Using a next-generation (massively parallel Illumina-based) strategy, we sequenced promoter, 3' untranslated, intronic and coding regions of 10 Wnt pathway genes (WNT1
) in genomic DNA obtained from a first set of 198 Caucasian ASD patients and 240 healthy Caucasian controls, arranged in pools of 38–40 samples each. The number of reads was homogeneously distributed over the different amplicons and pools (and therefore between ASD and controls), reads mapped 88% of the targeted regions and coverage averaged 865-fold/individual. To validate the next-generation sequencing strategy employed, we compared results obtained using this strategy to results obtained using Sanger sequencing of the exons in a control locus (DACT1
) in a subset of samples (40 ASD). Every DACT1
=8), including several singleton variants (n
=3), found by Sanger sequencing in this set was also detected by the next-generation sequencing strategy, and the allele frequencies of all SNPs were identical using each methodology (Pearson's correlation test; r
<0.0001; Supplementary Table S1
In the 10 genes under study, we identified 652 SNPs overall in this set of 438 combined ASD and control samples. We restricted most of our attention to nonsynonymous sequence variants found only in our own sample set (not referenced in dbSNP, 1000 Genomes, the Exome Variant Server (EVS) or the Exome Chip Design Consortium). To determine whether any of the putative SNPs found by this method were artifacts produced by the next-generation sequencing technology, we confirmed each of them by individual genotyping using either IPlex Sequenom or Sanger sequencing. Using this approach we confirmed 12 novel singleton missense variants distributed across 8 of the 10 loci included in our study (Supplementarty Table S2
). Moreover, genotyping in a second set of 91 ASD and 144 control samples failed to identify any additional occurrences of each of these rare deleterious missense SNPs. Although each of these SNPs were found only once among the probands in our samples and had not been reported in any public database, genotyping of parents demonstrated that none of them had occurred de novo
; all were transmitted from either the mother or father. Moreover, sequencing of family trios showed that each was also present in an unaffected sibling. This verifies that each of these variants is not an idiosyncratic artifact arising during lymphoblastoid cell line derivation, and suggests that they are extremely rare alleles (single-nucleotide variants) in the Caucasian population represented by our samples.
Although we found a greater number of extremely rare missense variants in our ASD sample compared with our control sample (4% in ASD vs 1.7% in controls; ), this difference in the ‘rare missense variant burden' did not reach statistical significance (Fisher's exact test, P=0.11). Nonetheless, when considering only those rare missense variants that were also predicted to be deleterious by PolyPhen-2, the difference in ASD versus control groups did become significant (3.5% in ASD vs 0.8% in controls; Fisher's exact test, OR=4.37, P=0.04; ). We also examined the distribution of less informative variants (that is, synonymous, untranslated regions and intronic variants) in each group, and found a very similar distribution for all of them (synonymous: 3.5% in ASD vs 2.9% in controls; untranslated regions: 11.1% in ASD vs 9.6% in controls; intronic: 24% in ASD vs 21% in controls; ). The similarity in the incidence of less informative variants argues against the possibility that the difference in the incidence of rare deleterious variants can be attributed to differences in recent ancestry between the ASD and control sample groups.
Figure 1 Burden of novel unique variants in the Wnt pathway in autism spectrum disorders (ASD). Percentage of individuals carrying a novel unique variant that was nonsynonymous and predicted to be deleterious by the PolyPhen-2 algorithm, compared with nonsynonymous, (more ...)
In addition to these novel SNPs, we found one previously referenced rare (<1%) missense SNP that was overrepresented in our ASD group versus our control group. This SNP (rs61758378) occurred in WNT1, the gene encoding the first-discovered mammalian Wnt ligand. To eliminate the possibility of a sequencing artifact we confirmed this result by genotyping all samples using IPlex Sequenom. We also genotyped an additional 91 ASD and 144 control samples to replicate and potentially extend these findings. In this way, we not only confirmed MAF results for rs61758378 obtained using the Illumina platform in the original sample set, we further found that the rs61758378 minor allele remained significantly overrepresented in the combined ASD sample compared with the combined control sample (8 A/T in 267 ASD (MAF=1.69%) vs 1 A/T in 377 controls (MAF=0.13%); Fisher's exact test, allelic P=0.0048; ). Furthermore, when we compared the rs61758378 MAF in our combined ASD sample with the reported MAF in the EVS for a much larger group of individuals, the association also remained significant (8 A/T in 267 in our combined ASD sample (MAF=1.69%) vs 45 A/T in 4300 European/American individuals in the EVS (MAF=0.5%), Fisher's exact test, allelic P=0.011). Because different allele frequencies are reported in European populations (TSI, Tuscans in Italy, MAF=3.0% CEU, Northern Europeans in Utah, MAF=0.5%), we wanted to evaluate and address possible effects of subtle stratification in ancestry between cases and controls. We therefore performed Multidimensional Scaling using genome-wide SNP genotype data for a subset of the total samples used in this analysis (235 SSCs and 329 controls) using PLINK. Multidimensional Scaling using reference CEU and TSI data showed that the ratio of individuals clustering with CEU and TSI was not significantly different between ASD and controls. Nevertheless, we noted that one of the first four dimensions was significantly different between ASD and controls, indicating possible stratification. We sought to account for this using a logistic regression analysis including the first four dimensions as covariates. In this analysis, the odds ratio (OR) remained high (OR=8.17) and the P-value remained close to significance (P=0.053). We note that our analyses have low power for SNPs with such a low MAF and in a modest number of samples.
Association of WNT1 rs61758378 with ASD
The rs61758378 minor allele changes a conserved serine at position 88 into arginine. Although this change is not predicted to be deleterious by PolyPhen-2 (HumDiv score=0.174), serine 88 is nonetheless a very highly conserved residue across species (PhasCons score=0.998), and the change to arginine at this residue is predicted to be deleterious by Grantham score, which categorizes codon replacements into classes of increasing chemical dissimilarity (Grantham score=110, moderately radical change).38
Given the potential association between the WNT1
rs61758378 minor allele and ASD in our sample set, we therefore tested for functional effects of this variant on Wnt signaling pathway activation. To accomplish this, we engineered this variant into a human WNT1 cDNA (WNT1-S88R) using site-directed mutagenesis and then tested it alongside WT Wnt1 in several assays for Wnt signaling activity.
First, we performed a plasmid-based Wnt reporter assay in human embryonic kidney (HEK293T) tissue culture cells. This is a well-established method to quantify transcriptional activation downstream of the Wnt/β-catenin signaling pathway, based in this case on transient transfection of the pBAR-Luc reporter plasmid that expresses luciferase downstream of a β-catenin-responsive promoter.36
Cells were co-transfected with pBAR-Luc and a dose range (1–50
ng) of plasmids encoding WNT1-S88R or WT WNT1 to test for the ability of these proteins to activate Wnt/β-catenin pathway-dependent transcription. At all doses tested, WNT1-S88R showed increased activity in this assay compared with WT WNT1 (). Moreover, this difference was statistically significant at the 20
ng dose (WT WNT1, 1.00±0.09 vs WNT1-S88R, 1.37±0.087, t
=0.02) and at the 50
ng dose (WT WNT1, 1.00±0.18 vs WNT1-S88R, 1.86±0.22, t
=0.016; ). Similar results were obtained when experiments were replicated in a different laboratory by a different investigator using independently prepared DNA constructs and reagents (pSuperTop; data not shown).
Figure 2 Functional analysis of S88R. (a) Plasmid-based Wnt/β-catenin pathway reporter assay in HEK293T cells. HEK293T cells were transfected with the pBAR-Luc reporter plasmid together with plasmids encoding either WNT1-S88R or wild-type (WT) WNT1 at (more ...)
The aforementioned assays rely on transient transfection of reporter plasmid, and on this basis they are conceptually more variable than assays based on read-out of an endogenous Wnt pathway target gene or of a single-copy Wnt pathway reporter integrated into the genome. Accordingly, we confirmed these results in the same (HEK293T) cell line using quantitative reverse transcription-PCR to directly measure transcriptional activation of the endogenous Wnt/β-catenin pathway target gene AXIN2
Mirroring results obtained with the Wnt reporter assay, compared with WT WNT1, WNT1-S88R showed significantly increased transcriptional activation of AXIN2
(WT Wnt1, 1.00±0.025 vs Wnt1-S88R, 2.16±0.43, t
=0.02; ). We confirmed these results via a third in vitro
assay that relies on SH-SY5Y BAR-Luc cells. This is an independently derived immortalized cell line with a single copy of a Wnt/β-catenin pathway luciferase reporter stably integrated into its genome.36
Consistent with data obtained by the plasmid-based Wnt reporter assays and by quantitative reverse transcription-PCR in HEK293T cells, SH-SY5Y BAR-Luc cells showed significantly increased activation of their reporter after transfection of WNT1-S88R (WT Wnt1, 1.00±0.04 vs Wnt1-S88R, 1.28±0.03, t
Taken together, our data from cell-based transcription target assays suggest that the rs61758378 minor allele associated with ASD in our sample (WNT1-S88R) encodes a modestly more active form of the WNT1 ligand compared with the most prevalent WNT1 allele found in Caucasians.