The results of the long-awaited California Autism Twin Study (CATS) have been published. The article, Genetic Heritability and Shared Environmental Factors Among Twin Pairs With Autism appears in the Archives of General Psychiatry and is open-access (i.e. free). The study set out “To provide rigorous quantitative estimates of genetic heritability of autism and the effects of shared environment.”
The basic idea of a twin study is fairly straighforward: “identical” (or monozygotic, MZ) twins share 100% of their DNA, “fraternal” (dizygotic, or DZ) twins share about 50% of their DNA. If a condition is purely genetic, identical twins will both have the condition or not have the condition. The percentage of twins with a condition is called concordance. High concordance implies a highly genetic condition.
Early twin studies showed a high concordance. These studies were relatively small. The first study, Infantile autism: a genetic study of 21 twin pairs, was published in 1977 and included, as you might guess, only 21 twin pairs. Of these, 10 DZ and 11 MZ twins. There were three twin studies which made up the main body of knowledge suggesting a strong heritability of autism, including a total of 56 pairs.
The California Twin study sought to take a much closer look. First by including a much greater number of twin pairs, and second by including modern diagnostic methods. They used the California Department of Developmental Services as a resource to identify twin pairs with at least one autistic. The CDDS database has advantages and disadvantages. First, the database is large and covers fairly diverse state. This is important when you consider that (a) only about 1% of the population is autistic and (2) only about 3.2% of the population are twins. That means about 1 in 3000 are twins+autistic.
The disadvantages of the CDDS as a resource include the fact that not everyone with an ASD qualifies for services or is correctly classified. Of those who qualify, inclusion in the CDDS is voluntary. Technically, the CDDS serves only those with autistic disorder or those in the “fifth category” which includes “…disabling conditions found to be closely related to mental retardation or requiring treatment similar to that required for individuals with mental retardation.” This would limit the ability to identify ASD concordance. For example, twins who both have Asperger syndrome.
One major strength of the study was the use of clinical assessments plus parent interviews for diagnosis. They did not rely upon the CDDS reports for diagnoses. Also, the study determined twin status (MZ or DZ) through genetic testing, something which wasn’t available when the first twin studies were performed.
Here is the abstract for the study:
Context Autism is considered the most heritable of neurodevelopmental disorders, mainly because of the large difference in concordance rates between monozygotic and dizygotic twins.
Objective To provide rigorous quantitative estimates of genetic heritability of autism and the effects of shared environment.
Design, Setting, and Participants Twin pairs with at least 1 twin with an autism spectrum disorder (ASD) born between 1987 and 2004 were identified through the California Department of Developmental Services.
Main Outcome Measures Structured diagnostic assessments (Autism Diagnostic Interview–Revised and Autism Diagnostic Observation Schedule) were completed on 192 twin pairs. Concordance rates were calculated and parametric models were fitted for 2 definitions, 1 narrow (strict autism) and 1 broad (ASD).
Results For strict autism, probandwise concordance for male twins was 0.58 for 40 monozygotic pairs (95% confidence interval [CI], 0.42-0.74) and 0.21 for 31 dizygotic pairs (95% CI, 0.09-0.43); for female twins, the concordance was 0.60 for 7 monozygotic pairs (95% CI, 0.28-0.90) and 0.27 for 10 dizygotic pairs (95% CI, 0.09-0.69). For ASD, the probandwise concordance for male twins was 0.77 for 45 monozygotic pairs (95% CI, 0.65-0.86) and 0.31 for 45 dizygotic pairs (95% CI, 0.16-0.46); for female twins, the concordance was 0.50 for 9 monozygotic pairs (95% CI, 0.16-0.84) and 0.36 for 13 dizygotic pairs (95% CI, 0.11-0.60). A large proportion of the variance in liability can be explained by shared environmental factors (55%; 95% CI, 9%-81% for autism and 58%; 95% CI, 30%-80% for ASD) in addition to moderate genetic heritability (37%; 95% CI, 8%-84% for autism and 38%; 95% CI, 14%-67% for ASD).
Conclusion Susceptibility to ASD has moderate genetic heritability and a substantial shared twin environmental component.
The concordance values are 77% for male MZ twins and 50% for female MZ twins. These values are high. But the heritability estimates are influenced not just by the MZ twin concordance, but by the difference between MZ and DZ concordance. Consider if 100% of “identical” twins had a condition. This doesn’t necessarily mean that the condition is fully caused by hertiablity if, say, 100% of the “fraternal” twins also had the condition. In the California Twin study, DZ concordance values were relatively high (31% for males and 36% for females).
Based on these values, the authors calculate that shared environment accounts for 55% of the risk of autism and 37% for genetic heritability. There are big “error bars” (or confidence intervals) for the concordance and heritablity values. It is worth noting that the concordance values here are similar to those reported recently by Bearman’s team at Columbia (40-50% pairwise concordance for MZ twins. This study also used the CDDS as a resource, but did not use clinical assessments) and Goldsmith’s group at the University of Wisconsin (43% pairwise concordance). These studies also concluded that genetic heritability was lower than previously thought.
The authors concluded their paper:
Our study provides evidence that the rate of concordance in dizygotic twins may have been seriously underestimated in previous studies and the influence of genetic factors on the susceptibility to develop autism, overestimated. Because of the reported high heritability of autism, a major focus of research in autism has been on finding the underlying genetic causes, with less emphasis on potential environmental triggers or causes. The finding of significant influence of the shared environment, experiences that are common to both twin individuals, may be important for future research paradigms. Increasingly, evidence is accumulating that overt symptoms of autism emerge around the end of the first year of life. Because the prenatal environment and early postnatal environment are shared between twin individuals, we hypothesize that at least some of the environmental factors impacting susceptibility to autism exert their effect during this critical period of life. Nongenetic risk factors that may index environmental influences include parental age, low birth weight, multiple births, and maternal infections during pregnancy. Future studies that seek to elucidate such factors and their role in enhancing or suppressing genetic susceptibility are likely to enhance our understanding of autism.
The authors can be heard on KQED in San Francisco discussing the paper.
In a companion commentary on Archives of General Psychiatry, Is Autism, at Least in Part, a Disorder of Fetal Programming?, Peter Szatmari comments on the shift in perception based on the new heritability estimates: “The autism field can now join the chorus and ask, “Where did all the heritability go?” We now appear to have an answer, at least in part: those original estimates were inflated.”
“Inflated” seems a bit of a loaded word. There are big error bars on the current study, there were even bigger ones in the old studies. What I found very interesting was this peak at an upcoming paper by the Baby Siblings Research Consortium:
The high DZ concordance rate is consistent with estimates reported from the Baby Siblings Research Consortium. In these studies, infant siblings of children with ASD are followed up from birth to age 36 months so that risk can be calculated in a prospective fashion. The latest report from the Baby Siblings Research Consortium suggests that the risk is upwards of 20%, which is slightly less than the DZ rate provided by Hallmayer and colleagues (31%-36% for ASD). The confidence intervals of the estimates overlap to be sure, but this may also suggest potential ascertainment bias for concordant as opposed to nonconcordant DZ twin pairs or that twinning itself is a risk factor for ASD.
(reference: Ozonoff S, Young G, Carter AS, Messinger D, Yirmiya N, Zwaigenbaum L, Bryson SE, Carver L, Constantino J, Dobkins K, Hutman T, Iverson J, Landa R, Rogers S, Sigman M, Stone W. Recurrence risk for autism spectrum disorders: a Baby Siblings Research Consortium Study. Pediatrics. In press.)
They commentary poses some good questions about potential bias in the California Autism Twin Study:
One must ask whether there was any bias artificially reducing concordance in the study by Hallmayer and colleagues. One possibility is missing data, which are substantial. Another possibility is differential misclassification. It is not inconceivable that parents are more likely to rate MZ twins with a disability as more dissimilar than DZ twins. There may be an unconscious effort on their part to see one twin as less affected than the other. Such reporting effects, known as sibling deidentification, have been reported for twin studies of temperament.8 It would be extremely interesting to see whether concordance rates differ by instrument, ie, the Autism Diagnostic Interview–Revised, which is based on parental report, vs the Autism Diagnostic Observation Schedule, which is based on an independent observer. If the sensitivity of the measurement tool is less in MZ twins than in DZ twins, a smaller difference in observed concordance rates would be expected compared with true concordance rates based on no measurement error.
A third threat to the validity of the findings is that twinning itself might be a risk factor for ASD, so that the heritability estimates generated would not be generalizable to the population of nontwin children with ASD. There may be some factor associated with twinning such as maternal age, coming from a monochoreonic placenta, prematurity, or in vitro fertilization that could place twins at risk for ASD. There is in fact some evidence that twins have a higher rate of autism than nontwins,9 but further work needs to be undertaken to provide better evidence.
It’s taken 34 years since the first autism twin study until the much larger California Autism Twin Study. This has been a time consuming and expensive undertaking. I doubt there will be a chance in the near future for a larger study to address the issues laid out above. This doesn’t preclude some other studies to clarify some questions, but the California Autism Twin Study will likely stand for some time before being challenged or confirmed.