* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Beata_Tick_Thesis_Feb_2016
Survey
Document related concepts
History of psychiatry wikipedia , lookup
Facilitated communication wikipedia , lookup
History of mental disorders wikipedia , lookup
Dissociative identity disorder wikipedia , lookup
Controversy surrounding psychiatry wikipedia , lookup
Classification of mental disorders wikipedia , lookup
Diagnostic and Statistical Manual of Mental Disorders wikipedia , lookup
Pyotr Gannushkin wikipedia , lookup
Child psychopathology wikipedia , lookup
Spectrum disorder wikipedia , lookup
Autism therapies wikipedia , lookup
Heritability of autism wikipedia , lookup
Epidemiology of autism wikipedia , lookup
Transcript
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 (EmoASD) or that the reverse is true and ASD causes Emotional symptoms (EmoASD). 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 (SDQCAST). 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 (SDQCAST) or r’ (SDQCAST) 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