PreprintPDF Available
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

Background Tuberous Sclerosis Complex (TSC), a multi-system genetic disorder, is associated with a wide range of TSC-Associated Neuropsychiatric Disorders (TAND). Individuals have apparently unique TAND profiles, challenging diagnosis, psycho-education, and intervention planning. We proposed that identification of natural TAND clusters could lead to personalized identification and treatment of TAND. Two small-scale studies showed cluster and factor analysis could identify clinically meaningful natural TAND clusters. Here we set out to identify definitive natural TAND clusters in a large, international dataset. Method Cross-sectional, anonymized TAND Checklist data of 453 individuals with TSC were collected from six international sites. Data-driven methods were used to identify natural TAND clusters. Mean squared contingency coefficients were calculated to produce a correlation matrix, and various cluster analyses and exploratory factor analysis were examined. Statistical robustness of clusters was evaluated with 1000-fold bootstrapping, and internal consistency calculated with Cronbach’s alpha. Results Ward’s method rendered seven natural TAND clusters with good robustness on bootstrapping. Cluster analysis showed significant convergence with an exploratory factor analysis solution, and, with the exception of one cluster, internal consistency of the emerging clusters was good to excellent. Clusters showed good clinical face validity. Conclusions Our findings identified a data-driven set of natural TAND clusters from within highly variable TAND Checklist data. The seven natural TAND clusters could be used to train families and professionals and to develop tailored approaches to identification and treatment of TAND. Natural TAND clusters may also have differential aetiological underpinnings and responses to molecular and other treatments.
Content may be subject to copyright.
Page 1/20
Multivariate Data Analysis Identies Natural
Clusters of Tuberous Sclerosis Complex Associated
Neuropsychiatric Disorders (TAND)
Petrus J de Vries ( petrus.devries@uct.ac.za )
University of Cape Town https://orcid.org/0000-0002-8915-1571
Loren Leclezio
University of Cape Town
Sugnet Gardner-Lubbe
Stellenbosch University
Darcy Krueger
Cincinnati Children's Hospital Medical Center
Mustafa Sahin
Boston Children's Hospital
Steven Sparagana
Texas Scottish Rite Hospital for Children: Scottish Rite for Children
Liesbeth De Waele
UZ Leuven: Katholieke Universiteit Leuven Universitaire Ziekenhuizen Leuven
Anna Jansen
UZ Brussel: Universitair Ziekenhuis Brussel
Research Article
Keywords: Tuberous Sclerosis Complex, TAND, natural TAND clusters, neuropsychiatric, autism spectrum
disorder, cluster analysis, factor analysis, precision medicine
DOI: https://doi.org/10.21203/rs.3.rs-835934/v1
License: This work is licensed under a Creative Commons Attribution 4.0 International License. 
Read Full License
Page 2/20
Abstract
Background
Tuberous Sclerosis Complex (TSC), a multi-system genetic disorder, is associated with a wide range of
TSC-Associated Neuropsychiatric Disorders
(
TAND). Individuals have apparently unique TAND proles,
challenging diagnosis, psycho-education, and intervention planning. We proposed that identication of
natural TAND clusters could lead to personalized identication and treatment of TAND. Two small-scale
studies showed cluster and factor analysis could identify clinically meaningful natural TAND clusters.
Here we set out to identify denitive natural TAND clusters in a large, international dataset.
Method
Cross-sectional, anonymized TAND Checklist data of 453 individuals with TSC were collected from six
international sites. Data-driven methods were used to identify natural TAND clusters. Mean squared
contingency coecients were calculated to produce a correlation matrix, and various cluster analyses
and exploratory factor analysis were examined. Statistical robustness of clusters was evaluated with
1000-fold bootstrapping, and internal consistency calculated with Cronbach’s alpha.
Results
Ward’s method rendered seven natural TAND clusters with good robustness on bootstrapping. Cluster
analysis showed signicant convergence with an exploratory factor analysis solution, and, with the
exception of one cluster, internal consistency of the emerging clusters was good to excellent. Clusters
showed good clinical face validity.
Conclusions
Our ndings identied a data-driven set of natural TAND clusters from within highly variable TAND
Checklist data. The seven natural TAND clusters could be used to train families and professionals and to
develop tailored approaches to identication and treatment of TAND. Natural TAND clusters may also
have differential aetiological underpinnings and responses to molecular and other treatments.
Background
Tuberous Sclerosis Complex (TSC) is a multi-system genetic disorder associated with a range of physical
manifestations1. The main burden of the disorder is, however, linked to the neurological and
neuropsychiatric features of TSC2. TSC-associated neuropsychiatric disorders (TAND) is seen in ~ 90% of
individuals across the lifespan3. Previous studies suggested that each individual with TSC appears to
present with their own unique prole or ‘TAND signature4. The complexity and perceived uniqueness of
TAND proles is, however, a major barrier to screening, diagnostic work-up, intervention planning, and
psycho-education. Thus, identication of natural TAND clusters - predictable groupings of specic
Page 3/20
neuropsychiatric characteristics - may be a powerful strategy to reduce the perceived overwhelming
heterogeneity of individual TAND proles and resulting ‘treatment paralysis’, and could lead to the
development of a personalized approach to management and treatment of individuals with TSC in
clinical settings.
Machine-based data reduction methods have been used in humans to reduce the multi-dimensionality of
behavioural characteristics to identify previously unrecognized clusters of behaviours. For example, in a
study of the behavioural phenotype of humans with Cornelia de Lange Syndrome, categorical principal
component analysis (PCA) was used as a data reduction tool and to describe relationships between a
large number of behavioural variables5. This methodology allowed for new insights into the relationships
between physical and behavioural characteristics and into genotype-phenotype correlations. In a
feasibility study, we examined the viability of using multivariate data analysis techniques as a novel
strategy in TSC to identify natural TAND clusters. Cluster analysis of 29 variables in 56 individuals with
TSC rendered six natural TAND clusters with good face validity and signicant convergence with a 6-
factor exploratory factor analysis solution6. The natural TAND clusters identied included a ‘Scholastic’
cluster, anAutism Spectrum Disorder-like’ cluster, a ‘Dysregulated behaviour’ cluster, a
‘Neuropsychological’ cluster, a ‘Hyperactive/Impulsive’ cluster, and a ‘Mixed/Mood’ cluster. The feasibility
study, however, had a small sample from only two centres (South Africa and Australia) and, whilst
informative from a methodological perspective, clearly required replication and expansion. Next, we
therefore used the same methodology in a new sample of n = 85 individuals with TSC from 7 European
countries7. The study also found six natural TAND clusters and replicated the majority of the earlier
ndings. Even though there is no consensus on sample size for cluster analysis research, we
acknowledged the need for signicantly larger sample size, and for evaluation of the statistical
robustness and internal consistency of TAND clusters.
In this study, we used cluster and factor analysis methods in a large international sample, in search of
denitive natural TAND clusters. In addition, we evaluated the robustness and internal consistency of
identied natural TAND clusters.
Methods
Subjects
Participants for this study (n = 453) were recruited from ve expert TSC centres: Cincinnati, USA (365
participants), Boston, USA (25 participants), Brussels, Belgium (25 participants), Dallas, USA (14
participants) and Leuven, Belgium (9 participants). An additional 16 participants were recruited through
Tuberous Sclerosis International (TSCi). To be eligible, participants had to meet clinical criteria for TSC1.
Anonymized data deliberately included participants with a wide age and ability range. All procedures
contributing to this work complied with the ethical standards of the relevant national and international
committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. The
protocol was peer-reviewed in the Department of Psychiatry at the University of Cape Town and
Page 4/20
submitted for ethical approval by the Faculty of Health Sciences, Human Research Ethics Committee
(Ethics Ref 340/2015). Additional study sites obtained ethical approval or waivers from their respective
HREC/IRB bodies.
Procedures
The TAND Checklist was administered to parents and caregivers of individuals with TSC by the resident
TSC coordinator and/or treating physician. The TAND Checklist is a short pen-and-paper checklist that
captures the high frequency neuropsychiatric diculties in TSC. It was developed in partnership with
family and professional stakeholders in the TSC community. It typically takes about 10–20 minutes to
complete. For details about TAND and the TAND Checklist, including a downloadable version of the
Checklist, please see de Vries
et al,. (2015)
3.
Data Analysis
Statistical analysis was performed on anonymized TAND Checklist data using a series of steps as
outlined below and in Fig. 1.
STEP 1. Select TAND Checklist variables
The following sections of the TAND Checklist were included in the analysis: Sect.3, behavioural
challenges (19 questions/variables); Sect.5, academic skills (4 variables); and Sect.7,
neuropsychological skills (6 variables). This equated to 29 dichotomous TAND variables.
STEP 2. Compute Correlation Matrix
Mean squared contingency coecient8 was used to compute a correlation matrix for the 29 binary
variables selected from the TAND Checklist. Where missing values were present, these were omitted
pairwise in correlation computations.
STEP 3. Cluster Analysis
Several clustering solutions were compared. Hierarchical clustering methods provide a clustering tree
visually representing the merging of TAND variables and suggesting a suitable number of clusters.
Hierarchical clustering methods including complete linkage, average linkage, Ward’s method and
McQuitty’s method were applied with the R functions hclust() in base R9. Although hierarchical clustering
has often been used with great success, the algorithm is fairly naïve and some more recent methods in
the R package cluster10 were therefore also investigated. The FANNY (fuzzy clustering) method applied to
the data allocates a probability for belonging to each cluster rather than simply allocating each item to a
single cluster. DIANA is a divisive analysis hierarchical clustering method, whereas the other hierarchical
methods are all agglomerative.
STEP 4. Bootstrapping
Following cluster analysis, 1000-fold bootstrapping11 was applied to assess the statistical robustness of
the clustering solution. Based on the bootstrap sample of patients, a squared contingency coecient
Page 5/20
correlation matrix was computed, and hierarchical cluster analysis with Ward’s method was performed.
The number of clusters was xed at 7 and a bootstrap replicate clustering solution was obtained. In order
to assess the robustness or stability of the original observed clustering solution, the number of times
each pair of variables clustered together was calculated.
STEP 5. Exploratory Factor Analysis
After suitable clustering solutions were obtained and bootstrapping applied, exploratory factor analysis
was employed with the fa() R function in the package psych12. The factor analysis was also performed
on the mean squared contingency coecient correlation matrix. All the different options of factor
extraction and rotation available in the fa() function were investigated. These combinations were applied
to solutions with between four and seven factors. In order to nd the factor solution that best matched
the cluster analysis solution, the Tucker index of factor congruence13 was used. Optimal rotations were
achieved with Orthogonal Procrustes Analysis14 and the overall congruence summarised by the sum of
the diagonal values. For cases other than 6 factors, the congruence matrix was padded with zeros to
obtain a square matrix before rotation. Full algebraic details are available in the supplemental data of
Leclezio, Gardner-Lubbe & de Vries6.
STEP 6. Test Internal Consistency
Reliability analysis15 was used to test the internal consistency of the TAND variables both the clusters
identied and factors generated and in the nal proposal for clusters.
STEP 7. Compare Cluster Analysis ndings with Exploratory
Factor Analysis Results
Here we compared the data-driven cluster solutions with the exploratory factor analysis in order to
examine similarities and differences of the two approaches.
Step 8. Description of nal natural clusters
Finally, we integrated all results to generate a nal set of natural TAND clusters.
Results
Cluster Analysis
Hierarchical clustering with Ward’s method produced a seven-cluster solution. A dendrogram of the Ward
cluster analysis shows detail of the natural clustering of the TAND variables examined (Fig. 2).
The rst cluster included diculties with mathematics, spelling, writing, and reading suggesting a natural
‘Scholastic’ cluster. The second cluster included memory diculties, getting disorientated, attention
diculties at both a behavioural and neuropsychological level, diculty with visuo-spatial tasks, dual-
task diculties, and executive skills diculties. These suggested a natural ‘Neuropsychological’ cluster.
Page 6/20
The third cluster included extreme shyness, depressed mood, anxiety, and mood swings, suggesting a
natural ‘Mood/Anxiety’ cluster. The fourth cluster included inexibility, unusual language (repeating words
or phrases over and over again), repetitive behaviours, poor eye contact, and peer diculties, suggesting
an autism spectrum disorder (ASD)-like cluster. The fth cluster included aggressive outbursts, temper
tantrums, self-injury, and absent or delayed onset of language. The rst three items supported a natural
‘Dysregulated behaviour’ cluster, but the delayed language item did not overtly seem to t the natural
cluster. The sixth cluster contained impulsivity, overactivity, and restlessness, suggesting a natural
‘Overactive/Impulsive’ cluster. The seventh cluster contained two biological items - diculties with eating
and sleep-related problems. We refer to this as the ‘Eat/Sleep’ cluster.
Bootstrapping
Bootstrapping results are shown in Fig. 3. Black boxes are shown around the clusters as shown in the
dendrogram (see Fig. 2). Red boxes are shown around other items that showed relationships between
variables outside clusters. The Scholastic cluster was very stable with all four items clustering together
96% of the time. Three neuropsychological items and absent/delayed language showed association with
the scholastic cluster (Visuo-spatial skills, 32–47%; Dual-tasking, 21–27%; Executive skills, 19–21%;
Absent/Delayed language, 15–28%). In the Neuropsychological cluster, items showed bootstrap values
ranging from 100% (behavioural and neuropsychological attention diculties; dual-tasking and executive
skills), to 76% (Memory and Disorientation), while other pairings were less stable (Behavioural and
neuropsychological attention decits in relation to visuo-spatial decits, 8–17%). Items in the
Overactive/Impulsive cluster clustered together 94% (overactive, restless), 45% (impulsivity, overactive)
and 37% of the time (impulsivity, restlessness). In the Mood/Anxiety cluster extreme shyness showed
bootstrapping values between 15% and 32%, depressed mood 32–70%, anxiety 27–70%, and mood
swings 15–58%. Depressed mood also associated with aggression and temper tantrums 59% and 30% of
the time, respectively. The fth cluster, Dysregulated behaviour, showed that temper tantrums and
aggressive outbursts clustered together 100% of the time and that self-injury and absent/delayed
language clustered together 60% of the time. Self-injury clustered with aggression and temper tantrums
19–25% of the time, but absent/delayed language showed very low bootstrapping values (3–5%).
Instead, absent/delayed language showed higher bootstrapping values with items in other clusters,
particularly the scholastic (15–28%) and ASD-like cluster (6–40%). The ASD-like cluster items ranged
between 12% and 89% of which unusual language and repetitive behavior showed the highest
bootstrapping values (89%). The two items of the Eat/Sleep cluster clustered together 26% of the time.
Factor Analysis
Factor loadings (cut-off > 0.35) from exploratory factor analysis are shown in Fig. 4. The factor analysis
solution that most closely matched Ward’s hierarchical cluster analysis was the principal components
factor extraction method with clustering method rotation. Results supported a seven-factor solution very
similar to the cluster solution outlined above. Three items cross-loaded onto more than one factor. These
were self-injurious behavior (ASD-like factor and Eat/Sleep factor), sleeping diculties (Eat/Sleep factor
and Mood/Anxiety factor), and disorientation (ASD-like factor and the Neuropsychological factor).
Page 7/20
Internal Consistency
Five clusters had Cronbach alpha values 0.7 indicating good to excellent internal consistency:
Scholastic (0.97), Neuropsychological (0.87), ASD-like (0.76), Dysregulated behaviour (0.74), and
Overactive/Impulsive (0.70). The remaining 2 clusters had lower alpha values: Mood/Anxiety (0.69) and
Eat/Sleep (0.48). Similarly, in the factor analysis solutions 6 factors showed good to excellent internal
consistency: Scholastic (0.97), Neuropsychological (0.86), ASD-like (0.79), Dysregulated behaviour (0.75),
Mood/Anxiety (0.74) and Overactive/Impulsive (0.70). Only the Eat/Sleep factor scored < 0.7 with an
alpha = 0.54, suggesting poor internal consistency.
Comparison of cluster analysis and factor analysis ndings
Factor analysis conrmed a prole similar to cluster analysis, but with some slight variance between
clusters and factors (Fig. 5). Cluster and factor analysis showed the ‘Overactive/Impulsive’ and
‘Scholastic’ clusters to be clearly distinct, the ‘Mood/Anxiety’ and ‘Neuropsychological’ clusters to be fairly
distinct, while the other three clusters/factors showed more evidence of cross-loading between items in
different clusters/factors.
Integrated ndings of natural TAND clusters
Expert review of all ndings identied one problematic item for cluster placement (absent/delayed
language). The item was linked with self-injury (in the dysregulated behaviour cluster) in hierarchical
clustering. However, bootstrapping values for the item were very low inside the cluster (3–5%), but higher
with items in the scholastic and ASD-like cluster. After thorough review of all data, the item was moved to
the ASD-like cluster. Taking together all results, we propose 7 natural TAND clusters. Table 1 shows the
integrated natural clusters, component items and internal consistency for each natural TAND cluster (see
Table 1).
Page 8/20
Table 1
Seven natural TAND clusters identied in this study. The table shows the clusters with items contained
and internal consistency of each cluster
Natural TAND Cluster No. of
Items TAND Checklist Items Internal consistency
(Alpha)
Scholastic 4 • Reading
• Writing
• Spelling
• Mathematics
0.97*
Neuropsychological 7 • Memory
• Disorientation
• Attention diculties
(behaviour)
• Neuropsychological attention
decits
• Visuo-spatial
• Dual-tasking
• Executive skills
0.87*
Autism Spectrum
Disorder-Like 6 • Inexible
• Unusual language
• Repetitive behaviour
• Poor eye contact
• Peer diculties
• Delayed language
0.79*
Dysregulated behaviour 3 • Aggressive outbursts
• Temper tantrums
• Self-injury
0.73*
Overactive/Impulsive 3 • Overactive
• Impulsive
• Restless
0.70*
* Cronbach alpha  0.7 indicates good internal consistency
Page 9/20
Natural TAND Cluster No. of
Items TAND Checklist Items Internal consistency
(Alpha)
Mood/Anxiety 4 • Mood swings
• Anxiety
• Depressed mood
• Extreme shyness
0.69
Eat/Sleep 2 • Eating diculties
• Sleep diculties
0.48
* Cronbach alpha  0.7 indicates good internal consistency
Discussion
TSC-Associated Neuropsychiatric Disorders (TAND) have largely gone undiagnosed and untreated
despite affecting 90% of individuals with TSC. This has been attributable to lack of awareness, and lack
of expertise in TAND, but most fundamentally, the overwhelming uniqueness of TAND proles. We
proposed that the identication of naturally occurring TAND clusters may improve identication and
intervention4. In a feasibility study, we showed that a data-driven approach was able to identify natural
clusters of TAND6, and these ndings were replicated in a second study7. However, ndings required
larger-scale replication and extension, particularly to evaluate the robustness of proposed natural TAND
clusters. In this study, various cluster analysis techniques and exploratory factor analysis was applied in
a large and diverse international sample (n = 453). In addition, bootstrapping and internal consistency
analyses were performed.
Results identied seven natural TAND clusters, and bootstrapping showed clusters to be reasonably
stable. The scholastic cluster showed the highest robustness in terms of replicability on bootstrapping,
while other clusters showed a degree of agreement, suggesting that, with the exception of one item, the
identied cluster solution as shown in the dendrogram (Fig.2) is suciently replicable and stable to use
in next-step work. The one problematic item in terms of cluster placement (absent/delayed language)
was moved from the dysregulated cluster to the ASD-like cluster after expert statistical and clinical review
of constructs. Exploratory factor analysis showed that a 7-factor solution mapped well onto the majority
of clusters. The Scholastic, Neuropsychological, Overactive/Impulsive, and Eat/Sleep clusters showed
good agreement between clusters and factors, but signicant cross-loading was observed between the
ASD-like, Dysregulated behaviour and Mood/Anxiety clusters. Internal consistency for clusters and
factors was good to excellent for 5 of the original 7 clusters generated (except for the Mood/Anxiety and
Eat/Sleep clusters) and for 6 of the 7 factors identied (except for the Eat/Sleep factor). In slight contrast
to the feasibility studies that suggested 6 natural TAND clusters6,7, this larger-scale study identied 7
TAND clusters. In this study, the two biological (vegetative) items (sleeping/eating) grouped together,
whereas in the feasibility studies they were incorporated into the ASD-like cluster (diculties with eating)
Page 10/20
and the Mood/Anxiety cluster (sleep diculties). The remaining 6 clusters, however, were remarkably
similar to the ndings from the small-scale feasibility and replication studies6,7.
Seven Natural TAND Clusters
As outlined in Table 1, integration of the multivariate ndings led to our proposal of 7 natural clusters for
further validation and potential implementation.
Cluster 1. Scholastic cluster
The rst of the seven natural TAND clusters identied is a ‘Scholastic’ cluster indicating diculties
relating to reading, writing, spelling and mathematics. The items in the Scholastic cluster (rendered by
both cluster analysis and factor analysis) showed high bootstrapping, very high factor loadings and
alpha scores, indicating the close relationship and reliability between items. Findings highlight the need
for assessment in this cluster if an individual shows signs of diculty across any one of the four items.
Academic diculties are a common concern in TSC2,3,16,17 and not only affect school-aged children, but
also have long-term consequences in adulthood.
Cluster 2. Overactive/Impulsive cluster
Both cluster analysis and factor analysis includes overactivity, restlessness, and impulsivity in this
cluster. Bootstrapping and internal consistency were high, indicative of the reliability of items and how
they group together. This cluster appears clinically meaningful given the high rates of Attention Decit
Hyperactivity Disorder (ADHD) reported in TSC16–18. However, it is of interest that the cluster did not
include attentional diculties, which were grouped in the neuropsychological cluster. This may suggest
ADHD in TSC to be more typically of the ‘predominantly hyperactive/impulsive subtype’ or could suggest
that there may be differential pathways to the attentional and hyperactive/impulsive decits seen in TSC.
Cluster 3. Neuropsychological cluster
This cluster includes memory decits, disorientation, neuropsychological attention decits as well as
attention decits in daily life, dual task decits, executive decits, and visuo-spatial decits. Whilst visuo-
spatial decits were grouped within the ASD factor, cluster analysis grouped visuo-spatial decits with
the other neuropsychological skills. Bootstrapping supported the clustering with neuropsychological skills
but conrmed a frequent co-occurrence with the scholastic cluster. Based on the existing TSC literature,
the cluster maps very well onto the high rates of a range of neuropsychological attentional, executive, and
memory decits reported16,17,19−21.
Cluster 4. Mood/Anxiety cluster
Four items are included in this cluster – anxiety, depressed mood, mood swings and extreme shyness. We
observed that factor analysis included inexibility and sleep-related problems with the four other items.
However, bootstrapping classied these two items in the mood/anxiety cluster only 12–15% (sleep) and
Page 11/20
19% − 26% (inexibility) of the time. Given the cluster analysis and bootstrapping observed, inexibility
was therefore retained with ASD-like features, and ‘sleep diculties’ with the Eat/Sleep cluster. The four
mood/anxiety items (mood swings, anxiety, depressed mood, extreme shyness) are commonly seen in
children and adults with TSC16–18, 17,22.
Cluster 5. Dysregulated behaviour cluster
The dysregulated behaviour cluster includes aggressive outbursts, temper tantrums and self-injurious
behaviour. Cluster analysis also included absent/delayed language in the cluster, but, as outlined earlier,
bootstrapping did not support the robustness of this item in the cluster. One of the biggest concerns to
families is the high rate of ‘behaviors that challenge’ seen in TSC specically with regards to aggression
and temper tantrums, self-injury and damage to property16–18, 23,24. It was therefore of interest that a
specic and distinct cluster of dysregulated behaviours was identied here.
Cluster 6. Autism Spectrum Disorder (ASD)-like cluster
This natural TAND cluster includes six items - inexibility, unusual language, repetitive behaviour, poor
eye contact, peer diculty and delayed/absent language. As outlined above, initial cluster analysis did
not include absent/delayed language in the ASD-like cluster, but bootstrapping and factor analysis
suggested these characteristics to be more likely to co-occur with ASD-like rather than with other TAND
behaviours. TSC is one of the medical conditions most strongly associated with ASD and the symptoms
of ASD in TSC seems to map very well with symptoms observed in non-syndromic ASD2,3, 16–18,25,26. It
was therefore of interest to see the natural emergence of an ASD-like cluster of behaviours from a clinical
perspective.
Cluster 7. Eat/Sleep cluster
This cluster includes eating and sleeping diculties. Given that sleeping and eating are fundamental
biological/vegetative functions it was not surprising to see them cluster together. High rates of sleep
problems have been reported in individuals with TSC27, and decits in circadian rhythm are now
described in animal models of this disorder28. This cluster was not identied in the feasibility study, but
this much larger sample suggested that these concerns group together. Importantly, both sleep and eating
diculties cross-load with other clusters, underlining the fact that they often co-occur with other
neuropsychiatric diculties. However, the fact that they cluster independently suggests the need to
investigate these in their own right, not only in the context of other so-called co-morbid conditions.
Study Limitations
Firstly, anonymized TAND Checklist data were used here to identify potential natural TAND clusters, and
no other sources of information that may be relevant in cluster analysis or factor analysis, such as
clinical evaluations or neuropsychological assessments, were included. We acknowledge that it is
therefore theoretically possible that other clusters may be identied using different kinds of multi-level
Page 12/20
data. However, we specically wanted to use the TAND Checklist for this purpose, given that it is a simple,
yet systematic and freely available tool that could easily be implemented in real-life settings around the
globe. Pilot validation of the TAND Checklist29 indicated that the TAND Checklist was a valid tool in
extrapolating multi-level neuropsychiatric manifestations in TSC. It would be important to include
external validation of the natural TAND cluster ndings based on TAND Checklist data in relation to
expert clinical data as a logical next step. Secondly, we acknowledge that all the data were ‘lifetime’ data.
We have therefore to date not examined the developmental pattern of natural TAND clusters, which may
have a more dynamic nature than captured here. However, our ndings should allow longitudinal
examination of natural TAND clusters in future large-scale studies. This should also include examination
of the association between age, gender, intellectual ability and other potential correlates of TAND. Thirdly,
the current TAND Checklist collects data in a dichotomous fashion. The study was therefore not able to
explore the subtleties of severity that may be important to examine in natural TAND clusters in future.
Development of a quantied version of the TAND Checklist is currently underway.
Conclusions
Based on the largest collection of TAND Checklist data to date, our analyses generated seven natural
TAND clusters that were statistically robust and had good clinical face validity. We therefore propose the
identication of these clusters to be a useful rst step as a guide to further assessment and treatment
options in clinical practice. Next steps could include the development of targeted teaching and training of
professionals and individuals with TSC about the seven natural TAND clusters, and identication of
appropriate evidence-based resources and interventions that map onto these clusters. At a scientic level,
we propose that identication of these naturally-occurring clustering of the neuropsychiatric phenotype of
TSC may be the rst step towards a more dimensional, data-driven approach to the study of the etiology
and treatments (molecular and otherwise) 2,17,30 of individuals with TSC and related neurodevelopmental
disorders.
Declarations
Ethics approval and consent to participate
All procedures contributing to this work complied with the ethical standards of the relevant national and
international committees on human experimentation and with the Helsinki Declaration of 1975, as
revised in 2008. The protocol was peer-reviewed in the Department of Psychiatry at the University of Cape
Town and submitted for ethical approval by the Faculty of Health Sciences, Human Research Ethics
Committee (Ethics Ref 340/2015). Additional study sites obtained ethical approval or waivers from their
respective HREC/IRB bodies.
Consent for publication
Not applicable
Page 13/20
Availability of data and materials
The data that support the ndings of this study are available from the corresponding author [PJdV] upon
reasonable request.Full algebraic details of analysis methodologies are available in the supplemental
data of Leclezio, Gardner-Lubbe & de Vries6.
Competing interests
This work was supported by a Ph.D. studentship from the Tuberous Sclerosis Association in the United
Kingdom (Grant: 2014-S01) (LL), a project grant from the King Baudoin Foundation Fund Dr & Mrs
Charles Tournay-Dubisson to PJdV and AJ (2019-J1120010-213544), and by the National Research
Foundation and Struengmann Fund, University of Cape Town (to PJdV). The funding bodies played no
role in the design of the study in collection, analysis, and interpretation of data, or in writing the
manuscript. PJdV was a study steering committee member of three phase III trials sponsored by Novartis.
PJdV and AJ are on the scientic advisory group of the TOSCA international disease registry sponsored
by Novartis. MS reports grant support from Novartis, Roche, Biogen, Astellas, Aeovian, Bridgebio, Aucta
and Quadrant Biosciences unrelated to this project. He has served on Scientic Advisory Boards for
Roche, Celgene, Regenxbio, Alkermes and Takeda.
Funding
The work was funded through a PhD studentship from the Tuberous Sclerosis Association (UK) for the
studentship to LL (Grant: 2014-S01), a project grant from the King Baudouin Foundation Fund Dr & Mrs
Charles Tournay-Dubisson to PJdV and AJ (2019-J1120010-213544), the National Research Foundation
for incentive funding to PJdV, and the Struengmann Fund for endowment of the professorship to PJdV.
DK and MS are supported by the Developmental Synaptopathies Consortium (NIH U54-NS092090), which
is a part of the National Center for Advancing Translational Sciences (NCATS) Rare Diseases Clinical
Research Network (RDCRN). RDCRN is an initiative of the Oce of Rare Diseases Research (ORDR),
NCATS, funded through collaboration between NCATS, National Institute of Mental Health, NINDS and
National Institute of Child Health and Human Development (NICHD). None of the funders had any role in
study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Authors’ contributions
PdV and LL generated the idea of natural TAND clusters. LL collected and collated data from all
participating sites. DK, MS, SS, LdW and AJ provided anonymized data from their TSC clinical research
samples. LL and SGL performed the data analysis and visualization of results. PdV, LL, SGL and AJ
contributed to interpretation of the data. All authors contributed to writing and reviewing the manuscript
and approved the nal manuscript.
Acknowledgements
Page 14/20
The authors thank the funders for nancial support of the project. We thank Molly Valle and colleagues
for conducting the TAND Checklist interviews. Finally, the authors thank all study participants.
Note
After completion of preparation of this manuscript, Dr Loren Leclezio sadly passed away. In honour of her
work, we have opted to retain her as a co-author.
References
1. Northrup H, Krueger DA, International Tuberous Sclerosis Complex Consensus Group. Tuberous
sclerosis complex diagnostic criteria update: recommendations of the 2012 international Tuberous
Sclerosis Complex Consensus Conference.
Pediatr Neurol.
2013; 49 :243-254.
2. Curatolo P, Moavero R, de Vries PJ. Neurological and neuropsychiatric aspects of tuberous sclerosis
complex.
Lancet Neurol.
2015; 14: 733-745.
3. de Vries PJ, Whittemore VH, Leclezio L
et al.
Tuberous sclerosis associated neuropsychiatric
disorders (TAND) and the TAND Checklist.
Pediatr Neurol.
2015; 52: 25-35.
4. Leclezio L, de Vries PJ. Advances in the treatment of tuberous sclerosis complex.
Curr Opin
Psychiatry
2015; 28: 113-120.
5. Wulffaert J, van Berckelaer-Onnes I, Kroonenberg P, Scholte E, Bhuiyan Z, Hennekam R. Simultaneous
analysis of the behavioural phenotype, physical factors, and parenting stress in people with Cornelia
de Lange syndrome.
J Intellect Disabil Res.
2009; 53: 604-619.
. Leclezio L, Gardner-Lubbe S, de Vries PJ. Is It Feasible to Identify Natural Clusters of TSC-Associated
Neuropsychiatric Disorders (TAND)?
Pediatr Neurol.
2018; 81: 38-44.
7. de Vries P.J., Belousiva E., Benedik M., et al., Natural clusters of tuberous sclerosis complex (TSC)-
associated neuropsychiatric disorders (TAND): new ndings from the TOSCA TAND research project.
J Neurodev Disord.
2020; 12: 24.
. Cramér H. Mathematical methods of statistics. Princeton: Princeton University Press; 1946.
9. R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R
Foundation for Statistical Computing; 2017. https://www.R-project.org/.
10. Maechler, M., Rousseeuw, P., Struyf, A., Hubert, M., Hornik, K. cluster: Cluster Analysis Basics and
Extensions. R package version 2.0.5.; 2016. https://CRAN.R-project.org/package=cluster.
11. Efron B, Tibshirani RJ.
An introduction to the bootstrap
. Boca Raton: Chapman & Hall/CRC; 1993.
https://cds.cern.ch/record/526679/les/0412042312_TOC.pdf
12. Revelle W. psych: Procedures for Personality and Psychological Research, Evanston, Illinois:
Northwestern University; 2017. https://CRAN.R-project.org/package=psych Version = 1.7.5.
13. Lorenzo-Seva U, ten Berge JMF. Tucker's congruence coecient as a meaningful index of factor
similarity.
Methodology.
2006;2:57-64.
Page 15/20
14. Gower JC, Dijksterhuis GB.
Procrustes problems.
Oxford Statistical Science Series, 30. Oxford, UK:
Oxford University Press; 2004.http://www.oup.com/uk/catalogue/?ci=9780198510581
15. Cronbach LJ. Coecient alpha and the internal structure of tests.
Psychometrika
.1951; 16(3): 297-
334.
1. Prather P, de Vries PJ. Behavioral and cognitive aspects of tuberous sclerosis complex.
J Child Neurol
2004; 19: 666-674.
17. de Vries PJ, Wilde L, de Vries MC, Moavero R, Pearson DA, Curatolo P. A clinical update on tuberous
sclerosis complex-associated neuropsychiatric disorders (TAND).
Am J Med Genet.
2018; 178C: 309-
320.
1. de Vries PJ, Hunt A, Bolton PF. The psychopathologies of children and adolescents with tuberous
sclerosis complex (TSC): a postal survey of UK families.
Eur Child Adolesc Psychiatry
2007; 16: 16-
24.
19. de Vries PJ, Gardiner J, Bolton PF. Neuropsychological attention decits in tuberous sclerosis
complex (TSC).
Am J Med Genet.
2009; 149A: 387-395.
20. Ridler K, Suckling J, Higgins NJ
et al.
Neuroanatomical correlates of memory decits in tuberous
sclerosis complex.
Cereb Cortex
2007; 17: 261-271.
21. Tierney KM, McCartney DL, Serfontein JR, de Vries PJ. Neuropsychological attention skills and
related behaviours in adults with tuberous sclerosis complex.
Behav Genet.
2011; 41: 437-444.
22. Lewis JC, Thomas HV, Murphy KC, Sampson JR. Genotype and psychological phenotype in tuberous
sclerosis.
J Med Genet
2004; 41: 203-207.
23. Eden KE, de Vries PJ, Moss J, Richards C, Oliver C. Self-injury and aggression in tuberous sclerosis
complex: cross syndrome comparison and associated risk markers.
J Neurodev Disord
2014; 6: 10.
24. Wilde L, Eden K, de Vries P, Moss J, Welham A, Oliver C. Self-injury and aggression in adults with
tuberous sclerosis complex: Frequency, associated person characteristics, and implications for
assessment.
Res Dev Disabil
2017; 64: 119-130.
25. Bolton PF, Park RJ, Higgins JN, Griths PD, Pickles A. Neuro-epileptic determinants of autism
spectrum disorders in tuberous sclerosis complex.
Brain
2002; 125: 1247-1255.
2. Jeste SS, Varcin KJ, Hellemann GS, Gulsrud AC, Bhatt R, Kasari C, Wu J, Sahin M, Nelson CA III.
Symptom proles of autism spectrum disorder in tuberous sclerosis complex.
Neurol.
2016; 87(8):
766-772.
27. van Eeghen AM, Numis AI, Staley BA, Therrien SE, Thibert RL, Thiele EA. Characterizing sleep
disorders of adults with tuberous sclerosis complex: a questionnaire-based study and review.
Epilepsy Behav.
2011; 20: 68-74.
2. Lipton JO, Boyle LM, Yuan ED
et al.
Aberrant Proteostasis of BMAL1 Underlies Circadian
Abnormalities in a Paradigmatic mTOR-opathy.
Cell Rep
2017; 20: 868-880.
29. Leclezio L, Jansen A, Whittemore VH, de Vries PJ. Pilot validation of the tuberous sclerosis-
associated neuropsychiatric disorders (TAND) checklist.
Pediatr Neurol.
2015; 52: 16-24.
Page 16/20
30. de Vries PJ. Targeted treatments for cognitive and neurodevelopmental disorders in tuberous
sclerosis complex.
Neurotherapeutics
2010; 7: 275-282.
Figures
Figure 1
Schematic overview of data analysis in the study.
Page 17/20
Figure 2
Cluster analysis ndings of the study (n = 453). The gure shows the dendrogram illustrating the 7
natural TAND clusters generated with Ward’s cluster analysis method.
Page 18/20
Figure 3
1000-Fold Bootstrapping applied to WARD’s cluster analysis. Results are expressed as the proportion of
time that two individual items cluster together.
Page 19/20
Figure 4
Factor analysis ndings of the study (n = 453). The gure shows the seven-factor solution generated by
exploratory factor analysis (EFA).
Page 20/20
Figure 5
Relationship between natural TAND Clusters and Factor Analysis. Results show two highly distinct
clusters (Scholastic and Overactive/Impulsive), three fairly distinct clusters (Mood/Anxiety,
Neuropsychological, Eat/Sleep), and two clusters showing more signicant cross-loading (ASD-like and
Dysregulated behaviour).
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Background: Tuberous sclerosis complex (TSC)-associated neuropsychiatric disorders (TAND) have unique, individual patterns that pose significant challenges for diagnosis, psycho-education, and intervention planning. A recent study suggested that it may be feasible to use TAND Checklist data and data-driven methods to generate natural TAND clusters. However, the study had a small sample size and data from only two countries. Here, we investigated the replicability of identifying natural TAND clusters from a larger and more diverse sample from the TOSCA study. Methods: As part of the TOSCA international TSC registry study, this embedded research project collected TAND Checklist data from individuals with TSC. Correlation coefficients were calculated for TAND variables to generate a correlation matrix. Hierarchical cluster and factor analysis methods were used for data reduction and identification of natural TAND clusters. Results: A total of 85 individuals with TSC (female:male, 40:45) from 7 countries were enrolled. Cluster analysis grouped the TAND variables into 6 clusters: a scholastic cluster (reading, writing, spelling, mathematics, visuo-spatial difficulties, disorientation), a hyperactive/impulsive cluster (hyperactivity, impulsivity, self-injurious behavior), a mood/anxiety cluster (anxiety, depressed mood, sleep difficulties, shyness), a neuropsychological cluster (attention/concentration difficulties, memory, attention, dual/multi-tasking, executive skills deficits), a dysregulated behavior cluster (mood swings, aggressive outbursts, temper tantrums), and an autism spectrum disorder (ASD)-like cluster (delayed language, poor eye contact, repetitive behaviors, unusual use of language, inflexibility, difficulties associated with eating). The natural clusters mapped reasonably well onto the six-factor solution generated. Comparison between cluster and factor solutions from this study and the earlier feasibility study showed significant similarity, particularly in cluster solutions. Conclusions: Results from this TOSCA research project in an independent international data set showed that the combination of cluster analysis and factor analysis may be able to identify clinically meaningful natural TAND clusters. Findings were remarkably similar to those identified in the earlier feasibility study, supporting the potential robustness of these natural TAND clusters. Further steps should include examination of larger samples, investigation of internal consistency, and evaluation of the robustness of the proposed natural clusters.
Article
Full-text available
Tuberous sclerosis complex (TSC) is associated with a wide range of behavioral, psychiatric, intellectual, academic, neuropsychological, and psychosocial difficulties, which are often underdiagnosed and undertreated. Here, we present a clinical update on TSC‐associated neuropsychiatric disorders, abbreviated as “TAND,” to guide screening, diagnosis, and treatment in practice. The review is aimed at clinical geneticists, genetic counselors, pediatricians, and all generalists involved in the assessment and treatment of children, adolescents and adults with TSC, and related disorders. The review starts with a summary of the construct and levels of TAND, before presenting up‐to‐date information about each level of investigation. The review concludes with a synopsis of current and future TAND research.
Article
Full-text available
What this paper adds: This paper adds specific examination of behavioural difficulties in adults with tuberous sclerosis complex who also have intellectual disability, a population at heightened risk of adverse behavioural outcomes which has received limited focussed examination to date. Findings support existing suggestions that there is relatively high risk for both self-injury and aggression, and provide novel insight into characteristics that may be associated with the presence of these behaviours.
Article
Background: Tuberous sclerosis complex (TSC) is a genetic disorder with multisystem involvement. The lifetime prevalence of TSC-Associated Neuropsychiatric Disorders (TAND) is in the region of 90% in an apparently unique, individual pattern. This "uniqueness" poses significant challenges for diagnosis, psycho-education, and intervention planning. To date, no studies have explored whether there may be natural clusters of TAND. The purpose of this feasibility study was (1) to investigate the practicability of identifying natural TAND clusters, and (2) to identify appropriate multivariate data analysis techniques for larger-scale studies. Methods: TAND Checklist data were collected from 56 individuals with a clinical diagnosis of TSC (n = 20 from South Africa; n = 36 from Australia). Using R, the open-source statistical platform, mean squared contingency coefficients were calculated to produce a correlation matrix, and various cluster analyses and exploratory factor analysis were examined. Results: Ward's method rendered six TAND clusters with good face validity and significant convergence with a six-factor exploratory factor analysis solution. The "bottom-up" data-driven strategies identified a "scholastic" cluster of TAND manifestations, an "autism spectrum disorder-like" cluster, a "dysregulated behavior" cluster, a "neuropsychological" cluster, a "hyperactive/impulsive" cluster, and a "mixed/mood" cluster. Conclusions: These feasibility results suggest that a combination of cluster analysis and exploratory factor analysis methods may be able to identify clinically meaningful natural TAND clusters. Findings require replication and expansion in larger dataset, and could include quantification of cluster or factor scores at an individual level.
Article
Tuberous sclerosis complex (TSC) is a neurodevelopmental disorder characterized by mutations in either the TSC1 or TSC2 genes, whose products form a critical inhibitor of the mechanistic target of rapamycin (mTOR). Loss of TSC1/2 gene function renders an mTOR-overactivated state. Clinically, TSC manifests with epilepsy, intellectual disability, autism, and sleep dysfunction. Here, we report that mouse models of TSC have abnormal circadian rhythms. We show that mTOR regulates the proteostasis of the core clock protein BMAL1, affecting its translation, degradation, and subcellular localization. This results in elevated levels of BMAL1 and a dysfunctional clock that displays abnormal timekeeping under constant conditions and exaggerated responses to phase resetting. Genetically lowering the dose of BMAL1 rescues circadian behavioral phenotypes in TSC mouse models. These findings indicate that BMAL1 deregulation is a feature of the mTOR-activated state and suggest a molecular mechanism for mitigating circadian phenotypes in a neurodevelopmental disorder.
Article
Objective: To determine the extent to which deficits associated with autism spectrum disorder (ASD) in toddlers with tuberous sclerosis complex (TSC) overlap with those in toddlers with nonsyndromic ASD (nsASD) and to examine cognitive function and epilepsy severity in toddlers with TSC and comorbid ASD. This is the endpoint analysis from a longitudinal investigation of ASD risk factors in children with TSC. Methods: Measures included the Autism Diagnostic Observation Schedule (ADOS), the Mullen Scales of Early Learning, and clinical epilepsy variables. A repeated-measures analysis of variance was performed with between-subjects factor of group (typically developing, TSC/no ASD, TSC/ASD, nsASD) and within-subjects factors of individual ADOS item scores in the social communication and repetitive behavior/restricted interest domains. Within the TSC group, comparisons of epilepsy characteristics and cognitive domains were performed using independent-samples t tests. Results: Children with TSC/ASD demonstrated a profile of social communication impairment that had complete convergence with nsASD. Measured social communication impairments included gestures, pointing, eye contact, responsive social smile, and shared enjoyment. This convergence was observed despite the high comorbidity between ASD and cognitive impairment in TSC. Conclusions: This study supports the clinical diagnosis of ASD in young children with TSC and demonstrates remarkable convergence of autism symptoms between TSC/ASD and nsASD. Our results strongly suggest the need for early intervention in toddlers with TSC, with treatment strategies targeting social communication function as well as broader developmental domains, before the onset of autism symptoms.
Article
Tuberous sclerosis (also known as tuberous sclerosis complex [TSC]) is a multisystem genetic disorder that affects almost every organ in the body. Mutations in the TSC1 or TSC2 genes lead to disruption of the TSC1-TSC2 intracellular protein complex, causing overactivation of the mammalian target of rapamycin (mTOR) protein complex. The surveillance and management guidelines and clinical criteria for tuberous sclerosis were revised in 2012, and mTOR inhibitors are now recommended as treatment options for subependymal giant cell astrocytomas and renal angiomyolipomas-two common features of the disease. However, most morbidity and mortality caused by tuberous sclerosis is associated with neurological and neuropsychiatric manifestations. Treatment of epilepsy associated with tuberous sclerosis remains a major challenge, with more than 60% of patients having ongoing seizures. Tuberous-sclerosis-associated neuropsychiatric disorders (TAND) are multilevel and occur in most individuals with the disorder, but are rarely assessed and treated. Clinical trials of mTOR inhibitors to treat seizures and TAND are underway. Management of the neurological and neuropsychiatric manifestations of the disorder should be coordinated with treatment of other organ systems. In view of the age-related expression of manifestations from infancy to adulthood, continuity of clinical care and ongoing monitoring is paramount, and particular attention is needed to plan transition of patient care from childhood to adult services. Copyright © 2015 Elsevier Ltd. All rights reserved.
Article
Tuberous sclerosis complex (TSC) is a multisystem genetic disorder with physical and neuropsychiatric manifestations and significant research progress has been made in recent years. Here, we focus on the key advances over the last 18 months. Three main themes were identified in the literature. Firstly, the diagnostic criteria and surveillance guidelines for TSC were revised, incorporating a genetic criterion alongside clinical criteria, and making a positive step towards evidence-based treatment of TSC. Secondly, a new term - TSC-associated neuropsychiatric disorders (TAND) - was introduced as an umbrella term for all possible neuropsychiatric difficulties seen in TSC, and a TAND Checklist was developed as a screening tool. Thirdly, the risks and benefits of molecularly targeted treatments of the neuropsychiatric manifestations of TSC are being debated. The updated diagnostic criteria and management guidelines, the new concept of TAND and the TAND Checklist should lead to significant improvements in the quality of care for individuals with TSC. The promise of mammalian target of rapamycin inhibitors and other molecular treatments are still to be confirmed. We suggest that great care should be taken to identify 'optimal mammalian target of rapamycin signalling' in the therapeutic approach to the neuropsychiatric features of the disorder.