Heritability of autism spectrum disorders: a meta-analysis of twin studies

Abstract

Background: The etiology 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 meta-analysis of all twin studies of ASD published to date and explores the etiology along the continuum of a quantitative measure of ASD.

Methods: A PubMed Central, Science Direct, Google Scholar, Web of Knowledge structured search conducted online, 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. In addition, a quantile DF extremes analysis was carried out on Childhood Autism Spectrum Test scores measured in a population sample of 6,413 twin pairs including affected twins.

Results: The meta-analysis correlations for monozygotic twins (MZ) were almost perfect at .98 (95% Confidence Interval, .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 Phenotype of ASD) and increased to .67 (95% CI .61-.72) when applying a prevalence rate of 1%. The meta-analytic heritability estimates were substantial: 64-91%. Shared environmental effects became significant as the prevalence rate decreased from 5-1%: 07-35%. The DF analyses show that for the most part, there is no departure from linearity in heritability.

Conclusions: We demonstrate that: (a) ASD is due to strong genetic effects; (b) shared environmental effects become significant as a function of lower prevalence rate; (c) previously reported significant shared environmental influences are likely a statistical artefact of overinclusion of concordant DZ twins.

Keywords: Autism spectrum disorders; DF extremes analysis; heritability; meta-analysis; twin studies.

Figure 1

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 Autism Spectrum Disorder 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, 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)

Figure 2

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 Autism Spectrum Disorder to 5% in Study 6 & Study 9. In Meta‐analysis 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)

Figure 3

DF group heritabilities of the Childhood Autism Spectrum Test scores at age 8. The x‐axis denotes the cut off points applied to generate the groups for the DF analyses: the proportions represent the cumulative probability at the right‐hand side of the distribution, using the z‐values below as cut off points. The horizontal line drawn at and above the upper 95% CI of the group heritabilities at the higher end of the distribution (5–1%), serves as a visual guide to illustrate the groups which the high end estimates are not overlapping with