Rethinking the nature of genetic vulnerability to autistic spectrum disorders

Rethinking the nature of genetic vulnerability to autistic spectrum disorders

Opinion TRENDS in Genetics Vol.23 No.8 Rethinking the nature of genetic vulnerability to autistic spectrum disorders David H. Skuse Behavioural and...

168KB Sizes 0 Downloads 5 Views


TRENDS in Genetics

Vol.23 No.8

Rethinking the nature of genetic vulnerability to autistic spectrum disorders David H. Skuse Behavioural and Brain Sciences Unit, Institute of Child Health, 30 Guilford Street, London WC1N 1EH, UK

Autism is a common and genetically heterogeneous disorder, with an estimated heritability of >90%. Its specific underlying causes are largely unknown. Here, I propose that low levels of autistic vulnerability, reflected in social-cognitive processing differences, do not necessarily manifest in a behavioural phenotype but are usually compensated for during development. They are more likely to lead to a recognizable syndrome among individuals of low intelligence, who are male or have independent neurodevelopmental vulnerability owing to a wide range of gene mutations, chromosomal anomalies or environmental insults. Consequently, the apparent association between mental retardation and autistic syndromes is not because they usually have common causes, but rather because the presence of both features greatly increases the probability of clinical ascertainment. Autism: identification and proposed causes Autism is generally regarded as a severe and distinct disorder, with neurodevelopmental origins in early childhood. It affects the ability to engage in the normal to-andfro of social interaction and is associated with restricted or abnormal use of language; by definition, these symptoms are accompanied by a range of ritualistic behaviours, restricted interests and motor stereotypies (Box 1). The ultimate cause of the condition is, in most cases, thought to be genetic susceptibility but, despite a decade of research, little progress has been made in finding any candidate genes that contribute to susceptibility in idiopathic cases [1]. Recently, Sebat et al. [2] provided evidence that de novo copy number variation has an important role in at least 10% of cases; this might indicate a new research direction. Meanwhile, linkage studies aimed at revealing genomic regions harbouring susceptibility genes rarely replicate Glossary Autism Diagnostic Interview (ADI)-Revised: a complex interview for administration by trained interviewers to parents of suspected autistic children; the ‘gold standard’ assessment procedure for research purposes. It has an algorithmic scoring system that summarizes the degree of impairment in the three main dimensions of autistic symptomatology (Box 1) [40].

Corresponding author: Skuse, D.H. ([email protected]). Available online 13 July 2007.

Autism Diagnostic Observation Schedule: a complex observational system for administration by those trained to use it. It involves structured interactions between an observer and an individual with a suspected autistic disorder. The scoring system (recently revised [51]) summarizes impairments in terms of social interaction and the use of language for social communication [41]. Broader Phenotype Autism Symptom Scale (BPASS): a brief interview for administration by trained clinicians to parents or teachers. Four domains of symptomatology are assessed. Only preliminary findings on its psychometric properties are available [36]. Child Asperger Syndrome Test (CAST): a questionnaire (37 items) rated by parents, which enquires systematically about the three main domains of autistic impairment and does not seem to have a onedimensional structure on statistical evaluation in a general population-based sample [32]. Endophenotype: this term should refer to cognitive, neurophysiological, neuroanatomical or other biological processes that are potentially closer to the action of genes than overt behaviour or symptomatic psychopathology. In practice, endophenotypes are usually assumed to be quantitative traits that reflect individual differences in allelic variation with functional consequences. Heritability: used here in the narrow sense, heritability refers to the proportion of phenotypic variance that is additive in nature (i.e. is due wholly to allelic variation). The heritability of a quantitative (or dimensional trait) phenotype is the proportion of phenotypic variance in that trait that is accounted for by genetic effects. For a categorical phenotype (present or absent), it is the proportion of variance in liability (on the basis of the liability-threshold model) that is accounted for by genetic effects [77]. LOD score: a statistical estimate of whether two genetic loci are likely to lie near each other on a chromosome and therefore stand an increased chance of being inherited together. If the position of one locus is known (because it is a marker), the position of the second (which might be a gene) can be inferred. LOD scores >3 imply that the likelihood of observing the result if the two loci are not linked is less than 1 in 1000, and is conventionally regarded as the lower limit of ‘significance’. Multiplex family: families containing more than one autistic individual. In practice, the autistic individuals are usually siblings or, at least, first-degree relatives. Quantitative trait: this term refers to traits that are continuous characteristics in a population, rather than characteristics of the phenotype that are qualitatively different from normal. Such traits are influenced by more than one gene and their interaction with the environment. Social and Communication Disorders Checklist (SCDC): a questionnaire (12 items) rated by parents or teachers, which summarizes autistic symptoms along one dimension and key social, language and behavioural characteristics of the autistic spectrum of disorders [31]. Social Responsiveness Scale (SRS): a questionnaire (65 items) rated by parents or teachers, which enquires about a wide range of autistic symptoms but has been shown statistically to indicate that there is a one-dimensional distribution of scores in the autistic and comparison populations [8].

0168-9525/$ – see front matter ß 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.tig.2007.06.003



TRENDS in Genetics

Vol.23 No.8

Box 1. Autism: definitional issues Autistic spectrum disorders are defined by three main domains of impairment [48,49]. These domains comprise difficulties in reciprocal social interaction, abnormalities in the use of language for social communication (including nonverbal communication and play) and restricted, repetitive and stereotyped patterns of behaviour, interests and activities. The term ‘pervasive developmental disorders’, which was formerly used to encompass all shades of autistic symptomatology has recently fallen out of favour, in part because it is unfamiliar to parents and in part because it is inaccurate (the associated handicaps are not all-pervasive). There is no evidence that any aspect of the autistic phenotype is qualitatively distinct from normal development. Accordingly, the term ‘autistic traits’ is often used to mean mild autistic symptoms that are (probably) not of clinical significance but can be detected in normal populations or those with other medical conditions. For research purposes, there is nearly a monopoly on the assessment process that mandates (for both funding and publication purposes) the use of the Autism Diagnostic Interview – Revised (ADIR) [40] (a parent-report instrument) and the Autism Diagnostic Observation Schedule (ADOS) [41] (a structured method of child observation). There is no evidence that the findings of genetic studies of autism have shown increased consistency if such strict methods of phenotyping have been used. In fact, the opposite seems to be true; broadening the phenotypic definition seems to yield better evidence for risk loci [50,3]. The current debate about the best approach for phenotyping in genetic studies centres on the conventional view that autism comprises three distinct domains of impairment. It seems clear that these domains should be regarded as dimensions of impairment, rather than qualitatively distinct categories. The use of conventional diagnostic criteria is a valuable heuristic in clinical practice. But for genetic studies, it is important to be aware that the phenotypic qualities of autistic spectrum disorders are not incontrovertible and are potentially subject to revision [51]. There might be more than the three dimensions of autistic impairment (as defined above), two dimensions of autism (i.e. difficulties in reciprocal social interaction and problems in the use of language) [13] or autism might be one-dimensional [8]. I strongly advise geneticists who are engaged in autism research to exercise more than a little skepticism about the claims made about the validity of phenotypic characteristics, captured by interviews or questionnaires. In the absence of independent evidence of aetiology, symptomatic phenotypes reflect primarily the characteristics of the instrument used to measure them.

key loci. Increasing the sample sizes has not improved discrimination [3,4], although there might be mathematical reasons for shrinking LOD scores (see Glossary) that could be solved by using a Bayesian approach (posterior probability of linkage) [5] or increasing the genetic homogeneity of the sample [6]. Identifying individuals at genetic risk in extended families could be a profitable method of increasing the homogeneity of datasets [7]. Ascertainment for genetic studies usually entails strict case identification, but it is arguable that the criteria for diagnosis are arbitrary. An alternative framework proposes that susceptibility should be conceptualized as one or more quantifiable and dimensional traits [8]. At present, it is unclear whether the best method of measuring such traits for efficient genomic mapping is by behavioural ratings, cognitive testing or other means. Recent evidence from a general population-based twin study [9,10] suggests different (sets of) genes influence key components of the autistic phenotype, thus the hypothesis that autism has a unitary aetiology could be fallacious [11]. Genetic studies of autism almost invariably recruit samples that are selected according to the strictest and most uniform case criteria in psychiatric genetics [12].

Adherence to case-definition rules has influenced both funding and publication decisions (Box 1). Individuals with moderate-to-severe mental retardation and low verbal abilities predominate. On initial consideration, there are real difficulties reconciling findings from symptom-screening studies of general populations, which claim autistic symptoms reflect genetically heterogeneous quantitative traits [8–10], with the research focus of many psychiatric geneticists, who seek strong evidence of specific causative genes [13,14]. Key questions for future research include the following: (i) In the presence of moderate-to-severe mental retardation, is an autistic phenotype simply more prominent than in children of normal-range intelligence? (ii) Do the same genes that cause autism also cause mental retardation? (iii) Are children with neurodevelopmental vulnerability (of whatever aetiology) more likely to express autistic traits in ways that are obvious to an observer? (iv) If subtle social-cognitive deficits of an autistic type are relatively common, would ‘autistic endophenotypes’ be found in people with mild, or even absent, autistic symptoms and normal-range intelligence if the means existed to measure them reliably [15]? (v) Could the apparently strong association between autism and mental retardation reflect ascertainment bias [16] and be due, in part, to ‘diagnostic substitution’ of children previously ascertained as mentally retarded [17]? Herein, I will attempt to provide answers to these questions by reviewing the assumptions that led to the current conceptualization of autism as a distinct neurodevelopmental disorder, with both a potentially identifiable neuropathology and a coherent genetic substrate. I will argue that studies aiming to map genes that increase autism susceptibility have often, perhaps usually, inadvertently identified loci or candidates that are important for the development of elements of general intelligence needed to compensate for underlying autistic traits. Effective compensation almost certainly reflects a subtle gene– environment interaction. I suggest, in light of recent findings on the dimensionality of autistic traits, that several key elements of the autistic phenotype assort independently of one another. Furthermore, I propose these elements are common, and evidence for them could potentially be found in up to one-third of the general population (Box 2). Confounding between autism and mental retardation in twin studies Evidence that autism is strongly influenced by genetic susceptibility comes largely from twin studies [18–22], which have shown substantially greater concordance for autistic symptoms in identical [monozygotic (MZ)] twins than dizygotic (DZ) twins (Table 1). However, the same twin studies have also shown that autistic symptoms are most prominent if there is associated mental retardation. The greater the degree of retardation, the more severe the autistic phenotype. A close relationship exists between genetic liability, the diagnosis of autism and broader learning difficulties (especially verbal intelligence) in both MZ


TRENDS in Genetics

Vol.23 No.8

Box 2. Autistic traits in single-gene and chromosomal disorders There is increasing evidence that autism and autistic traits are found in a wide variety of genetic and chromosomal disorders. It has been known for many years that certain ‘single-gene’ conditions often have distinct autistic-like features, especially Fragile X and Rett syndromes, in which their prevalence is between 30% and 40%. Theories have evolved to explain why such associations could lead to a revealing of the neural substrate of idiopathic autism [52–54]. When only a handful of genetic syndromes had been studied systematically, this conclusion seemed plausible. Now, because of the diversity of conditions with a similar prevalence of autistic traits, the time has come to reappraise earlier optimistic conjectures. Duchenne muscular dystrophy is associated with autistic behaviours (social and communication deficits) in at least one-third of patients [55]. Tuberous sclerosis (especially TSC2 mutations) is associated with clinically significant autistic features in up to 50% of patients [56]. Children with Williams syndrome commonly have poor theory-of-mind skills and lack social reciprocity [57]. Prader– Willi syndrome critical-region duplications and overexpression of maternally-derived genes are associated with autistic spectrum disorders in up to 25% of patients [58]. Over one-third of children with velocardiofacial syndrome have autistic spectrum disorders [59,60]. Children with Down’s syndrome frequently display social withdrawal and stereotypies of an autistic nature [61]. All sexchromosome aneuploidies have autistic behaviour at a rate far in excess of that expected by chance, including Turner syndrome [62], Klinefelter syndrome [63] and, possibly, 47,XYY and 47,XXX syndromes [64,65]. Autistic traits are often found in genetically mediated metabolic disorders, such as Smith–Lemli–Opitz syndrome [66], but have not been widely investigated yet. Specificity between syndrome manifestations in association with these single gene disorders has not been established and it is premature to conclude that such metabolic conditions share neurobiological mechanisms with idiopathic autism. The more closely we look for autistic traits, the more often they are discovered.

and DZ samples. Even MZ and DZ co-twins with less severe symptoms (a broader phenotype) are reported to have a mean verbal intelligence quotient (IQ) nearly two standard deviations below co-twins with no autistic symptoms at all (Table 1). This could be due, in part, to confounding between the criteria for diagnosis of autism (communication skills) and impaired verbal intelligence, with which several items used for diagnostic purposes are directly correlated. Twin studies strongly implied that genes relevant to the clinical diagnosis of autism function through a profound effect on neurodevelopment, although they can occasionally manifest as a highly variable, milder phenotype in the absence of generalized learning difficulties. An alternative hypothesis is that, within these twin samples, lower general intelligence diminished the possibility of cognitive compensation for independently inherited autistic traits. DZ co-twins who seemed to lack autistic symptoms generally lacked the associated nonspecific learning difficulties, whereas MZ co-twins who had one set of inherited characteristics invariably also inherited the others. Revisiting estimates of heritability Heritability estimates make certain assumptions. An often-quoted estimate of autism heritability is 90% [13]. In deriving this estimate, it was necessary to consider the findings of contemporaneous prevalence surveys of autism in the general population. Assuming a base rate of 0.0175% (derived from UK national statistics of the


proportion of autistic children in educational facilities at the time the estimate was made), heritability was estimated to be 93%; however, this was reduced to 91% if the prevalence was 0.1%. In recent community surveys, the prevalence of autism is much higher, in the range of 0.2% [12] to 0.4% [23], indicating that the estimate should be revised downwards. In twin studies, estimates of heritability also rely on the proportionate risk to DZ co-twins. This was reported as effectively zero for the full-autism phenotype and just 10% for a broader phenotype in the UK twin samples [20]. If the risk to siblings had been underestimated, heritability would be reduced accordingly; recent research suggests that a partial phenotype is much more common than previously suspected in the siblings of autistic individuals. Current estimates put the recurrence risk of the full-autism phenotype between 2% and 6% [24]. If a more relaxed set of criteria is used to define risk, the proportion of siblings who have several features of a broader phenotype rises to 20% [25]. Recent investigations of subtle characteristics of autistic-like dysfunction, such as pragmatic language skills, have found even greater proportions [26], and in families with more than one autistic child up to half of ‘unaffected’ siblings are now reported to have significant traits [27]. What is inherited? These observations raise important questions about what is being inherited. If autism, as conventionally defined, is so heritable, why do so few siblings (who share half of the genetic risk) display the full (as distinct from a partial) phenotype? One possibility is that the risk of displaying the full phenotype is greatly increased if another independent risk factor is present (which could be genetic, epigenetic, stochastic or environmental in origin). There are potentially many factors that influence the overt expression of autistic symptoms, which could be unrelated to the underlying risk of inheriting one or more ‘autistic’ characteristics, in terms of social and communication deficits or stereotyped or inflexible cognitions or behaviours. The major influence on whether a behavioural phenotype is observed is general cognitive ability. In a large sample of multiplex families with autism, Spiker et al. [28] found that the social and communication symptoms used to diagnose autism correlated with nonverbal IQ (but not verbal ability), independent of the relationship between verbal status and the diagnostic criteria. The heritable severity gradient found in sibling pairs diagnosed as autistic could have been the consequence of normal variation in cognitive ability superimposed on children who were genetically at risk of autism. Starr et al. [29] queried whether family-based genetic risk differs according to whether autism is accompanied by severe intellectual disability and concluded that the same genetic liability applies to idiopathic autism regardless of IQ. These findings seem consistent with the hypothesis that a predisposition to autistic features of behaviour could be relatively common, and autism is not an ‘all-or-nothing’ phenomenon, but rather its clinical manifestations are influenced (among many other factors) by general cognitive ability. How common is a genetic predisposition to autistic behaviour in the general population? One indication might


Table 1. Summary of findings on concordance in autistic symptoms and cognitive disability among MZ and DZ twins MZ twin pair-wise concordance for cognitive disability

DZ twin concordance for autism

DZ twin concordance for cognitive disability

Cognitive abilities of twins diagnosed with autism

11 8 male 3 female

10 7 male 3 female





23 18 male 5 female

17 9 male 1 female 7 mixed sex



11 Unclear MZ/DZ distribution of sex. Overall, 11 pairs male, 10 pairs female. 25 21 male 4 female

10 All samesex pairs


No information given, but by definition all autistic cases had to have ‘gross deficits in language development’ 10/11 Mean social quotient (Vineland) for MZ twins with autism was 46

Mean IQ in typical cases of autism (n = 14) was 52.9. Overall, half of autistic twins had an IQ <50 No information given on level of ability

0/10 Mean social quotient (Vineland) for DZ twins with autism was 57

3/10 concordant for delayed speech or reading disability, but all had a normal IQ

Overall, half of autistic twins had an IQ <50

20 13 male 7 female


Concordance for poor verbal abilities in autistic twin pairs was high. Overall, mean verbal IQ difference within pairs was 30.


2/20 DZ pairs were concordant for communication impairment

Twins with autism or atypical autism had a mean verbal IQ of 35

The numbers of males and females are given for each study referred to.

Cognitive abilities of twins diagnosed with atypical autism or normal Mean IQ for atypical cases of autism (n = 11) was 61.4

7 co-twins with several autistic symptoms (5 MZ and 2 DZ) had a mean verbal IQ of 81. 15 unaffected cotwins had a mean verbal IQ of 110

Neuropathological abnormalities

Provenance of twin sample


Among the autistic sample, 4 had epilepsy and 11 had an abnormal EEG

Multiple sources of information were sought nationally to identify known twin pairs


No information given

28 families were selected from an advertisement in the National Autistic Society newsletter. 12 families from UCLA outpatients or clinical referrals


12% of autistic twins were reported to have epilepsy

Scandinavian medical and other specialists in child development were surveyed, from 5 countries


One-third of autistic individuals had epilepsy, of which 18/19 had a verbal IQ <30: in 6 MZ pairs, concordant for autism, one twin was also epileptic, as were 4/6 co-twins

One-third of MZ twin pairs were from the Folstein and Rutter 1977 sample A nationwide search for twin pairs was conducted. The authors state significant ascertainment bias was ‘extremely unlikely’. The authors comment that milder symptoms of autism differ from autism per se because the former phenotype is not associated with mental retardation or epilepsy


Vol.23 No.8

MZ twin pair-wise concordance for autism

TRENDS in Genetics

DZ twins a



MZ twins a


TRENDS in Genetics

Vol.23 No.8

be provided by the observation that autism and autistic traits are found in up to half of children with diverse chromosomal and genetic anomalies (especially those associated strongly with mental retardation), supporting the hypothesis that there is a variable ‘threshold of expression’ dependent upon IQ (Box 2). These findings, of diverse genetically unrelated conditions associated with an autistic-like behavioural phenotype, imply the following reinterpretation of ‘zero DZ concordance’ in twin data might be appropriate. If cases are ascertained for twin studies on the basis of a combination of low intelligence and the presence of autistic features (both of which are strongly influenced by what are arguably independent genetic factors, environmental risk events having been excluded by design), a high pair-wise concordance for both autistic features and low IQ should be expected among MZ twin pairs, which is what has been reported. Depending on the relative heritability of low IQ (of whatever genetic aetiology) and features of autism, it is not expected that both members of every pair will be concordant for both features, but the severity of autistic traits should be strongly associated with overall cognitive ability in this highly selected sample. DZ pairs will only be fully concordant for autism if the co-twin is affected by (arguably independently inherited) learning difficulties or another neurodevelopmental insult, by chance a rare event. There could, therefore, be many unidentified twin pairs in which one, or both, has a milder phenotype, with several autistic symptoms; however, because neither twin suffers from generalized learning difficulties, they have never been ascertained clinically. Categorical versus dimensional aspects of the phenotype Evidence has emerged during the past decade to indicate that clinically defined autism is not a qualitatively distinct syndrome, but it is variable in severity, independent of general intelligence, both within and between families. Several different methods of measuring autistic traits have been developed to capture the full spectrum of severity, which were extended into the general population. The first such study [30] used a questionnaire later known as the ‘Social and Communication Disorders Checklist (SCDC)’ [31]. Heritability, in the narrow sense, was found to be 0.76 for a sample of 656 twin pairs, with no significant role for a shared environment (effectively family influences). Modelling for sex differences in the response to genetic and environmental influences showed no significant effects, although males had more marked symptoms than females, on average. The checklist used in this survey was designed to measure a one-dimensional trait and did not attempt to emulate the three-dimensional structure of conventional autistic diagnostic criteria (Box 1). A second study of 788 twin pairs, using a more detailed phenotypic scale [Social Responsiveness Scale (SRS)], reported similar findings [8]. The crucial finding of this latter study was the continuous distribution of the trait, which also seemed to be onedimensional, despite incorporation of questions about social behaviour, communication and (to a degree) other aspects of the autistic phenotype. Another autism screening instrument, the Child Asperger Syndrome Test (CAST) [32], was used in a survey of 3419 general population-based


twin pairs [9,10]. Males scored higher than females for autistic traits and the findings for narrow heritability (0.8) were similar in both sexes, with no significant influence of a shared environment. The three CAST symptom dimensions (corresponding to the main dimensions of autistic symptomatology) were only moderately correlated and there was a substantial, but largely independent, genetic specificity for each component of the tripartite phenotype. Although there might be technical reasons for this result, because of the method of estimation employed (DeVries– Fulker extremes analysis), the result is intriguing and provocative. In a fourth study, the Broader Phenotype Autism Symptom Scale (BPASS) [33] was used to estimate heritability in 201 multiplex families. The outcome variables comprised four traits (expressiveness, conversational skills, social motivation and flexibility or restricted interests); how they map onto the three conventional dimensions of autism is unclear. The findings confirmed earlier evidence for familiality of impairment in social interaction skills [34] and repetitive behaviours [35]. However, the construct validity of BPASS seems to differ substantially from the other measures discussed. The authors report a high genetic correlation between the ‘social motivation’ and the ‘flexibility’ domains, which is inconsistent with earlier findings [9,10]. In addition, whereas other population studies report no significant correlation between IQ and autistic traits [8,31], the BPASS [36] was correlated to IQ on most dimensions. Estimates of heritability were also substantially lower than had been found by twin studies. All these symptom scales had been validated in studies of conventionally diagnosed autistic children and, to that extent, they should all have been measuring the same latent variables, which indexed vulnerability to develop the clinically identified condition. All the scales are derived from a similar set of presumptions. First, that the genetic basis of autism is complex, with multiple genes contributing to autism susceptibility. Second, the assumption that the failure of linkage studies to reveal consistent signals is probably related to the reliance on categorical diagnoses for the selection of participants. Third, that there is a need to develop measures of autistic traits that can be applied to the general population, in addition to clinically identifiable cases. All the scales are completely reliant on accurate parental report and cannot have any validity that extends beyond the set of questions posed and veracity of parental responses. This is a little discussed, but crucial, issue, because parents are becoming increasingly well informed about the nature and range of autistic symptoms. It is too early to tell which approach will work best in genetic studies in which autism is considered as a quantitative trait. Genetic heterogeneity of autism An outstanding question, for those engaged in research to find the genetic basis of autistic disorders, is whether we should expect the same set of genes to influence the risk of developing the following characteristics: (i) Conventional autism, strictly clinically defined. (ii) Autistic spectrum disorders, defined according to conventional criteria but with less strict degrees of severity.



TRENDS in Genetics

Vol.23 No.8

(iii) Autistic traits, defined according to one or more dimensions that are continuous between clinical cases and general population-based samples. (iv) Subgroups of conventional symptoms or characteristics of development (such as age at onset of language). The issue of genetic heterogeneity is discussed at some length in reports from early twin studies [22]. The finding that there can be substantial variation within MZ pairs, in terms of severity and the pattern of behavioural manifestations of autism, compared with between-family case variation, was believed to indicate that little heterogeneity should be expected. In other words, this seemed to be evidence that the same genetic influences could manifest as a highly variable phenotype [37]. On the other hand, if intellectual variation within pairs is genetically unrelated to autistic traits, but itself manifests as a highly variable phenotype, the apparent differences in severity autistic traits could be explained by an alternative mechanism. In summary, several studies using general population-selected samples of twins [8–10] and singletons [38] have now claimed, first, that the distribution of the severity of autistic traits is continuous and, second, that there is no clear relationship between symptom severity and verbal or nonverbal intelligence. For these assertions to be valid, it must be assumed that the identification of autistic individuals in most genetic studies to date has been subject to enormous bias. Consistency in estimates of population prevalence among epidemiological studies [39] would then simply reflect the fact that similar, if not identical, criteria were being used for making the diagnosis. To date, these criteria are almost invariably based on the Autism Diagnostic Interview (ADI) [40] and Autism Diagnostic Observation Schedule [41]. Self-evidently, the use of highly reliable measures by trained interviewers will identify roughly equal proportions of ‘cases’ in general population-based samples, assuming there is an equivalent liability in those populations. But the ability to identify cases reliably, without any external validation of the disorder, apart from the symptoms used to define it, cannot illuminate the genetic basis of the syndrome. Nor does it demonstrate that autism, as conventionally defined, is anything other than an arbitrary constellation of symptoms that are most readily identifiable among children with some degree of generalized learning difficulty. It is to address these well-recognized problems with diagnostic ascertainment that research has, in recent years, attempted to identify cognitive, neurophysiological or other endophenotypes that lie closer to the action of susceptibility genes than symptoms or observable behaviour [42]. Concluding remarks Autism is a set of heritable traits, as indicated by twin and family studies, but it is probably not as heritable as previously claimed. Recent research calls into question the assumption that genes predisposing to autism also predispose to low IQ and neuropathological processes resulting in phenotypes such as epilepsy. Revised estimates of heritability, derived from conventional diagnoses in twin or family studies, would yield approximately the same values

as those found by screening instruments designed to measure autistic traits (not amounting to a clinical diagnosis) in general population-based samples (i.e. 0.75–0.8). Although this value is still large, its implications are conceptually of considerable significance. In this review I have argued that we should not expect to find any greater genetic homogeneity in clinically defined autism than in samples with autistic traits that were drawn from the general population. Further, I suggest that the distribution of traits in the general population, which appears non-normal when measured in terms of symptom severity [8,38], might fit a Gaussian curve if measured in terms of endophenotypes. This is because symptomatic compensation occurs in many individuals of normal-range intelligence. Strategies for genetic research in autism The logic of recruiting increasingly larger samples of multiplex families with cases of ‘typical’ autism, which are dominated by children with moderate-to-severe mental retardation, for the purpose of genomic-linkage or association studies is, therefore, questionable. First, because of the relatively small samples that are available for study, even if many different groups with access to such data pool their resources (to date, typical linkage and association studies have recruited, at most, a few hundred families). Second, because there is immense expense involved in such studies, in which participating subjects are widely spread throughout the USA, Europe and beyond. Third, there is an attendant risk of mapping loci and/or genes that have more association with heritable forms of generalized mental retardation than with specific features of an autistic spectrum disorder. The true prevalence of autism Most epidemiological surveys indicate that approximately 70% of autism cases have significant mental retardation [12,43]. Recent evidence suggests this claim might reflect a bias in ascertainment, in the sense that suspected autistic children with moderate-to-severe learning difficulties are more likely to be selected for second-stage assessment, especially if surveyed in the preschool period. Reconciling this observation with evidence that there is no overall association with IQ in population surveys of autistic traits, using screening questionnaires, requires reconsideration of the widely held assumption that genes increasing susceptibility to autism also cause mental retardation and abnormal brain development, manifesting in electroencephalogram abnormalities or epilepsy [44]. I have proposed an alternative explanation: in this reconceptualization, neuropathological findings are simply markers for a brain that is less likely to compensate for the susceptibility to autistic traits, and such markers are often associated with global mental retardation. Recent reports indicating that abnormal head growth during early childhood implies a common neuropathological process for autism do not hold up to scrutiny in large, heterogeneous samples [45]. Another risk factor is male sex: males are substantially more likely to be ascertained as having an autistic disorder than females, and curiously the sex ratio is much larger among children with higher abilities [43]. One interpretation of this observation


TRENDS in Genetics

Vol.23 No.8

is that females are equally at risk, in terms of genetic predisposition, but a factor relating to genetic or hormonal sex differences enables them to compensate for that risk [46]. They are, therefore, less likely to manifest the full range of autistic symptoms, as conventionally measured (Box 3). Autism in chromosomal and genetic disorders Emerging studies of miscellaneous chromosomal or single-gene disorders associated with low IQ indicate that about one-third of subjects have significant autistic traits. There is reasonable consistency in the proportion affected, between syndromes, whether the anomaly is a 22q11.2 microdeletion syndrome, Smith–Lemli–Opitz syndrome or Duchenne muscular dystrophy. Rather than revealing potential candidate genes or neuropathological anomalies that are specific to autism, a more parsimonious explanation is that the moderate-to-severe mental retardation associated with these conditions merely reveals autistic traits that were already present. This explanation is supported by the observation that equivalent autistic traits can be found in about one-third of idiopathic cases of mental retardation too. In a survey using an autism screening questionnaire, the Box 3. Sex differences and autistic traits The male preponderance of identified individuals with autism (a ratio of male:female of at least 4:1) is one of the few undisputed facts that could indicate a mechanism for biological vulnerability. There was evidence from early studies of a higher proportion of females relative to males among autistic children with severe-toprofound mental retardation, although the reduced sex ratio might apply more to atypical varieties of autism and cases that are associated with medical disorders, rather than low IQ [29]. However, it seems beyond dispute that among those with high IQ, males predominate and the ratio is nearer 10:1 [67]; this finding cannot be explained by medical complexities. Essentially, there are three competing theories for sex differences in the prevalence of autism. First, there is a possibility that X- or Y-linked genes are involved. The excess of males could indicate an X-linked disorder [68], and there has been some evidence of X-linkage in genome-wide scans, but this is unconvincing. An excess of males with X-linked mental retardation does not provide an explanation, because the male:female autism ratio decreases with IQ, rather than increases. There is no evidence of Y-linkage either [69]. Cases of apparent male-to-male transmission of autism in multiplex families ‘rule out’ X-linkage as the predominant mode of inheritance in these families [70], although this assertion assumes that the phenotypic manifestations of the genetic predisposition are the same in males and females. Maternal imprinting of the X chromosome could lead to femalespecific expression of X-linked genes (from the paternal X chromosome), and there is evidence that the threshold for expression of autistic traits is influenced in this way [71]. The preponderance of Xlinked genes that, if mutated, leads to mental retardation in males could be another factor contributing to the overt manifestations of autism in the male population. Hormonal mechanisms might be operative, and there is preliminary evidence to suggest that androgenization of the male foetus increases vulnerability [72]. Finally, other mechanisms of risk might be different in males and females [73]. There has been recent interest in the apparent sexlimited expression of autosomal genes from candidate regions on chromosomes 7q [74] and 17q11.2 [75]. Perhaps females with autism will prove to be more genetically informative than males [3,76]. Females might have less genetic heterogeneity than males because they are less likely to have concomitant mental retardation, and hence a confounded phenotype. Therefore, should the criteria for diagnosis be the same in females and males or do females have a modified phenotype?


SCDC [31], in a general population-based sample of 6082 8-year-olds, 62 children were found to have a verbal IQ <70. Of these, 30.6% had SCDC scores in the ‘probably autistic’ range. The equivalent proportion was 41.2% for those who had a verbal IQ <60 (D.S., unpublished). In cases where low IQ is in the severe-to-profoundly impaired range, there may be a different aetiological mechanism in some cases. Whereas the ability to compensate for independently inherited autistic traits will be very limited in such children, there will be others in which an autistic phenotype is merely a non-specific outcome of inefficient functioning of the ‘social brain’ [29,47]. Accordingly, statements concerning recruitment into genetic studies, such as ‘participants met current diagnostic criteria for autism and were included only if they had a minimal developmental level of 18 months on the Vineland Adaptive Behavior Scale Score, or an IQ equivalent >35. These minimal developmental levels assure that ADI-R results are valid and reduce the likelihood of including individuals with only severe mental retardation’ [1], must be regarded as misleading: there is no assurance of the ‘validity’ of the ADI-R (Box 1) [40] at such low levels of attainment, whatever that concept means in terms of neuropathological processes. If the criteria by which autism is diagnosed are arbitrary, at which point should the line be drawn for clinical purposes and genetic studies? In the clinic, it is important to know what degree of severity of autistic symptoms is associated with significant functional impairment. But this information is not currently available with regard to the effect on functioning of the various components of the phenotype, as conventionally defined. For genetic studies, theory suggests that it would be better to choose quantitative traits for gene mapping, rather than categorically defined cases, preferably using extended pedigrees of families that are stratified to be genetically more homogeneous. How can genetic homogeneity be ensured? One approach has been to search for endophenotypes that might be more closely associated to underlying genetic liability than symptoms. However, if the aim is to search for genes that specifically influence susceptibility to the neurocognitive deficits associated with the autistic phenotype, it would make more sense to focus on samples of individuals with normal-range intelligence and good structural language skills. Whether genetic risk factors are different in females and males is an issue yet to be resolved. The remarkable sex difference in prevalence has, to date, been an uncomfortable, yet obvious, fact that must have relevance to aetiology but is rarely discussed. Genetic homogeneity, although elusive, could be easier to detect in high-functioning autistic individuals, who are arguably in the majority (although still rarely detected). Acknowledgements I thank Tony Charman and Dorothy Bishop for their helpful and incisive comments on various drafts of the manuscript and Rebecca Chilvers whose clinical insights have contributed to the development of many of the ideas expressed herein. Particular thanks are due to the families of autistic children who have contributed to our research during the past decade, work that is supported by the Wellcome Trust, Nancy Lurie Marks Family Foundation, National Alliance for Autism Research and Simons Foundation.



TRENDS in Genetics

Vol.23 No.8

References 1 Ma, D.Q. et al. (2006) Dissecting the locus heterogeneity of autism: significant linkage to chromosome 12q14. Mol. Psychiatry 12, 376–384 2 Sebat, J. et al. (2007) Strong association of de novo copy number mutations with autism. Science 316, 445–449 3 Szatmari, P. et al. (2007) Mapping autism risk loci using genetic linkage and chromosomal rearrangements. Nat. Genet. 39, 319–328 4 Bartlett, C.W. and Vieland, V.J. (2007) Accumulating quantitative trait linkage evidence across multiple datasets using the posterior probability of linkage. Genet. Epidemiol. 31, 91–102 5 Vieland, V.J. (2006) Thermometers: something for statistical geneticists to think about. Hum. Hered. 61, 144–156 6 Shao, Y. et al. (2002) Phenotypic homogeneity provides increased support for linkage on chromosome 2 in autistic disorder. Am. J. Hum. Genet. 70, 1058–1061 7 Terwilliger, J.D. and Goring, H.H. (2000) Gene mapping in the 20th and 21st centuries: statistical methods, data analysis, and experimental design. Hum. Biol. 72, 63–132 8 Constantino, J.N. and Todd, R.D. (2003) Autistic traits in the general population: a twin study. Arch. Gen. Psychiatry 60, 524–530 9 Ronald, A. et al. (2006) Genetic heterogeneity between the three components of the autism spectrum: a twin study. J. Am. Acad. Child Adolesc. Psychiatry 45, 691–699 10 Ronald, A. et al. (2006) Phenotypic and genetic overlap between autistic traits at the extremes of the general population. J. Am. Acad. Child Adolesc. Psychiatry 45, 1206–1214 11 Happe, F. et al. (2006) Time to give up on a single explanation for autism. Nat. Neurosci. 9, 1218–1220 12 Chakrabarti, S. and Fombonne, E. (2005) Pervasive developmental disorders in preschool children: confirmation of high prevalence. Am. J. Psychiatry 162, 1133–1141 13 Freitag, C.M. (2007) The genetics of autistic disorders and its clinical relevance: a review of the literature. Mol. Psychiatry 12, 2–22 14 State, M.W. (2006) A surprising METamorphosis: autism genetics finds a common functional variant. Proc. Natl. Acad. Sci. U. S. A. 103, 16621–16622 15 Skuse, D. et al. (2005) Measuring social-cognitive functions in children with somatotropic axis dysfunction. Horm. Res. 64 (Suppl. 3), 73–82 16 Blaxill, M.F. et al. (2003) Commentary: Blaxill, Baskin, and Spitzer on Croen et al. (2002), the changing prevalence of autism in California. J. Autism Dev. Disord. 33, 223–226 17 Croen, L.A. et al. (2002) The changing prevalence of autism in California. J. Autism Dev. Disord. 32, 207–215 18 Folstein, S. and Rutter, M. (1977) Infantile autism: a genetic study of 21 twin pairs. J. Child Psychol. Psychiatry 18, 297–321 19 Ritvo, E.R. et al. (1985) Concordance for the syndrome of autism in 40 pairs of afflicted twins. Am. J. Psychiatry 142, 74–77 20 Bailey, A. et al. (1995) Autism as a strongly genetic disorder: evidence from a British twin study. Psychol. Med. 25, 63–77 21 Steffenburg, S. et al. (1989) A twin study of autism in Denmark, Finland, Iceland, Norway and Sweden. J. Child Psychol. Psychiatry 30, 405–416 22 Le Couteur, A. et al. (1996) A broader phenotype of autism: the clinical spectrum in twins. J. Child Psychol. Psychiatry 37, 785–801 23 Baird, G. et al. (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 24 Newschaffer, C.J. et al. (2002) Heritable and nonheritable risk factors for autism spectrum disorders. Epidemiol. Rev. 24, 137–153 25 Bolton, P.F. et al. (1998) Autism, affective and other psychiatric disorders: patterns of familial aggregation. Psychol. Med. 28, 385–395 26 Bishop, D.V. et al. (2006) Characteristics of the broader phenotype in autism: a study of siblings using the children’s communication checklist-2. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 141, 117–122 27 Constantino, J.N. et al. (2006) Autistic social impairment in the siblings of children with pervasive developmental disorders. Am. J. Psychiatry 163, 294–296 28 Spiker, D. et al. (2002) Behavioral phenotypic variation in autism multiplex families: evidence for a continuous severity gradient. Am. J. Med. Genet. 114, 129–136 29 Starr, E. et al. (2001) A family genetic study of autism associated with profound mental retardation. J. Autism Dev. Disord. 31, 89–96

30 Scourfield, J. et al. (1999) Heritability of social cognitive skills in children and adolescents. Br. J. Psychiatry 175, 559–564 31 Skuse, D.H. et al. (2005) Measuring autistic traits: heritability, reliability and validity of the Social and Communication Disorders Checklist. Br. J. Psychiatry 187, 568–572 32 Allison, C. et al. (2007) The Childhood Asperger Syndrome Test (CAST): Test-retest reliability in a high scoring sample. Autism 11, 173–185 33 Sung, Y.J. et al. (2005) Genetic investigation of quantitative traits related to autism: use of multivariate polygenic models with ascertainment adjustment. Am. J. Hum. Genet. 76, 68–81 34 Kolevzon, A. et al. (2004) Familial symptom domains in monozygotic siblings with autism. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 129, 76–81 35 Silverman, J.M. et al. (2002) Symptom domains in autism and related conditions: evidence for familiality. Am. J. Med. Genet. 114, 64–73 36 Dawson, G. et al. (2006) Quantitative assessment of autism symptomrelated traits in probands and parents: Broader Phenotype Autism Symptom Scale. J. Autism Dev. Disord. 37, 523–536 37 Folstein, S.E. et al. (1999) Predictors of cognitive test patterns in autism families. J. Child Psychol. Psychiatry 40, 1117–1128 38 Posserud, M.B. et al. (2006) Autistic features in a total population of 79-year-old children assessed by the ASSQ (Autism Spectrum Screening Questionnaire). J. Child Psychol. Psychiatry 47, 167–175 39 Williams, J.G. et al. (2006) Systematic review of prevalence studies of autism spectrum disorders. Arch. Dis. Child. 91, 8–15 40 Le Couteur, A. et al., eds (2003) The Autism Diagnostic InterviewRevised (ADI-R), Western Psychological Services 41 Lord, C. et al. (2000) The autism diagnostic observation schedulegeneric: a standard measure of social and communication deficits associated with the spectrum of autism. J. Autism Dev. Disord. 30, 205–223 42 Viding, E. and Blakemore, S.J. (2007) Endophenotype approach to developmental psychopathology: implications for autism research. Behav. Genet. 37, 51–60 43 Honda, H. et al. (2005) Cumulative incidence of childhood autism: a total population study of better accuracy and precision. Dev. Med. Child Neurol. 47, 10–18 44 Tuchman, R. and Rapin, I. (2002) Epilepsy in autism. Lancet Neurol. 1, 352–358 45 Lainhart, J.E. et al. (2006) Head circumference and height in autism: a study by the collaborative program of excellence in autism. Am. J. Med. Genet. A. 140, 2257–2274 46 Skuse, D. (2006) Genetic influences on the neural basis of social cognition. Philos. Trans. R. Soc. Lond. B Biol. Sci. 361, 2129–2141 47 Frith, C.D. (2007) The social brain? Philos. Trans. R. Soc. Lond. B Biol. Sci. 362, 671–678 48 American Psychiatric Association (2000) Diagnostic and Statistical Manual of Mental Disorders. American Psychiatric Association 49 World Health Organization (1993) The ICD-10 Classification of Mental and Behavioural Disorders. World Health Organization 50 Duvall, J.A. et al. (2007) A quantitative trait locus analysis of social responsiveness in multiplex autism families. Am. J. Psychiatry 164, 656–662 51 Gotham, K. et al. (2006) The autism diagnostic observation schedule: revised algorithms for improved diagnostic validity. J. Autism Dev. Disord. 37, 613–627 52 Harvey, C.G. et al. (2007) Sequence variants within exon 1 of MECP2 occur in females with mental retardation. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 144, 355–360 53 Belmonte, M.K. and Bourgeron, T. (2006) Fragile X syndrome and autism at the intersection of genetic and neural networks. Nat. Neurosci. 9, 1221–1225 54 LaSalle, J.M. et al. (2005) Rett syndrome: a Rosetta stone for understanding the molecular pathogenesis of autism. Int. Rev. Neurobiol. 71, 131–165 55 Hinton, V.J. et al. (2006) Social behavior problems in boys with Duchenne muscular dystrophy. J. Dev. Behav. Pediatr. 27, 470–476 56 Bolton, P.F. (2004) Neuroepileptic correlates of autistic symptomatology in tuberous sclerosis. Ment. Retard. Dev. Disabil. Res. Rev. 10, 126– 131 57 Laws, G. and Bishop, D. (2004) Pragmatic language impairment and social deficits in Williams syndrome: a comparison with Down’s








64 65


TRENDS in Genetics

Vol.23 No.8

syndrome and specific language impairment. Int. J. Lang. Commun. Disord. 39, 45–64 Veltman, M.W. et al. (2005) Autism spectrum disorders in Prader-Willi and Angelman syndromes: a systematic review. Psychiatr. Genet. 15, 243–254 Vorstman, J.A. et al. (2006) The 22q11.2 deletion in children: high rate of autistic disorders and early onset of psychotic symptoms. J. Am. Acad. Child Adolesc. Psychiatry 45, 1104–1113 Antshel, K.M. et al. (2006) Autistic spectrum disorders in velo-cardio facial syndrome (22q11.2 Deletion). J. Autism Dev. Disord, DOI: 10.1007/s10803-006-0308-6 Carter, J.C. et al. (2007) Autistic-spectrum disorders in down syndrome: further delineation and distinction from other behavioral abnormalities. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 144, 87–94 Creswell, C. and Skuse, D. (2000) Autism in association with Turner syndrome: implications for male vulnerability. Neurocase 5, 511–518 van Rijn, S. et al. (2006) X Chromosomal effects on social cognitive processing and emotion regulation: a study with Klinefelter men (47,XXY). Schizophr. Res. 84, 194–203 Geerts, M. et al. (2003) The XYY syndrome: a follow-up study on 38 boys. Genet. Couns. 14, 267–279 Patwardhan, A.J. et al. (2002) Reduced size of the amygdala in individuals with 47,XXY and 47,XXX karyotypes. Am. J. Med. Genet. 114, 93–98 Sikora, D.M. et al. (2006) The near universal presence of autism spectrum disorders in children with Smith-Lemli-Opitz syndrome. Am. J. Med. Genet. A. 140, 1511–1518

67 Gillberg, C. et al. (2006) Brief report: ‘the autism epidemic’. The registered prevalence of autism in a Swedish urban area. J. Autism Dev. Disord. 36, 429–435 68 Muhle, R. et al. (2004) The genetics of autism. Pediatrics 113, e472– e486 69 Jamain, S. et al. (2002) Y chromosome haplogroups in autistic subjects. Mol. Psychiatry 7, 217–219 70 Pickles, A. et al. (2000) Variable expression of the autism broader phenotype: findings from extended pedigrees. J. Child Psychol. Psychiatry 41, 491–502 71 Skuse, D.H. et al. (1997) Evidence from Turner’s syndrome of an imprinted X-linked locus affecting cognitive function. Nature 387, 705–708 72 Knickmeyer, R. et al. (2006) Fetal testosterone and empathy. Horm. Behav. 49, 282–292 73 Schellenberg, G.D. et al. (2006) Evidence for multiple loci from a genome scan of autism kindreds. Mol. Psychiatry 11, 1049–1060 74 Bonora, E. et al. (2005) Mutation screening and association analysis of six candidate genes for autism on chromosome 7q. Eur. J. Hum. Genet. 13, 198–207 75 Sutcliffe, J.S. et al. (2005) Allelic heterogeneity at the serotonin transporter locus (SLC6A4) confers susceptibility to autism and rigid-compulsive behaviors. Am. J. Hum. Genet. 77, 265– 279 76 Skuse, D.H. (2000) Imprinting, the X-chromosome, and the male brain: explaining sex differences in the liability to autism. Pediatr. Res. 47, 9–16 77 Owen, M.J. et al. (2005) Schizophrenia: genes at last? Trends Genet. 21, 518–525

Elsevier joins major health information initiative Elsevier has joined with scientific publishers and leading voluntary health organizations to create patientINFORM, a groundbreaking initiative to help patients and caregivers close a crucial information gap. patientINFORM is a free online service dedicated to disseminating medical research. Elsevier provides voluntary health organizations with increased online access to our peer-reviewed biomedical journals immediately upon publication, together with content from back issues. The voluntary health organizations integrate the information into materials for patients and link to the full text of selected research articles on their websites. patientINFORM has been created to enable patients seeking the latest information about treatment options online access to the most up-to-date, reliable research available for specific diseases.

For more information, visit