (p.Ser376Leu) is located in the ion channel domain of the protein, which is highly intolerant to missense variants⁴³ and harbors other pathogenic variants.^41,42 This patient has had cervical dystonia since the late 20s, without additional features reported, which only partially aligns with previous findings and may broaden the phenotypic spectrum of KCNN2 to include later-onset isolated dystonia.

In 2020, two missense variants in CHD6 were linked to infancy-onset generalized dystonia and developmental delay upon exome sequencing.⁴ We also identified a novel frameshift variant in a patient with adolescence-onset generalized dystonia with myoclonus and ataxia. This variant was classified as likely pathogenic due to its absence in control databases and CHD6’s intolerance to LoF variants (pLI-score of 1). The same study also identified a presumably disease-causing missense variant in KLC1 in a patient with childhood-onset generalized dystonia and developmental delay.⁴ In our sample, we found a novel splice site variant in KLC1 in a patient with late-onset cervical dystonia with no reported additional features, as well as five missense VUS in patients with adult-onset upper limb dystonia. Additionally, ZMYND11 was proposed as a cause of childhood-onset focal dystonia and intellectual disability.⁴ In our study, we identified two patients with presumably pathogenic variants in ZMYND11: a single amino acid duplication in a patient with cervical dystonia, tremor, and polyneuropathy of unknown AAO, and a stop-gain variant in a patient with infancy-onset generalized dystonia with no additional features reported.

Although no pathogenic variants were identified in the recently proposed candidate gene SPTBN1, we discovered nine patients with rare missense or in-frame VUS. The affected patients shared adolescence or adult onset (median AAO: 40 years, range: 16–46) focal or segmental dystonia that mainly affected the neck or upper extremities. In contrast, the initial report described a splice site variant in a patient with adolescent-onset segmental dystonia and developmental delay.⁴

Furthermore, we identified a novel, likely pathogenic missense variant in NR4A2, a gene recently associated with levodopa-responsive dystonia and developmental delay.^44,45 Notably, our patient’s presentation was primarily characterized by cervical dystonia with no additional features reported, and information on levodopa responsiveness was unavailable. Additionally, four other patients harbored missense variants in NR4A2 classified as VUS, all located within the ligand-binding domain at locations intolerant to missense changes.⁴³ These patients all presented with adult-onset focal dystonia affecting the neck or upper limbs, with AAO ranging from 27 to 44 years.

It remains to be seen whether variants in CHD6, KLC1, NR4A2, SPTBN1, and ZMYND11 significantly contribute to dystonia pathogenesis and if different types of variants (e.g., missense vs. truncating) are linked to varying clinical manifestations, as observed in other disorders. Our data suggest a possible role of these genes in dystonia without neurodevelopmental features.