This study used behavioral genetic structural equation modeling to explore the etiology of nonverbal communication and social dysfunction data from individuals 18 years old or younger who participated in AGRE and had a PDD diagnosis. The analyses suggested that common environmental factors were not influential and that the variation between siblings/twins was better explained by genetic and unique environmental factors for both phenotypes. Dominant genetic effects were stronger than additive genetic effects. Broad heritability estimates were 45% and 28% for nonverbal communication deficits and social dysfunction, respectively. The role of genetic factors is consistent with findings that PDDs themselves are highly genetic (Rutter et al., 1999
) and adds to results from correlational studies on the familiality of specific characteristics. The strong contribution of unique environmental factors, particularly for social interaction scores, highlights the potential importance of treatment.
Unfortunately, because this was only the fourth study to use behavior genetic modeling with a sample that included individuals with PDDs, there is limited basis for comparison of our heritability estimates. The other studies were interested in overall liability of autistic traits and symptoms in the general population; dependent variables were based on parent report questionnaire data; and they included primarily unaffected and subclinical individuals (Constantino & Todd, 2003
; Ronald et al., 2006a
; Sung et al., 2005
). In contrast, the present study focused on only autism-affected individuals and examined liability of clinical characteristics relevant to treatment response from a gold-standard interview measure. In addition, Sung et al.’s (2005)
study did not include twins, Ronald et al.’s (2006a)
study only included 44 MZ twins with unconfirmed PDD, and Constantino and Todd’s (2003)
study included 13 twins (no report of how many were MZ) with unconfirmed PDD, whereas our study was based on 94 MZ twins with confirmed PDD. Therefore, differences between studies can be attributed to different aims and thus differing methodologies, as well as power. Sung et al. (2005)
found much lower heritability estimates for their scales that were most similar to our nonverbal communication and social measures. This is expected, given that their focus was only on siblings and unaffected family members. Our results are more similar to those in Ronald et al.’s (2006a)
study, which also indicated no evidence for shared environmental influences yet somewhat higher heritability estimates for different measures of social and communication impairment (which included impairment in the verbal domain). Constantino and Todd (2003)
differed from both our study and Ronald et al.’s (2006a)
study in that they found evidence for shared environment for total social responsiveness scores.
Our analyses also tested the degree to which the phenotypes were influenced by the same
genetic and environmental factors. Results indicated a very strong phenotypic correlation between nonverbal communication and social scores that was explained by both common genetic and environmental influences. This finding differs from Ronald et al’s (2006a)
study, which found low correlations between areas of autistic traits and limited evidence for shared etiological factors. This difference may be explained by Ronald et al’s (2006a)
use of mostly unaffected individuals, given that cognitive, language, and social abilities tend to be more closely correlated in children with PDD than the general population (Carpenter et al., 2002
; Dyck et al., 2006
). Ronald et al. (2006b)
recently completed an additional analysis of their data to examine individuals at the extreme who showed many autistic traits. These results were more consistent with our current findings, in that they found a stronger relationship between domains (correlations of .58 for males and .51 for females) as well as some genetic overlap, in their more extreme sample than the earlier sample comprised of primarily unaffected individuals (Ronald et al., 2006a
). The fact that our data produced an even stronger correlation in an even more extreme sample (clinically diagnosed individuals), with significant evidence for common genetic and unique environmental influences between social interaction and nonverbal communication, further suggests differences in the interrelatedness and etiology of autistic symptom domains in clinically affected versus unaffected individuals.
Results indicated that the most parsimonious and best-fitting explanation of the data did not include gender differences. This should be considered preliminary because of the fairly small number of female MZ twins, which might have limited our power to detect small differences. However, the pattern of correlations was similar for males and females, so the lack of significant gender effects is consistent with the pattern of data. Our results are different from those in one study, which found that equating genders significantly reduced the fit of genetic models for autistic traits; yet, that study had an even smaller number of female MZs with PDDs and also measured the traits in unaffected individuals (Ronald et al., 2006a
Our correlational results can be compared to three previous studies that tested ICCs on the same scales among multiplex families. The magnitude of the relationship between MZ twins for ADI nonverbal communication scores found in this study (.57 for males and .47 for females) corroborates a study of 33 MZ twins that found a nearly identical ICC for this same measure (.56; Kolevzon et al., 2004
). The MZ correlation for ADI social dysfunction was somewhat lower compared to Kolevzon et al.’s (2004)
findings. The low correlation between DZ twins and siblings found in our study for ADI social scores is consistent with previous research that has found this measure to have low, non-significant correlations between affected siblings (MacLean et al., 1999
; Silverman et al., 2002
). Our correlations for ADI nonverbal communication scores among affected siblings/DZ twins were somewhat lower than previous studies that found ICCs of .39 and .19 for sample sizes of 94 and 457 respectively, compared to our correlations of .12 for males and .05 for females.
Although this study is one of the largest to date of individuals with PDDs, the number of MZ twins was still relatively small for structural equation modeling. Therefore, there was limited power, particularly for gender comparisons. This may be why heritability estimates from the modeling were somewhat lower than would be expected based on the correlation patterns between MZ twins and DZ twins/siblings. In addition, variance for ADI scores was restricted, given that all participants exceeded certain cut-offs on these scales to meet criteria for a PDD. Although restricted variance can reduce power by reducing effect size, it also means that even small to moderate effects are noteworthy.
Aspects of the sample should be considered when interpreting and generalizing findings. The authors recognize that use of a selected sample (e.g. selecting children with diagnoses of PDD as opposed to an unselected population sample) when conducting this type of behavior genetic analysis limits generalizability of findings; estimates can be generalized to the population of children with PDDs/autism but would be biased if generalized to the general population. This is consistent with the study aim to explore the genetic and environmental underpinnings of clinical symptom domains in affected individuals with autism. Studies focused explicitly on clinically diagnosed individuals are important given accumulating evidence of different etiologic and phenotypic patterns regarding social interaction and communication development in clinically diagnosed individuals. It should also be taken into consideration that the sample included predominantly multiplex families.
We were unable to account for some potential covariates. Given power considerations and the complexity of the data analyses, age corrections were not implemented. The ADI scores were based on responses for when the participant was four or five years old, but there might have been differences in recall of parents based on the participant’s current age. Finally, AGRE did not collect data on whether a participant had undergone treatment.
Despite the aforementioned limitations, this study has several important strengths. We employed a methodology fairly new to PDD studies; and it is, to our knowledge, one of only five studies to investigate phenotypic congruence among AGRE siblings (Constantino et al., 2006
; Goin-Kochel, Mazefsky, Riley, & AGRE, under review; Kolevzon et al., 2004
; Silverman et al. 2002
) compared to over 85 AGRE-based molecular studies. Phenotypic studies are critical in order to clarify how varying patterns of genetic and environmental inheritance correspond to different phenotypes among the population. Such studies are necessary so that more informed, empirically based decisions can be made for sample selection in molecular studies by choosing stratification characteristics with a large genetic influence. Our study begins to address whether ADI nonverbal communication and social domains may be an important endophenotype for molecular genetic studies by demonstrating that the etiology of these phenotypes involves genetic factors as well as the presence of unique environmental factors. Results indicate that nonverbal communication skills may be more genetically driven, and therefore a better stratification choice than social interaction.
These phenotypes are also important because of their relation to treatment response (Sallows & Graupner, 2005
). Developing a better understanding of etiological factors related to treatment-response predictors is critical, given that poor post-treatment outcomes are correlated with poor intake measures, suggesting that there is a group of children with autism for whom intensive behavioral treatment is not effective on its own. Developing a better idea of etiological factors and characteristics related to treatment response will also help improve decisions regarding intervention type. The large (72%) unique environmental influence on social interaction that was found in this study suggests that treatment may have a relatively greater impact on social skills than nonverbal communication which is more genetically driven.
Future studies can build on our results by replicating them in samples that include more MZ twins and females. Next steps should also include genetic analyses of subgroups of children who have participated in various treatment programs and have an identified response pattern.