Key role for gene dosage and synaptic homeostasis in autism spectrum disorders

Autism spectrum disorders (ASD) are characterized by impairments in reciprocal social communication, and repetitive, stereotyped verbal and non-verbal behaviors. Genetic studies have provided a relatively large number of genes that constitute a comprehensive framework to better understand this complex and heterogeneous syndrome. Based on the most robust findings, three observations can be made: first, genetic contributions to ASD are highly heterogeneous with most likely a combination of alleles with low and high penetrance. Second, the majority of the mutations apparently affect a single allele, suggesting a key role for gene dosage in the susceptibility to ASD. Finally, the broad expression and function of the causative genes suggest that alteration of synaptic homeostasis could be a common biological process associated with ASD. Understanding the mechanisms that regulate synaptic homeostasis should shed new light on the causes of ASD and may provide means to modulate the severity of the symptoms.


Introduction
The diagnosis of autism is based on impairments in two major domains -reciprocal social communication, and repetitive, stereotyped and ritualistic verbal and non-verbal behaviors. The term "autism spectrum disorders" (ASD) is used as a shorthand to refer to any patient that meets these diagnostic criteria. But beyond this unifying definition lies an extreme degree of clinical heterogeneity, ranging from profound to moderate impairments, but always with functional disability. Indeed, autism is not a single entity, but rather a complex phenotype thought to be caused by different types of defects in common pathways, producing similar behavioral phenotypes. The prevalence of ASD overall is about 1/100, but closer to 1/300 for typical autism 1 . ASD are more common in males than females with a 4:1 ratio 2, 3 . Twin and family studies have conclusively described ASD as the most "genetic" of neuropsychiatric disorders, with concordance rates of 82-92% in monozygotic twins versus 1-10% in dizygotic twins; sibling recurrence risk is 6% 2, 3 .
From 15 to 70% of children diagnosed as suffering from ASD have intellectual disabilities 4 , and it is now understood that, like intellectual disability, autism symptoms can be caused either by gene mutations or by chromosomal aberrations (Box 1). In approximately 10-25% of the affected individuals, autism is "syndromic", ie. occurring in a child with a known genetic or environmental toxin disorder, such as fragile X, tuberous sclerosis, neurofibromatosis, valproic syndrome, or autism caused by brain herpes simplex infection 2, 4 .
In the last years, various independent studies and large scale international efforts have identified a growing number of candidate genes for ASD and suggest a set of mechanisms that could underlie the ASD phenotype.

Genetic variations and the modes of inheritance of ASD
Due to the absence of classical Mendelian inheritance, ASD were first thought to be a polygenic trait involving many loci. Therefore, model free linkage studies, such as affected sib-pair analyses, were performed to identify susceptibility genes. Many genomic regions were detected, but only a restricted number of loci were replicated in independent scans (e.g., chromosome 7q31 and 17q11). To homogenize the genetic and phenotypic data and to gain higher statistical power, collaborative efforts 4 were initiated, such as the autism genome project (AGP), that genotyped 1496 sib-pair families using the Affymetrix 10K single nucleotide polymorphisms (SNP) array 5 . Nevertheless, no genome-wide significant loci could be detected, and the signals on chromosome 7q31 and 17q11 were lost. The absence of relevant targets identified by linkage studies prompted geneticists to use an alternative method: association studies with dense SNP arrays. In theory, association studies are sensitive to allelic heterogeneity whereas linkage studies are not. Nevertheless, association studies provide major advantages compared with linkage studies. First, studies can include large sample of patients since they are not restricted to multiplex families with two or more affected children. Second, the genomic regions associated with the trait are much narrower than in linkage studies, due to loss of strong linkage disequilibrium between relatively close genomic regions. In addition, SNP arrays can be used to detect structural variants such as copy number variations (CNVs) 6 .
By using these approaches, several genes were associated with ASD. A list of 190 genes is available at AutDB, a public, curated, web-based, database for autism research (http://www.mindspec.org/autdb.html). However, it should be noted that most of these genes remain only candidates since their association was not always confirmed by replication and/or functional validation. In Tables 1 and 2, we present a list of some of these genes, and we consider the mechanisms that might be affected. Depending on the penetrance of the mutation on the risk for ASD, we derive two main categories of genes or loci. In the first (Table 1), genes or loci appear to have a high penetrance, but are mutated in a limited number of individuals (sometime a single individual). In this category, variations are mostly composed of de novo or rare point mutations, CNVs and cytogenetically detected deletions/duplications (Box 1). The second category of genes includes the socalled susceptibility genes to ASD (Table 2). Here, the variations are mostly composed of SNPs or inherited CNVs observed in the general population and associated with low risk for ASD. Especially, in this category of genes, the association with ASD should be taken with great care, since the three largest genome-wide association studies (GWAS) performed on more than 1000 patients in each study could not detect the same genes associated with ASD 7-9 .
Two opposite models have been proposed to explain the inheritance of ASD. Some researchers have considered autism as a polygenic trait (i.e., a group of patients with a relatively 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 5 homogeneous phenotype produced by multiple genes, each of low effect). Others have considered autism as a very phenotypically heterogeneous group of different disorders (autisms), thought to be caused by different genes in common pathways (in this model, eventually a single highly penetrant mutation would be sufficient to produce autism). The heterogeneous model was recently supported by both the absence of strong linkage and/orassociation loci replicated by whole genome scans and by the identification of apparently monogenic forms of autism, each time affecting a limited number of patients, 1-2% for the most replicated genes. Nevertheless, the polygenic model of autism cannot be excluded for several reasons. First, it is now well established that the clinical outcome of most of the monogenic disorders are actually modulated by additional genetic variations. Second, in polygenic traits such as blood pressure and height for example, the effect of a single common variant on the phenotype seems to be very low (OR<2) and requires large sample size (>15000) to be detected 10 . In ASD, the current genome scans -the largest were performed on cohorts of less than 2000 casesmight lack the statistical power necessary to detect such common alleles with low effect. Finally, the number of deleterious mutations within the human genome of one individual remains difficult to establish, but it is most likely that, in a large proportion of patients, the combination of multiple rare variants such as CNVs, coding and regulatory variations could affect specific biological pathways and therefore increase the risk for autism.

Abnormal gene dosage in ASD
In recent years, enormous progress has been achieved in detecting rare and frequent structural variants of the human genome, such as deletions, duplications and inversions 11 . At least eight studies have searched for such genomic imbalances in patients with ASD 5, 12-19 . Differences in patient inclusion criteria, genotyping methodologies and algorithms to detect deletions and duplications make the comparison of these results difficult. Nevertheless, it seems that there is a significant increase in rare inherited and de novo CNVs in the ASD population compared with the general population. The overall rate of de novo CNVs in ASD could range from 5-10% compared with 1% in the general population 15 .
Not surprisingly, the frequency of rare and de novo CNVs increases for simplex families with one affected child (7-10%) compared with multiplex families with at least two affected children (2-3%) 15 .

6
The presence of dysmorphic features in the patient also increases the odds of detecting a rare or de novo CNV in up to 27.5% 12 . Based on the current findings, the majority of the CNVs apparently affect only one copy of the gene (which can be either deleted or duplicated), suggesting that abnormal gene dosage or expression might play a key role in the susceptibility to ASD.
Several mechanisms could explain why abnormal gene dosage would increase the risk for ASD. First, deletions may reveal a mutation on the second allele (Fig. 1A), as it has been recently shown for NRXN1 and CNTNAP2, two genes associated with ASD. Patients with Pitt-Hopkins-like syndrome (severe mental retardation, autistic behavior, epilepsy, and breathing anomalies) have been reported to carry one CNV affecting either NRXN1 or CNTNAP2, and a deleterious point mutation affecting the same genes on the second allele 20 . Therefore, this apparently dominant trait was actually a recessive trait when a mutation screening of the remaining allele was performed. In theory, the second mutation could be present only in some tissues, maybe in specific parts of the brain (Fig. 1B).
This situation has not been observed in ASD yet, but is well known in familial forms of cancer where both an inherited and a somatic mutation are required to develop the disease (the "two hit" or Knudson model). Gene dosage problems may also appear even if the second copy does not present a mutation, but is silenced by epigenetic hallmarks, as it has been observed to be the case in disorders affecting imprinted loci, such as Angelman or Prader-Willi syndromes 21 . Similarly, allelic exclusion (i.e., the expression of a single allele in one cell) could turn off the unaffected allele in all cells or in specific cell types (Fig. 1C). Allelic exclusion was well characterized for immunoglobulins and olfactory receptors and is one of the mechanisms ensuring that a single antibody or receptor is expressed per cell 22 . This mechanism, first considered as rare, could be in fact more frequent than originally thought 23 . Especially in the brain, allelic exclusion has been observed for cell adhesion molecules such as protocadherins 24 . If the cadherins and protocadherins associated with ASD 7, 14, 25 were subject to allelic exclusion, then it would be possible that turning off the expression of the unaffected copy could totally deplete the cell of the protein product.
Second, abnormal gene dosage might by itself represent a risk factor for ASD. It is known from work in yeast that genes with high sensitivity to dosage are enriched in regulatory and structural proteins 26 , that need to interact with precise stoichiometry. Thus, lower or higher levels of a single 7 component of these complexes might disorganize the assembly of the machinery and alter its biological function. In particular, scaffolding synaptic proteins such as SHANK involved in specific protein-protein interactions are expected to be especially sensitive to dosage ( Figure 1D). Finally, some mutation associated with ASD could act as gain of function. Indeed, the knock-in mice carrying the Nlgn3 R451C mutation -previously identified in one patient with autism and his brother with Asperger syndrome 27 -displayed different synaptic and behavioral phenotypes compared to the complete Nlgn3 knock-out 28 .

Abnormal level of synaptic proteins
Several lines of evidence indicate that mutations in genes regulating various aspects of synaptogenesis and neuronal circuit formation (Fig. 2) are associated with an increased risk for ASD. Among these, several genes seem to regulate the level of proteins at the synapse. Two X-linked genes, MeCP2 and FMR1, are involved in autism "secondary" to Rett and fragile X syndromes, respectively. MeCP2 ( Fig   2B) is a protein that directly and/or indirectly regulates neurotrophic factors, such as Brain Derived Neurotrophic factor (BDNF), by binding to methylated DNA 29 . Deletions or mutations of MECP2 are associated with Rett syndrome in females, whereas duplications of MeCP2 are associated with mental retardation and ASD in males and psychiatric symptoms, including generalized anxiety, depression, and compulsions in females 30 . FMRP (Fig 2A, B) is a selective RNA-binding protein that transports mRNA into dendrites and regulates the local translation of some of these mRNAs at synapses in response to activation of metabotropic glutamate receptors (mGluRs). In the absence of FMRP, there is an excess and a dysregulation of mRNA translation leading to altered protein synthesis dependent plasticity 31 .
Mutations of other genes associated with ASD seem to affect the level of synaptic proteins by dysregulating overall cellular translation 31 . Patients with neurofibromatosis, tuberous sclerosis, or Cowden/Lhermitte Duclos syndromes have higher risk than the general population to have ASD.
Consistent with the hypothesis of a relationship between abnormal levels of synaptic proteins in ASD, many studies have reported mutations in genes involved in synaptic protein ubiquitination, including UBE3A, PARK2, RFWD2 and FBXO40 13 ( Fig. 2A). Protein degradation through ubiquitination proceeds through the ligation of ubiquitin to the protein to be degraded. This posttranslational modification directs the ubiquitinilated proteins to cellular compartments or to degradation into the proteasome. The ligation of ubiquitin is reversible and could be used to regulate specific protein levels at the synapse. In mice, many proteins of the post-synaptic density, including the mouse orthologs of the ASD-associated SHANK proteins, have been demonstrated to be targeted by ubiquitination in an activity-dependent homeostatic manner 35 . Ubiquitination involves activating enzymes (E1), conjugating enzymes (E2) and ligases (E3). Substrate specificity is usually provided by the E3 ligases, which typically have substrate-binding sites. UBE3A (also called E6-AP) is an E3 ligase encoded by an imprinted gene (only expressed from the maternal copy) and is responsible for Angelman syndrome 21 . In ASD, de novo maternal duplications of chromosome 15q11-q13 including UBE3A have been observed in 1-3 % of the patients 21 . It is still not clear whether UBE3A alone contributes to the risk of ASD, since other candidate genes are also duplicated on chromosome 15q11-q13, however, its role at the synapse has been recently demonstrated in mice 36, 37 . In cultured hippocampal neurones Ube3a is localized at the pre-and post-synaptic compartments, but also at the nucleus. Experience-driven neuronal activity induces Ube3A transcription, and then, Ube3A regulates excitatory synapse development by controlling the degradation of Arc, a synaptic protein that promotes the internalization of the AMPA subtype of glutamate receptors 37 . This might have many consequences for synaptic structure, as suggested by Ube3a maternal-deficient mice, which exhibit abnormal dendritic spine development, including spine morphology, number and length 36 , and a   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 9 reduced number of AMPA receptors at excitatory synapses 37 .
Finally, the transcription factor MEF2C (Fig. 2B), involved in the regulation of the number of synapses appears to be a risk factor for intellectual disability 38 , and could therefore also be associated with ASD. Taken together, the genetic results obtained in humans and the functional studies mostly obtained in mice suggest that different independent mechanisms could alter the level of synaptic proteins; however, the actual nature of the impaired synaptic function(s), and its association with ASD phenotype remains to be characterized.

Abnormal formation of neuronal circuits in ASD
The main category of genes associated with ASD is related to the development and the function of neuronal circuits 39 . At the synaptic membranes, cell adhesion molecules, such as NLGNs and NRXNs By contrast, mutant knockin Nlgn3 and knockout Nlgn4 displayed enhanced to normal learning compared with wild-type mice 28,48 . Furthermore, in the mouse model for fragile X, an enhanced 10 Nlgn1 expression improved social behavior, whereas no effect on learning and memory was observed 49 . Finally, in the honeybee, sensory deprived animals had a lower level of Nlgn1 expression, but a generally increased level of the Nlgn2-5 and NrxnI expression compared with hive bees 50 .
The postsynaptic density plays a major role in the organization and plasticity of the synapse, and mutations affecting scaffolding proteins, such as SHANK2, SHANK3 and DLGAP2, are recurrently found in ASD 19, 51, 52 . Deletions at 22q13 and mutations of SHANK3 could be present in more than 1-2% of ASD patients (Box 1) 51-53 . Shank proteins are a family of three members, which are crucial components of the postsynaptic density. Together with their binding partners, they have been shown, in vitro, to regulate the size and shape of dendritic spines 54 . They also link glutamate receptors to the cytoskeleton and variations in genes regulating cytoskeletal dynamics were associated with mental retardation and ASD 19, 55 .
The role of neurotransmitter transporters and receptors in the susceptibility to ASD is still unclear. Because of the abnormally high levels of serotonin in ASD patients 33 , the serotonin transporter SLC6A4 was extensively analyzed, and the results pointed toward dimensional rather than categorical roles for SLC6A4 in stereotypic behaviors 56 . For glutamate, only weak associations for GRIK2 were detected 57 , and a duplication of the X-linked GRIA3 receptor gene was observed in a patient presenting typical autism 12 . Concerning GABA, the most robust findings concern the duplication of the GABA receptor subunit gene-cluster on chromosome 15q11-13 and the observation of maternal over-transmission of a rare variant of the GABA(A) receptor beta3 subunit gene (GABRB3) 58 .
Finally, proteins, related to axonal growth and synaptic identity, are now also suspected to play a role in ASD. Semaphorins are membrane or secreted proteins (Fig. 2) that influence axon outgrowth and pruning, synaptogenesis and the density and maturation of dendritic spines. SNPs located close to the semaphorin SEMA5A were associated with ASD in a large cohort 8 . Independently, the level of SEMA5A mRNA was found to be lower in brain-tissue and B-lymphoblastoid cell lines from patients with ASD compared with controls 59 . The contactin family of proteins is involved in axonal guidance as well as in the connection between axons and glial cells, and ASD patients have been found to have deletions of the contactin genes CNTN3 and CNTN4 and the contactin associated 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65   11 protein CNTNAP2 25, 60-63 . In addition, inherited CNVs or SNPs have been found in other cell-adhesion proteinscadherins (CDH9, CDH10, CDH18) and protocadherins PCDH9 and PCDH10 7, 14, 25which might contribute to the susceptibility to ASD by altering neuronal identity. If synaptic homeostasis is altered in ASD, environmental factors that influence this regulatory process could also modulate its severity. As reviewed elsewhere 33, 70 , abnormal serotonin and/or melatonin levels and altered sleep or circadian rhythms might constitute risk factors for ASD 71 . Sleep has been proposed as an important mechanism to regulate synaptic homeostasis. During wakefulness there appears to be a global increase in the strength of excitatory synapses, scaled down during sleep to a baseline level 65, 72 , a mechanism that can play an important role in learning and memory 73 . In addition to mutations of genes directly involved in synaptic processes, we have recently proposed that ,  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 12 in some cases, ASD could result from the interplay between abnormalities in synaptic and clock genes, and that restoring circadian rhythms might therefore be beneficial for the patients and their families 70 .

Abnormal synaptic homeostasis in ASD
Most of the genes considered in this review are thought to be expressed throughout the brain, however, neuroimaging studies seem to converge into a stereotypical network of brain regions where differences between ASD and control populations can be detected. These two results would not need to be in contradiction, if different brain networks were differently resilient to variations in synaptic homeostasis (Fig. 3). From an evolutionary standpoint, brain networks involved in more recently acquired cognitive skills, such as language or complex social behavior, might have less compensatory mechanisms compared with more ancient biological functions that have been shaped by a much stronger selective pressure.

Concluding remarks and perspectives
It is a matter of time for geneticists to be able to obtain whole genome sequences of ASD patients.