A total of 24 GABA receptors, 19 GABA-associated coding genes, and 19 additional autism candidates were targeted for aCGH assessment (Table ). After exclusion of outlier samples, a total of 168 case and 149 control individuals were examined by CGH arrays in the initial discovery set. In all, 6,143 CNV events calls were identified across our tested population, representing 1,223 potentially unique variations. The overall number of CNV events did not exhibit significant differences between ASD affected and unaffected individuals, although both the average and total kilobases impacted did show statistical significance (perm P
< 0.001; 10,000 permutations; one-sided), consistent with previous reports indicating pathogenic CNVs tend to be larger in size and over-represented in affected individuals (for example [35
]). The full list of CNV calls from the Nexus Rank Segmentation algorithm is provided in Additional file 4
: Table S4. The logistic regression analysis of case and control allele frequencies were used, in conjunction with additional biological annotation criteria, to prioritize a subset of regions for confirmation via a second molecular platform, TaqMan copy number quantitative PCR assays (Applied Biosystems). In addition to the ASD candidate loci prioritized for follow-up, five control genomic regions, representing known polymorphic locations with minor allele frequencies >10%, were also targeted with TaqMan assays to test the sensitivity of our laboratory technique. The 29 targeted regions and five control loci targeted for follow-up are detailed in Additional file 3
: Table S3, along with confirmation results. A diagram of the work flow for discovery and validation is provided in Additional file 5
: Figure S1. The sample set used for molecular confirmation and additional population screening was comprised of the individuals in which the CNVs were initially observed (for confirmation), along with an additional 170 Caucasian cases and 170 unaffected ethnically-matched controls that were not present in the original discovery set (Table ). A subset of independently confirmed loci that remained interesting was run on a larger set of 755 case and 1,809 unaffected family members. GEE analysis results for these data, accounting for relationships between individuals, are provided in Table .
Targeted CNV call events were confirmed with quantitative PCR at 13 of 29 loci, 45%. This suggests a minimum false-positive rate (at the locus level) of 55% (addressed below).
Medical records were examined for a subset of individuals harboring rare CNV events at loci of interest (ABAT, NRXN1, JAKMIP1, GABBR2, NLGN2, OXTR). We examined salient phenotypic and medical history characteristics for these individuals to look for commonalities among carriers of CNVs at particular loci. Characteristics examined included developmental abnormalities in speech and motor development, learning disabilities, and evidence of dysmorphologies. Additional co-morbidities (for example irritable bowel syndrome, asthma) and neuropsychiatric features were also noted. No obvious correspondence between aberrant CNV state and specific phenotypic traits was observed. A summary of available phenotypic and medical characteristics for the individuals examined is provided in Additional file 6
: Table S5.
In addition to these rare variants, common structural polymorphisms were observed and validated at multiple loci, including NRXN3, CNTNAP2, GABBR2, SLC38A3, and NLGN2. Of these more frequent events, only the GABBR2 locus approached nominal significance (GEE; P = 0.029; OR = 1.16; 95% CI: 0.7-1.93) in our larger validation set, but the result was not significant at α = 0.025 after correction for multiple testing (Table ).
The role of structural variation, particularly rare and de novo CNVs, in autism has been established in a number of previous studies. Here we examined a set of autism candidate loci using high-resolution aCGH analysis. Our principal objective was to detect structural variants at these loci that may have been potentially missed by other screening platforms, such as SNP-based arrays. Among the limitations impacting our study was the presence of a high false-positive rate (in excess of 50%) in our initial aCGH discovery set. The most likely explanation for this elevated false-positive rate is that the majority of genes in our study, including both GABA and non-GABA-related targets (for example MBD1-MBD4), belong to families of homologous genes which can lead to aberrant probe hybridization. As the majority of loci in our study belonged to homology groups, we could not formally test for this effect. In addition, the modest overall sample numbers used in this study limited our ability to accurately estimate population allele frequencies for rarer events. Even with these complicating factors, several promising events were detected and independently confirmed by quantitative PCR. Rare structural variations were detected at ABAT, NLGN4-Y, NRXN1, and JAKMIP1 (Table ). Most notable among these results, statistically significant enrichment of deletions (GEE; P = 0.0215; OR = 2.75; 95% CI: 1.161-6.5; α = 0.025) in affected vs. unaffected individuals was observed at NRXN1 within our larger family-based validation set with after correction for multiple testing (Table ).
We observed and confirmed an approximate 4 kb deletion in a female autistic indvidual within the JAKMIP1 gene that removes two exons. JAKMIP1, a member of a family of microtubule interacting proteins, is known to interact with GABBR1 protein [36
] and knockdown of its expression via siRNA results in increased GABBR2 expression, suggesting a possible role for JAKMIP1 in GABBR2 regulation. Misregulation of JAKMIP1 has also been observed in lymphoblastoid cell lines from autistic individuals [37
]. We note, however, that no significant enrichment for the deletion was observed in our larger validation set. More detailed characterization of these deletions in the affected and unaffected individual is warranted before further inference can be made, as the precise boundaries of the deletion events may be different in case and control individuals.
A large (approximately 70 kb) deletion encompassing two exons of the NRXN1 genes was detected in an autistic male (Figure ). NRXN1 is a neuronal adhesion molecule that has been implicated in a number of neurological disorders, including autism (for example [38
]. Suggestive SNP association results at the NRXN1 locus were reported in a large collaborative study, but genome-wide significance was not reached [7
]. Significant enrichment was previously reported for rare structural events [42
]. The removal of this exonic region within an autistic male in our dataset, along with the detection of significant enrichment of NRXN1 deletions in cases vs
. unaffected individuals in this study, provides further evidence for NRXN1 as an ASD risk locus. Furthermore, the location of our CNV, along with the placement of previously reported NRXN1 events (Figure ), suggest that disruptions of the alpha form of NRXN1. The beta form of NRXN1, which is transcribed from a secondary downstream promoter, is less frequently observed to be disrupted by CNV in ASD cases. Combined with the aforementioned genetic association results, our data suggests that both rare and common variants may be contributing to ASD risk at this genomic location.
Figure 1 A deletion of approximately 70 kb encompassing two exons (3,4) from the NRXN1 gene (alpha-1-isoform). Previously reported NRXN1 CNV events in ASD cases are shown mapped below [38-40,42-49]. A subset of NRXN1 structural variants reported in individuals (more ...)
A deletion of the entire Neuroligin 4Y gene within an autistic male was also observed and confirmed. In the larger dataset, the deletion was observed in the original individual's unaffected father as well as within an unaffected male sibling. Two additional unaffected carriers (unrelated) were also observed in the larger dataset. Interestingly, however, we observed two duplication events within the larger validation set, both of which were present in unrelated ASD individuals. Neuroligin 4Y, which resides on the non-recombining portion of the Y chromosome in modern humans was, in the evolutionary past, allelic with Neuroligin 4X (NLGN4X
) gene. Several studies have previously implicated NLGN4X in autism (for example [50
]). Nucleotide mutations in NLGN4Y have been reported in an autistic male and his father, who possessed learning disabilities [53
]. While there is currently no data regarding the functional status and possible role of NLGN4Y, EST evidence from the NCBI database suggests that NLGN4Y retains a molecularly active promoter and is expressed in the brain. A recent report demonstrated a potential role for some pseudogenes in regulating the expression of their parent or source genes by providing decoy targets for miRNAs [54
]. It is possible that, through this or some other mechanism, NLGN4Y is involved in regulating the expression of the NLGN4X locus in males. Our finding of copy number aberrations of this gene in autistic individuals suggests that the locus warrants further investigation.
We observed a deletion at the ABAT gene locus within multiple ASD individuals, one in our discovery dataset and a second case ASD case in the validation set. Further examination of the locus within the larger, family based set revealed 7 of 755 affected individuals carrying the deletion and five (of 1,809) unaffected individuals (unrelated to the seven affected individuals). Although we observed a trend of enrichment in affected vs. unaffected individuals in our dataset, statistical significance was not met at α = 0.05. The observed trend, however, suggests that ABAT locus, and the GABA system as a whole, warrant further scrutiny for their potential role in ASD risk. ABAT is involved in the catabolism of GABA, has previously been implicated in autism by genetic association, although replication of the association was not observed in the small validation dataset (n
= 91 autism trios) [55
]. In the one deleted individual from our dataset for which CGH array data was available at the locus, the ABAT deletion removes exon 1 of the gene (Additional file 7
: Figure S2). We note that the nearby A2BP1 and GRIN2A loci, which have also been implicated in autism [55
], were not included in our deleted region (Additional file 8
: Figure S3).
We also observed a previously reported deletion in the OXTR receptor in a single male ASD individual [56
]. A smaller duplication region in OXTR was also detected in a separate ASD case.
In addition to the rare copy number events described above, we also observed and confirmed common copy number polymorphisms at SLC38A3, GABBR2, NRXN3, SHANK3, and NLGN2 (Table ). Among these, only duplication events in GABBR2 achieved nominal statistical significance at α = 0.05, but this result failed to hold up after multiple correction for multiple testing. Furthermore, the relatively high level of background noise of assays targeting this locus makes us approach the GABBR2 finding with caution. Additional examination of the locus through alternative platforms (for example long-read high throughput sequence technologies) will be necessary before further conclusions can be drawn.