Biological Overlap of Attention-Deficit/Hyperactivity Disorder and Autism Spectrum Disorder: Evidence From Copy Number Variants

Objective Attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD) often co-occur and share genetic risks. The aim of this analysis was to determine more broadly whether ADHD and ASD share biological underpinnings. Method We compared copy number variant (CNV) data from 727 children with ADHD and 5,081 population controls to data from 996 individuals with ASD and an independent set of 1,287 controls. Using pathway analyses, we investigated whether CNVs observed in individuals with ADHD have an impact on genes in the same biological pathways as on those observed in individuals with ASD. Results The results suggest that the biological pathways affected by CNVs in ADHD overlap with those affected by CNVs in ASD more than would be expected by chance. Moreover, this was true even when specific CNV regions common to both disorders were excluded from the analysis. After correction for multiple testing, genes involved in 3 biological processes (nicotinic acetylcholine receptor signalling pathway, cell division, and response to drug) showed significant enrichment for case CNV hits in the combined ADHD and ASD sample. Conclusion The results of this study indicate the presence of significant overlap of shared biological processes disrupted by large rare CNVs in children with these 2 neurodevelopmental conditions.

A ttention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD) show strong comorbidity at the level of both symptoms and disorder. 1,2 Although ADHD and ASD are distinctive in terms of core diagnostic symptoms, both have onset in early childhood, present more commonly in males, and are associated with similar cognitive, developmental, and neurological problems. 3 Twin studies have consistently shown that shared inherited factors explain a large proportion of the comorbidity between ADHD and ASD, as well as comorbidity with other neurodevelopmental problems. 4,5 Although shared common risk variants for ADHD and ASD have not been identified thus far, 6,7 this could reflect the relatively small sample sizes in the genome-wide association studies (GWAS) of each of these disorders. However, recent studies have suggested that rare (<1% frequency) chromosomal deletions and duplications, known as copy number variants (CNVs), occurring in children with ADHD show significant overlap with those already implicated in ASD. [8][9][10] It is not yet known whether ADHD and ASD also more broadly share biological underpinnings.
In this study, we set out to investigate whether large rare CNVs found in individuals with each of these clinical phenotypes index disruption of shared biological pathways in the disorders. The first aim was to determine whether biological Supplemental material cited in this article is available online.
JOURNAL OF THE AMERICAN ACADEMY OF CHILD & ADOLESCENT PSYCHIATRY VOLUME 53 NUMBER 7 JULY 2014 www.jaacap.org pathways disrupted by CNVs in individuals with ADHD, as compared with ethnically matched controls, showed statistically significant enrichment for CNV hits in participants with ASD, as compared to a separate set of controls. The second aim was to meta-analyze the ADHD and ASD samples to increase the power of detecting specific shared biological pathways disrupted in individuals with these 2 neurodevelopmental conditions.

METHOD Participants With ADHD
The sample consisted of 799 young persons of white ethnicity from Cardiff, Wales (n ¼ 559), St. Andrews, Scotland (n ¼ 44), and Dublin, Ireland (n ¼ 196). All children were recruited from community clinics and had a diagnosis of DSM-IV/DSM-III-R ADHD or International Statistical Classification of Diseases and Related Health Problems-Tenth Revision (ICD-10) hyperkinetic disorder. Exclusion criteria were intellectual disability (ID; IQ <70), major medical or neurological conditions, ASD, psychosis, and bipolar disorder. Approval was obtained from North West England, Wales, National Health Service Tayside, and Eastern Regional Health Authority research ethics committees. Written informed consent was obtained from parents, and assent/consent was gained from the young persons.

Clinical Measures
ADHD and other psychiatric diagnoses were assessed by trained psychologists using the Child and Adolescent Psychiatric Assessment (CAPA) parent version, 11 a semi-structured interview. Confirmation of pervasiveness of symptoms in school was obtained using the Child Attention-Deficit Hyperactivity Disorder Teacher Telephone Interview (CHATTI) 12 or the Conner's Teacher Questionnaire. 13 The Wechsler Intelligence Scale for Children-III/IV was used to assess IQ. 14,15 The age range was 4 through 18 years, with a mean age of 10 years 3 months (SD ¼ 3 years). The sample was 87.4% male.

Genome-wide Data: Individuals With ADHD and Controls
DNA for all participants with ADHD was extracted from saliva or peripheral blood samples, as described previously. 16 Control genetic data were obtained from the Wellcome Trust Case-Control Consortium-Phase 2 (WTCCC2). 17 Quality control (QC) procedures and CNV detection protocols were identical to those described previously. 16 Analysis was based on single nucleotide polymorphisms (SNPs) that were present on genotyping chips in both participants with ADHD and controls. After QC, genome-wide data for 502,702 SNPs from 727 participants with ADHD and 5,081 controls were used for analysis. Analyses of CNVs were limited to those that were large (>500 kb) and rare (<1% frequency in the combined group of participants with ADHD and controls) because they have better concordance across different genotyping platforms, are determined with greater accuracy, and are more robustly associated with neurodevelopmental disorders. 8 There were 78 large, rare CNVs within the control sample (as previously published 16 ) and 85 from participants with ADHD. Parental genotype data were not available for most of the sample.

CNV Data: Individuals With ASD
CNVs for participants with ASD and independent controls were obtained from the publicly available supplementary data of a study comparing CNVs in 996 individuals of white ethnicity with ASD to 1,287 matched controls. 18 In this dataset, ASD diagnosis was confirmed using the Autism Diagnostic Interview-Revised (ADI-R) and Autism Diagnostic Observation Schedule (ADOS). 19,20 Control samples were obtained from the Study on Addiction: Genetics and Environment (SAGE) and from HapMap CEPH Utah (HapMap CEU). 18,21 CNVs were selected if present at <1% frequency in the total sample and having length >30kb, giving a set of 5,478 CNVs, as described previously. 18 This CNV set contains 215 CNVs >500 kb in controls and 133 in participants with ASD. Of these 133 CNVs in participants with ASD, 13 were de novo (i.e., confirmed not to be transmitted from either parent) and 120 were confirmed to be inherited.

Method for Testing Pathway Enrichment
Pathways. The following 5 sets of pathways were used in the enrichment analyses (the same as used previously 16 ): Gene Ontology (GO), 22  For reasons of power, analyses were restricted to pathways containing between 3 and 1,500 genes (16,569 in total). Furthermore, pathways required at least 10 hits in the total sample to be counted (10,240 in total). This was to reduce the chance of small pathways showing apparent enrichment based on a small number of CNV hits.
Testing Pathway Enrichment of Case CNV Hits. Each CNV was assigned a binary variable ("participant" or "control") according to whether it came from a participant with ADHD or ASD or from a control. A CNV was considered to "hit" a gene if any part of the CNV lay between the start and end points of the longest transcript of the gene (as defined by the National Center for Biotechnology Information [NCBI]). Both CNV and gene positions use build 36.3.
The following logistic regression model was fitted to the sample of CNVs for each pathway separately: participants with ADHD or ASD/control w CNV length þ number of genes hit outside pathway þ hit gene(s) in pathway (yes/no), and the deviance was compared to the following: participants with ADHD or ASD/control w CNV length þ number of genes hit outside pathway to give a (1-sided) test of enrichment of case CNV hits on genes in the pathway.
This method is similar to that proposed by others 27 and has been applied to de novo schizophrenia CNVs. 28 CNV length is fitted in the model because long CNVs are more likely to hit any set of genes than small ones, and CNV length may differ systematically between participants with ADHD or ASD and controls. The number of genes hit outside pathway is fitted to allow for case CNVs influencing disease status by hitting genes other than those in the pathway being tested. A binary variable (yes/no) is used to indicate whether a CNV hits gene(s) in a pathway rather than the number of genes in the pathway hit by the CNV, to allow for some pathways having several genes that are physically close together (and thus likely to be hit by the same CNV).
The same analysis approach was also used to obtain tests of gene-specific enrichment for participant CNV hits, by defining pathways containing single genes.
Primary Analyses. To test whether the pathways with nominally significant enrichment in the ADHD sample showed greater than expected enrichment for case CNV hits in the ASD sample, enrichment analyses were run in the ASD sample restricted to the pathways enriched at various levels (p < .05, p < .01, p < .001) in the ADHD sample. The number of enriched pathways in the ASD sample at the same significance level as that used to select pathways from the ADHD sample was compared to that obtained when the participant/control labels were randomly permuted in the ASD sample. This procedure was repeated 1,000 times, and the p value for the number of pathways significantly enriched in both samples was estimated as the proportion of replicates, where the number of significantly enriched pathways was at least as great as that observed in the actual data. This analysis allows for overlap between pathways in terms of their gene membership. Analyses were carried out using all participants with ASD CNVs (versus control CNVs), and also for de novo and inherited CNVs (versus all control CNVs), separately.
A combined pathway enrichment analysis was performed by meta-analyzing the ADHD and ASD case CNVs compared to their corresponding control CNVs. Enrichment p values for both genes and pathways were obtained by adding a 2-level factor coding for sample (ADHD/ASD) to the regression models (to allow for possible differences in CNV calling and/or ethnic differences between samples) and applying these to the combined ADHD and ASD data.
Secondary Analyses. To assess the extent to which the observed overlap in enriched pathways is driven by known loci for autism susceptibility, the enrichment analysis was repeated on the complete pathway set, omitting genes in 14 8 Moreover, it is possible that any observed overlap in pathways enriched in both samples is due to the physical overlap between case CNVs from the 2 samples, rather than shared biology per se (although larger than expected physical overlap may, of course, be due to shared biology). We tested whether ASD participants' CNVs are more likely than their control peers' CNVs to overlap participants with ADHD participants' CNVs, and vice versa, by fitting the following logistic regression model to the CNVs from each sample: Overlap (y/n) w participants with ADHD or ASD/control þ CNV length. A CNV was defined as overlapping if any part of that CNV overlapped any case CNV from the other sample. A significant positive regression coefficient for the case/control term is taken as evidence that case CNVs in 1 sample are more likely than their corresponding control CNVs to overlap case CNVs in the other sample, allowing for CNV size (large CNVs being more likely to overlap than small ones). This analysis was initially carried out genome-wide, using all CNVs, and then repeated omitting CNVs in the autism regions listed above to see if these regions accounted for any significant overlap.
The pathway enrichment analyses were repeated on the complete pathway set after removing from both sets of cases any CNV occurring in a participant with ADHD that overlapped a CNV in the ASD sample (leaving 102 ASD CNVs and 55 ADHD CNVs). A significant excess of pathways enriched in both samples would provide evidence of shared biology even among CNVs that do not hit the same genes.
To determine whether any observed overlaps in significant pathways were driven primarily by deletions or duplications, the pathway enrichment analyses were repeated on the complete pathway set, using deletions or duplications alone. Table 1 shows the number of pathways achieving differing significance levels (p < .05, p < .01, p < .001) in the ASD sample that were also significant at the same significance level in the ADHD sample. It can be seen that a significant overlap in enriched pathways for ASD and ADHD is observed for both ASD de novo and inherited CNVs, with the most significant overlap being observed in the analysis of all ASD CNVs together.

Enriched Pathways and Genes in the Combined ADHD and ASD Dataset
Of the 100 pathways that were significantly (p < .05) enriched for case CNV hits in both the ADHD and ASD samples, the 20 pathways that were most significantly enriched in the combined ADHD and ASD (all CNV) samples are shown in Table 2, together with the number of gene hits by case CNVs, and the genes that are individually significantly (nominal p < .05) enriched for case CNV hits in the combined ADHD and ASD sample. Note that 3 of these pathways (nicotinic acetylcholine receptor signaling, cell division, and response to drug) have enrichment p values in the combined ADHD and ASD data that are significant even after Bonferroni correction for 10,240 pathways tested (p < 4.88 Â 10 À6 ). This correction is conservative, as the tested pathways are not independent because of shared genes. The genes hit by case CNVs in these 3 pathways are listed in Table S1 (available online).

Removal of CNVs in Known ASD Regions and
Overlapping Case CNVs Secondary analyses assessed whether these observed results were driven by specific loci previously implicated in ASD and shown to overlap with CNVs in a subsample of the current ADHD sample, 8 or by overlapping CNV regions in the 2 groups, including CNVs other than those falling within these known loci. Analyses were repeated omitting these regions. Results are shown in Table 3.
For the analysis omitting the specific regions, it can be seen that the overlap in pathways is still significant, although, not surprisingly, the level of significance is reduced, particularly for analyses based on the ASD de novo CNVs. For the analysis omitting all overlapping case CNVs, results show a modestly significant overlap of enriched pathways in the ASD de novo CNVs but not in the inherited CNVs or the total CNV set.
Pathway enrichment p values in the absence of CNVs overlapping known ASD loci or overlapping a case CNV from the other disorder are shown in Table S2 (available online) for all 100 pathways significantly (p < .05) enriched in both ADHD and ASD when all CNVs were analyzed (primary analysis). Removing the known ASD regions makes little difference to the pathway enrichment, whereas removing case CNVs that physically overlap with case CNVs from the other disorder generally reduces enrichment significance considerably. Thus, most of the overlap in enriched pathways can be attributed to case CNVs in the disorders hitting the same loci (and  BIOLOGICAL OVERLAP OF ADHD AND ASD thus genes) but not necessarily in regions previously implicated in ASD.

Analysis of Deletions and Duplications Separately
Of the 85 case CNVs in the ADHD sample, 21 were deletions and 64 were duplications. Among the 78 control CNVs in the ADHD dataset, 13 were deletions and 65 were duplications. Of the 133 case CNVs in the ASD dataset, 34 were deletions and 99 were duplications. Among the control CNVs in the ASD dataset, 65 were deletions and 150 were duplications. The numbers of pathways significantly enriched in both ADHD and ASD are provided in Table 4. Analyzing duplications and deletions separately reduces both the number of pathways significantly enriched in both ADHD and ASD and the significance of any excess. There was no evidence that either deletions or duplications separately account for the observed pathway overlap between ADHD and ASD. Pathway-specific enrichment p values for the 100 pathways significantly enriched (p < .05) in both ADHD and ASD when deletions and duplications are analyzed separately are shown in Table S3 (available online).

List of CNVs Used in the Analyses
A complete list of the case CNVs >500 kb (ADHD and ASD) used in the analyses is given in Table S4 (available online).

DISCUSSION
The results show that the biological pathways enriched (p < .05) for CNVs in the ADHD sample (relative to controls) as a group show more enrichment for CNVs in the ASD sample (relative to an independent set of controls) than expected by chance. A similar enrichment was observed (results not shown) when the analyses were performed in the reverse direction (i.e., the pathways enriched in the ASD sample tested in the ADHD sample). This finding indicates the presence of common biological liability for ADHD and ASD. Significant overlap was observed for both de novo and inherited CNVs in the ASD sample, although these results are not independent because the same control CNVs were used. Given that an earlier study using part of the current ADHD sample found enrichment in ADHD CNV loci that had been previously implicated in ASD, 8 those loci were omitted from the current analysis to obtain independent replication of the earlier finding that ASD CNV loci were also found in children with ADHD. The continued significant overlap of CNVs and pathways suggests that other CNV loci are also contributing to this effect, although, given the small number of loci in this additional analysis, the significance of overlap with ADHD and de novo ASD CNVs is reduced. Moreover, omitting all ADHD case CNVs overlapping at all with ASD case CNVs and vice versa from the analyses also shows more generally that shared biological pathways are implicated in ADHD and ASD above and beyond overlap of specific CNV regions. Given that performing this strict analysis with no physically overlapping CNVs substantially reduces the pool of CNVs in the analysis, it is remarkable that there is still demonstrable overlap in biological pathways tapped into by CNVs from participants with ADHD and ASD. To highlight which specific pathways contain enrichment evidence in both ADHD and ASD, the 2 samples were combined in a joint analysis. Three pathways showed significant enrichment after correction for multiple testing ("nicotinic acetylcholine receptor signalling pathway," "cell division," and "response to drug"). Owing to the definition of pathway categories, many of the analyzed pathways overlap with one another, including pathways embedded in one another. The 3 significant pathways contain different significant genes ( Table 2 and Table S1; the latter is available online) despite each pathway containing at least 1 significant gene from the same region on chromosome 16 (MYH11, NDE1, ABCC1, ABCC6), and a significant (or nearly significant) gene from the same region on chromosome 17 (MYO19, AATF, ACACA). It is unclear which gene(s) in these regions is responsible for the CNV enrichment. It should be noted that even when these regions are removed from the analysis, all 3 pathways still show significant enrichment (nicotinic acetylcholine receptor signaling pathway: p ¼ 1.17 Â 10 À3 , cell division: p ¼ 6.06 Â 10 À3 , response to drug: It is interesting to note that the neurobiology encompassed by these pathways enriched in both ADHD and ASD has been implicated in a previous pathway analysis of the current ADHD sample, which explored the overlap of common (SNPs) and rare (CNVs) variants. 16 The most significantly enriched pathway in the combined ADHD and ASD samples (Table 2) is the PANTHER pathway "nicotinic acetylcholine receptor signalling," which is also significantly enriched in both ADHD and ASD separately. This pathway contains 2 genes, MYH11 and CHRNA7, of potential interest. The gene CHRNA7 encodes the alpha 7 nicotinic acetylcholine receptor, which has a role in calcium signaling in the brain. This gene was previously shown to be enriched for both CNV hits and GWAS signal in the current ADHD sample. 16 It has been found also to have duplications spanning it in a genome-wide analysis of CNVs, a finding that was replicated in independent ADHD samples. 10 CHRNA7 is located at the chromosomal locus of 15q13.3, and deletions at this locus have been found to be associated with different neurodevelopmental abnormalities and neuropsychiatric disorders. 29 There is evidence from a small case series that deletions and duplications at this locus could also be associated with features of ASD. 30 MYH11 is at the chromosomal locus of 16p13.1, a region that has previously been shown to be enriched for CNVs in a subsample of participants with ADHD, at a genome-wide level relative to controls. 8 This region has also been implicated in autism, 31 schizophrenia, 32 and intellectual disability/multiple congenital anomalies. 33 It should be noted that MYH11 is involved in numerous other enriched pathways, suggesting that it may influence ADHD and ASD susceptibility through multiple biological processes.
Interestingly, the current analysis has not implicated the types of biological pathways previously reported in pathway analyses in ASD samples that are related to synaptic and neuronal plasticity and those involved in neurotransmission or synapse formation and maintenance. 34,35 This is likely because such pathways are not significantly enriched for case CNV hits in the ADHD sample.
Given the very high level of comorbidity and symptom correlation between ADHD and ASD, 1 it is arguable as to whether it is clinically meaningful to attempt to distinguish "pure" ASD or ADHD cases. Although children with a clinician's diagnosis of ASD were excluded from the ADHD sample, previous clinical analyses have reported subthreshold ASD traits in this sample. 36 Moreover, clinical data on levels of ADHD symptoms were not available for the ASD cases for this analysis. 18 Thus, it is possible that the overlap in biological pathways detected in this study may be reflecting the presence of subthreshold ADHD and ASD traits in the samples. However, given the strong relationship of the 2 conditions, the value of attempting to control for subthreshold traits is unclear. Also, despite overlaps, ADHD and ASD are clinically distinctive. Furthermore, disruption of synaptic function by rare CNVs may play a more important role in some neurodevelopmental processes, such as those involved in ASD and ID, but not all. Thus far, synaptic functions have not been implicated by pathway analysis of SNP data in ADHD. 16 Although the participants with ADHD and the WTCCC2 controls were genotyped on different chips, CNV calling used only the SNPs common to both genotyping chips, with the same QC procedures used to filter the SNPs. Thus, the CNV calls are comparable between participants with ADHD or ASD and controls, particularly for large (>500-kb) CNVs that are called with high accuracy and reliability. The participants with ASD and controls were genotyped on the same chips, using the same QC protocols, so again the CNV calls are directly comparable between participants with ASD and controls. The participants with ADHD and WTCCC controls are of similar ethnic backgrounds (UK individuals of white ethnicity), as are the participants with ASD and controls (US white individuals). Note that, in the combined analysis, differences in ethnicity and CNV calling between the ADHD and ASD samples are controlled by including "sample" as a covariate. Thus, significant pathway enrichments are unlikely to have arisen because of population stratification or differences in CNV calling.
One limitation of this study is that the biological pathway categories were defined based on the GO, KEGG, PANTHER, MGI, and MSigDB databases, which depend on the accuracy of the annotations in these databases. As the contents of these databases come to reflect the growing knowledge of these processes, pathway analyses such as those described in this article will be better able to implicate specific biological processes in disease etiology. Furthermore, future studies need to assess the functional effects of genes implicated in specific overlapping pathways, to determine the exact nature and extent of the shared underlying biology of ADHD and ASD.
It is of note that the overlap in pathways was detected for de novo as well as inherited CNVs. Both inherited and de novo CNVs are enriched in participants with ASD relative to controls, with a higher frequency of de novo CNVs in females than in males and in sporadic ASD cases (i.e., families with a single affected child) than in "multiplex" families (i.e., with more than 1 affected child). [37][38][39][40] De novo CNVs have also been reported in individuals with ADHD, although at a lower rate than that reported in ASD and schizophrenia. 9 Unfortunately, because of the unavailability of complete parental genetic data, it is not known whether the CNVs in the ADHD sample were inherited or de novo.
Although the majority of the CNVs in this sample (both in participants with ADHD or ASD and in controls) were duplications, the strength of the signal was reduced by analyzing deletions and duplications separately. This suggests that both types of CNVs likely contribute to disrupting biological processes underlying ADHD and ASD.
There is evidence that large, rare CNVs are more likely to occur in children with ADHD who have comorbid ID (IQ <70). 8 Similarly, there is a somewhat greater rate of CNVs in children with ASD with ID relative to those without ID. 41 Although children with ID were excluded from the ADHD analyses, the ASD sample did not make such exclusions. However, this would serve to reduce biological overlap, and it means that the results cannot be explained by comorbid ID.
In conclusion, this study provides evidence that ADHD and ASD show significant overlap of shared biological processes being disrupted by CNVs. This finding gives preliminary evidence of the mechanisms that may underpin observed phenotypic overlap 1 and shared heritability. 4 The findings would benefit from replication and further investigation using larger collaborative samples and future updated versions of pathway annotation databases to determine the specific biological pathways that are being affected by these rare variants. These findings further strengthen the conceptual grouping of ADHD and ASD as related neurodevelopmental disorders. Further research in this area has the potential to shed light on heterogeneity of ADHD and ASD clinical phenotypes and the subtyping of child neurodevelopmental disorders. &