Rare deleterious mutations of the gene EFR3A in autism spectrum disorders

Autism spectrum disorders (ASDs) are defined by persistent deficits in social communication and social interaction and restricted repetitive patterns of behavior, interests or activities [1]. These syndromes are common in the population, with a prevalence of approximately 1% [2], and demonstrate both considerable phenotypic and extensive genetic heterogeneity [3]. High-throughput sequencing approaches have provided substantial insight into the genomic architecture of ASDs. For example, multiple analyses of whole-exome sequencing data demonstrate an over-representation of de novo, loss-of-function mutations in brain-expressed genes in affected individuals and point to half a dozen new ASD genes [3‐6]. These have been identified based on the clustering of mutations in the same gene in unrelated individuals, providing strong evidence for association [3]. However, a large number of compelling, rare, de novo missense mutations are also found in probands, though a clear threshold for identifying the association of these mutations with ASD is less obvious. Both the rarity of the individual mutations and the small size of current exome discovery cohorts suggest that clarifying which of these de novo mutations point to bona fide ASD genes will require alternative approaches. Large-scale, targeted, case/control sequencing as a complement to de novo mutation discovery in ASD is one such strategy.

In a whole-exome analysis of 238 families [3], we identified a single proband carrying two novel de novo missense mutations in synaptic genes, one each in EFR3A (Eighty-five Requiring 3A [NCBI Reference Sequence: NM_015137]) and CASK (Calcium/Calmodulin-dependent Serine/Threonine Kinase [NCBI Reference Sequence: NM_003688]). Both yeast EFR3[7] and mammalian EFR3A and EFR3B[8] have been linked to the control of phosphoinositide metabolism, a pathway demonstrated to play a role in ASD [9]. CASK is implicated in X-linked intellectual disability [10]. Neither the occurrence of one or several de novo missense mutations in a single affected individual is a statistically significant finding. However, our overall analysis of 599 simplex ASD quartets suggests that approximately 20% of de novo missense mutations in brain-expressed genes found in cases will prove to be true ASD loci, representing an approximately fourfold increase over a brain-expressed gene chosen at random [11]. Given an increased prior probability based on the exome results and the strong biological plausibility of both genes, we conducted a targeted analysis of EFR3A and CASK in large cohorts using both Sanger sequencing and whole-exome data. We found that rare deleterious mutations in EFR3A are associated with ASD using an experiment-wide significance threshold.

To test our hypothesis that mutations in EFR3A and/or CASK confer risk for ASD, we performed Sanger sequencing of all coding exons and splice sites of both genes in 705 comprehensively phenotyped European cases from the SSC. All rare (<1% frequency) nonsynonymous variants were confirmed by a second round of Sanger sequencing. We focused on novel alleles seen only once (singleton variants) and not present in two large databases, dbSNP137 and Exome Variant Server, the latter of which had 6,503 exomes. We reasoned that this strategy would most likely identify deleterious substitutions subject to purifying selection and provide, along with case/control matching for ancestry, the most robust protection against population stratification [22].

In CASK, only two variants met these criteria among all 705 cases (Additional file 11: Table S6). In light of the low cumulative allele frequency and anticipated low power to detect an effect, we did not pursue this gene further. We identified six novel nonsynonymous singleton mutations in EFR3A (Table 1 and Additional file 12: Table S7) and, consequently, we proceeded to screen this gene in several cohorts for which we had access to DNA or whole-exome sequencing data. We identified an additional 1,491 European cases: (1) 452 from the SSC were subjected to Sanger sequencing and (2) 1,039 from the AASC had whole-exome sequencing data, for a total of 2,196 cases. We identified a total of 3,389 European controls: (1) 912 NINDS neurologically normal European controls matched to SSC cases via PCA of genotyping data and subjected to Sanger sequencing, (2) 1,614 neuropsychiatrically unscreened controls of NE origin matched to SSC cases and who had whole-exome sequencing data, and (3) 863 from the AASC with whole-exome sequencing data. For the NE control exomes, a minimum read threshold of only one independent read was used to identify variants in an effort to minimize false negatives. For the AASC dataset, Broad cases/controls and Baylor cases/controls were matched and evaluated using identical variant-calling approaches and filtering criteria within each site. All rare nonsynonymous variants in SSC cases, NINDS controls and NE controls were confirmed by Sanger sequencing; confirmations were not available for case and control AASC variants.

The analysis of a total of 2,196 cases and 3,389 controls demonstrated that novel nonsynonymous singleton mutations in EFR3A were twice as frequent in cases compared to controls. However, the P value was not statistically significant when corrected for the investigation of two genes since we initially analyzed CASK as well as EFR3A (16/2,196 cases and 12/3,389 controls; P = 0.084, odds ratio = 2.065, 95% confidence interval = 0.924 to 4.652, Fisher exact test, right-tailed). Since the combination of low allele frequency and high conservation has been shown to provide high sensitivity and specificity for predicting functionality in rare variant studies, in contrast to in silico prediction programs [23], we evaluated conservation with three widely used informatics tools: PhyloP, GERP and ConSurf. All found significantly more case variants mapping to conserved positions (Tables 1 and 2). Furthermore, using 10,000 permutations of the cases and controls to test the significance of the enrichment of deleterious mutations in cases, we calculated a P value of 0.0077. We evaluated family data from SSC subjects and found that all newly identified variants were transmitted. Whole-exome data for only two of these subjects (11379.p1 and 11808.p1) have now been reported; neither has a de novo loss-of-function mutation which might contribute to their phenotype. SSC case 11808.p1 does have a novel de novo missense mutation (N160S) in DGCR14 (DiGeorge Syndrome Critical Region Gene 14), which has not been associated with ASD or intellectual disability [24]. We also determined that all of the SSC subjects except one (13507.p1) have undergone genome-wide copy number variant (CNV) analysis; none have de novo CNVs that might better explain their phenotype [25].

We took advantage of the recently available crystal structure (Protein Data Bank ID 4N5A) of the N-terminal fragment of S. cerevisiae Efr3 [18] to map the human mutations (Figure 1) and determine their potential to disrupt protein structure and function, blinded to case/control status. Because the C-terminal portion is not well conserved, only mutations up to and including amino acid 451 could be evaluated with high confidence. (The full length EFR3A protein has 821 amino acid residues.) Every mutation that was assessed to be deleterious as informed by the crystal structure was a case variant, except R161*, which is assumed to be damaging (Table 1 and Additional file 13: Table S8). R161* was found in an NE control for whom neuropsychiatric information is unavailable, so we cannot determine if this mutation is associated with any neuropsychiatric condition. Thus, crystal structure analysis also identifies a significantly greater number of deleterious mutations in cases than controls (P = 0.017, odds ratio = 9.282, 95% confidence interval = 1.119 to 204.784, Table 2). As would be expected, all of these deleterious mutations are also at highly conserved positions as per PhyloP, GERP and ConSurf. Interestingly, the reverse is not always true, i.e., there are mutations at highly conserved positions which were assessed to be benign in light of the crystal structure. Therefore, having knowledge of the three-dimensional structure of Efr3 enriches our analysis by providing more biological information to evaluate the deleteriousness of mutations. We also noted for subjects with family data, deleterious mutations as per the crystal structure are not shared by the unaffected siblings (Table 1).

EFR3A is a member of the EFR3 family of genes, conserved throughout eukaryotes and essential for viability [7]. The Drosophila melanogaster homologue, rolling blackout (RBO), is highly expressed in the nervous system [26], is enriched at the neural synapse [27], and was proposed to regulate phospholipase C signaling [26]. RBO has also been proposed to function as a transmembrane lipase [26], but structural analysis of Efr3 does not support this hypothesis (Additional file 14: Table S9) [18]. Instead, it shows that EFR3/RBO has a scaffold function with the majority of the protein comprising alpha-helical HEAT (Huntington, Elongation factor 3, regulatory subunit A of protein phosphatase 2A, and Target of rapamycin) repeats.

The tissue expression of EFR3A has not been described, so we performed Western blot analysis of several mouse tissues and found that EFR3A is broadly expressed, including in the brain (Additional file 15: Figure S6). We also analyzed its expression using exon-array data from a study of the spatio-temporal transcriptome of the human brain [19]. There is a steady increase in EFR3A mRNA levels in multiple brain regions through fetal development and into adolescence (Figure 2A). In situ hybridization of adult human dorsolateral prefrontal cortex revealed the presence of EFR3A in cortical neurons including pyramidal neurons (Figure 2B). This pattern is consistent with prior data on the expression of ASD genes [9, 19], as well as functional annotation of genes that are highly co-expressed with ASD genes, showing enrichment for a category associated with the development of cortical projection (pyramidal) neurons [19].

We next identified the top 100 genes co-expressed with EFR3A (Additional file 16: Table S10) using the same dataset [19]. Gene ontology enrichment analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID v6.7) [28, 29] revealed synaptic genes, including SYNJ1, the major PtdIns(4,5)P₂ phosphatase in the brain [30], as the most significant finding (Figure 2C). We also compared the expression profile of EFR3A with a discrete module of co-expressed genes (M12) significantly associated with ASD in a prior transcriptome analysis of post-mortem autism and control brains [20]. M12 is enriched for genes involved in synaptic function, vesicular transport and neuronal projection and is downregulated in the autistic brain. We compared the distribution of expression correlation coefficients between EFR3A and M12 genes (Figure 2D) and between EFR3A and all brain-expressed genes [19] (Figure 2E). We found that the distribution between EFR3A and M12 genes was significantly skewed toward positive correlation coefficients compared to the distribution between EFR3A and all brain-expressed genes (P < 2.2 × 10⁻¹⁶, Wilcoxon test). When a similar analysis was performed on the homologue EFR3B (which is largely brain-expressed) and eight genes strongly associated with ASD from recent CNV and exome studies [3‐6, 25], EFR3A was the most strongly correlated with M12 expression (Figure 2D). We repeated this process with three additional modules of co-expressed genes (M2, M3 and M16) identified by a prior analysis of BrainSpan transcriptome data from normal brains [21]. All are significantly associated with ASD candidate genes, although M2 and M3 are enriched for early fetal transcriptional regulators affected by de novo loss-of-function mutations in ASD, while M16, which has significant overlap with M12, is enriched for synaptic genes upregulated during late fetal/early postnatal stages and genes harboring inherited common variants in ASD. As might be expected given its developmental expression pattern and synaptic function, EFR3A is positively correlated with M16 (Additional file 17: Figure S7A; P < 2.2 × 10⁻¹⁶, Wilcoxon test) and negatively correlated with M2 and M3 (Additional file 17: Figure S7B,C; P < 2.2 × 10⁻¹⁶, Wilcoxon test).

Our conservation, structure-based functional and expression analyses suggest a role for rare deleterious EFR3A mutations in the risk for ASD, adding to the emerging data on specific synaptic functions, including phosphoinositide metabolism, relevant to these disorders. Multiple resequencing projects for ASD have revealed numerous rare variants in both cases and controls. Given the over-representation of de novo loss-of-function mutations in cases, it is implausible that a subset of damaging missense mutations does not carry risk as well. However, differentiating relevant functional mutations from the large collection of neutral background variation remains a challenge. We have approached this issue by following up an observation of a de novo mutation in an ASD proband with a relatively large case/control analysis, relying on diverse approaches to identify putatively deleterious variants. While the overall burden of singleton variants was not impressive, the use of multiple conservation measures and crystal structure analysis to segregate functional variation showed consistent evidence for experiment-wide association with ASD.

Our results would clearly not survive correction for genome-wide comparisons. Of course, given the distribution of singleton mutations across the genome, this statistical threshold, if applied to every targeted analysis, would demand implausibly large case/control samples. In an effort to skirt this problem, we used an initial observation in an unbiased exome-wide study to establish a narrow hypothesis and then relied on an experiment-wide P value threshold for our case/control analysis. At present, this seems a reasonable approach to evaluating single gene association. Additional data on the distribution of de novo missense mutations in the genome and the integration of ASD risk associated with varying classes of mutations [31] with co-expression network data [11, 21] will shed significant light on the contribution of any one gene to ASD.

Our expression data, combined with evidence for the involvement of EFR3A in synaptic phosphoinositide metabolism [8], suggest that EFR3A may play an important role in synaptic function during human fetal brain development. In addition to the significant conservation and structure-based findings, our analysis comparing the expression profile of EFR3A with M12 and M16 further suggests that this gene is associated with ASD. Not only are EFR3A and M12/M16 expression strongly correlated but EFR3A is also the most strongly correlated in the context of ASD-associated genes and its homologue EFR3B. Although co-expression data do not prove that a gene causes a disorder, they can provide another piece of supportive evidence [11, 21]. The determination of when in development and in what cell types EFR3A is expressed provides insight into how EFR3A mutations might contribute to the pathophysiology of ASD.

A potential limitation of this study is that we combined data from Sanger sequencing (SSC cases and NINDS controls) and whole-exome sequencing (AASC cases/controls and NE controls). It is possible that the two techniques can yield different sets of variants. As described under Methods, for the NE controls, we determined that >98% of the coding and splice site sequences were covered by at least eight independent reads. To minimize false negatives in controls that might bias toward an excess of rare mutations in cases, a minimum of only one independent read was used to identify variants for confirmation. Regarding the AASC samples, case and control exome data were subjected to identical variant-calling approaches and filtering criteria within each site and were, therefore, treated equally, suggesting that any error should be randomly distributed between these groups.

Another limitation is that we did not find additional de novo mutations in the subjects for whom family DNA was available (only SSC cases), which would strengthen the association of EFR3A mutations with ASD. However, there is abundant evidence that inherited mutations also contribute to ASD [31]. The presence of SSC case mutations in unaffected parents and/or siblings points to incomplete penetrance, as expected in complex genetic disorders such as ASD. The crystal structure analysis was able to stratify the mutations further by determining that potentially deleterious variants were generally not shared by siblings. Although this is an interesting observation, it is based on a very small number of events (five SSC case mutations with both sibling data and crystal structure information) and cannot be assigned statistical significance. We did observe one premature stop codon mutation in an NE control as well as in an SSC case, indicating that EFR3A mutations are not sufficient to cause ASD. However, neuropsychiatric data was not available for this control cohort. Moreover, the identification of well-established ASD-associated variants in unscreened controls is so commonplace as to be expected in a study such as this one.

EFR3A is a critical component of a complex containing a phosphatidylinositol 4-kinase that synthesizes the plasma membrane pool of the phosphoinositide PtdIns4P, the direct precursor of PtdIns(4,5)P₂[8]. PtdIns(4,5)P₂ has a wide variety of direct functions in the central nervous system, including regulation of exo/endocytosis, ion channel function, neurotransmitter receptors, and transporters and nucleation of the actin cytoskeleton [32, 33]. Additionally, PtdIns(4,5)P₂ is a precursor to numerous signaling metabolites: diacylglycerol and InsP₃ (via phospholipase C activity), which are key regulators of Ca²⁺ signaling, and PtdIns(3,4,5)P₃ (via PI 3-kinase activity), which mediates many cellular processes such as activation of the Akt/mTOR signaling pathway [30]. Mutations in PTEN, which encodes a PtdIns(3,4,5)P₃ phosphatase, and in TSC1 and TSC2, which are key effectors in the PtdIns(3,4,5)P₃ signaling pathway, have demonstrated the importance of synaptic phosphoinositide signaling in syndromic forms of autism (Figure 3) [9, 34, 35]. Common polymorphisms and rare CNVs in MET, another gene involved in phosphoinositide metabolism, have implicated this pathway in idiopathic ASD as well [36, 37].

The identification of rare deleterious mutations in EFR3A, a gene linked to PtdIns4P synthesis (Figure 3), further strengthens the role of phosphoinositide metabolism in ASD. The precise effects of EFR3A on the levels of various phosphoinositides still have to be determined, and an EFR3A knock-out mouse is not yet available. Delineating the molecular details and functional significance of interactions between EFR3A and its binding partners will allow the development of in vitro assays to assess further the severity of the variants we report here. Importantly, phosphoinositide metabolizing enzymes are pharmacologically targetable [38‐40]. The applicability of this approach towards ASD has been shown for the closely connected mTOR pathway in mouse models of tuberous sclerosis [41, 42]. Therefore, mutations in EFR3A and perturbations in phosphoinositide metabolism may point to a potential avenue for treatment in a subset of ASD patients.

Rare nonsynonymous mutations in EFR3A are significantly more common among ASD cases than controls at positions that are conserved and positions that would be disruptive to protein structure and function based on analysis of the Efr3 crystal structure. These results further implicate phosphoinositide metabolism in the pathophysiology of ASD, a pathway that is pharmacologically targetable. Exactly how EFR3A mutations contribute to that pathophysiology will have to await further delineation of how the protein functions and the development of specific assays to test their severity.

NINDS Neurologically Normal Caucasian Control Panel: [http://​ccr.​coriell.​org/​Sections/​Collections/​NINDS/​DNAPanels.​aspx?​PgId=​195&​coll=​ND/​]

AASC controls: [https://​www.​nimhgenetics.​org/​available_​data/​controls/​]

NCBI dbSNP: [http://​www.​ncbi.​nlm.​nih.​gov/​snp]

EVS: [http://​evs.​gs.​washington.​edu/​EVS/​]

UCSC Genome Browser: [http://​www.​genome.​ucsc.​edu/​]

SeattleSeq: [http://​snp.​gs.​washington.​edu/​SeattleSeqAnnota​tion138/​]

ConSurf: [http://​consurf.​tau.​ac.​il/​]

DAVID v6.7: [http://​david.​abcc.​ncifcrf.​gov/​]