News The latest developments in autism research.

Twin study suggests autism traits inherited independently

by  /  29 September 2011
Double trouble: Genetic and environmental factors may both contribute to the uneven distribution of autism-associated traits in twins.

A new study of nearly 6,000 pairs of twins suggests that the three core traits of autism are are inherited separately, and to varying degrees, both in individuals with autism and in the general population.

The study, which appeared online 7 September in Behavioral Genetics,is based on a sample of 12-year-old participants from the Twins Early Development Study, a longitudinal study of twins born in the U.K. between 1994 and 19961.

Of the nearly 12,000 children in the new study, 80 children have a suspected autism spectrum disorder and score above the 95th percentile on each domain of the Childhood Autism Spectrum Test (CAST). The CAST is a 31-item screening tool filled out by parents that assesses symptoms of the three core features — social deficits, communication impairments and restricted and repetitive behaviors — associated with autism.

Each of these three domains is 72 to 76 percent heritable in males, and 58 to 74 percent heritable in females, the researchers found. These estimates are based on the degree of concordance, or agreement, in core autism traits between twins.

Males show more genetic overlap than females do on social and communication impairments in the sample as a whole, as well as among individuals with the highest scores on the CAST, the researchers report.

“If you measure the three parts of the triad in the general population, you get kids with autism who have all three sets of symptoms, children who have none and children who have one or two,” says lead investigator Angelica Ronald, assistant professor at the University of London’s Center for Brain and Cognitive Development in Birbeck. “The behaviors are fractionated out, autism being the almost rare example when you get all three together.”

Unmatched sets:

Ronald and her colleagues Francesca Happé and Robert Plomin first proposed the ‘fractionable autism triad’ in 20062. This hypothesis suggests that there are not only different genes for different autism-related symptoms, but that the behaviors associated with the disorder are also ‘fractionable,’ appearing at different frequencies and to varying severity in the general population.

Results from the new study validate those from two 2006 twin studies, conducted when the twins were 7 to 8 years old. One of those studies reported limited genetic overlap between symptoms of social impairment and restricted and repetitive behavior3. The other looked at communication impairments and found that most genetic influences are specific to each type of impairment4. The new study includes many of the same twin pairs, but is larger, and investigates gender differences in the heritability of the three domains.

The researchers calculated the degree of concordance in each set of symptoms between both monozygotic, or identical, twins, who share the same genetic code, and dizygotic, or fraternal, twins, who are as genetically similar as any other siblings, but who share the same uterine environment.

Their aim was to tease out the degree of genetic and environmental influences for each component of the autism triad between twins. “You can split environment into two types, shared and non-shared,” Ronald says. “It’s usually the non-shared effects that influence autism and autism traits.”

These can include everything from biochemical events in utero that affect only one twin, to birth complications and childhood accidents, she says.

Studies of twins first established the heritability of autism, and still play an important role in pointing to possible causes of the disorder, though they do not typically include DNA analysis.

“We need twin studies to tackle this problem from both a genetic and environmental standpoint,” says Joachim Hallmayer, associate professor of psychiatry at Stanford University in Palo Alto, California who was not involved with the new work.

In another twin study published earlier this year, Hallmayer reported that 38 percent of autism spectrum disorders are caused by genetic factors and 58 percent by environmental factors5. Other researchers challenged those results, saying that 192 twin pairs, of whom at least one twin has autism, is far too small a sample on which to base such a broad conclusion.

Geneticists also say that the study underplayed the importance of rare new mutations that might appear in only one twin.

Traveling companions:

Hallmayer says the results from the new study support his findings. “Clearly, there is quite a substantial amount of environmental factors involved,” he says.

The new study’s findings also raise a number of questions that still need explanations, Hallmayer says. “Why do these traits co-occur in autism?” he asks. “There must be something that ties them together.”

Ronald says molecular studies of genes such as CNTNAP2, which has been linked to the timing of language development in toddlers, as well as to autism and specific language impairment, provide experimental support for the fractionable autism hypothesis.

“CNTNAP2 is not necessarily associated with autism as a whole, but with language difficulties in autism, which is exactly in line with the fractionable autism triad hypothesis,” she says.

However, other epidemiological studies with a similarly large number of participants have shown that although each cluster of symptoms associated with autism does exhibit some degree of independence, the traits tend to “travel together,” says John Constantino, professor of psychiatry and pediatrics at Washington University in St. Louis.

Children with disorders such as specific language impairment and obsessive-compulsive disorder may also score high on one or more of the triad of symptoms, he points out. This might have contributed to the independent occurrence of symptoms in each category in the new study, he says.

Still, the high correlation between social and communication traits in males in the sample is a provocative finding, he says. “It may be an important and interesting lead to pursue in order to understand gender disparities in the [disorder].”



1: Robinson E.B. et al. Behav. Genet. Epub ahead of print (2011) PubMed

2: Happé F. et al. Nat. Neurosci. 9, 1218-1220 (2006) PubMed

3: Ronald A. et al. J. Am. Acad. Child Adolesc. Psychiatry 45, 691-699 (2006) PubMed

4: Ronald A. et al. J. Am. Acad. Child Adolesc. Psychiatry 45, 1206-1214 (2006) PubMed

5: Hallmayer J. et al. Arch. Gen. Psychiatry Epub ahead of print (2011) PubMed

14 responses to “Twin study suggests autism traits inherited independently”

  1. Jon Brock says:

    I spotted this paper the other day on your Papers of the Week (thanks for that, by the way, a great resource).

    The big problem with the paper for me is that it relies on a questionnaire measure (the CAST), which the authors have divided up (based it seems on their own intuitions) to give three separate scales, corresponding to the three arms of the autistic triad. However, by their own admission, the questionnaire can’t be used in this way.

    In this paper, they state that:

    “In this sample the domain-specific subscales exhibited the following internal consistency reliabilities: SI 0.56 (11 items); CI 0.65 (12 items); RRBI 0.48 (7 items).”

    These values are considered to be “poor”, “questionable”, and “unacceptable” respectively. Essentially, none of the three subscales are measuring behaviours that hang together.

    In an earlier review, Happe and Ronald (2008) report that:

    “For both the full and abbreviated CAST versions, exploratory principal component factor analyses led to between four and eight factors.”

    Yet they still argue for a “separable” autistic triad (three factors), and continue to use the three CAST “subscales” as the basis for evaluating heritability models.

    I don’t get it.

  2. RAJensen says:

    The Plomin group in the UK (Happe, Arnold and others) use the classical twin study design to calculate their heritability estimates. The classical twin study design method is over one hundred years old and is based on a flawed assumption which is that ‘genetics’ always equals ‘inheritance’. The study of Downs Syndrome should have laid this flawed assumption to rest years ago. In Downs Syndrome MZ twins are concordant and DZ twins are discordant yet if you calculated a ‘heritability’ estimate based on the difference in concordance rates between MZ and DZ twin pairs then Downs Syndrome, not autism, would be the most heritable of the developmental disorders. The study that has driven the genetic models since 1995 is the British twin study which reported a concordance rate for MZ twin pairs for narrowly defined autism of 60% and when concordance for a broader definition was included the study calculated a concordance rate and a heritability estimate of over 90%. The British twin study included twins who participated in a previous twin study (1977) and added new twins recuited from clinics throughout the UK. Any of the newly recruited twins pairs where one or both twins had an identified medical cause were disqualified from participating. Three pairs of newly recruited twin pairs were disqualified, one pair each with Downs Syndrome, congenital rubella syndrome and Williams Syndrome, none of which is a heritable congenital syndrome. Williams Syndrome is the only syndrome among the three pairs disqualified where transgenerational inheritance does occur, but thansgenerational inheritance in Williams Syndrome represents a tiny fraction of cases and when it does occur the parent(s) are always unaffected as far as autism is concerned.

  3. RAJensen says:

    Furthermore, the article states that of the 12,000 children studied 80 had a suspected autism spectrum disorder which represents 0.00666 of the total group, remarkeably far below the usual and somewhat suspect concordance rates of 1.0 % in the general population. Either this was a misprint or the prevelance rates of 1% must be considered to be highly suspect since this was a population based study.

  4. RAJensen says:

    Correction to previous post.
    ‘somewhat suspect concordance rates of 1.0 % in the general population’.

    Should read ‘somewhat suspect ‘prevelance’ rates of 1.0 % in the general population’.

  5. RAJensen says:

    I apologize for the typo errors. Unforunately, posting to SFARI does not have an edit function after posting.
    To recapitulate, of the 12,000 children in the study 80 had a ‘suspected’ Autism Spectrum Disorder which is a .67 prevelance rate which is contradictory to the widely held belief that prevelance of auitsm in the general population is 1.0%. That represents 2/3rds of the putative 1% prevelance rates. This study should have found at least 120 cases of ‘suspected’ Autism Spectrum Disorder. Why this huge variance? The broadening of diagnostic criteria that began with the introduction of DSM-IV (1994) and ICD-10 (1994) has increased the ambiguity of autism definition to the point where there are no distinct boundaries between disorders (autism, schizophrenia, ADHD, childhhood language disorders,bi-polar disorder)and no distinct boundaries between disorder and normalcy.
    Given the growing ambiguity of autism definition no one can state exactly what the prevelance of autism is in the general population.
    This is a huge problem and DSM-5 doesn’t appear to have any answers towards solving the problem of autism definition.

  6. Elise Robinson says:

    The CAST total scale shows good reliability and validity (see Scott et al., 2002; Williams et al. 2005). However, as noted above, the reliabilities of the CAST subscales are low. The total CAST was split into three subscales to reflect the domains of the DSM-IV criteria for autism (social impairment, communication impairment, restricted/repetitive behaviors and interests). The low reliabilities are secondary to the heterogeneity of the symptoms in each of the DSM subdomains. For example, both motor stereotypies and repetitive interests fall within the domain of restricted and repetitive behaviors, but it is common for them to appear independently within a single individual. In a more extreme example, both absence of language and speech abnormalities qualify as communication impairments, but they are mutually exclusive. The CAST is designed to assess the broad variety of behaviors that fall under the ASD umbrella as defined in the DSM-IV. However, as those behaviors often appear unclustered within subdomains, the internal consistency reliability of the subscales is low.

    While inevitable to a certain extent, we recognize this to be a limitation of the research, as noted in the paper. Low reliabilities attenuate statistical associations (estimated correlations are lower than they would be if the scale was perfectly reliable – see discussion section for explanation of how this may have affected our results). However, the univariate subscale heritabilities were still very high. Accordingly, the low cross-domain heritabilities were not simply a function of measurement. A challenge of future research will be to consider how to measure autistic traits in a manner that both maximizes reliability and includes the full spectrum of behaviors that are associated with ASDs.

    • Jon Brock says:

      Hi Elise

      Thanks for the response and clarification.

      I’m not an expert on twin studies so I’m happy to be corrected. However, the issue really is one of validity rather than reliability.

      For example, I could make up a scale of completely unrelated phenotypes – hair colour, shoe size, nonverbal IQ etc and compile them into a single scale. I would still expect identical twins to have more similar scores to one another than non-identical twins, but it wouldn’t mean that the scale was meaningful in any way.

      I’m certainly not arguing against the general idea that different autism “symptoms” can be inherited independently. But by lumping together symptoms that don’t necessarily go together (as in the case of repetitive behaviours and restricted interests), you may well miss out on dependencies between symptoms that *are* inherited together.

      Say, for example, there was a perfect association between hand flapping and speech delay and say there was a single CNV that led to both behaviours. Obviously that’s not true (and is a gross oversimplification of what twin studies are looking for) but the point is that your analyses would almost certainly conclude that the two behaviours were inherited independently because both behaviours were pre-assigned to different domains.

      As I say, I’d love to be wrong (or be proved wrong in the future).

      • Elise Robinson says:

        Thanks Jon,

        We wonder about these questions as well, particularly the best manner in which to evaluate sources of phenotypic clustering. There seem to be circumstances under which the degree of etiologic overlap varies, both between measured genetic risks and between subgroups within the population (as we saw here).

        There are reasons to feel reassured about the validity of the CAST subscales, however, particularly as they were employed among extreme-scoring individuals in this analysis. Participants at or above the 95% of each subscale had mean scores almost as high as the 80 measured individuals with ASDs. The extreme scorers within each domain, on average, displayed much greater clustering within their domain(s) of impairment than across domains.

        Your point is taken with regard to the continuum, though, and is certainly one of the limitations of taking a DSM-based or top down approach to subphenotype definition. Angelica Ronald and colleagues recently published a paper in Molecular Psychiatry in which this very issue was addressed. They conducted a factor analysis of the autism–tics, AD/HD and other comorbidities inventory (A-TAC) and estimated etiologic overlap using the autism items as they clustered naturally. They similarly found that social impairment, communication impairment, and restricted/repetitive behaviors and interests (the factors that emerged) were predominantly genetically independent.

        • Anonymous says:

          Hi Elise again.

          I didn’t know about the Molecular Psychiatry paper (for anyone else reading, it’s available here and Table 1 on p25 is the relevant bit).

          You’re right that the clusters do fall out quite nicely there. I guess the obvious question is what happens to your data when you take the same approach? And why the CAST doesn’t give you a similar factor structure? Is it because the CAST has more items and so will support more factors?

          Not only do I not know about twin studies, I also don’t know a huge amount about factor analysis. So sorry for stupid questions – and thanks for a really interesting discussion.

  7. Just a parent says:

    I am stunned by the colossal waste of time and resources devoted to these ‘studies’.

    You guys are like someone standing by the river bank with a pen and paper, watching children drawn and making notes on the changing colour and shape of their clothes as they sink. You are playing with numbers. Nice.

    Next time you try to group our children, instead of unfalsifiable, useless prattling on about surface symptoms (do you actually get to MEET many children with autism I wonder?) please do us all a favor and focus on grouping things like types of immune dysfunction (anyone of you heard of massive immune abnormalities found in autism, things like low NK cells, abnormal cytokine profile, T cell abnormalities, autoimmune markers?), microglial inflammation, abnormalities in autonomic function, abnormal vasoconstriction, vascular endothelial inflammation, impaired blood flow to the brain, impaired methylation, depletion of glutathione, oxidative stress, mitochondrial dysfunction, impaired carbohydrate digestion, endotoxemia, celebral folate deficiency. How do those findings fit into your hereditability groups?

    In which hereditability subgroup would you put the child who stops self-harming after being treated for reflux? In which subgroup would you fit a child who stops ticking and OCD behaviors overnight after being treated with antihistamines?

    Our children are suffering, dying prematurely of seizures, wondering off, getting lost in the woods, getting run over by cars, killing themselves, killing their parents, drowning every day. You stand there and make notes.

    • Anonymous says:

      I might ask you whether you get to MEET many children with autism? As an autistic adult, I can tell you that I have none of those immune dysfunctions you list. What I do have are ‘surface symptoms’ that you seem to think are so irrelevant, as well as anxiety that is partly brought on by people like you grouping autistics as killers (which you do so disrespectfully in your last sentence).

  8. Just another parent says:

    At least Prof Nicholson has done some background reading of literature.

    “Autism is what we call a mosaic disease, so it has many different facets to it. I discovered that by reading a lot of papers on the subject; about 500 in the last year or so, covering all the different areas of autism research. That’s actually one of the things that most people working in autism DON’T TEND TO DO. THEY WORK IN THEIR OWN AREA AND THEY DON’T ACTUALLY LOOK ACROSS THE WHOLE SPECTRUM. And if you look into the literature, you’ll find that autism isn’t just a sort of neuropsychiatric, behavioural, and social disorder, but it’s also associated with gastrointestinal problems in the majority of children. Maybe 70% of the children at some time or other have quite severe gastrointestinal disturbances. But also, there are problems that have been recorded in cardiovascular systems. They have a raised median diastolic blood pressure. They have an abnormal QRS complex. That’s the actual electrical control of the heart. It looks as though they have some renal kidney transport defects as well. So, IT’S A SYSTEMIC DISEASE, but the most obvious effect is the social and behavioural, and so it tends to be associated with that. But it doesn’t mean to say that the other parts of the system aren’t deeply involved in the mechanism of action, and I think what we have to do now using our modern technology is to take a step back, look at the whole problem as a systemic problem, …”

Leave a Reply

Your email address will not be published. Required fields are marked *