Download Beata_Tick_Thesis_Feb_2016

The aetiology of Autism Spectrum Disorders
and the relationship with associated
psychiatric difficulties
Beata Magdalena Tick, MSc
Social, Genetic and Developmental Psychiatry MRC Centre
Institute of Psychiatry, Psychology and Neuroscience
King’s College London
A thesis submitted to King’s College London for the degree of
Doctor of Philosophy (PhD)
1
Abstract of Thesis
Studies of the aetiology of Autism Spectrum Disorders (ASD) have been limited so far due to
the rare nature of samples allowing detection of the total genetic and environmental
influences. Most studies to date point towards strong genetic influences. However, two recent
reports found that shared environmental influences mattered more than genetics.
The first two empirical chapters of this thesis describe a set of innovative analyses
designed to provide new evidence on the aetiology of ASD, and to make sense of the existing
contradictory evidence. The new evidence comes from the Social Relationships Study (SRS).
For the first time, a single model was fitted to data recognising the Broad Spectrum as an ASD
subgroup. The results revealed strong genetic influences and little support for shared
environmental factors. The second empirical chapter reports on the first meta-analysis of ASD
twin studies to redress which familial effects are more important: genetic or shared
environmental factors? Again, the results showed the importance of strong genetic influences.
Moreover, the strength of shared environmental factors was dependent on the prevalence
values used to correct for selecting individuals on ASD affection status. These results re-affirm
that ASD is under strong genetic influences and that detecting shared environmental factors
may be a statistical artefact.
The second half of this thesis deals with the aetiology of associated psychiatric
difficulties in autism, the study of which has gained significant momentum in the last decade.
To extend this work, the third and fourth empirical chapters utilise data from the SRS as well as
data on autism traits from the Twins Early Development Study (TEDS). Similar aetiology
profiles were found for comorbid Emotional symptoms and Hyperactivity across the two
samples. However, the aetiology of comorbid Conduct problems differed and appeared to be
dependent on the severity of autism manifestations. This could possibly be explained by the
fact that these problems are conceptualised differently in children with ASD and those with
milder manifestations.
2
There are two main implications of the findings from this thesis for future research.
First, the next-generation sequencing studies ought to increase their efforts to include samples
across the whole autism spectrum as the genetic risk is very likely to be shared across clinical
cases, those showing Broad Phenotype and individuals with mild autistic behaviours. Second,
validation is needed of the observed relationships of comorbid psychiatric difficulties in ASD,
using independent cohorts; such data are important for the future design of ASD-appropriate
comorbidity measures and interventions.
3
Statement of Work
This thesis utilised data from two different studies. Chapters 3, 5 and 6 used data from
the SRS, headed by Professor Francesca Happé and Professor Patrick Bolton [MRC grant
G0500870] and Chapters 3 and 6 used data from the TEDS headed by Professor Robert Plomin
[MRC grant G0901245, previously G0500079]. I developed a project proposal to obtain the
TEDS data for analyses presented in Chapter 6. Throughout this thesis, I had the responsibility
for the development of hypotheses, data quality checks and twin analyses. The twin modelfitting was carried out entirely by me under supervision of Dr Frühling Rijsdijk.
Chapter 4 presents meta-analysis on the combined dataset of all autism twin studies
prepared by me and Dr Frühling Rijsdijk. This involved systematic review of all published autism
twin studies to extract summary data on number of concordant and discordant twin pairs, as
well as determining mode of ascertainment across included studies. Although meta-analysis of
this data required using scripts in the classic Mx programme (led by Dr Frühling Rijsdijk), I have
a very good conceptual understanding of the method used.
Finally, any written work and interpretation of results presented in this thesis is solely
designed by me, inclusive of manuscripts submitted to peer-reviewed journals for publication.
Supervision of these aspects of the thesis was equally shared between Dr Frühling Rijsdijk and
Professor Francesca Happé.
4
Acknowledgments
In the first instance, I would like to thank my first supervisor – Dr Frühling Rijsdijk, as
without her I would have never arrived where I am now. Frühling is the most passionate and
incredible statistician I have ever met and I really hope that both of these traits have rubbed off
on me, at least in small percentage. She is the top expert in her field and I was very lucky to have
met her at the airport to travel together to twin-modeling course in Boulder. Since then, she
took me under her wing and guided, inspired and supported my work and made me the
academic I am today.
A huge thank you goes to Professor Francesca Happé as my second supervisor, as
without her expertise on autism this thesis would not be possible. Franky always has a good
word of advice, will find the time for you in her incredibly busy schedule as the director of the
SGDP and will even make you tea and offer biscuits. She is truly inspiring in combining her roles
as a world-recognised researcher and a mother to three children. It has been an incredible
honour to spend my time at the SGDP under Frühling’s and Franky’s supervision. Without the
Medical Research Council’s funding and the SRS and TEDS participants, we would be able to
embark on this amazing journey together. Thank you!
I have been very fortunate to share the last three years with Steven and he has
supported me during the PhD on many occasions. As a scientist himself, he always understood
what it takes to complete a PhD and it is his deep friendship, advice, compassion and most
importantly his love that made it all a lot more enjoyable. Thank you, my darling, I am eternally
grateful!
I would like to thank my mother and wonderful siblings – Marcin, Arleta and Adrian, for
cheering me on and their words of praise. To Pawel, Sebastian and Bartek for entertaining and
injecting our family with joy. A big thank you to the Kiddles: Pete, Sue and Emma. You never
stopped believing in my ability and are so welcoming at every opportunity. I am very lucky to
be a part of your family.
5
Rachel is a very special friend and her continuous support, kindness and optimism
always renewed my motivation. She supplied me with (brilliant) non-scientific books that
distracted me, when necessary, and the Maccabees to get me through my R analyses. She
shared with me so many wonderful moments when pregnant and as a new mom. Together with
James, Sebastian and cat Tora always made joint dinners a success. Thank you for your
continuous support!
Gratuities to the ladies from my year (in no particular order): Anna, Ruth, Helen,
Susanna, Rebecca, Vicki and Magda. You are the most diverse bunch I have ever met and I hope
that we continue to remain in contact, one way or another. Especially, I thank Monika my office
mate, with whom I shared every step of the way! We have had an unforgettable time together.
Thank you to all the remaining SGDP friends, you all know who you are.
Thanks to my non-academic friends, Kirsty and Steve Sinky, for ‘come dine with me’,
patiently listening about science, the board games and the snowboarding trips. They are also
part of the ‘Plover Way’ quiz team, along with Helen, Fran, Simon, Rowan, Chris, Ziggy and
Jenny - I thank you all for the fun and the wins we made! Steve’s band mates: Rowan, Loz, Steve
Stall and Nick as the Panda Party, whom provided such fun evenings when signing about Arnold
Strong and Bonny and Clyde. If I decide on a new career, you must interview me as your
manager… Also, thanks to Christina for being so friendly and ready to exchange experiences on
completing a PhD vs. becoming a lawyer.
The last thank you goes to my past bosses, Martin and Alison. I will never forget your
support when I decided to pursue the undergraduate and postgraduate education. My
undergraduate supervisor, Naz, opened many doors for me and encouraged to be fearless,
hardworking and not settling for the mediocre. It is because of Naz that I discovered my passion
for statistics and teaching.
6
Publications
Published (* equally contributing joint first author):
Colvert, E.*, Tick, B.*, McEwen, F., Stewart, C., Curran, S., Woodhouse, E., Gillan, N., Hallett,
V., Lietz, S., Garnett, T., Ronald, A., Plomin, R., Rijsdijk, F., Happé, F., & Bolton, P. Heritability
of Autism Spectrum Disorder in a UK Population-Based Twin Sample, (2015). JAMA Psychiatry,
72 (5), 415-423.
Tick, B., Colvert, E., McEwen, F., Stewart, C., Woodhouse, E., Gillan, N., Hallett, V., Lietz, S.,
Garnett, T., Simonoff, E., Ronald, A., Bolton, P., Happé, F., & Rijsdijk, F. Autism Spectrum
Disorders and other mental health problems: exploring etiological overlaps and phenotypic
causal associations, (2016). Journal of the American Academy of Child and Adolescent Psychiatry,
55 (2), 106–113.
Tick, B., Bolton, P., Happé, F., Rutter, M. & Rijsdijk, F. Heritability of Autism Spectrum
Disorders: a meta-analysis of twin studies, (2016). Journal of Child Psychology and Psychiatry,
57 (5), 585–595.
O'Nions, E.*, Tick, B.*, Rijsdijk, F., Happé, F., Plomin, R., Ronald, A., & Viding, E. Examining
the genetic and environmental associations between autistic social and communication deficits
and psychopathic callous-unemotional traits, (2015). PLoS One 10(9): e0134331.
7
Table of Contents
Abstract of Thesis .................................................................................................................... 2
Statement of Work .................................................................................................................. 4
Acknowledgments....................................................................................................................5
Publications .............................................................................................................................. 7
Table of Contents .................................................................................................................... 8
List of Tables .......................................................................................................................... 14
List of Figures .........................................................................................................................16
Chapter 1 General Introduction to Autism Spectrum Disorders .............................................. 17
1.1 Overview ....................................................................................................................... 17
1.2 Autism Spectrum Disorders: concept and diagnosis ...................................................... 17
1.3 Beyond the diagnostic criteria for ASD – the Broader Autism Phenotype ..................... 24
1.4 Prevalence of ASD in the general population ................................................................ 26
1.4.1 Autism/Autistic Disorder ......................................................................................... 27
1.4.2 Asperger Syndrome (AS) and Childhood Disintegrative Disorder (CDD) ................ 28
1.4.3 Pervasive Developmental Disorder-Not Otherwise Specified (PDD-NOS).............. 28
1.4.4 Combined Autism Spectrum Disorders (ASD) ........................................................ 28
1.4.5 The Broader Autism Phenotype (BAP) ................................................................... 29
1.5 Is autism heritable? ........................................................................................................30
1.5.1 Overview .................................................................................................................30
1.5.2 What is heritability?.................................................................................................30
1.5.3 Deriving heritability for categorical disorders .......................................................... 31
1.5.4 Ascertainment Correction for Selected Samples ..................................................... 32
1.5.5 Heritability of autism ............................................................................................... 33
8
1.6 Molecular genetic studies of autism...............................................................................38
1.6.1 Common genetic variants .......................................................................................39
1.6.2 Rare genetic variants.............................................................................................. 40
1.6.3 Genetic effects in simplex and multiplex families ................................................... 42
1.7 Environmental factors in autism ................................................................................... 44
1.7.1 Effects of chemical toxins ....................................................................................... 44
1.7.2 Vaccinations ............................................................................................................45
1.7.3 Pregnancy-related complications ........................................................................... 46
1.7.4 The hormonal in utero influences ............................................................................. 47
1.8 Psychiatric comorbidity in autism ..................................................................................50
1.8.1 Comorbidity – an overview ......................................................................................50
1.8.2 Causal models of comorbidity ................................................................................. 51
1.8.3 Autism and associated psychiatric difficulties..........................................................52
1.8.4 Early studies ............................................................................................................ 53
1.8.5 Modern studies .......................................................................................................54
1.9 Thesis goals and structure .............................................................................................56
1.9.1 Aim 1 ....................................................................................................................... 57
1.9.2 Aim 2....................................................................................................................... 57
1.9.3 Aim 3 ....................................................................................................................... 57
1.9.4 Aim 4.......................................................................................................................58
Chapter 2 Methods .................................................................................................................59
2.1 Overview .......................................................................................................................59
2.2 Samples.........................................................................................................................59
2.2.1 The Twins Early Development Study (TEDS) sample...............................................59
9
2.2.2 The Social Relationships Study (SRS) sample......................................................... 60
2.3 Measures .......................................................................................................................63
2.3.1 Autism traits - Childhood Autism Spectrum Test (CAST) .........................................63
2.3.2 The Development and Well-being Assessment (DAWBA) ...................................... 64
2.3.3 Autism Diagnostic Interview-Revised (ADI-R) .........................................................65
2.3.4 Autism Diagnostic Observation Schedule (ADOS) ................................................. 66
2.3.5 Strengths and Difficulties Questionnaire (SDQ) ...................................................... 67
2.4 Statistical Approaches .................................................................................................. 68
2.4.1 The Biometric Theory of Inheritance – from single-gene to polygenic model ......... 68
2.4.2 The Classical Twin Method...................................................................................... 70
2.4.3 The Univariate ACE Model ...................................................................................... 71
2.4.4 Estimating ACE of the liability for a disorder – model-fitting to ordinal data........... 73
2.4.5 The Multivariate Approach...................................................................................... 74
2.4.6 The Cholesky Decomposition ................................................................................. 75
2.4.7 The Gaussian Decomposition .................................................................................. 76
2.4.8 The Bivariate Genetic Model (Correlated-Factors Interpretation) ........................... 77
2.4.9 The Bivariate Genetic Model - combining continuous and ordinal data ................... 78
2.4.10 The Sex-Limitation Model ..................................................................................... 79
2.4.11 The Direction of Causation (DoC) Model ............................................................... 81
2.5 Assumptions and implications of the twin design ......................................................... 82
2.5.1 Assortative Mating ................................................................................................. 82
2.5.2 Equal Environments assumption .............................................................................83
2.5.3 Gene-Environment (G x E) association ................................................................... 84
2.5.4 Generalising from twin to population samples ....................................................... 86
10
Chapter 3 Heritability of Autism Spectrum Disorder in a UK Population-Based Twin Sample . 87
3.1 Abstract ........................................................................................................................ 88
3.2 Introduction & Methods................................................................................................ 89
3.3 Results .......................................................................................................................... 92
3.4 Discussion......................................................................................................................93
3.5 Manuscript-specific references ......................................................................................95
3.6 Supplementary Online Materials ................................................................................... 97
Chapter 4 Heritability of Autism Spectrum Disorders: a meta-analysis of twin studies ........ 106
4.1 Abstract ....................................................................................................................... 106
4.2 Introduction ................................................................................................................ 107
4.3 Methods ...................................................................................................................... 109
4.3.1 Sample .................................................................................................................. 109
4.3.2 Statistical Analysis ................................................................................................ 114
4.3.3 Ascertainment Correction ..................................................................................... 114
4.4 Results......................................................................................................................... 116
4.4.1 Tetrachoric correlations ........................................................................................ 116
4.4.2 The ratio of discordant/concordant MZ and DZ twins ........................................... 118
4.4.3 A, C and E estimates ............................................................................................. 119
4.5 Discussion ................................................................................................................... 120
4.5.1 Main Findings and Conclusion ............................................................................... 120
Chapter 5 Autism Spectrum Disorders and other mental health problems: exploring etiological
overlaps and phenotypic causal associations ....................................................................... 124
5.1 Abstract ....................................................................................................................... 124
5.2 Introduction................................................................................................................. 125
5.3 Methods ...................................................................................................................... 127
5.3.1 Participants ........................................................................................................... 127
5.3.2 Measures ............................................................................................................... 128
11
5.3.3 Statistical Analyses................................................................................................ 129
5.3.4 Results .................................................................................................................. 131
5.4 Discussion ................................................................................................................... 136
5.4.1 Aetiology of ASD and SDQ problems .................................................................... 137
5.4.2 Direction of Causation........................................................................................... 138
5.4.3 Limitations ............................................................................................................ 140
5.4.4 Conclusions ........................................................................................................... 141
5.5 Supplementary Online Materials ................................................................................. 142
Chapter 6 The aetiology of autism traits and other mental health problems in a general
population twin sample ........................................................................................................ 144
6.1 Overview ..................................................................................................................... 144
6.1.1 Internalising problems and autism traits ............................................................... 145
6.1.2 Externalising problems and autism traits .............................................................. 147
6.2 Methods ...................................................................................................................... 150
6.2.1 Participants ........................................................................................................... 150
6.2.2 Measures .............................................................................................................. 150
6.3 Statistical Analyses ...................................................................................................... 152
6.3.1 Twin Correlations .................................................................................................. 152
6.3.2 Testing for sex differences – Sex Limitation Models .............................................. 152
6.3.3 The Bivariate Genetic (no sex-differences) Models ................................................ 155
6.3.4 The Direction of Causation (DoC) Models.............................................................. 156
6.4 Results ........................................................................................................................ 156
6.4.1 Descriptive Statistics ............................................................................................. 156
6.4.2 Phenotypic analyses ............................................................................................. 158
6.4.3 Sex-limitation analyses ......................................................................................... 160
12
6.4.4 Bivariate Genetic and Direction of Causation (DoC) no-sex differences Model Results
...................................................................................................................................... 163
6.5 Discussion ................................................................................................................... 167
6.5.1 Suggestive cut-offs findings .................................................................................. 167
6.5.2 Analyses of the total spectrum of CAST and SDQ measures ................................. 168
6.5.3 Diagnostic vs population sample results ................................................................ 169
Chapter 7 General Discussion ............................................................................................... 171
7.1 Summary of aims and findings ..................................................................................... 171
7.1.1 Re-evaluating the aetiology of ASD ....................................................................... 171
7.1.2 The relationship of ASD with associated psychiatric difficulties............................. 175
7.1.3 Limitations and future directions ........................................................................... 179
7.1.4 Final remarks ......................................................................................................... 180
References............................................................................................................................ 182
Appendices .......................................................................................................................... 204
13
List of Tables
Table 1-1 The diagnostic criteria in DSM 5 for Autism Spectrum Disorders. ........................... 20
Table 1-2 Summary of all twin studies of clinical autism published to date. ............................ 35
Table 1-3 Summary of all twin studies of autism traits published to date, adapted from Ronald
& Hoekstra, (2014). ................................................................................................................. 37
Table 2-1 The cut off scores for each of the diagnostic ADOS categories................................ 67
Table 2-2 Suggestive SDQ cut offs, adapted from Goodman, (1997). .................................... 68
Table 4-1 List of primary twin studies on ASD....................................................................... 111
Table 4-2 MZ and DZ twin correlations for individual studies and meta-analytic estimates based
on 6 configurations. .............................................................................................................. 118
Table 4-3 Estimates of the genetic and environmental variance components for individual
studies and meta-analytic estimates based on 6 configurations. .......................................... 122
Table 5-1 Bivariate MZ and DZ within-trait and cross-trait twin correlations of three SDQ
problems and ASD-BeD. ....................................................................................................... 134
Table 5-2 Phenotypic overlap due to genetic and environmental effects. ............................. 136
Table 5-3 Concordance/discordance on SDQ measures among MZ and DZ pairs, split by
concordance on ASD-BeD. ................................................................................................... 142
Table 5-4 Comparison of fit indices of the Bivariate AE model, the reciprocal and unidirectional
Causal models for SDQ problems and ASD-BeD................................................................... 143
Table 6-1 Percentage of males and females (low/high on CAST) meeting SDQ suggestive cut
offs. ...................................................................................................................................... 157
Table 6-2 Means (M) and standard deviation (SD, in bracket) for SDQ and CAST measures by
sex & zygosity groups. .......................................................................................................... 158
Table 6-3 Phenotypic results for the three SDQ problems and CAST. ................................... 159
Table 6-4 Model-fit indices for bivariate sex-limitation models fitted to each SDQ-CAST
combination. ........................................................................................................................ 161
14
Table 6-5 Estimates for each SDQ-CAST combination as estimated in the full heterogeneity
sex-limitation model and the best-fitting Variance Inequality model.................................... 162
Table 6-6 Phenotypic overlap due to A and E as estimated in the genetic bivariate model for
each SDQ-CAST combination............................................................................................... 165
Table 6-7 Model-fit indices for the bivariate genetic and Direction of Causation models fitted to
each SDQ-CAST combination............................................................................................... 165
15
List of Figures
Figure 2-1 The SRS sample selection, adapted from Colvert & Tick et al., (2015). .................. 62
Figure 2-2 Example of inheritance of Huntington’s disease. .................................................. 69
Figure 2-3 ACE Path diagram. ................................................................................................. 72
Figure 2-4 The liability-threshold model of disorders.............................................................. 73
Figure 2-5 The bivariate Cholesky decomposition path diagram for a twin pair. ..................... 76
Figure 2-6 The bivariate Gaussian decomposition path diagram for a twin pair. ..................... 77
Figure 2-7 The correlated-factors interpretation of the Cholesky decomposition. .................. 78
Figure 2-8 The bivariate genetic model on combined continuous-ordinal data. ...................... 79
Figure 2-9 The bivariate sex-limitation model including paths for DZ opposite-sex twins...... 80
Figure 2-10 The bivariate direction of causation model. ......................................................... 81
Figure 4-1 Meta-analytic tetrachoric correlations. ................................................................ 117
Figure 4-2 Ratio of concordant and discordant MZ/DZ twins across studies. ........................ 119
Figure 4-3 Forest plots. ......................................................................................................... 123
Figure 5-1 Means and Standard Error of the Mean bars. ....................................................... 131
Figure 5-2 Associated SDQ problems in the SRS sample. ..................................................... 133
Figure 5-3 Standardized estimates of the three AE bivariate models. ................................... 135
Figure 6-1 rA and rC calculated across the males and females in the opposite-sex DZ twin pairs,
adapted from Neale, Røysamb, & Jacobson, (2006). ............................................................ 153
Figure 6-2 Raw means (and Standard Error of the Mean bars) for SDQ problems across the
TEDS sample. ....................................................................................................................... 157
Figure 6-3 The best fitting (no sex-differences) models and derived parameter estimates. ..164
16
Chapter 1 General Introduction to Autism Spectrum Disorders
1.1 Overview
This chapter introduces the concept of autism, its diagnosis and prevalence, aetiology and
comorbidities. Background information and overview of literature that is relevant to the studies
presented in this thesis, are given, as well as a sketch of the chapters that will follow.
1.2 Autism Spectrum Disorders: concept and diagnosis
Autism, originating from the Greek word ‘autos’ meaning ‘self’, was first coined by Eugene
Bleuler in 1911 to describe the symptom of deliberate turning-away from the external world to
lead a solitary inner life (Bleuler, 1951). This feature was first described in the context of
schizophrenia and perhaps because of this behavioural overlap ‘infantile autism’ was initially
conceptualised as ‘childhood psychosis’ (Kolvin, 1971). However, the two were separated as
independent disorders in 1971 on the basis of core symptom differences, age of onset, familial
history, and most importantly, differential drug responsiveness (Kolvin, Ounsted, Humphrey,
& McNay, 1971).
Infantile autism (IA) was first formally reported as a neuropsychiatric condition by Leo
Kanner in 1943 (Kanner, 1968). Additionally, Hans Asperger reported on the schizophrenia-like
social withdrawal in children with pedantic and stereotyped speech, clumsiness and
obsessional behaviours (Asperger, 1944 translated by Frith, 1991). IA entered the third edition
of the Diagnostic and Statistical Manual (DSM III, American Psychiatric Association, 1980)
under the heading of Pervasive Developmental Disorders (PDD), along with Atypical Autism
(equal to PDD-NOS [Not Otherwise Specified], assigned when the child did not display all of
the IA behavioural criteria) and Childhood Onset PDD - capturing children who displayed IA
behaviours later than the 30-months cut off for IA diagnosis (but no later than 12 years old). The
revision to DSM III (DSM IIIR; American Psychiatric Association, 1987) introduced Autistic
Disorder with 16 item criteria (instead of 6 in DSM III) and Atypical Autism and Childhood Onset
17
PDD were combined under PDD-NOS. DSM IV (+DSM IV TR [text revision], (American
Psychiatric Association, 1994, 2000)) - the longest-lasting editions of the manual from 1994 to
2013 - further redefined Autistic disorder, and added Asperger Disorder (commonly known as
Asperger syndrome), Childhood Disintegrative Disorder (for late onset regression) and Rett
Syndrome. The DSM IV additions marked the emergence of the unofficial umbrella term of
Autism Spectrum Disorders (ASD) (Prior et al., 1998), discussed next. Note that throughout this
thesis the terms ‘autism’ and ‘ASD’ are used in different capacity. ‘Autism’ is used in the context
of autism behaviours found in clinical and non-clinical populations. ‘ASD’ is used in the context
of a clinical diagnosis.
Even though Kanner and Asperger revealed their observations around the same time,
it was Lorna Wing (1981) who first described the similarities of Kanner’s infantile autism and
Asperger’s ‘autistic psychopathy’. She alluded to the fact that they are not distinct, but
representations of differential levels of severity of the same disorder (Wing, 1981). The 1979
Camberwell study provided the first set of data in favour of an ‘autism spectrum’ within the
clinical population: of 35,000 children known to special education/clinical services, 17 children
met Kanner’s criteria for infantile autism, a few met Asperger syndrome criteria (number not
specified) and 74 children displayed a wide range of manifestations reflecting some of the
features of both syndromes (Wing & Gould, 1979). The authors concluded that their findings on
the heterogeneity of cardinal features of autism clearly demonstrated that autism is better
defined as a spectrum than as strict categories, with Kanner’s infantile autism only explaining a
fraction of cases on the clinical spectrum with social and communication difficulties.
In the 90’s, Christopher Gillberg in his comprehensive review commented on not
finding ‘a shred of evidence’ (page 815, (Gillberg, 1992)) for a clear separation of cardinal
features of Kanner’s infantile autism from Asperger syndrome, and that the latter was far more
common than Kanner’s infantile autism. He also made the point that ASD is seen in the
presence of all levels of intelligence, from severely intellectually impaired to average or above
average IQ. It must be noted, however, that IQ can also be a confounding factor influencing
18
one’s ability to recognise autistic traits as an observable behavioural entity (Skuse, 2007). This
is related to the idea that individuals with high levels of intelligence develop sufficient sociocognitive skills to compensate for inherited autistic vulnerabilities, leading to under-recognition
of autistic traits at the behavioural level. In comparison, children with intellectual disability are
less likely to develop such compensatory mechanisms, hence express readily observable
autistic behaviours and, as a result, receive clinical diagnosis (Skuse, 2007). Two studies
investigated whether the relationship between autistic traits and intelligence is underpinned by
common aetiological factors. A general population twin study revealed only a modest genetic
overlap during childhood suggesting genetic independence, within the normal range
(Hoekstra, Happé, Baron-Cohen, Ronald, 2010) and at the extremes, for ASD and IQ (Hoekstra,
Happé, Baron-Cohen, Ronald, 2009).
Autism awareness is ever increasing and led to knowledge that ASD diagnosis is a
lifelong condition (Matson, Mayville, Lott, Bielecki, & Logan, 2003). Males are affected more
than females, reported ratio of 4:1, with the caveat that the underdiagnosis in females may be
due to the stereotype of autism as a male disorder (Blumberg, Bramlett, Kogan, Schieve, &
Jones, 2013). ASD is quite often accompanied by co-occurring mental (Leyfer et al., 2006) and
physical health (Bauman, 2010) disorders.
The enormous efforts to uncover the aetiology of ASD at the molecular level have
begun to reap rewards (Geschwind, 2011), but the medical profession is still some distance from
the diagnosis based on specific genetic variants. Until then, they will continue to rely on
behavioural manifestations. Because of this realisation, psychiatrists and the research
community strive to improve their diagnostic practices to include constantly emerging research
evidence on the validity of diagnostic tools. By no means will this be an easy task, as arriving at
an autism diagnosis can be reached through hundreds of combinations of behavioural patterns
that are organised differently for each individual. This means that each individual went through
a different developmental pattern made of complex interactions of multiple risks. As such,
autism is a multidimensional construct underpinned by the principle of ‘multiple determinism’
19
– the development of the condition is likely to be due to combination of multiple causes, none
of which work in isolation of the other (Pelaez-Nogueras, 1996). The ‘multiple causes’ principle
has already been demonstrated at the genetic level in a study of large general population of
twins reporting lower than expected phenotypic correlations and only modest genetic overlap
between the triad of impairments: social, communications and repetitive behaviours (Ronald,
Happe, Plomin, 2005; Ronald et al., 2006). This evidence has led to the coining of ‘fractionated
triad hypothesis’ – different features of autism are due to separate genetic influences,
associated with different brain regions and exhibit as differential cognitive impairments
(Happe, Ronald, Plomin, 2006).
The latest version, the DSM 5 (American Psychiatric Association, 2013), merges the
domains of social and communication impairments into one ‘social communication’ domain,
both for clinical utility (i.e. most symptoms of communication difficulties involve social
situations and vice versa) and in light of factor analytic studies suggesting a single factor
(Frazier et al., 2014; Mandy, Charman, & Skuse, 2012). The repetitive restricted behaviours and
interests (RRBI’s) form the second domain included in the DSM 5 criteria, with the addition of
sensory sensitivities. Children that meet the social communication but not the RRBI’s criteria
are to be assigned a Social Communication Disorder and not ASD. Table 1-1 below provides a
full breakdown of inclusion criteria.
Table 1-1 The diagnostic criteria in DSM 5 for Autism Spectrum Disorders.
A. Persistent deficits in social communication and social interaction across multiple
contexts, as manifested by the following, currently or by history (examples are illustrative,
not exhaustive, see text):
1. Deficits in social-emotional reciprocity, ranging, for example, from abnormal social
approach and failure of normal back-and-forth conversation; to reduced sharing of interests,
emotions, or affect; to failure to initiate or respond to social interactions.
2. Deficits in nonverbal communicative behaviours used for social interaction, ranging, for
example, from poorly integrated verbal and nonverbal communication; to abnormalities in
eye contact and body language or deficits in understanding and use of gestures; to a total
lack of facial expressions and nonverbal communication.
3. Deficits in developing, maintaining, and understanding relationships, ranging, for
example, from difficulties adjusting behaviour to suit various social contexts; to difficulties in
sharing imaginative play or in making friends; to absence of interest in peers.
Specify current severity: see section F
20
B. Restricted, repetitive patterns of behaviour, interests, or activities, as manifested by at
least two of the following, currently or by history (examples are illustrative, not exhaustive;
see text):
1. Stereotyped or repetitive motor movements, use of objects, or speech (e.g., simple motor
stereotypies, lining up toys or flipping objects, echolalia, idiosyncratic phrases).
2. Insistence on sameness, inflexible adherence to routines, or ritualized patterns or verbal
nonverbal behaviour (e.g., extreme distress at small changes, difficulties with transitions,
rigid thinking patterns, greeting rituals, need to take same route or eat food every day).
3. Highly restricted, fixated interests that are abnormal in intensity or focus (e.g., strong
attachment to or preoccupation with unusual objects, excessively circumscribed or
perseverative interest).
4. Hyper- or hyporeactivity to sensory input or unusual interests in sensory aspects of the
environment (e.g., apparent indifference to pain/temperature, adverse response to specific
sounds or textures, excessive smelling or touching of objects, visual fascination with lights or
movement).
Specify current severity: see section F
C. Symptoms must be present in the early developmental period (but may not become fully
manifest until social demands exceed limited capacities, or may be masked by learned
strategies in later life).
D. Symptoms cause clinically significant impairment in social, occupational, or other
important areas of current functioning.
E. These disturbances are not better explained by intellectual disability (intellectual
developmental disorder) or global developmental delay. Intellectual disability and autism
spectrum disorder frequently co-occur; to make comorbid diagnoses of autism spectrum
disorder and intellectual disability, social communication should be below that expected for
general developmental level.
Note: Individuals with a well-established DSM-IV diagnosis of autistic disorder, Asperger’s
disorder, or pervasive developmental disorder not otherwise specified should be given the
diagnosis of autism spectrum disorder. Individuals who have marked deficits in social
communication, but whose symptoms do not otherwise meet criteria for autism spectrum
disorder, should be evaluated for social (pragmatic) communication disorder.
Specify if:
With or without accompanying intellectual impairment
With or without accompanying language impairment
Associated with a known medical or genetic condition or environmental factor
Associated with another neurodevelopmental, mental, or behavioural disorder
With catatonia (refer to the criteria for catatonia associated with another mental disorder,
pp. 119-120, for definition) (Coding note: Use additional code 293.89 [F06.1] catatonia
associated with autism spectrum disorder to indicate the presence of the comorbid
catatonia.)
F. Severity levels for autism spectrum disorder:
Severity Level
Social communication
Restricted, repetitive
behaviours
Level 3
Severe deficits in verbal and
Inflexibility of behaviour,
"Requiring very
nonverbal social communication extreme difficulty coping
substantial support”
skills cause severe impairments
with change, or other
in functioning, very limited
restricted/repetitive
initiation of social interactions,
behaviours markedly
and minimal response to social
interfere with functioning in
overtures from others. For
all spheres. Great
example, a person with few
21
Level 2
"Requiring substantial
support”
Level 1
"Requiring support”
words of intelligible speech who
rarely initiates interaction and,
when he or she does, makes
unusual approaches to meet
needs only and responds to only
very direct social approaches
Marked deficits in verbal and
nonverbal social communication
skills; social impairments
apparent even with supports in
place; limited initiation of social
interactions; and reduced or
abnormal responses to social
overtures from others. For
example, a person who speaks
simple sentences, whose
interaction is limited to narrow
special interests, and how has
markedly odd nonverbal
communication.
Without supports in place,
deficits in social communication
cause noticeable impairments.
Difficulty initiating social
interactions, and clear examples
of atypical or unsuccessful
response to social overtures of
others. May appear to have
decreased interest in social
interactions. For example, a
person who is able to speak in
full sentences and engages in
communication but whose toand-from conversation with
others fails, and whose attempts
to make friends are odd and
typically unsuccessful.
distress/difficulty changing
focus or action.
Inflexibility of behaviour,
difficulty coping with
change, or other
restricted/repetitive
behaviours appear frequently
enough to be obvious to the
casual observer and interfere
with functioning in a variety
of contexts. Distress and/or
difficulty changing focus or
action.
Inflexibility of behaviour
causes significant
interference with functioning
in one or more contexts.
Difficulty switching between
activities. Problems of
organization and planning
hamper independence.
One of the most significant changes of the DSM 5 criteria is replacing the term
‘Pervasive Developmental Disorders’ with ‘Autism Spectrum Disorders’ reflecting that ASD is
not pervasive but has a specific set of behavioural criteria (Kaufmann, 2012). Secondly, both
Asperger’s syndrome and PDD-NOS have been subsumed by a single diagnostic category of
ASD based on the observations of an overlap between the two leading to confusion as to which
is appropriate to use. Albeit there is a risk that individuals with these diagnoses might end up
with the Social Communication Disorder (SCD) and not an ASD diagnosis, as there is no
operational definition for SCD as yet (Skuse, 2012).
22
The removal of Asperger’s syndrome was met with anxiety by the parents and those
with the diagnosis, fearing they would no longer meet criteria under DSM-5 and lose eligibility
for support (Skuse, 2012). For this reason, the DSM-5 committee aimed to back up their
decision-making by scientific evidence. For example, despite the wide-spread use of the term,
e at least five studies showed that the Asperger’s syndrome criteria in DSM-IV were virtually
impossible to meet (Ghaziuddin, et al., 1995; Manjiviona & Prior, 1995; Szatmari, 1995;
Eisenmajer et al., 1996; Mayes, Calhoun, Crites, 2001). There is concern, however, that the
DSM-5 ASD criteria has poor sensitivity for Asperger’s syndrome and may exclude a substantial
proportion of individuals with intact cognition yet in need of support (McPartland, Reichow,
Volkmar, 2012; Mayes, Calhoun, Crites, 2001). This is problematic in light of the fact that these
days most individuals receiving diagnosis have normal-range IQ.
Rett syndrome has been removed as evidence shows that patients grow out of ASDrelated social impairment. For those Rett’s patients who still meet the ASD criteria, a specifier
is added to mark that there is a relevant genetic or other medical condition. For example, in
some cases, an ‘ASD associated with MeCP2 mutation’ can be given to reflect the genetic
aetiology accounting for Rett’s that was not known when the syndrome was included in the
DSM IV (De Weerdt, 2011). The removal of Childhood Disintegrative Disorder was, again, in
light of accumulated literature showing important differences from ASD on acuity and severity
of regression (loss of previously acquired skills) (Kaufmann, 2012).
Even though the DSM 5 was met with a degree of criticism for not being specific
enough, there is evidence that it outperformed the DSM IV on the diagnosis of ASD in two
European cohorts (Mandy, Charman, Puura, & Skuse, 2014). Because clear genetic markers and
effective treatments are yet to be discovered, the aim for now is to continue to improve
behavioural diagnostic practices and promote autism inclusiveness in society (Moore &
Goodson, 2003).
23
1.3 Beyond the diagnostic criteria for ASD – the Broader Autism Phenotype
The historical account of ASD shone the spotlight on individuals that met, or would currently
meet, the diagnostic criteria. However, these were not the only contributions of Kanner and
Asperger, as their observations also included families of children with ASD. For example,
Kanner referred to first- and second-degree relatives that acquired language later than usual,
were obsessive and uninterested in people (Kanner, 1943). Asperger used the adjectives of
‘withdrawn, ‘pedantic’, and ‘eccentric’ when reflecting on some parents (Asperger, 1944
translated by Frith, 1991). The Broader Autism Phenotype (BAP) describes the milder
manifestations of impairment of social communication skills and/or rigidity and idiosyncratic
personality characteristics at the ‘sub-threshold’ levels of clinically diagnosed ASD (Sucksmith,
Roth, & Hoekstra, 2011). The most prominent effect of reporting the BAP in parents is that it
alerts to the possibility of ASD being hereditary, at least in part.
The early studies on the genetics of ASD referred to the BAP as cognitive impairment
(Folstein & Rutter, 1977) or cognitive disorder (Steffenburg et al., 1989). Mention of the Broader
Autism Phenotype as a clinical entity for the first time can be found in a seminal 1995 study
(Bailey et al., 1995). Studies in the 90’s providing supplementary evidence for the BAP showed
that siblings of an ASD proband (the first family member identified with a disorder) had
cognitive disorders (Bolton et al., 1994; Bolton & Rutter, 1990), and severe social disorders
(Piven et al., 1990) more often than controls, and that parents had increased social
communication abnormalities (Landa et al., 1992; Wolff, Narayan, & Moyes, 1988).
The first comprehensive review of the BAP highlighted the need to define the BAP
boundaries in relation to ASD for it to be meaningful in molecular genetic studies (Bailey,
Palferman, Heavey, & Le Couteur, 1998). Importantly, the authors warned that the phenotypic
variability from the BAP to ASD is not likely to be expressed as pattern-like gradations, i.e. there
is no clear behavioural cut-off between manifestations of the BAP and ASD and that they are
not easily quantifiable. Moreover, despite of selective parallel autism impairments and
characteristics found in both ASD and the BAP, individuals displaying the BAP have a lot higher
24
chance of leading an independent life than ASD individuals. This is because their intelligence is
intact, their language ability is normal and they do not suffer epileptic seizures (Mazzone, Ruta,
& Reale, 2012).
The acknowledgment of the BAP has been one of the driving forces for
conceptualisation of the autism phenotype not as a qualitatively discrete categorical construct
but a quantitative trait (a genetically determined characteristic), influenced by multiple genetic
and environmental influences. The current consensus is that the genetic liability distribution
underpins the spectrum of autism behaviours and is merged in a continuous fashion between
the general and clinical populations (Sucksmith et al., 2011).
Aside from using the DSM and ICD (International Classification of Diseases, developed
by the World Health Organisation; synonymous in use to the DSM) instruments to define
autism in categorical terms, the idea that autism is a quantitative trait led to the development
of the term ‘autistic/autistic-like traits’, as observed and measured across clinical and general
populations. The BAP is a measure of autistic traits and is assessed in individuals in two ways:
as a categorical or a continuous trait.
An example of a categorical measure of BAP is the Broad Autism Phenotype
Questionnaire (BAPQ), designed to reflect the phenotypic expression of the genetic liability to
autism in adults, especially for biological parents of children with ASD (Sasson, Lam, Childress,
et al., 2013). No other instruments have been designed for such use (Gerdts & Bernier, 2011). A
recent factor analysis study revealed a three factor structure to the BAPQ, consisting of
withdrawn behaviours, difficulties in understanding of pragmatic language use and rigidity,
which broadly map onto the cardinal autism features but are less profound (Sasson, Lam,
Childress, et al., 2013). The study also estimated a new set of cut offs to reliably classify the
‘unaffected’ and ‘affected’ individuals on the BAPQ measure amongst parents with and without
children with ASD, as previously reported cut offs relied on parents previously known to exhibit
the BAP (Hurley, Losh, Parlier, Reznick, & Piven, 2007).
25
The questionnaires measuring the BAP as a continuous trait can be defined in two ways. The
first set takes the same approach as the BAPQ and defines autistic traits as a three factor
structure albeit measured in continuous manner, while complying with the DSM/ICD criteria.
The second set does not use the DSM/ICD criteria and simply aims to assess the BAP features
quantitatively. The examples of the former set are the Social and Communication Disorders
Checklist (Skuse, Mandy, & Scourfield, 2005), the Quantitative Checklist for Autism in Toddlers
(Q-CHAT) (Allison et al., 2008), the Childhood Autism Spectrum Test (CAST) (Williams et al.,
2005), and the Autism Spectrum Screening Questionnaire (ASSQ) (Posserud, Lundervold, &
Gillberg, 2006). The latter set is exemplified by questionnaires such as the Social
Responsiveness Scale (Constantino, 2011) or the Autism Spectrum Quotient (AQ) (BaronCohen, Wheelwright, Skinner, Martin, & Clubley, 2001). The development of all of the above
continuous autistic traits measures’ allowed for the definition of the Broader Autism Phenotype
in psychometric terms (Lai et al., 2013; Sucksmith et al., 2011; Wheelwright, Auyeung, Allison,
& Baron-Cohen, 2010).
The analyses conducted in this thesis, as described in Chapters 3 and 5, include a
subgroup of individuals defined as ‘affected’ on the Broad Spectrum Disorder (equivalent
conceptually to the BAP) by using categorical approach. Recruited as part of the Social
Relationship Study (SRS), these individuals met this criteria as they: 1) were assigned
ASD(other) category on Development and Well-Being Assessment (DAWBA) or 2) were
assigned Broad Spectrum Disorder on the Autism Diagnostic Interview – R (ADI-R) or scored
just below the cut off (-2 points) on the Autism Diagnostic Observation Schedule (ADOS) to
correspond to the Broad Spectrum Disorder ADI-R diagnosis (see Chapter 3 for more details).
By including sub-diagnostic categories in our analyses we increased the information with
regard to the distributional assumptions underlying the liability to ASD.
1.4 Prevalence of ASD in the general population
Prevalence is an indicator of the number of individuals with a particular condition in a
population (Centres for Disease Control and Prevention, 2010). Recording of prevalence of ASD
26
is challenging due to the changing/inconsistent diagnostic criteria and the variable approaches
of sample ascertainment. For example, studies rely either on clinical records within national
registries and hospital entries, or use unsystematic data records from social services. Only one
study to date reports on the prevalence of ASD in adult populations, which in effect are not
different from rates observed in childhood populations (Brugha et al., 2011). Recording of
prevalence of ASD is often directed by the historical changes of the DSM diagnostic criteria.
More precisely, studies prior to the broadening of the cardinal autism features in the 90’s used
the ‘strict’ autism criteria, whereas from the 90’s onwards the ‘broad’ autism or the ASD criteria
were generally accepted. For this reason, prevalence rates are presented below separately for
Autism/Autistic Disorder, Asperger Syndrome, Childhood Disintegrative Disorder, PDD-NOS
and of combined Autism Spectrum Disorders.
1.4.1 Autism/Autistic Disorder
A comprehensive review of the epidemiology of autistic disorder revealed that, at that time,
there were 53 studies published from 1966 to 2008 in 17 different countries (Fombonne, 2006;
Fombonne, 2009). The most striking feature while compiling the information across countries
was the enormous heterogeneity of the population coverage, which significantly impacts the
derivation of prevalence rate. While bearing this caveat in mind, prevalence of autistic disorder
estimates varied from 0.7 to 72.6 per 10,000 and averaged at 20.6 per 10,000 (Fombonne,
2009). The rates prior to 1987 hovered at ~5 cases per 10,000 but it increases to between 7 and
40 in studies conducted after year 2000.
A follow up epidemiological review aimed to supplement these rates with information
from countries other than Northern Europe (Elsabbagh, Divan, Koh, Kim, Kauchali, Marcín, et
al., 2012). Recorded autistic disorder prevalence rates were as followed: Western Pacific 2.33 to
94/10,000; South East Asia 11.7/10,000 (only 1 study available), and America 0.7 to 40.5/10,000.
Most recently the Global Burden of Disease Study, using regional prevalence models, estimated
the global burden of ASD expressed as number of years individuals live with the disability
(Baxter et al., 2014). As a result, autistic disorder prevalence was estimated at 2.4 per 1000
27
(0.2%) and accounted for more than 58 cases under the age of 5 living with the disability in
100,000 of the population (rounded to .06% of the population). Despite the claims that
prevalence of autistic disorder is increasing, Baxter et al. (2014) did not find significant changes
in prevalence across the globe between 1990 and 2010.
1.4.2 Asperger Syndrome (AS) and Childhood Disintegrative Disorder (CDD)
Both AS and CDD were introduced under the umbrella of ASD criteria in 1994. Epidemiological
data on the AS is, however, very scarce (Fombonne, 2009). The limited number of welldesigned studies to detect the true prevalence of AS diminishes the validity of obtained
estimates: 6 per 10,000, while recognising that AS prevalence was always lower than for autistic
disorder (estimated ratio of 1:4). This ratio of prevalence estimates goes against the opinion of
Gilberg (1992), who suggested that Asperger syndrome was more common than the strict
autism manifestation. An even lower prevalence rate has been recorded for CDD, with the
lower bound estimate of 2.0 an upper estimate of 4.0 per 100,000 (Fombonne, 2009).
1.4.3 Pervasive Developmental Disorder-Not Otherwise Specified (PDD-NOS)
PDD-NOS, also known as atypical autism after the publication of DSM IV, was assigned when
the individual fell short of the strict autistic disorder criteria (American Psychiatric Association,
1994). The prevalence of PDD-NOS was found to be higher than that of autistic disorder: 37.1
per 10,000 compared to 20.6 per 10,000.
1.4.4 Combined Autism Spectrum Disorders (ASD)
Fombonne (2009) calculated the overall prevalence for ASD at 63.7 per 10,000 (synonymous to
Pervasive Developmental Disorders [PDD] term used in that review) by pulling together the
estimates of autistic disorder, AS, CDD and PDD-NOS across the reviewed studies. Two studies
that followed Fombonne’s converge on this finding, as Elsabbagh et al. (2012) provides a
median estimate of 62 cases and Baxter et al. (2014) of 76 cases per 10,000. For the first time a
global burden of ASD (expressed as Disability Adjusted Life Years) is provided: 52 million, with
111 persons per 100,000 classified as disabled due to ASD (Baxter et al., 2014). A study of Korean
28
school-children estimated an even higher prevalence rates – 3.74% for boys and 1.47% for girls
(Kim et al., 2011). The most recent US survey reports ASD prevalence approaching 1.5% of the
population (15 per 1000 or 1 in 68 children) (Developmental Disabilities Monitoring Network
Surveillance, 2014).
All of the evidence demonstrates that ASD affects a significant proportion of the
population. Yet it is recognised that many cases are unaccounted for (due to methodological
limitations of quoted surveys) or not recognised/diagnosed in time to enter these surveys due
to age limit criteria. This is particularly true for older adults. The advent of technological
developments and data storage brings hope that information on affected individuals can one
day be effectively pulled together to provide more accurate estimates. In addition, the everincreasing awareness and societal inclusiveness of autism will hopefully encourage individuals
without the diagnosis to seek help and open up about their condition, allowing them to
eventually access the support they are statutorily offered.
1.4.5 The Broader Autism Phenotype (BAP)
Producing reliable prevalence rates in the general population for the BAP is problematic as it is
not a clinical condition and therefore not widely assessed, although a very important aspect of
the autism spectrum. Family studies provide some indication on the prevalence in adult
relatives, most often parents. The oldest report quotes that 15-45% of relatives display the BAP
(Bailey et al., 1998) and this rate is of moderate value when measured with the BAPQ: 14-23%
of parents met the BAP diagnostic cut off (Sasson, Lam, Childress, et al., 2013). The evidence
on the prevalence of BAP in child populations shows a different picture. In a sample of unrelated
1925 children at mean age of 8 years, the BAP was measured with Children Autism Spectrum
Test (CAST) (Williams et al., 2005). Authors found that at least 5.8% of children scored above
the cut off of 15 points and 4.8% scored borderline of that cut off (12 to 14 points). As CAST’s
ability to predict ASD diagnosis was 50% in this epidemiological sample, a 5% prevalence rate
of the BAP can be estimated by halving the 10% of children scoring highly on CAST.
29
1.5 Is autism heritable?
1.5.1 Overview
This section of Chapter 1 takes a step back to briefly explain the concept of heritability before
moving to explanation of methods that must be applied in order to derive heritability estimates
for ASD as a categorical disorder. Then, studies of heritability of autism in the clinical and
general population samples are introduced.
1.5.2 What is heritability?
In 1903 Wilhelm Johannsen introduced the terms ‘genotype’ and ‘phenotype’, to distinguish
between the material an organism inherits from parents (the genotype) and their observable
characteristics (the phenotype) (Johannsen, 1911). The transmission of the genetic properties
from parents to children was the essence of ‘heredity’ and ‘heritability’ is the quantitative
expression of heredity – the degree to which a phenotype can be inherited from previous
generations.
A fuller account of methods calculating heritability is provided in the Methods section
(Chapter 2). Behaviour genetics as a field (by investigating latent genetic and environmental
influences on behaviour) has provided an incredible amount of evidence for strong genetic
influences on complex traits like mental health disorders. In lay terms, ‘heritability’ is the
proportion of the observed behaviour’s variance explained by the genetic differences amongst
people (Plomin, DeFries, Knopik, & Neiderhiser, 2013).
Twin studies are the most popular way to estimate heritability of complex behaviours.
The concept is rather straightforward, as it involves comparison of the derived correlations for
monozygotic (MZ) twin pairs (i.e. children conceived from the same ovum, who share ~100% of
segregating genes) with that of dizygotic (DZ) twin pairs (children conceived from separate ova,
who share 50% of their segregating genes, on average, like siblings) (Plomin et al., 2013). For
the actual estimation of heritability the distribution of the trait will be a decisive factor for the
statistical methods we use. For example, estimating heritability for a categorical disorder
requires additional assumptions and statistical adjustments that deal with non-continuous
30
data. Additionally, ascertainment approaches to select affected twins from the population will
require corrections in order to get unbiased estimates of heritability. As both these issues are
important for twin research on ASD, we discuss both of them in more detail below.
1.5.3 Deriving heritability for categorical disorders
One of the most popular methods to provide an index of heritability for categorical disorders is
to present probandwise concordance (agreement) for MZ and DZ twin pairs, recruited because
at least one of twins is affected. There are two possible outcomes: the pair is concordant (both
twins are affected) or discordant (one twin affected and one unaffected). Using this
information, the index of genetic risk is estimated by the proband-wise concordance rate
following the formula:
[(2 x number of concordant pairs)] / [2 x (number of concordant pairs) + discordant pairs] (McGue,
1992; Rijsdijk & Sham, 2002).
Higher MZ than DZ concordance rates indicate genetic influences on the trait. However,
probandwise concordance method does not take into account the population prevalence for
the disorder, making it impossible to reliably estimate the genetic and environmental
influences, which is one of the advantages of structural equation ‘liability threshold’ modelling.
A detailed description of this model is provided in the Method section (Chapter 2,
2.4.4). The main advantage is the translation of proportions of concordance/discordance into
tetrachoric correlations between underlying continuous (normal) distributions of liability.
These liabilities are hypothesised to underpin the categorical diagnoses (van Dongen,
Slagboom, Draisma, Martin, & Boomsma, 2012), with a threshold distinguishing between
unaffected and affected individuals (i.e. those with enough risk to exhibit symptoms of the
disorder) (Rijsdijk & Sham, 2002).
The model can be extended further to assume multiple thresholds. For example,
heritability for affective disorders was estimated in a model including broad and strict disorder
31
criteria to reflect that twins were ascertained from both psychiatric registry as well as the
general population (Kendler, Pedersen, Neale, & Mathe, 1995). However, no twin studies of
ASD to date have used multiple thresholds methodology, despite the growing evidence that
autism is a quantitative trait and that the Broad Autism Phenotype is a meaningful component
of the underlying genetic liability (Sucksmith et al., 2011).
1.5.4 Ascertainment Correction for Selected Samples
Studies of human genetics rely on sampling of individuals with the specific phenotype in order
to determine its aetiology. Twin studies typically sample only a subset of the population, i.e.
the pairs with at least one affected twin (McGue, 1992). An affected twin becomes a proband
(member of the family with the genetic disorder that came into contact with the medical
community) only when s/he was independently ascertained into the sample (Morton, 1959).
Whether an affected twin becomes a proband will depend on whether both twins are
affected and the thoroughness of sampling (or catchment area of a hospital/clinic). In a
population with a reliable track record of all affected individuals, ascertainment is complete as
every affected twin entering the sample is a proband (i.e. each is independently ascertained).
This type of ascertainment is therefore also called ‘double’ ascertainment.
In ‘double’
ascertainment the probability (π) of being a proband is 1, as affected individuals have every
chance of entering the sample (Stene, 1977). However, when records are kept in multiple
registers, hospitals or health institutions across several regions and the catchment area per
clinic is small, the probability of being a proband diminishes and is less than 1 (i.e. π<1). This
means that the concordant affected pairs enter the sample due to identification of a single
proband within the pair and it is during the follow up that their co-twin can be found to be
affected as well. This type of ascertainment is defined as incomplete or ‘single’ ascertainment.
Due to rarity of twin samples with the specific disorder, inclusion of probands occurs through a
mixture of both complete and incomplete ascertainment methods, classified as ‘multiple
incomplete’ with an associated probability of 0<π<1 (Rijsdijk & Sham, 2002).
32
Even though the π constant can be applied as a correction weight when estimating
heritability, the need for systematically selected twin samples to meet the assumption of being
representative of the general population cannot be stressed enough. For example, parents of
twins where serious medical condition is present in both children are more likely to seek medical
attention, therefore have higher chance to be invited to participate in genetic studies. This
over-inclusion of concordant pairs could lead to observed stronger phenotypic associations,
which in the case of MZ twins could lead to inflated heritability rates. Conversely, over-inclusion
of DZ concordant twins could lead to inflated rates of shared environmental effects.
1.5.5 Heritability of autism
1.5.5.1 Clinical samples
When autism was formally reported as a neuropsychiatric condition by Leo Kanner in the 40’s,
the prevailing perception of its aetiology was one hinting towards family environmental factors.
Kanner proposed autism to be due to lack of expression of parental warmth towards the child
(Kanner, 1949). In Kanner’s original 1943 paper explaining autism, he ascribed the children’s
difficulties to an ‘inbuilt’ disorder of ‘affective contact’. However, under the influence of
Bettelheim’s later psychogenic account, Kanner later accepted the ‘refrigerator mother’ notion
(Bettelheim, 1967). This perception was dominant in the medical establishment through the
1960’s.
The seminal study by Folstein and Rutter (1977) was the first scientific attempt to
establish whether infantile autism was indeed due to environmental or genetic influences. In
this first twin autism study, they reported probandwise concordance rates for 11 MZ pairs of
53% and for 10 DZ pairs of 0%, indicating for the first time the strong importance of genetic
factors in autism (Folstein & Rutter, 1977). Eleven twin studies followed to date (Table 1-2). The
evolution of the DSM is demonstrated by the earlier studies following the strict autism criteria,
whereas studies after 1995 reflect the broadening of the cardinal autism features and diagnosis
of ASD. All studies are summarised in Table 1-2 for ease of presentation.
33
It is apparent that twin studies of clinical diagnosis of autism are rare. This is not
surprising, as collecting data on twins with clinically confirmed or a research diagnosis is of
considerable difficulty. As seen in Table 1-2, the more recent studies referred to formal twin
modeling methods (described in detail in the Methods section, Chapter 2) to derive the A
(genetic), C (shared environmental factors) and E (non-shared environmental factors)
components of autism. The earlier studies report probandwise concordance rates which can
only be taken as a rough indication of influences of genetic risk. When probandwise
concordance rates are considered on their own, the ratio between MZ and DZ twins is
characterised by a less than 2 to 1 ratio, which on average points to strong genetic influences
(disregarding sex differences). Moreover, the DZ rates are never half or above that of MZ rates,
which would indicate influence of shared environmental effects.
However, comparison of twin modeling results paints a more complicated picture as
two (Frazier, et al., 2014; Hallmayer et al., 2011) of the five published studies suggest that
autism is under stronger shared environmental than genetic influences, with the remaining
three studies showing probandwise concordance ratios supporting high heritability
(Lichtenstein et al., 2010; Sandin et al., 2014; Taniai et al., 2008). The Hallmayer et al., (2011)
publication suggesting strong shared environmental effects on the development of autism has
been positively received, as environmental influences are considered to be more readily
malleable than genetic predispositions.
Szatmari concluded in his commentary paper that ‘the aetiology of ASD has received
renewed impetus’ (Szatmari, 2011). The commentary states that the Hallmayer results provide
the answer to the reason of the slow progress in the discovery of specific genetic variants and
the existence of missing heritability (the difference between the total genetic variance
[estimated from twin studies] accounted for by known genome-wide associations (Plomin et
al., 2013)), even though they go against 30 years of autism research. However, since Szatmari’s
remark new genetic discoveries are beginning to characterise autism as a complex genetic trait.
34
Table 1-2 Summary of all twin studies of clinical autism published to date.
Study
1. Folstein & Rutter, 1977
2. Ritvo, Freeman, MasonBrothers, Mo, & Ritvo, 1985
3. Steffenburg et al., 1989
4. Bailey et al., 1995
5. Le Couteur et al., 1996
n,
Probandwise
MZ/DZ concordance,
pairs
MZ/DZ
Autistic Disorder (AD)
11/10
.58/.0
23/17
.96/.24
A, C, E estimate
-
11/10
.95/.0
25/20
.60/.0
28/20
.73/.0
Autism Spectrum Disorders (ASD)
6. Taniai, Nishiyama, Miyachi,
19/26
.95/.31
♂ A: 73%, C:0%, E: 27%
Imaeda, & Sumi, 2008
♀ A: 87%, C:0%, E: 13%
7. Rosenberg et al., 2009
67/210
♂: .86/.40
♀: .100/.20*
8. Lichtenstein, Carlström,
29/88
♂: .47/.0
A: 80%, C: 0%, E: 20%
Råstam, Gillberg, &
♀: .14/.20&
Anckarsäter, 2010
9. Hallmayer et al., 2011
54/138
♂: .77/.31
AD: A: 37%, C: 55%,
♀: .50/.36
E: 8%
ASD: A: 38%, C: 58%, E:4%
10. Frazier, Thompson, et al.,
128/440
A: 21%, C: 78%, E: 1%
2014
11. Sandin et al., 2014£
A: 50%, C: 0%, E: 50%
12. Nordenbæk, Jørgensen,
13/23
.95/.40
Kyvik, & Bilenberg, 2014
Abbreviations: ♂ male; ♀ female; A – genetic factors; C- shared environmental factors; E – non-shared
environmental factors.
* The pair has at least 1 female, meaning that DZ opposite sex twin pairs contribute to this calculation.
& concordance for DZ opposite sex twins was zero.
£ Number of twin pairs was not provided, albeit most likely the same cohort as in Lichtenstein et al.
(2010) was analysed.
It is now understood that rare genetic variance is important in many cases, as discussed in detail
in section 1.6. More importantly, studies also suggest that common (additive) genetic variance
aggregates as a background risk and accounts for between 17% (Cross Disorder Group of the
Psychiatric Genomics et al., 2013) to 52.4% of the total autism genetic variance (Gaugler et al.,
2014). As genetic samples increase in size, these estimates are predicted to become more
precise.
Finally, Szatmari reignites the idea of autism as a disorder of foetal programming (Bale
et al., 2010), which suggests that maternal environment during the pregnancy linked to
ingestion of drugs, toxic chemicals and maternal illnesses as diabetes and infections, are the
35
causes of autism. It is obvious that environments and other factors do interact with the
mother’s and child’s genotype. However, epidemiological findings to date have not come up
with a definitive autism pathogenesis including these factors (Lai, Lombardo, & Baron-Cohen,
2014), as discussed in section 3 of this chapter.
1.5.5.2 General population samples
Table 1-3 provides a summary of all community twin studies published to date on autism traits.
As explained in section 1.3, relatives of individuals with the diagnosis of ASD exhibit higher
levels of autism traits than individuals from families without ASD incidence.
This suggests that sub-ASD threshold, continuously distributed autism traits’
expressions share genetic influences with the ASD diagnosis. Understanding the sources of
individual differences in autism traits may pave the way towards understanding of the clinical
diagnoses of ASD (Sucksmith et al., 2011).
The autism traits genetic model-fitting findings can be grouped into two broad themes.
First, they suggest that modest shared environmental influences play a role in the development
of autism traits during the early (age 2) and the middle childhood periods (7-17 years)
(Constantino & Todd, 2005; Constantino & Todd, 2003; Edelson & Saudino, 2009; Stilp et al.,
2010). Second, the remaining nine studies covering middle childhood to late teen years (ages
17-18) support the overall theme of autism traits to be highly heritable, with no support for
shared environmental influences (Constantino et al., 2000; Hoekstra et al., 2007; Lundstrom et
al., 2012; Robinson et al., 2011; Ronald et al., 2005, 2008, 2011; Ronald, Happé, Price, BaronCohen, & Plomin, 2006; Skuse, Mandy, & Scourfield, 2005).
Studies by Ronald et al., (2006), Robinson et al., (2011) and Lundstrom et al., (2012)
extended the twin model design to investigate the aetiology both in the general population and
at the extreme end of the autism traits distribution. In particular, Ronald et al., (2006)
demonstrated for the first time high group heritabilities, both at beyond as well as above the
extreme cut offs. Robinson et al., (2011) utilised a multiple threshold approach to report
comparably high heritability of autism traits at quantitative cut offs of 1%, 2.5% and 5% in
36
Table 1-3 Summary of all twin studies of autism traits published to date, adapted from Ronald &
Hoekstra, (2014).
Study
1. (Constantino, Przybeck,
Friesen, & Todd, 2000)
2. (Constantino & Todd,
2003)
3. (Constantino & Todd,
2005)
n, MZ/DZ
pairs
98/134
268/320
89/196
Measure; age
(years)
Parent SRS; 715
Parent SRS; 715
Parent & child
SRS; 8-17
4. (Ronald, Happe, &
Plomin, 2005)
T: 1144/1994
P: 1456/2540
5. (Skuse, Mandy, &
Scourfield, 2005)
6. (Ronald et al., 2006)
278/378
1221/2198
Teacher &
parent DSM-IV
items; 7
Parent SCDC; 517
Parent CAST; 8
7. (Hoekstra, Bartels,
Verweij, Boomsma, 2007)
8. (Ronald, Happe, &
Plomin, 2008)
81/113
AQ; 18
T: 934/1645
P: 1204/2042
S: 1123/1907
Teacher,
parent, selfreport CAST; 9
9. (Edelson & Saudino,
2009)
10. (Stilp, Gernsbacher,
Schweigert, Arneson, &
Goldsmith, 2010)
11. (Ronald, Larsson,
Anckarsäter, Lichtenstein,
2011)
12. (E B Robinson et al.,
2011)
145/168
Parent CBCL; 2
414/797
Parent MCHAT; 2-3
1788/3752
Parent A-TAC;
9-12
2126/3842
Parent CAST; 12
13. (Lundstrom et al., 2012)
3229/8073
Parent A-TAC;
9-12
A, C, E estimate; other
worthy conclusions
A: 76%, C:0%, E: 24%,
male sample only
A: 48%, C:32%, E: 20%
♂ A: 87%, C:13%, E: 0%,
♀ A: 73%, C:10%, E: 17%,
largely female sample
A: 62-76%, C: 0%, E: 2538%, across raters and
social & non-social items
A: 74%, C:0%, E: 26%
Extreme cut off analysis
A: 64-92%, C:0%, E: 836%; Continuous analysis
A: 78-81%, C:0%, E: 1922%
A: 57%, C:0%, E: 43%
T: A: 69%, C:0%, E: 31%
P: A: 82-87%, C:0%, E: 1318%
S: A: 36-47%, C: 18% (♂),
E: 46-53%
A: 40%, C: 20%, E: 40%
A: 44%, C: 32%, E: 24%
More stringent threshold:
A: 74%, C: 19%, E: 7%
A: 49-76%, C: 0%, E: 2451%
A: (♀)53 -(♂)72%, C: 0%, E:
47-28%
Extreme cut off analysis
A: 88-90%, C: 0%, E: 1012%
Several cut offs analysis
A: 59-88%, C: 0-23%*, E:
12-23%
(♂ only) A: 74-86%, C: 08%*, E: 14-21%
Abbreviations: SRS – Social Responsiveness Scale; A – genetic factors; C- shared environmental factors;
E – non-shared environmental factors; SCDC – Social and Communication Disorders Checklist; CAST –
Children Autism Spectrum Test; AQ- Autism Spectrum Quotient; T – teacher, P – parent, S – selfreport; CBCL – Child Behaviour Checklist; M-CHAT – Modified Checklist for Autism in Toddlers; A-TAC
– Autism, Tics, AD/HD, and other Comorbidities inventory.
* Confidence Intervals span 0, non-significant estimate.
37
adolescent sample. Lundstrom et al., (2012) added to this evidence by reporting a significant
genetic overlap between uniformly high heritability estimates derived from low, high, 10th and
15th percentile cut off groups. Altogether, these studies demonstrate that autism traits and ASD
are related to the same genetic susceptibility, further supporting the idea that autism is a
quantitative trait (Constantino & Todd, 2003; Gillberg, 1992; Sucksmith et al., 2011).
In summary, findings from both clinical and general population twin samples appear to
point to the conclusion that autism is highly heritable, although a few clinical studies do suggest
that shared environmental influences might also be playing an important role in the aetiology
of autism. However, it is possible that these conflicting findings are due to small sample size in
clinical studies. General population samples provide a good example of exploring the
differential cut offs of the autism distribution. This approach would be particularly useful given
the emerging evidence of the effects of differential prevalences of autism, as well as the
subgroups that are part of the autism distribution. There are no twin studies to date exploring
these issues in diagnostic samples and their impact on derived heritability estimates from
categorical diagnosis data.
1.6 Molecular genetic studies of autism
While twin studies have proved to be extremely useful in estimating the total genetic influence
on autism, it is the molecular genetics field that searches for specific genes associated with that
trait (a genetically determined characteristic) (Plomin et al., 2013). As autism is a quantitative
trait many molecular studies focus on identification of multiple locations (i.e. Quantitative Trait
Loci, QTL) of genetic variants affecting the disorder (Geschwind, 2011; Lai et al., 2014). The idea
is that individuals with autism differ from the rest of the population in respect to an allele type
(the smallest unit of genetic information) at certain locations. The search for QTL’s is divided
into finding of common and rare genetic variants (alleles) (De Rubeis & Buxbaum, 2015).
Frequency of a specific allele (for example A1) is determined by dividing the number of times
A1 is observed in the population by the total copy number of all the alleles at that particular
38
genetic location in the population (Kwok, 2000). The alternate allele forms at a locus are
commonly described as Single Nucleotide Polymorphisms (SNPs) (De Rubeis & Buxbaum,
2015).
1.6.1 Common genetic variants
SNPs estimated to have allele frequency higher than 5% in the population are considered
indicators of common genetic variance (Lai et al., 2014). The method of Genome-Wide
Association (GWA) searches the ever-increasing coverage provided by improved SNP arrays for
SNPs associated with the disorder. Thus far, three GWA autism studies have been conducted
(De Rubeis & Buxbaum, 2015). Typically the strength of association between a genetic variant
and a phenotype is summarised using odds ratio, calculated as the odds of an allele in cases
divided by the odds of an allele in controls. An odds ratio of 1 indicates no difference between
cases and controls (Plomin et al., 2013)
A study of individuals with European ancestry has identified six significant SNPs, with
the strongest association with ASD being for SNP rs4307059 on chromosome 5p14 (odds
ratio=1.19) (Wang et al., 2009). However, this SNP association was not replicated in the second
GWA autism study. It nevertheless identified suggestive associations on chromosomes 6q27
and 20p13 (Weiss, Arking, & Consortium, 2009). The third GWA study identified SNP rs4141463
as significantly associated with autism, although its signal did not meet the significance
threshold when replicated in a smaller sample (Anney et al., 2010). Altogether, the GWA studies
in the general populations have not yet identified SNPs with effect sizes big enough to be
deemed causal of autism (Lai et al., 2014).
In recognition that relative contributions of common genetic influences differ across
families in relation to the rate of incidence of autism within each family, Klei et al., (2012)
estimated the proportion of autism liability due to additive genetic effects (see Methods
section, Chapter 2 for a detailed discussion). Using data on simplex (one member of the family
has autism) and multiplex (several members are affected) families authors showed that, when
combined, common SNPs exert substantial additive genetic effect on liability to autism. Of the
39
total genetic variation, the heritability due to additive effects (narrow sense heritability) was
estimated at 60% in multiplex and 40% for simplex families (Klei et al., 2012). Furthermore, a
study of unrelated individuals in Sweden using Genome-Wide Complex Trait Analysis (GCTA)
(Yang, Lee, Goddard, & Visscher, 2011), estimated narrow sense heritability precisely at 52.4%
while concluding that most of it is due to common SNPs variation (Gaugler et al., 2014).
Furthermore, a UK study suggested that common genetic effects vary developmentally, as
heritability estimated ranged between 24-45% in childhood/adolescence and dropped to 16%
in middle adolescence (St Pourcain et al., 2014).
Combined together, these findings show that common SNPs can explain the
emergence of autism and milder subtypes (including the Broader Autism Phenotype)
(Sucksmith et al., 2011); the increased prevalence of autism in children from parents with
autistic traits (Baron-Cohen, 2012), as well as why the spectrum of autistic behaviours is seen in
the general population (Chakrabarti et al., 2009).
1.6.2 Rare genetic variants
The Swedish study has demonstrated that rare genetic variants are also important as these
variants were seen to substantially explain individual-wise liability to autism (Gaugler et al.,
2014). SNPs of allele frequency less than 5% in the population are considered as rare genetic
variants, also called rare mutations (Lai et al., 2014). These are known as chromosomal
abnormalities in the form of deletions or duplications of a chromosomal region or even an entire
chromosome, accounting for 5% of cases with autism (Betancur, 2011).
The most frequent chromosomal abnormalities identified in autism are 15q11-q13
duplications and 22q11.2 and 22q13.3 deletions (De Rubeis & Buxbaum, 2015). The
deletions/duplications can also happen at the DNA level when a cell has an abnormal number
of copies of DNA sections. These are called copy number variations (CNVs) and have been
consistently observed in individuals with autism, either through inheritance or occurring de
novo (anew). The inherited CNVs include recessive and chromosome X-linked deleterious
mutations (Lim et al., 2013; Yu et al., 2013). The de novo mutations arise as germline mutations
40
during the meiotic division of gametogenesis (De Rubeis & Buxbaum, 2015) and are thought to
have large effect size (3.5), especially in simplex families (Kong et al., 2012; Michaelson et al.,
2012; Neale et al., 2012; O’Roak et al., 2012). De novo mutations are detected in 5-10% of
individuals with autism (Pinto et al., 2014).
Because of the high impact on the individual liability to autism, de novo mutations have
been the main focus for gene discovery using the next-generation sequencing that collectively
identified nine autism risk genes: ANK2, CHD8, CUL3, DYRK1A, GRIN2B, KATNAL2, POGZ,
SCN2A, TBR1 (Neale et al., 2012; O’Roak et al., 2012; Sanders et al., 2012). One particular
mutation - in the SHANK group of genes - were linked quite strongly to the whole spectrum of
autism symptoms and differential severity of intellectual disability (Leblond et al., 2014). As
exome identification (sequencing of all the protein coding genes in the genome) is proving its
potential to identify additional risk genes because of large sample sizes (De Rubeis & Buxbaum,
2015), two recent studies have further identified 40 autism risk genes (De Rubeis et al., 2014;
Iossifov et al., 2012).
As demonstrated, studies of the genetics of autism bring a lot of promise, especially in
respect to rare mutations, but these only capture the genetic aetiology of autism in a fraction
of cases. As large-scale collaborations become the norm and the sample sizes increase as a
result, the discovery of common variants will become more realistic, as it has been shown in
schizophrenia (Ripke et al., 2014). At this stage, around 80 rare mutations genes are known,
although geneticists estimate that ~1000 genes are implicated in autism, meaning that it will
be some time before all genes are identified (De Rubeis & Buxbaum, 2015). A recent review
summarises in detail all of the genetic discoveries and comments on the links between the
genes and biological pathways implicated in autism (Geschwind and State, 2015). In particular,
chromatin modification, synaptic function, fragile X mental retardation and early embryonic
development.
Even when genetic associations are discovered, they will not provide all of the
information needed to understand the causality of autism as the action of genes may frequently
41
be dependent on non-genetic and environmental factors (Kendler, 2005). Indeed, twin studies
show that MZ twins are never perfectly correlated on a trait, which implicates that non-genetic
and environmental factors are always at play. For example, a study of MZ twins discordant for
ASD showed that differential DNA methylation processes were responsible for the diagnostic
discordance amongst most twin pairs (Wong et al., 2014). There is also evidence that the higher
risk of autism in males compared to females could be due to imprinted chromosome X-linked
genes, which males invariably inherit from their mother. In contrast, females that have
inherited the X chromosome from their father appear to be affected a lot less and therefore are
protected from developing autism (Skuse, 2000). In regards to environmental factors, studies
of the last four decades produced several hypotheses of the interaction between genes and
environments in autism, discussed in section 1.7.
1.6.3 Genetic effects in simplex and multiplex families
Searching for specific genetic variants for autism as a complex genetic psychiatric trait involves
the already mentioned case-control studies (Genome Wide Association [GWA] studies) but also
family-based design studies (Sullivan, Daly, O’Donovan, 2012). The latter design is used to
evaluate how genotypes and phenotypes segregate within families and allow identification of
mutations arising de novo – something GWA studies cannot do.
Studying multiplex pedigrees (2 or more siblings have ASD in the family, equal to
concordant twin pair) provides an advantage to discover classes of genes related to the disorder
as they are over-represented (enriched) in such families, therefore can reveal the causal genetic
variation with high penetrance (several family members are carrying the variant associated
with the disease). In the case of autism this approach led to identification of large effect de novo
mutations, however, the disadvantage is that these genetic events might be family specific
(‘private’) and not entirely reflective of the general population risk (Sullivan, Daly, O’Donovan,
2012). In addition, children from multiplex families are believed to be at a higher risk of
inheriting autism. For this reasons, scientists began looking more favourably at simplex
pedigrees.
42
Simplex family design (1 sibling has ASD and 1 is unaffected, discordant twin pair) also
evaluates the influence of de novo mutations, but in the context of reduced fecundity because
of the disorder, as well as a proven de novo structural variation due to recent high penetrance
mutations in the large number of genes. Most recent study does indeed show that at least 30%
of ASD in simplex families is due to de novo mutations, which in turn create nonsense, splicesite or small frame-shift structural variants (Iossifov et al., 2015). What is unexpected is that the
rate of de novo mutation rates impacting structural variation is fairly similar for both simplex
and multiplex ASD families (Sanders et al., 2011; Pinto et al., 2010). In addition, it has been
shown that 40% of simplex families from Simons Simplex Collection could be in fact classified
as ‘high risk’, i.e. transmission genetics are playing a strong causative role (Ronemus, Iossifov,
Wigler, 2014). This means that the two pedigree types are perhaps not that different in terms
of genetic risk factors predisposing to autism. Additionally, autism heritability estimates
derived from twin samples (by default including simplex and multiplex families) for both
diagnostic and continuous trait dimensions (e.g. CAST) are comparable in magnitude.
The weakness of both designs is that enriched environmental causes of the disorder are
not measured and yet multiple psychiatric disorders are due to different and rare environmental
risks. Furthermore, these influences can differ between the pedigrees. In the case of simplex
families, detecting the specific environmental risk that led to a diagnosed disorder in only one
of the siblings is nearly impossible. The risk may be individual-specific and not interact with
genetic risk in the same way for other individuals in that family, and is unlikely to be found in
other simplex families. Not being able to measure these individual gene-environment
interactions limits our understanding of the aetiology of autism. In the case of the multiplex
pedigree design, it is possible that these families are enriched for shared environmental risk
factors (Sullivan, Daly, O’Donovan, 2012). These two types of pedigrees may therefore require
different research methods to understand the environmental risks predisposing to autism.
43
1.7 Environmental factors in autism
As autism symptoms can occur as early as 14 months of age (Landa, 2008), a number of
hypotheses have been postulated to explain this early onset. Specifically, these hypotheses
relate to exposures during the prenatal and perinatal stages, with a number of factors linked to
later developmental stages (Gardener, Spiegelman, & Buka, 2009; Grabrucker, 2013;
Landrigan, 2010). These can be grouped into five broad areas: exposure to toxic chemicals,
vaccinations, pregnancy-related complications and hormonal in utero influences.
1.7.1 Effects of chemical toxins
The hypothesis of the effect of toxins in autism has been suggested in line with understanding
of the general vulnerability of the foetal brain to toxic exposures (Bondy & Campbell, 2005).
The complex interplay between toxic environments and genetic susceptibility may cause
autism, as genetic influences on its own are insufficient to explain clinical and within-family
symptom heterogeneity and discordant MZ twins (Landrigan, 2010).
There are a number of studies apparently linking chemicals (medication & insecticides)
with increased incidence of autism in the last two decades. For example, Thalidomide (antisickness treatment given to pregnant women in the 1950’s and 60’s) has been reported to affect
5000+ children worldwide, leading to limb & eye malformations as well as autism when taken
20-24 days after conception (Miller & Stromland, 2011). Four children out of 100 affected by
this embryopathy met the full diagnostic criteria for ASD (Stromland, Nordin, Miller,
Akerstrom, & Gillberg, 1994). Valproic acid, an anticonvulsant, led to similar effects. Out of 57
children born to mothers that took valproic acid in the first 4 weeks of pregnancy, 34 reported
two or more autistic symptoms, four received diagnosis of autistic disorder and two of
Asperger’s syndrome (Moore et al., 2000). A derivate of valproic acid, valproate, used as
medication for epileptic seizures has also been found to lead to increased autism incidence in
children born to mothers with epilepsy (Christensen et al., 2013). A drug preventing gastric
ulcers, misoprostol, has also been used as an abortifacient. A Brazilian case study reported that
44
four out of seven children prenatally exposed to misoprostol had ASD (Bandim, Ventura, Miller,
Almeida, & Costa, 2003).
Chlorpyrifos, used to control insects, has been banned from use in schools and homes
in 2001 but is still heavily applied in the US agriculture (Landrigan, 2010). It is reported to have
strong effects on new-born rats and humans, leading to reduced number of neurons, decreased
intelligence and behavioural alterations (Levin et al., 2001; Rauh et al., 2006). The follow up of
children later in life revealed significant developmental delays, cognitive deficits and an
increased incidence of PDD-NOS (Rauh et al., 2006). Combined, toxins’ exposure evidence is
compelling and warrants further research to understand the processes through which the
occurrence of autism emerges.
1.7.2 Vaccinations
The first suggestion that vaccination of infants against the measles, mumps and rubella (the
MMR) led to increased autism prevalence emerged in 1998 (Wakefield, Murch, & Anthony,
2010). It resulted in a media frenzy that has diminished parents’ confidence in the MMR, which
ultimately protects children from life threatening diseases. Upon the release of the 1998 paper,
research bodies across the globe performed epidemiological studies in several countries and
found no support for the MMR jabs leading to higher incidence of autism in the UK (Kaye,
Melero-Montes, & Jick, 2001; Taylor et al., 1999), the US (Dales, Hammer, & Smith, 2001),
Japan (Honda, Shimizu, & Rutter, 2005), Denmark (Madsen et al., 2002), and Finland (Makela,
Nuorti, & Peltola, 2002). Following this trend, there were suggestions of thimerosal, ethyl
mercury used to preserve vaccination vials to prevent microbial contamination, as linked with
autism. But similarly to the MMR jab, no evidence to date exists to support this claim (Barile,
Kuperminc, Weintraub, Mink, & Thompson, 2012; Heron, Golding, & Team, 2004). The vaccines
story incident not only raised the awareness of autism but also pushed for empirical studies that
indefinitely quashed the hypothesis that immunisation leads to autism.
45
1.7.3 Pregnancy-related complications
The functional brain abnormalities that have been observed in autism patients has led to a
hypothesis that the pathogenesis could begin as early as ‘in utero’ stages (for a detailed
discussion see DiCicco-Bloom et al., 2006). A comprehensive meta-analysis identified over fifty
prenatal factors as the subject of sixty four epidemiological studies (Gardener et al., 2009).
After a thorough review, authors identified several factors of significance.
Increased parents’ age at birth was identified as one of the factors associated with
autism (Gardener et al., 2009). The most recent meta-analysis of almost 26,000 autism cases
and 8.7 million controls supported the finding of the advanced maternal age as a risk factor for
autism; this effect remained significant while controlling for the paternal age (Sandin et al.,
2012). However, paternal advanced age has also been previously reported as a risk factor
(Reichenberg et al., 2006), leading to conclusions that the maternal and paternal advanced age
act as independent risk factors for increased incidence of autism through different
mechanisms. For older mothers, this could be due to a higher risk of chromosomal
abnormalities in the ova of advanced age (Kolevzon, Gross, & Reichenberg, 2007); in fathers the
risk is linked to the accumulation of de novo spontaneous mutations in spermatogonia as
paternal age advances (Kong et al., 2012; Reichenberg et al., 2006).
Gestational hypoxia (inadequate levels of oxygen supply) is another factor that may
lead to increased incidence of autism (Gardener et al., 2009; Newschaffer et al., 2007). For
example, the complications of maternal bleeding, maternal hypertension, lengthened labour,
cord complications and Caesarean section have all been linked to hypoxia and therefore autism
(Croen, Grether, Yoshida, Odouli, & de Water, 2005; Glasson et al., 2004; Hallmayer et al.,
2002). A study of labour complications found that associated high stress levels led to hypoxia
as well as a cerebral haemorrhage in the new-born. A follow up of children surviving the
haemorrhagic injury documented an increased risk for developing autism (Limperopoulos et
al., 2007).
46
There is also evidence for a twofold increase of risk for autism in children of mothers
with gestational diabetes (Gardener et al., 2009; Grabrucker, 2013). This type of diabetes
develops spontaneously in response to the heightened demand for insulin during pregnancy.
The mechanism of this physical pathology in relation to autism remains unknown. A recent
report expanded the metabolic conditions to include hypertension and obesity as predictors of
autism. The authors reported a significant association for autism and obesity alone (Krakowiak
et al., 2012). A suggestive trend was reported for hypertension and gestational diabetes, further
highlighting the need for systematic explorations of these risk factors in relation to autism.
Several other factors have been linked to increased incidence of autism and include
viral infections, zinc deficiency, or melatonin secretion. However, studies reporting on these
factors are often underpowered to define them as truly associated with autism (Gardener et al.,
2009; Grabrucker, 2013). For example, even though rubella as a viral infection has been linked
to autism, this relationship can also be explained by the more general immune predispositions
of the mother and the foetus (Grabrucker, 2013). Speaking more generally, the possibility of
alternative explanation of above findings is not the only limitation of epidemiological studies of
autism. Measurement imprecision, small sample size, broad disease definitions and
retrospective recall of experiences are the methodological limitations noted for all of the above
studies, calling for a better-structured research in this area (Gardener et al., 2009).
1.7.4 The hormonal in utero influences
The epidemiological research presented thus far has not considered the persistence of autism
in males in comparison to females (ratio of occurrence is generally estimated at 4:1) and this
gender imbalance is yet to be understood in genetic terms (Newschaffer et al., 2007).
Considering the cardinal features of social communication impairment and obsessive
need for routines more often observed in boys has led to the ‘extreme male brain’ theory of
autism (Baron-Cohen, 2002), which was informally mentioned as early as 1944 (Asperger, 1944;
translated by Frith in 1991). It suggested that autism is an extreme of the normal male profile
of ‘systemising’ – the need to predict behaviour as a system of quantitative variables,
47
manipulation of which will result in predictable and controllable outcome (Baron-Cohen, 2002).
The female brain is defined as the ‘empathising’ brain that cares about others.
The suggested process that leads to the dimorphic development of male and female
brains is that of in utero exposure to the sex hormone, testosterone. At first, proving of this
hypothesis was difficult due to challenges in obtaining of the amniotic fluid and the validity of
morphologic markers for increased testosterone levels. However, the recent developments of
mass spectrometry as well as blood collection from the umbilical cord allowed for evidence
gathering.
In one of the first studies of this kind, autism traits were reported by mothers in 6 to 10
year olds. The link between the foetal testosterone and the severity of autism trait scores was
explored. As a result, a positive relationship between the foetal testosterone and higher autism
trait scores was reported. Moreover, this effect was actually not dependent on gender and
reported to be true in both sexes (Auyeung et al., 2009). Using the same design, this association
was explored in a lot younger children - 18 to 24 months old. Elevated rates of autism traits
were this time reported only for boys and were positively correlated with increased foetal
testosterone levels (Auyeung, Taylor, Hackett, & Baron-Cohen, 2010). However, an
independent study measuring autistic traits in early adulthood (20 years of age) did not
replicate this association (Whitehouse et al., 2012). Males were reported to have higher number
of autism traits than females, although autism trait levels in either of the group were not related
to increased testosterone levels, collected from umbilical cord blood samples.
The most recent study, for the first time, explored this hypothesis in individuals with a
clinical ASD diagnosis. Males between ages of 15-21 were identified from the Danish birth
records as receiving an ICD 10 ASD diagnosis (equivalent to DSM IV) and having amniotic fluid
samples available. Concentration of sex steroids (progesterone, 17α-hydroxy-progesterone,
androstenedione and testosterone) as well as cortisol was also collected in a sample of controls.
Comparison analysis reported that the ASD group had elevated levels of all four hormones and
48
cortisol in comparison to controls, regardless whether the male was later assigned autistic
disorder, Asperger syndrome or PDD-NOS diagnosis (Baron-Cohen et al., 2014).
The above studies cumulatively provide mixed support for the extreme male brain
hypothesis. The evidence on autism traits associated with elevated testosterone levels in males
comes from two studies (Auyeung et al., 2009, 2010), one of which also found elevated levels
in females (Auyeung et al., 2009). Only one study to date explored this hypothesis in individuals
with clinical diagnosis and an effect of elevated amniotic fluid steroid hormones on later
diagnosis of autism (Baron-Cohen et al., 2014). What is not clear about the foetal testosterone
hypothesis is whether increased levels are sufficient to lead to autism or how they interact with
the known markers of genetic vulnerability, such as family history of autism (Skuse, 2010).
Secondly, Skuse rightly points out the lack of explanation as to how the theory is aligned with
the concept of autism as multidimensional construct (section 1.2). Considering the fractionated
triad hypothesis of the autistics traits and the small genetic overlap between the core
components, it would be important to explain the mechanisms through which the foetal
exposure increases the risk to every phenotypic component of the diagnosis (Skuse, 2010).
The epidemiological research into other than genetic effects on the development of
autism and associated features illustrates the difficulties this area of research faces – multiple
phenotypic presentations, increasingly complex genetic susceptibility, and multiple nongenetic agents that seem predictive of autism onset later in life. Identifying of these aetiologies
actually challenges the future genetic and epidemiological studies as researchers are forced to
build interdisciplinary models to account for the overall phenotypic and aetiological
heterogeneity of autism. But as George Bernard Shaw said: ‘science never solves a problem
without creating ten more’.
The increasing awareness of autism was met with concurrently increasing
governmental funding across the globe to come closer to finding the aetiology and ‘cure’.
Somewhat incidentally, however, the subsequent characterisation of what it means to live with
autism for the individual and their family has uncovered another area that also requires
49
scientific scrutiny – the comorbidity of other mental health problems in autism, briefly
discussed in the last section of this chapter.
1.8 Psychiatric comorbidity in autism
Chapters 5 and 6 provide a lengthy introduction to the concept of psychiatric comorbidity in
autism in twin samples. This section will primarily focus on introducing the broad key concepts.
Psychiatric comorbidity in autism has gained recognition mainly in the last decade, considering
that proper care should and must include amelioration of conditions that are treatable, while
autism itself thus far is treatment-resistant. The most recent autism review reports that higher
morbidity of individuals with autism is mainly due to comorbid conditions (Lai et al., 2014).
1.8.1 Comorbidity – an overview
The term ‘comorbidity’ was introduced by Feinstein (1970) to describe an ‘additional clinical
entity’ observed to manifest itself alongside the ‘index disease under study’. In psychiatry, cooccurring mental health disorders are reported to be very frequent. For example, an early study
on the US National Comorbidity survey reported that fifty one percent of patients with clinical
diagnosis of major depression also met the diagnostic criteria for at least one co-occurring
anxiety disorder (Kessler et al., 1994). Another example is that of the Australian National Survey
of Mental Health and Wellbeing reporting that twenty one percent of patients meeting the
DSM IV criteria for any mental health disorders also met criteria for three or more co-occurring
disorders (Andrews, Slade, & Issakidis, 2002).
As mentioned in the beginning of this chapter, the DSM has undergone regular reviews
that had a significant impact on the evolution of ASD diagnosis, and other mental health
domains have been similarly affected. It is generally acknowledged that the updates to the DSM
and ICD manuals, characterised by a considerable increase in the number of categories, are the
reasons for the increase in psychiatric comorbidity rates (Maj, 2005). The latest edition of DSM
recognises that comorbidity is very common and hence encourage diagnosis of all relevant
50
mental disorders in order to enable clinicians to devise a treatment plan that will increase
patient’s wellbeing and functioning (Pincus, Tew, & First, 2004).
The flip side of allowing for multiple diagnoses is the potential splitting up of complex
clinical condition into several parts that could potentially lead to a less holistic treatment of the
patient and an unwarranted polypharmaceutical intervention (Maj, 2005). However,
pharmacology is not the only treatment available; Cognitive Behavioural Therapy has proven
to be an extremely effective way in improving symptoms of many mental health disorders. This
means that diagnosis of multiple disorders has merit as it may pave the way for appropriate
behavioural therapies.
1.8.2 Causal models of comorbidity
The causes of co-occurrence of disorders can be investigated with models of comorbidity (Kim,
2010). An exhaustive review and quantitative description of these models is covered in two
seminal papers by Klein and Riso (Klein & Riso, 1993) and Neale and Kendler (Neale & Kendler,
1995).
The four potential reasons for comorbidity are: 1) sampling and base rates; 2) artefacts
of diagnostic criteria; 3) drawing incorrect diagnostic boundaries and 4) aetiological
relationships between disorders (Klein & Riso 1993). In terms of an aetiological link, two
explanations are given. First, that one psychiatric disorder is a risk factor/has a causal role for
another disorder (e.g. conduct disorder leading to an increased risk of substance abuse); and
second, that the two disorders co-occur due to overlapping (genetic or environmental)
processes (e.g. co-occurrence of major depression and anxiety due to common genes (Kendler,
Neale, Kessler, Heath, & Eaves, 1992)).
The risk factor/causal effects models are usually studied using longitudinal methods as
they can provide information on the course and stability of psychiatric disorders and their
comorbidity. Investigating correlating (genetic & environmental) risk factors is best facilitated
in behaviour genetics studies, which allow differential predictions of cross-family member
cross-disorder correlations based on differential genetic relatedness (discussed in detail in
51
Chapter 2, section 2.3.5). As such, twin studies are particularly informative in understanding and
decomposing the genetic and (shared and non-shared) environmental aetiology of comorbidity
(Neale & Kendler, 1995). Chapters 5 and 6 provide a detailed summary of twin studies on autism
traits and associated mental health problems. It must be noted that no studies to date
performed model-fitting analyses for this aetiological profile in a sample of twins with a clinical
or research diagnosis of ASD.
An alternative behavioural genetics’ model is the Direction of Causation (DoC) model,
which explores the relationship between variables in terms of phenotypic reciprocal or
unidirectional mechanisms (phenotypic = an observable characteristic determined by genetic
and environmental factors) (Heath et al., 1993; Simonoff, 2000), described in detail in Chapter
2, section 2.3.11. The advantage of the DoC model is that it can be fitted to cross-sectional
datasets. For example, an Australian study of female twins aged 18-45 years old measured
recently experienced psychological distress and its relationship with the self-perceived
childhood environment (Gillespie, Zhu, Neale, Heath, & Martin, 2003). In effect, authors found
support for a unidirectional phenotypic mechanism between the two variables – memories of
cold, overprotective and autonomy-limiting parents seemed to increase the level of
psychological distress (Gillespie et al., 2003).
Knowing the mechanisms of comorbidity is of high scientific and practical importance,
especially in regards to treatment (Rutter, 1994). Considering that there is no cure for autism,
the scientific community has made a concerted effort in the last decade to highlight the
incidence of comorbid psychiatric disorders and to minimise their impact, especially when they
are treatable. The next section will provide an overview of the recent findings on associated
psychiatric difficulties in individuals with autism.
1.8.3 Autism and associated psychiatric difficulties
The observations of associated psychiatric difficulties go back as far as the 1970s, however, the
tendency was to attribute these difficulties in children and adults to autism itself, as ASD
52
diagnosis encompasses a set of extremely severe symptoms and other difficulties were not
deemed as pivotal (Skokauskas & Gallagher, 2012).
1.8.4 Early studies
The earliest observations relate to onset of depressive states as individuals with ASD
entered the puberty stage of development (Rutter, 1970). Further reports described the
development of self-destructive behaviours, outbursts of rage and bodily self-harm and lashing
out at others during the adolescent years (Gillberg & Schaumann, 1981). What became clear
was that adolescents with the most severe forms of autism displayed the most destructive
behaviours. Additionally, the acquired academic skills obtained pre-puberty were irrevocably
lost during the puberty and individuals never retained the pre-puberty functionality (Gillberg &
Schaumann, 1981).
In another case study of Asperger’s syndrome, Gillberg considers that perhaps the
syndrome should leave the spectrum as a subgroup of ASD and be instead considered a
personality disorders clearly distinguishable from infantile autism, due to the high cooccurrence of affective disorders and suicidal attempts (Gillberg, 1985). He provided a detailed
account on the feelings of inadequacy amongst the brighter ASD individuals whom longed for
closer friendships but were incompetent in finding them due to their primary diagnosis. A
separate study reported on 34 Asperger individuals to show symptoms of affective disorder
(23%), psychotic symptoms (17%) and attempted suicide (11%) during puberty (Wing, 1981).
Taking these findings into account, Gillberg strongly advocated treatment of autistic
youngsters and paying special attention to them during puberty. Failing to do so, would mean
that ‘full psychiatric health in adult life [would be] an exception rather than the rule’ (Gillberg,
1984). He also noted that in the face of autism and its prime features of lack of ability to
communicate, finding out of youngsters’ internal feelings becomes even more difficult as they
are unable to express their anxiety or depression, and for this feature to be a particular
challenge in ASD and associated psychiatric difficulties research (Gillberg & Schaumann, 1981).
53
Subsequent studies started to provide fuller accounts of the ASD individuals and those
displaying autism traits, as well as their family characteristics. Lainhart and Folstein (1994)
reported on 8 out of 17 cases to have depressive episodes (without mania) and on 7 cases
meeting criteria for major affective disorder (a major disturbance of emotions or extreme mood
swings, not otherwise assigned to a detectable organic abnormality). Moreover, ASD
individuals that were reported to show the additional difficulties often became more
noncompliant, uncooperative and unmanageable, while displaying high levels of hyperactivity
and aggression (Lainhart & Folstein, 1994).
Lainhart and Folstein have also commented that in 50% of the cases there is a family
history of existing affective disorders and suicide, indicating a genetic link. Moreover, they also
reported a greater risk of development of affective disorder in first-degree relatives of autistic
probands (Lainhart & Folstein, 1994). Further two studies found similar patterns for major
depressive disorder (Bolton, Pickles, Murphy, & Rutter, 1998; DeLong & Dwyer, 1988).
The seminal paper by Lainhart in 1999 took the evidence provided by all previous
researchers to present a coherent account of comorbidity in autism (Lainhart, 1999). While
reporting on increased rates of major depression and social phobia in first-degree relatives,
Lainhart expressed that the additional problems escalate due to core defining features of
autism and associated impairment, medical disorders and life experiences related to having
autism. She proposed that future studies ought to determine whether the risk of developing
psychiatric disorder in autism are over and above the risk of these problems in the general
population or other development disorders. If the risk is indeed increased, potential risk factors
like genetics, neurologic, cognitive, and environmental effects need to be identified and
understood for provision of better detection and treatment (Lainhart, 1999).
1.8.5 Modern studies
One of the first modern studies exploring the associated psychiatric difficulties in
autism was that by Leyfer et al (2006). The authors modified the Kiddie Schedule for Affective
Disorders and Schizophrenia in order to reflect other psychiatric manifestations in ASD. In a
54
sample of 109 children with ASD diagnosis aged 5-17, 13% of the sample met criteria for
depressive disorder, 8% for manic disorders, 23% for anxiety disorders and 30% for Attention
Deficit Hyperactivity Disorder (ADHD). Overall, 30% of children had a diagnosis of two
additional mental health disorders (Leyfer et al., 2006).
In the second study of this kind, one comorbid disorder was reported by 70% of
participants (of 112, age 12) and 41% at least two (Simonoff et al., 2008). Similar rates to that
of Leyfer et al were reported: 29% had social anxiety disorder, 28% had ADHD and 28% had
oppositional defiant disorder. A follow up of the same cohort at age 16 revealed that the
additional problems were persistent and domain-specific from childhood to adolescence
(Simonoff et al., 2013). There is also evidence on comorbid conduct disorders in ASD, whose
persistence is moderated by the presence of comorbid ADHD symptoms, as the child grows
older (Flouri, Midouhas, Charman, & Sarmadi, 2015). The above studies provide the first insight
into the high prevalence of comorbid internalising and externalising disorders in individuals
with ASD since the early studies.
In a review of 27 studies, rates for comorbid disorders in autism were: 0-6% for
schizophrenia, 0-50% for affective disorders or symptoms, 5-35% for generalised anxiety, 1064% for simple phobias and 1-37% for obsessive compulsive disorders (authors highlighted
associated psychiatric difficulties and ASD diagnostic and sample size heterogeneity across all
studies) (Skokauskas & Gallagher, 2012). A separate review discussed the comorbidity between
ASD and ADHD diagnoses and reported that a significant proportion of children with ASD, 3195%, also report symptoms of inattention/hyperactivity/impulsivity (Antshel, Zhang-James, &
Faraone, 2013). The interesting feature of the relationship between ASD and ADHD is that up
to DSM IV diagnosis of both disorders in one individual was not allowed, despite a large body
of evidence that they are often comorbid.
The early and modern studies of comorbidity in autism clearly demonstrate that cooccurring disorders and mental health problems should be taken very seriously and reveal how
important it is to regularly screen those at risk of ASD diagnosis for associated psychiatric
55
difficulties. There seems to be a slight contrast, however, between types of associated
psychiatric difficulties reported by both sets of studies. The early accounts speak often of
affective and psychotic-like disorders, without mentioning explicitly ADHD or anxiety
symptoms. The modern studies, on the contrary, highlight these problems most often, without
putting so much emphasis on psychotic symptoms.
There are a number of studies that attempted the use of Cognitive Behavioural Therapy
to alleviate high levels of internalising symptoms and report moderate effectiveness (Spain,
Sin, Chalder, Murphy, & Happe, 2015). A recent report provides evidence that use of risperidone
is beneficial in treating aggression symptoms in ASD, although it should only be prescribed on
a short-term basis (Dinnissen, Dietrich, van den Hoofdakker, & Hoekstra, 2015). Both of these
findings clearly demonstrate that treatments that work well in the general population do not
readily translate to a similar level of effectiveness in ASD individuals, further highlighting the
urgent need for better understanding of the aetiology of associated psychiatric difficulties.
As highlighted in Chapters 5 and 6, several twin studies to date aimed to provide clues
whether associated psychiatric difficulties and autism traits share common genetic or
environmental influences. However, no study to date performed model-fitting analyses to
elucidate the aetiology of this relationship in a twin sample with ASD diagnosis. This fact in
itself highlights the early stages of comorbidity research in autism, which aims to identify the
pathways that would meaningfully explain why individuals with autism display such high levels
of associated psychiatric difficulties. Understanding of the aetiology of this overlap will aid the
development of better-suited treatments.
1.9 Thesis goals and structure
The content of the proposed thesis revolves around two primary goals:
(1) To re-evaluate the importance of genetic and environmental influences on ASD and the
BAP, considering the recent findings on significance of shared environmental influences.
56
(2) To investigate the aetiology of comorbidity of associated psychiatric difficulties in an ASD
twin sample, as no studies to date looked at the underlying genetic and environmental
aetiology in this group using twin model-fitting techniques. As ASD twin samples are small, the
same set of analyses is conducted in a general population twin sample to see if relationships at
clinical extreme also hold for individual differences at subclinical levels.
The current thesis is structured as followed:
1.9.1 Aim 1
The evaluation of genetic and environmental influences on ASD is made through a novel
statistical modeling-fitting approach in an ASD twin sample ascertained from a larger, general
population-representative birth cohort. Using the sample of twins with ASD diagnosis and
those categorised as Broad Spectrum, genetic and environmental contributions are calculated
for the range of autistic behaviours measured with continuous as well as categorical
instruments. The peer-reviewed paper reporting these findings has been published in the May
edition of JAMA Psychiatry (Colvert & Tick, et al. 2015).
1.9.2 Aim 2
Inspired by the controversy that still exists around the significance of shared-environmental
effects in favour of genetic effects (e.g. Frazier et al., 2014; Hallmayer et al., 2011), a systematic
review and meta-analysis was carried out on all existing ASD twin studies to date. The aims of
this study were not only to produce less biased estimates by applying appropriate sample
selection and ascertainment correction methods, but also to increase statistical power to
detect effects of small size by leveraging a larger sample size. A scientific manuscript based on
this work has been submitted for publication in the Journal of Child Psychology and Psychiatry
and is currently under peer review (Tick, B., Happé, F., Bolton, P., Rutter, M. & Rijsdijk, F.).
1.9.3 Aim 3
The aetiological overlap between ASD and externalising and internalising problems as
identified by the Strengths and Difficulties Questionnaire (SDQ, indicative of associated
57
psychiatric difficulties) was measured in the SRS sample, as used for Aim 1. The association
between psychiatric difficulties and ASD was tested for the presence of correlated genetic and
environmental factors. Furthermore, using the twin structure of the data and differential
aetiology across these variables, more specific direction-of-causation models were explored. A
scientific manuscript has been submitted to the Journal of American Academy of Child
Psychology and Psychiatry and is currently under peer review (Tick, B., Colvert, E., McEwen, F.,
Stewart, C., Woodhouse, E., Gillan, N., Hallett, V., Lietz, S., Garnett, T., Simonoff, E., Ronald,
A., Bolton, P., Happé, F., Rijsdijk, F.).
1.9.4 Aim 4
Using the population twin sample, the aim was to deal with the aetiological overlap between
autism traits and internalising and externalising traits from the SDQ measure, as in Aim 3. The
size of this sample allowed exploration of sex-differences in aetiological overlap. Again, in
addition to a basic correlated risk-factor model, more specific direction-of-causation models
were explored and possible direction of causation effects discussed.
58
Chapter 2 Methods
2.1 Overview
This chapter will provide an overview of the samples, measures and statistical methods that
were used in chapters 3-6. It will explain in detail the use of twin studies, whether based on
diagnostic or general population samples, and how they help to answer the questions regarding
genetic and environmental influences on complex traits. The chapter will conclude with
assumptions (and their implications) of the classical twin model.
2.2 Samples
The data from two separate samples were used in the analyses described in Chapters 3, 5 and
6. The Twins Early Development Study (TEDS) sample will be introduced first as from this
population birth-cohort the clinical sample was derived - the Social Relationship Study (SRS).
2.2.1 The Twins Early Development Study (TEDS) sample
The TEDS (Haworth, Davis, & Plomin, 2012) is a unique longitudinal population birth-cohort of
over 10,000 active twin pairs (www.teds.ac.uk). The twins followed in this sample were born in
England and Wales between January 1994 and December 1996. At first, all families with the
record of a live twin birth were contacted by the Office for National Statistics (ONS) to require
parental consent for participation in the TEDS, resulting in 16810 responding families. The first
data collection was performed when twins were 2 years old and as of December 2014 every twin
would have turned 18 years old. The phenotypic data is available at different intervals for
language, cognition, academic achievement, and common behavioural problems in the context
of normal development. The data used in Chapters 3 and 5 was derived when twins were 8 years
old and the data used in Chapter 6 was derived when twins were 12 years old (more detail is
provided in those chapters).
The TEDS is established as highly representative of the UK population: 92.8% are
white, 40.1% of parents had an A-level or higher educational status, and 46.4% mothers were
59
employed as compared to statistics reported by Walker et al (Walker, Maher, Coulthard,
Goddard, & Thomas, 2001). Zygosity of the twins was assigned using a standard parent-rated
zygosity questionnaire (Goldsmith, 1991) that proved to be accurate in 95% in comparison to
DNA marker method (Price et al., 2000). For the remaining 5% of pairs, genetic marker testing
was performed. For every wave of assessment, prior parental consent was always obtained and
an ethical approval has been provided by the King’s College London ethics committee
(reference: 05/Q0706/228). Additional collaborative spin-off studies were conducted in the
TEDS to explore more specific hypotheses, generally performed on the TEDS sub-sample of
twins. The SRS was one of such samples, discussed next.
2.2.2 The Social Relationships Study (SRS) sample
Information provided in this section is presented in Chapter 3 as a published journal article with
an accompanying supplementary materials section (Colvert & Tick et al., 2015). The main
involvement of the student in the SRS sample was cleaning, quality control and analysis of the
dataset as described in each Chapter of this thesis. She did not collect the data as the study was
completed prior to her PhD.
The SRS sample underwent a two-stage selection process. The main aim was to include
all families in which one or both twins were suspected or confirmed to have an Autism Spectrum
Disorder (ASD). The first stage involved identification of families who had at least one twin
scoring at or above 15 points on the Child Autism Screening Test (CAST) completed when twins
were 8 years old (Scott, Baron-Cohen, Bolton, & Brayne, 2002).
To ensure that there were no systematic biases in the employed sampling technique
and to quantify any selective attrition, letters and CAST questionnaires were also sent to 1,900
families from the original TEDS sample that had ended their participation in the study at an
early stage; this was done at the SRS time point (ages 12-15). These included families who were
excluded from the main study due to severe medical and genetic conditions, such as severe
developmental delay. This process yielded 34 families where at least one twin scored at or
above 15 or where an ASD was reported. The two CAST mail-outs yielded 289 families. In
60
addition, 210 families had reported an ASD diagnosis to TEDS (via phone/mail). Some of those
reporting a diagnosis were also in the group identified as above the CAST cut-off, and the pool
of potential families with suspected ASD was 412.
Of these families, 82 (20%) could not be contacted, either due to address changes or because
they had subsequently dropped out of the TEDS, or refused participation. See Figure 2-1.
In total, 330 families were then asked to complete the ASD module of the Development
and Well-Being Assessment (DAWBA) (Goodman, Ford, Richards, Gatward, & Meltzer, 2000)
via telephone interview, as the second stage of the SRS sample selection. As a result, after
exclusion of 10 pairs on the basis of missing zygosity information or other medical conditions
(e.g. Down's syndrome and profound deafness) (Dworzynski, Happe, Bolton, & Ronald, 2009),
the DAWBA identified 230 families with at least one child who met criteria for ASD.
To increase the likelihood of capturing the most complete sample, all child psychiatrists
in the UK were sent a letter asking for details of twins born between 1994 and 1996 and
suspected of ASD. These were checked to ensure that they were not already part of the TEDS
and if not they were sent information packs and CAST questionnaires. In addition adverts were
placed in the Twins and Multiple Births Association newsletter and on the National Autistic
Society’s website. These additional recruitment methods yielded five further families who were
not part of the main TEDS population, bringing the total SRS sample to 235 families with at
least one twin suspected of, or diagnosed with, ASD. From this group, 89 families could not be
contacted or declined to participate in the study; 17 families opted to complete only the
questionnaire section of the project. 129 families (62%) had home (or research centre) visits.
In order to categorise the sample, the gold standard diagnostic tools Autism Diagnostic
Observation Schedule (ADOS) and Autism Diagnostic Interview-Revised (ADI-R) were used.
Two researchers worked with each family, one carrying out the ADI-R and the other the ADOS
for one twin and then swapping for the second twin. This design meant that different assessors
carried out the ADI-R and ADOS assessments within each pair in order to minimize any effects
61
TEDS families screened, n=8,941
Screening Stage 1: “At risk” families selected, n=412
- Families with at least one twin scoring ≥15 points on CAST, n=289
- Families with at least one twin reported to have existing ASD diagnosis, n=210
(NB some twins met both criteria so numbers add up to less than 499).
Families that refused/no
response n=82 (20%).
Screening Stage 2: Home visits selection
DAWBA telephone
interviews
completed in n=330
families.
At least one child met
DAWBA criteria for ASD,
n=230.
Families
invited
to take
part in
SRS
n=235.
Families meeting eligibility
criteria for TEDS but not
recruited during earlier
phases of study; identified
via UK psychiatrists mailout and adverts, n=5.
Families
that
refused/no
response
n=89 (38%).
Face-to-face assessment
Families eligible for
ASD home assessment,
n=146.
“Low risk” (CAST<12)
comparison families that
underwent home
assessment, n=80 (excluded
n=1), of whom
n=29 completed DAWBA.
Families that
underwent home
assessment, n=129
(excluded n=1).
Families
completing
questionnaire
booklet
ONLY,
excluded from
analysis,
n=17 (12%).
Twin analysis
DAWBA n=359 pairs
MZ=91, DZ SS=133, DZ OS=135 pairs
Best-estimate Diagnosis n=207 pairs
MZ=56, DZ SS=77, DZ OS=74 pairs
ADOS n=203 pairs
MZ= 55, DZ SS=77, DZ OS=71 pairs
ADI-R n=205 pairs
MZ=56, DZ SS=76, DZ OS=73 pairs
Figure 2-1 The SRS sample selection, adapted from Colvert & Tick et al., (2015).
62
of rater bias. In total, ADOS assessments were conducted for 249 individual twins (spread over
124 pairs) and ADI-R interviews were carried out for 253 individual twins (spread over 126 pairs).
The advantage of using different diagnostic tools was that it allowed comparison of parent and
observer rated measures of autistic symptoms. For 89 cases (37%), they did not lead to the
same diagnosis. All cases with diagnostic disagreement were referred to a team of psychiatrists
who reviewed all available sources of information and reached a consensus decision (the Bestestimate Diagnosis). The weighted kappa statistic for the ADI-R and ADOS was .67, indicating
a substantial agreement and in keeping with the weighted kappa of .79 (Bolte & Poustka, 2004).
One of the limitations of the SRS sample is the lack of the IQ data in light of the observation
that different levels of IQ can impact the recognition of ASD symptoms, as discussed in section
1.2.
A comparison group was also included in the study, consisting of 79 families from the
TEDS sample who scored below 12 on the CAST at age 8 and who lived in the South East of
England. This group was matched to the suspected ASD group in terms of gender, zygosity,
age and Socioeconomic Status (SES). They completed the same battery of assessments (e.g.,
measures of Intelligence Quotient) as the suspected ASD sample but, because they were
selected to be at low risk for ASD, did not complete the diagnostic assessments (i.e. ADOS and
ADI-R). A subsample (n = 29) completed the ASD module of the DAWBA either online or by
telephone interview. Within the final sample those with ASD were broadly comparable to those
eligible for participation (CAST ≥ 15 or suspected ASD) but who did not take part, with the
exception of gender (Zygosity χ2(1) = 1.5, p=.23; SES t(397) = -1.2, p=.25; CAST t(420) = -1.5,
p=.14; Gender χ2(1) = 20.1, p<.001, 36% of the high CAST/suspected ASD group were female,
versus 17% of the final sample).
2.3 Measures
2.3.1 Autism traits - Childhood Autism Spectrum Test (CAST)
The Childhood Autism Spectrum Test (CAST; formerly known as the “Childhood Asperger’s
Syndrome Test”) is designed for parents to indicate levels of autism-like symptoms in children
63
as delineated in the DSM IV and ICD 10 (Scott et al., 2002). The complete questionnaire contains
39 items to be evaluated with a ‘yes’ or ‘no’ answer. Twelve items each form the social and the
communication scale (24 items in total) and 7 items form the non-social scale (see Appendix 1).
The remaining 8 items enquire about the general development. The total cut off for the three
scales is 15 or above (out of 31 items) and has been shown to demonstrate a 100% sensitivity,
97% specificity, a positive predictive value of 50% for an ASD diagnosis (Williams et al., 2005)
and a good test-rest reliability (Williams et al., 2006). Numbers of twins with CAST data are
provided in each chapter separately.
2.3.2 The Development and Well-being Assessment (DAWBA)
The DAWBA is a battery of mixed assessment techniques developed to identify a range of
psychiatric diagnoses in children and adolescents (5-17 year olds) aligned with ICD 10 and DSM
IV diagnostic criteria (Goodman et al., 2000). It has been widely used in both the
epidemiological and clinical sample surveys to report on child psychopathology (Meltzer,
Gatward, Goodman, & Ford, 2000). The DAWBA has showed to have good validity by displaying
89% specificity in a community sample and 92% sensitivity derived from a clinical sample
(Goodman et al., 2000).
Application of the DAWBA in the TEDS sub-sample of 10 to 12 year old twins was to
further elucidate families with at least one twin at risk of ASD diagnosis. Therefore, only the
section related to autism spectrum conditions of the DAWBA battery was used – 15 questions
relating to social difficulties (Cronbach’s α = .92), 14 questions about the repetitive and
restricted behaviours and interests (RRBI) (α = .87) and 3 questions about developmental
language milestones (α = .74) (Dworzynski et al., 2009). The original social difficulties scale was
divided into the social and communication difficulties sub-domains of satisfactory internal
validity (social α = .81, communication α = .83). The affected individuals received the DAWBA
diagnosis as followed: autism, Asperger syndrome and ASD Other. An autism diagnosis was
given when the individual met the operational criteria delineated in ICD 10 and DSM IV. When
the individual met these criteria but showed no language delay, an Asperger syndrome
64
diagnosis was assigned. To receive an ASD Other diagnosis, the child displayed a minimum of
three probable or two definite symptoms within the social sub-domain, two probable or one
definite symptom within the communication sub-domain and two probable or one definite
symptom within the RRBI sub-domain.
2.3.3 Autism Diagnostic Interview-Revised (ADI-R)
The ADI-R is considered as one of the gold-standard clinical diagnostic instruments for
evaluation of autism in children and adults, assessing social reciprocal interactions, language
and communication and RRBI’s (Lord, Rutter, & Couteur, 1994). Information on current
behaviours is given by the caregiver during a semi-structured clinical interview consisting of 93
questions. It also contains an ‘age of onset’ item to specify if the abnormalities were present
prior to 36 months. The interview begins with the description of the child’s early development
and is followed by 41 questions about the verbal and nonverbal communication. Questions 50
to 66 cover the social development and play, followed by 13 questions describing the RRBI’s.
Finally, 14 questions are asked about the general behaviour as well as the cognitive and physical
skills. The ADI-R diagnostic algorithm classification is provided by the Autism Genetics
Resource Exchange (AGRE, www.agre.org, (Geschwind et al., 2001)) and has been applied in
the SR sample in the following manner:
1. Autism is identified using the well-validated ADI-R scoring algorithm (Lord et al., 1994).
2. NQA (Not Quite Autism) represents individuals who are no more than one point away from
meeting the autism criteria on any or all of the 3 "content" domains (i.e., social, communication,
and/or behaviour), and meet the criteria on the “age of onset” domain; or, individuals who meet
criteria on all 3 "content" domains, but do not meet the criteria on the "age of onset" domain.
3. Broad Spectrum defines individuals who show patterns of impairment along the spectrum of
the pervasive developmental disorders. This is a broad diagnostic category that encompasses
individuals ranging from mildly- to severely-impaired. This category potentially includes such
pervasive developmental disorders as the PDD-NOS and Asperger's syndrome, which are used
in many genome scans; however, this classification is not based on any validated algorithms.
65
Autism Genetics Resource Exchange (www.agre.org) affected status algorithms
I. "Autism" classification: 1. social >= 10, and 2. communication: verbal >= 8; nonverbal >= 7, and
3. behaviour >= 3 PLUS age of onset >= 1.
II. "Not Quite Autism (NQA)" classification:
(A) Meets the cut-offs on all 3 "content" domains, but not the age of onset domain: 1. social >=
10, and 2. communication: verbal >= 8; nonverbal >= 7, and 3. behaviour >= 3 PLUS age of onset
= 0.
OR: (B) Is no more than 1 point below the cut-off on any, or all, of the 3 "content" domains, and
meets the “age of onset” domain: 1. social >= 9, and 2. communication: verbal >= 7; nonverbal
>= 6, and 3. behaviour >= 2 PLUS age of onset >=1.
III. "Broad Spectrum" classification: Age of onset >= 0; PLUS does not meet the criteria for
Autism or NQA; PLUS meets one or more of the following (A, B, or C):
(A) Shows a severe deficit on at least one domain; severe is defined by the scores at one or more
of the following levels (e.g., 1 or 2 or 3): 1. social >= 8, or 2. communication: verbal >= 7;
nonverbal >= 6, or 3. behaviour >= 3.
(B) Shows moderate deficits in at least two domains; moderate is defined by the scores at two
or more of the following levels (e.g., 1 + 2, or 2 + 3, or 1 +3): 1. social >= 4; 2. communication >=
3 (nonverbal or verbal); 3. behaviour >= 2.
(C) Shows only minimal deficits, but in all three domains at the following levels: 1. social >= 3,
and 2. communication >= 2 (nonverbal or verbal), and 3. behaviour >= 1.
2.3.4 Autism Diagnostic Observation Schedule (ADOS)
Often, the ADI-R and the ADOS assessments jointly form the basis for an overall ASD diagnosis.
The ADOS is a 35-40 minute-long standardised observational assessment in a form of a semistructured activities led by the examiner, providing standard contexts in which child’s social
interaction and communication skills (and other behaviours) emerge (Lord et al., 1989). Each of
the four modules accommodates a range of languages abilities and chronological age: Module
1 assumes no speech or single words only; Module 2 assumes some speech but not verbal
66
fluency; Module 3 assumes verbal fluency in children and Module 4 assumes verbal fluency
typical for adolescents and adults.
The SRS participants were evaluated with Module 3 of the ADOS as over 90% of the
sample exhibited verbal fluency. Modules 1 or 2 were employed in the remaining 10% of the
sample (Lord et al., 2000). An updated ADOS algorithm was used (provided by communication
with C. Lord, 2008) to yield the scores for communication, social interaction and RRBIs. The cut
offs for ASD and Autism diagnostic groups are provided in Table 2-1 below. For diagnostic
classification purposes, these diagnostic groups were merged to create one ASD category. The
additional Broad Spectrum diagnostic group included individuals scoring below the cut offs (-2
points, column 5 in Table 2-1), hence corresponding to the Broad Spectrum category on the
ADI-R.
Table 2-1 The cut off scores for each of the diagnostic ADOS categories.
Module
Language ability
Autism
ASD
Broad Spectrum
1
No speech
16
11
9
1
Single words
12
8
6
2
Single words (>5 years)
9
8
6
3
Verbal fluency
9
7
5
2.3.5 Strengths and Difficulties Questionnaire (SDQ)
Strengths and Difficulties Questionnaire (Goodman, 1997) is a well-regarded dimensional
(Goodman & Goodman, 2009) measure of mental health, used in clinical as well as school
settings to measure the psychological wellbeing of children between the ages 2-17 (Becker,
Woerner, Hasselhorn, Banaschewski, & Rothenberger, 2004). It has been translated into 100
languages worldwide. The main asset of the SDQ is that although a brief instrument, it is
comprehensive while covering behavioural manifestations related to different mental health
disorders.
The SDQ has been showed to have a comparative predictive validity in separating the
psychiatric and nonpsychiatric samples (Goodman, 1997) and that it can be used as a screening
67
questionnaire to identify children with a mental health disorder (Goodman, Renfrew, & Mullick,
2000; Goodman, 2001), in particular clinical cases of ASD and ADHD (Russell, Rodgers, & Ford,
2013). The SDQ showed an 80% specificity and 85% sensitivity, based on carers and teachers
reports (Goodman, Ford, Corbin, & Meltzer, 2004).
The SDQ is made of five subscales, each containing 5 items measuring: Emotional
symptoms, Conduct problems, Hyperactivity, Peer relationships and Pro-social behaviours
(control scale). Severity of problems is assessed using a three-point rating: 0=Not True,
1=Somewhat True, 2=Certainly True (see Appendix 2). As already mentioned, the SDQ is
designed as a dimensional measure although suggestive cut offs for Borderline and Abnormal
mean levels have been suggested (see Table 2-2 below).
Table 2-2 Suggestive SDQ cut offs, adapted from Goodman, (1997).
Parent completed
Total Difficulties Score
Emotional Symptoms Score
Hyperactivity Score
Conduct Problems Score
Peer Problems Score
Prosocial Behaviour Score
Teacher Completed
Total Difficulties Score
Emotional Symptoms Score
Hyperactivity Score
Conduct Problems Score
Peer Problems Score
Prosocial Behaviour Score
Normal
Borderline
Abnormal
0-13
0-3
0-5
0-2
0-2
6-10
14-16
4
6
3
3
5
17-40
5-10
7-10
4-10
4-10
0-4
0-11
0-4
0-5
0-2
0-3
6-10
12-15
5
6
3
4
5
16-40
6-10
7-10
4-10
5-10
0-4
2.4 Statistical Approaches
2.4.1 The Biometric Theory of Inheritance – from single-gene to polygenic model
Every chapter in this thesis is underpinned by the theory of hereditary transmission of
phenotypes (Plomin et al., 2013). These rules were first explained by Mendel (1866), whom
observed that there are two hereditary elements of a trait, which can interchangeably separate
or segregate during a reproductive cycle of pea plants. Mendel deducted that both of these
elements are passed from the parents and that one of the elements can overrule the other,
therefore is “dominant”. Having one of those elements is sufficient for a certain trait to be
68
expressed (for example a smooth skin on the pea plant). The overruled, in other words the
“recessive” element, only expresses when both elements are of “recessive” type at that location
within the genome. Nowadays, these elements are recognised as the basic units of heredity
and simply referred to as genes and the alternative dominant and recessive forms are the alleles.
Figure 2-2 Example of inheritance of Huntington’s disease.
Abbreviations: d=dominant, r=recessive. Gametes=the eggs and sperm containing one allele, which is
then recombined (source: Plomin et al., 2013).
The best example of a single genes disorder in humans, albeit with devastating
consequences, is the Huntington’s disease (see Figure 2-2 above) (Gusella et al., 1983). The
differential genotypes (unique allele combinations) of both parents explain the fundamental
aim of ‘genetics’ as a science – to explain to what extent the differences in the genotype account
for the differences in the phenotype (an observable characteristic determined by both genes
and the environment). In this example, the mother is the carrier of the dominant allele, which
results in an estimated risk of 50% for her child to inherit this lethal disease. The only reason
why Huntington’s prevails in the population is because the onset of symptoms begins in middle
adulthood and past the carriers’ reproductive years.
In contrast to Huntington’s disease, the genetic aetiology of behavioural
traits/disorders is a lot more complex as demonstrated by the recent genome-wide association
study (GWA) finding of at least 108 schizophrenia-associated genetic loci (multiple locations in
the genome) (Ripke et al., 2014). However, we are some distance from being able to identify
the remaining genes for schizophrenia, or for any other psychiatric disorder, as it requires
69
enormous sample sizes (at least 50,000 individuals) to detect the small effect each gene is
known to have, commonly explained as polygenic effects (Plomin et al., 2013).
Under the polygenic model, the non-additive (dominant) and additive genetic effects
are summed up across the loci and explained as the total genetic effect on the phenotype. For
a locus to have an additive genetic effect means that an allele is assigned a genotypic value of
0, 1 or 2 reflecting that none, one or both alleles are associated with a trait. When a child inherits
an allele with a value of 1, it means that its additive effect will contribute to the expression of
child’s phenotype as much as it did to their parent’s phenotype. Consequently, child’s
resemblance of their parents increases, regardless of other allele at that locus or other loci
(Plomin et al., 2013). This quantitative genetic approach allows estimating the heritability of
behavioural traits/disorders by looking at the phenotypic similarity of related individuals, which
is the hallmark of twin studies.
2.4.2 The Classical Twin Method
The quantitative genetic approach aims to estimate the genetic and environmental effects that
influence the population variation of a trait. More simply, it tries to answer the following
question: what are the reasons for individual differences in a trait within human populations?
Humans are quite homogenous as they share 99% of the DNA (a molecule carrying genetic
code). Thanks to the 1% of the genetic code, we become unique. However, the MZ twins are an
exception to this rule as they share 100% of their DNA and show identical additive and nonadditive genetic effects across all loci. In comparison, the DZ twins share on average 50% of the
genetic variants. The classical twin design takes this resemblance difference and uses it in the
estimation of genetic and environmental components of the variance within a trait as well as
between traits (Plomin et al., 2013). Therefore, the greater phenotypic similarity on a trait in
MZ twins compared to DZ twins arises as a function of their greater genetic relatedness.
The phenotypic similarity of twins growing up in the same family can be attributed to
additive genetic (A), dominant genetic (D) and shared environmental (common, C) factors. The
non-shared environmental (E) factors are the influences that lead to dissimilarity (and contain
70
the error measurement). Both MZ and DZ twin similarity is calculated by intra-class correlations
(most appropriate for group data structures) that are treated as coefficients of twin similarity
(Shrout & Fleiss, 1979). The MZ:DZ intra-class correlation ratio can indicate to what extent
genetics are important for a trait. For example a ratio of 2:1 would indicate strong additive
genetic and no shared environmental influences. If the DZ correlation is higher than half that of
MZ correlation, then the increased similarity can only be due to C effects. The DZ correlation
lower than half that of MZ correlation indicates dominance effect.
The heritability (including both additive and dominant effects) and environmental
effects statistics, h2, c2 and e2 can be derived with pen and paper using the similarity coefficients
as per Falconer’s formula (Rijsdijk & Sham, 2002):
h2 (heritability)= 2 (rMZ – rDZ)
c2 (shared environments)= rMZ – h2
e2 (non-shared environments)= 1 – h2 + c2
These formulae, however, only apply to univariate models and are only useful in certain simple
cases (i.e. it does not work for separating additive from dominant genetic effects), cannot take
into account missingness in the data and cannot provide confidence intervals to indicate
whether an estimate spans zero. These problems are addressed by structural equation models
(SEM) using full information maximum likelihood, both of which provide the basis of
sophisticated model-fitting techniques, described in detail in Chapters 3, 5 and 6.
2.4.3 The Univariate ACE Model
The ACE model is the most common causal model fitted to twin data, for this reason, it will be
used in all of the examples from this point onward. The formal estimation of A, C and E
components utilises matrix algebra in program OpenMx (Boker et al., 2011). The method of
path analysis, illustrated by a path diagram below (Figure 2-3), decomposes the covariation
within twin pairs and then across traits, assuming an ACE causal model (Wright, 1918).
71
Figure 2-3 ACE Path diagram.
The observed data of Variable 1 (square) is regressed against the latent (unmeasured, circles) causal A, C
and E variance components; a + c + e are the path coefficients. Single headed arrow indicates partial
regression statistic and double headed arrow indicates covariance (correlation) statistic. The latent A, C
and E variables can only covary with itself, illustrated by the set value of 1.0 for the first latent additive
variable (left); the same applies for the remaining latent variables. Grey dotted arrows signify path
tracing rules, described in the text.
The drawn arrows in Figure 2-3 represent the statistical effects of one variable on
another, which are independent of any other variable. The covariation, whether measured
within twin pairs or across variables, is the sum of all legitimate paths. The double headed arrow
between the latent variables signifies the covariance between them and is set to 1.0 for additive
effects for MZ twins and 0.5 for DZ twins, reflecting the correlation coefficient of their genetic
relatedness. All of the latent shared environmental factors are assumed to be equal within pairs,
regardless of zygosity, signified by the covariance coefficient of 1.0. No link between latent
non-shared environmental factors indicates covariance of 0, as these factors contribute only to
twins’ dissimilarity. The variance for Twin 1 and Twin 2 is obtained by path tracing (illustrate by
dotted grey arrows in Figure 2-3) from a, trace up through the double headed arrow to make a
loop (accounting for variance of A) and then trace back down through a again, repeating the
same process for c and e. In effect, mathematically it is equal to a*1.0*a + c*1.0*c + e*1.0*e =
a2 + c2 + e2. Then, the path coefficients are traced from Twin 2 to Twin 1 (order is immaterial) to
72
obtain a*1.0*a + c*1.0*c for MZ twins and a*0.5*a + c*1.0*c for DZ twins, resulting in the
expected covariance of a2 + c2 for MZ and 0.5a2 + c2 for DZ twins.
2.4.4 Estimating ACE of the liability for a disorder – model-fitting to ordinal data
Despite the growing evidence that psychiatric disorders are quantitative traits, the diagnoses
for most disorders are based on dichotomous or ordinal categorisation. For this reason,
behavioural geneticists developed a model, which assumes that the liability for the disorder is
underpinned by a normally distributed continuum of risk (see Figure 2-4 below).
The indicative threshold for the disorder is a hypothetical construct acting as a border
between those that display some symptoms of the disorder and those with symptoms severe
enough to meet the diagnostic criteria (Plomin et al., 2013). In practice, this threshold is set as
a z-score with the cumulative probability at the right-hand side reflecting the prevalence rate
for that disorder in the general population. To obtain the variance-covariance matrix in order
to estimate the genetic and environmental influences on the disorder, the tetrachoric (for
binary categorisation) or polychoric (for ordinal categorisation) correlations are derived
(Pearson & Lee, 1900), instead of the intra-class correlation appropriate for continuous data.
The ACE model described above can be applied to ordinal data in the sense that the variance
decomposition is applied to the assumed underlying normal distribution of liability and based
on the MZ:DZ ratios of the tetrachoric correlations.
Figure 2-4 The liability-threshold model of disorders.
The dotted line indicates the threshold, which discriminates between individuals displaying some of the
disorder’s symptoms (below the threshold) and those that exceed it (above the threshold) by displaying
symptoms severe enough to warrant a psychiatric diagnosis.
73
2.4.5 The Multivariate Approach
The univariate ACE model-fitting can be easily extended to derive the genetic and
environmental factors influencing the variance of and covariance between multiple measures –
a twin-model fitting technique that enables the exploration of shared aetiologies between
(psychiatric) disorders.
The simplest multivariate model (the bivariate model), which analyses the genetic &
environmental basis of the covariance between two traits (Plomin et al., 2013), will be used to
illustrate the logic of the model. The principle of this method is not only to estimate the
expected variance and covariance for the MZ and DZ twin pairs for each trait, but also the
expected cross-covariances in twins for the two traits. If we call the measures A and B and
annotate the first twin as 1 and the second as 2 (A1 would annotate measure A for the first twin),
the variance and the covariance would be contained within the following symmetric matrix
(Plomin et al., 2013; Rijsdijk & Sham, 2002):
𝐓𝐰𝐢𝐧 𝟏 𝐭𝐫𝐚𝐢𝐭 𝐀 𝐓𝐰𝐢𝐧 𝟏 𝐭𝐫𝐚𝐢𝐭 𝐁 𝐓𝐰𝐢𝐧 𝟐 𝐭𝐫𝐚𝐢𝐭 𝐀 𝐓𝐰𝐢𝐧 𝟐 𝐭𝐫𝐚𝐢𝐭 𝐁
𝐌𝐙
Within − Twin
Within − Twin
Cross − Twin
Cross − Twin
𝐓𝐰𝐢𝐧 𝟏 𝐭𝐫𝐚𝐢𝐭 𝐀
𝑉𝑎𝑟(𝐴1)
𝐶𝑜𝑣(𝐵1𝐴1)
𝐶𝑜𝑣(𝐴2𝐴1)
𝐶𝑜𝑣(𝐵2𝐴1)
𝐓𝐰𝐢𝐧 𝟏 𝐭𝐫𝐚𝐢𝐭 𝐁
𝐶𝑜𝑣(𝐴1𝐵1)
𝑉𝑎𝑟(𝐵1)
𝐶𝑜𝑣(𝐴2𝐵1)
𝐶𝑜𝑣(𝐵2𝐵1)
𝐌𝐙
Cross − Twin
Cross − Twin
Within − Twin
Within − Twin
𝐓𝐰𝐢𝐧 𝟐 𝐭𝐫𝐚𝐢𝐭 𝐀
𝐶𝑜𝑣(𝐴1𝐴2)
𝐶𝑜𝑣(𝐵1𝐴2)
𝑉𝑎𝑟(𝐴2)
𝐶𝑜𝑣(𝐵2𝐴2)
𝐓𝐰𝐢𝐧 𝟐 𝐭𝐫𝐚𝐢𝐭 𝐁
𝐶𝑜𝑣(𝐴1𝐵2)
𝐶𝑜𝑣(𝐵1𝐵2)
𝐶𝑜𝑣(𝐴2𝐵2)
𝑉𝑎𝑟(𝐵2)
The above variance-covariance symmetric matrix is divided into two sections of
calculation: within-twin cross-trait variance/covariance (marked yellow) and cross-twin crosstrait covariance (marked green). The matrix provides ten pieces of information (that is the
diagonal and lower off diagonal pieces; the upper off-diagonal pieces represent the same
information as the lower off-diagonal pieces): the four within-twin diagonal pieces
74
Var(A1+B1+A2+B2) define the variance for both measures for both twins within a single pair (in
this example MZ twin pair); the within-twin off-diagonal piece Cov(A1B1) + Cov(B1A1) express
the phenotypic covariances between traits A and B for the first twin and Cov(A2B2) + Cov(B2A2)
for the second twin; the off-diagonal cross-twin pieces Cov(A1A2) + Cov(B1B2) define the crosstwin covariance for each trait and pieces Cov(A1B2) and Cov(B1A2) are the cross-twin crosstrait covariance.
The last two pieces ((Cov(A1B2) and Cov(B1A2)) expressed as cross-twin cross-trait
correlations are very informative when estimated in MZ and the DZ twin pairs separately.
Comparing the ratio of these correlations would indicate the source of the phenotypic overlap
between traits A and B. If the ratio is about 2:1, the two traits overlap because of common
genetic influences. If the DZ cross-twin cross-trait correlation is higher than half of that of MZ,
shared environmental factors jointly influence both traits.
The Bivariate model can be fitted to the data in two ways: using a Cholesky
decomposition (but interpreted as standardised correlated-factors) or using a Gaussian
decomposition (less traditional, but sometimes necessary). The two are described briefly in the
sections below.
2.4.6 The Cholesky Decomposition
In multivariate analyses, a Cholesky decomposition is usually fitted to the data. In a Cholesky,
ordering of the variables matters as the latent components of variable A are given precedence
over the components estimates for the subsequent variables (see Figure 2-5). The variablespecific variance components of variable B are uncorrelated with variable A. Therefore, the first
set of components accounts for the genetic and environmental influences on variable A and
those shared with variable B. The genetic influences estimated for variable B indicate the
residual influences not accounted for and therefore not correlated with those estimated for the
variable A (Loehlin, 1996).
75
Figure 2-5 The bivariate Cholesky decomposition path diagram for a twin pair.
The Cholesky decomposition is usually fitted to the data because it has much better
structure for mathematical optimisation of estimated parameters. However, the way we
interpret the output depends on the hypothesis at hand, discussed in detail by Loehlin (1996).
If the order of variables is immaterial, we present and interpret a standardised solution, the so
called ‘correlated-factors’ interpretations (Figure 2-7). We discuss this model further in section
2.4.8. Sometimes it is more convenient to fit a Gaussian decomposition to the data, discussed
in the next section.
2.4.7 The Gaussian Decomposition
The Gaussian decomposition is illustrated in Figure 2-6. In this decomposition, actual
correlation paths are fitted between the latent ACE factors and individual paths are fitted from
the latent factors to the observed variables (in squares). Even though this structure is more
problematic in terms of the mathematical optimisation, it is more convenient to the Cholesky
decomposition when applying constraints across groups. Application of constraints of e.g. the
correlational paths is particularly useful when fitting sex-limitation models on multivariate data
when DZ opposite-sex twin pairs are included (see section 2.4.10).
76
Figure 2-6 The bivariate Gaussian decomposition path diagram for a twin pair.
2.4.8 The Bivariate Genetic Model (Correlated-Factors Interpretation)
The advantage of the correlated-factors interpretation is that the order of the variables is
immaterial. The standardised model for only one twin (Figure 2-7) is represented as the two
variables with their individual standardised variance components (a21, c21, e21 and a22, c22, e22).
The strength of overlap between the factors is represented as the additive genetic (rA), shared
environmental (rC) and non-shared environment (rE) correlation, indicating the extent to which
the same A, C or E factors influence the two traits. The extent to which the phenotypic
relationship between variable 1 and 2 is due to these correlated factors is weighted by the
square-root of the actual standardized effect sizes (a21, c21, e21 and a22, c22, e22). The
standardized correlated-factor solution is interpreted such that the path from the A1 factor to
the Variable 1 and the A2 factor to Variable 2 is the square root of their respective standardized
path estimates (heritabilities=h2) and the correlation path between A1 and A2 is the genetic
correlation between these variables (rA). The same principle is applied to non-shared
environmental effects (E). Then, the proportion of the phenotypic correlation (rPH) is calculated
as due to correlated additive genetic effects A (rPH_A = h12 * rA * h22) and due to correlated nonshared environmental effects E (rPH_E = e12 * rE * e22) expressed as proportions of rPH. Figure
77
2-7 is accompanied by a mathematical explanation as to how the genetic correlation is derived
using matrix algebra.
Figure 2-7 The correlated-factors interpretation of the Cholesky decomposition.
2.4.9 The Bivariate Genetic Model - combining continuous and ordinal data
The models fitted in Chapter 6 examined the (genetic) relationship between two continuous
traits from the general population. Chapters 3 and 5 examine the genetic relationships between
continuous CAST/SDQ variables and the ordinal ASD variable, respectively, assessed in the
diagnostic SRS twin sample. In the fitted Cholesky decomposition, the ordinal variable is always
entered as second due to specifics of the software, rather than theoretical reasons. In all these
cases, we interpret the correlated-factors model (Figure 2-8).
When fitting to combined continuous-ordinal data, the principles of the bivariate model
outlined above hold, with the additional assumptions we make about the liability distributions
underlying the ordered categories. Thus, for the ordinal ASD diagnosis variable, a liabilitythreshold model was assumed (explained in section 2.4.4) with a standard normal distribution
underlying the ordered categories, which are discriminated by thresholds (z-values on the
standard normal distribution).
In the population, these thresholds correspond to the
prevalence of the (sub-categories of the) disorder. However, an additional complication of the
SRS data is that the twins were selected on having ASD or not (the controls) which means that
78
from the data we are not able to estimate the actual prevalences of these categories. The
thresholds on the liability were therefore fixed to population ‘known’ values of ASD prevalence:
1st threshold of 5% separated the unaffected and twins with Broad Spectrum (Baird et al., 2006);
2nd threshold of 1% separated the Broad Spectrum and ASD twins (Brugha et al., 2011). Not
fixing the thresholds would lead to biased results.
Figure 2-8 The bivariate genetic model on combined continuous-ordinal data.
This is a conceptual depiction for visualisation purposes and path tracing rules cannot be applied.
2.4.10 The Sex-Limitation Model
Sex-limitation models are only considered in Chapter 6, in which the analysis involves general
population data. Full sex-limitation models allow investigation of the presence of qualitative
and/or quantitative sex differences influencing a trait. Qualitative differences between males
and females relate to the idea that the genetic and environmental influences are separate for
each gender. For example, colour blindness is caused by a recessive allele on the X
chromosome. Because males inherit one X chromosome and females inherit two, males have a
higher statistical probability of suffering from colour blindness as they require only one and not
two alleles for the trait to express (Plomin et al., 2013). Quantitative differences indicate that
the magnitude of genetic environmental influences on the variance of a trait is genderdependent, after correcting for the mean differences between males and females. Detail on
how these differences are tested in sub-models is given in Chapter 6.
79
The general principle is that qualitative differences are suspected for a trait if the DZ
opposite-sex pairs have a much smaller correlation than the same-sex DZ pairs, which in the
model will result in genetic correlations (rAm-f) between males and females in DZ opposite-sex
twin pairs smaller than 0.5. In the same way, the shared environmental factors between males
and females in these pairs (rCmf) are estimated to correlate less than 1.0 (both assumptions are
tested in separate models). See Figure 2-9 for the bivariate ACE model for opposite-sex twin
pairs (E not depicted for simplicity). Quantitative sex differences are indicated when the MZ:DZ
correlation ratios are different across gender, indicating different aetiologies across gender and
resulting in genetic and environmental path coefficients (am, cm, em and af, cf, ef) which cannot
be constrained to be equal.
Figure 2-9 The bivariate sex-limitation model including paths for DZ opposite-sex twins.
80
2.4.11 The Direction of Causation (DoC) Model
This model provides an alternative approach of defining the source of covariance and was
utilised in Chapters 5 and 6. In this twin model, the phenotype itself is the independent variable
(rather than the genetic and environmental influences) as the phenotypic relationship between
two (or more) variables is conceptualised as a reciprocal or unidirectional causality (Heath et al.,
1993). An example of the direction of causation (depicted by an arrow) is given in Chapter 5
testing the hypothesis that, for example, Emotional symptoms cause ASD (EmoASD) or that
the reverse is true and ASD causes Emotional symptoms (EmoASD). A reciprocal relationship
is illustrated in Figure 2-10. In that example, trait 1 is aetiologically different (ACE model) from
trait 2 (AE model) therefore warrants phenotypic causation model-fitting as the differentially
predicted cross-twin cross-trait covariance form the basis for the DoC model. As such, because
the latent genetic and environmental factors across the two traits are independent of each
other, the only mechanism that generates the correlation between them are either reciprocal
(r + r’) or unidirectional (r or r’) causal phenotypic effects (Heath et al., 1993).
Figure 2-10 The bivariate direction of causation model.
The reciprocal causal paths r and r’ indicate the phenotypic causation of trait 1 on trait 2 and the reverse,
adapted from Heath, (1993).
In Chapter 5 this model is fitted on the combined continuous-ordinal data of the selected
SRS sample so that all specifics outlined in section 2.4.9 apply (i.e. a liability threshold model
81
for ASD with fixed threshold to correct for selection on ASD). In Chapter 6 this model is fitted
to continuous SDQ and CAST population TEDS data.
2.5 Assumptions and implications of the twin design
Twin studies have been very useful in psychology (and behavioural science in general) to help
understand how individual differences in traits are explained by genetic and environmental
differences between individuals. However, like all statistical models, there are certain
assumptions that apply when considering findings of twin studies (Rijsdijk & Sham, 2002). The
next sections outline the most important ones.
2.5.1 Assortative Mating
The twin design assumes that individuals mate randomly. Assortative mating takes into
account the fact that people couple with others that are more alike phenotypically, but
sometimes people are also attracted to the ‘opposites’ of themselves. Some research suggests
that the former happens more often than the latter (Neale & Cardon, 1992). Furthermore, it is
suggested that people tend to seek mates of similar intelligence.
For the DZ twins, if assortative mating was true, the range of variation would decrease
resulting in higher similarity of DZ sibling, potentially inflating the estimates of shared
environments at the cost of the lowered heritability estimates. In 2011, a theory was proposed
by Simon Baron-Cohen that the rise of autism prevalence in the recent decade is potentially
due to ‘geeks seeking out geeks’ and more regular passing of autism-like traits genes to
offspring. As explained in section 1.8.4 in the Introduction, Baron-Cohen thought of autism as
an expression of hyper-systematising (Baron-Cohen, 2002), therefore those on the spectrum
have higher propensity for subjects like maths and physics that utilise them.
However, this theory would only make sense when referring to those individuals on the
spectrum that are highly functioning, but even in that instance this claim is still only a
speculation. Individuals with the most severe autism condition often require 24-hour care and
are unable to lead independent life, therefore have diminished chances of mating and
82
reproducing. Early twin studies have found that when assortative mating does play a role in
psychiatric traits, the overall influence on the estimates is at best negligible (Neale & Maes,
2001).
There are several studies that investigated assortative mating in autism using
continuous measures of autistic traits. Using the Social Responsiveness Scale (SRS), one of the
earliest studies reported a significant correlation between spouses for social impairment in the
order of .38 (Constantino & Todd, 2005). Using the same measure, De La Marche et al (2015)
reported a stronger correlation of .53 (De La Marche et al., 2015). However, two further studies
that examined the same question using the Autism Spectrum Quotient (AQ) did not replicate
these results (Hoekstra et al., 2007; Pollman et al., 2010). It is somewhat hard to generalise from
these mixed findings. Further studies should assess this issue, preferably with independent
measures and multiple informants.
2.5.2 Equal Environments assumption
This assumption relates to the idea that environmentally driven similarity of the MZ and DZ
twins is because both types of twin pairs respond to environments equally (Rijsdijk & Sham,
2002). This is in line with observations that, regardless of zygosity, twins concurrently
experience the environments of the same womb, the same family and age in a similar manner.
However, if for example MZ twins are treated as more similar by their parents, higher MZ
correlations relative to DZ correlations could result in inflated genetic estimates and an
underestimation of the shared environmental influences.
There is indeed evidence that MZ twins are treated more similarly than DZ twins
(Loehlin & Nicholls, 1976). In a classic study, however, it has been shown that even when DZ
twins thought they were MZ twins, their self-perceived scores on anxiety and mood disorders
did not increase as a function of perceived resemblance (Kendler, Neale, Kessler, Heath, &
Eaves, 1993). Another study has further supported this claim for other psychiatric traits
(Hettema, Neale, & Kendler, 1995).
83
2.5.3 Gene-Environment (G x E) association
It is well established that behavioural traits are driven by nature AND nurture, meaning that
genes act in conjunction with the environment. The interplay between the two is suggested to
occur via a gene-environment (G x E) correlation as well as gene-environment (G x E) interaction
(Rutter, Moffitt, & Caspi, 2006).
Gene-environment correlation can be also described as genetic control of exposure to
environments (Rijsdijk & Sham, 2002). By implication it means that individuals with select
certain environments in relation to their genetic propensity. The most common examples are
the active, passive and evocative G x E correlations. Active G x E correlation occurs when
environments are sought out by the individual as a function of their genotype, for example
those with genetic liability for depression having a higher probability for experiencing stressful
life events (Kendler et al., 1995). Passive G x E correlation occurs when both genetic and
environmental propensities are passed onto their offspring, for example a football loving family
will deliver a higher number of football-related activities and conversations. Finally, evocative
G x E correlation occurs when individual’s phenotype (due to inherited genes) evokes a reaction
from others in the environment. For example, a hyperactive child may lead at times to parents
temporarily withdrawing from the child, potentially making the child to feel neglected and thus
evoking hyperactive behaviour to draw parental attention.
Specific implications are noted for the mentioned G x E correlations. A positive active
G x E correlation in the MZ twins could lead to inflated genetic estimates. This is because the
increased genetic resemblance would lead the MZ twins to seek out more similar environments
and enhancing further their total similarity in comparison to DZ twins (Rijsdijk & Sham, 2002).
Conversely, negative active G x E correlations will in turn decrease the genetic estimates. A case
of positive passive G x E correlation could lead to shared environmental influences to be
inflated, as inherited environments would lead to heightened DZ twin similarity, over and
above that assumed by the genetic similarity.
84
G x E interaction can be explained as genetic control of susceptibility to environments
(Rijsdijk & Sham, 2002). In lay terms, the differential genotypic predispositions will lead to a
varied response to the same environment. Or, that certain genotypes are more susceptible to
the effects of changes in the environment than other genotypes. For example, a higher
heritability was estimated for depressive symptoms in children when the peer rejection levels
were high (Brendgen et al., 2009). Using the same sample, heritability for aggressive
behaviours was lower when the child experienced a positive relationship with the teacher
(Brendgen et al., 2011). However, measuring of environments in the twin samples is only a
recent occurrence and at present there little knowledge on fitting of G x E interaction models
to multivariate data. Implications of not modelling of G x E interaction can lead to inflated
genetic estimates, especially when there is a positive interaction between genes and the shared
environmental influences (Rijsdijk & Sham, 2002). As the MZ and DZ twins are assumed (under
the equal environments assumption) to share 100% of shared environments, the level of
interaction with these environments will be greater in MZ twins due to their increased genetic
resemblance in comparison to DZ twins (that share on average 50% of genes). Conversely,
interaction of genes with the environments specific to individual twins (non-shared, the only
source of dissimilarity in the twin design), the genetic estimates will not be affected as these
environments are assumed to not lead to similarity in either zygosity. Statistically, if this
interaction exists, it will be subsumed under the E estimate accounting for the decreased
MZ/DZ similarity.
Overall, the limitations set out above require further exploration in order to quantify
their impact on the heritability and environmental estimates derived from twin studies. Until
such evidence emerges, the estimates ought to be taken as indicative. It is the study of specific
factors, rather than total genetic and environmental effects that may eventually provide us with
more absolute answers.
85
2.5.4 Generalising from twin to population samples
As twin studies are performed on twins, and twin births have characteristic differences from
conventional births, some have questioned the extent to which twin findings generalise to the
wider population. This is not surprising, considering that twins are more likely to be premature
than non-twin births and twin mothers do experience more birth-related difficulties (Rijsdijk &
Sham, 2002). Birth complications have been particularly under the scrutiny in relation to
psychiatric traits, considering the evidence that the development of schizophrenia has been
linked to obstetric difficulties (Kotlicka-Antczak, Pawelczyk, Rabe-Jablonska, Smigielski, &
Pawelczyk, 2014).
In relation to autism, a host of studies attempted to discern whether twinning leads to
an increased incidence of ASD diagnosis (Ronald & Hoekstra, 2011). Some studies suggest that
this might be the case (Betancur et al., 2002; Greenberg et al., 2001). However, a response using
population statistics method demonstrated that increased rates of autism traits were an effect
of ascertaining pairs of affected siblings rather than due to multiple birth (Visscher, 2002).
Furthermore, large population-based studies found no increase in incidence of ASD diagnosis
in twins (Croen et al., 2002; Hallmayer et al., 2002; Hultman et al., 2002).
Other studies have examined whether rates of autistic traits (and probable ASD diagnosis)
are higher in twins compared to singletons/non-twin siblings. Looking specifically at males, one
study reported slightly higher levels of autism traits in twins compared to male singletons,
although participants were recruited from separate samples and not matched for age, IQ and
socio-economic status (Ho et al., 2005). In contrast, a family study that took into account these
potentially confounding variables found no differences on self-reported autistic traits between
twins and non-twin siblings (Hoekstra et al., 2007a). Two large population-based cohort studies
also found overall no difference in twin-sibling mean levels of autistic traits (Ronald et al., 2006;
Curran et al., 2011). In summary, the majority of the reported studies do not support the
proposal that being a twin increases the possibility of receiving an ASD diagnosis.
86
Chapter 3 Heritability of Autism Spectrum Disorder in a UK
Population-Based Twin Sample
This chapter is presented in this thesis as published in JAMA Psychiatry and is cited as followed:
Colvert, E.*, Tick, B.*, McEwen, F., Stewart, C., Curran, S., Woodhouse, E., Gillan, N., Hallett,
V., Lietz, S., Garnett, T., Ronald, A., Plomin, R., Rijsdijk, F., Happe, F., Bolton, P. Heritability of
Autism Spectrum Disorder in a UK Population-Based Twin Sample, JAMA Psychiatry, 2015 In:
JAMA Psychiatry, 72 (5), 415-423.
* joint first author
87
3.1 Abstract
88
3.2 Introduction & Methods
89
90
91
3.3 Results
92
3.4 Discussion
93
94
3.5 Manuscript-specific references
95
96
3.6 Supplementary Online Materials
97
98
99
100
101
102
103
104
105
Chapter 4 Heritability of Autism Spectrum Disorders: a meta-analysis
of twin studies 1
4.1 Abstract
The aetiology of Autism Spectrum Disorder (ASD) has been recently debated due to emerging
findings on the importance of shared environmental influences. However, two recent twin
studies do not support this and instead re-affirm strong genetic effects on the liability to ASD,
a finding consistent with previous reports. This study conducts a systematic review and metaanalysis of all twin studies of ASD published to date.
A PubMed Central, Science Direct, Google Scholar, Web of Knowledge structured
search was conducted to identify all twin studies on ASD published to date. Thirteen primary
twin studies were identified, seven were included in the meta-analysis by meeting systematic
recruitment criterion; correction for selection and ascertainment strategies, and applied
prevalences were assessed for these studies.
The meta-analysis correlations for monozygotic twins (MZ) were almost perfect at .98
(95% Confidence Intervals, .96-.99). The dizygotic (DZ) correlation, however, was .53 (95% CI
.44-.60) when ASD prevalence rate was set at 5% (in line with the Broad Autism Phenotype of
ASD) and increased to .67 (95% CI .61-.72) when applying a prevalence rate of 1%. The metaanalytic heritability estimates were substantial: 64%-91%. Shared environmental effects
became significant as the prevalence rate decreased from 5% to 1%: 07%-35%.
We demonstrate that: 1) ASD is due to strong genetic effects; 2) shared environmental
effects become significant as a function of lower prevalence rate; 3) previously reported
significant shared environmental influences are likely a statistical artefact of over-inclusion of
concordant DZ twins.
1
Chapter adapted from Tick, B., Bolton, P., Happé, F., Rutter, M.* & Rijsdijk, F.* (2015). Heritability of
Autism Spectrum Disorders: a meta-analysis of twin studies. Journal of Child Psychology and Psychiatry
(in press). * joint senior author
106
4.2 Introduction
Autism is known as a severe pervasive neurodevelopmental disorder with poor prognosis. It has
a considerable impact on the family as well as social, educational and health care systems. A lot
of research effort has gone into understanding the causes of individual differences in autistic
behaviour, with clear evidence for genetic effects.
Twin studies of the heritability of Autism Spectrum Disorders (ASD; an umbrella term
denoting autism, Asperger syndrome and Pervasive Developmental Disorder Other) have been
reviewed and summarized most recently by Ronald & Hoekstra (Ronald & Hoekstra, 2011).
Their review included the 7 primary studies published up to 2011, annotated in Table 4-1
(below).
Ronald & Hoekstra (2011) demonstrated that heritability estimates were high and
largely comparable across the published studies. This was true even when the diagnostic
criteria for autism were broadened to include ASD – median estimate of probandwise
concordance for the former was 76% in MZ twins and 0% in DZ, and for the latter 88% and 31%,
consistent with a high proportion of heritable effects on ASD. Five further studies have been
published since: Hallmayer et al., 2011, Frazier et al., 2014, Nordenbæk, Jørgensen, Kyvik, &
Bilenberg, 2014, Sandin et al., 2014 and Colvert & Tick et al., 2015. However, findings from the
recent five studies following the Ronald & Hoekstra review have suggested a more complicated
aetiological picture.
Both Hallmayer et al., (2011) and Frazier et al., (2014) reported significant influences of
shared environmental effects, steering the debate towards the higher importance of the
environment rather than a genetic predisposition to ASD. In Hallmayer et al., (2011) the
variance of the liability to ASD in a clinical sample was significantly accounted for by shared
environmental factors (58%) and only moderately by genetic effects (38%). In Frazier et al.,
(2014), an ever higher estimate of shared environmental effects was reported: 64%-78%,
depending on symptom measure (albeit no well-established diagnostic tools were applied). In
contrast, using a population-based cohort of ~2 million individuals, Sandin et al., (2014) showed
107
that the individual risk for ASD increased with genetic relatedness, with no effects of shared
environment. Two further twin studies (Nordenbaek et al., 2014 and Colvert & Tick et al., 2015)
again confirmed the importance of genetic effects on ASD by showing high MZ concordance
rates of 95% and 94% compared to and 4% & 46% for DZ pairs, respectively, and little support
for shared environmental effects.
The potential reasons for the differences across the recent studies are discussed in our
latest publication (Colvert & Tick et al., 2015) and include issues regarding sample
ascertainment and measurement differences. Ronald & Hoekstra (2011) highlighted the fact
that diagnoses across studies were often based on unstandardised or proxy measures of ASD
rather than the conventional in-person assessments such as the Autism Diagnostic Observation
Schedule (ADOS) (Lord et al., 1989) and Autism Diagnostic Interview – Revised (ADI-R) (Lord
et al., 1994).
Secondly, since heritability estimates across studies are often derived from selected
clinical twin samples prior knowledge of the prevalence (threshold on the liability) is required
for statistical modelling. The prevalence for ASD is considered to be 1% in the general
population (Baird et al., 2006; Brugha et al., 2011; Elsabbagh, Divan, Koh, Kim, Kauchali,
Marcin, et al., 2012). However, it is well recognised that the Broad Autism Phenotype cases just
falling short of the diagnostic cut-off are part of the underlying continuous liability distribution
of ASD (Maxwell, Parish-Morris, Hsin, Bush, & Schultz, 2013). This category is captured by a
lower threshold on the liability consistent with a prevalence of around 5%, supported by the
fact that 5.8% of general population score above the cut off on the Childhood Autism Spectrum
Test and 1%/5.8% of these individuals receive an ASD diagnosis (Williams et al., 2005). The use
of different thresholds (assumed prevalences) and multiple vs single threshold models could be
a potential source of discrepancies in estimates of heritability and environmental effects across
studies.
108
The aims of the current study are four-fold:
1. to reconsider the inconsistent findings, especially with respect to the evidence for
shared environmental influences on ASD;
2. to independently estimate twin correlations and heritability estimates for each study
while correcting for selection and ascertainment strategy (especially when the original study
did not do so);
3. to conduct a meta-analysis of the published studies using appropriate corrections for
selection and ascertainment strategy for each individual study;
4. to study the effects of assumed prevalence rates (fixed thresholds) on twin
correlations and heritability estimates in each individual study and on the combined sample.
To summarize, in this paper we report results of a quantitative meta-analysis of the
combined data of published twin studies of ASD to date. Many primary twin studies on low
prevalence disorders such as ASD are based on ascertained samples of relatively small size. The
benefit of the present analysis is not only to produce the most unbiased estimates by applying
appropriate ascertainment and selection correction methods, but also to increase statistical
power to detect effects of small size by leveraging a larger total sample size.
4.3 Methods
4.3.1 Sample
To identify all published studies on heritability of ASD, a PubMed, Science Direct, Google
Scholar, and Web of Knowledge computerised search was undertaken to identify any prior
reviews of ASD research as well as independent investigations of the topic. This produced all
studies reviewed by Ronald & Hoekstra (2011) as well as five studies published after their
review, providing a total of 13 eligible studies (Table 4-1). They were geographically oriented in
Northern Europe (UK + Scandinavia), Japan and the United States.
Regarding general inclusion/exclusion criteria, we followed the protocol outlined by
Sullivan, Kendler, & Neale, 2003. Due to rarity of diagnostic samples of twins with ASD, we
aimed to include as many studies as possible. For this reason the criterion for conformity on
109
measurement instruments was loosened, although most included studies employed DSM/ICD
diagnostic criteria. To maximise meta-analysis sample size, data on opposite-sex twin pairs
were also included where reported. When several publications reported on the same sample,
we included the most recent report (exclusion criteria for studies 1 and 4 as annotated in Table
4-1). Study 11 (Sandin et al., 2014) was excluded on the basis of lack of information on the
specific number of concordant pairs, and the fact that the majority of twins in this extended
family study are most likely reported on in a previous study (Lichtenstein, Carlström, Råstam,
Gillberg, & Anckarsäter, 2010), included in the meta-analyses.
Of importance was the systematic recruitment criterion, which decisively excluded
studies 7 and 10 as they did not systematically select the probands from the general population.
In addition, twins in these two studies did not undergo any in-person screening to validate their
diagnosis, a practice followed by every other study included in this meta-analysis. Study 2 was
excluded, since, although using systematic recruitment, it was one biased in favour of families
with multiple cases of autism.
Diagnosis blind to zygosity and co-twin’s status mostly features in studies conducted
after 2000, in line with recent practice in twin research (Sullivan et al., 2003). However, this is
not applicable to Random Population Ascertainment (study 8) and in study 9, where proband
selection was on the basis of electronic records - an alternative source for systematic
recruitment due to technological developments in patient data storage. We retained both of
these studies since they met all other criteria.
110
Table 4-1 List of primary twin studies on ASD.
Source
Country
Systematic
Blind to zygosity
Basis for diagnosis
Diagnostic Range of diagnostic AscerRecruitment;
& co-twin status;
Criteria
outcomes
tainment
(n pairs)
DZ OS included
Type
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< Included studies >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
3. Steffenburg et Nordic
Yes; included
No; No
Records and interview; ABC,
DSM III R
Autistic Disorder1
CA
al., 1989
Regions
triplets; (21)
the Lotter checklist & DIPBEC
5. Le Couteur et
United
Yes; included
No; No
Records and interview; ADI,
ICD 10 &
Broader Phenotype2, CA
al., 1996 **
Kingdom triplets; (48)
ADOS
DSM IV
Autistic Disorder3,
Atypical Autism4
6. Taniai et al.,
Japan
Yes; (45)
Yes; Yes
Records and semi-structured
DSM IV
Autistic Disorder3,
CA
5
2008
interview; CARS-TV
Asperger Syndrome ,
PDD-NOS6
8. Lichtenstein et Sweden
Yes; (7982)
n/a; Yes
Records and telephone
DSM IV
Pervasive DevelopRPA
7
al., 2010
interview; A-TAC
mental Disorders
9. Hallmayer et
United
Yes; (192)
n/a; Yes
Records and interview; ADI,
DSM IV
Autistic Disorder3,
IA (π=.96)
8
al., 2011
States
ADOS
ASD
12. Nordenbæk
Denmark Yes; (36)
Yes; No
Records and interview; ADOS,
ICD 10 &
Autistic Disorder3,
IA (π=.76)
et al., 2014
DISCO
DSM IV
Asperger Syndrome5,
PDD-NOS6
13. Colvert & Tick United
Yes; (127)
Yes; Yes
Records and interview; CAST,
ICD 10 &
Autism Spectrum
CA
8
et al., 2015
Kingdom
DAWBA, ADI, ADOS, BestDSM IV
Disorder , Broad
estimate Diagnosis
Spectrum Disorder9
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< Excluded studies >>>>>>>>>>>>>>>>>>> <<<<< Reason for exclusion >>>>>
1. Folstein &
United
Yes
Yes; No
Records and interview
DSM
Superseded by Le
CA
Rutter, 1977 **
Kingdom
released in
Couteur et al 1996
1980
2. Ritvo et al.,
United
No
Not specified; Yes
Records and interview
DSM III
Biased Systematic
IA (π=.86)
1985 *
States
Recruitment
111
4. Bailey et al.,
1995 **
7. Rosenberg et
al., 2009 *
United
Kingdom
United
States
Yes; included
triplets
No
Yes; No
Records and interview
ICD 10
No; Yes
Voluntary Registry: care-giver
reported diagnosis of ASD
DSM IV TR
10. Frazier et al.,
2014 *
United
States
No
No; Yes
Voluntary Registry: care-giver
reported diagnosis of ASD
DSM IV TR
11. Sandin et al.,
2014 **
Sweden
Yes
n/a; Yes
National Registry Diagnostic
information
ICD-10
Superseded by Le
Couteur et al 1996
Did not meet
Systematic
Recruitment
criterion
Did not meet
Systematic
Recruitment
criterion
No concordance
information, and,
Lichtenstein et al
(2010) already
reported on at least
part of this twin
sample
CA
n/a*
n/a*
CA
112
Abbreviations: CA = complete (or double) ascertainment (π=1); IA = incomplete ascertainment (0<π<1), π is calculated; RPA = random population ascertainment; DSM =
Diagnostic and Statistical Manual of Mental Disorders; ICD = International Classification of Diseases.
Note: Study number is assigned chronologically based on publication year. Studies 1 and 4 (**) were excluded on the basis that they were superseded by a more recent study of
the same research group; study 11 (**) did not provide twin concordance data, additionally, Lichtenstein et al (2010) have already reported on twins included in study 11. Studies
2, 7 and 10 (*) did not meet criterion of Systematic Recruitment.
Definition of diagnostic outcomes:
1 Autistic Disorder (DSM II R): onset prior to age 5. Criteria: 8 out of 16 items across three categories: at least 2 difficulties in social interaction category and at least 1 difficulty in
communication and restricted, repetitive and stereotyped patterns of behaviour categories.
2 The Broad Phenotype: measured behavioural domains of communication impairment and social dysfunction. Meeting the cut-off for either deficit alone or in combination was
required for diagnosis. RRB's not included in this criterion.
3 Autistic Disorder (DSM-IV): onset prior to age 3. Criteria: at least 2 difficulties in social interaction category and at least 1 difficulty in communication and restricted, repetitive
and stereotyped patterns of behaviour categories. Delays or abnormal functioning in at least 1 of the following: social interaction, social language communication, and
symbolic/imaginative play.
4 Atypical Autism: atypical clinical features and a loosened age criterion than used in Autistic Disorder (DSM IV) diagnosis.
5 Asperger Syndrome (DSM IV): core triad of symptoms criteria as in Autistic Disorder. However, no clinically significant delay is observed in areas of language development,
cognitive development, age-appropriate self-skills or adaptive behaviour (other than in social interaction).
6 Pervasive Developmental Disorder- Not Otherwise Specified (DSMIV): presentations that do not meet the criteria for autistic disorder because of late age of onset, atypical
symptomatology, or subthreshold symptomatology, or all of these. This category includes ‘atypical autism’.
7 Pervasive Developmental Disorders or Autism Spectrum Disorder (DSM IV): includes conditions of Autistic Disorder, PDD-NOS, Asperger Disorder, Rett’s Disorder, and
Childhood Disintegrative Disorder.
8 ASD: included individuals with Autistic Disorder diagnosis and those who met a broader definition of ASD based on published criteria for combining information from the ADI-R
and ADOS (see Hallmayer et al., 2011 for more details).
9 The Broad Spectrum: classification not based on any validated algorithms but includes individuals that have just missed diagnostic threshold cut offs for ASD diagnosis and
exhibit high levels autism traits [see Colvert & Tick, et al, 2015 for more details].
113
4.3.2 Statistical Analysis
In the present analysis ASD is treated as a discrete trait and analysed using a liability threshold
model. The assumption is that the risk of ASD follows the standard normal distribution with the
disorder only manifesting itself when a certain threshold is exceeded. The joint distribution of
twin liabilities is assumed to follow the bivariate normal, and the strength between the
liabilities is measured by tetrachoric correlations (based on the relative proportions of
concordant and discordant pairs).
The differences in MZ and DZ correlations provide information on the relative
importance of genetic and environmental variance as specified in a standard biometrical
genetics model (Neale & Cardon, 1992; Plomin, DeFries, Knopik, & Neiderhiser, 2013; Rijsdijk
& Sham, 2002). The resemblance of MZ and DZ twin pairs is specified as reflecting latent
additive genetic factors (A), shared environmental effects (C) and non-shared environmental
effects (E). The covariance of MZ pairs is specified as A+C and that of DZ pairs as .5*A + C (MZ
twins share 100% of segregating genes and DZ twins 50%; and the correlation of 1 for C reflects
growing up in the same family).
We assumed a single-threshold model, with one cut-off on the liabilities corresponding
to the prevalence of Autism Spectrum Disorder including Asperger syndrome, PDD-NOS and
individuals that score highly on autism symptoms but miss the diagnostic criteria cut off
(Colvert & Tick et al., 2015) (see eData in Appendix 4). Analyses were conducted in the program
Mx (Neale, Boker, Xie, & Maes, 2003).
4.3.3 Ascertainment Correction
Different ascertainment of subjects across the primary studies requires a different correction
method (Sullivan et al., 2003). When complete information is available for the sample (i.e. for
Random Population Ascertainment, RPA) the normal probability density function is given by:

  ( x , x )dx dx
1
2
1
2
(eq 1)

114
where Φ is the bivariate normal probability density function of the two liabilities for each twin.
The integral signs -∞ to +∞ indicate that the entire distribution is considered. The Mx frequency
fit function multiplies the count of each response category by their -2 log likelihood to obtain
the overall likelihood of the data.
For non-randomly selected samples, ascertainment corrections adjust the bivariate
normal distribution for the unobserved response categories (therefore the probability of the
observed cells increases), provided that the threshold is known. This is achieved by dividing the
RPA samples likelihood function by the probability density of the remaining cells (1 minus the
probability density of the missing cells Ã). This is equivalent to multiplying the likelihood
function by 1/ the probability density of the remaining cells: 1/1 – Ã, which is accomplished by
including a weight model.
Under Double (Complete) Ascertainment the correction factor used to multiply the
likelihood function (eq 1) by is:
t
1/1 
t
  ( x , x
1
2
)dx1 dx 2 (eq 2)
  
where the integral denotes Ã, part of the distribution reflecting both twins scoring below
threshold for the disorder, i.e. the concordant unaffected pairs.
Under Single Ascertainment the correction factor used to multiply the likelihood
function (eq 1) by is:
t
1/1 
 ( x )dx
1
1
(eq 3)

where the integral denotes Ã, part of the distribution reflecting the first twin (proband) is below
threshold for the disorder (unaffected), i.e. individual which come to the attention of the study
(probands) must be affected.
The correction for Incomplete Ascertainment (mix of ‘singly’ (S) and ‘doubly’ (D)
ascertained concordant pairs) is dependent on π. Then, the ascertainment probability is the
proportion of formally diagnosed probands who were originally identified as ‘at risk’, or
115
2D/2D+S. In practice (following Sullivan et al., 2003), the likelihood of the discordant pairs as
well as the singly ascertained concordant pairs is corrected with the following weight function,
incorporating π:

 / 2  ( x1 )dx1   2
t

  ( x , x
1
t
2
)dx1dx2 (eq 4)
t
The likelihood of the doubly ascertained concordant pairs is corrected by weight:

2   / 2  ( x1 )dx1  
2
t

2
  ( x , x )dx dx
1
2
1
2
(eq 5)
t t
Estimates of π for IA are either provided by the studies or computed by utilising information
available in the publications.
Data files were generated for each of the primary studies in Table 4-1, including the
frequencies of each available response category, π and the threshold z-values corresponding
to the reported prevalences in the individual studies. When not provided, we used a prevalence
of 5% [fixed z-value of 1.65] for ASD (Baird et al., 2006). The eData shows reported and
assumed prevalences and can be found in Appendix 4. Individual study analyses were followed
by meta-analytic analyses, by fitting one overall MZ and DZ correlation or one overall set of A,
C and E parameters to the data, while applying appropriate weight corrections and using fixed
thresholds for each study if needed. An example Mx eScript is available in Appendix 3.
4.4 Results
4.4.1 Tetrachoric correlations
Point estimates for MZ and DZ tetrachoric correlations for individual studies as well as metaanalytic results (corrected for incomplete ascertainment and selection) are shown in Figure 4-1
and presented with 95% Confidence Intervals [95% CI] in Table 4-2.
To deal with uncertainties concerning the definitions used to select the ASD phenotype
and corresponding prevalence to fix the thresholds, we conducted several meta-analyses. The
first is on all selected studies using fixed thresholds based on prevalences as reported in each
study (see eData), apart from study 8 for which the threshold is always estimated: r MZ=0.98
116
(95% CI 0.97-0.99), rDZ=0.62 (95% CI 0.55-0.68). In the second meta-analysis the thresholds of
study 6 and 9 are fixed to a 5% prevalence (z-value 1.65) which is more in line with the Broad
Phenotype definition: rMZ=0.98 (95% CI 0.96-0.99), rDZ=0.52 (95% CI 0.44-0.60). Next, only
studies conducted after 1995, signifying the awareness of the Broad Phenotype definition, were
included. In that meta-analysis we fixed thresholds as reported in each study but estimated the
threshold for study 8: rMZ=0.98 (95% CI 0.96-0.99), rDZ=0.62 (95% CI 0.55-0.68). Finally, again
considering the studies conducted after 1995, we fixed all prevalences to 5%, 3% and 1%,
respectively to test the range of values reported for ASD and the Broad Phenotype in the
literature. In effect, we found that as the prevalence rate decreased from 5% to 3% to 1%, the
DZ correlations increased. See Table 4-2 for all rMZ and rDZ estimates along with 95% CI.
Figure 4-1 Meta-analytic tetrachoric correlations.
Maximum likelihood MZ and DZ tetrachoric correlation coefficients for each of the studies individually
as well as meta-analysis results using 6 different configurations (M1-M6). Meta-analysis 1: using all data
and reported prevalence as fixed thresholds. Meta-analysis 2: as in 1 but changing prevalence of ASD to
5% in Study 6 & Study 9. In Meta-analysis 3-6 only studies after 1995 using the Broad Phenotype
definitions were considered. Meta-analysis 3: using reported prevalence as fixed thresholds, Metaanalysis 4: fixing all thresholds to 5%; Meta-analysis 5: fixing all thresholds to 3% and Meta-analysis 6:
fixing all thresholds to 1%. Note that in all analyses, the threshold of study 8 (Random Population
Ascertained sample) was estimated (z-value around 2.4 corresponding to a 0.08% prevalence).
117
Table 4-2 MZ and DZ twin correlations for individual studies and meta-analytic estimates based on 6
configurations.
Studies
Reference
RMZ
RDZ
Study 3
Steffenburg et al., 1989
.99 (.98/1.00)
-.76 (-.98/.77)
Study 5
Le Couteur et al., 1996
.99 (.99/1.00)
.31 (-.09/.75)
Study 6_1
Taniai et al., 2008, prev 2% ^
approaches 1
approaches .83
Study 6_2
Taniai et al., 2008, prev 5% ^
approaches 1
approaches .78
Study 8
Lichtenstein et al., 2010 *
.82 (.65/.92)
.43 (.20/.61)
Study 9_1
Hallmayer et al., 2011, prev 0.6%
.97 (.94/.99)
.67 (.56/.77)
Study 9_2
Hallmayer et al., 2011, prev 5%
.94 (.89/.98)
.46 (.30/.60)
Study 12
Nordenbaek et al., 2014, prev 5%
.99 (.97/1.00)
.08 (-.39/.53)
Study 13
Colvert, Tick et al., 2015
.99 (.95/1.00)
.60 (.46/.71)
Study
3, 5, 8, 12, 13,
6_1, & 9_1
Study
3, 5, 8, 11, 12,
6_2, & 9_2
Meta-analysis, using reported prevalence as
fixed TH (TH in St8 estimated)
.98 (.97/.99)
.62 (.55/.68)
Meta-analysis, changing prevalence of ASD to
5% in St6 & St9 (TH in St8 estimated)
.98 (.96/.99)
.52 (.44/.60)
Study
5, 8, 12, 13
6_1 & 9_1
Study
5, 6, 8, 9, 12, 13
Meta-analysis, studies after 1995 using broader .98 (.96/.99)
.62 (.55/.68)
phenotype, using reported prevalence as fixed
TH (TH in St8 estimated)
Meta-analysis, studies after 1995 using broader .97 (.96/.99)
.53 (.44/.60)
phenotype, using prevalence of 5% for all (TH
st8 estimated)
Study
Meta-analysis, studies after 1995 using broader .98 (.96/.99)
.58 (.51/.65)
5, 6, 8, 9, 12, 13
phenotype, using prevalence of 3% for all (TH
St8 estimated)
Study
Meta-analysis, studies after 1995 using broader .98 (.97/.99)
.67 (.61/.72)
5, 6, 8, 9, 12, 13
phenotype, using prevalence of 1% for all (TH
St8 estimated)
95% Confidence Intervals in brackets.
* Threshold for Study 8 estimated at around 2.40 z-score, equivalent to prevalence value of 0.08%.
^ Due to non-convergence of the correlation model (MZ correlations approaching 1) the tetrachoric
correlations are derived from the ACE estimates.
4.4.2 The ratio of discordant/concordant MZ and DZ twins
There is a possibility that ascertainment methods of twin samples may lead to biased aetiology
estimates, i.e. oversampling of MZ concordant/undersampling of MZ discordant pairs may lead
to high heritability estimates. Conversely, oversampling of DZ concordant pairs could lead to
high shared-environmental estimates, as reported in the Hallmayer et al (2011) study. To
explore whether any particular study included in the meta-analysis was prone to bias, number
of concordant/discordant MZ and DZ twins were compared. However, from the information in
Figure 4-2 (below) it does not appear that reported heritability and shared-environmental
118
estimates have been greatly influenced by ascertainment methods, comparing across
published studies.
200
180
18
MZ discordant
MZ concordant
DZ discordant
DZ concordant
160
140
120
4
120
11
100
30
80
84
77
60
2
18
40
20
0
0
10
10
1
Steffenburg
8
18
26
2
Le Couteur
18
1
Taniai
32
7
70
1
22
17
24
12
8
3
1
Lichtenstein Hallmayer Nordenbaek Colvert,Tick^ Colvert,Tick*
22
22
Figure 4-2 Ratio of concordant and discordant MZ/DZ twins across studies.
Number of discordant (MZ in blue and DZ in grey) and concordant (MZ in orange and DZ in yellow) twin
pairs on ASD. Colvert,Tick^ proportions are based on twins with ASD diagnosis only, whereas
Colvert,Tick* includes twins with ASD and Broad Spectrum Diagnosis. Unaffected twins are excluded
from the figure but this information can easily be found in Appendix 4.
4.4.3 A, C and E estimates
The A, C and E point estimates (corrected for incomplete ascertainment and selection) can be
found in Table 4-3 (below). The forest plot (Figure 4-2, below), depicts in the first panel the
additive genetic effects (A) and in the second the shared-environmental effects (C) for
individual studies as well as for the meta-analyses. Estimates for additive genetic effects are
generally high with exception of two studies that showed significant proportions of C (studies
6 and 9). However, when the threshold for study 9 was changed to a prevalence rate of 5%, the
C estimate dropped to zero.
The first meta-analysis on all selected studies, using fixed thresholds as reported in
each study, yielded a heritability of 74% (95% CI 0.70-0.87), with a significant proportion of
shared environmental effects: 25% (95% CI 0.12-0.37). In the second meta-analysis, when the
119
thresholds of study 6 and 9 are fixed to 5% prevalence, the heritability increases to 93% (95%
CI 0.77-0.99) and C becomes non-significant. A detailed investigation of studies conducted after
1995, in which the Broad Phenotype definition for ASD was included, applying thresholds as
reported in each study gave estimates similar to when all studies were considered. However,
when we subsequently fix all thresholds to 5%, 3% or 1%, we see an increase in the proportion
of C (consistent with the observed increase in DZ correlations relative to the MZ correlations,
Figure 4-1). This is a significant finding that stresses the importance of the assumed prevalences
of the disorder in the population when using model-fitting analysis on ascertained samples.
Note that in all analyses the threshold for study 8 (RPA sample) was estimated.
4.5 Discussion
4.5.1 Main Findings and Conclusion
Using a quantitative meta-analytic approach, we estimated the heritability of ASD using
studies of twins with a diagnosis of autism spectrum disorder. Applying appropriate
ascertainment corrections and maximum-likelihood estimation, our study produced
tetrachoric twin correlations and heritability estimates for all studies published to date,
inclusive of studies that previously only reported proband-wise concordant rates. The metaanalytic heritability estimates ranged between 64% and 91% (despite diagnostic
heterogeneity) and are in line with previous reports (Bailey et al., 1995; Colvert & Tick et al.,
2015; Folstein & Rutter, 1977; Le Couteur et al., 1996; Lichtenstein et al., 2010; Nordenbæk et
al., 2014; Ritvo, Freeman, Mason-Brothers, Mo, & Ritvo, 1985; Rosenberg et al., 2009;
Steffenburg et al., 1989; Taniai, Nishiyama, Miyachi, Imaeda, & Sumi, 2008).
The most important statistical finding concerns the assumptions we make about the
underlying distribution of the phenotype as a discrete trait, which is a standard normal
distribution with a threshold discriminating between affected and non-affected individuals. A
statistical correction necessary for selected (diagnostic) samples is fixing the threshold in the
model to the prevalence rate of the disorder in the general population. The different diagnostic
120
outcomes across ASD studies prove problematic to derive the correct overall prevalence in
meta-analysis.
When looking at the individual studies as well as the meta-analysis results, it is apparent
that detecting significant heritable variance is quite robust, but detecting significant C effects
depends on the assumed prevalence of the disorder. Effectively, pushing the threshold to
higher values (i.e. decreasing the assumed prevalence rate) will not affect the MZ twin
correlations as much since they are already at the top of their statistical bound (upper 95% CI
values approaching 1). However, the DZ correlations will increase relative to the MZ
correlations, consistent with increasing the effect of shared environment. Given the
importance of this effect, fitting multiple-threshold models including a Broader Phenotype as
a meaningful sub-category with a lower (fixed) threshold on the spectrum to ASD (Colvert &
Tick et al., 2015; Sasson, Lam, Parlier, Daniels, & Piven, 2013) might perhaps be a better
method, albeit there is no generally agreed measure of such category.
A second important point to note from the meta-analyses is that even when sharedenvironmental effects become significant, they never explain the majority of the variance in
ASD (as claimed by Hallmayer et al., 2011 and Frazier et al., 2014). We therefore conclude that
significance of shared environments (C) in ASD is likely to be a statistical artefact as a result of
the assumptions made of the prevalence of autism in the population. The meta-analysis results
are in line with the results from the largest extended family population study (Sandin et al.,
2014), (showing no effects of shared environment) as well as results from the only random
population twin study, which did not need to rely on fixed threshold correction (Lichtenstein et
al, 2010). Nevertheless, we note here the limitation of random population studies, which use
proband selection based on electronic records within registries rather than using cases with
individually confirmed diagnosis of ASD.
In conclusion, using an appropriate meta-analytic statistical approach we
demonstrated that the aetiology of ASD in a combined sample is more consistent with strong
genetic influences. Secondly, we can reject the claim that there is a strong shared
121
environmental effect on autism spectrum disorders accounting for the majority of variance and
alert to the danger of placing too much weight on findings from a single study, such as
Hallmayer et al (2011). At the same time, we do not exclude the possibility that environmental,
or at least non-genetic, effects influence ASD. But unless a suitably powered and well-designed
new study comes forward, this claim should be put to one side for now.
Table 4-3 Estimates of the genetic and environmental variance components for individual studies and
meta-analytic estimates based on 6 configurations.
Studies
Reference
A
C
E
Study 3
Steffenburg et al., 1989
.99 (.54/.99)
.00 (.00/.46)
.01 (.00/.05)
Study 5
Le Couteur et al., 1996
.99 (.49/.99)
.00 (.00/.51)
.01 (.00/.01)
Study 6_1
Taniai et al., 2008, prev 2%
.33 (.12/.71)
.67 (.29/.88)
.00 (.00/.01)
Study 6_2
Taniai et al., 2008, prev 5%
.44 (.17/.92)
.56 (.08/.83)
.00 (.00/.01)
Study 8
Lichtenstein et al., 2010 *
.79 (.29/.92)
.03 (.00/.44)
.18 (.08/.36)
Study 9_1
Hallmayer et al., 2011, prev 0.6%
.59 (.40/.82)
.38 (.15/.56)
.03 (.01/.06)
Study 9_2
Hallmayer et al., 2011, prev 5%
.95 (.68/.98)
.00 (.00/.26)
.05 (.02/.12)
Study 12
Nordenbaek et al., 2014, prev 5%
.99 (.69/.1.00)
.00 (.00/.31)
.00 (.00/.00)
Study 13
Colvert, Tick et al., 2015
.79 (.56/.98)
.19 (.00/.43)
.01 (.00/.05)
Study
3, 5, 8, 12, 13,
6_1, & 9_1
Study
3, 5, 8, 12, 13,
6_2, & 9_2
Meta-analysis, using reported
prevalence as fixed TH (TH in St8
estimated)
Meta-analysis, changing
prevalence of ASD to 5% in St6 &
St9 (TH in St8 estimated)
.74 (.70/.87)
.25 (.12/.37)
.01 (.01/.03)
.93 (.77/.99)
.06 (.00/.21)
.01 (.01/.03)
Study
5, 8, 12, 13
6_1 & 9_1
Meta-analysis, studies after 1995
.72 (.60/.86)
.26 (.13/.39)
.02 (.01/.03)
using broader phenotype, using
reported prevalence as fixed TH
(TH in St8 estimated)
Study
Meta-analysis, studies after 1995
.91 (.76/.99)
.07 (.00/.22)
.02 (.01/.04)
5, 6, 8, 9, 12, 13 using broader phenotype, using
prevalence of 5% for all (TH st8
estimated)
Study
Meta-analysis, studies after 1995
.81 (.67/.96)
.18 (.03/.32)
.02 (.01/.03)
5, 6, 8, 9, 12, 13 using broader phenotype, using
prevalence of 3% for all (TH St8
estimated)
Study
Meta-analysis, studies after 1995
.64 (.53/.77)
.35 (.22/.46)
.01 (.01/.02)
5, 6, 8, 9, 12, 13 using broader phenotype, using
prevalence of 1% for all (TH St8
estimated)
* Threshold for Study 8 estimated at around 2.40 z-score, equivalent to prevalence value of 0.08%.
122
Figure 4-3 Forest plots.
Upper panel = additive genetic effects (A), lower panel = shared environmental effects (C), calculated for
each study individually as well meta-analysis estimates, using 6 different configurations. Horizontal lines
represent the 95% confidence intervals. Meta-analysis 1: using all data and reported prevalence as fixed
thresholds. Meta-analysis 2: as in 1 but changing prevalence of ASD to 5% in Study 6 & Study 9. In Metaanalysis 3-6 only studies conducted after 1995 using broader phenotype definitions were considered.
Meta-analysis 3: using reported prevalence as fixed thresholds, Meta-analysis 4: fixing all thresholds to
5%; Meta-analysis 5: fixing all thresholds to 3% and Meta-analysis 6: fixing all thresholds to 1%. Note that
in all analyses, the threshold of study 8 (Random Population Ascertained sample) was estimated (z-value
around 2.4 corresponding to a 0.08% prevalence).
123
Chapter 5 Autism Spectrum Disorders and other mental health
problems: exploring etiological overlaps and phenotypic causal
associations 2
5.1 Abstract
Recent studies highlight the impact of co-existing mental health problems in ASD. No twin
studies of ASD co-morbidities to date have reported on individuals meeting diagnostic criteria
of ASD. This twin study is therefore the first to report on the aetiological overlap between
clinically diagnosed ASD and Emotional symptoms, Hyperactivity and Conduct problems.
Genetic and environmental influences on the covariance between ASD and co-existing
problems were estimated, in line with the ‘correlated risks model’ prediction. Phenotypic
causality models were also fitted to explore alternative explanations of comorbidity; that coexisting problems are the result of or result in ASD symptoms, that they increase recognition
of ASD, or that they arise due to an over-observation bias/confusion when differentiating
between disorders.
Fifty percent of twins with Broad Spectrum/ASD experienced increased levels of
Emotional symptoms or Hyperactivity when compared to unaffected twins; a quarter of cases
met these criteria for the three reported problems concurrently. The phenotypic correlation
between ASD and Emotional symptoms was entirely explained by genetic influences and
accompanied by a moderate genetic correlation (.42). The opposite was true for the overlap
with Conduct problems as non-shared environmental factors had the strongest impact. For
Hyperactivity, the best fitting model suggested a unidirectional phenotypic influence of
Hyperactivity on ASD.
2
Chapter adapted from Tick, B., Colvert, E., McEwen, F., Stewart, C., …..& Rijsdijk, F. (2015) Autism
Spectrum Disorders and other mental health problems: exploring etiological overlaps and phenotypic
causal associations. Journal of the American Academy of Child and Adolescent Psychiatry (in press)
124
Our findings suggest a possible effect of Hyperactivity on identification of ASD. The
lack of genetic influences on Conduct problems-ASD overlap further supports the genetic
independence of these two phenotypes. Finally, the co-occurrence of Emotional symptoms in
ASD, compared to other co-occurring problems, is completely explained by common genetic
effects.
5.2 Introduction
ASD is a highly heritable neurodevelopmental disorder characterised by impaired social
communication and restricted and repetitive behaviours and interests (Colvert & Tick et al.,
2015). In 2010 52 million individuals worldwide were estimated to have ASD (Baxter et al.,
2014). Research over the last decade highlighted the high rates and severe impact of additional
(comorbid) mental health problems in ASD (Leyfer et al., 2006; Lundström et al., 2014; Matson
& Kozlowski, 2011; Simonoff et al., 2013). This is acknowledged in the most recent version of
the APA’s Diagnostic and Statistical Manual of Mental Disorders (DSM 5; APA, 2013), where
additional diagnoses such as ADHD or Anxiety disorders are allowed alongside ASD for the first
time (American Psychiatric Association, 2013; Grzadzinski, Huerta, & Lord, 2013).
In genetic epidemiology, there are two common reasons for comorbidity: 1) one
disorder has a causal role for another disorder (e.g. chronic kidney disease as an increasing risk
for coronary heart disease (Aronow, Ahn, Mercando, & Epstein, 2000)); 2) the two disorders cooccur because of overlapping genetic, or environmental, reasons (e.g. major depression and
alcoholism may co-occur in adulthood due to childhood maltreatment (Holmes & Robins,
1987)). Twin studies are particularly informative in decomposing the genetic and (shared/nonshared) environmental influences on comorbidity with potentially high scientific and practical
importance for diagnosis and treatment (Neale & Kendler, 1995; Rutter, 1994).
Commonly reported mental health comorbidity in clinical ASD samples include
internalising problems such as excessive fears and worry (anxiety) and depression (Pugliese,
White, White, & Ollendick, 2013; Simonoff et al., 2013; Vasa et al., 2013). Internalising problems
have been postulated as a cause of ASD symptoms (e.g., “rigid preferences for sameness”
125
resulting from extreme anxiety), or an effect (e.g., communication difficulties lead to isolation,
sadness and anxiety) (Kanner, 1968).
General population twin studies have shown a phenotypic correlation between autisticlike traits and internalising problems of .30-.35, which was stable across a 5-year period
(Lundström et al., 2011; Scherff et al., 2014). This covariance was explained modestly by shared
genetic factors and mostly by shared and non-shared environmental factors. Longitudinal
direction of causation investigations from 7/8 to 12 years showed that autistic-like traits had a
modest phenotypic influence on internalising problems over time, while the reverse effect was
of smaller magnitude (Hallett, Ronald, Rijsdijk, & Happé, 2010). Closer examination of
internalising subdomains showed that the greatest genetic overlap occurred between
generalised anxiety and negative affect and repetitive behaviours and communication
problems (Hallett, Ronald, Rijsdijk, & Happé, 2012). Conversely, stronger genetic correlation
(.53) on autism and anxiety traits was reported in a large Swedish cohort (Lundström et al.,
2011). The moderate phenotypic overlap between the traits is somewhat surprising considering
the high rates of internalising problems in clinical ASD samples (Leyfer et al., 2006; Simonoff
et al., 2013). Replication of these results in twins with a diagnosis of ASD is much needed
(Scherff et al., 2014).
ADHD-like traits are also commonly reported in ASD. Twin findings show a consistent
pattern of moderate to strong genetic overlap between ADHD and autism traits (Larson et al.,
2010; Rommelse, Franke, Geurts, Hartman, & Buitelaar, 2010; Ronald, Simonoff, Kuntsi,
Asherson, & Plomin, 2008). Longitudinal direction of causation cross-lagged models showed
that ADHD traits at age 8 were strongly predictive of ASD traits at age 12 (particularly
communication difficulties) with a significantly less strong converse effect (Taylor et al., 2013).
An examination of subdomains of ADHD revealed that hyperactivity correlated less strongly
than inattentiveness and impulsivity with autism subdomains (Ronald, Larsson, Anckarsäter, &
Lichtenstein, 2014). Possible pleiotropic genetic effects (i.e. same genes predisposing to ASD
and ADHD) were tested using common SNPs (Single Nucleotide Polymorphisms) in genome126
wide analysis to estimate SNP co-heritabilities (Cross-Disorder Group of the Psychiatric
Genomics et al., 2013). The genetic correlation between the SNPs for ASD and ADHD was nonsignificant, but further research is needed with much larger samples.
Comorbidity of ASD with conduct-like problems is not easily interpretable and might
be explained by confusion over the phenotype and its appropriate definition. ASD-associated
disruptive behaviours may resemble conduct problems. However, ‘meltdowns’ reflect the
intrinsic developmental difficulties of ASD; whereas conduct problems in non-ASD individuals
reflect different psychological processes (Moffitt, 2011). Very few studies have explored
conduct-like problems in ASD to date. Those that exist, report an aetiological independence
and show small to moderate genetic overlap (larger in boys than girls), with most covariance
explained by shared and non-shared environmental factors (O’Nions & Tick et al., 2015, in press,
Jones et al., 2009; Kerekes et al., 2014; Lundström et al., 2011).
Previous studies demonstrate modest to strong genetic influences on autism traits and
comorbid internalising and ADHD-related problems, and an aetiological independence for
conduct-like problems. Here, for the first time to our knowledge, we examine in this study the
aetiology of these three comorbidities in population-based ASD twin sample diagnosed with
gold-standard instruments. First, we examined whether comorbidities are due to correlated
genetic and environmental risks. Secondly, we tested phenotypic causal effects, with the
direction of causation (ASD – internalising/externalising) to be determined. Each of these
hypotheses has implications for potential intervention strategies (Simonoff, 2000).
5.3 Methods
5.3.1 Participants
Twins ‘at risk’ of ASD were identified from the Twins Early Development Study (TEDS)
(Haworth et al., 2012). Best-estimate Diagnosis (BeD) was based on in-person assessments
using gold-standard instruments, parent interview and observational information (for full
details see Colvert & Tick et al., 2015). In addition to families with one or both twins meeting
127
criteria for ASD, the SRS recruited control twins who were low in ASD traits (scoring <12 on the
Childhood Autism Spectrum Test [CAST] (Williams et al., 2005)). Overall, BeD data were
available for 207 twin pairs (MZ=56, DZ Same-Sex=77 and DZ Opposite-Sex=74) with a mean
age of 13.16 years. Information on other mental health problems was collected from parents
using the Strengths and Difficulties Questionnaire (SDQ) (Goodman, 1997). SDQ scores were
available for 166 (80%) of the 207 twin pairs (MZ=45, DZ Same-Sex=57 and DZ OppositeSex=64).
5.3.2 Measures
5.3.2.1 Best-estimate Diagnosis (BeD) for Autism Spectrum Disorder
The full procedure related to establishing Best-estimate Diagnosis (forthwith referred to as
‘ASD-BeD’) can be found in Colvert & Tick et al (2015). The SRS participants were evaluated for
ASD with a range of diagnostic measures: DAWBA, ADOS, and ADI-R. ASD-BeD was made
according to DSM IV and ICD 10 criteria and based primarily on scores from the ADI-R and
ADOS. The additional sub-category of Broad Spectrum was included in the classification to
capture individuals with high-level autism traits that fell just short of an ASD diagnosis. ASDBeD, therefore, consists of 3 ordinal classes: 0=no ASD/controls, 1=Broad Spectrum, 2=ASD.
5.3.2.2 Strengths and Difficulties Questionnaire (SDQ)
The SDQ (Goodman, 1997) is a well-regarded measure of mental health for 2-17 year-olds and
widely used in clinical settings. It contains five subscales (5 items each) adapted to measure
Emotional symptoms, Conduct problems, Hyperactivity, Peer relationship problems and Prosocial behaviours, with each item given a three-point rating: 0=Not True, 1=Somewhat True,
2=Certainly True (maximum score=10).
Information on Emotional symptoms, Hyperactivity and Conduct problems were used
as continuous measure in the genetic model fitting analyses. We disregarded the Peer relations
problems as they are already captured by an ASD diagnosis and the Prosocial scale as it
measures positive and not negative co-morbidities (the main focus of this study).
128
For descriptive statistics and phenotypic analyses the lower (Borderline) threshold is
used to indicate elevated rates of each SDQ problem; relatively few participants passed the
Borderline but not the Abnormal thresholds, but inspection of the data suggested these
participants were more properly grouped with Abnormal than with unaffected cases. Cut-offs
for Borderline and Abnormal levels (out of 10 points), respectively, are: Emotional symptoms 4,
5; Hyperactivity 6, 7; and Conduct problems, 3, 4. SDQ revealed good internal consistency levels
in the SRS sample: Emotional symptoms Cronbach’s α=.72, Hyperactivity α=.81 and Conduct
problems α=.67.
5.3.3 Statistical Analyses
5.3.3.1 Twin Correlations
The principles of the twin model fitting are provided elsewhere (Rijsdijk & Sham, 2002).
Structural Equation Modeling (SEM) was performed in OpenMx (Boker et al., 2011) using full
information maximum likelihood estimation for the (genetic and environmental) variance and
covariance decomposition. Continuous SDQ scores and ordinal ASD-BeD were analysed jointly.
For ASD, a liability threshold model was assumed with a standard normal distribution
underlying the ordered categories with individuals receiving the diagnosis as they cross the
disease threshold on this liability distribution (Falconer, 1965).
The joint multivariate normal distribution assumed between each measure of SDQ and
ASD-BeD in a pair of twins allows estimation of the within- and across-twin (MZ/DZ)
correlations. The ratio of MZ/DZ correlations indicate whether genetic or sharedenvironmental influences are responsible for variation within traits and on the covariance
between them, which are formally estimated in the bivariate genetic twin model (explained
below). Due to the selected nature of the sample, the thresholds on the ASD-BeD liability were
fixed to population ‘known’ values of ASD prevalence: 1st threshold of 5% (Baird et al., 2006)
separated the unaffected and twins with Broad Spectrum; 2nd threshold of 1% (Brugha et al.,
2011) separated the Broad Spectrum and ASD twins.
129
5.3.3.2 The Bivariate Genetic Model
The (co)variance of each SDQ subscale and ASD-BeD was partitioned into additive genetic (A),
shared-environmental (C) and non-shared environmental (inclusive of measurement error) E
effects (Rijsdijk & Sham, 2002). In the absence of a specific order of traits in the model, the
standardized correlated-factor solution is interpreted such that the path from the A1 factor to
the SDQ subscale and the A2 factor to ASD-BeD are the square roots of their respective
standardized path estimates (heritabilities) and the correlation path between A1 (SDQ) and A2
(ASD-BeD) is the genetic correlation between the variables (rA) (Rijsdijk et al., 2005). The same
principle is applied to non-shared environmental effects (E). We then calculated the proportion
of the phenotypic correlation (rPH) due to correlated additive genetic effects A (rPH_A = √h12 * rA
* √h22) and due to correlated non-shared environmental effects E (rPH_E= √e12 * rE * √e22)
expressed as proportions of rPH (Plomin et al., 2013). Note, as shared environmental factors did
not influence ASD-BeD (Colvert & Tick et al., 2015), they cannot explain the covariance with
comorbid traits, hence excluded from above calculations.
5.3.3.3 The Direction of Causation (DoC) Models
The specific genetic and environmental effects on each of the SDQ measures and ASD-BeD are
modelled as in the full bivariate genetic model, but with the genetic and environmental
correlation between the traits substituted by causal effects of one trait on the other (Heath et
al., 1993; Rijsdijk & Sham, 2002). In Figure 5-3, a reciprocal model is presented with r and r’
reflecting both alternative causal pathways.
Within twin data, the differentially predicted cross-twin cross-trait covariance is the
basis of the DoC model, and will only work if the relative proportions of variance (i.e. the h2) are
sufficiently different across traits. For example, if SDQ problems are caused by ASD-BeD
(SDQ ←ASD-BeD), then the expected cross-trait cross-twin covariance will be dominated by
additive genetic effects. Thus because of the high heritability for ASD-BeD, a higher MZ
compared to DZ covariance will be observed compared to if the relationship was the reverse
(SDQ →ASD-BeD). To establish the direction of causation, the reciprocal (bidirectional) model
130
is fitted first and the unidirectional models (r and r’) are the nested sub-models compared by
using likelihood ratio and fit statistics (AIC and BIC) (Burnham, Anderson, & Huyvaert, 2011;
Raftery, 1995). Theoretically, if the variables only have two sources of variance (A and E) we can
only proceed with causal hypothesis testing if we assume that both traits are measured without
error or that the measurement errors are equal in magnitude (Heath et al., 1993).
5.3.4 Results
5.3.4.1 Descriptive Statistics
Figure 5-1 shows the mean (and Standard Errors) for each SDQ subscale, divided into males and
females meeting criteria for ASD, Broad Spectrum and unaffected (comprising control twins
and the unaffected co-twins of probands).
Figure 5-1 Means and Standard Error of the Mean bars.
Means for each SDQ subscale in each diagnostic ASD-BeD group (n=332 of twin individuals).
ASD=Autism Spectrum Diagnosis, BS=Broad Spectrum, UN=Unaffected. The range of scores for SDQ
was 0 to 10 points; Borderline/Abnormal category cut off for Hyperactivity was 6 points, for Emotional
symptoms 4 and for Conduct problems 3 points.
5.3.4.2 Comorbid SDQ problems
In Figure 5-2, below, we define ‘proband’ as any individual assigned an ASD or Broad Spectrum
Best-estimate Diagnosis. Of 135 individual probands, 71 (52%) passed the cut off for Emotional
symptoms, 72 (53%) for Hyperactivity and 45 (33%) for Conduct problems. Twenty-four (18%)
probands met the cut off for all three mental health problems. Thirty three probands (24%) had
no comorbid mental health problems. Of 40 ASD-BeD concordant proband pairs (MZ n=18, DZ
131
n=22), 45% were also concordant for Emotional symptoms (n=18); 40% for Hyperactivity
(n=16); 18% for Conduct problems (n=7), and 22% were concordant for any two or more
problems (n=9). Table 5-3 (in Supplementary Materials section 5.5) shows that MZ twins
concordant on ASD-BeD display a higher concordance rate on SDQ problems, in comparison to
DZ twins.
5.3.4.3 Comorbid SDQ problems among discordant twins
Six of the 55 ASD-BeD discordant pairs (11%) were concordant on Emotional symptoms, four
(7%) on Hyperactivity, and 4 (7%) on Conduct problems. Three pairs (5%) were concordant for
any two or more problems. Among ASD-BeD discordant pairs, 3 were MZ and 52 were DZ. One
MZ proband and 2 co-twins displayed no SDQ problems by parent report. In the DZ group, both
individuals in 12 pairs, and co-twin only in 23 pairs were SDQ problems free; 23% (n=12) of pairs
were concordant on at least one SDQ problem.
5.3.4.4 Twin Correlations
Table 5-1 (below) shows polychoric and polyserial correlation estimates from the three bivariate
models. The cross-twin (MZ/DZ) within-trait correlations (column 1 and 2) for Emotional
symptoms (.70/.19) and Hyperactivity (.66/.08) indicate dominance rather than additive genetic
effects. However, our small sample is underpowered to detect dominance effects, thus we
report broad sense heritability by fitting additive genetic effects only. Estimates for Conduct
problems were .63/.25 and for ASD-BeD .91/.46, consistent with an AE model. The highest
phenotypic overlap (rPH) was between Emotional symptoms and ASD-BeD (.33), followed by
Hyperactivity (.28) and Conduct problems (.13) (all significant at 95% Confidence Intervals [95%
CI]).
132
MZ
E
m
o Co-twin
E
m
Probands o
C
o
n
C
o
n
H
y
p
E
m
o Co-twin
C
o
n
H
y
p
E
m
o Co-twin
Disconcord Concordant
Disconcordant
Concordant
1 ♂
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31 ♀
32
33
34
35
E
m
Probands o
1 ♂
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23 ♀
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
H
y
p
C
o
n
Concordant
DZ OS
Disconcordant
Figure 5-2 Associated SDQ problems in the
SRS sample.
MZ=Monozygotic, DZ SS=Dizygotic Same
Sex, DZ OS=Dizygotic Opposite Sex twin pairs.
Concordant = ASD/Broad Spectrum in both
twins; Discordant = twin 1 is ASD/Broad
Spectrum, twin 2 is unaffected. Headings
‘Proband’ and ‘Co-twin’ represent an
individual. When squares to the right of
Proband and to the left of Co-twin are
coloured, it means this individual met the cut
off for Borderline OR Abnormal category for
Emotional symptoms [Emo] (=>4 points, green
fill), Hyperactivity [Hyp] (=>6, yellow fill) and
Conduct problems [Con] (=>3, blue fill). Scores
for concordant unaffected are not included in
this figure, but can be found in Table 5-3.
H
y
p
Concordant
H
y
p
Concordant
Concordant
C
o
n
Disconcordant
20 ♀
21
C
o
n
Disconcordant
1 ♂
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
H
y
p
Concordant
E
m
Probands o
DZ SS
133
Table 5-1 Bivariate MZ and DZ within-trait and cross-trait twin correlations of three SDQ problems and
ASD-BeD.
cross-twin
cross-twin
Variable
rMZ 1
rDZ 1
cross-trait (MZ) cross-trait (DZ)
ASD-Best-estimate Diagnosis
Emotional symptoms
3
.91 (.84-.95)
.70 (.53-.80)
.46 (.36-.55)
.19 (.02-.27)
2
2
n/a
.26 (.15-.36)
n/a
.17 (.07-.27)
Hyperactivity
.66 (.46-.77) .08 (.00-.24) .20 (.09-.29)
.06 (.00-.15)
Conduct problems
.63 (.45-.75) .25 (.07-.41)
.06 (.00-.18)
.01 (.00-.11)
1
Maximum likelihood within-trait twin correlations (rMZ and rDZ) estimated in a model with two
thresholds on the liability to ASD fixed to population values of Broad Spectrum Diagnosis (5%) and ASD
(1%) prevalence.
2
Maximum likelihood cross-twin cross-ASD-BeD correlations, obtained for each SDQ measure and
ASD-BeD separately.
3
For ASD-BeD, 3 sets of MZ and DZ correlations are available as 3 bivariate analyses were performed:
here only one is provided (the other two were of values identical to one decimal place and with
overlapping 95%CI).
Significant estimates (i.e. 95%CI not spanning zero) are given in bold.
5.3.4.5 Bivariate Genetic Model Results
Table 5-4 (supplementary materials) shows the model-fit results and fit statistics (2, AIC, BIC)
for the three bivariate models in separate sections. The goodness-of-fit test of two competing
models was compared using the chi-square statistic (2) and the difference in degrees of
freedom (DF) (Rijsdijk & Sham, 2002). A non-significant 2 value (p>0.05) means the tested
model is consistent with the data. A significant 2 value (p<0.05) means the model poorly fits
the data and is rejected. For 2, an increase of 3.84 or more for 1DF indicates a significantly
worse fit. In each section the first 2(DF) concerns the difference in -2LL values of model 3 and
2 (testing the significance of parameter C), and the second of model 3a and 3 (testing the
significance of either parameter a21 or e21).
Models 2 to 3a were consistent with the twin correlations and showed that shared
environmental factors did not influence the (co)variance of SDQ and ASD-BeD (model 3). The
AE models could be further reduced for the Emotional symptoms–ASD-BeD analysis (AE model
3a with path e21 removed) and Conduct problems (AE model 3a with a21 removed). The
estimates of genetic influences (Figure 3) ranged from 50% for Hyperactivity, 60% for Conduct
problems, 62% for Emotional symptoms, to 91% for ASD-BeD across the three models. The
strongest phenotypic overlap (rPH=.32, column 2, Table 5-2) between Emotional symptoms and
134
ASD-BeD was entirely accounted for by genetic factors (column 3, Table 5-2), as well as
revealing a strong genetic correlation (rA=.42, Figure 5-3).
The Hyperactivity-ASD-BeD overlap was of similar size (rPH=.29) and strongly
influenced by genetics (62%) and by non-shared environmental influences (38%; column 4,
Table 5-2). However, the non-shared environmental factors showed a much stronger
correlation (rE=.51) than the genetic factors (rA=.27). A small but significant phenotypic
association (rPH=.10) between Conduct problems and ASD-BeD was entirely explained by
strongly associated non-shared environmental influences (rE=.50). Overall, looking at the most
parsimonious models, the results provide considerable support for a genetic basis for the
comorbidity of Emotional symptoms/Hyperactivity, and ASD-BeD.
Figure 5-3 Standardized estimates of the three AE bivariate models.
Abbreviations: Emotional symptoms [Emo]; Hyperactivity [Hyp]; Conduct problems [Con].
Abbreviations continued: Prevalences for the ASD liability threshold were fixed at 5% (Broad Spectrum
Disorder) and 1% (ASD).
A1/2 & E1/2 denote latent genetic and environmental factors on each trait individually;
rA & rE denote the genetic and environmental correlations between the A and E factors influencing ASDBeD and each SDQ measure.
Note: The genetic correlation was fixed at zero for ASD-BeD and Conduct problems, as this path was
non-significant. Conversely, the environmental correlation was fixed at zero for ASD-BeD and Emotional
symptoms, as this path was non-significant (dashed lines).
r & r’ – phenotypic causal paths calculated in the Direction of Causation Models instead of the r A & rE
paths.
135
Table 5-2 Phenotypic overlap due to genetic and environmental effects.
Variable
RPH2
RPH_A3
RPH_E3
Emotional symptoms 1
Hyperactivity 1
Conduct problems 1
.32 (.23-.40)
.29 (.20-.37)
.10 (.04-.16)
.32 (.23-.40) [100%]
.18 (.06-.30) [62%]
-
.11 (.03-.20) [38%]
.10 (.04-.16) [100%]
1
Thresholds for the ASD liability were set at 5% (Broad Spectrum Disorder) and 1% (ASD).
rPH - phenotypic correlation between ASD-BeD and each SDQ measure (95%CI);
3
rPH_A, rPH_E - extent to which the phenotypic correlation (rPH) is due to correlated genetic and nonshared environmental influences (95%CI). Values in square brackets are percentages of rPH.
2
5.4 Discussion
The aetiology of three co-occurring mental health problems was examined in a sample of twins
assessed for ASD. Overall, half of probands met the criteria for Borderline/Abnormal levels for
Emotional symptoms or Hyperactivity, and a quarter met the cut-off for all three problems.
Only one other twin study to date has reported on multiple comorbid disorders in ASD using
health record data (Lundström et al., 2014). In a sample of 272 twins 50% had four or more
coexisting disorders (although these were not identical to those considered here); on average
65% of non-ASD co-twins of probands had at least one other disorder, suggesting genetic
origins.
We found lower Hyperactivity but similar Conduct problems levels when compared to
the Oppositional Defiant Disorder levels reported in Lundstrom et al’s sample, clustering all
three under Externalising problems. Internalising problems were not measured in the Swedish
study. It is worth noting that the current study sample was identified via several screening
stages, followed by face-to-face assessment, whereas Lundstrom et al used parent telephone
interview data/national registry information as a ‘proxy’ clinical diagnosis; a limitation noted by
the authors. This could result in under- or overestimation of probands with coexisting
difficulties.
In comparison to non-twin samples, our prevalence rates for co-existing disorders in
ASD exceeded those of Leyfer et al., (2006), but were comparable to Simonoff et al., (2008).
This could be due to use of different measures. However, recent reports suggest the same
136
psychometric properties as the SDQ’s performance in ASD adolescents and adult samples was
comparable to non-ASD samples. The SDQ shows good external validity when compared to
several clinically utilized measures of anxiety, depression, Obsessive Compulsive Disorder, and
ADHD (Findon et al, 2015, under review).
5.4.1 Aetiology of ASD and SDQ problems
Before moving onto the discussion of aetiological results, it is necessary to mention that results
of the current study are often compared to results derived from the Twins Early Development
Study (TEDS), from which the SRS sample was derived. This will be highlighted, where
necessary.
5.4.1.1 Emotional symptoms
The heritability of Emotional symptoms (62%) was comparable to estimates in typically
developing TEDS sample (50-60%) (Gregory & Eley, 2007). The phenotypic Emotional
symptoms- ASD-BeD overlap (.32) was comparable to previous studies and entirely explained
by genetic influences. We found no influence of shared environment on this overlap in contrast
to previous TEDS study (Scherff et al., 2014), but this might reflect limited power. The genetic
correlation was comparable to typically developing TEDS & Swedish samples of similar ages
(Hallett et al., 2012; Lundström et al., 2011; Scherff et al., 2014). Our unique data provide
evidence that internalising problems in ASD populations are partly due to genetic influences.
5.4.1.2 Hyperactivity
The heritability of Hyperactivity (50%) was closer to the lower range of those reported on
typically developing TEDS sample (50-80%) (Ronald et al., 2008). Genetic influences explained
62% of the phenotypic correlation (rPH=.29) between Hyperactivity and ASD-BeD, with the
remainder explained by non-shared environmental influences. The moderate genetic
correlation (.27) is comparable to that reported in typically-developing adolescent TEDS
samples (.12-.33) for social difficulties and repetitive behaviours and ADHD traits (Taylor,
Charman, & Ronald, 2015), but stands in contrast to data from middle childhood TEDS &
137
Swedish studies (~.50) (Larson et al., 2010; Ronald et al., 2008). It has been demonstrated by a
TEDS study that the communication difficulties aspects of the ASD triad, but not others,
correlate more strongly with ADHD-like traits (Taylor et al., 2015). Future analysis should take
this into account.
5.4.1.3 Conduct problems
The heritability of Conduct problems was comparable to those reported in the general
literature (overlapping CI’s) (Salomone et al., 2014). The small phenotypic overlap (rPH=.10)
between Conduct problems and ASD-BeD was explained entirely by non-shared environmental
effects, which is in agreement with O’Nions & Tick et al (2015) that showed almost complete
genetic independence of autism traits from callous-unemotional traits in the general twin
population (TEDS). This, however, stands in contrast to non-TEDS studies of larger samples
reporting genetic overlap of .14 in boys for Conduct Problems and ASD, and .35 for Oppositional
Defiant Disorder and ASD – measures derived from parent telephone interviews (Kerekes et al.,
2014). These differences could be explained as informant-dependent effects. A recent report
suggests that the severity of conduct problems varies as a function of language ability as ASD
children without phrase speech exhibited the highest levels of such behaviours (Scourfield, Van
den Bree, Martin, & McGuffin, 2004). The low levels of Conduct problems and little overlap with
ASD in the current sample could be explained by the high proportion of diagnosed twins with
verbal fluency (90%).
5.4.2 Direction of Causation
In twin studies, covariation predominantly due to E (indicated by a significant within person
cross-trait correlation but a non-significant cross-trait cross-twin correlation) could indicate a
phenotypic causal relationship (Heath et al., 1993). From our data, following that logic, the
relationship most likely to be phenotypically causal would be that between Conduct problems
and ASD-BeD (all E), and the one least likely to be causal would be that between Emotional
problems and ASD-BeD (all A).
138
However, the most convincing statistical evidence for direction of causation is seen for
the Hyperactivity-ASD-BeD relationship, where the unidirectional model with causal path Hyp
→ ASD-BeD showed the best overall fit across all models (including the correlated risk models)
based on lowest BIC fit index and, importantly, significance of this causal path indicated by chisquare test. Similar direction of causation effects were found in the TEDS general population
twin study of ADHD traits at age 8 predicting autistic traits at age 12 (Taylor et al., 2013).
Furthermore, a study in the independent epidemiological ALSPAC sample showed that
children exhibiting high probability for abnormal SDQ hyperactive-inattentive symptoms were
at greater probability of persistent social communication deficits and that this interrelationship
was not reciprocal (St Pourcain et al., 2011).
These findings can be explained in two ways. First, it is possible that greater
hyperactivity raises the likelihood of ASD symptoms being noticed and reaching diagnostic
threshold. In previous work with the TEDS sample, we have shown that low IQ and/or teachernoted externalising problems increase the likelihood that females with high ASD traits meet
diagnostic criteria for ASD (Dworzynski, Ronald, Bolton, & Happé, 2012). Secondly, it is possible
that the findings reflect the existence of a combined clinical entity of autistic/hyperactiveinattentive syndrome (St Pourcain et al., 2011).
Alternatively, this ‘phenotypic causality’ is not a direct effect, but an association due to
correlated independent factors that we have not accounted for in our analysis. For example, a
family history of alcohol abuse was suggested to be a potential risk factor for both autism and
ADHD diagnoses in a recent Swedish study (Sundquist, Sundquist, & Ji, 2014). Another example
is maternal tobacco smoking during the first trimester and teenage pregnancy which was
identified as a shared prenatal/perinatal predictors of ASD and ADHD-like trait trajectories
across the child’s lifespan (St Pourcain et al., 2011). Future (longitudinal) studies are necessary
to test these alternative accounts.
139
5.4.3 Limitations
Sampling bias is important to consider in studies of comorbid conditions, since individuals with
more than one mental disorder are more likely to become part of a clinical sample (Neale &
Kendler, 1995), leading to artificially increased comorbidity rates when using clinically
ascertained groups. Our sample is population-based rather than clinic-ascertained. However, a
child’s ASD may still affect the likelihood that other problems are picked up by parents and
clinicians. ‘Over-observation’ is a possibility for inflating reports of additional problems.
Diagnostic overshadowing, conversely, may mean that all difficulties are attributed to the ASD
or Hyperactivity and co-existing problems are missed. However, if either of these general biases
were driving our findings, uniformly high or low correlations might be expected between ASD
and all three SDQ problems, which was not the case in our sample.
The fact that the measurement error could not be estimated in the direction of
causation models could potentially mean that certain variable relationships could be explained
as due to other than causal factors. As the BeD diagnosis was based on multiple diagnostic
instruments and opinions of diagnostic experts whereas the SDQ concerns only 5 item ratings
obtained solely from parents, their measurements errors are possibly not equal, as stated in the
Methods section. The reason behind this assumption is the fact that it is not possible to
specifically model measurement error when there are only two sources of variance in the data
– A and E in this instance (no shared environmental influences, C, influenced the variance in the
bivariate model). If it is assumed that the BeD variable has an SE lower or near zero (which is
not possible to detect) compared to SDQ, then the causal path pointing from BeD to SDQ might
be overestimated relative to the causal path pointing to BeD from SDQ (Neale & Cardon, 1992).
The only causal combination that could be potentially affected is the Emotional symptoms –
BeD model in chapter 5, as this causal path (r’) could be dropped when estimating the best
fitting model. Therefore, the accepted model where BeD causally influences Emotional
symptoms could be alternatively explained as due to differences in SE’s.
140
Other limitations include limited power to detect possible shared environmental
influences and reliance on parent-report SDQ ratings rather than direct assessment of the
children. Ideally, future studies ought to validate the SDQ against the functional impairment
characteristic in ASD while using in-person diagnostic ratings (Simonoff et al., 2013).
Additionally, our finding could be explained as ‘correlated error variance’ (Rijsdijk & Sham,
2002), reflecting the same raters’ difficulties evaluating ambiguous communication styles
resulting in erroneous evaluations of the child with ASD. To exclude potential rater biases, SDQ
measures should ideally be collected from multiple informants in future studies.
5.4.4 Conclusions
Comorbidity of other mental health problems in ASD is important to identify because it is often
treatable. Our findings suggest a possible effect of Hyperactivity on identification of ASD,
which could be the focus of future studies to clarify behavioural patterns as predictive of ASD
or other diagnoses. Conversely, relatively little overlap between ASD and Conduct problems
found here suggests the importance of discriminating apparently similar (e.g. socially
disruptive) behaviours due to different causes. Finally, the overlap of ASD and Emotional
symptoms explained as completely due to genes further strengthened the genetic aetiology
demonstrated by previous twin studies. Clinicians may wish to be alert to possible
internalising/externalising difficulties in siblings of those with ASD.
141
5.5 Supplementary Online Materials
Table 5-3 Concordance/discordance on SDQ measures among MZ and DZ pairs, split by concordance
on ASD-BeD.
BeD
Twin
Pairs with SDQ
Concordant
Concordance
Discordant on
Unaffected on
pairs
available (166
affected on
status (207
SDQ category
SDQ category
missing
pairs)
SDQ category
pairs)
SDQ
‘Narrow’ ASD
Emo MZ=13 &
MZ=8, DZ=4
MZ=3, DZ=4
MZ=2, DZ=1
6
MZ=17 &
DZ=9
(62%) (45%)
(23%) (45%)
(15%) (10%)
DZ=11
Hyp MZ=13 &
MZ=6, DZ=3
MZ=2, DZ=3
MZ=5, DZ=3
6
DZ=9
(46%) (33%)
(15%) (33%)
(39%) (33%)
Con MZ=13 &
MZ=1, DZ=1
MZ=3, DZ=4
MZ=9, DZ=4
6
DZ=9
(8%) (11%)
(23%) (44%)
(69%) (44%)
‘Narrow’ ASD
Emo MZ=5 &
MZ=4, DZ=2
MZ=1, DZ=6*
MZ=0, DZ=5
8
+ Broad
DZ=13*
(80%) (15%)
(20%) (46%)
(0%) (39%)
Spectrum
Hyp MZ=5 &
MZ=2, DZ=5*
MZ=2, DZ=5
MZ=1, DZ=3
8
MZ=7* &
DZ=13*
(40%) (39%)
(40%) (39%)
(20%) (22%)
DZ=19*
Con MZ=5
MZ=2, DZ=3*
MZ=3, DZ=5
MZ=0, DZ=5
8
&DZ=13*
(40%) (22%)
(60%) (39%)
(0%) (39%)
‘Narrow’ ASD
Emo MZ=2 &
MZ=1, DZ=4
MZ=0, DZ=18
MZ=1, DZ=21
17
+ unaffected
DZ=43
(50%) (9%)
(0%) (42%)
(50%) (49%)
MZ=2 &
Hyp MZ=2 &
MZ=0, DZ=4
MZ=1, DZ=25
MZ=1, DZ=14
17
DZ=60
DZ=43
(0%) (9%)
(50%) (58%)
(50%) (33%)
Con MZ=2 &
MZ=0, DZ=4
MZ=0, DZ=12
MZ=2, DZ=27
17
DZ=43
(0%) (9%)
(0%) (28%)
(100%) (63%)
Broad
Emo MZ=1 &
MZ=0, DZ=1
MZ=0, DZ=3
MZ=1, DZ=5
1
Spectrum +
DZ=9
(0%) (11%)
(0%) (33%)
(100%) (56%)
Unaffected
Hyp MZ=1 &
MZ=0, DZ=0
MZ=0, DZ=2
MZ=1, DZ=7
1
MZ=1 &
DZ=9
(0%) (0%)
(0%) (22%)
(100%) (78%)
DZ=10
Con MZ=1 &
MZ=0, DZ=0
MZ=1, DZ=7
MZ=0, DZ=2
1
DZ=9
(0%) (0%)
(100%) (78%)
(0%) (22%)
Unaffected
Emo MZ=24 &
MZ=1, DZ=1
MZ=6, DZ=10
MZ=17, DZ=36
9
MZ=29 &
DZ=47
(4%) (2%)
(25%) (21%)
(71%) (77%)
DZ=51
Hyp MZ=24 &
MZ=1, DZ=1
MZ=2, DZ=11
MZ=21, DZ=35
9
DZ=47
(4%) (2%)
(8%) (23%)
(88%) (75%)
Con MZ=24 &
MZ=2, DZ=2
MZ=4, DZ=13
MZ=18, DZ=32
9
DZ=47
(8%) (4%)
(17%) (28%)
(75%) (68%)
*1 MZ and 2 DZ pairs were concordant on Broad Spectrum; of these 3 pairs, 1 DZ pair had SDQ scores
available.
142
Table 5-4 Comparison of fit indices of the Bivariate AE model, the reciprocal and unidirectional Causal models for SDQ problems and ASD-BeD.
Variable
Model
-2LL
DF
np
AIC
BIC
p-value
2(DF)
1
Emotional
1 Correlations
2770.477
737 9
1296.477
-1159.736
symptoms
2 ACE
2775.600
737 9
1301.600
-1154.614
3 AE
2775.642
740 6
1295.642
-1170.570
0.04(3)
0.99
3a AE (e21=0)
2777.029
741 5
1295.029
-1174.515
1.39(1)
0.24
4 Reciprocal AE 2
2775.642
740 6
1295.642
-1170.570
&
3
2775.849
741 5
1293.849
-1175.695
0.21(1)
0.65
5 Drop r: Emo  BeD
2778.214
741 5
1296.214
-1173.330
2.57(1)&
0.11
6 Drop r’: Emo  BeD
Hyperactivity
1 Correlations 1
2883.043
737 9
1409.043
-1047.171
2 ACE
2890.503
737 9
1416.503
-1039.711
3 AE
2890.559
740 6
1410.559
-1055.653
0.06(3)
0.99
4 Reciprocal AE 2
2890.559
740 6
1410.559
-1055.653
2894.847
741 5
1412.847
-1056.698
4.29(1)&
0.04
5 Drop r: Hyp  BeD
2891.150
741 5
1409.150
-1060.395
0.59(1)&
0.44
6 Drop r’: Hyp  BeD 3
1
Conduct
1 Correlations
2623.63
737 9
1149.63
-1306.58
problems
2 ACE
2624.25
737 9
1150.25
-1305.96
3 AE
2624.32
740 6
1144.32
-1321.89
0.07(3)
0.99
3a AE (a21=0)
2625.53
741 5
1143.53
-1326.01
1.28(4)
0.86
4 Reciprocal AE 2
2624.32
740 6
1144.32
-1321.89
&
2627.10
741
5
1145.10
-1324.43
2.79(1)
0.10
5 Drop r: Con  BeD
&
3
2624.53
741 5
1142.53
-1327.01
0.21(1)
0.65
6 Drop r’: Con  BeD
1
Constrained correlation model: combined continuous SDQ problems and ordinal BeD analysis. To correct for ascertainment, threshold 1 and 2 on the liability to ASD are fixed to
correspond to 5% and 1% prevalence. Correlations are further constrained to obtain: 1 overall within-person cross-trait correlation; 1 MZ cross-trait cross-twin correlation and 1
DZ cross-trait cross-twin correlation.
2
Because all C parameters were non-significant and estimated to be close to zero, the AE model was selected as the base reciprocal model.
3
Best-fitting model (i) as compared to the Reciprocal AE model, and based on (ii) the lowest AIC and BIC values.
Note: Number of observed statistics (os) = 746 (same across all models); DF=os-np (number of parameters) AIC and BIC are DF-penalised; & compared to Reciprocal Causal
model.
-value concerns the difference in -2LL value of model 3 and 2 (testing the significance of parameter C); the second concerns the difference in
-2LL value of model 3a and 3 (testing the significance of either parameter a21 or e21); the third concerns the difference in -2LL value of model 5 and 4 (testing the significance of
parameter r); the fourth concerns the difference in -2LL value of model 6 and 4 (testing the significance of parameter r’).
143
Chapter 6 The aetiology of autism traits and other mental health
problems in a general population twin sample
Chapters 3 and 4 provided an extensive overview of the evidence that Autism Spectrum
Disorders are highly heritable with a moderate to small addition of non-shared environmental
influences. Chapter 5 investigated to what extent the aetiological factors that influence other
mental health problems are associated with those acting on ASD, using a ‘Correlated Risks’
model. An alternative view was also explored and tested the idea that these associations exist
because of a reciprocal (or unidirectional) phenotypic influence of one variable on another (the
Direction of Causation model).
The findings of that Chapter were novel. Particularly, a correlated risk model fitted the
relationship between both internalising problems and conduct problems with research
diagnosis of ASD better than phenotypic causality model. However, whereas for internalising
problems the relationship with ASD was totally explained by genetic effects, for conduct
disorder this was totally due to non-shared environmental effects. Finally, the relationship with
Hyperactivity was better predicted by the model assuming a phenotypic influence of
Hyperactivity on ASD, which has been previously reported but never in a diagnostic sample.
6.1 Overview
In this Chapter the aim is to extend the results found in the diagnostic SRS sample by examining
the aetiology of autism-like traits and associated comorbidities in a general population TEDS
twin sample. This is possible because of the innovative approaches of behavioural geneticists
studying psychological traits in large twin birth-cohort populations, such as the TEDS (Haworth
et al., 2012).
The notion that it is possible to study ‘autism’ in the general population has a long
history (see section 1.3 of the Introduction). Both Kanner (Kanner, 1943) and Asperger
(Asperger, translated by Frith 1991) made the early observations that autism-like traits were
144
present in the first- and second-degree relatives of children with the clinical condition
(Sucksmith et al., 2011). These sub-threshold expressions of autism-like behaviours in
unaffected family members provided the evidence that autism is a continuous (quantitative)
trait that prevails in the general population (Hoekstra, Bartels, Cath, & Boomsma, 2008).
On the molecular level, this notion is now supported by the demonstration that it is
common genetic variants (found across the whole population) that confer autism susceptibility
in the vast majority of cases (Gaugler et al., 2014). Knowing the aetiology and the processes
leading to individual differences in autism traits in the population is thought to enrich the
understanding of clinical diagnoses (Ronald & Hoekstra, 2014). Therefore, access to data from
thousands of TEDS families on the same phenotypes provides a unique opportunity for testing
the comorbidity models in the general population and comparing them to the results from the
sample of twins meeting diagnostic criteria for ASD, provided in Chapter 5.
6.1.1 Internalising problems and autism traits
There is a growing interest for understanding internalising problems in children with high ASD
traits (i.e. subclinical). Internalising is an inward expression of negative emotions in response to
external situations, resulting in symptoms of anxiety, depression, somatic complaints and
excessive fears/worry (Rieffe et al., 2011). Several general population twin studies have
explored the aetiology of these problems in relation to autism traits across developmental
stages.
A Swedish twin study looked at co-occurring anxiety traits in late childhood (ages 9-12),
and at co-occurring anxiety and depression traits in adults (Lundström et al., 2011). All traits
were moderate to highly heritable in childhood (51%-71%) and low to modest in adulthood
(13%-36%). None were influenced by shared environmental effects. The genetic correlations
with autism traits were between .51-.53 across both ages. Non-shared environmental
correlations were low (anxiety rE=.07-.10 & .21 for anxiety and for depressive traits,
respectively).
145
A similar approach was taken in the TEDS sample at age 8 when parents and teachers
rated children on autism traits and co-occurring anxiety traits (Hallett, Ronald, & Happé, 2009).
Similar rates of heritability were reported in the Swedish study, with some shared
environmental effects (13% for autism and 14% for anxiety traits), which overlapped entirely.
The correlations were modest for both genetic (parent rA=.12 & teacher rA=.19) and non-shared
environmental effects (rE=.07 & rE=.25). These statistics indicate little aetiological overlap
compared to the Swedish study. Later, an alternative domain-specific analysis in the same
TEDS sample revealed that it was communication difficulties and repetitive/restricted
behaviours which correlated most with generalised anxiety and negative affect. These
phenotypic correlations were due to modest genetic influences (Hallett et al., 2012).
A follow up TEDS study more specifically explored the direction of causation using
longitudinal modeling of anxiety and autism traits using the same measures collected at age 8
and 12 (Hallett et al., 2010). In this study, which was the first of its kind, the authors
demonstrated a bidirectional pattern of transmission, as autism traits at age 8 modestly
influenced internalising behaviours at age 12. The reverse path (internalising behaviours at 8
influencing autism traits at 12), although significant, was of lesser strength (.06 vs. .10-.15). The
small genetic overlap (rA=.17) across traits at both ages indicated influences specific to each
trait.
The latest of TEDS studies investigated the genetic and environmental overlap of
autism traits and comorbid internalising traits in 12-14 year-old adolescent twins (Scherff et al.,
2014). Internalising traits showed moderate heritability (~50%), low shared environmental and
moderate non-shared environmental effects when compared to findings at age 8 (Hallett et al.,
2009). In contrast, autism traits were 60% heritable with moderate shared and low non-shared
environmental effects. There was a stronger genetic overlap for males (rA=.30) vs. females (.12),
albeit confidence intervals overlapped. Non-shared environmental influences were mostly
independent. Finally, a Dutch study of 18 year old twins revealed that Withdrawn Behaviours
and Social Problems, known to be characteristic of anxiety/depressive symptoms, explained
146
23% of the variance of autistic traits, with most of this overlap due to common genetic aetiology
(Hoekstra et al., 2007).
In summary, the internalising traits and autism traits relationship in the literature to
date (mostly based on the TEDS sample) appears to be mainly due to shared genetic effects,
varying in strength across developmental stages. These effects are low in childhood and
moderate to strong in adolescence and adulthood. Possibly, there is an indication that higher
autism traits predispose to later internalising traits. When reported, shared environmental
effects are small to moderate but correlate strongly; non-shared environmental effects are
largely independent. Lastly, quantitative but not qualitative sex differences were found for
autism traits but not internalising traits.
6.1.2 Externalising problems and autism traits
6.1.2.1 ADHD
To externalise means to express negative emotions by directing them towards other people or
objects. Externalising or challenging behaviours are often seen in individuals on the autism
spectrum, in response to sensory stimuli or because that is the only way for them to
communicate their frustration. These behaviours involve physical aggression, emotional
outbursts, inability to sit still/concentrate and repetitive movements (e.g. hand flapping), which
are also seen in Attention Deficit Hyperactivity Disorder (ADHD).
Twin studies explored this comorbidity in children as young as 2 year old in a TEDS
sample of same-sex twins (Ronald, Edelson, Asherson, & Saudino, 2010). Variance in both traits
was best explained by an ACE model: autism traits were due to additive genetic (19%), sharedenvironmental (35%) and non-shared environmental effects (46%) and ADHD traits were due
to additive genetic (54%), shared-environmental (17%) and non-shared environmental effects
(30%). The genetic correlation between autism and ADHD was modest (r A=.27), while nonshared environmental contributions did not significantly overlap (rE=-.17). An independent
study of community sample of 7-15 year-old male twins using the same instrument (Children
Behavioural Checklist, CBCL) found that behavioural problems explained most of the
147
proportion of variance (43%) in autism traits, with CBCL Attention problems contributing most
strongly. Despite the high covariation, specific genetic factors for autism traits were found
(Constantino, Hudziak, & Todd, 2003). Note, that these studies were conducted on relatively
small samples (n=312 and n=219).
Ronald et al (Ronald, Simonoff, et al., 2008) examined the overlap between autismADHD traits in the general population TEDS sample, including twins at the quantitative
extreme. Teacher and parent reported childhood data showed high heritability for both traits
(59% to 89%) and a strong genetic association (rA=.50-.60). This effect was even stronger at the
quantitative extreme (rA=.62). Quantitative and qualitative sex differences were found for
teacher-reported data, but only the former for parent-reported data. When complemented
with the direction of causation model at ages 8 to 12 in the TEDS sample, ADHD traits modestly
predicted autism traits (.15) but the reverse effect was small (.05). Modest (rA female=.23) to
moderate (rA male=.41) genetic correlations were estimated at age 12 (Taylor et al., 2013).
However, findings of this study ought to be interpreted with caution as the cross-lag model
fitted to data poorly when compared to longitudinal Cholesky decomposition.
In an independent Swedish sample at ages 9 to 12 (Lichtenstein et al., 2010), both traits
were shown to be highly heritable at 80% & 79% and highly correlated (rA=.87). A similarly high
genetic correlation (rA=.72) was reported in an Australian community sample of adult twins
(ages 18-33) (Reiersen, Constantino, Grimmer, Martin, & Todd, 2008). Furthermore, a study in
a combined Dutch sample of adults aimed to elucidate which behavioural components of each
trait are most strongly associated (Polderman et al., 2013). They found that ADHD and autistic
traits
overlapped
primarily
because
of
shared
attention-related
problems
(inattention/attentional switching capacity), which were entirely genetic in origin. In summary,
studies to date show that the relationship between ADHD and autism traits is genetic in nature,
illustrated by high genetic correlations reported in most studies.
148
6.1.2.2 Conduct traits
Conduct problems are also thought of as a form of externalising problems due to the antisocial
and delinquent behaviours displayed. To date, there have been only 2 twin studies examining
the relationship between autism traits and behaviours related to conduct problems. A TEDS
study of 9 year-old twins revealed a similar aetiology for both antisocial and autism traits (~50%
heritability, ~20% shared environmental influences that correlated perfectly) and a moderate
genetic correlation (rA=.43). A negative correlation was found for non-shared environmental
effects (rE=.02, ns), indicating that these effects act on each trait independently (Jones et al.,
2009). A separate TEDS study of callous unemotional traits at age 7 and social-communication
impairments at age 8, reported a modest genetic correlation (~rA=.28) and small non-shared
environmental correlation (~rE=.09) (O’Nions & Tick et al., 2015). These findings and reported
modest phenotypic correlations (.21-.27) suggest that the genetic and environmental effects
are predominantly trait-specific.
In summary, there is strong evidence that ADHD and autism traits overlap due to
genetic reasons (that may act differently across sexes) with some indication that ADHD
phenotypically influences expression of autism traits. Limited evidence on conduct traits
provides a mixed picture that seems to be dependent on the measure employed, although
there is some suggestion of aetiological independence between autism and conduct traits.
6.1.2.3 Aims of the current study
This study was carried out to see if relationships at clinical extreme reported in Chapter 5 also
hold for individual differences for relevant traits at subclinical levels. For this reason, models
fitted to data follow the same order as in Chapter 5, with an addition of sex limitation models
to detect possible sex differences in aetiology. The ASD research diagnosis data used in Chapter
5 was replaced by the Childhood Autism Spectrum Test (CAST) scores at age 12 measuring
autism traits in the TEDS general population. However, the same set of the SDQ subscales of
Emotional symptoms, Hyperactivity and Conduct problems were available as measures of
associated psychiatric difficulties. Of the 207 SRS pairs included in the analysis in Chapter 5, 115
149
pairs had CAST score available at age 12 and were included in the analyses in the current
Chapter. No TEDS studies to date conducted analyses on this combination of SDQ-CAST
variables.
6.2 Methods
6.2.1 Participants
The analyses were performed on twin data derived from the TEDS (Haworth et al., 2012), a
general population representative birth-cohort of twins born in England and Wales between
1994-1996, described in depth in Chapter 2. Consent forms requesting participation when twins
were 12 years old were sent to 8439 families. Following from that, data on Childhood Autism
Spectrum Test ‘CAST’, (Scott, Baron-Cohen, Bolton, & Brayne, 2002) was obtained from 5767
families (data available for at least one twin in the pair). Strengths and Difficulties
Questionnaire 'SDQ' (Goodman, 1997) data were available for 5660 families for the three
subscales: Emotional symptoms, Hyperactivity and Conduct problems. The analyses described
below made use of all available data using full-information maximum likelihood estimation
procedures.
6.2.2 Measures
6.2.2.1 Childhood Autism Spectrum Test – CAST
CAST is a 31-item questionnaire designed to detect autistic-like traits in the general population
(Scott et al., 2002), based on behavioural descriptions of ASD as outlined in the ICD 10 and the
DSM IV. It is utilised as a screening tool in identification of individuals potentially ‘at risk’ of ASD
diagnosis, as demonstrated in Chapter 3. TEDS CAST measurement at age 12 (mean age=11.28,
SD=.70) involved a 30-item version (see Appendix 1), which assessed a full range of autism traits
while discarding an age-inappropriate item from the item list used for TEDS CAST
measurement at age 8 (Taylor et al., 2013). Parents provided a ‘yes’ or ‘no’ answer and
responses were summed.
150
A twin pair was included in the analysis if the parent provided answers for at least half
of the thirty items for both twins (Robinson et al., 2012). Exclusion criteria were applied to twin
pairs known to have serious medical syndromes (except ASD) or severe pregnancy
complications or a lack of parental consent signature. As reported in Chapter 2, an overall CAST
cut-off score of 15 points or greater has a positive predictive value of 50% (Williams et al., 2005).
In the current age group, 4.5% of the individual twins (258 individuals) met this cut-off criterion,
exactly the same prevalence rate reported for CAST at age 8 (Colvert & Tick et al., 2015).
Individual items analysis revealed an overall strong internal consistently for CAST for age 12 (α
=.73; Taylor et al., 2013).
6.2.2.2 Strengths and Difficulties Questionnaire – SDQ
SDQ (Goodman, 1997) is a widely used screening tool for positive and negative psychological
traits (Stone, Otten, Engels, Vermulst, & Janssens, 2010). In the current analysis, the sub-scales
of negative psychological traits were used: Emotional symptoms, Hyperactivity and Conduct
problems. We disregarded the Prosocial Behaviours and Peer relations as the aim of the current
chapter was similar to that of Chapter 5 – to explore the relationship of autism traits and the
three mentioned SDQ measures, but in a general population twin sample. Additionally, Peer
relations are putatively tapped by CAST (i.e. social impairment).
Each sub-scale contains five questions (see Appendix 2) and parents were asked to
provide ratings on a three-point Likert scale (0=’Not True’, 1=’Somewhat True’, 2=’Certainly
True’). Item scores were unavailable in the current dataset to explore scale reliability. However,
these are reported in other published studies from TEDS and show an overall moderate to
strong internal consistency across the subscales: Emotional symptoms Cronbach’s α = .67
(Scherff et al., 2014), Hyperactivity α = .76 (Merwood et al., 2013) and Conduct problems α = .55
(Jaffee, Hanscombe, Haworth, Davis, & Plomin, 2012).
151
6.3 Statistical Analyses
6.3.1 Twin Correlations
Chapter 2 provides detailed information on univariate and multivariate twin model fitting. Full
information maximum likelihood estimation was performed using Structural Equation
Modeling (SEM) in OpenMx (Boker et al., 2011) in R (URL http://www.R-project.org) to partition
the genetic and environmental influences on the variance and covariance of each SDQ problem
and CAST. In contrast to model-fitting performed in Chapter 5 involving joint ordinalcontinuous measures analysis, models in the current chapter are fitted to continuous SDQ &
CAST data.
For each sex-by-zygosity group a 4 x 4 correlation matrix can be estimated for each
combination of SDQ problem and CAST variable in Twin 1 and Twin 2. To make the results more
interpretable, it is customary to apply constraints on these correlations to obtain: 1. one crosstrait correlation for males and females separately; 2. cross-twin within-trait (twin) correlations
for SDQ and CAST for each of the 5 sex-zygosity groups; 3. and finally, one cross-twin crosstrait correlation (i.e. rSDQ1-CAST2 = rSDQ2-CAST1) for each of the 5 sex-zygosity groups. The
ratio of obtained MZ/DZ correlations provides an indication of the genetic (A) and/or shared(C) and non-shared environmental factors (E) that potentially influence the variation within
traits and the between-trait covariation. The strength of these influences is formally tested in
the bivariate genetic model.
6.3.2 Testing for sex differences – Sex Limitation Models
The rationale of sex limitation models is discussed in detail in Chapter 2. The availability of a
large dataset provides an opportunity in this Chapter to explore the potential differential
genetic and environmental effects in males and females. This kind of analysis proved to be
problematic in the SRS (Chapter 3 and 5), as the sample was not sufficiently powered to report
any meaningful results.
152
In first instance, a full sex-limitation bivariate ACE model was fitted to the data. This
model allows both quantitative sex differences for the A, C and E variance components (i.e.
estimate the genetic and environmental influences separately for males and females) as well as
qualitative sex differences for either the A or C components (Neale & Maes, 2001). Information
on quantitative sex differences comes from differential MZ/DZ ratios in males and females.
Information on qualitative sex differences comes from the opposite-sex DZ twin pair
correlation. Qualitative differences are indicated if these correlations are substantially lower
than those of the DZ same-sex pairs. In this case either the genetic (rAm-f) or shared
environmental factor correlations (rCm-f, estimated in a separate model) are allowed to vary
from those obtained in same-sex DZ pairs (which are fixed to their theoretical values of .5 and
1, respectively). See the path diagram for the opposite-sex pair twins in Figure 6-1 below,
illustrating both the qualitative and quantitative aspects of the model for the A and C
components. For simplicity the E component was omitted.
Figure 6-1 rA and rC calculated across the males and females in the opposite-sex DZ twin pairs, adapted
from Neale, Røysamb, & Jacobson, (2006).
153
In the current Chapter, because we are fitting multiple phenotypes (without a specific
order) and the sample includes opposite-sex pairs, an adjusted correlated-factors model
instead of the usual Cholesky decomposition was fitted. This model, outlined in Neale et al
(Neale, Røysamb, & Jacobson, 2006), is necessary to prevent the order of the variables affecting
the ability of the model to account for the DZ opposite-sex twins data. In this model for each
variance component we specify a diagonal matrix to estimate the effects on each of the traits
(SDQ and CAST) while allowing a correlation between the factors. A set of six models aimed to
establish the potential effects of gender on the aetiology of the three psychological difficulties
(SDQ) and autism traits (CAST) and their genetic and environmental overlap:
1a. ACE Heterogeneity (rAm-f free) – quantitative and qualitative sex differences are allowed for
the male and female parameters, while estimating the rAm-f in the opposite-sex DZ pairs (rAm-f
can be lower than .50 as assumed for the same-sex DZ pairs). The A, C and E correlations
between SDQ and CAST are also allowed to differ for males (notation: rAm, rCm, rEm) and females
(notation: rAf, rCf, rEf).
1b. ACE Heterogeneity (rAm-f free) - quantitative and qualitative sex differences are allowed for
the male and female parameters, while estimating the rCm-f in the opposite-sex DZ pairs (rCm-f
can be lower than 1 as assumed for the same-sex DZ pairs). The A, C and E correlations between
SDQ and CAST are also allowed to be different for males and females.
2. ACE Common Effects – quantitative differences between males and females are allowed but
NOT qualitative differences - the genetic correlations rAm-f are fixed to .50 and the shared
environmental correlations rCm-f to 1 in the opposite-sex DZ twins. The difference in fit of the
Common Effects model is tested against the fit of the heterogeneity models 1a and 1b to
indicate whether qualitative sex differences affect the aetiology of SDQ and CAST and their
overlap.
3. ACE Scalar Effects – quantitative but NOT qualitative sex differences are allowed (i.e. rAm-f is
fixed to .50 and rCm-f is fixed to 1 in the opposite-sex DZ twins). The A, C and E correlations
154
between SDQ and CAST are equated to be the same across males and females (rA, rC, rE). The
fit of this model is compared to the fit of the Common Effects model.
4. ACE Homogeneity – assumes no difference between the sexes in the aetiology of SDQ and
CAST and their overlap, therefore male and female parameters are equated. The fit of this
model is compared to the fit of the Common Effects model.
5. ACE Variance Inequality – assumes no differences between sexes as outlined in the ACE
Homogeneity model above, however, the expected male variances for SDQ and CAST are
specified as the female variances multiplied by a phenotypic scalar. If the variance of a trait is
larger in males than in females, the estimated scalar value will be >1, if it is smaller the scalar
value will be <1. The fit of this model is compared to the fit of the Common Effects model.
6.3.3 The Bivariate Genetic (no sex-differences) Models
The bivariate genetic model (not incorporating sex-differences) fitting the usual Cholesky
Decomposition is described in detail in Chapter 2. The set of models fitted in this Chapter is the
same as those in Chapter 5. The fitted Cholesky decomposition (Neale & Cardon, 1992) is
interpreted as the standardised correlated-factors solution because order of the variables is
immaterial.
The A (additive genetic) factors for continuous SDQ and CAST measures are the square
roots of their respective standardised path estimates, indicating each measure’s heritability.
The correlation path between the A factors for SDQ and CAST is the genetic correlation
between them (rA). Correlations between shared environmental (C) and non-shared
environmental (E) SDQ-CAST factors are calculated in the same manner (rC, rE). The obtained
A, C and E standardised path estimates allow to calculate the proportion of the phenotypic
correlation (rPH) due to correlated additive genetic effects (rPH_A = h12 * rA * h22), correlated
shared environmental (rPH_C = c12 * rC * c22) and correlated non-shared environmental effects
(rPH_E = e12 * rE * e22) (Owens et al., 2012). For ease of interpretation, these are expressed as
proportions of the phenotypic correlation (rPH_A/rPH; rPH_C/rPH; rPH_E/rPH) (Plomin et al., 2013).
155
6.3.4 The Direction of Causation (DoC) Models
The direction of causation model is described in detail in Chapter 2 (section 2.4.11). This model
was also fitted to the data in Chapter 5 to explore alternative explanations of the phenotypic
associations between two traits other than those examined in bivariate genetic model as per
‘correlated risks’ theory (Simonoff, 2000). In this model, the genetic and environmental
correlation between traits is replaced with estimates of causal effect of one trait on the other.
Using twin data, the DoC model will only work in instances when the relative
proportions of variance (h2, c2) sufficiently differ between traits, as the differentially predicted
cross-twin cross-trait covariances form the basis of the DoC model. As such, if the DoC model
covariance mimics the pattern of transmission of SDQ, for example, then SDQ causes CAST
(SDQCAST). The direction of causation between SDQ and CAST is established by fitting
unidirectional nested sub-models to a base reciprocal (bidirectional) model (which is derived
from the best fitting bivariate genetic model). Structural Equation Modeling is used to compare
models’ fit using 2 statistic as well as AIC and BIC fit indices (Burnham et al., 2011; Raftery,
1995).
6.4 Results
6.4.1 Descriptive Statistics
Figure 6-2 depicts the mean (and Standard Error bars) for the males and females that met the
cut off criteria for autism traits (CAST=>15 points) or scored within the normal range (<15) and
available SDQ data. Table 6-2 provides means (and Standard Deviation) for SDQ and CAST
measures by sex & zygosity groups.
Table 6-1 provides information on the percentage of the TEDS sample meeting SDQ
cut offs (borderline or abnormal, see Methods chapter section 2.2.5), provided separately for
males and females that met the CAST cut off and those scoring within the normal range.
156
10
9
Emotional symptoms
Hyperactivity
Conduct problems
8
6.59
7
5.49
6
5
4.02
4
4.28
3.59
3.18
3.29
3
1.6
2
1.89
1.39
2.29
1.18
1
0
CAST =>15
(N=167) Males
CAST <15
(N=5176) Males
CAST =>15
(N=66) Females
CAST <15
(N=5904) Females
Figure 6-2 Raw means (and Standard Error of the Mean bars) for SDQ problems across the TEDS
sample.
Table 6-1 Percentage of males and females (low/high on CAST) meeting SDQ suggestive cut offs.
CAST status
Emotional
symptoms
CAST =>15 (N=167)
Borderline: 55%
Abnormal: 42%
Borderline: 14%
Abnormal: 8%
Hyperactivity
Conduct problems
MALES
CAST < 15 (N=5176)
Borderline: 65%
Abnormal: 54%
Borderline: 17%
Abnormal: 10%
Borderline: 65%
Abnormal: 44%
Borderline: 20%
Abnormal: 8%
FEMALES
CAST =>15
(N=66)
CAST < 15 (N=5904)
Borderline: 65%
Abnormal: 52%
Borderline: 18%
Abnormal: 11%
Borderline: 55%
Abnormal: 46%
Borderline: 7%
Abnormal: 4%
Borderline: 59%
Abnormal: 41%
Borderline: 15%
Abnormal: 6%
Notes: Suggestive cut offs for Emotional symptoms: borderline = 4, abnormal = 5-10;
Hyperactivity: borderline = 6, abnormal = 7-10;
Conduct problems: borderline = 3, abnormal = 4-10.
157
Table 6-2 Means (M) and standard deviation (SD, in bracket) for SDQ and CAST measures by sex &
zygosity groups.
Measure
MZM
DZM
DZOm
MZF
DZF
DZOf
CAST
M=5.23
M=5.37
M=6.05
M=4.26
M=4.55
M=4.39
(3.77),
(3.68),
(3.81),
(3.12),
(3.14),
(3.17),
N=1850
N=1734
N=1968
N=2265
N=1950
N=1767
Emo
M=1.67
M=1.68
M=1.69
M=1.91
M=1.94
M=1.90
(1.82),
(1.87),
(1.90),
(1.96),
(1.95),
(1.97),
N=1848
N=1731
N=1766
N=2263
N=1944
N=1765
Hyp
M=3.39
M=3.25
M=3.52
M=2.30
M=2.51
M=2.15
(2.29),
(2.40),
(2.51),
(1.97),
(2.14),
(1.89),
N=1848
N=1732
N=1764
N=2263
N=1944
N=1765
Con
M=1.45
M=1.51
M=1.41
M=1.16
M=1.21
M=1.24
(1.47),
(1.59),
(1.56),
(1.34),
(1.41),
(1.41),
N=1847
N=1732
N=1765
N=2263
N=1944
N=1765
Abbreviations: MZM=Monozygotic Males, DZM=Dizygotic Males, DZOm=Dizygotic Opposite-Sex
Males, MZF=Monozygotic Females, DZF=Dizygotic Females, DZOf=Dizygotic Opposite-Sex Females.
CAST=Children Autism Spectrum Test, Emo=SDQ Emotional Symptoms, Hyp=SDQ Hyperactivity,
Con=SDQ Conduct Problems.
6.4.2 Phenotypic analyses
In preparation for model fitting analyses, the raw data scores were regressed on age and sex to
prevent an over-estimation of shared environmental influences due to inflated similarity of
same-sex twins compared to opposite-sex non-twin siblings (McGue & Bouchard, 1984). The
obtained residual scores were [log(x+constant)] transformed in R (URL http://www.Rproject.org) to correct for the positively skewed distribution due to high proportion of twins
scoring low across the measures. The regressed/transformed measures then met the
assumption of normal distribution indicated by skewness and kurtosis values falling within the
±1 range. Table 6-1 provides mean (Standard Deviation) descriptive statistics for each sex &
zygosity group.
Table 6-3 (below) shows the MZ and DZ within-pair as well as the cross-twin crossCAST MZ and DZ correlations and the strength of the phenotypic overlap. There was a modest
phenotypic overlap between Emotional symptoms and CAST (rPH) that varied between .28-.30.
For externalising problems the association with CAST was moderate, ranging between .32-.37.
Doubling the difference of the MZ and DZ within-trait correlations provides an indication of the
aetiological profile for each measure. DZ correlations higher or lower than half of that of MZ
158
indicate shared environmental influences in the former and dominant genetic effects in the
latter.
Table 6-3 Phenotypic results for the three SDQ problems and CAST.
Measure
All Twins
CAST
Emo
cross-twin
within-trait
rMZ
cross-twin
within-trait
rDZ
cross-twin
cross-CAST
MZ1
cross-twin
cross-CAST
DZ1
rPH
.78 (.77-.80)
.62 (.59-.64)
.41 (.39-.44)
.36 (.33-.38)
n/a
.24 (.22-.26)
n/a
.17 (.14-.19)
n/a
.29 (.27-.31)
Hyp
.78 (.77-.80)
.27 (.24-.30)
.32 (.30-.34)
.22 (.20-.24)
.35 (.33-.37)
Con
.78 (.76-.79)
.51 (.49-.53)
.29 (.27-.31)
21 (.18-.23)
.32 (.30-.34)
MZ/DZ Males
CAST
.78 (.75-.80)
.34 (.29-.40)
n/a
n/a
n/a
Emo
.61 (.57-.64)
.32 (.26-.38)
.26 (.23-.29)
.15 (.10-.19)
.30 (.27-.33)
Hyp
.78 (.75-.80)
.23 (.17-.29)
.34 (.31-.36)
.22 (.18-.27)
.37 (.35-.40)
Con
.76 (.74-.79)
.49 (.44-.53)
.28 (.25-.31)
.20 (.16-.34)
.32 (.30-.35)
MZ/DZ Females
CAST
.78 (.76-.80)
.49 (.45-.54)
n/a
n/a
n/a
Emo
.62 (.59-.65)
.35 (.30-.40)
.21 (.18-.24)
.17 (.13-.21)
.28 (.25-.30)
Hyp
.76 (.74-.78)
.28 (.23-.33)
.31 (.28-.33)
.23 (.19-.27)
.34 (.31-.36)
Con
.79 (.77-.81)
.53 (.49-.57)
.29 (.26-.32)
.23 (.19-.27)
.32 (.29-.34)
DZ Opposite Sex
CAST
.44 (.41-.48)
n/a
n/a
n/a
Emo
.37 (.33-.41)
.18 (.15-.21)
.30 (.27-.33)
Hyp
.26 (.22-.31)
.22 (.19-.25)
.37 (.35-.40)
Con
.53 (.50-.56)
.21 (.18-.34)
.32 (.29-.35)
Abbreviations: rMZ & rDZ – maximum likelihood within trait Monozygotic Dizygotic twin correlations; r PH
– phenotypic correlation; CAST=Children Autism Spectrum Test, Emo=SDQ Emotional Symptoms,
Hyp=SDQ Hyperactivity, Con=SDQ Conduct Problems.
1
For SDQ-CAST, 3 sets of correlations are calculated for each group as 3 bivariate analyses were
performed; here, for simplicity, only one set of CAST estimates is provided (the other two set of values
were identical to the first decimal place and with overlapping 95%CI).
The ‘All Twins’ values in Table 6-3 indicate moderate to strong heritability and shared
environmental influences (C) for CAST, Emotional symptoms and Conduct problems.
Hyperactivity also appears to be strongly genetic, but unlike other measures that are driven
mainly by additive genetic (A) factors, there is an indication of dominance (D) rather than
shared environmental influences. These indications were taken into account when performing
multivariate sex-limitation models. For Emo-CAST and Con-CAST combinations, an overall
ACE model was assumed. However, for Hyp-CAST combination, a hybrid model was fitted that
159
allowed estimation of dominant genetic effects (D) specific to Hyperactivity and shared
environmental effects specific to CAST (C), discussed next.
6.4.3 Sex-limitation analyses
Table 6-4 provides the fit indices (2, AIC, BIC) of the three sets of bivariate sex-limitation
models. The goodness-of-fit of a competing model (compared to the base model) was assessed
with a chi-square statistic (2) and change in estimated degrees of freedom (DF), where a nonsignificant 2 (p>0.05) indicates that the suggested (more restricted) model fits the data nonsignificantly worse. An increase in chi square value of 3.84 per each DF indicates a poor fit and
the model is rejected. Conversely, the smaller the AIC/BIC values the better the fit of the model,
while accounting for the number of parameters estimated (AIC) and BIC indices have proven to
outperform other fit statistics in large samples and complex models. (Markon & Krueger, 2004).
For this reason, the BIC indices are used as the main source of information when deciding on
the best fitting model results presented in Table 6-4.
The indices are provided for the AC(D)E Heterogeneity as the base model and then the
overall best fitting model, chosen on the basis of parsimony and BIC indices. Table 6-5 provides
the estimates and their CI derived from these models. Overall, no qualitative sex differences in
ACE parameters can be expected across the three SDQ-CAST combinations as the Common
Effects model fitted better than the full Heterogeneity model (meaning that genetic and shared
environmental effects [for Emo and Con SDQ only] could be equated across males and females
in DZ opposite-sex pairs). Similarly, no quantitative sex differences were found, albeit variance
size inequality was found to be significant across the sexes for Hyperactivity, Conduct problems
and CAST as the Scalar values >1 (see Table 6-5). Altogether, these findings do not indicate that
further model fitting on males and females separately should be considered. The aetiology
specific to each SDQ-CAST combination is the same for males and females and sufficiently
explained by one set of ACE/ADE parameters.
160
Table 6-4 Model-fit indices for bivariate sex-limitation models fitted to each SDQ-CAST combination.
Emo-CAST
-2LL
np
AIC
BIC
p-value
comparison
2(DF)
1a ACE Heterogeneity (rAm-f free)
31439.98
26
-14094.02
-170955.46
.79 (4)
.94
model 2 to 1a
1b ACE Heterogeneity (rCm-f free) &
31440.75
26
-14093.25
-170954.69
.02 (4)
.99
model 2 to 1b
2 ACE Common Effects
31440.76
22
-14101.24
-170990.23
3 ACE Scalar Effects
31453.00
19
-14095.00
-171004.70
12.24 (3)
<.01
model 3 to 2
4 ACE Homogeneity
31893.85
13
-13666.15
-170617.15
453.08 (9)
<.001
model 4 to 2
5 ACE Variance Inequality**
31459.78
15
-14096.22
-171033.44
19.02 (7)
<.01
model 5 to 2
Con-CAST
1a ACE Heterogeneity (rAm-f free)
22234.99
26
-23297.01
-180151.55
3.93 (4)
>.41
model 2 to 1a
1b ACE Heterogeneity (rCm-f free)
22234.99
26
-23297.01
-180151.55
3.93 (4)
>.41
model 2 to 1b
2 ACE Common Effects
22238.93
22
-23301.07
-180183.17
3 ACE Scalar Effects
22243.90
19
-23302.10
-180204.90
4.97 (3)
>.17
model 3 to 2
4 ACE Homogeneity
22753.38
13
-22804.62
-179748.73
514.45 (9)
<.001
model 4 to 2
5 ACE Variance Inequality**
22249.52
15
-23304.48
-180234.81
10.59 (7)
>.16
model 5 to 2
Hyp-CAST
1 ACDE Heterogeneity (rAm-f free)
8221.044
24
-37030.96
-192755.09
0 (4)
=1
model 2 to 1
2 ACDE Common Effects
8221.044
20
-37038.96
-192790.62
3 ACDE Scalar Effects
8223.51
18
-37040.49
-192805.95
2.47 (2)
=.29
model 3 to 2
4 ACDE Homogeneity
8842.847
12
-36433.15
-192239.88
621.80 (8)
<.001
model 4 to 2
5 ACDE Variance Inequality
8241.683
14
-37030.32
-192823.28
20.64 (6)
<.002
model 5 to 2
5a ADE Variance Inequality**
8242.045
13
-37031.96
-192831.80
21 (7)
<.01
model 5a to 2
Best fitting model: ACE/ADE Variance Inequality - Models one set of A, C/D and E paths (np=6), one set of A, C (for Emo and Con only) and E correlations between the variables
(rA, rC [for Emo or Con only], rE), but the male variances for SDQ and CAST are multiplied by a scalar, to allow for sex differences in variance of the scales (np=2). Total np=15 (for
Hyp-CAST np=13).
Note: Male and Female means for SDQ and CAST measures are estimated freely in all models (np=4).
&
Model 1b is fitted for Emo-CAST and Con-CAST only, as no C influences could be estimated in the hybrid Hyp-CAST model.
** Each of these models is compared to an AC(D)E Common Effects model (assumes quantitative but NOT qualitative sex differences) given that it does not fit the data worse
than AC(D)E Heterogeneity (rGm-f free) model, which is the case for the three SDQ-CAST combinations. Abbreviations: -2LL – minus twice the log likelihood; np – number of
estimated parameters; AIC – Akaike Information Criterion; BIC – Bayesian Information Criterion; 2(DF) and p-value likelihood-ratio testing (difference in -2LL and DF).
161
Table 6-5 Estimates for each SDQ-CAST combination as estimated in the full heterogeneity sex-limitation model and the best-fitting Variance Inequality model.
Model/Measure
Male (SDQ/CAST)
Female (SDQ/CAST)
Male (SDQ-CAST)
Female (SDQ-CAST)
AC(D)E
Heterogeneity
a 2m
d2/c2m
e 2m
a 2f
d2/c2f
e 2f
rAm
rCm
rEm
rAf
rCf
(rG)
.52 (.41-.60) .10 (.03-.19) .39 (.35-.42) .50 (.39-.59) .12 (.05-.21) .38 (.35-.42) .32 (.25-.43) 1 (-.04-1)
.13 (.06-.19) .15 (.03-.28) .91 (.45-.1)
Emo-CAST $
Hyp-CAST $**
Con-CAST $
Best Fitting
Model:
Variance
Inequality
Emo/CAST
Hyp-CAST**
Con-CAST
.73 (.67-.77)
.18 (.13-.28)
.75 (.70-.78)
.53 (.45-.60)
.72 (.66-.75)
a2
.50 (.44-.57)
.70 (.65-.74)
.13 (.12-.23)
.78 (.76-.79)
.52 (.47.57)
.71 (.65-.75)
.04 (.01-.09)
.59 (.50-.65)
.02 (.00-.06)
.23 (.17-.30)
.05 (.02-.10)
d2/c2
.11 (.06-.17)
.08 (.04-.12)
.63 (.54-.66)
.26 (.21-.30)
.07 (.03-.12)
.23 (.21-.26)
.23 (.20-.25)
.23 (.21-.25)
.24 (.21-.26)
.23 (.21-.26)
e2
.39 (.36-.41)
.22 (.21-.24)
.23 (.22-.25)
.22 (.21-.24)
.22 (.21-.24)
.22 (.21-.24)
.61 (.52-.70)
.25 (.17-.40)
.68 (.60-.77)
.51 (.43-.58)
.61 (.52-.70)
rA
.17 (.08-.26)
.51 (.36-.59)
.10 (.02-.18)
.29 (.21-.36)
.17 (.08-.26)
rC
.22 (.20-.24)
.24 (.22-.26)
.22 (.20-.24)
.20 (.19-.23)
.22 (.20-.24)
rE
rEf
.21 (.15-.26)
.91 (.73-1)
n/a
.14 (.08-.20)
.74 (.57-.91)
n/a
.12 (.06-.17)
.29 (.22-.36)
1 (.67-1)
.16 (.09-.22)
.22 (.11-.33)
.75 (.49-1)
.13 (.08-.19)
rPH
rPH_A
rPH_C
rPH_E
Scalar
SDQ/
CAST
.24 (.18-.29)
1 (.72-1)
.17 (.13-.21)
.29 (.27-.31)
.14 (.10-.18)
.10 (.06-.13)
.05 (.04-.07)
.90/1.27
.99 (.77-1)
-
.35 (.33-.37)
.32 (.30-.34)
-
.03 (.02-.04)
1.25/1.26
.27 (.21-.33)
.96 (.68-1)
-.12 (-.16-.08)
.15 (.11-.19)
.32 (.30-.34)
.15 (.11-.19)
.14 (.10-.17)
.03 (.02-.04)
1.17/1.27
Abbreviations: a2, d2/c2, e2 - standardized parameter estimates for SDQ and CAST scores; m – males, f - females; rA, rC, rE – genetic, C and E correlations between SDQ-CAST
scores; rPH – phenotypic overlap between SDQ/CAST; rPH_A – phenotypic overlap due to A factors; rPH_C – phenotypic overlap due to C factors; rPH_E – phenotypic overlap due to E
factors.
**ADE model
$
Freely estimated genetic correlations (rGm-f) in opposite-sex pairs in the Heterogeneity model:
Emo-CAST: rAm1-f1 =.50 (.43-.50); rAm2-f2 =.50 (.46-.50); rAm1-f2 =.04 (-.10-.15); rAm2-f1 = .21 (.12-.28).
Hyp-CAST: rAm1-f1 =.50 (.33-.50); rAm2-f2 =.50 (.48-.50); rAm1-f2 =.50 (.41-.50); rAm2-f1 =.50 (.39-.50).
Con-CAST: rAm1-f1 =.50 (.45-.50); rAm2-f2 =.50 (.45-.50); rAm1-f2 =-.03 (-.16-.09); rAm2-f1 =.15 (.06-.23).
162
6.4.4 Bivariate Genetic and Direction of Causation (DoC) no-sex differences Model
Results
Based on the results of the sex-limitation models that showed no sex differences in the
aetiology (apart from relatively small variance differences), a new set of (no-sex-differences)
bivariate models were fitted using the preferred Cholesky Decomposition. The reason for this
is that this decomposition is more favourable in terms of optimization and is preferred over the
‘correlation approach’ if not strictly necessary (Neale, Røysamb, & Jacobson, 2006). This does
mean that the estimates below might be slightly different than those provided in Table 6-4.
The fit indices for both types of models can be found in Table 6-7. The bivariate genetic
models formally tested the proportion of A, C, D and E influences on the (co)variance of the
three SDQ and CAST combinations as suggested by phenotypic correlations results. Just like in
Chapter 5, DoC models were fitted as an alternative to genetic model in order to determine the
most likely mode of transmission of additional mental health difficulties in populations
displaying varied levels of autism traits. The genetic and DoC models were compared using AIC
and BIC fit statistics and revealed that the best model was the one assuming the genetic and
environmental influences on the covariance between each SDQ problem and autism traits.
A visual representation of these models fitted to the overall data and the derived
estimates is given in Figure 6-3. Unlike in the sex-limitation models (and the no-sex-difference
variance inequality model using the correlation approach), specific C influences could be
successfully removed for CAST (c22) across the three SDQ-CAST combinations without
significant worsening in fit. Table 6-5 shows the proportion of the phenotypic overlap for each
SDQ-CAST combination explained by additive genetic (A) and non-shared environmental
factors (E).
Overall, a differential aetiology is observed across the individual traits as well as for the
overlap between autism traits and SDQ measures. All traits displayed moderate to strong
heritability accounted for by additive genetic effects, except from Hyperactivity, which was
strongly influenced by dominant genetic effects. Shared environmental influences were found
163
for Emotional symptoms and Conduct problems. The non-significant C estimate in Emo-CAST
model could not be dropped (as well as path c21). The only explanation is that despite the nonsignificant specific C effects on CAST, there is a small but significant C effect on CAST due to
the overlap between Emotional symptoms and CAST. Similarly, for Con-CAST, path c21 also
could not be dropped from the model. It is very likely that these small effects are detected
because of the large sample size.
Figure 6-3 The best fitting (no sex-differences) models and derived parameter estimates.
In terms of the aetiology of the overlap, moderate genetic correlations were reported for both
Emotional symptoms and Conduct problems when combined with CAST. The almost perfect
genetic correlation between Hyperactivity and CAST can be explained as due to pleiotropic
effects – the small proportion of additive genetic effects influencing Hyperactivity will almost
certainly affect expression of CAST. Weak to moderate non-shared environmental effects as
well as weak bivariate associations were found. Finally, the weak to moderate phenotypic
correlations between SDQ problems and CAST were mostly explained by genetic influences.
164
Table 6-6 Phenotypic overlap due to A and E as estimated in the genetic bivariate model for each SDQ-CAST combination.
Model
Emo-CAST
Hyp-CAST
Con-CAST
rPH
rPH_A
rPH_E
.28 (.26-.30)
.34 (.33-.36)
.30 (.29-.32)
.24 (.22-.26)
.32 (.30-.34)
.28 (.26-.30)
.04 (.03-.05)
.02 (.01-.04)
.02 (.01-.03)
rPH_A /rPH
86%
94%
93%
rPH_E /rPH
14%
6%
7%
Table 6-7 Model-fit indices for the bivariate genetic and Direction of Causation models fitted to each SDQ-CAST combination.
Variable
Emo-CAST
Hyp-CAST
Con-CAST
Model
1 Correlations 1
2 ACE
3 AcE (c11 for SDQ) 2
4 Reciprocal AcE 3
5 Drop Emo  CAST^
6 Drop Emo  CAST^
1 Correlations 1
2 Hybrid ACDE
3 ADE 4
4 Reciprocal ADE 3
-2LL
18833.27
18834.51
18834.51
18855.09
18863.32
18937.56
23870.56
23901.51
23901.51
23901.87
DF
22766
22766
22767
22768
22769
22769
22765
22766
22767
22767
np
11
11
10
9
8
8
11
10
9
9
AIC
-26698.73
-26697.49
-26699.49
-26680.91
-26674.68
-26600.44
-21659.44
-21630.49
-21632.49
-21632.13
BIC
-190132.42
-190131.19
-190140.37
-190128.96
-190129.91
-190055.67
-185085.96
-185064.19
-185073.37
-185073.01
2(DF)
0(1)
8.23(1)
82.47(1)
0(2)
-
p-value
1
0.004
0.001
1
-
5 Drop Hyp  CAST^
6 Drop Hyp  CAST^
1 Correlations 1
2 ACE
3 AcE (c11 for SDQ) 2
4 Reciprocal AcE 3
5 Drop Con  CAST^
6 Drop Con  CAST^
24014.36
24103.47
22649.84
22650.06
22650.06
22743.47
22762.88
22726.51
22768
22768
22765
22765
22766
22767
22768
22768
8
8
11
11
10
9
8
8
-21521.64
-21432.53
-22880.16
-22879.94
-22881.94
-22790.53
-22773.12
-22809.49
-184969.69
-184880.58
-186306.67
-186306.46
-186315.64
-186231.40
-186221.17
-186257.54
112.49(1)
201.61(1)
0(1)
19.41(1)
57.27(1)
0.001
0.001
1
0.001
0.001
165
Table 6-7 Abbreviations: -2LL – minus twice the log likelihood; np – number of estimated parameters; AIC – Akaike Information Criterion; BIC – Bayesian Information Criterion;
2(DF) and p-value likelihood-ratio testing (difference in -2LL and DF).
1
Constrained correlation model: correlations are constrained to obtain (in addition to the MZ and DZ within-trait correlations): 1 overall within-person cross-trait correlation; 1
MZ cross-trait cross-twin correlation and 1 DZ cross-trait cross-twin correlation.
2
For both Emo-CAST and Con-CAST combinations paths c11 and c21 could not be removed, leading to worsened model fit. The significance of path c21 will not be regarded any
further since we primarily focused on estimating the strength of associations of aetiological factors common for both measures and are of substantial effect size in the bivariate
model, in this case factors A and E (the specific C factor is not significant for CAST in both models).
3
Reciprocal DoC model is based on the best fitting genetic model. In this model, the a 21, c21, e21 paths are replaced with r (SDQCAST) or r’ (SDQCAST) phenotypic causal
paths, hence reduced number of parameters (np) compared to the genetic model for Emo-CAST and Con-CAST combinations. Hyp-CAST reciprocal
model is not affected by this reduction as there is an assumption of differential aetiology on each measure from the start (no d21 or c21 paths are calculated in the hybrid model).
4
The specific C factor for CAST could be removed from the hybrid ACDE model, leading to the best fitting ADE model in which dominant and additive genetic effects are allowed
to influence the variance of Hyperactivity SDQ. As explained under subscript 2 above, only associations between the A and E factors for both measures are relevant for the
covariance of the variables.
^ For each SDQ-CAST combination, the uni-directional model fits are compared to the Reciprocal model fit (the -2LL of model 4), providing the 2statistics for these models in
the last two columns. Similarly, the 2statistics for models 3 in the last columns are relative to the fit (-2LL) of models 2.
166
6.5 Discussion
Using the same measures of internalising and externalising traits as in Chapter 5, the analyses
reported in the current Chapter examined the overlap with autism traits in the subclinical range,
and tested whether any overlap was due to shared genetic and/or environmental risk factors.
The overlap due to causal phenotypic effects between the measured behaviours was also
explored.
6.5.1 Suggestive cut-offs findings
Before moving on to discussion of the fitted genetic models, we discuss some results using
clinical cut-offs on the population data measures of this sample. First of all, it is worth noting
that there was an approximate 2½:1 ratio of males to females meeting the CAST cut-off criteria
for elevated autism traits (and thus at risk of an ASD diagnosis). Secondly, both males and
females that met the CAST cut-off displayed higher levels of the three measured co-occurring
problems (mean scores met borderline cut off for all three problems in both gender groups)
when compared to twins within the normal range of autism traits.
We then looked at the SDQ suggestive cut-offs. Considering twins scoring highly on
autism traits, there are marked differences in the percentage of males vs. females meeting
either cut off criteria (borderline or abnormal) for Emotional symptoms and Hyperactivity. The
results show that larger proportion of males met both cut offs for Hyperactivity in comparison
to females. Conversely, more females met borderline and abnormal cut offs for Emotional
symptoms in comparison to males. Both genders were comparable on the proportion of each
group meeting cut offs for Conduct problems. Moreover, a similar pattern of results held for
twins scoring within the normal range of autism traits, with the exception of a larger proportion
of males meeting cut offs not only for Hyperactivity but also for Conduct problems when
compared to females. Comparison of the high vs. low- scoring twins on CAST, irrespective of
gender, showed that three-times as many twins with high CAST scores met the borderline cut
off compared to twin scoring within the normal range. This ratio was even higher for the
abnormal cut off as a four to five-fold difference was noted between high and low scoring twins.
167
Combined, these findings show that twins meeting the CAST cut off were affected by
internalising and externalising problems far more than twins scoring within the normal range.
This follows the pattern of findings in Chapter 5, as the levels of other mental health
problems were observed to increase almost linearly as a function of ASD symptom severity.
That is, twins in the Broad Spectrum category scored higher than unaffected twins on all three
SDQ measures and twins in the ASD category scored the highest of all.
6.5.2 Analyses of the total spectrum of CAST and SDQ measures
The aetiology of the overlap for the three SDQ problems and autism traits were uniform as the
‘Correlated risks’ model proved to be a better fit to the data in comparison to the Direction of
Causation models. Looking first at the aetiology of autism traits measured by the CAST at age
12, individual differences were mainly due to strong additive genetic effects and shared
environmental influences were non-significant, mirroring the results reported in the Swedish
study (Lundström et al., 2011). It is somewhat distinct from the aetiology of the same measure
at age 8 and Autism Quotient at age 12-14 in the same sample, as these two studies found small
but significant shared environmental effects (Hallett et al., 2009; Scherff et al., 2014). It is
possible that this is due to each study’s assumption of sex differences and the way findings are
reported, as we found shared environmental effects to be also small but significant in the sex
limitation analyses but report the results of the (better-fitting, more parsimonious) no-sexdifferences models here.
The aetiology of internalising problems was characterised here by moderate additive
genetic influences and non-shared environmental factors comparable to the results of previous
studies, even though different measures are used (Hallett et al., 2009; Scherff et al., 2014).
However, the genetic correlation with autism traits is stronger in our analyses when compared
to parent-reported childhood measures (Hallett et al., 2009) and adolescent ratings (Scherff et
al., 2014) and comparable to the Swedish sample, with a caveat that their genetic correlation is
based on a lower heritability estimate for anxiety (Lundström et al., 2011). Finally, when
compared to findings from diagnosed ASD twins in Chapter 5, genetic associations are
168
comparable and the confidence intervals partially overlap. An important limitation of the
emotional symptoms measure is that it is not possible to specify whether in the current samples
these problems are reflective of underlying depressive or anxiety symptoms, taking into
account the heterogeneity of environmental risks that lead to each of these conditions
(Kendler, Neale, Kessler, Heath, Eaves, 1992).
Because Hyperactivity was strongly influenced by dominant genetic effects, it presents
a different aetiology profile from the previous studies reported in the introduction, but it is in
keeping with another TEDS study, which provided support for an ADE model for SDQ
Hyperactivity across different raters (Merwood et al., 2013). For this reason, meaningful
comparisons cannot be made and analyses in the TEDS sample did not replicate the phenotypic
directional effect reported in Chapter 5.
Conduct problems were strongly heritable in the current analyses, with a moderate
influence of shared environmental factors. The overlap with autism traits was characterised by
moderate genetic and low non-shared environmental correlation, which is in keeping with
previous reports (Jones et al., 2009; O’Nions & Tick et al., 2015).
6.5.3 Diagnostic vs population sample results
When comparing the results across samples (i.e. current results with the results reported in
Chapter 5), the main conclusions are as followed:
(i) Overall, the heritability estimates of the SDQ measures as well as ASD/CAST are the
same across clinical and population samples.
(ii) In terms of phenotypic comorbidity, the results for Emotional symptoms and
Hyperactivity are similar across the clinical (.32 & .29 respectively) and general population (.28
& .34) samples, respectively, based on overlapping 95% CI.
(iii) In terms of the aetiology of this comorbidity in both the clinical and general
population samples, although seemingly different, the results for Emotional symptoms and
Hyperactivity are again fairly similar in the sense that most of the phenotypic overlap is
169
explained by genetic factors for clinical versus population samples: 100% vs 86% for Emotional
symptoms and 62% vs 94% for Hyperactivity.
(iv) The biggest differences in results from the clinical and general population samples
are seen for the comorbidity of autism traits and Conduct problems. First of all, the phenotypic
comorbidity is significantly smaller in the clinical sample (.10) compared to the population
sample (.30), with non-overlapping 95% CI. In addition, in terms of the aetiology of this
comorbidity, most (93%) is explained by genetic effects in the population sample, compared to
0% in the clinical sample.
The difference in phenotypic overlap could be due to the fact that conduct problems
measured in the population and clinical samples are not entirely the same. It has been
suggested that conduct problems are conflated with autism-like behaviours (Moffitt, 2011). It
is perhaps easier to conflate these behaviours in individuals that display mild versions of both
behavioural problems. If parents are not able to clearly distinguish e.g. an emotionally cold
comment from a slight lack of understanding of a social situation, or conduct problems due to
naughtiness from problems due to anxiety or social situations, they might be inclined to inflate
scores on these behaviours as well as the CAST, leading to a significantly higher phenotypic
correlation. It is also possible that having a diagnosis of ASD helps parents to interpret their
child’s (mis)behaviour.
Overall, the results, for Emotional symptoms and Hyperactivity do compare well across
populations, meaning that not only CAST but also other mental health problems and their
comorbidity are well represented by a spectrum of severity.
170
Chapter 7 General Discussion
This section of the thesis will briefly recap the overarching goals of the thesis. Next, the main
findings are summarised and general conclusions and implications are provided.
Finally, general limitations and future directions are presented. The chapter will conclude with
final remarks.
7.1 Summary of aims and findings
7.1.1 Re-evaluating the aetiology of ASD
The first goal of this thesis was to re-evaluate the aetiology of ASD. Chapter 3 focused on
estimating genetic and environmental influences on ASD in the SRS sample ascertained from
a larger, general population-representative birth cohort, the TEDS. The two novel aspects of
this study were: first, the diagnostic sample included twins with high levels of autism traits
(defined as Broad Spectrum) and selected low-risk twins, as well as those with diagnosed ASD,
to capture the entire range of the underlying liability to autism. Second, using a bivariate
model-fitting approach, the genetic/environmental relationship between a dimensional autism
trait measure and categorical diagnostic constructs of ASD was explored.
The findings confirmed that liability to autism was mostly attributable to additive
genetic influences and very little support was found for the influence of shared environmental
factors. This trend held across continuous and categorical measures of autism (CAST, DAWBA,
ADOS, Best-estimate diagnosis). The ADI-R, as an exception, was shown to have some
significant shared-environmental variation. However, this could be due to rater bias effects as
ADI-R is uniquely based on parental information. Since the same parent rated both twins in the
SRS sample, this could have inflated their phenotypic similarity. The second major finding was
that the aetiological relationship between the dimensional trait measure and the categorical
constructs of autism was predominantly genetic, indicated by a substantial genetic correlation.
This finding provides strong support for the notion of inclusion of the Broad Spectrum category
171
when estimating the genetic liability to autism. In summary, based on a careful systematic
sample selection procedure, including individuals with different levels of severity of autism
traits, the findings of Chapter 3 did not support the suggestion of major shared environmental
influences, as reported by two recent twin studies (Frazier, Thompson, et al., 2014; Hallmayer
et al., 2011).
In light of the findings of Chapter 3, and the controversy surrounding the findings by
Hallmayer et al and Frazier et al reporting significant and predominant shared environmental
influences on ASD, we sought to establish the sources of discrepant findings in the literature.
Therefore, the main aim of Chapter 4 was to systematically review all ASD twin studies
published to date and to carry out a meta-analysis, in order to gain statistical power to derive
the most objective genetic and environmental estimates. Applying appropriate sample
selection and ascertainment corrections, which was not always the case in the original studies,
twin correlations and heritability estimates were independently derived for each study. Several
meta-analyses were then conducted on the combined data. The effect of assumed prevalence
rates (used as fixed threshold to correct for sample ascertainment) on individual studies and
meta-analytic estimates was investigated.
The findings of Chapter 4 demonstrated that the individual-study and meta-analytic
heritability estimates ranged from strong to substantial. The main finding, however, was that
applying different prevalence rates, and hence thresholds, significantly affected the estimates
of shared environmental effects. Assuming a prevalence rate of 5% yielded substantial
heritability and non-significant shared environmental effects. Strikingly, as the threshold was
pushed from 5 to 3 and then to 1% prevalence rate, shared environmental influences became
increasingly significant. This effect is due to the fact that more stringent prevalence rates
increase the phenotypic similarity of the DZ twins relative to the MZ twins, which increases the
effects of shared environment. The MZ correlations are relatively unaffected by this process as
they are already at the top of their statistical bound. In summary, these results indicate that
172
previously reported significant shared environment effects are likely a statistical artefact due
to prevalence assumptions, as well as possible oversampling of DZ concordant pairs.
7.1.1.1 Conclusions and implications
Chapters 3 and 4 complement one another in application of novel design and sampling methods
when deriving estimates of the genetic and environmental effects on autism. By doing so, it is
hoped that these can provide benchmarks for future studies. Cumulatively, they provide further
support for the model of substantial genetic influences on the development of autism, across
the whole spectrum. At this stage, the claim of strong shared environment factors looks
relatively weak.
Chapter 3 provides the first set of evidence of the impact of incorporating the Broad
Spectrum category on the same liability as the ASD diagnosis in genetic models. No other study
to date included this meaningful subgroup as part of the liability to ASD in their design, despite
the long-standing evidence that some family members of autism probands display high levels
of autism traits and therefore are genetically part of the autism spectrum distribution (see
section 1.3 of Chapter 1). By demonstrating that inclusion of a Broad Spectrum category is
possible in twin model-fitting, it is hoped that future studies will consider following this practice
to represent better the underlying structure of autism. The Broad Autism Phenotype is
conceptualised in this thesis categorically, however, this does not mean that future studies
cannot include it as a quantitative measure instead. To pursue the development of this research
area, and following the explicit recognition in DSM 5 that observed symptoms of ASD fall on a
continuum, it will be necessary for the autism community to be clear how to operationalise the
spectrum notion.
Use of different fixed thresholds, synonymous with different prevalence rates of
autism, showed an impact on twin correlations (and heritability estimates), but also highlights
uncertainty about the ‘real’ prevalence of ASD or the Broad Autism Phenotype. The suggestive
prevalence of 5% for the latter agrees with the rate obtained from data used within this thesis
on two occasions. First, when the 6423 eligible TEDS pairs were assessed at age 8 for autism
173
traits, 4.5% scored above the CAST cut off (Chapter 3). Secondly, the same prevalence was
observed in the dataset describing TEDS adolescent twins (12-14 years old) in Chapter 6. A
recent study demonstrated an even higher rate of the BAP: 14-23%, estimated in an adult
sample (Sasson, Lam, Parlier, Daniels, & Piven, 2013). As long as studies rely on observational
data and not biological indices to understand the aetiology of autism, future model-fitting
designs ought to implement a ‘differential fixed threshold’ approach (i.e., modelling the effect
of taking different putative thresholds), to reflect that prevalence rates are dependent on
changing definitions and cut-offs used to capture diagnosed ASD.
For now, and until better- powered and well-designed studies come forward, the
cumulative evidence supports the genetic rather than the shared environmental effects as the
main aetiologic component of individual differences in autism. However, the research
community ought to remain open-minded and not exclude the possibility of environmental, or
at least non-genetic, effects on the development of autism. Heritability estimates apply to the
specific population examined, in the current environmental setting. Other populations, or
changes of environmental factors, will affect heritability estimates. Even the meta-analytic
estimates derived in this thesis are based on data from a tiny fraction of all twins that are
affected by autism worldwide. As seen in Table 4-1 (page 106), most clinical/diagnostic twin
studies published to date are from the US or Western and Northern Europe, with the exception
of one study originating from Japan. More evidence is urgently needed to understand the
aetiology of autism in the remaining parts of the world, which represent a huge cultural as well
as economic diversity. Diagnostic as well as population data on autism traits will verify to what
extend they differ from wealthy countries’ data. It is not implausible that the aetiology profiles
will not be uniform and specific to regions or countries.
These findings have strong implications on the potential research directions in autism
as a field. Twin studies can provide an indication about the area of research that financial
resources should be diverted to. As the genetic causes appear to be a lot stronger risk factor for
autism than environmental causes, investing in the next-generation sequencing studies to look
174
at genetic variant-specific causes would seem appropriate. However, as genes interact with the
environments, their expression can be changed through environmental interventions. The best
example of that is the dietary changes that prevent the development of a highly genetic
condition – the phenylketonuria (PKU). As autism is also under genetic influences, studying of
environmental interventions should also be high on the autism research agenda. As shown in
section 1.7.3, there is an increased incidence of autism following birth complications. It is
therefore important to investigate whether better provision of maternity care can prevent birth
complications and consequently decrease incidence of autism.
7.1.2 The relationship of ASD with associated psychiatric difficulties
The second goal of the thesis was to study the aetiology of the relationship between autism
and associated psychiatric difficulties. Chapter 5 focused on exploring this overlap in twins with
an ASD diagnosis, recruited as part of the SR study. To date, no formal model-fitting analyses
of multiple psychiatric conditions have been performed in such a group. Using a twin modelfitting approach, the study tested for the first time whether comorbid SDQ Emotional
symptoms, Hyperactivity and Conduct problems arise due to an aetiological overlap with ASD
or because of causal phenotypic associations.
The findings showed that half of twins identified as probands (either assigned to Broad
Spectrum or ASD under the Best-estimate Diagnosis category [ASD-BeD]) met the criteria for
borderline/abnormal levels for Emotional symptoms or Hyperactivity, and a quarter of
probands met these criteria for the three SDQ comorbid difficulties, indicating a high level of
additional psychiatric difficulties. At a phenotypic level, both Emotional symptoms and
Hyperactivity showed moderate phenotypic overlap with ASD-BeD, comparable to estimates
derived from general population twin studies of autism traits. In contrast, there was little
overlap between Conduct problems and ASD-BeD. Model fitting analyses revealed that the
relationship of Emotional symptoms (and Conduct problems) with ASD-BeD was best
explained by an ‘aetiological overlap’ model. For the former, the overlap was entirely explained
by genetic influences and for the latter by non-shared environmental factors. Even though
175
genetic influences explained more than half of the phenotypic overlap between Hyperactivity
and ASD-BeD, the model predicting that Hyperactivity phenotypically influences ASD-BeD
symptoms was the best-fitting model. This finding may suggest that heightened hyperactivity
raises the likelihood of ASD symptoms being noticed and reaching diagnostic thresholds. Or, it
could be taken as evidence for the existence of a combined clinical entity of
autistic/hyperactive-inattentive syndrome, as suggested by previous studies (Pourcain et al.,
2011).
Chapter 6 aimed to investigate the aetiology of comorbidity of the relevant traits at
subclinical levels in a large, general population birth-cohort – the TEDS sample. Similar sets of
models exploring the aetiology of potential overlap were fitted to data as in Chapter 5, with the
difference that ASD research diagnosis was replaced by a continuous measure of autism traits
(CAST). In addition, due to bigger sample size and power, sex limitation models were fitted to
the data to detect possible sex differences in the shared aetiology of autism traits and
Emotional symptoms, Hyperactivity and Conduct problems.
Regarding gender differences in phenotypic means, there were at least twice as many
males in comparison to females that met the clinical cut-off for elevated autism traits (therefore
at risk of an ASD diagnosis). Members of both genders that met the CAST cut-off displayed
higher levels of the three associated psychiatric difficulties, compared to individuals within the
normal range of autism traits. There were marked differences in the proportion of each gender
meeting cut-offs for specific associated problems, as more males than females met both
borderline and abnormal cut offs for Hyperactivity, whereas more females than males met both
cut offs for Emotional symptoms. Despite these differences between the two genders, sex
limitation models did not find sex differences in the aetiology across the three difficulties and
autism traits, and the best-fitting model described these relationships as due to overlapping
aetiological factors.
For Emotional symptoms the findings converged with those from the diagnosed ASD
SR sample as a moderate genetic correlation with autism trait was reported, with genetic
176
factors explaining most of that phenotypic overlap. The aetiology of Conduct problems was in
keeping with previous reports, and the overlap with autism traits was characterised by a
moderate genetic and a low non-shared environmental correlation, departing from findings in
Chapter 5. Finally, we found that the relationship between Hyperactivity and autism traits was
best explained by shared additive genetic and non-shared environmental effects, with the
caveat that Hyperactivity in the TEDS sample was mostly due to dominant genetic effects and
only a fraction of its variance was explained by additive genetic effects. These findings are quite
different to those obtained in the diagnostic sample, as the relationship between ASD and
Hyperactivity was best explained by the direction of causation model. The differences in
findings across the two samples are likely due to threshold effects imposed by the diagnosis of
ASD, which may alter parents’ understanding (and hence report) of externalising behaviours in
diagnosed vs. normal range autism traits samples. Parents that care for children with ASD may
have more insight when separating (mis)behaviours specific to ASD diagnosis and those that
are due to hyperactivity/inattention. Conversely, valid separation of hyperactive/inattentive
behaviours and autistic-like symptoms by parents of children with subclinical/non-diagnosed
forms of either behaviour may be more problematic. This may be because both are seen by
parents as ‘challenging’ behaviours; it may be hard to know whether the child struggles with
social situations because s/he does not understand vs. has a short attention span. Where a child
has not received a diagnosis of ASD, despite high ASD traits, parents’ attributions regarding
the reasons for difficult behaviour, may be different.
7.1.2.1 Conclusions and implications
Chapters 5 and 6 provided compatible evidence in the sense that heritability estimates for
autism traits and ASD measures, as well as for the associated psychiatric difficulties, are
comparable across the diagnostic and population samples. The phenotypic overlap of
Emotional symptoms and Hyperactivity with autism spectrum/ traits across the two samples
was revealed to be of moderate strength, based on overlapping confidence intervals across
estimates. Furthermore, despite the differences in the best-fitting model for Emotional
177
symptoms-ASD-BeD and Hyperactivity-ASD-BeD, these overlaps are explained largely by
genetic influences.
The biggest difference across the diagnostic and general population samples was in
relation to associated Conduct problems and spectrum of autism traits. The phenotypic overlap
was three times higher (although still moderate) in the population compared to the diagnostic
sample and the confidence intervals did not overlap. Moreover, in the former the overlap was
mostly accounted for by genetic influences whereas none were found in the diagnostic sample
and instead explained by non-shared environmental factors. The conclusion is that these
differences could stem from difficulties over correct identification of conduct problems vs.
autism-related behaviours, especially in the general population samples. Conflating of both
behavioural manifestations could lead to inflated scores on both Conduct problems and autism
traits measures, leading to significantly higher phenotypic correlations and differences in
derived aetiologies.
There are several implications of these findings. Researchers in the past were
somewhat baffled by the relatively low genetic associations of internalising problems and
autism traits in the general population samples, and called for evidence from diagnosed
samples. Yet the evidence of this thesis continues to support the notion that internalising
problems overlap only moderately, and because of common genes, with ASD diagnosis.
Looking outside of the moderate genetic commonality that can be explained by pleiotropy, it
is implied that in the current sample these traits are at least to some level independent
aetiologically. It is therefore possible that even though we observe high overall levels of
internalising, each individual might be developing these problems in a unique way, highlighting
further the complexity with which associated difficulties develop in ASD. Additionally, SDQ
Emotional symptoms scale is a short measure therefore has limited power to detect alternative
sources of worry and depressive moods, such as social unease.
If internalising problems are developed uniquely from person to person then they will
require individual-tailored treatment. As twin studies typically rely on remotely-collected and
178
questionnaire data, it will be important in future to design acceptable tools that nonetheless
capture associated difficulties with greater precision, and account for functional impairment in
ASD. More effort should also go into data collection from multiple sources, which could become
easier given the advent of online-based assessments.
The different best-fitting models for associated Hyperactivity across the diagnostic and
general population samples of autism traits highlight the possibility of different phenotypic
manifestations and aetiological processes. A lot more research is necessary to validate/refute
these findings and the current study should be treated as an indicator in need of further
replication to elucidate the aetiology of associated psychiatric problems in ASD. If hyperactivity
or inattention problems in ASD are truly a part of an underlying new syndrome that combines
these two traits, then allowing diagnosis of both ADHD and ASD in DSM 5 is a progressive step.
7.1.3 Limitations and future directions
Limitations of the twin study methodology are discussed at length in section 2.4 as well as in
the relevant chapters. The lack of IQ data means that this confounding factor potentially
affecting the recognition of ASD was not controlled for. A major limitation not previously
discussed in this thesis is the low number of female participants in the diagnostic samples. As
already mentioned in Chapter 1, it is possible that the current diagnostic system and practices
are more appropriate for males and not necessarily good at recognising subtle autism
manifestations in females, resulting in female underdiagnosis. This would limit the
generalisability of the present findings to male populations only. When conducting the metaanalysis study, it was hoped that by pulling the samples together the collective number of
females would allow sex limitation models to be fitted to the diagnostic data. However, it was
not the case and even combining all existing studies there were still too few females to conduct
this analysis. It can only be suggested that future studies attempt more balanced sampling of
males and females, to make future findings applicable to both genders.
Another way to address this, which stems from the finding of the current thesis that
autism is indeed dimensional, is to continue to supplement autism research with studies on
179
males and females with subclinical manifestations. These populations are more abundant and
less difficult to recruit. Involving families with children with an ASD diagnosis is an arduous task.
The SRS sample was not immune to this problem despite best efforts to capture as many UK
twins as possible within the carefully curated TEDS sample. The fact that we see a third of
eligible families not participating in the SRS could be potentially explained by the extreme
pressure of raising twins who have a severe neurodevelopmental condition. It is also worth
noting that entering adolescence (SRS sample focused on 12-14 year olds) may be particularly
stressful for families of children with ASD; children may be undergoing emotional, hormonal,
educational and residential transitions (Smith, Greenberg, Mailick, 2013).
It must also be highlighted that in the SRS sample SDQ data was available for only
eighty percent of the sample, further limiting the generalisability of findings. Secondly, we
were not able to fit any longitudinal direction of causation models to detect the dynamic
psychological changes that individuals with ASD diagnosis experiences. This is because the SRS
sample data was only available when twins were between ages 12-14. However, the future aim
of the study is to collect data when twins are approaching adulthood, which is currently
underway. The issues surrounding first diagnosis of ASD in adulthood are indeed very
interesting, but beyond the scope of this thesis, given that all diagnostic assessments were
carried out in SRS by ages 12-14. Availability of data at both time points will allow analyses
detecting any potential changes in the aetiology of associated psychiatric difficulties.
7.1.4 Final remarks
This thesis can be summed up by three major contributions to the current knowledge on the
aetiology of autism:
1. Autism is highly heritable in the studied populations.
2. Genetic influences on ASD are largely shared with those on autistic traits.
3. Associated psychiatric difficulties are highly comorbid in autism, but can have different
phenotypic/aetiological relationships with ASD and autism traits.
180
It is hoped that the work reported here will contribute to better understanding of the causes of
ASD and comorbid difficulties, in order to help address these challenges and improve the lives
of those with ASD and their families.
181
References
Allison, C., Baron-Cohen, S., Wheelwright, S., Charman, T., Richler, J., Pasco, G., & Brayne, C.
(2008). The Q-CHAT (Quantitative CHecklist for autism in toddlers): A normally
distributed quantitative measure of autistic traits at 18-24 months of age: Preliminary
report. Journal af Autism and Developmental Disorders, 38, 1414–1425.
American Psychiatric Association. (1980). Diagnostic and statistical manual of mental
disorders (Third edition). Washington D.C.
American Psychiatric Association. (1987). Diagnostic and statistical manual of mental
disorders (Revised third edition). Washington D.C.
American Psychiatric Association. (1994). Diagnostic and statistical manual of mental
disorders (Fourth edition). Washington D.C.
American Psychiatric Association. (2000). Diagnostic and statistical manual of mental
disorders (Revised fourth edition, text revision). Washington D.C.
American Psychiatric Association. (2013). Diagnostic and Statistical manual of mental
disorders (Fifth edition). Washington, DC.
Andrews, G., Slade, T., & Issakidis, C. (2002). Deconstructing current comorbidity: data from
the Australian National Survey of Mental Health and Well-Being. The British Journal of
Psychiatry : The Journal of Mental Science, 181, 306–314.
Anney, R., Klei, L., Pinto, D., Regan, R., Conroy, J., Magalhaes, T. R., … Hallmayer, J. (2010). A
genome-wide scan for common alleles affecting risk for autism. Human Molecular
Genetics, 19, 4072–4082.
Antshel, K. M., Zhang-James, Y., & Faraone, S. V. (2013). The comorbidity of ADHD and
autism spectrum disorder. Expert Review of Neurotherapeutics, 13, 1117–1128.
Aronow, W. S., Ahn, C., Mercando, A. D., & Epstein, S. (2000). Prevalence of coronary artery
disease, complex ventricular arrhythmias, and silent myocardial ischemia and incidence
of new coronary events in older persons with chronic renal insufficiency and with normal
renal function. American Journal of Cardiology, 86, 1142+.
Asperger, H. (1991). Autistic psychopathy’ in childhood (translated by Frith U). In Frith U ed
Autism and Asperger syndrome (pp. 37–92).
Auyeung, B., Baron-Cohen, S., Ashwin, E., Knickmeyer, R., Taylor, K., & Hackett, G. (2009).
Fetal testosterone and autistic traits. British Journal of Psychology, 100, 1–22.
Auyeung, B., Taylor, K., Hackett, G., & Baron-Cohen, S. (2010). Foetal testosterone and
autistic traits in 18 to 24-month-old children. Molecular Autism, 1:11.
182
Bailey, A., Lecouteur, A., Gottesman, I., Bolton, P., Simonoff, E., Yuzda, E., & Rutter, M.
(1995). Autism as a Strongly Genetic Disorder - Evidence From a British Twin Study.
Psychological Medicine, 25, 63–77.
Bailey, A., Palferman, S., Heavey, L., & Le Couteur, A. (1998). Autism: The phenotype in
relatives. Journal of Autism and Developmental Disorders, 28, 369–392.
Baird, G., Simonoff, E., Pickles, A., Chandler, S., Loucas, T., Meldrum, D., & Charman, T.
(2006). Prevalence of disorders of the autism spectrum in a population cohort of children
in South Thames: the Special Needs and Autism Project (SNAP). Lancet, 368, 210–215.
Bale, T. L., Baram, T. Z., Brown, A. S., Goldstein, J. M., Insel, T. R., McCarthy, M. M., …
Nestler, E. J. (2010). Early life programming and neurodevelopmental disorders.
Biological Psychiatry, 68:4, 314–319.
Bandim, J. M., Ventura, L. O., Miller, M. T., Almeida, H. C., & Costa, A. E. S. (2003). Autism and
Mobius sequence - An exploratory study of children in northeastern Brazil. Arquivos De
Neuro-Psiquiatria, 61, 181–185.
Barile, J. P., Kuperminc, G. P., Weintraub, E. S., Mink, J. W., & Thompson, W. W. (2012).
Thimerosal Exposure in Early Life and Neuropsychological Outcomes 7-10 Years Later.
Journal of Pediatric Psychology, 37, 106–118.
Baron-Cohen. (2002). The extreme male brain theory of autism. Trends in Cognitive Sciences,
6, 248–254.
Baron-Cohen, S. (2012). Autism and the Technical Mind. Scientific American, 307, 72–75.
Baron-Cohen, S., Auyeung, B., Nørgaard-Pedersen, B., Hougaard, D. M., Abdallah, M. W.,
Melgaard, L., … Lombardo, M. V. (2014). Elevated fetal steroidogenic activity in autism.
Molecular Psychiatry, 1–8.
Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The AutismSpectrum Quotient (AQ): Evidence from Asperger syndrome/high-functioning autism,
males and females, scientists and mathematicians. Journal of Autism and Developmental
Disorders, 31, 5–17.
Bauman, M. L. (2010). Medical Comorbidities in Autism: Challenges to Diagnosis and
Treatment. Neurotherapeutics, 7, 320–327.
Baxter, a J., Brugha, T. S., Erskine, H. E., Scheurer, R. W., Vos, T., & Scott, J. G. (2014). The
epidemiology and global burden of autism spectrum disorders. Psychological Medicine,
1–13.
Becker, A., Woerner, W., Hasselhorn, M., Banaschewski, T., & Rothenberger, A. (2004).
Validation of the parent and teacher SDQ in a clinical sample. European Child and
Adolescent Psychiatry, Supplement, 13:II, 11-16.
Betancur, C., Lebover, M., & Gillberg, C. (2002). Increased rate of twins among affected sibling
pairs with autism. American Journal of Human Genetics, Elsevier (Cell Press), 70(5): 1381-3.
Betancur, C. (2011). Etiological heterogeneity in autism spectrum disorders: More than 100
genetic and genomic disorders and still counting. Brain Research, 1380, 42–77.
183
Bettelheim, B. (1967). The empty fortress: infantile autism and the birth of the self. The Free
Press (London: Collier-McMillan Ltd.).
Bleuler, E. (1951). Autistic thinking. In Organization and pathology of thought. Selected sources
(pp. 399–437).
Blumberg, S. J., Bramlett, M. D., Kogan, M. D., Schieve, L. A., & Jones, J. R. (2013). Changes in
Prevalence of Parent-reported Autism Spectrum Disorder in School-aged U.S. Children:
2007 to 2011– 2012. National Health Statistics Reports, 1–12.
Boker, S., Neale, M., Maes, H., Wilde, M., Spiegel, M., Brick, T., … Fox, J. (2011). OpenMx: An
Open Source Extended Structural Equation Modeling Framework. Psychometrika, 76,
306–317.
Bolte, S., & Poustka, F. (2004). Diagnostic Observation Scale for Autistic Disorders: initial
results of reliability and validity. Z Kinder Jugendpsychiatr Psychother., Feb;32(1):
Bolton, P. F., Pickles, A., Murphy, M., & Rutter, M. (1998). Autism, affective and other
psychiatric disorders: patterns of familial aggregation. Psychological Medicine, 28, 385–
395.
Bolton, P., Macdonald, H., Pickles, A., Rios, P., Goode, S., Crowson, M., … Rutter, M. (1994). A
case-control family history study of autism. Journal of Child Psychology and Psychiatry
and Allied Disciplines, 35, 877–900.
Bolton, P., & Rutter, M. (1990). Genetic Influences in Autism. International Review of
Psychiatry, 2, 67–80.
Bondy, S. C., & Campbell, A. (2005). Developmental neurotoxicology. Journal of Neuroscience
Research, 81, 605–612.
Brendgen, M., Boivin, M., Dionne, G., Barker, E. D., Vitaro, F., Girard, A., … Pérusse, D. (2011).
Gene-Environment Processes Linking Aggression, Peer Victimization, and the TeacherChild Relationship. Child Development, 82, 2021–2036.
Brendgen, M., Vitaro, F., Boivin, M., Girard, A., Bukowski, W. M., Dionne, G., … Pérusse, D.
(2009). Gene-environment interplay between peer rejection and depressive behavior in
children. Journal of Child Psychology and Psychiatry, and Allied Disciplines, 50, 1009–1017.
Brugha, T. S., McManus, S., Bankart, J., Scott, F., Purdon, S., Smith, J., … Meltzer, H. (2011).
Epidemiology of autism spectrum disorders in adults in the community in England.
Archives of General Psychiatry, 68, 459–465.
Burnham, K. P., Anderson, D. R., & Huyvaert, K. P. (2011). AIC model selection and
multimodel inference in behavioral ecology: Some background, observations, and
comparisons. Behavioral Ecology and Sociobiology.
Centres for Disease Control and Prevention. (2010). How Prevalence is calculated. Retrieved
from
http://www.cdc.gov/pednss/how_to/read_a_data_table/calculating_prevalence.htm
Chakrabarti, B., Dudbridge, F., Kent, L., Wheelwright, S., Hill-Cawthorne, G., Allison, C., …
Baron-Cohen, S. (2009). Genes Related to Sex Steroids, Neural Growth, and Social184
Emotional Behavior are Associated with Autistic Traits, Empathy, and Asperger
Syndrome. Autism Research, 2, 157–177.
Christensen, J., Grønborg, T. K., Sørensen, M. J., Schendel, D., Parner, E. T., Pedersen, L. H., &
Vestergaard, M. (2013). Prenatal valproate exposure and risk of autism spectrum
disorders and childhood autism. JAMA : The Journal of the American Medical Association,
309, 1696–703.
Colvert, E., Tick, B., McEwen, F., Stewart, C., Curran, S., Woodhouse, E., … Bolton, P. (2015).
Heritability of Autism Spectrum Disorder in a UK population-based twin sample. JAMA
Psychiatry, 72(5):415.
Constantino, J. N. (2011). The quantitative nature of autistic social impairment. Pediatric
Research, 69 (5 Pt 2), 55-62.
Constantino, J. N., Hudziak, J. J., & Todd, R. D. (2003). Deficits in Reciprocal Social Behavior in
Male Twins: Evidence for a Genetically Independent Domain of Psychopathology.
Journal of the American Academy of Child & Adolescent Psychiatry, 42, 458–467.
Constantino, J. N., Przybeck, T., Friesen, D., & Todd, R. D. (2000). Reciprocal social behavior in
children with and without pervasive developmental disorders. J Dev Behav Pediatr., 21,
2–11.
Constantino, J. N., & Todd, R. D. (2003). Autistic traits in the general population: a twin study.
Archives of General Psychiatry, 60, 524–530.
Constantino, J. N., & Todd, R. D. (2005). Intergenerational transmission of subthreshold
autistic traits in the general population. Biological Psychiatry, 57, 655–660.
Croen, L. a., Grether, J. K., & Selvin, S. (2002). Descriptive Epidemiology of Autism in a
California Population: Who Is at Risk? Journal of Autism and Developmental Disorders, 32(3),
217–224.
Croen, L. A., Grether, J. K., Yoshida, C. K., Odouli, R., & de Water, J. (2005). Maternal
autoimmune diseases, asthma and allergies, and childhood autism spectrum disorders A case-control study. Archives of Pediatrics & Adolescent Medicine, 159, 151–157.
Cross-Disorder Group of the Psychiatric Genomics, C., Lee, S. H., Ripke, S., Neale, B. M.,
Faraone, S. V, Purcell, S. M., … Wray, N. R. (2013). Genetic relationship between five
psychiatric disorders estimated from genome-wide SNPs. Nat Genet, 45, 984–994.
Curran, S., Dworzynski, K., Happe, F., Ronald, A., Allison, C., Baron-Cohen, S., … Bolton, P. F.
(2011). No major effect of twinning on autistic traits. Autism Research, 4, 377–382.
Dales, L., Hammer, S. J., & Smith, N. J. (2001). Time trends in autism and in MMR immunization
coverage in California. JAMA-Journal of the American Medical Association, 285, 1183–
1185.
De La Marche, W., Noens, I., Kuppens, S., Split, J. L., Boets, B., & Steyaert, J. (2015). Measuring
quantitative autism traits in families: informant effect or intergenerational transmission?
European Child & Adolescent Psychiatry, 24(4), 385–395.
185
De Rubeis, S., & Buxbaum, J. D. (2015). Recent Advances in the Genetics of Autism Spectrum
Disorder. Current Neurology and Neuroscience Reports, 15:36.
De Rubeis, S., He, X., Goldberg, A. P., Poultney, C. S., Samocha, K., Cicek, A. E., …
Consortium, A. S. (2014). Synaptic, transcriptional and chromatin genes disrupted in
autism. Nature, 515, 209–U119.
De Weerdt, S. (2011). Reclassification of Rett syndrome diagnosis stirs concerns. Simons
Foundation Autism Research Inititative. https://sfari.org/news-andopinion/news/2011/reclassification-of-rett-syndrome-diagnosis-stirs-concerns
DeLong, G. R., & Dwyer, J. T. (1988). Correlation of family history with specific autistic
subgroups: Asperger’s syndrome and bipolar affective disease. Journal of Autism and
Developmental Disorders, 18, 593–600.
Developmental Disabilities Monitoring Network Surveillance. (2014). Prevalence of autism
spectrum disorder among children aged 8 years - autism and developmental disabilities
monitoring network, 11 sites, Unites States, 2010. MMWR Surveillance Summary, 63 (2):
1-21.
DiCicco-Bloom, E., Lord, C., Zwaigenbaum, L., Courchesne, E., Dager, S. R., Schmitz, C., …
Young, L. J. (2006). The developmental neurobiology of autism spectrum disorder. The
Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 26, 6897–
6906.
Dinnissen, M., Dietrich, A., van den Hoofdakker, B. J., & Hoekstra, P. J. (2015). Clinical and
pharmacokinetic evaluation of risperidone for the management of autism spectrum
disorder. Expert Opinion on Drug Metabolism & Toxicology, 11, 111–124.
Dworzynski, K., Happe, F., Bolton, P., & Ronald, A. (2009). Relationship Between Symptom
Domains in Autism Spectrum Disorders: A Population Based Twin Study. Journal of
Autism and Developmental Disorders, 39, 1197–1210.
Dworzynski, K., Ronald, A., Bolton, P., & Happé, F. (2012). How different are girls and boys
above and below the diagnostic threshold for autism spectrum disorders? Journal of the
American Academy of Child and Adolescent Psychiatry, 51, 788–797.
Edelson, L. R., & Saudino, K. J. (2009). Genetic and Environmental Influences on Autistic-Like
Behaviors in 2-Year-Old Twins. Behavior Genetics, 39, 255–264.
Eisenmajer, R., Prior, M., Leekam, S., Wing, L., Gould, J., Welham, M., Ong, B. (1996).
Comparison of clinical symptoms in autism and Asperger’s disorder. Journal of the
American Academy of Child and Adolescent Psychiatry, 35, 1523-1531.
Elsabbagh, M., Divan, G., Koh, Y.-J., Kim, Y. S., Kauchali, S., Marcín, C., … Fombonne, E.
(2012). Global Prevalence of Autism and Other Pervasive Developmental Disorders.
Autism Research, 5, 160–179.
Falconer, D. S. (1965). The inheritance of liability to certain diseases, estimated from the
incidence among relatives. Annals of Human Genetics, 29, 51–76.
186
Findon, J., Cadman, T., Stewart, C., Woodhouse, E., Eklund, H., Hayward, H., Golden, D.,
Chaplin, E., Glaser, K., Simonoff, E., Murphy, D., Bolton, P, & McEwen, F. (2015).
Screening for co-occuring conditions in adults with autism spectrum disorder using the
Strengths and Difficulties Questionnaire, Autism Research, under review.
Feinstein, A. R. (1970). The pre-therapeutic classification of co-morbidity in chronic disease.
Journal of Chronic Diseases, 23:7, 455-468.
Flouri, E., Midouhas, E., Charman, T., & Sarmadi, Z. (2015). Poverty and the Growth of
Emotional and Conduct Problems in Children with Autism With and Without Comorbid
ADHD. Journal of Autism and Developmental Disorders, Epub ahead.
Folstein, S., & Rutter, M. (1977). Infantile autism: a genetic study of 21 twin pairs. Journal of
child psychology and psychiatry, and allied disciplines, 297–321.
Fombonne, E. (2006). Epidemiology of Pervasive Developmental Disorders. In New
Developments in Autism, The Future is Today (pp. 14–27).
Fombonne, E. (2009). Epidemiology of pervasive developmental disorders. Pediatric Research,
65(6), 591-598.
Frazier, T. W., Ratliff, K. R., Gruber, C., Zhang, Y., Law, P. a, & Constantino, J. N. (2014).
Confirmatory factor analytic structure and measurement invariance of quantitative
autistic traits measured by the Social Responsiveness Scale-2. Autism : The International
Journal of Research and Practice, 18, 31–44.
Frazier, T. W., Thompson, L., Youngstrom, E. A., Law, P., Hardan, A. Y., Eng, C., & Morris, N.
(2014). A Twin Study of Heritable and Shared Environmental Contributions to Autism.
Journal of Autism and Developmental Disorders, pp. 1–13.
Gardener, H., Spiegelman, D., & Buka, S. L. (2009). Prenatal risk factors for autism:
comprehensive meta-analysis. The British Journal of Psychiatry : The Journal of Mental
Science, 195, 7–14.
Gaugler, T., Klei, L., Sanders, S. J., Bodea, C. a, Goldberg, A. P., Lee, A. B., … Buxbaum, J. D.
(2014). Most genetic risk for autism resides with common variation. Nature Genetics, 46,
881–885.
Gerdts, J., & Bernier, R. (2011). The Broader Autism Phenotype and Its Implications on the
Etiology and Treatment of Autism Spectrum Disorders. Autism Research and Treatment.
Geschwind, D. H. (2011). Genetics of autism spectrum disorders. Trends in Cognitive Sciences,
15(9), 409-416.
Geschwind, D. H., Sowinski, J., Lord, C., Iversen, P., Shestack, J., Jones, P., … Spence, S. J.
(2001). The autism genetic resource exchange: a resource for the study of autism and
related neuropsychiatric conditions. American Journal of Human Genetics, 69(2), 463-466.
Geschwind, D. H., State, M. (2015). Gene hunting in autism spectrum disorder: on the path to
precision medicine. The Lancet Neurology, 14(11), 1109-1120.
187
Ghaziuddin, M., Tsai, L., Ghaziuddin, N. (1992b). Brief report: a comparison of the diagnostic
criteria for Asperger syndrome. Journal of Autism and Developmental Disorders, 22, 643649.
Gillberg, C. L. (1984). Autistic-Children Growing Up - Problems During Puberty And
Adolescence. Developmental Medicine and Child Neurology, 26, 125–129.
Gillberg, C. L. (1985). Asperger’s syndrome and recurrent psychosis--a case study. Journal of
Autism and Developmental Disorders, 389–397.
Gillberg, C. L. (1992). Autism and autistic-like conditions: Subclasses among disorders of
empathy. In Journal of Child Psychology and Psychiatry and Allied Disciplines (Vol. 33, pp.
813–842).
Gillberg, C. L., & Schaumann, H. (1981). Infantile Autism and Puberty. Journal of Autism and
Developmental Disorders, 365–371.
Gillespie, N. A., Zhu, G., Neale, M. C., Heath, A. C., & Martin, N. G. (2003). Direction of
causation modeling between cross-sectional measures of parenting and psychological
distress in female twins. Behavior Genetics, 33, 383–396.
Glasson, E. J., Bower, C., Petterson, B., de Klerk, N., Chaney, G., & Hallmayer, J. F. (2004).
Perinatal factors and the development of autism - A population study. Archives of
General Psychiatry, 61, 618–627.
Goldsmith, H. H. (1991). A zygosity questionnaire for young twins: a research note. Behavior
Genetics, 21, 257–269.
Goodman, A., & Goodman, R. (2009). Strengths and Difficulties Questionnaire as a
Dimensional Measure of Child Mental Health. Journal of the American Academy of Child
and Adolescent Psychiatry, 48, 400–403.
Goodman, R. (1997). The strengths and difficulties questionnaire: A research note. Journal of
Child Psychology and Psychiatry and Allied Disciplines, 38, 581–586.
Goodman, R. (2001). Psychometric properties of the strengths and difficulties questionnaire.
Journal of the American Academy of Child and Adolescent Psychiatry, 40, 1337–1345.
Goodman, R., Ford, T., Corbin, T., & Meltzer, H. (2004). Using the Strengths and Difficulties
Questionnaire (SDQ) multi-informant algorithm to screen looked-after children for
psychiatric disorders. European Child & Adolescent Psychiatry, 13, 25–31.
Goodman, R., Ford, T., Richards, H., Gatward, R., & Meltzer, H. (2000). The Development and
Well-Being Assessment: Description and Initial Validation of an Integrated Assessment
of Child and Adolescent Psychopathology. Journal of Child Psychology and Psychiatry, 41,
645–655.
Goodman, R., Renfrew, D., & Mullick, M. (2000). Predicting type of psychiatric disorder from
Strengths and Difficulties Questionnaire (SDQ) scores in child mental health clinics in
London and Dhaka. European Child & Adolescent Psychiatry, 9, 129–134.
Grabrucker, A. M. (2013). Environmental factors in autism. Frontiers in Psychiatry, 3:118.
188
Greenberg, D. a, Hodge, S. E., Sowinski, J., & Nicoll, D. (2001). Excess of twins among affected
sibling pairs with autism: implications for the etiology of autism. American Journal of
Human Genetics, 69(5), 1062–1067.
Gregory, A. M., & Eley, T. C. (2007). Genetic influences on anxiety in children: What we’ve
learned and where we're heading. Clinical Child and Family Psychology Review, 10, 199–
212.
Grzadzinski, R., Huerta, M., & Lord, C. (2013). DSM-5 and autism spectrum disorders (ASDs):
an opportunity for identifying ASD subtypes. Molecular Autism, 4, 12.
Gusella, J. F., Wexler, N. S., Conneally, P. M., Naylor, S. L., Anderson, M. A., Tanzi, R. E., …
Martin, J. B. (1983). A Polymorphic DNA Marker Genetically Linked to HuntingtonsDisease. Nature, 306, 234–238.
Hallett, V., Ronald, A., & Happé, F. (2009). Investigating the association between autistic-like
and internalizing traits in a community-based twin sample. Journal of the American
Academy of Child and Adolescent Psychiatry, 48, 618–627.
Hallett, V., Ronald, A., Rijsdijk, F., & Happé, F. (2010). Association of autistic-like and
internalizing traits during childhood: a longitudinal twin study. The American Journal of
Psychiatry, 167, 809–817.
Hallett, V., Ronald, A., Rijsdijk, F., & Happé, F. (2012). Disentangling the associations between
autistic-like and internalizing traits: A community based twin study. Journal of Abnormal
Child Psychology, 40, 815–827.
Hallmayer, J., Cleveland, S., Torres, A., Phillips, J., Cohen, B., Torigoe, T., … Risch, N. (2011).
Genetic Heritability and Shared Environmental Factors Among Twin Pairs With Autism.
Archives of General Psychiatry, 68(11), 1095-102.
Hallmayer, J., Glasson, E. J., Bower, C., Petterson, B., Croen, L., Grether, J., & Risch, N. (2002).
On the twin risk in autism. American Journal of Human Genetics, 71, 941–946.
Happe, F., Ronald, A., Plomin, R. (2006). Time to give up on a single explanation for autism.
Nature Neuroscience, 9, 1218-1220.
Haworth, C. M. A., Davis, O. S. P., & Plomin, R. (2012). Twins Early Development Study
(TEDS): A Genetically Sensitive Investigation of Cognitive and Behavioral Development
From Childhood to Young Adulthood. Twin Research and Human Genetics, 16(1), 117-25.
Heath, A. C., Kessler, R. C., Neale, M. C., Hewitt, J. K., Eaves, L. J., & Kendler, K. S. (1993).
Testing hypotheses about direction of causation using cross-sectional family data.
Behavior Genetics, 23, 29–50.
Heron, J., Golding, J., & Team, A. S. (2004). Thimerosal exposure in infants and
developmental disorders: A prospective cohort study in the United Kingdom does not
support a causal association. Pediatrics, 114, 577–583.
Hettema, J. M., Neale, M. C., & Kendler, K. S. (1995). Physical similarity and the equalenvironment assumption in twin studies of psychiatric disorders. Behavior Genetics, 25,
327–335.
189
Hoekstra, R. A., Bartels, M., Cath, D. C., & Boomsma, D. I. (2008). Factor structure, reliability
and criterion validity of the autism-spectrum quotient (AQ): A study in Dutch population
and patient groups. Journal of Autism and Developmental Disorders, 38, 1555–1566.
Hoekstra, R. A., Bartels, M., Verweij, C. J. H., & Boomsma, D. I. (2007). Heritability of autistic
traits in the general population. Archives of Pediatrics & Adolescent Medicine, 161, 372–
377.
Ho, A., Todd, R. D., Constantino, J. N. (2005. Brief report: autistic traits in twins vs. non-twins
– a preliminary study. Journal of Autism and Developmental Disorders, 35: 129-133.
Hoekstra, R. A., Bartels, M., Hudziak, J. J., Van Beijsterveldt, T. C., Boomosma, D. I. (2007).
Genetic and environmental covariation between austistic traits and behavioural
problems. Twin Research and Human Genetics, 10(6): 853-60.
Hoekstra, R. A., Bartels, M., Verweij, C. J. H., Boomsma, D. I. (2007a). Heritability of autistic
traits in the general population. Archives of Pediatric Adolescent Medicine, 161: 372-377.
Hoekstra, R. A., Happé, F., Baron-Cohen, S., Ronald, A. (2010). Limited genetic covariance
between autistic traits and intelligence: findings from a longitudinal twin study.
American Journal of Medical Genetics Part B: Neuropsychiatric Genetics, 135B(5): 9941007.
Hoekstra, R. A., Happé, F., Baron-Cohen, S., Ronald, A. (2009). Association between extreme
autistic traits and intellectual disability: insights from a general population twin study.
The British Journal of Psychiatry, 195(6): 531-536.
Holmes, S. J., & Robins, L. N. (1987). The Influence of Childhood Disciplinary Experience on
the Development of Alcoholism and Depression. Journal of Child Psychology and
Psychiatry, 28, 399–415.
Honda, H., Shimizu, Y., & Rutter, M. (2005). No effect of MMR withdrawal on the incidence of
autism: a total population study. Journal of Child Psychology and Psychiatry, 46, 572–579.
Hultman, C. M., Sparen, P., & Cnattinguis, S. (2002). Perinatal risk factors for infantile autism.
Epidemiology, 13(4): 417-423
Hurley, R. S. E., Losh, M., Parlier, M., Reznick, J. S., & Piven, J. (2007). The broad autism
phenotype questionnaire. Journal of Autism and Developmental Disorders, 37, 1679–1690.
Iossifov, I., Ronemus, M., Levy, D., Wang, Z., Hakker, I., Rosenbaum, J., … Wigler, M. (2012).
De Novo Gene Disruptions in Children on the Autistic Spectrum. Neuron, 74, 285–299.
Iossifov, I., Levy, D., Allen, J., Ye, K., Ronemus, M., Lee, Y., Yamrom, B., Wigler, M. (2015).
Low load for disruptive mutations in autism genes and their biased transmission. PNAS,
E5600–E5607, doi: 10.1073/pnas.1516376112.
Jaffee, S. R., Hanscombe, K. B., Haworth, C. M. A., Davis, O. S. P., & Plomin, R. (2012). Chaotic
Homes and Children’s Disruptive Behavior: A Longitudinal Cross-Lagged Twin Study.
Psychological Science, 23(6), 643-50.
Johannsen, W. (1911). The genotype conception of heredity. American Naturalist, 45, 129–159.
190
Jones, A. P., Larsson, H., Ronald, A., Rijsdijk, F. V, Busfield, P., McMillan, A., … Viding, E.
(2009). Phenotypic and aetiological relationships between psychopathic tendencies,
autistic traits, and emotion attribution. Criminal Justice And Behavior, 36(11), 1198-1212.
Kanner, L. (1943). Child psychiatry - Mental deficiency. American Journal of Psychiatry, 99,
608–610.
Kanner, L. (1949). Problems of Nosology and Psychodynamics of Early Infantile Autism.
American Journal of Orthopsychiatry, 19, 416–426.
Kanner, L. (1968). Autistic disturbances of affective contact. Acta Paedopsychiatr, 35, 100–136.
Kaufmann, W. E. (2012). DSM-5: The New Diagnostic Criteria For Autism Spectrum Disorders.
http://www.autismconsortium.org/symposiumfiles/WalterKaufmannAC2012Symposium.pdf
Kaye, J. A., Melero-Montes, M. D., & Jick, H. (2001). Mumps, measles, and rubella vaccine and
the incidence of autism recorded by general practitioners: a time trend analysis. British
Medical Journal, 322, 460–463.
Kendler, K. S. (2005). A gene for.{’'}: The nature of gene action in psychiatric disorders.
American Journal of Psychiatry, 162, 1243–1252.
Kendler, K. S., Kessler, R. C., Waiters, E. E., MacLean, C., Neale, M. C., Heath, A. C., & Eaves,
L. J. (1995). Stressful life events, genetic liability, and onset of an episode of major
depression in women. American Journal of Psychiatry, 152, 833–842.
Kendler, K. S., Neale, M. C., Kessler, R. C., Heath, A. C., & Eaves, L. J. (1992). Major depression
and generalized anxiety disorder. Same genes, (partly) different environments? Archives
of General Psychiatry, 49, 716–722.
Kendler, K. S., Neale, M. C., Kessler, R. C., Heath, A. C., & Eaves, L. J. (1993). A test of the
equal-environment assumption in twin studies of psychiatric illness. Behavior Genetics,
23, 21–27.
Kendler, K. S., Pedersen, N. L., Neale, M. C., & Mathe, A. A. (1995). A Pilot Swedish Twin
Study of Affective-Illness including Hospital-Ascertained and Population-Ascertained
Subsamples - Results of Model-Fitting. Behavior Genetics, 25, 217–232.
Kerekes, N., Lundström, S., Chang, Z., Tajnia, A., Jern, P., Lichtenstein, P., … Anckarsäter, H.
(2014). Oppositional defiant- and conduct disorder-like problems: neurodevelopmental
predictors and genetic background in boys and girls, in a nationwide twin study. PeerJ, 2,
e359.
Kessler, R. C., McGonagle, K. A., Zhao, S., Nelson, C. B., Hughes, M., Eshleman, S., … Kendler,
K. S. (1994). Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the
United States. Results from the National Comorbidity Survey. Archives of General
Psychiatry, 51, 8–19.
Kim, Y. S., Leventhal, B. L., Koh, Y.-J., Fombonne, E., Laska, E., Lim, E.-C., … Grinker, R. R.
(2011). Prevalence of autism spectrum disorders in a total population sample. The
American Journal of Psychiatry, 168, 904–912.
191
Kim, Y.-K. (2010). Handbook of behavior genetics (1st softco.). Springer: New York.
Klei, L., Sanders, S. J., Murtha, M. T., Hus, V., Lowe, J. K., Willsey, A. J., … Devlin, B. (2012).
Common genetic variants, acting additively, are a major source of risk for autism.
Molecular Autism, 3:9.
Klein, D. N. & R. L. P. (1993). Psychiatric disorders: Problems of boundaries and comorbidity.
In In C. G. Costello (Ed.), Basic issues in psychopathology (pp. 19–66).
Kolevzon, A., Gross, R., & Reichenberg, A. (2007). Prenatal and perinatal risk factors for
autism - A review and integration of findings. Archives of Pediatrics & Adolescent
Medicine, 161, 326–333.
Kolvin, I. (1971). Studies in the Childhood Psychoses: Diagnostic Criteria and Classification.
The British Journal of Psychiatry, 118, 381–384.
Kolvin, I., Ounsted, C., Humphrey, M., & McNay, A. (1971). The Phenomenology of Childhood
Psychoses. The British Journal of Psychiatry, 118, 385–395.
Kong, A., Frigge, M. L., Masson, G., Besenbacher, S., Sulem, P., Magnusson, G., … Stefansson,
K. (2012). Rate of de novo mutations and the importance of father’s age to disease risk.
Nature, 488, 471–475.
Kotlicka-Antczak, M., Pawelczyk, A., Rabe-Jablonska, J., Smigielski, J., & Pawelczyk, T.
(2014). Obstetrical complications and Apgar score in subjects at risk of psychosis. Journal
of Psychiatric Research, 48, 79–85.
Krakowiak, P., Walker, C. K., Bremer, A. A., Baker, A. S., Ozonoff, S., Hansen, R. L., & HertzPicciotto, I. (2012). Maternal Metabolic Conditions and Risk for Autism and Other
Neurodevelopmental Disorders. Pediatrics, 129(5), e1121–e1128.
Kwok, P. Y. (2000). Approaches to allele frequency determination. Pharmacogenomics, 1, 231–
235.
Lai, M.-C., Lombardo, M. V, & Baron-Cohen, S. (2014). Autism. Lancet, 383, 896–910.
Lai, M.-C., Lombardo, M. V, Chakrabarti, B., & Baron-Cohen, S. (2013). Subgrouping the
Autism “Spectrum”: Reflections on DSM-5. PLoS Biol, 11, e1001544.
Lainhart, J. E. (1999). Psychiatric problems in individuals with autism, their parents and
siblings. International Review of Psychiatry, 11, 278–298.
Lainhart, J. E., & Folstein, S. E. (1994). Affective disorders in people with autism: A review of
published cases. Journal of Autism and Developmental Disorders.
Landa, R. J. (2008). Diagnosis of autism spectrum disorders in the first 3 years of life. Nature
Clinical Practice Neurology, 4, 138–147.
Landa, R., Piven, J., Wzorek, M. M., Gayle, J. O., Chase, G. A., & Folstein, S. E. (1992). Social
Language Use in Parents of Autistic Individuals. Psychological Medicine, 22, 245–254.
Landrigan, P. J. (2010). What causes autism? Exploring the environmental contribution.
Current Opinion in Pediatrics, 22, 219–225.
192
Larson, T., Anckarsäter, H., Gillberg, C., Ståhlberg, O., Carlström, E., Kadesjö, B., … Gillberg,
C. (2010). The autism--tics, AD/HD and other comorbidities inventory (A-TAC): further
validation of a telephone interview for epidemiological research. BMC Psychiatry, 10, 1.
Le Couteur, A., Bailey, A., Goode, S., Pickles, A., Gottesman, I., Robertson, S., & Rutter, M.
(1996). A broader phenotype of autism: The clinical spectrum in twins. Journal of Child
Psychology and Psychiatry and Allied Disciplines, 37, 785–801.
Leblond, C. S., Nava, C., Polge, A., Gauthier, J., Huguet, G., Lumbroso, S., … Bourgeron, T.
(2014). Meta-analysis of SHANK Mutations in Autism Spectrum Disorders: A Gradient of
Severity in Cognitive Impairments. PLOS GENETICS, 10(9):e1004580.
Levin, E. D., Addy, N., Nakajima, A., Christopher, N. C., Seidler, F. J., & Slotkin, T. A. (2001).
Persistent behavioral consequences of neonatal chlorpyrifos exposure in rats.
Developmental Brain Research, 130, 83–89.
Leyfer, O. T., Folstein, S. E., Bacalman, S., Davis, N. O., Dinh, E., Morgan, J., … Lainhart, J. E.
(2006). Comorbid psychiatric disorders in children with autism: Interview development
and rates of disorders. Journal of Autism and Developmental Disorders, 36, 849–861.
Lichtenstein, P., Carlström, E., Råstam, M., Gillberg, C., & Anckarsäter, H. (2010). The
genetics of autism spectrum disorders and related neuropsychiatric disorders in
childhood. The American Journal of Psychiatry, 167, 1357–1363.
Lim, E. T., Raychaudhuri, S., Sanders, S. J., Stevens, C., Sabo, A., MacArthur, D. G., … Project,
N. E. S. (2013). Rare Complete Knockouts in Humans: Population Distribution and
Significant Role in Autism Spectrum Disorders. Neuron, 77, 235–242.
Limperopoulos, C., Bassan, H., Gauvreau, K., Robertson, R. L., Sullivan, N. R., Benson, C. B., …
duPlessis, A. J. (2007). Does cerebellar injury in premature infants contribute to the high
prevalence of long-term cognitive, learning, and behavioral disability in survivors?
Pediatrics, 120, 584–593.
Loehlin, J. C. (1996). The Cholesky approach: A cautionary note. Behavior Genetics, 26, 65–69.
Loehlin, J. C., & Nicholls, J. (1976). Heredity, environment and personality. Austin: University of
Texas.
Lord, C., Risi, S., Lambrecht, L., Cook, E. H., Leventhal, B. L., Dilavore, P. C., … Rutter, M.
(2000). The Autism Diagnostic Observation Schedule-Generic: A standard measure of
social and communication deficits associated with the spectrum of autism. Journal of
Autism and Developmental Disorders, 30, 205–223.
Lord, C., Rutter, M., & Couteur, A. L. (1994). Autism diagnostic interview-revised: A revised
version of a diagnostic interview for caregivers of individuals with possible pervasive
developmental disorders. Journal of Autism and Developmental Disorders, 24, 659–685.
Lord, C., Rutter, M., Goode, S., Heemsbergen, J., Jordan, H., Mawhood, L., & Schopler, E.
(1989). Austism diagnostic observation schedule: A standardized observation of
communicative and social behavior. Journal of Autism and Developmental Disorders, 19,
185–212.
193
Lundström, S., Chang, Z., Kerekes, N., Gumpert, C. H., Råstam, M., Gillberg, C., …
Anckarsäter, H. (2011). Autistic-like traits and their association with mental health
problems in two nationwide twin cohorts of children and adults. Psychological Medicine,
41(11), 2423-33.
Lundström, S., Chang, Z., Rastam, M., Gillberg, C., Larsson, H., Anckarsater, H., &
Lichtenstein, P. (2012). Autism Spectrum Disorders and Autisticlike Traits: Similar
Etiology in the Extreme End and the Normal Variation. Archives of General Psychiatry,
69(1), 46-52.
Lundström, S., Reichenberg, A., Melke, J., Råstam, M., Kerekes, N., Lichtenstein, P., …
Anckarsäter, H. (2014). Autism spectrum disorders and coexisting disorders in a
nationwide Swedish twin study. Journal of Child Psychology and Psychiatry, 56(6), 702-10.
Madsen, K. M., Hviid, A., Vestergaard, M., Schendel, D., Wohlfahrt, J., Thorsen, P., … Melbye,
M. (2002). A population-based study of measles, mumps, and rubella vaccination and
autism. New England Journal of Medicine, 347, 1477–1482.
Maj, M. (2005). `Psychiatric comorbidity’: an artefact of current diagnostic systems? British
Journal of Psychiatry, 186, 182–184.
Makela, A., Nuorti, J. P., & Peltola, H. (2002). Neurologic disorders after measles-mumpsrubella vaccination. Pediatrics, 110, 957–963.
Mandy, W., Charman, T., Puura, K., & Skuse, D. (2014). Investigating the cross-cultural validity
of DSM-5 autism spectrum disorder: evidence from Finnish and UK samples. Autism : The
International Journal of Research and Practice, 18, 45–54.
Mandy, W. P. L., Charman, T., & Skuse, D. H. (2012). Testing the construct validity of
proposed criteria for DSM-5 autism spectrum disorder. Journal of the American Academy
of Child and Adolescent Psychiatry, 51, 41–50.
Manjiviona, J., Prior, M. (1995). Comparison of Asperger syndrome and high-functioning
autistic children on a test of motor impairment. Journal of Autism and Developmental
Disorders, 25, 25-39.
Matson, J. L., & Kozlowski, A. M. (2011). The increasing prevalence of autism spectrum
disorders. Research in Autism Spectrum Disorders, 5(1), 418-425.
Matson, J. L., Mayville, E. A., Lott, J. D., Bielecki, J., & Logan, R. (2003). A comparison of social
and adaptive functioning in persons with psychosis, autism, and severe or profound
mental retardation. Journal of Developmental and Physical Disabilities, 15, 57–65.
Maxwell, C. R., Parish-Morris, J., Hsin, O., Bush, J. C., & Schultz, R. T. (2013). The broad autism
phenotype predicts child functioning in autism spectrum disorders. Journal of
Neurodevelopmental Disorders, 5, 25.
Mayes, S. D., Calhoun, S. L., Crites, D. L. (2001). Does DSM IV Asperger’s disorder exist?
Journal of Abnormal Child Psychology, 29(3): 263-271.
Mazzone, L., Ruta, L., & Reale, L. (2012). Psychiatric comorbidities in asperger syndrome and
high functioning autism: diagnostic challenges. Annals of General Psychiatry, 11, 16.
194
McPartland, J. C., Reichow, B., Volkmar, F. R. (2012). Sensitivity and specificity of proposed
DSM-5 diagnostic criteria for autism spectrum disorder. Journal of the American Academy
of Child and Adolescent Psychiatry, 51(4): 368-383.
McGue, M. (1992). When assessing twin concordance, use the probandwise not the pairwise
rate. Schizophrenia Bulletin, 18, 171–176.
McGue, M., & Bouchard, T. J. (1984). Adjustment of Twin Data for the Effects of Age and Sex.
Behavior Genetics, 14, 325–343.
Meltzer, H., Gatward, R., Goodman, R., & Ford, T. (2000). Mental health of children and
adolescents in Great Britain. London: The Stationary Office.
Merwood, a, Greven, C. U., Price, T. S., Rijsdijk, F., Kuntsi, J., McLoughlin, G., … Asherson, P.
J. (2013). Different heritabilities but shared etiological influences for parent, teacher and
self-ratings of ADHD symptoms: an adolescent twin study. Psychological Medicine, 43,
1973–84.
Michaelson, J. J., Shi, Y., Gujral, M., Zheng, H., Malhotra, D., Jin, X., … Sebat, J. (2012). WholeGenome Sequencing in Autism Identifies Hot Spots for De Novo Germline Mutation.
Cell, 151, 1431–1442.
Miller, M. T., & Stromland, K. K. (2011). What can we learn from the thalidomide experience:
an ophthalmologic perspective. Current Opinion on Ophthalmology, 22, 356–364.
Moffitt, S. (2011). Conduct Disorder vs. Autism: Identifying the Differences. Retrieved from
http://www.autismkey.com/conduct-disorder-vs-autism-identifying-the-differences/
Moore, S. J., Turnpenny, P., Quinn, A., Glover, S., Lloyd, D. J., Montgomery, T., & Dean, J. C.
S. (2000). A clinical study of 57 children with fetal anticonvulsant syndromes. Journal of
Medical Genetics, 37, 489–497.
Moore, V., & Goodson, S. (2003). How Well Does Early Diagnosis of Autism Stand the Test of
Time? Autism, 7, 47 –63.
Morton, N. E. (1959). Genetic Tests Under Incomplete Ascertainment. American Journal of
Human Genetics, 11, 1–16.
Neale, B. M., Kou, Y., Liu, L., Ma’ayan, A., Samocha, K. E., Sabo, A., … Daly, M. J. (2012).
Patterns and rates of exonic de novo mutations in autism spectrum disorders. Nature,
485, 242–U129.
Neale, M. C., Boker, S. M., Xie, G., & Maes, H. H. (2003). Mx: Statistical Modeling. VCU Box
900126, Richmond, VA 23298: Department of Psychiatry.
Neale, M. C., & Kendler, K. S. (1995). Models of comorbidity for multifactorial disorders.
American Journal of Human Genetics, 57, 935–953.
Neale, M. C., & Maes, H. M. (2001). Methodology for Genetic Studies of Twins and Families.
Dodrecht, The Netherlands: Kluwer Academic Publishers B. V.
195
Neale, M. C., Røysamb, E., & Jacobson, K. (2006). Multivariate genetic analysis of sex
limitation and G x E interaction. Twin Research and Human Genetics : The Official Journal
of the International Society for Twin Studies, 9, 481–489.
Neale, M.C., & Cardon, L. R. (1992). Methodology for genetic studies of twins and families.
Dordrecht: Kluwer Academic Publishers.
Newschaffer, C. J., Croen, L. A., Daniels, J., Giarelli, E., Grether, J. K., Levy, S. E., … Windham,
G. C. (2007). The epidemiology of autism spectrum disorders. Annual Review of Public
Health, 28, 235–258.
Nordenbæk, C., Jørgensen, M., Kyvik, K. O., & Bilenberg, N. (2014). A Danish populationbased twin study on autism spectrum disorders. European Child and Adolescent
Psychiatry, 23, 35–43.
O’Nions, E., Tick, B., Rijsdijk, F., Happe, F., Plomin, R., Ronald, A., & Viding, E. (2015).
Examining the genetic and environmental associations between autistic social and
communication deficits and psychopathic callous-unemotional traits. Plos One, 10(9):
e0134331.
O’Roak, B. J., Vives, L., Girirajan, S., Karakoc, E., Krumm, N., Coe, B. P., … Eichler, E. E. (2012).
Sporadic autism exomes reveal a highly interconnected protein network of de novo
mutations. Nature, 485, 246–U136.
Owens, S. F., Picchioni, M. M., Ettinger, U., McDonald, C., Walshe, M., Schmechtig, A., …
Toulopoulou, T. (2012). Prefrontal deviations in function but not volume are putative
endophenotypes for schizophrenia. Brain, 135, 2231–2244.
Pearson, K., & Lee, A. (1900). Mathematical contributions to the theory of evolution VII - On
the application of certain formulae in the theory of correlation to the inheritance of
characters not capable of quantitative measurement. Proceedings of the Royal Society of
London, 66, 324–327.
Pelaez-Nogueras, M. (1996). Multiple influences in behavioural interactions. Behavioural
Developmental Bulleting, 2, 10-14.
Pincus, H. A., Tew, J. D., & First, M. B. (2004). Psychiatric comorbidity: is more less? World
Psychiatry : Official Journal of the World Psychiatric Association (WPA), 3, 18–23.
Pinto, D., Delaby, E., Merico, D., Barbosa, M., Merikangas, A., Klei, L., … Scherer, S. W. (2014).
Convergence of Genes and Cellular Pathways Dysregulated in Autism Spectrum
Disorders. American Journal of Human Genetics, 94, 677–694.
Pinto, D. et al. (2010). Functional impact of global rare copy number variation in autism
spectrum disorders. Nature, 466, 368–372.
Piven, J., Gayle, J., Chase, G. A., Fink, B., Landa, R., Wzorek, M. M., & Folstein, S. E. (1990). A
Family History Study of Neuropsychiatric Disorders in the Adult Siblings of Autistic
Individuals. Journal of the American Academy of Child And Adolescent Psychiatry, 29, 177–
183.
Plomin, R., DeFries, J. C., Knopik, V. S., & Neiderhiser, J. M. (2013). Behavioral Genetics (6th
Edition.). New York: Worth Publishers.
196
Polderman, T. J. C., Hoekstra, R. A., Vinkhuyzen, A. A. E., Sullivan, P. F., van der Sluis, S.,
Posthuma, D. (2013). Attentional switching forms a genetic link between attention
problems and autistic traits in adults. Psychological Medicine, 43: 1985-1996.
Pollmann, M. M. H., Finkenauer, C., & Begeer, S. (2010). Mediators of the link between autistic
traits and relationship satisfaction in a non-clinical sample. Journal of Autism and
Developmental Disorders, 40(4), 470–478.
Posserud, M. B., Lundervold, A. J., & Gillberg, C. (2006). Autistic features in a total population
of 7-9-year-old children assessed by the ASSQ (Autism Spectrum Screening
Questionnaire). Journal Of Child Psychology And Psychiatry, 47, 167–175.
Price, T. S., Freeman, B., Craig, I., Petrill, S. A., Ebersole, L., & Plomin, R. (2000). Infant
zygosity can be assigned by parental report questionnaire data. Twin Research : The
Official Journal of the International Society for Twin Studies, 3, 129–133.
Prior, M., Eisenmajer, R., Leekam, S., Wing, L., Gould, J., Ong, B., & Dowe, D. (1998). Are
there subgroups within the autistic spectrum? A cluster analysis of a group of children
with autistic spectrum disorders. Journal of Child Psychology and Psychiatry and Allied
Disciplines, 39, 893–902.
Pugliese, C. E., White, B. A., White, S. W., & Ollendick, T. H. (2013). Social anxiety predicts
aggression in children with ASD: Clinical comparisons with socially anxious and
oppositional youth. Journal of Autism and Developmental Disorders, 43, 1205–1213.
Raftery, A. E. (1995). Bayesian Model Selection in Social Research. Sociological Methodology,
25, 111–163.
Rauh, V. A., Garfinkel, R., Perera, F. P., Andrews, H. F., Hoepner, L., Barr, D. B., … Whyatt, R.
W. (2006). Impact of prenatal chlorpyrifos exposure on neurodevelopment in the first 3
years of life among inner-city children. Pediatrics, 118, e1845–e1859.
Reichenberg, A., Gross, R., Weiser, M., Bresnahan, M., Silverman, J., Harlap, S., … Susser, E.
(2006). Advancing paternal age and autism. Archives of General Psychiatry, 63, 1026–
1032.
Reiersen, A. M., Constantino, J. N., Grimmer, M., Martin, N. G., & Todd, R. D. (2008). Evidence
for shared genetic influences on self-reported ADHD and autistic symptoms in young
adult Australian twins. Twin Research and Human Genetics : The Official Journal of the
International Society for Twin Studies, 11, 579–585.
Rieffe, C., Oosterveld, P., Terwogt, M. M., Mootz, S., van Leeuwen, E., & Stockmann, L.
(2011). Emotion regulation and internalizing symptoms in children with autism spectrum
disorders. Autism : The International Journal of Research and Practice, 15, 655–70.
Rijsdijk, F. V, & Sham, P. C. (2002). Analytic approaches to twin data using structural equation
models. Briefings in Bioinformatics, 3, 119–133.
Rijsdijk, F. V, van Haren, N. E. M., Picchioni, M. M., McDonald, C., Toulopoulou, T., Hulshoff
Pol, H. E., … Sham, P. C. (2005). Brain MRI abnormalities in schizophrenia: same genes or
same environment? Psychological Medicine, 35, 1399–1409.
197
Ripke, S., Neale, B. M., Corvin, A., Walters, J. T. R., Farh, K.-H., Holmans, P. A., … Consor, W.
T. C.-C. (2014). Biological insights from 108 schizophrenia-associated genetic loci.
Nature, 511, 421+.
Ritvo, E. R., Freeman, B. J., Mason-Brothers, A., Mo, A., & Ritvo, A. M. (1985). Concordance
for the syndrome of autism in 40 pairs of afflicted twins. The American Journal of
Psychiatry, 142, 74–77.
Ritvo, E. R., Spence, M. A., Freeman, B. J., Masonbrothers, A., Mo, A., & Marazita, M. L.
(1985). Evidence for Autosomal Recessive Inheritance in 46 Families with Multiple
Incidences of Autism. American Journal of Psychiatry, 142, 187–192.
Robinson, E. B., Koenen, K. C., McCormick, M. C., Munir, K., Hallett, V., Happe, F., … Ronald,
A. (2011). Evidence That Autistic Traits Show the Same Etiology in the General
Population and at the Quantitative Extremes (5%, 2.5%, and 1%). Archives of General
Psychiatry, 68, 1113–1121.
Robinson, E. B., Koenen, K. C., McCormick, M. C., Munir, K., Hallett, V., Happé, F., … Ronald,
A. (2012). A multivariate twin study of autistic traits in 12-year-olds: Testing the
fractionable autism triad hypothesis. Behavior Genetics, 42, 245–255.
Rommelse, N. N., Franke, B., Geurts, H. M., Hartman, C. A., & Buitelaar, J. K. (2010). Shared
heritability of attention-deficit/hyperactivity disorder and autism spectrum disorder. Eur
Child Adolesc Psychiatry, 19, 281–295.
Ronald, A., Edelson, L. R., Asherson, P., & Saudino, K. J. (2010). Exploring the relationship
between autistic-like traits and ADHD behaviors in early childhood: Findings from a
community twin study of 2-year-olds. Journal of Abnormal Child Psychology, 38, 185–196.
Ronald, A., Happe, F., Bolton, P., Butcher, L. M., Price, T. S., Wheelwright, S., … Plomin, R.
(2006). Genetic heterogeneity between the three components of the autism spectrum: A
twin study. Journal of the American Academy of Child and Adolescent Psychiatry, 45, 691–
699.
Ronald, A., Happe, F., & Plomin, R. (2005). The genetic relationship between individual
differences in social and nonsocial behaviours characteristic of autism. Developmental
Science, 8, 444–458.
Ronald, A., Happe, F., & Plomin, R. (2008). A twin study investigating the genetic and
environmental aetiologies of parent, teacher and child ratings of autistic-like traits and
their overlap. European Child & Adolescent Psychiatry, 17, 473–483.
Ronald, A., Happé, F., Price, T. S., Baron-Cohen, S., & Plomin, R. (2006). Phenotypic and
genetic overlap between autistic traits at the extremes of the general population. Journal
of the American Academy of Child and Adolescent Psychiatry, 45, 1206–1214.
Ronald, A., & Hoekstra, R. A. (2011). Autism Spectrum Disorders and Autistic Traits: A Decade
of New Twin Studies. American Journal of Medical Genetics Part B-Neuropsychiatric
Genetics, 156B, 255–274.
Ronald, A., & Hoekstra, R. A. (2014). Progress in Understanding the Causes of Autism
Spectrum Disorders and Autistic Traits: Twin Studies from 1977 to the Present Day.
Behaviour Genetics of Psychopathology, 2, 33–65.
198
Ronald, A., Larsson, H., Anckarsäter, H., & Lichtenstein, P. (2011). A twin study of autism
symptoms in Sweden. Molecular Psychiatry, 16(10), 1039-47.
Ronald, A., Larsson, H., Anckarsäter, H., & Lichtenstein, P. (2014). Symptoms of autism and
ADHD: A swedish twin study examining their overlap. Journal of Abnormal Psychology,
123, 440–451.
Ronald, A., Simonoff, E., Kuntsi, J., Asherson, P., & Plomin, R. (2008). Evidence for
overlapping genetic influences on autistic and ADHD behaviours in a community twin
sample. Journal of Child Psychology and Psychiatry, 49, 535–542.
Ronemus, M., Iossifov, I., Levy, D., Wigler, M. (2014). The role of de novo mutations in the
genetics of autism spectrum disorders. Nature Reviews Genetics, 15, 133-141.
Rosenberg, R. E., Law, J. K., Yenokyan, G., McGready, J., Kaufmann, W. E., & Law, P. a.
(2009a). Characteristics and concordance of autism spectrum disorders among 277 twin
pairs. Archives of Pediatrics & Adolescent Medicine, 163, 907–14.
Rosenberg, R. E., Law, J. K., Yenokyan, G., McGready, J., Kaufmann, W. E., & Law, P. A.
(2009b). Characteristics and concordance of autism spectrum disorders among 277 twin
pairs. Archives of Pediatrics & Adolescent Medicine, 163, 907–914.
Russell, G., Rodgers, L. R., & Ford, T. (2013). The Strengths and Difficulties Questionnaire as a
Predictor of Parent-Reported Diagnosis of Autism Spectrum Disorder and Attention
Deficit Hyperactivity Disorder. Plos One, 8, 9.
Rutter, M. (1970). Autistic Children - Infancy To Adulthood. Seminars In Psychiatry, 2, 435–50.
Rutter, M. (1994). Comorbidity: Meanings and Mechanisms, Clin Psychol Sci Prac, 100-103.
Rutter, M., Moffitt, T. E., & Caspi, A. (2006). Gene-environment interplay and
psychopathology: multiple varieties but real effects. Journal of Child Psychology and
Psychiatry, 47, 226–261.
Salomone, E., Kutlu, B., Derbyshire, K., McCloy, C., Hastings, R. P., Howlin, P., & Charman, T.
(2014). Emotional and behavioural problems in children and young people with autism
spectrum disorder in specialist autism schools. Research in Autism Spectrum Disorders, 8,
661–668.
Sanders, S. J., Murtha, M. T., Gupta, A. R., Murdoch, J. D., Raubeson, M. J., Willsey, A. J., …
State, M. W. (2012). De novo mutations revealed by whole-exome sequencing are
strongly associated with autism. Nature, 485, 237–U124.
Sanders, S. J. et al. (2011). Multiple recurrent de novo CNVs, including duplications of the
7q11.23 Williams syndrome region, are strongly associated with autism. Neuron, 70, 863–
885.
Sandin, S., Hultman, C. M., Kolevzon, A., Gross, R., MacCabe, J. H., & Reichenberg, A. (2012).
Advancing Maternal Age Is Associated With Increasing Risk for Autism: A Review and
Meta-Analysis. Journal of the American Academy of Child And Adolescent Psychiatry, 51,
477–486.
199
Sandin, S., Lichtenstein, P., Kuja-Halkola, R., Larsson, H., Hultman, C. M., & Reichenberg, A.
(2014). The familial risk of autism. JAMA : The Journal of the American Medical
Association, 311, 1770–7.
Sasson, N. J., Lam, K. S. L., Childress, D., Parlier, M., Daniels, J. L., & Piven, J. (2013). The
Broad Autism Phenotype Questionnaire: Prevalence and Diagnostic Classification.
Autism Research, 6, 134–143.
Sasson, N. J., Lam, K. S., Parlier, M., Daniels, J. L., & Piven, J. (2013). Autism and the broad
autism phenotype: familial patterns and intergenerational transmission. Journal of
Neurodevelopmental Disorders, 5, 11.
Scherff, A., Taylor, M., Eley, T. C., Happe, F., Charman, T., & Ronald, A. (2014). What Causes
Internalising Traits and Autistic Traits to Co-occur in Adolescence? A Community-Based
Twin Study. Journal of Abnormal Child Psychology, 42, 601–610.
Scott, F. J., Baron-Cohen, S., Bolton, P., & Brayne, C. (2002). The CAST (Childhood Asperger
Syndrome Test) - Preliminary development of a UK screen for mainstream primaryschool-age children. Autism, 6, 9–31.
Scott, F. J., Baron-Cohen, S., Bolton, P., & Brayne, C. (2002). The CAST (Childhood Asperger
Syndrome Test): preliminary development of a UK screen for mainstream primaryschool-age children. Autism : The International Journal of Research and Practice, 6, 9–31.
Scourfield, J., Van den Bree, M., Martin, N., & McGuffin, P. (2004). Conduct problems in
children and adolescents: a twin study. Archives of General Psychiatry, 61, 489–496.
Shrout, P. E., & Fleiss, J. L. (1979). Intraclass correlations: uses in assessing rater reliability.
Psychological Bulletin, 86, 420–428.
Simonoff, E. (2000). Extracting meaning from comorbidity: genetic analyses that make sense.
Journal of Child Psychology and Psychiatry, and Allied Disciplines, 41, 667–674.
Simonoff, E., Jones, C. R. G., Baird, G., Pickles, A., Happ??, F., & Charman, T. (2013). The
persistence and stability of psychiatric problems in adolescents with autism spectrum
disorders. Journal of Child Psychology and Psychiatry and Allied Disciplines, 54, 186–194.
Simonoff, E., Jones, C. R. G., Baird, G., Pickles, A., Happe, F., & Charman, T. (2013). The
persistence and stability of psychiatric problems in adolescents with autism spectrum
disorders. Journal of Child Psychology and Psychiatry, 54, 186–194.
Simonoff, E., Pickles, A., Charman, T., Chandler, S., Loucas, T., & Baird, G. (2008). Psychiatric
disorders in children with autism spectrum disorders: Prevalence, comorbidity, and
associated factors in a population-derived sample. Journal of the American Academy of
Child and Adolescent Psychiatry, 47, 921–929.
Simonoff, E., Pickles, A., Charman, T., Chandler, S., Loucas, T., & Baird, G. (2008). Psychiatric
disorders in children with autism spectrum disorders: prevalence, comorbidity, and
associated factors in a population-derived sample. Journal of the American Academy of
Child and Adolescent Psychiatry, 47, 921–929.
Skokauskas, N., & Gallagher, L. (2012). Mental health aspects of autistic spectrum disorders in
children. Journal of Intellectual Disability Research, 56, 248–257.
200
Skuse, D. H. (2007). Rethinking the nature of genetic vulnerability to autistic spectrum
disorders. Trends in Genetics, 23(7): 387-395.
Skuse, D. H. (2000). Imprinting, the X-chromosome, and the male brain: explaining sex
differences in the liability to autism. Pediatric Research, 47, 9–16.
Skuse, D. H. (2012). DSM-5’s Conceptualization of Autistic Disorders. Journal of the American
Academy of Child And Adolescent Psychiatry, 51, 344–346.
Skuse, D. H., Mandy, W. P. L., & Scourfield, J. (2005). Measuring autistic traits: heritability,
reliability and validity of the Social and Communication Disorders Checklist. British
Journal of Psychiatry, 187, 568–572.
Skuse, D. H., Mandy, W. P. L., & Scourfield, J. (2005). Measuring autistic traits: heritability,
reliability and validity of the Social and Communication Disorders Checklist. The British
Journal of Psychiatry : The Journal of Mental Science, 187, 568–572.
Skuse, D. (2009). Is autism really a coherent syndrome in boys, or girls? British Journal or
Psychology, 100, 33-37.
Smith, L. E., Greenberg, J. S., Mailick, M. R. (2013). Adults with autism: outcomes, family
effects, and the multi-family group psychoeducational model. Current Psychiatry
Reports, 14(6): 732-738.
Spain, D., Sin, J., Chalder, T., Murphy, D., & Happe, F. (2015). Cognitive behaviour therapy for
adults with autism spectrum disorders and psychiatric co-morbidity: A review. Research
in Autism Spectrum Disorders, 9, 151–162.
St Pourcain, B., Skuse, D. H., Mandy, W. P., Wang, K., Hakonarson, H., Timpson, N. J., …
Smith, G. D. (2014). Variability in the common genetic architecture of socialcommunication spectrum phenotypes during childhood and adolescence. Molecular
Autism, 5-18.
St Pourcain, B. S., Mandy, W. P., Heron, J., Golding, J., Smith, G. D., & Skuse, D. H. (2011).
Links between co-occurring social-communication and hyperactive-inattentive trait
trajectories. Journal of the American Academy of Child and Adolescent Psychiatry, 50(9),
892-902.
Steffenburg, S., Gillberg, C., Hellgren, L., Andersson, L., Gillberg, I. C., Jakobsson, G., &
Bohman, M. (1989). A Twin Study of Autism in Denmark, Finland, Iceland, Norway and
Sweden. Journal of Child Psychology and Psychiatry and Allied Disciplines, 30, 405–416.
Stene, J. (1977). Assumptions for Different Ascertainment Models in Human-Genetics.
Biometrics, 33, 523–527.
Stilp, R. L. H., Gernsbacher, M. A., Schweigert, E. K., Arneson, C. L., & Goldsmith, H. H. (2010).
Genetic Variance for Autism Screening Items in an Unselected Sample of Toddler-Age
Twins. Journal of the American Academy of Child And Adolescent Psychiatry, 49, 267–276.
Stone, L. L., Otten, R., Engels, R. C. M. E., Vermulst, A. A., & Janssens, J. M. A. M. (2010).
Psychometric properties of the parent and teacher versions of the strengths and
difficulties questionnaire for 4- to 12-Year-olds: A review. Clinical Child and Family
Psychology Review, 13(3), 254–274.
201
Stromland, K., Nordin, V., Miller, M., Akerstrom, B., & Gillberg, C. (1994). Autism in
Thalidomide Embryopathy - a Population Study. Developmental Medicine And Child
Neurology, 36, 351–356.
Sucksmith, E., Roth, I., & Hoekstra, R. A. (2011). Autistic traits below the clinical threshold: Reexamining the broader autism phenotype in the 21st century. Neuropsychology Review,
21(4), 360-89.
Sullivan, P. F., Kendler, K. S., & Neale, M. C. (2003). Schizophrenia as a Complex Trait.
Archives of General Psychiatry, 60, 1187–1192.
Sullivan, P. D., Daly, M. J., O’Donovan, M. (2012). Genetic architecture of psychiatric
disorders: the emerging picture and its implications. Nature Review Genetics, 13, 537-551.
Sundquist, J., Sundquist, K., & Ji, J. (2014). Autism and attention-deficit/hyperactivity disorder
among individuals with a family history of alcohol use disorders. eLife, 3, e02917.
Szatmari, P. (2011). Is Autism, at Least in Part, a Disorder of Fetal Programming? Archives of
General Psychiatry, 68(11), 1091-2.
Szatmari, P., Archer, L., Fisman, S., Streiner, D. L., Wilson, F. (1995). Asperger’s syndrome and
autism: differences in behaviour, cognition, and adaptive functioning. Journal of the
American Academy of Child and Adolescent Psychiatry, 34, 1662-1671.
Taniai, H., Nishiyama, T., Miyachi, T., Imaeda, M., & Sumi, S. (2008). Genetic influences on the
broad spectrum of autism: Study of proband-ascertained twins. American Journal of
Medical Genetics Part B-Neuropsychiatric Genetics, 147B, 844–849.
Taylor, B., Miller, E., Farrington, C. P., Petropoulos, M. C., Favot-Mayaud, I., Li, J., & Waight,
P. A. (1999). Autism and measles, mumps, and rubella vaccine: no epidemiological
evidence for a causal association. Lancet, 353, 2026–2029.
Taylor, M. J., Charman, T., Robinson, E. B., Plomin, R., Happé, F., Asherson, P., & Ronald, A.
(2013). Developmental associations between traits of autism spectrum disorder and
attention deficit hyperactivity disorder: a genetically informative, longitudinal twin
study. Psychological Medicine, 43, 1735–46.
Taylor, M. J., Charman, T., & Ronald, A. (2015). Where are the strongest associations between
autistic traits and traits of ADHD. European Child & Adolescent Psychiatry, in press.
Van Dongen, J., Slagboom, P. E., Draisma, H. H. M., Martin, N. G., & Boomsma, D. I. (2012).
The continuing value of twin studies in the omics era. Nature Reviews Genetics, 13, 640–
653.
Vasa, R. A., Kalb, L., Mazurek, M., Kanne, S., Freedman, B., Keefer, A., … Murray, D. (2013).
Age-related differences in the prevalence and correlates of anxiety in youth with autism
spectrum disorders. Research in Autism Spectrum Disorders, 7, 1358–1369.
Visscher, P. (2002). Increased rate of twins among affected sib pairs. Letter to the Editor.
American Journal of Human Genetics, Elsevier (Cell Press), 71(4), 995-996.
202
Wakefield, A. J., Murch, S. H., & Anthony, A. (2010). Ileal-lymphoid-nodular hyperplasia, nonspecific colitis, and pervasive developmental disorder in children (Retraction of vol 351,
pg 637, 1998). Lancet, 375, 445.
Walker, A., Maher, J., Coulthard, M., Goddard, E., & Thomas, M. (2001). Living in Britain:
Results from 2000/2001 General Household Survey. London: TSO.
Wang, K., Zhang, H., Ma, D., Bucan, M., Glessner, J. T., Abrahams, B. S., … Hakonarson, H.
(2009). Common genetic variants on 5p14.1 associate with autism spectrum disorders.
Nature, 459, 528–533.
Weiss, L. A., Arking, D. E., & Consortium, J. H. & A. (2009). A genome-wide linkage and
association scan reveals novel loci for autism. Nature, 461, 802–U62.
Wheelwright, S., Auyeung, B., Allison, C., & Baron-Cohen, S. (2010). Defining the broader,
medium and narrow autism phenotype among parents using the Autism Spectrum
Quotient (AQ). Molecular Autism, 1:10.
Whitehouse, A. J. O., Mattes, E., Maybery, M. T., Dissanayake, C., Sawyer, M., Jones, R. M., …
Hickey, M. (2012). Perinatal testosterone exposure and autistic-like traits in the general
population: a longitudinal pregnancy-cohort study. Journal of Neurodevelopmental
Disorders, 4(1):25.
Williams, J., Allison, C., Scott, F., Stott, C., Bolton, P., Baron-Cohen, S., & Brayne, C. (2006).
The Childhood Asperger Syndrome Test (CAST) - Test-retest reliability. Autism, 10, 415–
427.
Williams, J., Scott, F., Stott, C., Allison, C., Bolton, P., Baron-Cohen, S., & Brayne, C. (2005).
The CAST (Childhood Asperger and Syndrome Test) - Test accuracy. Autism, 9, 45–68.
Wing, L., (1981). Asperger’s syndrome: a clinical account. Psychological medicine, 115–129.
Wing, L., & Gould, J. (1979). Severe impairments of social interaction and associated
abnormalities in children: epidemiology and classification. Journal of Autism and
Developmental Disorders, 9, 11–29.
Wolff, S., Narayan, S., & Moyes, B. (1988). Personality-Characteristics of Parents of AutisticChildren - a Controlled-Study. Journal of Child Psychology and Psychiatry and Allied
Disciplines, 29, 143–153.
Wong, C. C. Y., Meaburn, E. L., Ronald, A., Price, T. S., Jeffries, A. R., Schalkwyk, L. C., … Mill,
J. (2014). Methylomic analysis of monozygotic twins discordant for autism spectrum
disorder and related behavioural traits. Molecular Psychiatry, 19, 495–503.
Wright, S. (1918). On the nature of size factors. Genetics, 3, 367–374.
Yang, J., Lee, S. H., Goddard, M. E., & Visscher, P. M. (2011). GCTA: A Tool for Genome-wide
Complex Trait Analysis. American Journal Of Human Genetics, 88, 76–82.
Yu, T. W., Chahrour, M. H., Coulter, M. E., Jiralerspong, S., Okamura-Ikeda, K., Ataman, B., …
Walsh, C. A. (2013). Using Whole-Exome Sequencing to Identify Inherited Causes of
Autism. Neuron, 77, 259–273.
203
Appendices
Appendix 1: Childhood Autism Spectrum Test (CAST) 31-item questionnaire.
CAST Communication scale (12 items)
1. Does s/he tends to take things literally?
2. Does s/he find it easy to interact with other children?
3. Can s/he keep a two-way conversation going?
4. Does s/he enjoy joking around?
5. Does s/he have difficulty understanding the rules for polite behaviour?
6. Is his/her voice unusual (e.g. overly adult, flat, or very monotonous)?
7. Is s/he good at turn-taking in conversation?
8. Does s/he often do or say things that are tactless or socially inappropriate?
9. Does s/he sometimes say ‘you’ or ‘s/he’ when s/he means ‘I’?
10. Does s/he sometimes lose the listener because of not explaining what s/he
is talking about?
11. Does s/he often turn conversations to his/her favourite subject rather than
following what the other person wants to talk about?
12. Does s/he have odd or unusual phrases?
CAST Social scale (12 items)
13. Does s/he join in playing games with other children easily?
14. Does s/he come up to you spontaneously for a chat?
15. Is it important to him/her to fit in with the peer group?
16. When s/he was 3 years old, did s/he spend a lot of time pretending (e.g. playacting being a superhero, or holding teddy’s tea parties)?
17. Does s/he have friends, rather than just acquaintances?
18. Does s/he often bring you things s/he is interested in to show you?
19. Are people important to him/her?
20. Does s/he play imaginatively with other children, and engage in role-play?
21. Does s/he make normal eye contact?
22. Is his/her social behaviour very one-sided and always on his/her own terms?
23. Does s/he prefer imaginative activities such as play-acting or story-telling,
rather than numbers or lists of facts?
24. Does s/he care how s/he is perceived by the rest of the group?
CAST Non-social scale (7 items)
25. Does s/he appear to notice unusual details that other miss?
26. Does s/he like to do things over and over again, in the same way all the time?
27. Does s/he mostly have the same interests as his/her peers?
28. Does s/he have an interest which takes up so much time that s/he does little
else?
29. Does s/he appear to have an unusual memory for details?
30. Does s/he have any unusual and repetitive movements?
31. Does s/he try to impose routines on him/herself, or on others, in such a way
that it causes problems?
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
No
No
No
No
Yes
No
Yes
No
Yes
Yes
Yes
Yes
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
No
Yes
No
Yes
Yes
Yes
Yes
No
No
No
No
Yes
Yes
Yes
No
No
No
204
Appendix 2: Strengths and Difficulties Questionnaire – SDQ (25 items).
Emotional Symptoms Scale
1. Often complains of head-aches, stomach-ache or
sickness.
2. Many worries, often seems worried.
3. Often unhappy, down-hearted or tearful.
4. Nervous or clingy in new situations, easily loses
confidence.
5. Many fears, easily scared.
Hyperactivity Scale
1. Restless, overactive, cannot stay still for long.
2. Constantly fidgeting or squirming.
3. Easily distracted, concentration wanders.
4. Thinks things out before acting.
5. Sees tasks through to the end, good attention span.
Conduct Problems Scale
1. Often has temper tantrums or hot tempers.
2. Generally obedient, usually does what adults
request.
3. Often fights with other children or bullies them.
4. Often lies or cheats.
5. Steals from home, school or elsewhere.
Peer Problems Scale
1. Rather solitary, tends to play alone.
2. Has at least one good friend.
3. Generally liked by other children.
4. Picked on or bullied by other children.
5. Gets on better with adults than with other children.
Prosocial Behaviour Scale (control)
Not True
Somewhat
True
Certainly
True
Not True
Somewhat
True
Certainly
True
Not True
Somewhat
True
Certainly
True
Not True
Somewhat
True
Certainly
True
Not True
Somewhat
True
Certainly
True
1. Considerate of other people’s feelings.
2. Shares readily with other children (treats, toys,
pencils, etc.).
3. Helpful if someone is hurt, upset of feeling ill.
4. Kind to younger children.
5. Often volunteers to help others (parents, teachers,
other children).
* Items in italics are scored as reverse items.
205
Appendix 3: Mx eScript used in the meta-analysis.
! To estimate meta-analysis ACE components for ASD clinical diagnosis
#ngroups 16
G1: Model parameters
Calc
Begin Matrices;
X full 1 1 FREE ! A path
Y full 1 1 FREE ! c path
Z full 1 1 FREE ! E path
H full 1 1
U unit 1 1
! flag 1
O zero 1 1
! flag 0
P full 1 1
! definition var 'ascertainment probability' (pi-hat) of each study
T full 1 1
! definition var 'threshold' of each study
End Matrices;
Matrix H .5
Start .8 X 1 1
Start .5 Y 1 1
Start .5 Z 1 1
Begin Algebra;
A= X*X';
C= Y*Y';
E= Z*Z';
S= (P) _ (P) _ P+P-P*P ;
! Vector of ascertainment terms
L= A+C+E
| A+C _ A+C
| A+C+E ;
! Exp MZ cov matrix
M= A+C+E
| H@A+C _
H@A+C | A+C+E ;
! Exp DZ cov matrix
J= (P+P)*\mnor(U_O_T_T_U)-P*P*\mnor(L_O|O_T|T_T|T_U|U);
K= (P+P)*\mnor(U_O_T_T_U)-P*P*\mnor(M_O|O_T|T_T|T_U|U);
End Algebra:
interval A 1 1 C 1 1 E 1 1
Options RSidual
End
! *****************************************************
! Study 3 Steffenburg 1989 CA (pi=1) TH=2.55
! *****************************************************
G2: MZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St3.dat
Select if zygosity = 1;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
! definition var 'Frequency' for each pair type
G full 1 4
! definition var to select appropriate ascertainment term in mx S
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
L;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%J;
206
Options RSidual
End
G3: DZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St3.dat
Select if zygosity = 2;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
M;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%K;
Options RSidual
End
! ********************************************
! Study 5 Le Couteur 1996 CA (pi=1) TH=1.65
! ********************************************
G4: MZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St5.dat
Select if zygosity = 1;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
L;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%J;
Options RSidual
End
G5: DZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St5.dat
Select if zygosity = 2;
Select twin1 twin2 asc freq pihat threshold ;
Definition Asc Freq Pihat Threshold ;
207
Begin Matrices= Group 1 ;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
M;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%K ;
Options RSidual
End
! **********************************************
! Study 6 Taniai 2008 CA (pi=1) TH=2.06
! **********************************************
G6: MZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St6.dat
Select if zygosity = 1;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
L;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%J;
Options RSidual
End
G7: DZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St6.dat
Select if zygosity = 2;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
M;
208
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%K;
Options RSidual
End
! *******************************************************************
! Study 8 Lichtenstein 2010 RAP, no corrections AT ALL, TH estimated
! *******************************************************************
G8: MZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St8.dat
Select if zygosity = 1;
Select twin1 twin2 freq;
Definition Freq ;
Begin Matrices= Group 1;
V full 1 1 FREE ! Threshold
F full 1 1
End Matrices;
Specify F Freq
Covariances
L;
Thresholds
V|V ;
Frequency
F;
MA V 2.3
! start value for Threshold
Options RSidual
End
G9: DZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St8.dat
Select if zygosity = 2;
Select twin1 twin2 freq ;
Definition Freq ;
Begin Matrices= Group 1;
V full 1 1 =V8
F full 1 1
End Matrices;
Specify F Freq
Covariances
M;
Thresholds
V|V ;
Frequency
F;
Options RSidual
End
! *********************************************************
! Study 9 Hallmayer 2011 IA (pi=.92) TH=2.49
! *********************************************************
G10: MZ twin pairs
Data Ninput=10
Labels rep studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St9.dat
Select if zygosity = 1;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
209
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
L;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%J;
Options RSidual
End
G11: DZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St9.dat
Select if zygosity = 2;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
M;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%K;
Options RSidual
End
! *********************************************
! Study 12 Nordenbaek 2014 IA (pi=.76) TH=2.49
! *********************************************
G12: MZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St11.dat
Select if zygosity = 1;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
L;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%J;
210
Options RSidual
End
G13: DZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St11.dat
Select if zygosity = 2;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
M;
Thresholds
T|T;
Frequency
F;
Weight
\part(S,G)%K;
Options RSidual
End
! ****************************************************************************
! Study 13 Colvert & Tick 2014 PA no corrections but TH fixed to 1.65 (5% Prev)
! ****************************************************************************
G14: MZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St12.dat
Select if zygosity = 1;
Select twin1 twin2 asc freq pihat threshold;
Definition Asc Freq Pihat Threshold;
Begin Matrices= Group 1;
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
L;
Thresholds
T|T ;
Frequency
F;
Options RSidual
End
G15: DZ twin pairs
Data Ninput=9
Labels studyid studyyr zygosity twin1 twin2 asc freq pihat threshold
Ordinal File=St12.dat
Select if zygosity = 2;
Select twin1 twin2 asc freq pihat threshold ;
Definition Asc Freq Pihat Threshold ;
Begin Matrices= Group 1 ;
211
F full 1 1
G full 1 4
End Matrices;
Specify P Pihat
Specify T Threshold
Specify F Freq
Matrix G 1 1 1 1
Specify G Asc 0 Asc 0
Covariances
M;
Thresholds
T|T ;
Frequency
F;
Options RSidual nd=4
End
G16: constrain Total variance A+C+E to 1
Constraint NI=1
Begin Matrices = Group 1;
End Matrices;
Constraint U = A+C+E;
End
212
4: eData
used0=unaffected.
in the meta-analysis.
Twin 1 and TwinAppendix
2 annotation:
1=affected,
Study
ID
Year of
study
Reference
Zygosity
Twin 1
Twin 2
Ascertainment
type
N pairs
pi (π)
Threshold
3
1989
Steffenburg
MZ
1
0
1
1
1
2.33
3
1989
MZ
1
1
3
10
1
2.33
3
1989
DZ
1
0
1
10
1
2.33
3
1989
DZ
1
1
3
0
1
2.33
5
1996
MZ
1
0
1
2
1
1.65
5
1996
MZ
1
1
3
26
1
1.65
5
1996
DZ
1
0
1
18
1
1.65
5
1996
DZ
1
1
3
2
1
1.65
6
2008
MZ
1
0
1
1
1
2.06
6
2008
MZ
1
1
3
18
1
2.06
6
2008
DZ
1
0
1
18
1
2.06
6
2008
DZ
1
1
3
8
1
2.06
8
2010
MZ
0
0
-
2213
1
2.30
8
2010
MZ
1
0
1
11
1
2.30
8
2010
MZ
0
1
1
11
1
2.30
8
2010
MZ
1
1
3
7
1
2.30
8
2010
DZ
0
0
-
5652
1
2.30
8
2010
DZ
1
0
1
42
1
2.30
8
2010
DZ
0
1
1
42
1
2.30
8
2010
DZ
1
1
3
4
1
2.30
9
2011
MZ
1
0
1
22
0.9
2.49
9
2011
MZ
1
1
2
4
0.9
2.49
9
2011
MZ
1
1
3
28
0.9
2.49
9
2011
DZ
1
0
1
120
0.9
2.49
Le Couteur
Taniai
Lichtenstein
Hallmayer
Prevalence rate mentioned
Structural Equation
Modeling (SEM) was not
performed. 1% prevalence
for Autism Disorder is
assumed (z-value 2.33).
No SEM, include a broader
phenotype category hence
5% prevalence is assumed
(z-value 1.65).
Specified prevalence of ASD
for males at 3.3% and 0.82%
for females. A mean
prevalence is calculated of
2% (z-value 2.06).
Threshold value (2.3) is start
value for actual threshold,
which is estimated for this
data.
Specified prevalence of 1%
for males and 0.3% for
females. A mean prevalence
is calculated at .6% (z-value
2.49).
213
9
2011
DZ
1
1
2
5
0.9
2.29
9
2011
DZ
1
1
3
13
0.9
2.49
11
2014
MZ
1
0
1
1
0.76
1.65
11
2014
MZ
1
1
2
4
0.76
1.65
11
2014
MZ
1
1
3
8
0.76
1.65
11
2014
DZ
1
0
1
22
0.76
1.65
11
2014
DZ
1
1
2
1
0.76
1.65
11
2014
DZ
1
1
3
0
0.76
1.65
12
2015
MZ
0
0
-
29
1
1.65
12
2015
MZ
1
0
1
0
1
1.65
12
2015
MZ
0
1
1
3
1
1.65
12
2015
MZ
1
1
3
24
1
1.65
12
2015
DZ
0
0
-
51
1
1.65
12
2015
DZ
1
0
1
38
1
1.65
12
2015
DZ
0
1
1
32
1
1.65
12
2015
DZ
1
1
3
30
Ascertainment type: 1=discordant, 2=singly ascertained concordant, 3=doubly ascertained concordant.
1
1.65
Nordenbaek
Colvert, Tick
No SEM, prevalence set to
5%.
Prevalence set at the outset
at 5% for Broader
Phenotype.
214