Recent Genetic and Functional Insights in Autism Spectrum Disorder

Abstract

Purpose of this review

Recent advances in genetic technologies allowed researchers to identify large numbers of candidate risk genes associated with autism spectrum disorder (ASD). Both strongly penetrant rare variants and the accumulation of common variants with much weaker penetrance contribute to the etiology of ASD. To identify the highly-confident candidate genes, software and resources have been applied, and functional evaluation of the variants has provided further insights for ASD pathophysiology. These studies ultimately identify the molecular and circuit alteration underlying the behavioral abnormalities in ASD. In this review, we introduce the recent genetic and genomic findings and functional approaches for ASD variants providing a deeper understanding of ASD etiology.

Recent findings

Integrated meta-analysis that recruited a larger number of ASD cases has helped to prioritize ASD candidate genes or genetic loci into a highly-confidence candidate genes for further investigation. Not only coding but also non-coding variants have been recently implicated to confer the risk of ASD. Functional approaches of genes or variants revealed the disruption of specific molecular pathways. Further studies combining ASD genetics and genomics with recent techniques in engineered mouse models show molecular and circuit mechanisms underlying the behavioral deficits in ASD.

Summary

Advances in ASD genetics and the following functional studies provide significant insights into ASD pathophysiology at molecular and circuit levels.

Introduction

Autism spectrum disorder (ASD) is one of the most common neurodevelopmental disorders (NDDs) with prevalence estimated to be 1-2.47 % of children[1,2]. According to Diagnostic and Statistical Manual (DSM) 5, it is diagnosed by two core symptoms including impaired social communication and interaction, as well as repetitive and restrictive behaviors. Although individuals with ASD share core features, ASD is a clinically heterogeneous group of disorders. They often show a complex combination of medical, neurologic, and behavioral comorbid symptoms including seizure, impaired motor skills, intellectual disability (ID), speech delay, sleep disorder, and gastrointestinal problems. These symptoms generally appear before two years of age and continue throughout life. Although the research on ASD has made great progress during the past decades, no standard treatments are available likely due to the clinical and genetic heterogeneity of ASD[3].

Both genetic and environmental factors contribute to the etiology of ASD, however, the most progress has recently been made in understanding the genetic defects underlying the disorder. Early twin and family studies revealed that ASD is strongly genetically influenced with an estimated heritability of ASD of 40-90%[4-7]. In the past decade, large-scale genomic studies have identified hundreds of genetic defects including single-nucleotide variants (SNVs) and genomic copy number variations (CNVs) that are associated with ASD[8-10]. Moreover, recent studies have highlighted both rare and common variants, and coding and noncoding genetic changes that significantly contribute to the genetic basis of ASD, consistent with a heterogeneous and complex array of etiologies in ASD pathogenesis. Subsequent translational studies based on these genetic findings have uncovered specific molecular, circuit and behavioral deficits caused by these ASD-associated genetic changes.

In this review, we introduce the recent findings of genetic studies and the functional investigation of the consequences of these ASD-associated genetic and genomic defects using cells and genetically engineered animal models.

Identification of genetic variants involved in ASD

Over the past decade, there have been remarkable advances in genetic technologies such as next generation sequencing that has allowed researchers to identify large numbers of candidate risk variants in ASD. So far both rare and common variants have been suggested to confer risk for ASD but only the functional impact of a minor subset of the rare variants have been well studied. Whole-exome sequencing (WES) and microarray-based comparative genomic hybridization studies for CNVs have characterized the substantial impact of rare variants, especially newly arising de novo variants in ASD for a number of years[8,11-15]. Although hundreds of “ASD candidate genes” have been identified, the contribution of each individual candidate gene to the ASD population is very low (< 0.5%). To identify “high-confidence candidate genes” in which deleterious mutations affecting the same gene occur in many cases of ASD, a progressively greater numbers of ASD cases and families have been recruited for such analyses. Rubeis et al., analyzed 3,871 ASD cases and ancestry-matched controls and uncovered that ASD-risk genes were enriched in FMRP-targets, synaptic genes, and genes related to the regulation of transcription or chromatin remodeling[8]. Considering the great genetic heterogeneity and diversity of functions implicated by these ASD-associated genes, still larger cohorts were needed and were recently integrated with meta-analysis to provide further insights. A recent very large meta-analysis combined the de novo SNVs from 10,927 ASD, ID and developmental delay (DD) and analyzed 12,172 de novo variants[16]. They identified recurrent mutations in 253 genes and 124 of them reached genome-wide association study (GWAS) threshold suggesting a strong contribution to ASD disease etiology. In a network level analysis, some specific biological function modules were significantly enriched including the regulation of transcription from RNA polymerase II promoters, functions relating to neurotransmitter and synaptic signaling, transmembrane receptor protein serine/threonine kinase signaling, and c-Jun N-terminal kinase. These studies have provided solid evidence to prioritize specific candidate genes and molecular pathways for further investigation of ASD pathogenesis.

So far genetic studies have mostly recruited thousands of ASD cases and controls of European American ancestry. However, more recent studies have recruited substantial numbers of ASD and control individuals of other ethnicities. In the past year, two studies recruited Japanese cohorts: 262 trios with an individual affected by ASD or 1,108 ASD cases[17,18]. These studies basically replicated the previous findings reported for ASD individuals of European descent, such as the excess of deleterious de novo variants in ASD when consider all affected genes in aggregate, with an enrichment of mutations in FMRP targets, synaptic genes and genes essential in mice suggesting a shared pathway and etiology between Caucasians and Japanese[19]. Interestingly, among identified genes, the ATP2B2 gene that was not strongly implicated in previous studies in Caucasians was enriched for de novo mutations in the Japanese ASD cohort. Since differences of race and ethnicity in ASD etiology are still poorly understood, genetic analysis of various ethnicities may provide additional unique insights into ASD pathogenic mechanisms.

Estimating the pathogenesis of variants identified in individuals

Each person harbors millions of variants, which typically include more than 10,000 peptide-sequence altering variants, and more than 100 protein-truncating variants[20]. So the identification of the disease-causing gene variant(s) in an individual’s exome or genome remains a challenging problem. Interpreting CNVs that delete or duplicate the dosage of a gene is relatively easier but estimating the functional impact of gene sequence variants is particularly important to the interpretation of SNVs. Recent studies have applied several criteria and computational approaches to enrich potentially functional or pathogenic SNVs from among the numerous variants identified in each ASD case. One of the easiest criteria is the classification of the variant type. Among all SNVs, nonsense (a mutation in which a sense codon that corresponds to one of the twenty amino acids specified by the genetic code is changed to a chain-terminating codon), frameshift, and splice site nucleotide mutations are relatively easy to interpret and are expected to result in a loss-of-function (LOF) of the target gene. These are sometimes called “likely gene-disruptive variants” and are considered to be most deleterious. In the case of peptide-sequence altering missense variants, in silico prioritization tools have been developed to help predict their functional impact (Table 1). Although many tools have been developed, the SIFT, PolyPhen2 and Combined Annotation-Dependent Depletion (CADD) are the ones most frequently used in ASD studies[21-23]. The characteristics of recent commonly used software were well reviewed by Eilbeck et al.[24]. More recently, gene constraint metrics such as Residual Variation Intolerant Score (RVIS) and probability of LOF intolerance (pLI) are used to enrich ASD risk genes[25,26]. They were developed based on the recent population-scale variant databases such as ESP6500 and ExAC, and statistically model the tolerance of a gene to amino-acid change or LOF variations[27,28]. Since recent studies have suggested ASD-risk genes are less intolerant of mutations, this has become a favored approach for screening ASD candidate genes. Although it has been noted that the applications of these tools have missed some ASD candidate genes, these methods remain useful in setting a threshold for significance when identifying those genes to prioritize for further investigations[16].

Estimating the likely pathogenic variants associated with ASD

Criteria	Explanation	Interpretation
A. Allele frequency in healthy subjects
1000 genomes project	Database of 2,504 genomes sequenced from healthy subjects	Rare variants are more likely to have larger impact or pathogenic effects compared to common ones. We should take note that the criteria of healthy is varies among databases
Exome sequencing project 6500	Database of 6,503 exomes sequenced from healthy subjects
ExAc	Datasets of 60,706 exomes sequenced from unrelated healthy subjects
GnomAD	Datasets of 125,748 WES and 15,708 WGS from unrelated healthy subjects are available
B. Inheritance pattern
De novo	Newly arising mutations in patients.	De novo variants are more likely to be penetrant compared to inherited ones. The impact of maternally-inherited variants could be underestimated because of the female protective effect in ASD.
Inherited	Mutations inherited from father or mother to patients
C. Types of variants
Indel	Small insertions or deletions of bases	Nonsense, stoploss, splicing site mutations and indels are most likely to impact protein function. On the other hand, only subset of missense variants will impact protein function. Synonymous mutations do not alter amino acid sequence or protein function.
Nonsense	Mutations causing protein-truncation
Stoploss	Mutations disrupting the stop codon resulting in abnormal extention of proteins
Missense	Mutations causing a change to the amino acid
Splicing site	Mutations affecting the splicing sites possibly causing mis-splicing
Synonymous	Mutation which don't alter the amino acid sequence
D. Genetic intorelance
pLi	A gene score of the probability of loss-of-function intolerance determined by the number of observed variants and that of expected variants.	Mutations in intorerant genes are more likely to be deleterious
RVIS	A gene intolerance score determined by the number of observed nonsynonymous variants and that of synonymous variants
E. in silico tools to predict the impact of SNVs
SIFT	A prediction tool of the SNV impact based on the evolutional conservation of the protein's amino acid sequence	These tools score human variants and are usuful to estimate how deleterious a given variants will be to protein function. All of them can be applied to predict the impact of variants with amino-acid substitutions. CADD can be also used for indels.
PolyPhen2	A prediction tool of SNV impact based on the protein sequence and structure.
CADD	A prediction tool of the impact of SNVs and short indels. It is an integrative metric built from diverse genetic features such as evolutionary constraint, epigenetic status and the score of other prediction tools including SIFT and PolyPhen2.

Functional testing of ASD-associated variants by using cells and animals

In parallel with genetic approaches, functional evaluation of ASD-associated variants by in vitro and in vivo assays is crucial to facilitating an understanding of ASD pathophysiology and for the future rational design of therapeutic strategies. This functional testing approach provides insights that cannot be uncovered by genetic studies alone. We explain how functional studies are important in two types of ASD risk genes.

i). ASD risk genes associated with LOF

So far over 1000 genes with alterations in ASD have been identified in recent genetic studies and these are suggested to be high-confidence candidate genes based on the recurrent finding of gene-disruptive variants. This strategy successfully identifies genes using LOF in associated with disease. Over the past several years, strong ASD candidate genes with LOF mutations have been characterized in cell and model animal experimental systems. Chromodomain helicase DNA binding protein 8 (CHD8) is one of the genes most frequently mutated in ASD[8,29]. Recent mouse studies have identified synaptic, transcriptional, and behavioral abnormalities caused by CHD8 mutations and moreover the study of Chd8 point mutant mice suggests these abnormalities could be sexually dimorphic[30-32]. Haploinsufficiency of other strong candidate genes associated with ASD and ID, CHD2 and SETD2 have been recently reported to cause behavioral and synaptic abnormalities in the mutant mice[33-35]. Characterizing these genes especially using model animals provides insights in molecular, circuit, and behavioral abnormalities related with ASD that may lead to the future development of treatments for these specific genetic subtypes of ASD.

ii). ASD risk genes associated with gain-of-function (GOF) or specific pathways

Genetic studies have identified subsets of ASD genes associated with LOF, but this approach would miss genes containing possible gain-of-function (GOF) mutations. Most in silico tools are designed to predict nucleotide changes that give rise to a gene LOF; computational approaches that predict GOF variants remains difficult. Therefore, genes with GOF variants or variants causing unpredictable functional alterations tend to be undervalued despite the large number of missense variants that have been identified in individuals with ASD. For example, large numbers of missense but not gene-disrupting variants have been identified in CACNA1D encoding the voltage-gated L-type calcium channel in ASD cases. Because of this LOF bias, limited numbers of human genetic studies have supported its potential role in ASD. However, functional studies of the effects of these ASD-associated missense variants in CACNA1D have revealed that they result in a GOF increase in activity of the channel. In such cases, functional studies are critical to identifying these ASD candidate genes[36,37].

Functional assessment of ASD-associated variants can sometimes uncover unexpected molecular pathways involved in ASD pathogenesis. Recently an ASD-associated missense variant in SHANK3, a well-established ASD-candidate gene, was reported to disrupt a novel molecular pathway supported by this gene product. Researchers found that an ASD-associated S685I mutation in SHANK3 specifically diminishes Shank3-ABI1 interactions, which turned out to be critical for dendritic spine development and synaptic transmission. Moreover, behavioral assay of the knock-in mouse carrying this mutation caused ASD-associated behaviors[38]. So far several potential therapeutic strategies targeting the other SHANK3 pathways have been proposed[39,40]. However, the finding of Shank3 S685I indicates that the therapy aimed at correcting a specific Shank3-associated pathway may not be equally applicable to all ASD patients carrying pathogenic SHANK3 mutations. A recent study identifies that ASD-associated NLGN1 missense variants unexpectedly affected several distinct processes (e.g. protein misfolding and increased cleavage of extracellular domain)[41]. These findings indicate the complex functional outcome caused by different ASD-associated gene variants and the necessity of evaluating them with biological assays.

Functional analyses of ASD-related CNVs

Not only SNVs but also CNVs contribute to the pathogenesis of ASD with high penetrance. There have been substantial functional studies using mouse models of CNVs since the first CNV model of ASD was developed[10,42]. A recent study shows the significance of serotonin (5-HT) during a developmental stage in 15q11-13 duplication (15q dup) mice[43]. The 5-HT level in the brain of 15q dup mice is decreased[44] and 5-HT neural activity in the dorsal raphe nucleus (DRN) of 15q dup mice is also decreased[43]. These phenotypes may impair neocortical excitation/inhibition balance, correct sensory stimulus tuning and social behavior. Furthermore, restoration of normal 5-HT levels in 15q dup mice reveals the reversibility of ASD-related symptoms in the adult. Decreased DR 5-HT activity during social contact and reduced DR 5-HT neuron activity are also observed in 16p11.2 deletion mice[45]. The decrease in sociability in 16p11.2 deletion mice is rescued by activation of activation of DR 5-HT neurons and pharmacological activation of nucleus accumbens 5-HT1b receptors.

ASD genetics combined with genome engineering and AAV viral vectors in model organisms to resolve the circuit basis of behavioral problems in ASD as above. A recent study mapped the neuronal circuit deficits underlying impaired sociability produced by the increased dosages of the UBE3A gene found in a strongly penetrant CNV in ASD, maternal 15q11-13 triplication [extranumerary isodicentric chromosome 15q, idic(15)][46]. Glutamatergic synaptic transmission from ventral tegmental area glutamatergic neurons was impaired by the interaction of increased UBE3A and seizures [a frequent comorbidity in idic(15) and in human idiopathic ASD] because these repress expression of a gene encoding the synapse organizing protein CBLN1 that physically binds presynaptic NRXN1 and postsynaptic GluD1. Both NRXN1 and GRID1 (gene for GluD1) are frequently deleted genes in the ASD CNVs[47]. Consistent with the findings of the Krishnan et al.’ study[46], a recent study also finds the role for UBE3A in the nucleus and shows a function of UBE3A as a transcriptional regulator of the innate immune system in the brain[48].

Contributions of common variants

Although estimates vary across studies, common genetic variations, both coding and noncoding, are thought to account for approximately 20-60% of ASD liability[49-52]. In contrast, de novo, extremely rare SNV or CNV can have a larger effect but explain only <10% of overall liability[51]. Recent studies assessing both common and rare variants simultaneously suggest that the accumulation of both types of variants in an individual may have an etiologic role. It was previously thought that a single deleterious de novo variant in an individual may be sufficiently penetrant to fully explain the disease, and that common variants were significant only in cases without a strong acting variant. However, two recent large-scale studies suggested the significant contribution of common variants even in ASD cases with a known penetrant deleterious variant. Weiner et al., analyzed 6,454 ASD families and uncovered common polygenic variation still contributes to risk in ASD cases carrying a very deleterious de novo variant[53]. Furthermore, Niemi et al. examined more extreme cases[54]. They examined 6,987 cases of very severe NDD including ASD with morphological and/or physiological abnormalities in the central nervous system (CNS). Even in the extreme cases in whom monogenic causes were strongly suspected, they found part of the disease risk could be attributed to common variations. These studies show the effects of common variation are not negligible in most of ASD and severe NDD cases with or without highly penetrant rare mutations and highlight the complex genetic basis of ASD and NDD.

Recent studies of common variations, especially GWAS studies, report the genetic overlap between ASD and other psychiatric or NDDs. According to the most recent, largest GWAS study, the polygenic risk for ASD had significant overlap with schizophrenia, major depression and ADHD. Interestingly, common polygenic risk for ASD has been repeatedly suggested to have a positive correlation with educational attainment and IQ[55-59]. Considering that rare deleterious variants have the opposite associated, showing a lower IQ and more ID, the way by which rare and common variants confer the risk for ASD likely differs.

The contribution of non-coding variants in ASD

Recently whole-genome sequencing (WGS) is increasingly being used in ASD studies instead of WES mainly because of the falling cost of sequencing. Unlike WES that sequences only the protein-coding regions, WGS reads the entire genome enabling the study of noncoding genome sequences as well. The noncoding genome occupies ~98.5% of the genome, and contain the functional transcriptional regulatory elements that decides when and where or in which cell types a gene is expressed. In addition, human-specific regions (human accelerated regions) contained within the noncoding genome regions might be linked to human-specific traits and their disruption might be linked to neurological or cognitive dysfunction[60]. Researchers have found some evidences that noncoding sequence variations may account for ASD although the evidence is still weak compared to the evidence for the coding variants[61-65]. Brandley et al. focused the rare structural variants (SVs; e.g. deletions, duplications, insertions and inversions) affecting highly-intolerant cis-regulatory elements[66]. They found some recurrent rare noncoding SVs in ASD cases, such as Leo1 promoter disrupting variants. In addition, they show intolerant SVs affecting cis-regulatory elements (e.g. transcription start sites, fetal brain promoters and 3’UTRs) were over-transmitted from father to the ASD but not the control sibling. This observation may indicate that these SVs confer risk for ASD. Furthermore, the recent largest WGS of 1902 ASD quartet families provided significant insights in the contribution of de novo noncoding SNVs and small indels[67]. First, they assigned annotation categories to de novo noncoding variants and found no noncoding category was significantly associated with ASD after correction for multiple testing. However, further analysis using de novo risk score developed by machine learning detected a significant contribution of noncoding to ASD risk. In particular, noncoding variants in evolutionary conserved distal promoter regions showed the most robust signal and similar results were observed in a previous WGS study using a different analytic approach [62]. Overall, these studies show a weak but significant contribution of noncoding variants to ASD risk and future analysis of larger cohorts and improving the resources available to annotate of noncoding variants will provide further insights.

Conclusion

To elucidate the etiology of ASD, comprehensive analyses from genetic to translational studies are essential. Recent advances in genetics have provided increasing insights into the complex genetic basis of ASD. The evidence suggests ASD can be caused by genetic defects that include the following: 1) genomic segment CNVs that include micro-deletions, micro-duplications and even higher increases of genomic segment dosage that can involve multiple genes but often highlights a specific gene when assessing the overlap across many ASD cases; 2) strongly penetrance gene coding sequence SNV mutations that cause either a heterozygous (for steeply dose-sensitive genes) or homozygous LOF and in others cases heterozygous GOF mutations; 3) common variants that modify these penetrant CNV and SNV genetic changes; and 4) the possibility of polygenic mechanisms becoming fully penetrant only when two or more genetic changes occur in molecules in a common molecular pathway. The identification of non-coding variants in ASD is just beginning to emerge and the current evidence suggests these may interfere with conserved gene regulatory elements. Analysis of larger number of ASD cases with integrated meta-analysis has helped to prioritize ASD candidate genes or genetic loci into a highly-confident candidate set for further investigation. The identification of potential ASD candidate genes by genetic and computational approaches is still not perfect, but the development of new methods and online resources will continue to increase success in ASD genetics analytics. An important approach to obtaining further evidence that a gene defect found in ASD has an etiologic role in the disorder is the development of functional methods of validating each variant, for example, the highly efficient techniques of genome editing now possible using CRISPR/Cas9 is sure to accelerate the pace of discovery. Ultimately, these genetic defects can then be used to map the specific neuronal circuit defects that underlie behavioral deficits in ASD, providing a deeper understanding of ASD pathophysiology and pointing the way to candidate targets for therapeutic intervention.

Key points.

• Recent meta-analysis recruiting the largest number of ASD subjects or non-Caucasians identifies novel or promising candidate genes associated with ASD.

• Recent WGS studies suggest the significant contribution of not only coding but also non-coding variants in ASD.

• Functional characterization of GOF and LOF variants associated with ASD leads to a deeper understanding of the pathogenesis of ASD.

• Studies combining mouse models recapitulating genetic features in patients with ASD and recent genomic techniques show the specific defects of neuronal circuits underlying behavioral deficits in ASD and provide the potential therapeutic targets.