Abstract
Background
Tourette Disorder (TD), or Tourette Syndrome, is a highly heterogeneous childhood-onset neurodevelopmental disorder with a global prevalence of 0.5%. TD’s large phenotypic heterogeneity, variable heritability patterns, and frequent but varied comorbidity with other neurodevelopmental disorders indicate that different TD patients might have different etiologies.
Objectives
In this study, we performed unsupervised clustering of clinical data, such as categorical diagnoses and comorbidities, from 865 subjects with TD from the Tourette International Collaborative Genetics (TIC Genetics) study to detect Tourette Disorder subtypes.
Methods
We used two different clustering methods, K-Means and Bayesian Hierarchical Clustering to detect phenotypic subtypes.
Results
We identified five distinct, clinically relevant subtypes. These subtypes are characterized by both previously described TD comorbidities, such as Obsessive-Compulsive Disorder (OCD) and Attention Deficit/Hyperactivity Disorder (ADHD), as well as other characteristics, such as sex and region.
Conclusions
Defining clinically relevant TD subtypes could enable better diagnostics, treatment, and the understanding of TD etiology. In addition, stratified analysis of genetic data based on these phenotypic subtypes may help identify genes contributing to each TD subtype and provide insights into the disease variability.
Introduction
Tourette Disorder (TD), also called Tourette Syndrome, is characterized by multiple motor tics and at least one vocal tic, occurring many times a day (usually in bouts) nearly daily or intermittently for at least a year, with an onset before age 18 [1]. Most subjects with TD over the age of 10 report premonitory urges that are relieved by tics [2]. Coprophenomena (such as copropraxia or coprolalia) may occur when tics are most severe [2]. In addition, there are also sex-dependent differences in the phenotypic expression of TD, with males more often showing tic-related behaviors and females more often exhibiting Obsessive-Compulsive Symptoms/Behaviors (OCS/OCB) [3, 4]. Twin and family studies have established that TD bears a strong genetic component with an unclear inheritance pattern, probably due to polygenic cause [5, 6]. Large-scale genetic studies of TD have identified specific classes of genetic variation that contribute to the genetic risk of TD, including common variants [7], segregating variants, rare copy number variants, and de novo variants that impact protein-coding sequences [8–11].
Approximately 88% of TD subjects have one or more other Neuro Developmental Disorders (NDDs) as comorbidities [12], such as Obsessive-Compulsive Disorder (OCD, ~30–50%), Attention Deficit/Hyperactivity Disorder (ADHD, ~50–60%) and to a lesser extent, Autism Spectrum Disorder (ASD, ~10%), as well as anxiety, depression, and other conditions [13]. Given the substantial genetic and phenotypic heterogeneity of TD, identifying distinct subtypes could assist in better diagnosis and treatment, as well as enable a better understanding of genetic architecture through stratified analysis of genetic data.
Several prior studies (reviewed in [14]) have attempted to identify common factors and subtypes of TD. For example, one study examined heritability and polygenic load associated with TD, OCD, and ADHD using symptom-level factors [15]. Latent class analyses of subjects with TD identified three major TD-affected groups: 1) TD + OCS/OCB, 2) TD + OCD, and 3) TD + OCD + ADHD [16]. In two studies of TD comorbid disorder clusters and their heritability, TD + OCD and TD + OCD + ADHD were considered heritable while TD + ADHD was not [17, 18]. In another study of clinical factors of 139 subjects with TD, two subtypes of tic disorders with severe (21.6%) and mild (78.4%) symptoms were identified [19]. A 6-factor exploratory factor analysis of forty-nine motor and phonic tics in 3,494 individuals (1,191 TS probands and 2,303 first-degree relatives) identified social disinhibition as a heritable subtype of TD [20].
Although these studies provided useful information, they had limited ability to comprehensively identify TD subtypes, because many of them were based on cohorts with either limited subsets of available phenotypic information (i.e., comorbidities or symptoms), small sample size, and/or subjects that were demographically homogeneous. In this study, we identified subsets with TD using unsupervised clustering of clinical and demographic data from the Tourette International Collaborative Genetics (TIC Genetics) study, a large international study with over 20 sites across the United States, Europe, Israel, and South Korea [21]. This large, uniformly assessed clinical dataset enabled us to identify five clinically relevant subtypes that will enhance the understanding of the heterogeneity of TD.
Materials and Methods
Clinical Data
The clinical data were collected between 2011 and 2023 as part of the TIC Genetics Study [21]. The Institutional Review Board approved the study protocol at each local site. Informed consent was obtained from all participants (or in the case of minors, from their parents). The clinical assessment methods and definitions of TD, OCD, ADHD, and Trichotillomania have been previously described in detail [21], and are based on the Diagnostic and Statistical Manual of Mental Disorders—Fourth edition, Text Revision (DSM-IV-TR) [22] or Fifth edition (DSM-5) [1]. Clinicians were trained to record all symptoms and perform diagnoses in a consistent manner to ensure quality and reliability across all sites.
Detailed clinical data were acquired from an extensive written questionnaire which was completed by the participant (or the parent in the case of young probands); this information was validated in a semi-structured interview by an experienced clinician and recorded in a standardized fashion. Items included demographic information, details of tic, OCS/OCB/OCD, ADHD, and trichotillomania (including ages of onset), and other potentially relevant conditions. Based on the questionnaire and interview information, the clinician made lifetime diagnoses for 4,435 subjects, including probands with TD, both affected and unaffected parents, and some affected and unaffected siblings [21].
Data Preprocessing
To improve the precision of subtype identification, we excluded other tic disorders, such as other chronic, provisional, and transient forms. Among subjects that had TD, we excluded those that have any missing or “unable to rate” value for any of the Tic, Obsessive Compulsive, or Attention-disorder-Hyperactivity diagnoses. We also excluded subjects with potentially confounding factors flagged by clinicians, including “atypical presentation”, “psychosis”, “other severe neurological condition”, “congenital anomalies”, “genetic syndrome/chromosomal abnormality”, “other significant psychiatric history”, and “other significant medical history”. From each family, we selected only one member diagnosed with TD. 22 centers around the world were grouped into 3 regions: USA, Europe, and Asia. A description of the detailed clinical data variables for each subject included in this study is available in Supplementary Table S1.
Some parameters in our clinical dataset are mutually exclusive. For example, for OCD diagnosis, each subject belongs to one of the following categories: No OC disorder/symptoms, OC Subclinical, OC Symptoms, or OCD. In these cases, these mutually exclusive parameters were combined into a single categorical parameter (Supplementary Table S2). Age of onset for a given disorder was converted into two categories: “Early” (<= 10y) and “Late” (> 10y). In a few cases (<10%) where the age of onset was not available, age at diagnosis was used as a proxy. The age of onset of subjects with no diagnosis of the disorder was coded as “Never”. We used two categories for birth: “single” or “multiple” (>=2). After preprocessing, 14 categorical parameters were selected for clustering (Supplementary Table S2).
Exploratory data analysis and clustering method testing
Characteristics and distribution of individual parameters were analyzed to examine the dataset. Pairwise correlations between parameters were calculated as Cramer’s V statistics using the “CramersV” function in the R package “rcompanion” (version 2.5) [23]. Several clustering methods were tested (Supplementary Table S3), and two unsupervised clustering methods, K-Means and Bayesian Hierarchical Clustering (BHC) [24], were selected to identify TD subtypes. As the statistical methods chosen for our study are designed to handle highly correlated parameters, we retained the correlated parameters to provide additional insight into specific cluster characteristics.
K-Means clustering
K-means clustering is an unsupervised clustering method that partitions n observations into k clusters in which each observation belongs to the cluster with the closest cluster mean. Multiple Correspondence Analysis (MCA) [25] was used to convert categorical parameters into principal components (PCs). This enabled us to: 1. reduce noise by removing less significant PCs with small contribution to the overall variance, and 2. use clustering methods that require continuous variables. K-means clustering was performed using the K-Means function in R using the “stats” package (version 4.4.3). The number of clusters (centers = k) from 2 to 20 were tested. To ensure robust convergence and reproducibility, the algorithm was run with 5,000 iterations and 1,000 random starts.
Euclidean distance was used as the distance metric and the clustering method described by Hartigan and Wong (1979) [26]. The optimal number of clusters was identified based on the silhouette statistics (maximal inter-cluster / intra-cluster mean distance).
Bayesian Hierarchical Clustering (BHC)
The BHC method performs bottom-up hierarchical clustering, using a Dirichlet Process (infinite mixture) to model uncertainty in the data and Bayesian model selection, to identify clusters that can be merged at each step [24]. This method works with categorical data and optimizes hyperparameters to infer the optimal number of clusters. BHC clustering was performed using the BHC BioConductor package (version 1.56.0) in R with default parameters.
Cluster characteristics
Parameter importance was inferred using V-measure (or Normalized Mutual Information, calculated by “V-test” using “FactoMineR”, version 2.11, in R). V-measure is a single metric, computed based on percentage of the cluster that has a given characteristic and percentage of the population with a given characteristic that is in the cluster, which is an average of homogeneity and completeness. Key parameters that define the cluster characteristics have high V-measure. The relative importance of each parameter on each cluster was determined by Fisher’s exact test or V-Test.
Tools
R was used for all statistical analyses, and an overview of all R packages used in the analysis is listed in Supplementary Table S4.
Data availability
All data used in this study are provided in Supplementary Table S1.
Results
Data pre-processing and exploratory analyses
An overview of the data processing and analysis pipeline is provided in Figure 1. For clustering, we selected subjects diagnosed with TD (1,548) and excluded subjects with genetic abnormalities or severe psychiatric conditions (421) (see Methods for details). In addition, we excluded subjects with who were rated “Unable to rate” for Tic, Obsessive-compulsive, or Attention-deficit-Hyperactivity disorder diagnoses. To limit the effects of genetic relatedness, we selected one offspring from each family. The final data set used for identifying subtypes includes 865 offspring diagnosed with TD. Counts by category for each parameter are included in the Supplementary Table S2.
Figure 1. Analysis Pipeline overview.
We first performed exploratory data analysis to understand the characteristics of data, such as parameter distribution, completeness, and correlation. We then evaluated several unsupervised clustering methods (Supplementary Table S3) to identify methods that produce stable clusters. After testing, we selected two methods, K-Means and BHC for the final clustering and analysis of the data. After the cluster identification, we used V-Test to identify key factors that define the characteristics of each cluster and Fisher’s exact test to assess the impact of factors on each cluster.
To understand the relationship between selected parameters, we computed pairwise correlations among the parameters using Cramer’s V statistic (Supplemental Figure 1). The observed high correlations between OCD-related parameters, ADHD-related parameters, and Trichotillomania-related parameters are expected. There were no other significant correlations among the parameters.
After evaluating eight unsupervised clustering methods (Supplementary Table S3), and many different subsets of data (Supplementary Table S5), we selected two methods that produced stable clusters with our dataset: K-Means and Bayesian Hierarchical Clustering (BHC). Because these methods can accommodate correlated parameters, we included correlated parameters such as disease diagnosis, subtypes, and current symptom, in our analyses.
K-Means Clustering
To perform K-Means clustering, we used MCA to generate PCs from the dataset. When we provided all PCs to K-means clustering in our initial testing, it resulted in unstable clusters. These results are probably due to inherent variability in the clinical data. The later PCs capture mostly noise rather than meaningful structure in the clinical data. Therefore, we evaluated systematically excluding the least important PCs, one at a time, to reduce the noise in the dataset. For each set of PCs, we assessed the cluster number (K) from 2 to 20 and repeated the analysis 100 times to assess the stability. This resulted in the selection of the top three PCs (account for >35% of the variance), that produced five stable clusters (Silhouette Plot, Supplementary Figure 2).
Next, we identified the parameters that most significantly correlated (positively or negatively) with cluster identity (Figure 2A). For example, trichotillomania diagnosis is positively correlated only with cluster 2 (TrichDiag=TRUE), indicating grouping of subjects with trichotillomania in this cluster. ADHD is negatively correlated with cluster 3 (ADHDDiag=No ADHD). Plotting cluster memberships along the top three PCs confirmed clear separation of clusters along the three PCs, suggesting robust clustering and visually confirming the quality of clusters identified (Figures 2B–D). .
Figure 2. K-Means Clusters.
A. Major factors determining cluster characteristics. Positive values are positive correlations, while negative values are negative correlations. The larger the value, the stronger the correlation. B-D. Cluster plots on first three principal components.
The detailed contribution and significance of each parameter to each cluster are listed in Supplementary Data: K-Means.xlsx, including percentage of subjects in a category in each cluster, percentage of subjects in each cluster in a category, percentage of subjects in a category in the whole data set, and v-test for the significance of each parameter. These results provide details for each statistically significant parameter for each cluster, allowing us to fully understand the cluster characteristics. Overall, 12 out of 14 parameters show statistically significant contribution to the clusters, including all parameters associated with OCD, ADHD, trichotillomania, region, and sex (Supplementary Data: KMeans.xlsx – test.ch2 tab).
Based on the parameter contribution to the clusters and the individual membership of each cluster, we determined the defining characteristics of each cluster (Supplementary Table S6). To make the results below more comprehensible, we describe all diagnosis parameters for a given disorder as one category in the sections below (e.g., “ADHD” for ADHD diagnosis and ADHD current symptom) and included the p-value for the most statistically significant parameter.
Cluster 1 subjects (n=210) have ADHD (3.6E−71), early OCD age of onset (1.7E−46), no trichotillomania (p=1.1E−5), higher prevalence of males (2.2E−8) and ASD (p=7.9E−4). Cluster 2 subjects (n=40) have trichotillomania (p=6.7E−70), with higher prevalence of subjects from the USA (p=6.9E−4) and OCD (p=8.0E−4). Cluster 3 subjects (n=289) have high prevalence of OCD (p=4.0E−52), low levels of ADHD (p=4.0E−49), no trichotillomania (p=5.4E−8), higher prevalence of females (1.1E−6) and European subjects (1.8E−3). Cluster 4 subjects (n=229) have low NDD comorbidities, with no OCD (p=7.4E−127), very low ADHD (p=7.4E−45), no trichotillomania (p=3.3E−6), low prevalence of ASD (p=5.1E−6), and a high proportion of South Korean subjects (2.3E−6). Cluster 5 (n=97) comprises subjects with TD that have no OCD (p=5.6E−45), high prevalence of ADHD (p=2.6E−32), ASD (p=1.0E−3), and males (p=1.1E−4).
BHC
BHC clustering using the data set with 14 categorical parameters identified six clusters. We identified the parameters that most significantly correlated (positively or negatively) with cluster identity (Figure 3A). Plotting cluster memberships along the top three PCs confirmed clear separation, suggesting robust clustering (Figures 3B–D) and similarity between clustering produced by K-Means and BHC. The detailed contribution and significance of each parameter to each cluster are listed in Supplementary Data BHC.xlsx. Like the K-mean clustering analysis, 11 out of 14 parameters show a statistically significant contribution to the clusters (Supplementary Data: BHC.xlsx – test.ch2 tab).
Figure 3. BHC Clusters.
A. Cluster characteristics heatmap. Positive values are positive correlations, while negative values are negative correlations. The larger the value, the stronger the correlation. B-D. Cluster plots on first three principal components.
Next, we examined the characteristics of each of the six clusters (Supplementary Table S7). Cluster 1 (n=120) comprises subjects with TD that have a high prevalence of ADHD (p=8E−44), no OCD (p=5.2E−57), and no trichotillomania (p=2.2E−3). Cluster 2 subjects (n=206) have low NDD comorbidities such as no OCD (p=7.9E−110), very low ADHD (p=7.3E−60), and an absence of trichotillomania (p=1.4E−5), and a low prevalence of ASD (p=3.6E−5). Cluster 3 subjects (n=30) have trichotillomania (p=2.9E−47), with a higher prevalence of OCD (p=2.8E−8), and are more likely to be from the USA (1.2E−3). Cluster 4 subjects (n=10) have trichotillomania (p=1.4E−14) and no OCD (p=1.3E−3). Cluster 5 subjects (n=250) have a high prevalence of OCD (p=2.1E−49), low levels of ADHD (p=2.3E−64), an absence of trichotillomania (p=8.1E−7), and an increased proportion of females (4.2E−5). Cluster 6 subjects (n=249) have ADHD (3.4E−81), OCD (1.3E−42), no trichotillomania (p=8.7E−7), and a higher proportion of males (3.6E−5).
Comparison of K-Means and BHC’s results
K-Means uses a distance metric (Euclidean distance) and requires specification of the number of clusters. On the other hand, BHC uses an infinite mixture model to infer the optimal number of clusters and does not require specification of a distance metric. Fraction of members in each K-Means cluster that are in each BHC cluster is depicted in a heatmap in Figure 4. Both methods resulted in four large clusters of similar sizes and characteristics, as illustrated in PC plots (Figures 2 and 3) and Table 1. These large clusters differ from each other only in minor characteristics and slightly varying memberships. Besides the four large clusters, K-Means clustered all trichotillomania patients together in one cluster, while BHC split them into two clusters (with and without OCD). Most significant parameters (low p-value) that characterize each subtype and differentiate it from others are highlighted (Table 1).
Figure 4.
K-Means and BHC Cluster Comparison – Fraction of K-Means cluster members in each BHC Cluster in descending order of overlap. For example, 75% and 25% of the K-Means cluster 2 members are in BHC cluster 3 and 4, respectively.
Table 1.
Comparison of K-Means (KM) and BHC Clusters
| Kmclust | Kmsize+ | BHCclust | BHCsize* | Pos. Correlation | Neg. Correlation |
|---|---|---|---|---|---|
| 1 | 210 | 6 | 249 | ADHD | Trichotillomania |
| OCD | |||||
| Male | |||||
| ASD+ | |||||
| USA+ | |||||
| Tic Early Onset+ | |||||
| 2 | 40 | Trichotillomania | |||
| USA | |||||
| 3 | 30 | OCD+* | |||
| 4 | 10 | OCD* | |||
| 3 | 289 | 5 | 250 | OCD | ADHD |
| Female | Trichotillomania | ||||
| Europe+ | |||||
| 4 | 229 | 2 | 206 | South Korea+ | OCD |
| Tic Late Onset+ | ADHD | ||||
| Trichotillomania | |||||
| ASD | |||||
| 5 | 97 | 1 | 120 | ADHD | OCD |
| Male+ | TicSxCur | ||||
| ASD+ | Trichotillomania |
Note: Most statistically significant characteristics of each cluster are in bold
+ indicates of significance in Kmeans
indicates of significance in BHC
The differences are only in cluster characteristics with low statistical significance
Discussion
Tourette disorder has high phenotypic heterogeneity, high comorbidity with other NDDs, and varying rates of shared heritability with other NDDs, suggesting distinct underlying genetic causes for different TD subjects. In this study, we performed unsupervised clustering analysis of clinical data from 865 TD subjects in the TIC Genetics study. The analyses identified five clinically relevant subtypes and highlighted shared characteristics among TD patients. Prior attempts to identify specific genes or alleles contributing to disease risk have yielded only two genes with rare damaging variants that confer large effects and a single common variant of small individual effect [8, 9]. Stratified genetic analysis of TD subjects based on additional clinical features may help identify more genetically homogenous cohorts and improve power for gene discovery. Furthermore, identifying clinically relevant subtypes could enable better diagnostics and treatment. Earlier attempts to identify subtypes often relied on small sample sets from single cohort [2, 16, 28, 29]. In addition, these studies usually relied on prior hypotheses regarding a subset of factors that were the cause of differentiation. To overcome these limitations, here we performed unsupervised clustering analyses, using consistent clinical data from over 20 global sites, without filtering parameters using prior hypothesis. This analysis enabled us to build on prior knowledge and identify novel TD subtypes.
A major challenge in clinical data analysis lies in the presence of noise and occasional misclassification, which can compromise the reliability of findings. Therefore, data selection and cleanup are essential. In addition, we also evaluated several unsupervised clustering methods to identify the optimal methods for our data set (Supplementary Table S3). After evaluating a variety of data set selection criteria and clustering methods, we selected only TD subjects for our final dataset (Supplementary Table S1) and identified two methods, K-Means and BHC, that worked well with our dataset. Final dataset was selected to be as complete and homogeneous as possible and the methods were selected to ensure stable clustering.
The two methods are quite different from each other. K-Means clustering was performed with PCs from MCA using the Euclidean Distance metric. The number of clusters was determined using Silhouette Statistics, coupled with the stability of the clusters generated. On the other hand, BHC differs from traditional distance-based agglomerative clustering algorithms in several ways: (1) It defines a probabilistic model to compute the probability of a subject belonging to any of the existing clusters in the tree; (2) it uses a model-based criterion to decide on merging clusters rather than an ad-hoc distance metric; and (3) the number of clusters is automatically determined [24]. In our analysis, K-means identified five clusters while BHC identified six clusters. Interestingly, the largest four clusters from both methods have similar size and characteristics. Because BHC used all the (categorical) data, while K-Means clustering used the first three PCs, this might account for the slight differences in cluster sizes and minor characteristics. Nevertheless, as we started without any assumption regarding the number of subtypes, the identification of similar sets of clinically relevant clusters by two different methods (Table 1) increases our confidence in the results. In addition to confirming earlier findings on the importance of highly prevalent comorbidities such as ADHD and OCD [16], this large and diverse dataset identified new clusters that illustrate the importance of other clinically relevant factors such as trichotillomania and sex in TD subjects.
Trichotillomania is an often debilitating psychiatric condition characterized by recurrent hair-pulling, which can lead to disfiguring hair loss and significant distress and social impairment, and has been associated with elevated rates of anxiety, depression, ADHD, and OCD (DSM-5) [1]. In DSM-5, it is classified under OCD-related Disorders. The condition has significant heritability according to twin studies, a shared genetic liability with OCD and other OCD-related factors, as well as a distinct factor specific to hair-pulling and skin-picking [27]. Our study found subjects with TD and comorbid trichotillomania appear more likely to have OCD, reaffirming the earlier results.
Another advantage of the TIC Genetics cohort is its geographical diversity. Subjects were from the USA, Europe, Israel, and South Korea, which permitted us to assess the potential population-specific genetic contribution to TD subtypes. Subjects from TIC Genetics centers in South Korea include 69 Asians (~8.4% of the total TIC Genetics cohort). Only about 10% of all Asian subjects with a diagnosis of TD also had a diagnosis of OCD, as compared to 40–60% observed in Caucasian subjects with TD diagnosis. A prior cross-national study indicated a lower prevalence of OCD among Asians compared to the worldwide prevalence [28]. Determining whether the apparent difference in rates of OCD comorbidity among regions is due to genetic variation among populations or other factors is an important question requiring further study. In addition, there are higher proportion of TD subjects with minimal or no NDD comorbidity subjects from the USA and Europe than from South Korea, suggesting ancestry/genetic background and/or assessment differences might account for some of these findings.
While our dataset is larger than most prior studies and has more parameters, this study still has several limitations. First, compared with phase 1, TIC Genetics phase 2 focused on recruiting simplex trios. However, we confirmed that this did not affect the unsupervised clustering because adding study phase as a parameter did not affect the clustering results. Second, as subjects with TD made up the vast majority of the subjects with tic disorders in the study, we only included subjects with TD to improve cluster stability. A larger dataset, including subjects with other tic disorder diagnoses, could permit better clustering despite the noisiness of the data and provide better grouping of tic disorder subtypes in addition to TD. Third, our dataset might not be large or varied enough to identify all the underlying subtypes with high accuracy. Larger data sets will enable us to effectively use machine learning and other methods (e.g., Variational Auto Encoders) for unsupervised clustering. Lastly, in addition to observations by the clinicians, a portion of the data is self-reported by the subject or their parents and might produce an uncertain level of errors. Even though the clinicians in the TIC Genetics study were trained to record the symptoms and diagnosis in a consistent manner, there may still be undetected variations between individual clinicians and sites. Despite all these limitations, the large and diverse TIC Genetics cohort enabled us to enhance the current understanding by identifying new and potentially relevant factors and subtypes.
Future Direction
We are also exploring effective ways to include genetic data (e.g., SNP microarray and whole-exome sequencing data) available from the cohort to further identify subtypes. Another approach is to identify subtypes using genetic data and validate them using clinical data. The tools and methods we have evaluated and the workflow we developed here can also be used to evaluate hypotheses generated by domain experts. Ultimately, we plan to perform stratified analysis of genetic data based on the subtypes to identify new candidate genes and generate new insights into TD etiology, diagnosis, and treatment.
Supplementary Material
Acknowledgments
We thank the families who have participated in and contributed to this study. Bio samples are available through the NIMH Repository and Genomics Resource (U24MH068457 to J.A.T.). We gratefully acknowledge the contributions of Mayra Aldecoa, Diana Bok, Maria Cruz, Kayla Delapenha, Kristen Fleming, Laura Ibanez-Gomez, Jessica D. Leuchter, Adam Lombroso, Daniela Martinez, and Vessela Zaharieva to the TIC Genetics study. We are also grateful to the NJCTS for facilitating the inception and organization of the TIC Genetics study.
Funding:
This study was supported by grants from the National Institute of Mental Health (R01MH115958 to Gary Heiman and Jay Tischfield; R01MH115959 to Barbara J. Coffey; R01MH115960 to Alyssa Rosen; R01MH115961 to Samuel Kuperman; R01MH115962 to Donald L. Gilbert; R01MH115963 to Matthew State and Jeremy Willsey; and R01MH115993 to Samuel H. Zinner), from the Human Genetics Institute of New Jersey (to Gary Heiman and Jay Tischfield), and the New Jersey Center for Tourette Syndrome and Associated Disorders (to Gary Heiman and Jay Tischfield) and U24MH068457 (Jay Tischfield). Dr. Münchau has received grant from the Deutsche Forschungsgemeinschaft (DFG FOR 2698). Laura Muñoz-Delgado is financially supported by the “Juan Rodés” program (JR23/00016) from the Instituto de Salud Carlos III.
Conflicts of Interest:
Fernandez, T.V. has received research support or grants from the National Institutes of Health, Misophonia Research Fund, and Yale Child Study Center. He received an honorarium for participation in the 2025 Pediatric Psychopharmacology Update Institute by the American Academy of Child & Adolescent Psychiatry. He is paid for expert testimony and consultation by DLA Piper LLC. Dr. Gilbert has received compensation for expert testimony for the U.S. National Vaccine Injury Compensation Program, through the Department of Health and Human Services. He has received payment for medical expert opinions through TeladocHealth International. He has served as a paid consultant for PTC Therapeutics, Noema Pharma, and Emalex Biosciences and has received travel support to attend investigator meetings. He has provided educational lectures for Illumina, Inc and PTC Therapeutics, Inc. He has received salary compensation through Cincinnati Children’s for work as a clinical trial site investigator from Emalex Biosciences, Inc. (clinical trial, Tourette Syndrome), PTC Therapeutics (registry and clinical trial, Amino Acid Decarboxylase Deficiency), Neurocrine Biosciences (clinical trial, cerebral palsy), and Quince Therapeutics (clinical trial, ataxia telangiectasia). He has received book/publication royalties from Elsevier and Wolters Kluwer. Matthew State serves on the Scientific Advisory Board of MapLight Therapeutics.
Funding Statement
This study was supported by grants from the National Institute of Mental Health (R01MH115958 to Gary Heiman and Jay Tischfield; R01MH115959 to Barbara J. Coffey; R01MH115960 to Alyssa Rosen; R01MH115961 to Samuel Kuperman; R01MH115962 to Donald L. Gilbert; R01MH115963 to Matthew State and Jeremy Willsey; and R01MH115993 to Samuel H. Zinner), from the Human Genetics Institute of New Jersey (to Gary Heiman and Jay Tischfield), and the New Jersey Center for Tourette Syndrome and Associated Disorders (to Gary Heiman and Jay Tischfield) and U24MH068457 (Jay Tischfield). Dr. Münchau has received grant from the Deutsche Forschungsgemeinschaft (DFG FOR 2698). Laura Muñoz-Delgado is financially supported by the “Juan Rodés” program (JR23/00016) from the Instituto de Salud Carlos III.
Footnotes
Informed Consent and Institutional Review Board Statement:
All adult participants and parents of children provided written informed consent along with written or oral assent of their participating child. The Institutional Review Board (IRB) of each participating site approved the study.
Consortia:
Juliane Ball, Leonie F. Becker, Noa Benaroya-Milshtein, Kate Bornais, Danielle Cath, Keun-Ah Cheon, Barbara J. Coffey, Lenrine Dalmeijer, Erik M. Elster, Dana Feldman, Thomas V. Fernandez, Carolin Fremer, Lorena Garrote Espina, Donald L. Gilbert, Danea Glover, Cristina Gómez Rapela, Tammy Hedderly, Gary A. Heiman, Isobel Heyman, Hyun Ju Hong, Chaim Huijser, Christina Kappler-Friedrichs, Young Key Kim, Young Shin Kim, Robert A. King, Nadine Kirchen, Carolin Sophie Klages, Yun-Joo Koh, Florian Kraemer, Subramanian Krishnamurthy, Samuel Kuperman, Bennett L. Leventhal, Holan Liang, Maria Loreta Lopez, Marcos Madruga-Garrido, Veronika Mailänder Zelger, Osman Malik, Marieke Messchendorp, Malindi Mheen, van der, Dararat Mingbunjerdsuk, Pablo Mir, Astrid Morer, Kirsten Müller-Vahl, Alexander Münchau, Laura Muñoz-Delgado, Tara L Murphy, Cara Nasello, Elena Ojeda-Lepe, Veit Roessner, Alyssa Rosen, Guy Rouleau, Simon Schmitt, Chitra Shukla, Sara Sopena, Matthew W State, Tamar Steinberg, Frederika Tagwerker, Zsanett Tarnok, Joshua K. Thackray, Meitar Timmor, Jay A. Tischfield, Max A. Tischfield, Anne Uhlmann, Ana Vigil-Pérez, Frank Visscher, Susanne Walitza, Belinda Wang, Jinchuan Xing, and Samuel H. Zinner.
Financial Disclosures
See Conflicts of Interest section for details.
References
- 1.American Psychiatric Association: Diagnostic and statistical manual of mental disorders, Fifth Edition (DSM-5). Washington DC: American Psychiatric Association; 2013. [Google Scholar]
- 2.Eapen V, Robertson MM: Are there distinct subtypes in Tourette syndrome? Pure-Tourette syndrome versus Tourette syndrome-plus, and simple versus complex tics. Neuropsychiatr Dis Treat 2015, 11:1431–1436. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Garcia-Delgar B, Servera M, Coffey BJ, Lazaro L, Openneer T, Benaroya-Milshtein N, Steinberg T, Hoekstra PJ, Dietrich A, Morer A, group Ec: Tic disorders in children and adolescents: does the clinical presentation differ in males and females? A report by the EMTICS group. Eur Child Adolesc Psychiatry 2022, 31:1539–1548. [DOI] [PubMed] [Google Scholar]
- 4.Dy-Hollins ME, Chibnik LB, Tracy NA, Osiecki L, Budman CL, Cath DC, Grados MA, King RA, Lyon GJ, Rouleau GA, et al. : Sex Differences in Natural History and Health Outcomes Among Individuals With Tic Disorders. Neurology 2025, 104:e210249. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.O’Rourke JA, Scharf JM, Yu D, Pauls DL: The genetics of Tourette syndrome: a review. J Psychosom Res 2009, 67:533–545. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Price RA, Kidd KK, Cohen DJ, Pauls DL, Leckman JF: A twin study of Tourette syndrome. Arch Gen Psychiatry 1985, 42:815–820. [DOI] [PubMed] [Google Scholar]
- 7.Akingbuwa WA, Hammerschlag AR, Bartels M, Middeldorp CM: Systematic Review: Molecular Studies of Common Genetic Variation in Child and Adolescent Psychiatric Disorders. J Am Acad Child Adolesc Psychiatry 2022, 61:227–242. [DOI] [PubMed] [Google Scholar]
- 8.Willsey AJ, Fernandez TV, Yu D, King RA, Dietrich A, Xing J, Sanders SJ, Mandell JD, Huang AY, Richer P, et al. : De Novo Coding Variants Are Strongly Associated with Tourette Disorder. Neuron 2017, 94:486–499 e489. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Wang S, Mandell JD, Kumar Y, Sun N, Morris MT, Arbelaez J, Nasello C, Dong S, Duhn C, Zhao X, et al. : De Novo Sequence and Copy Number Variants Are Strongly Associated with Tourette Disorder and Implicate Cell Polarity in Pathogenesis. Cell Reports 2018, 24:3441–3454.e3412. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Cao X, Zhang Y, Abdulkadir M, Deng L, Fernandez TV, Garcia-Delgar B, Hagstrom J, Hoekstra PJ, King RA, Koesterich J, et al. : Whole-exome sequencing identifies genes associated with Tourette’s disorder in multiplex families. Mol Psychiatry 2021, 26:6937–6951. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.de Haan MJ, Delucchi KL, Mathews CM, Cath DC: Tic symptom dimensions and their heritabilities in Tourette’s syndrome. Psychiatr Genet 2015, 25:112–118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Pauls DL, Fernandez TV, Mathews CA, State MW, Scharf JM: The inheritance of Tourette Disorder: A review. Journal of Obsessive-Compulsive and Related Disorders 2014, 3:380–385. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Hirschtritt ME, Lee PC, Pauls DL, Dion Y, Grados MA, Illmann C, King RA, Sandor P, McMahon WM, Lyon GJ, et al. : Lifetime prevalence, age of risk, and genetic relationships of comorbid psychiatric disorders in Tourette syndrome. JAMA Psychiatry 2015, 72:325–333. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Yang K, Lei T, Jun J, Yang Q, Li J, Wang M, Cui Y: Advances in Clustering and Classification of Tic Disorders: A Systematic Review. Neuropsychiatric Disease and Treatment 2024, Volume 20:2663–2677. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Hirschtritt ME, Darrow SM, Illmann C, Osiecki L, Grados M, Sandor P, Dion Y, King RA, Pauls D, Budman CL, et al. : Genetic and phenotypic overlap of specific obsessive-compulsive and attention-deficit/hyperactive subtypes with Tourette syndrome. Psychol Med 2018, 48:279–293. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Grados MA, Mathews CA, Tourette Syndrome Association International Consortium for G: Latent class analysis of gilles de la tourette syndrome using comorbidities: clinical and genetic implications. Biol Psychiatry 2008, 64:219–225. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Mathews CA, Grados MA: Familiality of Tourette syndrome, obsessive-compulsive disorder, and attention-deficit/hyperactivity disorder: heritability analysis in a large sib-pair sample. J Am Acad Child Adolesc Psychiatry 2011, 50:46–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Brainstorm C, Anttila V, Bulik-Sullivan B, Finucane HK, Walters RK, Bras J, Duncan L, Escott-Price V, Falcone GJ, Gormley P, et al. : Analysis of shared heritability in common disorders of the brain. Science 2018, 360. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Yang K, Zhang W, Li Y, Wang X, Jiang Z, Hu S, Jun J, Yang Q, Li J, Hong X, et al. : Subtypes of tic disorders in children and adolescents: based on clinical characteristics. BMC Pediatrics 2025, 25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Hirschtritt ME, Darrow SM, Illmann C, Osiecki L, Grados M, Sandor P, Dion Y, King RA, Pauls DL, Budman CL, et al. : Social disinhibition is a heritable subphenotype of tics in Tourette syndrome. Neurology 2016, 87:497–504. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Dietrich A, Fernandez TV, King RA, State MW, Tischfield JA, Hoekstra PJ, Heiman GA, Group TICGC: The Tourette International Collaborative Genetics (TIC Genetics) study, finding the genes causing Tourette syndrome: objectives and methods. Eur Child Adolesc Psychiatry 2015, 24:141–151. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.American Psychiatric Association: Diagnostic and statistical manual of mental disorders, 4th Edition Text Revision (DSM-IV-TR). Washington DC: American Psychiatric Association; 2000. [Google Scholar]
- 23.Mangiafico SS: rcompanion: Functions to Support Extension Education Program Evaluation. 2.5.0 edition. Rutgers Cooperative Extension, New Brunswick, New Jersey; 2025. [Google Scholar]
- 24.Heller KA, Ghahramani Z: Bayesian hierarchical clustering. In; 2005. ACM Press; [Google Scholar]
- 25.Le S, Josse J, Husson F: FactoMineR: An R package for multivariate analysis. Journal of Statistical Software 2008, 25:1–18. [Google Scholar]
- 26.Hartigan JA, Wong MA: Algorithm AS 136: A K-Means Clustering Algorithm. Applied Statistics 1979, 28. [Google Scholar]
- 27.Monzani B, Rijsdijk F, Harris J, Mataix-Cols D: The structure of genetic and environmental risk factors for dimensional representations of DSM-5 obsessive-compulsive spectrum disorders. JAMA psychiatry 2014, 71:182–189. [DOI] [PubMed] [Google Scholar]
- 28.Weissman MM: Cross-national epidemiology of obsessive-compulsive disorder. CNS spectrums 1998, 3:6–9. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All data used in this study are provided in Supplementary Table S1.