Biomarkers for autism spectrum disorders

DNA Samples and Population Structure

The study included 426 ASD families. All of the index cases met Autism Diagnostic Interview-Revised (ADI-R) and Autism Diagnostic Observation Schedule (ADOS) criteria or on a clinical best estimate (Risi et al. J Am Acad Child Adolesc Psychiatry 2006; 45(9):1094-103). Thirty-two of these carried a cytogenetic chromosome rearrangement; 18 were detected by karyotyping 328 of 412 samples that originated from child diagnostic centres at the Hospital for Sick Children in Toronto and from St. John's, Newfoundland; 14 were already known to carry karyotypic anomalies (see Table 1 for information on these 32 patients). Affected and unaffected siblings were also assessed, and 56% (237/426) had one child (simplex) and 44% (189/426) had more than one child (multiplex) with ASD. Most cases were screened for fragile X mutations (75%) and if detected they were not included in the study. Most experiments were performed on blood genomic DNA (80%), otherwise the source was cell lines, e.g. lymphoblast cell lines. Population ancestry was estimated using STRUCTURE (Falush et al. Genetics 2003; 164(4):1567-87; Pritchard et al. Genetics 2000; 155(2):945-59).

Affymetrix GeneChip Human Mapping 500K Array Set

For each sample, approximately 500,000 SNPs were genotyped using the combined two-chip Nspl and Styl GeneChip® Human Mapping Commercial or Early Access Arrays (Affymetrix, Inc., Santa Clara, Calif.) according to the manufacturer's instructions and as described previously (Kennedy et al. 2003 Nat Biotechnol. 21:1233-7, the contents of which are incorporated herein by reference). Briefly, 250 ng of genomic DNA was digested with Nspl and Styl restriction enzyme (New England Biolabs, Boston, Mass.), ligated to an adaptor and amplified by PCR. The PCR products were then fragmented with DNaseI to a size range of 250 bp to 2,000 bp, labelled, and hybridized to the array. After hybridization, arrays were washed on the Affymetrix fluidics stations, stained, and scanned using the Gene Chip Scanner 3000 7G and Gene Chip Operating System. Data has been submitted to the Gene Expression Omnibus database (accession GSE9222). Karyotypes were generated using standard clinical diagnostic protocols.

Characterization of Copy Number Variation

Nspl and Styl array scans were analyzed for copy number variation using a combination of DNA Chip Analyzer (dChip) (Li and Wong 2001 Genome Biology 2: 0032.1-0032.11), Copy Number Analysis for GeneChip (CNAG) (Nannya 2005 Cancer Res. 65:6071-9) and Genotyping Microarray based CNV Analysis (GEMCA) (Komura 2006 Genome Res. 16:1575-84). Each of these references is incorporated herein by reference.

Analysis with dChip (www.dchip.org) was performed as previously described (Zhao et al 2005 Cancer Res. 65:5561-70) in batches of ˜100 probands. Briefly, array scans were normalized at the probe intensity level with an invariant set normalization method. After normalization, a signal value was calculated for each SNP using a model-based (PM/MM) method. In this approach, image artifacts were identified and eliminated by an outlier detection algorithm. For both sets of arrays, the resulting signal values were averaged across all samples for each SNP to obtain the mean signal of a diploid genome. From the raw copy numbers, the inferred copy number at each SNP was estimated using a Hidden Markov Model (HMM).

For analyses with CNAG version 2.0 (www.genome.umin.jp), the reference pool was set to include all samples and performed an automatic batch pair-wise analysis using sex-matched controls. Test samples were compared to all samples within the reference pool and matched based on signal intensity standard deviations. The scan intensities for each ‘test’ sample were compared to the average intensities of the reference samples (typically the average of 5-12 samples) and used to calculate raw copy number changes. Underlying copy number changes were then inferred using a Hidden Markov Model (HMM) built into CNAG.

GEMCA analysis was performed essentially as described (Komura et al. Genome Res 2006; 16(12):1575-84) with the exception that two designated DNA samples (NA10851 and NA15510) were used as references for pair-wise comparison to all proband experiments. These results were further filtered by only including those CNVs that were common to both pair-wise experiments.

CNVs were merged if they were detected in the same individual by more than one algorithm using the outside probe boundaries.

Controls and Autism Chromosome Rearrangement Database (ACRD)

Control samples consisted of (i) CNVs observed in 500 Europeans from the from the German PopGen project (Krawczak et al. Community Genet 2006; 9(1):55-61), and CNVs found in a cohort of 1000 Caucasian non-disease controls from the Ontario population (ref. 24). The ACRD that had 834 putative CNVs or breakpoints mapped to the genome was established. A CNV was considered ASD-specific if it was >10 kb, contained at least three probes and at least 20% of its total length was unique when compared to controls.

CNV Validation Experiments and Balance Rearrangement Breakpoint Mapping

PCR validation of CNV calls was performed using Quantitative Multiplex PCR of short fluorescent fragments (QMPSF) (Redon et al. Nature. 444:444-54) or SYBR-Green 1 based real-time quantitative PCR (qPCR) using controls at the ACCN1, CFTR or FOXP2 loci (PMID: 14552656). For both methods, primers were designed using the program PRIMER3 (http://frodo.wi.mit.edu/). Balanced rearrangements were mapped primarily using FISH (Nannya et al. Cancer Res 2005; 65(14):6071-9). The microdel program (Komura et al., ibid) was used to score CNV losses.

For QMPSF, short genomic sequences (140-220 bp) within putative CNVs were PCR amplified using dye-labelled primers corresponding to unique sequences. Each reaction also included co-amplified control amplicons corresponding to either ACCN1 or CFTR located at 17q11.2 and 7q31.2, respectively. Briefly, 40 ng of genomic DNA was amplified by PCR in a final volume of 25 μl using AmpliTaq® DNA polymerase (manufactured for Applied Biosystems by Roche Molecular Systems, Inc.) After an initial step of denaturation at 95° C. for 5 minutes conditions were as follows: 25 PCR cycles of 94° C. for 30 seconds, annealing at 60° C. for 45 seconds, and extension at 72° C. for 30 seconds. A final extension step at 72° C. for 15 minutes followed. QMPSF amplicons were separated on an ABI 3730xl DNA Analyzer (Applied Biosystems, Foster City, Calif.), and analyzed using ABI GeneMapper® software version 3.7 (Applied Biosystems). After adjustment of control amplicons to the same heights, the QMPSF pattern generated from test DNA was superimposed to that of the control DNA. For each putative CNV locus, the copy number ratio was determined by dividing the normalized peak height obtained from the test DNA by that of the control DNA. Peak ratios of >1.4 and <0.7 were indicative of copy number gains and losses, respectively. At least two independent QMPSF assays were required for CNV confirmation.

SYBR Green 1-based real-time qPCR amplification was performed using a Mx3005P quantitative PCR system (Stratagene, La Jolla, USA). Non-fluorescent primers were designed to amplify short genomic fragments (<140 bp) in putative CNV loci. Each assay also included amplification of a control amplicon corresponding to FOXP2 at 7q31.1 for comparison. After optimization of primer sets with control genomic DNA using ‘Brilliant® SYBR® Green QPCR Master Mix’ (Stratagene), test samples were assayed in 15 μl reaction mixtures in 96-well plates containing: 7.5 μl of reaction mix, 1.8 μl of primer, 6.0 ng of genomic DNA at 1.2 ng/μl, 0.225 μl of reference dye with 1:500 dilution, and 0.475 μl of water. PCR conditions consisted of 10 minutes of polymerase activation at 95° C., followed by 40 cycles of: 95° C. for 15 seconds and a single step at 60° C. for 1 minute for annealing and elongation. These steps were then followed by a final cycle of 95° C. for 1 minute, 55° C. for 30 seconds, and 95° C. for 30 seconds. Standard curve quantification was analyzed by MxPro-Mx3005P software (version 3.20 Build 340) to calculate copy number changes. Coefficient of variation (CV) was calculated on all sample Ct values to remove possible outlier when CV was greater than 1%. The average quantity of the putative CNV locus was divided by the average quantity of the control amplicon on FOXP2. Ratios of >1.4 and <0.7 were indicative of copy number gains and losses, respectively. Each putative CNV locus had at least two independent assays.

Results

Structural Variation Characteristics in ASD Cases

A total of 426 ASD index cases were tested for CNV content including 394 typical idiopathic cases and 32 others that were enrolled based on prior knowledge of having a cytogenetic abnormality. The Affymetrix 500 k SNP array was used because it provided the highest resolution screen available for both SNP genotype and CNV data. Using the SNPs, the ancestry of each sample was categorized (to guide selection of controls). Backgrounds of the samples were found to be: 90.3%, 4.5%, 4.5%, and 0.7%, European, European/mixed, Asian, or Yoruban, respectively.

To maximize CNV discovery, three calling algorithms were used as described above (see FIG. 1 ) and common results between them were merged to identify a ‘full’ dataset of 3389 independent CNVs (˜8 CNVs per genome, mean size 390 kb) (see Table 4 below). To minimize potential false positives, a second dataset was generated whereby a CNV needed to be detected by two or more algorithms and/or on both the NspI or StyI microarrays (Pinto et al. Hum Mol Genet 2007; 16 Spec No 2:R168-73).

This ‘stringent’ dataset contained 1312 CNVs (˜3 CNVs per genome, mean size 603 kb). Using q-PCR, 48% (12/26) and 96% (48/50) of random CNVs were validated in the full and stringent collections, respectively.

Five hundred European control samples were examined for their CNV content and similar numbers of CNVs (3695 in the full and 1558 in the stringent dataset) were found to those in the ASD cases (Table 4). This suggested germ-line chromosome instability was not a significant contributing mechanism. The ASD CNVs were then compared against the 500 European/Caucasian controls and the Database of Genomic Variants (a repository of structural variation in ‘non-disease’ populations) (Iafrate et al. Nat Genet 2004; 36(9):949-51) to establish autism-specific CNV datasets. The subsequent analysis then focused on the 276 CNVs in the stringent autism-specific category, which mapped across all 23 chromosomes ( FIG. 2 ), details of which are found in Table 3, below. Additional ASD-relevant CNV data is also found in the other categories in Table 5 (discussed below).

Wide-ranging prevalence frequencies of cytogenetically detectable chromosomal abnormalities in ASD, and the inability of microarray scans to find balanced abnormalities, prompted karyotyping to be performed. Karyotyping (and FISH) also provided the ability to characterize the chromosomal context (e.g. ring chromosomes) of some of the CNV regions, something not possible using microarrays alone. Therefore, 313 unbiased idiopathic cases where blood was available were examined and 5.8% (18/313) cases were found to have balanced (11) or unbalanced (7) karyotypes (all unbalanced karyotypic changes (7) were also found by microarray analysis and are included in the CNV statistics). The genomic characteristics of all CNVs are shown in the Autism Chromosome Rearrangement Database (see FIG. 3 ). In this study, CNV loss and gain will typically equate to a standard deletion or duplication. In some cases a duplication of only part of a gene could lead to its disruption (Table 5), and there are also positional effects on gene expression to consider.

De Novo, Overlapping/Recurrent, and Inherited Structural Variants

Structural variants found in ASD cases were initially prioritized to possibly be etiologic if they were not in controls and, (i) de novo in origin (25 cases) (see Table 5 below), (ii) overlapping (27 cases at 13 loci) in two or more unrelated samples (see Table 7 below), (iii) recurrent (same breakpoints) in two or more unrelated samples (four cases at two loci), (iv) or inherited the remainder). In a proof of principle analysis, CNVs were found at known ASD loci: NLGN4 and 22q, 15q, SHANK3 and NRXN1 in categories i, ii, iii, and iv, respectively. ASD structural variants found in controls (eg. NRX1) could also be involved.

By testing parental DNA and validating CNVs, a de novo mutation rate of 7.1% (4/56) and 2.0% (1/49) was observed in idiopathic simplex and multiplex families, respectively. There was parental information for 13 of 18 cases discovered to carry cytogenetic abnormalities and 7 (6 simplex, 1 multiplex) of these were de novo in origin. Since only 1/7 (from a simplex family) of these was balanced and directly interrupting a gene, it was estimated that this class of rearrangements had much less of a contribution than CNVs to the total rate of de novo and structural variation in the present cohort.

The collective data identified 25 de novo cases (Table 5) and in three, two or more events were identified. Notably, in family SK0152 ( FIG. 4 a ) there were four de novo events. In MM019 ( FIG. 4 b ) there were two de novo deletions, one leading to haplo-insufficiency of SHANK3.

The 13 loci where overlapping ASD-specific CNVs were found are likely indicative of ASD-susceptibility since they arise in two or more unrelated families. In six, gains and losses often encompassing entire genes were observed at the same locus (Table 6) suggesting general gene dysregulation to be involved.

Using q-PCR or by assessing SNP patterns, 196 inherited CNVs (90 maternal and 106 paternal) were confirmed. No sub-grouping of these demonstrated obvious parent-of-origin effects (the two chromosome 15q11-q13 duplications detected were both de novo in origin). A 160 kb deletion was detected in a male inherited from a carrier mother, leading to a null PTCHD1 in the proband and his dizygotic twin brother ( FIG. 4 c ). There were also instances where apparently balanced inherited translocations were accompanied by de novo deletions in the offspring (eg. DPYD) ( FIG. 4 d ).

Candidate ASD-Susceptibility Genes and Loci Identified

New ASD candidates identified were those with a structural change (either de novo or found in two or more unrelated ASD cases, or for the X chromosome an allele being transmitted maternally from an unaffected carrier) specific to that gene, including ANKRD11, DLGAP2, DPP6, DPP10, DPYD, PCDH9 and PTCHD1 (Tables 5 and 6). As previously noted, NLGN4, SHANK3 and NRXN1 were also identified. The PCDH9 and NRXN1 genes are also found as CNVs in controls in the DGV (Database of Genomic Variants).

Additional positional candidate genes identified were those found interrupted by balanced cytogenetic breakpoints including NEGR1, PIP5K1B, GABRG1, KLHL3, STK3, ST7, SATB2 (Table 1). Moreover, 77 CNVs in the stringent dataset overlapped with the Autism Chromosome Rearrangement Database providing a second line of evidence for involvement ( FIG. 2 ). For example, a 4.6 Mb de novo duplication at Xp11.23-11.22 was detected in a female SK0306-004 (Table 5) and a male in the database.

DPP6 and DPP10 emerge as being positional and functional candidates. DPP6 (˜1.5 Mb in size at 204.1) and DPP10 (˜1.3 Mb at 7q36.2) code for accessory trans-membrane dipeptidyl peptidase-like subunits that affect the expression and gating of Kv4.2 channels (KCND2). Kv4.2 channels function in regulation of neurotransmitter release and neuronal excitability in the glutamatergie synapse at the same sites where SHANK3 and the NLGN gene products are found. In addition, autism balanced breakpoints have been mapped near KCND2 at 7q31.

For DPP10 there are inherited CNV gains and losses (Table 5, FIG. 4 ). De novo and inherited CNVs were found at the multi-transcript DPP6 gene. A 66 kb de novo loss encompassing exons 2 and 3 is found in a male in family NA0002 ( FIG. 4 e ). In family SK0190, the male proband and an unaffected female sibling both carry a CNV gain inherited from an unaffected mother ( FIG. 4 f ) that encompassed the entire DPP6. A 270 kb gain was found in SK0115-003 that extends across the first exon (which may disrupt the functional gene) and SK0058-003 carries a maternally-inherited 16 kb intronic CNV gain (Table 1; FIG. 5 ).

Medical Genetics

Structural variants overlapping loci involved in medical genetic conditions including Waardenburg Type IIA (3p14.1), speech and language disorder (7q31), mental retardation (MR)(15q23-q24, 16p11.2) and velocardialfacial syndrome (VCFS) (22q13) were identified (Table 5), amongst others. Identification of the structural variant at these loci led to clinical re-assessment and either identification or refinement of the diagnosis, for additional syndromic features. Other instances (eg. SK0186-PTCHD1 deletion) ( FIG. 4 c ) prompted re-testing of the entire family and eventually a diagnosis of mild-ASD in a previously undiagnosed sibling. This family was then redesignated multiplex as opposed to simplex.

The identification of a de novo deletion (2.7 Mb) at 22q11.2 in two ASD brothers led to their re-examination and diagnosis for VCFS. The re-testing also further defined the siblings to be at opposite ends of the ASD spectrum ( FIG. 6 ). Larger duplications (4.3 Mb) of this same region in two other ASD families (SK0289 and SK0091) did not cause VCFS (Table 6); however, in SK0091 the variant was inherited from a normal father and not found in an affected male sibling.

A recurrent ˜500 kb duplication at 16p11.2 in two ASD families (SK0102 and NA0133) ( FIGS. 4 and 5 ) was also discovered. As with DPP6/DPP10 and 22q11.2, there were carriers of these structural variants without ASD. In a third family (MM0088), the proband has a larger 676 kb de novo deletion and it is only detected in one of two ASD siblings. ( FIG. 4 g ).

In sum, using the genome-wide scanning approach, numerous new putative-ASD loci (Tables 4 and 5, FIG. 2 ) were identified. Generally, ASD loci include (i) those that contain genes functioning in the PSD, (ii) and/or chromosomal regions previously shown to be involved in mental retardation, and (iii) involve dysregulation of gene expression.

CNVs that implicate ASD loci include the SHANK3, NLGN, and NRXN1-PSD genes and also identify novel loci at DPP6 and DPP10 (amongst others including PCDH9, RPS6KA2, RET from the full dataset) were identified.

Lastly, six unrelated ASD cases were identified (Table 6) that had either CNV gains or losses at the same locus which indicate that gene expression of genes in these regions are related to the development of speech and language and/or social communication in humans, as in SHANK3 and genes in the Williams-Beuren syndrome locus.

PTCHD1 as a Marker of ASD

As set out above, a genome scan with Affymetrix 500K SNP Arrays was used to identify a CNV deletion on chromosome Xp22.11 that spans exon 1 of the PTCHD1 gene. Exon 1 is shown bolded in FIG. 7 spanning nucleotide positions 1-359. The Cdna sequence of the PTCHD1 gene (NM_173495) as well as the amino acid sequence of the corresponding encoded protein is illustrated in FIG. 7 which illustrates a genomic size of: 59325, an exon/coding exon count of 3 encoding a protein of 783 amino acids.

The deletion was determined to be an ˜156 kb deletion on Xp22.11 on a male proband. The physical position of this CNV is chrX:22, 962,800-23, 119,000 (UCSC 2004 Assembly). The deletion is flanked by SNP probes rs7055928 and rs1918560 (at 22.956 and 23.133 Mb from the Xp terminus, respectively). The most proximal and distal SNPs (from the Affymetrix SNP microarrays) within the deleted region, as determined by the SNP microarray analysis, are rs7879064 (23.119 Mb) and rs4828958 (22.972 Mb). PCR amplicons from within the deleted region were used to confirm the deletion by Qpcr (PCR primers and locations are given below). This deletion spans the entire exon 1 of the PTCHD1 gene (NM_173495). Analysis of both Sty and Nsp chips data identified this event and was further validated using PCR and QPCR techniques. The following primers were used:

This CNV is autism specific as it was not present in the Database of Genomic Variants (DGV) and in other controls. Furthermore, the segregation of this deletion was characterized in family and it was identified that the deletion was transmitted from a heterozygous mother. A male sibling also had language deficits.

Mutation screening of PTCHD1 in N=400 autism patients was conducted in the usual manner. The following primers were used:

The mutation screening revealed an I173V mutation.

Identification of Additional Markers of ASD

By sequencing the entire coding region of PTCHD1 in 900 unrelated ASD cases, six missense mutations were identified in six unrelated ASD probands (Table 7, FIG. 8 ). For clinical details see Table 8.

All these mutations resulted in the substitution of highly conserved amino acids, and were inherited from unaffected carrier mothers. Based on in silico protein modeling, three mutations (L73F, I173V, V195I) are present in a predicted amino acid loop that sits outside of the cell membrane. This loop is posited to interact with the ligand, Hh. Another mutation, the 2-amino acid substitution ML336-337II was present within a predicted transmembrane domain. Finally, the E479G mutation was present within a predicted cytoplasmic amino acid loop. In five out of six families, these mutations segregated with the phenotype. Controls (439) were tested for the I173V and V195I mutations, 500 controls for ML336-337II, and 282 controls for L73F and E479G. None of these mutations were present in controls. Furthermore, the fact that these mutations were all maternally inherited to male probands, and were not observed in our control populations, indicates that the mutations are associated with ASD. In turn, it is reasonable to assume that these mutations contribute to the etiology of autism, and perhaps in-combination with other disease-related loci, give rise to the ASD phenotype.

Interestingly, in two of the ASD families reported in Tables 7/8 (Family-2 & Family-4), other ASD-related CNVs were identified. In family 2, in addition to I173V mutation, a de novo ˜1.0 Mb loss at 1p21.3 resulting in deletion of the entire DPYD gene (NM_000110.3) was identified. DPYD encodes a rate-limiting enzyme, dihydropyrimidine dehydrogenase (DPD), involved in pyrimidine metabolism. Complete DPD deficiency results in highly variable clinical outcomes, with convulsive disorders, motor retardation, and mental retardation being the most frequent manifestations. In Family-4, in addition to the V195I mutation, a 66 Kb de novo loss at 7q36.2 was identified resulting in deletion of DPP6 exon 3, and 33 amino acids towards the N-terminal end of the DPP6 protein. These cases evidence digenic involvement in ASD.

The ability of these PTCHD1-mutants to repress Gli2 expression was compared with wild type to determine if there was loss of function in the mutants. NIH10T1/2 fibroblasts were transfected with CMV-empty vector, a Gli-responsive promoter fused to the Luciferase gene (Gli2 pro), β-Gal (normalization) and PTCHD1 mutant expression plasmids. A mild loss of function of at least the E479G and ML336-711 mutants resulted in increased expression of Gli2 compared to wild type.