Identification of Developmental and Behavioral Markers Associated With Genetic Abnormalities in Autism Spectrum Disorder

Phenotypic assessments and biological samples were collected from 12 university-based centers as part of the SSC. Probands with ASD were included if they were between 4 years and 17 years, 11 months of age, did not have any first-, second-, or third-degree relatives with ASD, and met criteria for autism, ASD, or Asperger syndrome based on the standard SSC assessment (see reference 19). Participants provided written informed consent (and assent, as appropriate) after receiving a complete description of the study.

A recent comprehensive, integrated analysis of transmitted and de novo variation in ASD identified 65 ASD-associated genes and an additional six ASD-associated loci with high confidence (false discovery rate ≤0.1) (8). Most evidence for ASD association came from dnLoF mutations or dnCNVs. Based on the results of Illumina genotyping array and whole exome sequencing data to identify dnLoF mutations and dnCNVs, we divided the SSC probands into three groups: 1) 112 probands with at least one dnLoF mutation or dnCNV in, or including, a high-confidence ASD gene or locus (the high-confidence group); 2) 292 probands with a dnLoF mutation or dnCNV, but not in, or including, a high-confidence ASD gene or locus (the low-confidence group); and 3) 1,751 probands with no dnLoF mutations or dnCNVs in any gene or locus (the none group). An additional 702 probands were excluded from these groups because they did not have both genotyping array and whole exome sequencing data available, and therefore we could not be sure of their mutation status.

The main analyses were conducted between the high-confidence group and a subset of 112 cases from the none group, matched on nonverbal IQ, age, and sex. We refer to these cases as “matched autism controls.” Figure 1 depicts the process by which participants were included in the high-confidence group or the matched autism control group. The participants’ demographic characteristics are summarized in Table 1. A list of the specific genetic abnormalities represented in the high-confidence group is available in Table S1 in the data supplement that accompanies the online edition of this article. We examined the high-confidence group as a whole, and we also identified seven dnLoF mutations or dnCNVs found in at least four participants; these have previously been reported separately, and they include 16q11.2 deletions and duplications, 15q11.2–13 duplications, 1q21.1 duplication, and 7q11.23 duplications (7), as well as DYRK1A LoF mutations (20) and CHD8 LoF mutations (16). Supplementary analyses compared individuals from the low-confidence group, who may later be identified as high confidence as further studies are completed, to a separate group of matched controls from the none group (see Table S2 and Figure S1 in the online data supplement).

Matched groups were compared on a number of phenotypic domains. Cognitive ability was indexed using nonverbal IQ and verbal IQ, which were derived from standardized tests administered according to the child’s ability level. Standard scores from the daily living skills domain of the Vineland Adaptive Behavior Scales, 2nd Edition (Vineland II) (21) provide a measure of independent functioning that can be used alongside cognitive ability to index the presence and severity of intellectual disability. Motor skills were measured using item 5 from the Autism Diagnostic Interview–Revised (ADI-R) (22), which inquired about age at onset of independent walking, and the raw scores from the Purdue pegboard task. Language was measured using age at first words (item 9) and age at first phrases (item 10) from the ADI-R; the module of the Autism Diagnostic Observation Schedule (ADOS) (23) that provides a gross estimate of expressive language level (module 1, nonverbal/single words; module 2, flexible phrase speech; and modules 3 and 4, fluent speech/regular use of complex sentences); the standard score on the Peabody Picture Vocabulary Test, 4th Edition (24); and the standard score on the Vineland-II communication domain. We also report a language deficit variable, coded as present when the child’s ADOS module was lower than what would be expected based on his or her nonverbal mental age. Social communication deficits and restricted and repetitive behaviors associated with ASD were measured using total scores from the social, communication, and repetitive behavior domains of the ADI-R and the domain-calibrated scores from the ADOS (25). The ADI-R domain scores are based on behaviors retrospectively reported by the parent to have occurred when the child was between ages 4 and 5 or ever in the past, whereas the ADOS is based on currently observed behaviors. Current level of overall ASD symptoms was assessed using total scores from the Social Responsiveness Scale (SRS) (26), the ADOS overall calibrated severity scores (27), and a clinician-rated measure of ASD diagnostic certainty (the minimum score was 6 in the presence of an ASD diagnosis, so SSC scores ranged from 6 to 15). Behavior problems not specific to ASD were measured using T scores for externalizing and internalizing problems from the Child Behavior Checklist (28), a parent-report questionnaire. Presence of seizures was assessed using combined information from the SSC medical history form and ADI-R item 85. Family history of major psychiatric problems was determined from the SSC medical history form, based on presence or absence of schizophrenia, bipolar disorder, or depressive disorder in a family member with a level of genetic relatedness at least on the order of first cousin (see reference 29).

A randomized “nearest neighbor” approach was used to match probands with dnLoF mutations or dnCNVs in genes or loci with previously established ASD significance (high-confidence group) to probands with no such genetic events (none group) at a 1:1 ratio. Matching procedures were performed separately for male and female participants, using ranges of 10 nonverbal IQ points and 8 months of age. These ranges were selected as the narrowest ranges within which probands from the matched autism control group could be found for all probands from the high-confidence group. Matching procedures were performed using a macro in SAS (30). Case-control differences were evaluated using a mixed model with a random effect of the case-control pair (to reflect the correlated nature of the matched data) and a fixed effect of group for continuous variables, or a conditional logistic regression for categorical variables. In both types of models, an interaction with nonverbal IQ was included to determine whether group differences were moderated by cognitive level. We present both uncorrected and false-discovery-rate-corrected (31) p values. False discovery rate was calculated separately for the case-control differences and the moderator analyses, both using the total number of comparisons. Analyses were conducted in SAS, version 9.3 (32).

Limitations of the SSC data set for these types of comparisons include its rigid exclusion criteria for problems that are known to be associated with pathogenic genetic abnormalities, including very low mental age and birth trauma (e.g., perinatal incidents, prematurity), exclusion of individuals who did not meet stringent ASD criteria on standardized diagnostic instruments, and the lack of contemporaneously sampled controls from different families without ASD. Thus, we note the possibility that the present findings may vary when the full range of intellectual disability and associated features within ASD is represented. That the phenotypic data collection was blinded to genetic status is a major advantage over other comparisons between “syndromic” and “idiopathic” ASD, in which clinicians’ ratings on standardized instruments or measures of diagnostic certainty may be subconsciously affected by biases about, for example, whether ASD in fragile X syndrome or tuberous sclerosis is the same as “idiopathic” ASD. Therefore, our finding of lower clinician-rated diagnostic certainty for children with genetic abnormalities is robust and cannot be explained by clinician bias.

Another caveat is that although the high-confidence and matched autism control groups were matched on age, children in this study spanned a wide age range (4 to 17 years). A challenge to genetic studies requiring large samples is that it is difficult to interpret within-sample comparisons of children spanning the full range of ages and developmental stages. On the one hand, results of this study suggest that ASD symptoms are less impairing in those who are diagnosed with ASD with de novo mutations in high-confidence genes or loci than in peers with equivalent cognitive skills; on the other hand, the pattern of significant differences in early motor milestones (related to lower IQ) may suggest differences in the developmental trajectories or patterns of emergence of ASD symptoms. Indeed, the fact that the high-confidence group was characterized by later onset of independent walking than the matched autism control group indicates a very early phenotypic difference. Delayed walking is more frequently observed in individuals with intellectual disability, compared with the general population and compared with individuals with ASD, which suggests that it may serve as a marker of propensity toward later cognitive impairment. Considering that the high-confidence and matched autism control groups were matched on current nonverbal cognitive functioning, presence of this early developmental difference provides further evidence for different developmental trajectories (33). Such questions underscore the need to obtain genetic data in prospective longitudinal studies.

A third limitation of the study was the small sample sizes of participants with de novo mutations in, or including, the same ASD-associated genes and loci and our subsequent combination of all of these participants into a single group. Although a number of group-level findings still emerged as significant, Figure 3 clearly illustrates the limitations of combining individuals of such diversity. It also exemplifies the variability of phenotypic expression even within a known abnormality, already observed in many studies of these specific genetic disorders (17). While it would be interesting to make observations about the most common dnLoF mutations and dnCNVs, which included four CNV duplications, one CNV deletion, and two mutations all previously associated with ASD, there are published “genetics first” cohorts for each of these (17, 18, 20, 37–39). These studies describe wide within-cohort variability in phenotypic expression, based on type of mutation or CNV characteristics, such as deletion versus duplication, size of the error, and the specific genes involved (2). An obvious next step is to continue efforts to collect sufficient numbers of cases of specific genetic abnormalities to allow comparisons both within and across disorders, although the feasibility of this approach is limited by the relative rarity of any specific mutation. However, as our understanding of the underlying molecular neurobiology improves, grouping patients with mutations expected to affect the same pathway(s), and therefore potentially leading to a similar phenotypic outcome, may provide traction in this regard (40). Relatedly, future studies may identify common variants or familially transmitted genetic abnormalities that contribute to these biologically relevant groupings.

In conclusion, these results highlight the critical need to consider ASD-related symptoms and behaviors in the context of overall developmental level. The differences between individuals with and without de novo mutations were only revealed when sex, IQ, and age were carefully controlled in the analyses. Proper steps must be taken to account for these factors in future studies in order to advance our understanding of the range of phenotypic profiles associated with genetic findings in ASD. Studies such as these need to be replicated and extended as additional genetic abnormalities are found to be associated with ASD with high confidence. Findings from these studies will elucidate actual genotype-phenotype differences within ASD, which can be used to more carefully phenotype specific animal models for treatment targeting and to inform clinical genetic risk assessment and prognosis.