Lack of association of rare functional variants in TSC1/TSC2 genes with autism spectrum disorder

Autism spectrum disorders (ASDs) are a heterogeneous group of neurodevelopmental disorders characterized by impairments in reciprocal social communication and stereotyped behaviors. Though the genetic causes of ASD are diverse [1], mutations in many ASD genes, including NRXN1, NLGN3/4 and SHANK2/3, are associated with altered synaptogenesis, establishing aberrant synaptic protein synthesis and/or synaptic function as a common underlying mechanism in ASD [2, 3]. Single gene Mendelian disorders such as tuberous sclerosis complex (TSC), fragile X syndrome (FXS), Rett’s syndrome and Angelman syndrome show a high prevalence of autism [4]. An understanding of the underlying mechanisms of autism in such single gene disorders may offer insights into the pathogenesis of idiopathic ASD. TSC, characterized by benign hamartomas in multiple organs, is caused by mutations in either of the two tumor suppressor genes encoding hamartin (TSC1) and tuberin (TSC2). Features of ASD are present in 30 to 60% of individuals with TSC [5]. Hamartin and tuberin form a complex that functions as a critical negative regulator of mammalian target of rapamycin complex 1 (mTORC1)-mediated signaling [6, 7].

The TSC proteins act as a central hub in relaying signals from diverse cellular pathways to control mTORC1 activity (Figure 1). mTORC1 signaling in neuronal translation has been established as a regulator of long-lasting synaptic plasticity and memory as it integrates signals from many neuronal surface receptors/channels via ERK- and PI3K/Akt-mediated phosphorylation of the TSC1-TSC2 complex. Translational control is also essential for neuronal functions such as growth and axon guidance [8]. Furthermore, aberrant activation of mTORC1 signaling is a common feature in syndromes associated with autistic phenotypes and cognitive impairments such as TSC, fragile-X, neurofibromatosis-1 (NF1) as well as those with PTEN mutations seen in ASD with macrocephaly [9].

Consequently, we hypothesized that aberrant mTORC1 signaling may be a shared pathway between syndromic ASD and non-syndromic ASD, and that rare functional variants in genes that regulate mTORC1 signaling and/or play a role in synapse development and function may be associated with ASD. In this study, we chose TSC1, TSC2, MYCBP2, RHEB and FBXO45 as candidate genes for non-syndromic ASD and carried out resequencing in 300 ASD trios from the Simons Simplex Collection (SSC) using next-generation sequencing (NGS) technology. The rationale for the selection of these genes is based on their critical roles in neurodevelopment as briefly described below.

Loss of a single copy of Tsc1 in the mouse results in defects in neuronal morphology, including increased soma size, decreased spine density and increased spine size, thus implicating TSC proteins in synapse function in neurons [10]. Mice with a heterozygous inactivating mutation in Tsc2 exhibit abnormal hippocampal CA1 long-term potentiation (LTP), and consequently deficits in hippocampal-dependent learning [11]. Treatment with rapamycin, a specific inhibitor of mTORC1 can improve neuronal hypertrophy and improve behavioral phenotypes in Tsc2 ^+/− mutant mice [11]. Recent studies report rare variants in TSC1 and TSC2 genes to be associated with idiopathic ASD [12, 13].

The small GTPase Rheb (Ras homology enriched in brain) is a direct target of tuberin and hamartin in both Drosophila and mammals [14, 15]. Rheb is highly expressed in the hippocampus and is rapidly induced by growth factors and receptor-dependent synaptic activity, suggesting that Rheb may play an essential role in long term activity-dependent neuronal response [16]. We hypothesized that gain of function mutations in RHEB, resulting in aberrant mTORC1 activation and synaptic connectivity may occur in ASD.

We previously identified Pam (MYCBP2) as an interactor of the TSC protein complex in the mammalian CNS [17]. Human Pam and its homologs (the PHR family of proteins), including Phr1 in mouse, Highwire (HIW) in Drosophila, Regulator of Presynaptic Morphology (RPM-1) in Caenorhabditis elegans, and Esrom in zebrafish, are E3 ubiquitin ligases that have emerged as key regulators of axon guidance, outgrowth and synapse development [18]. Our studies show that Pam interacts with the TSC1-TSC2 complex, and regulates mTORC1 signaling [17, 19, 20]. FBXO45, an F-box protein restricted to the nervous system, is required for normal neuromuscular synaptogenesis, axon pathfinding and neuronal migration. Proteomic analysis has revealed that FBXO45 specifically associates with Pam, forming a novel FBXO45-Pam ubiquitin ligase complex that plays an important role in neural development [21].

We sequenced the coding regions of five candidate genes that regulate mTORC1 signaling and/or are implicated in synapse development and function (TSC1, TSC2, MYCBP2, RHEB and FBXO45) in 300 ASD trios from the SSC. We combined DNA from the 300 trios into 10 pools of 30 ASD probands and 10 pools of the corresponding 60 parents (Additional file 2: Figure S1). Each pool was PCR amplified to capture the 155 coding exons from the five target genes (Table 1). RHEB exon 1 and the first 125 bases of FBXO45 exon 1 could not be successfully amplified and hence are not included in the study. The 165 successful PCR amplicons were combined, concatenated and sheared to construct libraries. The 20 libraries were sequenced using the Illumina Genome Analyzer (Illumina, San Diego, CA, USA).

Sequencing yielded relatively uniform coverage distribution of each exon across all five candidate genes and 20 pools. A representative example of the uniform sequencing coverage obtained is depicted in Figure 2 for two of the largest genes sequenced, MYCBP2 (83 coding exons) and TSC2 (41 coding exons) for two proband pools and their corresponding parental pools. For all coding exons, read depth averaged at least 15,000x for each pool, or 200x for each individual proband/parent sample. Overall, sequence coverage was obtained from 99.4% of all targeted exons and 99.8% of all targeted bases (Table 1). Additional file 4: Table S3 provides the mean and range of sequence coverage over each exon for the five genes examined, with data for each proband and parental pool shown separately. Also, we were able to detect and confirm one variant in a pool of 30 samples (3.33% sensitive) and one variant in a pool of 60 samples (1.66% sensitive) by the analytical tools used.

After filtering for high quality variant calls using Syzygy [23], the variants were classified for analysis by (1) type/location, into nonsense, coding non-synonymous (missense), coding synonymous, intronic, 3'-UTR, 5'-UTR, and 5'- and 3'- splice variants, (2) distribution, into proband (proband pool only), ‘inherited’ (proband and corresponding parental pool) and non-transmitted (parental pool only) variants, and (3) presence/absence in either the dbSNP or 1000 Genomes databases, into reported and novel variants. We prioritized those proband, inherited or parental coding variants, not previously reported in the public databases that produced a premature stop codon (nonsense) or other non-synonymous change, as well as intronic variants that had a potential role in altering splicing based upon branch point sequence analysis (see Methods). The selected variants were confirmed by Sanger di-deoxy sequencing of individual DNA samples comprising the pool(s) in which the variant was identified.

Notably, there were no confirmed de novo events in the probands or inherited nonsense variants in any of the five genes. A total of ten missense and three non-coding intronic variants in MYCBP2, three missense variants in TSC1 and ten missense variants in TSC2 were confirmed. There were no confirmed variants of interest based on the above criteria in FBXO45 and RHEB. A recent study, which examined the metabotropic glutamate-receptor (mGluR) signaling pathway genes for rare genetic variants contributing to autism susceptibility also, did not find any significant coding variants in RHEB[12].

Of the ten missense variants confirmed in MYCBP2, four (K1325R, I1581V, D2437N and N2811D) were observed as variants inherited by the proband from a parent. The remaining six variants (V534I, K543R, V1285I, F1849C, P2533L and R2803H) were identified in different parent samples as singletons, not transmitted to the proband. The inherited variant K1325R and non-transmitted variant V534I were both found in the father of a single trio (Table 2). The three intronic variants (c.1625-19T>C, c.3982+31C>G and c.6727-16T>C) confirmed in MYCBP2 were all inherited (Table 2). None occurred at a classic donor/acceptor site but they were suspected instead to represent potential splicing defects due to their occurrence at putative branch/polypyrimidine sites. However, no exon skipping was observed for any of the three variants when the respective RNA samples were analyzed (data not shown).

For the TSC genes, we confirmed two inherited missense variants (T360N and P448S) in TSC1 and six (I64V, H152D, K533Q, K954R, E984Q and A1429S) in TSC2. Four of these (TSC1: T360N, P448S; TSC2: K533Q, Q984N) have been previously reported as potential risk factors for ASD [12, 13]. We also identified one non-transmitted parental missense variant (L116V) in TSC1 and four (C728F, P878S, S960F and T1330M) in TSC2 (Table 3). Two independent studies have reported differing roles for the TSC1 L116V variant, as an ASD specific variant [12] and as a control variant [13], though it appears here to be a rare polymorphism (Table 3). All the three TSC1 variants (inherited variants T360N and P448S, and non-transmitted variant L116V) are also reported in the tuberous sclerosis database - Leiden Open Variation Database (LOVD, http://​chromium.​liacs.​nl/​LOVD2/​TSC/​home.​php) - as probably non-pathogenic. However, all the TSC2 variants excepting, A1429S and T1330M are novel (Table 3). The variants T360N in TSC1 and A1429S in TSC2 were detected in both parents; however, the respective probands were heterozygous for these changes (Table 3).

As an additional large scale control dataset, we used data from the National Heart, Lung, and Blood Institute exome variant server (NHLBI EVS) and found five variants in MYCBP2 (V534I, K543R, V1285I, R2803H and N2811D), all three TSC1 variants (L116V, T360N and P448S) and a single TSC2 variant (A1429S) to be reported in the controls (Tables 2 and 3).

Overall, we did not find a significant enrichment of rare inherited variants in ASD probands as compared to the non-transmitted variants in parents for MYCBP2 (P > 0.05) or the TSC genes (P > 0.05).

An additional review of TSC missense mutations reported in LOVD with NHLBI EVS suggested that 19 TSC1 and 56 TSC2 variants were seen in controls. Furthermore, missense variants in TSC genes reported to have functional impact [30‐32] were catalogued and found not to be seen in idiopathic ASD cases, suggesting that mutations observed in frank TSC are not seen in ASD (Additional file 5: Table S4).

Dysregulation of mTORC1 signaling is correlated with aberrant synaptic protein synthesis and has emerged as a common biochemical pathway in many syndromes associated with ASD. Because approximately 50 to 60% of TSC patients exhibit autistic phenotypes, we hypothesized that functional rare variants in genes that regulate mTORC1 signaling may be frequent in non-syndromic ASD, which may explain the autistic phenotypes shared between TSC and ASD in general. We investigated five mTOR pathway candidates, TSC1, TSC2, MYCBP2, RHEB and FBXO45, based upon their critical roles in neurodevelopment. Mice lacking these genes exhibit defects in synapse development, axon navigation and hippocampal-dependent learning [20, 21, 33‐35]. The findings of our study however do not support the hypothesis and suggest that rare variants in the studied mTOR pathway candidate genes do not play a significant causal role in autism. We did not detect support for a genetic association between any of the studied genes and autism. No de novo variations, considered to be a hallmark of the risk-determining genes of greatest effect, were identified; neither did we identify any nonsense variants. In MYCBP2, TSC1 and TSC2, we observed a few rare inherited single nucleotide variants in ASD probands, but a comparable number of variants were also found in the parents indicating that the ASD variants are probably rare private sequence changes.

Next-generation sequencing of pooled samples is shown to be efficient and cost-effective in capturing variation in a limited target region that has been selectively amplified in multiple DNA samples as shown in studies to identify genes in type 1 diabetes, human complex 1 deficiency and inflammatory bowel diseases [23, 24, 36]. The approach is capable of deriving very deep coverage in targeted regions and as such can be sensitive to the detection of rare or private events within the pool [24]. However, it should be noted that we chose to validate only the subset of presumably deleterious and thus potentially pathogenic variants, a strategy that was not specific to detection and confirmation of the full mutational burden. Nonetheless, our negative findings are consistent with results from three recent studies that examined the patterns of de novo point mutations in ASDs by whole exome sequencing [37‐39]. None of the three studies, all of which reported overlapping results with respect to genes conferring susceptibility to autism, identified the mTOR pathway genes that we investigated as genuine autism risk factors, though de novo mutations were identified in two of the genes, MYCBP2 and TSC2. Neale et al.[37] identified a single de novo missense mutation in the myc-binding domain of MYCBP2. The mutation was however predicted to be benign by PolyPhen-2. These findings do not rule out Pam (MYCBP2) as a functional participant in autism as it is probable that Pam, an E3 ubiquitin ligase, may be functioning only at the translational level as evidenced by recent work from our laboratory on mTORC1 regulation by Pam in the brain using mouse models [20]. Pam has been identified as a potential substrate whose phosphorylation is directly or indirectly controlled by mTORC1 in two independent phosphoproteome studies [40, 41]. Furthermore, Pam has also been identified as a target transcript of fragile X mental retardation protein (FMRP) [42]. Both mTOR and FMRP play key roles in neuronal translation, and recent evidence indicates a link between mTORC1 signaling and FMRP. The role of Pam in regulating TSC/mTORC1 signaling in neurons, and more importantly the regulation of Pam by both mTORC1 and FMRP as well as the function of Pam in synapse development, suggest that either a direct or indirect involvement of this protein is possible in a subset of ASD.

Mutations in the TSC genes, TSC1 and TSC2 are known to cause syndromic autism [33]. In non-syndromic autism patients, TSC rates of 1.1% to 1.3% have been reported in many studies [43‐47]. Two recent studies looked at the possible role of TSC1 and TSC2 genes along with others in non-syndromic autism [12, 13]. Schaaf et al.[13] introduced the concept of ‘oligogenic heterozygosity’ of coding non-synonymous variants as a novel pathogenic mechanism for ASD risk. The study suggested that groups of variants with smaller effects in multiple genes acting together might be contributing to autism in patients with near normal/normal IQs and non-syndromic autism. Since our study did not examine the other candidate genes included in that model, we believe that we can neither support nor refute the ‘oligogenic heterozygosity’ model. Kelleher et al.[12] found a significant excess of TSC gene single nucleotide variants in the AGRE (Autism Genetic Resource Exchange) samples compared to controls. Table 4 summarizes TSC variants identified by our study and those by Kelleher et al.[12] and Schaaf et al.[13]. Of the 20 TSC variants identified as ASD specific by Kelleher et al.[12], 7 variants (35%) are reported in the NHLBI EVS (Table 4). Also, of the 12 variants identified as seen only in ASD probands and not controls by Schaaf et al.[13], 6 variants (50%) are reported in the NHLBI EVS (Table 4), raising the question of whether these variants produce pathogenic effects. A damaging de novo mutation R1580W has been identified in the GAP-domain of TSC2 in one of the recent autism exome sequencing studies [39] raising the question as to whether a subclinical TSC was present in that patient. The functional impact of this variant however needs to be assessed.

Given that we have demonstrated our approach to be sensitive to detection of a single private mutation in each pool, our study has comparable or greater power to detect an association of similar effect size to those reported from de novo mutations in each of the previous exome studies (Neale et al., n = 175 trios [37]; O'Roak et al., n = 209 families [39]; Sanders et al., n = 238 families [38]). With 300 trios, we calculate that we have approximately 80% power to detect a scenario where 2.5% of cases are caused by rare variants in TSC1 and 4.2% of cases are caused by rare variants in TSC2 at an experiment-wide significance threshold.

In our study, because we found a comparable number of transmitted and non-transmitted rare coding variants, we cannot provide support for an important role of rare variants in TSC genes in the pathogenesis of non-syndromic autism. Exome sequencing data of 1,000 cases and matched controls from Neale and Daly laboratories have observed TSC1 missense variants that are predicted to be damaging by PolyPhen-2 in 15 cases and 12 controls and a similar ratio of TSC2 missense variants in 63 cases and 60 controls (Daly M, unpublished data). This is consistent with our observation and those of previous exome sequencing studies that any ASD risk conferred by TSC gene variants is likely to be modest.

Our results suggest that in this sample of 300 trios ascertained, based on a proband with idiopathic ASD, there is no significant enrichment of ASD-risk variants in the candidate genes TSC1, TSC2, MYCBP2, RHEB and FBXO45. We did not observe any de novo or nonsense mutations with putative functional impact in any of the genes studied. Consistent with the null hypothesis, the number of rare missense variants transmitted to ASD cases was comparable to the untransmitted variants. Missense variants in TSC genes reported to have functional impact [30‐32] are not observed in any of the idiopathic ASD cases [12, 13, 37‐39] including those from our study and those in unpublished data from Neale and Daly. It remains essential to examine the functional impact of the missense variants detected in idiopathic ASD to conclude as to whether rare functional variants in TSC1/TSC2 could be a very rare cause of non-syndromic autism. Our data do not necessarily undermine the potential functional role of mTORC1 signaling in ASD. It remains possible that genes associated with ASD will ultimately be linked directly or indirectly to this pathway, known for its complexity, with many feed-forward and feed-back regulatory features.