Environmental factors

Twin studies suggest that 9–36% of the variance in autism predisposition might be explained by environmental factors (Tick, Bolton, Happé, Rutter, & Rijsdijk, Reference Tick, Bolton, Happé, Rutter and Rijsdijk2016). There is observational evidence for association with pre- and perinatal factors such as parental age, asphyxia-related birth complications, preterm birth, maternal obesity, gestational diabetes, short inter-pregnancy interval, and valproate use (Lyall et al., Reference Lyall, Croen, Daniels, Fallin, Ladd-Acosta, Lee and Newschaffer2017; Modabbernia, Velthorst, & Reichenberg, Reference Modabbernia, Velthorst and Reichenberg2017). Mixed results are reported for pregnancy-related nutritional factors and exposure to heavy metals, air pollution, and pesticides, while there is strong evidence that autism risk is unrelated to vaccination, maternal smoking, or thimerosal exposure (Modabbernia et al., Reference Modabbernia, Velthorst and Reichenberg2017). It is challenging to infer causality from observed associations, given that confounding by lifestyle, socioeconomic, or genetic factors contributes to non-causal associations between exposures and autism. Many putative exposures are associated with parental genotype (e.g. obesity, age at birth) (Gratten et al., Reference Gratten, Wray, Peyrot, McGrath, Visscher and Goddard2016; Taylor et al., Reference Taylor, Debost, Morton, Wigdor, Heyne, Lal and Robinson2019a, Yengo et al., Reference Yengo, Sidorenko, Kemper, Zheng, Wood, Weedon and Visscher2018), and some are associated both with maternal and fetal genotypes (e.g. preterm birth) (Zhang et al., Reference Zhang, Feenstra, Bacelis, Liu, Muglia, Juodakis and Muglia2017). Studies triangulating genetically informative designs are needed to disentangle these relationships (Davies et al., Reference Davies, Howe, Brumpton, Havdahl, Evans and Davey Smith2019; Leppert et al., Reference Leppert, Havdahl, Riglin, Jones, Zheng, Davey Smith and Stergiakouli2019; Thapar & Rutter, Reference Thapar and Rutter2019).

Linkage and candidate gene studies

Initial linkage studies were conducted to identify chromosomal regions commonly inherited in affected individuals. Susceptibility loci implicated a range of regions, but only two have been replicated (Ramaswami & Geschwind, Reference Ramaswami and Geschwind2018): at chromosome 20p13 (Weiss, Arking, Daly, & Chakravarti, Reference Weiss, Arking, Daly and Chakravarti2009) and chromosome 7q35 (Alarcón, Cantor, Liu, Gilliam, & Geschwind, Reference Alarcón, Cantor, Liu, Gilliam and Geschwind2002). Lack of replication and inconsistent findings were largely due to low statistical power (Kim & Leventhal, Reference Kim and Leventhal2015). Candidate gene association studies identified over 100 positional and/or functional candidate genes for associations with autism (Bacchelli & Maestrini, Reference Bacchelli and Maestrini2006). However, there was no consistent replication for any of these findings (Warrier, Chee, Smith, Chakrabarti, & Baron-Cohen, Reference Warrier, Chee, Smith, Chakrabarti and Baron-Cohen2015), likely due to limitations in study design (e.g. low statistical power, population diversity, incomplete coverage of variation within the candidate genes, and false positives arising from publication bias) (Ioannidis, Reference Ioannidis2005; Ioannidis, Ntzani, Trikalinos, & Contopoulos-Ioannidis, Reference Ioannidis, Ntzani, Trikalinos and Contopoulos-Ioannidis2001). The advancement of genome-wide association studies (GWAS) and next-generation sequencing techniques has significantly enhanced gene and variant discovery.

Common genetic variation

The SNP-heritability (proportion of variance attributed to the additive effects of common genetic variants) of autism ranges from 65% in multiplex families (Klei et al., Reference Klei, Sanders, Murtha, Hus, Lowe, Willsey and Devlin2012) to 12% in the latest Psychiatric Genomics Consortium GWAS (Fig. 2a) (Autism Spectrum Disorders Working Group of The Psychiatric Genomics Consortium, 2017; Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019). Variation is largely attributable to sample heterogeneity and differences in methods used to estimate SNP-heritability.

Fig. 2. Variance explained by different classes of genetic variants in autism. (a) Donut chart of the variance explained by different classes of variants. The narrow-sense heritability (82.7%, Nordic average, shades of green) has been estimated using familial recurrence data from Bai et al. (Reference Bai, Yip, Windham, Sourander, Francis, Yoffe and Sandin2019). The total common inherited heritability (12%) has been estimated using LDSC-based SNP-heritability (additive) from Grove et al. (Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) and the total rare inherited heritability (3%) has been obtained from Gaugler et al. (Reference Gaugler, Klei, Sanders, Bodea, Goldberg, Lee and Buxbaum2014). The currently unexplained additive heritability is thus 67.7% (total narrow-sense heritability minus common and rare inherited heritabilities combined). This leaves a total of 17.3% of the variance to shared and unique environmental estimates (Bai et al., Reference Bai, Yip, Windham, Sourander, Francis, Yoffe and Sandin2019). The term environmental refers to non-additive and non-inherited factors that contribute to variation in autism liability. Of this, de novo missense and protein-truncating variants (Satterstrom et al., Reference Satterstrom, Kosmicki, Wang, Breen, De Rubeis, An and Walters2020) and variation in non-genic regions (An et al., Reference An, Lin, Zhu, Werling, Dong, Brand and Sanders2018) together explain 2.5% of the variance. Whilst de novo variation can be inherited in some cases (germline mutation in the parent) and thus shared between siblings, it is unlikely that this will be shared by other related individuals, and thus unlikely to be included in the narrow-sense heritability in Bai et al. (Reference Bai, Yip, Windham, Sourander, Francis, Yoffe and Sandin2019). This is likely to be a lower-bound of the estimate as we have not included the variance explained by de novo structural variants and tandem repeats. Additionally, non-additive variation accounts for ~4% of the total variance (Autism Sequencing Consortium et al., Reference Doan, Lim, De Rubeis, Betancur, Cutler, Chiocchetti and Yu2019). Thus, ~11% of the total variance is currently unaccounted for, though this is likely to be an upper bound. (b) The variance explained is likely to change in phenotypic subgroups. For instance, the risk ratio for de novo protein-truncating variants in highly constrained genes (pLI > 0.9) is higher in autistic individuals with ID compared to those without ID (point estimates and 95% confidence intervals provided; Kosmicki et al., Reference Kosmicki, Samocha, Howrigan, Sanders, Slowikowski, Lek and Daly2017). (c) Similarly, the proportion of the additive variance explained by common genetic variants is higher in autistic individuals without ID compared to autistic individuals with ID (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019). Point estimates and 95% confidence intervals provided.

Early GWASs of autism were underpowered, partly due to overestimating potential effect sizes. Grove et al. (Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) conducted a large GWAS of autism combining data from over 18 000 autistic individuals and 27 000 non-autistic controls and an additional replication sample. They identified five independent GWAS loci (Fig. 3). Another recent study (Matoba et al., Reference Matoba, Liang, Sun, Aygün, McAfee, Davis and Stein2020) identified a further novel locus by meta-analyzing the results from Grove et al. (Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) with over 6000 case-pseudocontrol pairs from the SPARK cohort by employing a massively parallel reporter assay to identify a potential causal variant (rs7001340) at this locus which regulates DDH2 in the fetal brain. The sample sizes are still relatively small compared to other psychiatric conditions (Schizophrenia Working Group of the Psychiatric Genomics Consortium, Reference Ripke, Walters and O'Donovan2020; Howard et al., Reference Howard, Adams, Clarke, Hafferty, Gibson, Shirali and McIntosh2019), though ongoing work aims to double the sample size and identify additional loci.

Fig. 3. Karyogram showing the 102 genes implicated by rare variant findings at a false discovery rate of 0.1 or less (Satterstrom et al., Reference Satterstrom, Kosmicki, Wang, Breen, De Rubeis, An and Walters2020) and the five index SNPs identified in GWAS (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) of autism.

Using genetic correlations and polygenic score analyses, studies have identified modest shared genetics between autism and different definitions of autistic traits in the general population (Askeland et al., Reference Askeland, Hannigan, Ask, Ayorech, Tesli, Corfield and Havdahl2020; Bralten et al., Reference Bralten, van Hulzen, Martens, Galesloot, Arias Vasquez, Kiemeney and Poelmans2018; Robinson et al., Reference Robinson, St Pourcain, Anttila, Kosmicki, Bulik-Sullivan, Grove and Daly2016; Taylor et al., Reference Taylor, Martin, Lu, Brikell, Lundström, Larsson and Lichtenstein2019b). There is some evidence for developmental effects, with greater shared genetics in childhood compared to adolescence (St Pourcain et al., Reference St Pourcain, Robinson, Anttila, Sullivan, Maller, Golding and Davey Smith2018). These methods have also identified modest polygenic associations between autism and other neurodevelopmental and mental conditions such as schizophrenia, ADHD, and major depressive disorder, related traits such as age of walking, language delays, neuroticism, tiredness, and self-harm, as well as risk of exposure to childhood maltreatment and other stressful life events (Brainstorm Consortium et al., Reference Anttila, Bulik-Sullivan, Finucane, Walters, Bras, Duncan and Murray2018; Bulik-Sullivan et al., Reference Bulik-Sullivan, Finucane, Anttila, Gusev, Day, Loh and Neale2015; Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019; Hannigan et al., Reference Hannigan, Askeland, Ask, Tesli, Corfield, Magnus and Box2020; Lee et al., Reference Lee, Anttila, Won, Feng, Rosenthal, Zhu and Smoller2019Reference Lee, Anttila, Won, Feng, Rosenthal, Zhu and Smollerb; Leppert et al., Reference Leppert, Havdahl, Riglin, Jones, Zheng, Davey Smith and Stergiakouli2019; Cross-Disorder Group of the Psychiatric Genomics Consortium, Reference Lee, Ripke, Neale, Faraone, Purcell and Perlis2013; Warrier & Baron-Cohen, Reference Warrier and Baron-Cohen2019). Notably, autism is positively genetically correlated with measures of intelligence and educational attainment (EA) (Bulik-Sullivan et al., Reference Bulik-Sullivan, Finucane, Anttila, Gusev, Day, Loh and Neale2015; Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019), an observation supported by polygenic score association (Clarke et al., Reference Clarke, Lupton, Fernandez-Pujals, Starr, Davies, Cox and McIntosh2016). Polygenic Transmission Disequilibrium Tests have identified an over-transmission of polygenic scores for EA, schizophrenia, and self-harm from parents to autistic children, but an absence of such over-transmission to non-autistic siblings (Warrier & Baron-Cohen, Reference Warrier and Baron-Cohen2019; Weiner et al., Reference Weiner, Wigdor, Ripke, Walters, Kosmicki, Grove and Robinson2017), suggesting that these genetic correlations are not explained by ascertainment biases or population stratification. However, a genetic correlation does not necessarily imply a causal relationship between the two phenotypes and may simply index biological pleiotropy. Causal inference methods such as Mendelian randomization can be used to disentangle such relationships (Davies et al., Reference Davies, Howe, Brumpton, Havdahl, Evans and Davey Smith2019; Pingault et al., Reference Pingault, O'Reilly, Schoeler, Ploubidis, Rijsdijk and Dudbridge2018).

The relatively low SNP-heritability in autism compared to other psychiatric conditions may partly be due to phenotypic heterogeneity. In an attempt to reduce phenotypic heterogeneity, Chaste et al. (Reference Chaste, Klei, Sanders, Hus, Murtha, Lowe and Devlin2015) identified 10 phenotypic combinations to subgroup autistic individuals. Family-based association analyses did not identify significant loci, and SNP-heritability for the subgroups was negligent. It is unclear if reducing phenotypic heterogeneity increases genetic homogeneity, and investigating this in larger samples is warranted. Another study identified no robust evidence of genetic correlation between social and non-social (restricted and repetitive behavior patterns) autistic traits (Warrier et al., Reference Warrier, Toro, Won, Leblond, Cliquet, Delorme and Baron-Cohen2019). A few studies have investigated the common variant genetic architecture of social and non-social autistic traits in individuals with autism (Alarcón et al., Reference Alarcón, Cantor, Liu, Gilliam and Geschwind2002; Cannon et al., Reference Cannon, Miller, Robison, Villalobos, Wahmhoff, Allen-Brady and Coon2010; Cantor et al., Reference Cantor, Navarro, Won, Walker, Lowe and Geschwind2018; Lowe, Werling, Constantino, Cantor, & Geschwind, Reference Lowe, Werling, Constantino, Cantor and Geschwind2015; Tao et al., Reference Tao, Gao, Ackerman, Guo, Saffen and Shugart2016; Yousaf et al., Reference Yousaf, Waltes, Haslinger, Klauck, Duketis, Sachse and Chiocchetti2020) and in the general population (St Pourcain et al., Reference St Pourcain, Skuse, Mandy, Wang, Hakonarson, Timpson and Smith2014; Warrier et al., Reference Warrier, Toro, Chakrabarti, Børglum, Grove, Hinds and Baron-Cohen2018, Reference Warrier, Toro, Won, Leblond, Cliquet, Delorme and Baron-Cohen2019), but replication of the identified loci is needed.

Diagnostic classification is another source of heterogeneity: SNP-heritability of Asperger's syndrome (ICD-10 diagnosis) was twice (0.097 ± 0.001) that of childhood autism and unspecified pervasive developmental disorders (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) [due to overlap in subtype diagnoses, a hierarchy was used: childhood autism>atypical autism>Asperger's syndrome>unspecified subtypes (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019)]. Supporting this, polygenic scores for intelligence and EA had larger loadings in the Asperger's syndrome and childhood autism subgroups compared to other subgroups (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019). Additionally, the SNP-heritability of autism (all subtypes) without co-occurring ID diagnosis (0.09 ± 0.005) was three times that of autism with ID (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) (Fig. 2c).

Rare genetic variation

Rare genetic variants confer significant risk in the complex etiology of autism. They are typically non-Mendelian, with substantial effect sizes and low population attributable risk. It is estimated that ~10% of autistic individuals have been diagnosed with an identifiable rare genetic syndrome characterized by dysmorphia, metabolic, and/or neurologic features (Carter & Scherer, Reference Carter and Scherer2013; Tammimies et al., Reference Tammimies, Marshall, Walker, Kaur, Thiruvahindrapuram, Lionel and Fernandez2015). Associated syndromes include the 15q11-q13 duplication of the Prader-Willi/Angelman syndrome, fragile X syndrome, 16p11.2 deletion syndrome, and 22q11 deletion syndrome (Sztainberg & Zoghbi, Reference Sztainberg and Zoghbi2016). Prevalence estimates for autism vary widely between genetic syndromes; for example, 11% in 22q11.2 deletion syndrome and 54% in Cohen's syndrome (Richards, Jones, Groves, Moss, & Oliver, Reference Richards, Jones, Groves, Moss and Oliver2015). Of note, estimating the prevalence of autism in the context of genetic syndromes is complex (Havdahl et al., Reference Havdahl, Bal, Huerta, Pickles, Øyen, Stoltenberg and Bishop2016; Richards et al., Reference Richards, Jones, Groves, Moss and Oliver2015).

The rate of gene discovery in autism is a linear function of increasing sample size (De Rubeis et al., Reference De Rubeis, He, Goldberg, Poultney, Samocha, Ercument Cicek and Buxbaum2014). Early studies implicated nine genes in the first 1000 autism cases (Neale et al., Reference Neale, Kou, Liu, Ma'ayan, Samocha, Sabo and Daly2012; Sanders et al., Reference Sanders, Murtha, Gupta, Murdoch, Raubeson, Willsey and State2012), increasing to 27 and 33 associated genes from separate analyses of Simons Simplex Collection and Autism Sequencing Consortium (ASC) samples (De Rubeis et al., Reference De Rubeis, He, Goldberg, Poultney, Samocha, Ercument Cicek and Buxbaum2014; Iossifov et al., Reference Iossifov, O'Roak, Sanders, Ronemus, Krumm, Levy and Wigler2014). Integrating these samples using the TADA framework implicated a total of 65 autism genes (Sanders et al., Reference Sanders, He, Willsey, Ercan-Sencicek, Samocha, Cicek and State2015).

The MSSNG initiative analyzed whole genomes from 5205 individuals (N _cases = 2636), and identified 61 autism-risk genes, of which 18 were new candidates (Yuen et al., Reference Yuen, Merico, Bookman, Howe, Thiruvahindrapuram, Patel and Scherer2017). More recently, the largest whole-exome sequencing analysis to date conducted by the ASC (N = 35 584, N _cases = 11 986) identified 102 autism-associated genes (Fig. 3), many of which are expressed during brain development with roles in the regulation of gene expression and neuronal communication (Satterstrom et al., Reference Satterstrom, Kosmicki, Wang, Breen, De Rubeis, An and Walters2020). Rare CNVs and SNVs associated with autism have pleiotropic effects, thus increasing the risk for other complex disorders such as schizophrenia, ADHD, ID, and epilepsy (Gudmundsson et al., Reference Gudmundsson, Walters, Ingason, Johansson, Zayats, Athanasiu and Stefansson2019; Satterstrom et al., Reference Satterstrom, Walters, Singh, Wigdor, Lescai, Demontis and Daly2019, Reference Satterstrom, Kosmicki, Wang, Breen, De Rubeis, An and Walters2020).

Epigenetics

DNA methylation (DNAm), an epigenetic modification, allows for both genetic and environmental factors to modulate a phenotype (Martin & Fry, Reference Martin and Fry2018; Smith et al., Reference Smith, Kilaru, Kocak, Almli, Mercer, Ressler and Conneely2014). DNAm affects gene expression, regulatory elements, chromatin structure, and alters neuronal development, functioning, as well as survival (Kundaje et al., Reference Kundaje, Meuleman, Ernst, Bilenky, Yen, Heravi-Moussavi and Kellis2015; Lou et al., Reference Lou, Lee, Qin, Li, Gao, Liu and Yip2014; Peters et al., Reference Peters, Joehanes, Pilling, Schurmann, Conneely, Powell and Johnson2015; Sharma, Klein, Barboza, Lohdi, & Toth, Reference Sharma, Klein, Barboza, Lohdi and Toth2016; Yu et al., Reference Yu, Furukawa, Kobayashi, Shikishima, Cha, Sese and Toda2012; Zlatanova, Stancheva, & Caiafa, Reference Zlatanova, Stancheva and Caiafa2004). Additionally, putative prenatal environmental risk factors impact the offspring's methylomic landscape (Anderson, Gillespie, Thiele, Ralph, & Ohm, Reference Anderson, Gillespie, Thiele, Ralph and Ohm2018; Cardenas et al., Reference Cardenas, Gagné-Ouellet, Allard, Brisson, Perron, Bouchard and Hivert2018; Joubert et al., Reference Joubert, den Dekker, Felix, Bohlin, Ligthart, Beckett and London2016), thus providing a plausible molecular mechanism to modulate the neurodevelopmental origins of autism.

Autism Epigenome-Wide Association Study (EWAS) meta-analysis performed in blood from children and adolescents from SEED and SSC cohorts (N _cases = 796, N _controls = 858) identified seven differentially methylated positions (DMPs) associated (p < 10 × 10⁻⁰⁵) with autism, five of them also reported to have brain-based autism associations. The associated DMPs annotated to CENPM, FENDRR, SNRNP200, PGLYRP4, EZH1, DIO3, and CCDC181 genes, with the last site having the largest effect size and the same direction of association with autism across the prefrontal cortex, temporal cortex, and cerebellum (Andrews et al., Reference Andrews, Sheppard, Windham, Schieve, Schendel, Croen and Ladd-Acosta2018). The study reported moderate enrichment of methylation Quantitative Trait Loci (mQTLs) among the associated findings, suggesting top autism DMPs to be under genetic control (Andrews et al., Reference Andrews, Sheppard, Windham, Schieve, Schendel, Croen and Ladd-Acosta2018). These findings were further extended by the MINERvA cohort that added 1263 neonatal blood samples to the meta-analysis. The SEED-SSC-MINERvA meta-EWAS identified 45 DMPs, with the top finding showing the consistent direction of association across all three studies annotated to ITLN1 (Hannon et al., Reference Hannon, Schendel, Ladd-Acosta, Grove, Hansen, Andrews and Mill2018). The MINERvA sample was also used for EWAS of autism polygenic score, hypothesizing that the polygenic score-associated DNAm variation is less affected by environmental risk factors, which can confound case–control EWAS. Elevated autism polygenic score was associated with two DMPs (p < 10 × 10⁻⁰⁶), annotated to FAM167A/C8orf12 and RP1L1. Further Bayesian co-localization of mQTL results with autism GWAS findings provided evidence that several SNPs on chromosome 20 are associated both with autism risk and DNAm changes in sites annotated to KIZ, XRN2, and NKX2-4 (Hannon et al., Reference Hannon, Schendel, Ladd-Acosta, Grove, Hansen, Andrews and Mill2018). The mQTL effect of autism risk SNPs was corroborated by an independent study not only in blood, but also in fetal and adult brain tissues, providing additional evidence that autism risk variants can act through DNAm to mediate the risk of the condition (Hammerschlag, Byrne, Bartels, Wray, & Middeldorp, Reference Hammerschlag, Byrne, Bartels, Wray and Middeldorp2020).

Since autism risk variants impact an individual's methylomic landscape, studies that investigate DNAm in the carriers of autism risk variants are of interest to provide insight into their epigenetic profiles. A small blood EWAS performed in 52 cases of autism of heterogeneous etiology, nine carriers of 16p11.2del, seven carriers of pathogenic variants in CHD8, and matched controls found that DNAm patterns did not clearly distinguish autism of the heterogeneous etiology from controls. However, the homogeneous genetically-defined 16p11.2del and CHD8^+/− subgroups were characterized by unique DNAm signatures enriched in biological pathways related to the regulation of central nervous system development, inhibition of postsynaptic membrane potential, and immune system (Siu et al., Reference Siu, Butcher, Turinsky, Cytrynbaum, Stavropoulos, Walker and Weksberg2019). This finding highlights the need to combine genomic and epigenomic information for a better understanding of the molecular pathophysiology of autism.

It must be noted that a very careful interpretation of findings from peripheral tissues is warranted. DNAm is tissue-specific and therefore EWAS findings obtained from peripheral tissues may not reflect biological processes in the brain. Using the mQTL analytical approach may reduce this challenge, as mQTLs are consistently detected across tissues, developmental stages, and populations (Smith et al., Reference Smith, Kilaru, Kocak, Almli, Mercer, Ressler and Conneely2014). However, not all mQTLs will be detected across tissues and will not necessarily have the same direction of effect (Smith et al., Reference Smith, Kilaru, Kocak, Almli, Mercer, Ressler and Conneely2014). Therefore, it is recommended that all epigenetic findings from peripheral tissues are subjected to replication analyses in human brain samples, additional experimental approaches, and/or Mendelian randomization to strengthen causal inference and explore molecular mediation by DNAm (Walton, Relton, & Caramaschi, Reference Walton, Relton and Caramaschi2019).

EWASs performed in post-mortem brains have typically been conducted using very small sample sizes, due to limited access to brain tissue (Ladd-Acosta et al., Reference Ladd-Acosta, Hansen, Briem, Fallin, Kaufmann and Feinberg2014; Nardone et al., Reference Nardone, Sams, Reuveni, Getselter, Oron, Karpuj and Elliott2014). One of the largest autism EWAS performed in post-mortem brains (43 cases and 38 controls) identified multiple DMPs (p < 5 × 10⁻⁰⁵) associated with autism (31 DMPs in the prefrontal cortex, 52 in the temporal cortex, and two in the cerebellum) (Wong et al., Reference Wong, Smith, Hannon, Ramaswami, Parikshak, Assary and Mill2019), and autism-related co-methylation modules to be significantly enriched for synaptic, neuronal, and immune dysfunction genes (Wong et al., Reference Wong, Smith, Hannon, Ramaswami, Parikshak, Assary and Mill2019). Another post-mortem brain EWAS reported DNAm levels at autism-associated sites to resemble the DNAm states of early fetal brain development (Corley et al., Reference Corley, Vargas-Maya, Pang, Lum-Jones, Li, Khadka and Maunakea2019). This finding suggests an epigenetic delay in the neurodevelopmental trajectory may be a part of the molecular pathophysiology of autism.

Overall, methylomic studies of autism provide increasing evidence that common genetic risk variants of autism may alter DNAm across tissues, and that the epigenetic dysregulation of neuronal processes can contribute to the development of autism. Stratification of study participants based on their genetic risk variants may provide deeper insight into the role of aberrant epigenetic regulation in subgroups within autism.

Transcriptomics

Transcriptomics of post-mortem brain tissue

Although blood-derived transcriptome can be feasible to study due to easy access to the biological specimen, blood transcriptome results are not necessarily representative of the transcriptional machinery in the brain (GTEx Consortium, 2017). Hence, it is extremely hard to establish a causal relationship between blood transcriptional dysregulations and phenotypes in autism. A landmark initiative by Allen Brain Institute to profile human developing brain expression patterns (RNA-seq) from post-mortem tissue enabled neurodevelopmental research to investigate gene expression in the brain (Sunkin et al., Reference Sunkin, Ng, Lau, Dolbeare, Gilbert, Thompson and Dang2013). Analyzing post-mortem brain tissue, multiple studies identified dysregulation of genes at the level of gene exons impacted by rare/de novo mutations in autism (Uddin et al., Reference Uddin, Tammimies, Pellecchia, Alipanahi, Hu, Wang and Scherer2014; Xiong et al., Reference Xiong, Alipanahi, Lee, Bretschneider, Merico, Yuen and Frey2015), including high-resolution detection of exon splicing or novel transcript using brain tissue RNA sequencing (RNA-seq). High-resolution RNA-seq enabled autism brain transcriptome analysis on non-coding elements, and independent studies identified an association with long non-coding RNA and enhancer RNA dysregulation (Wang et al., Reference Wang, Zhao, Ju, Flory, Zhong, Jiang and Zhong2015; Yao et al., Reference Yao, Lin, Gokoolparsadh, Assareh, Thang and Voineagu2015; Ziats & Rennert, Reference Ziats and Rennert2013).

Although it is difficult to access post-mortem brain tissue from autistic individuals, studies of whole-genome transcriptome from autism and control brains have revealed significantly disrupted pathways (Fig. 4) related to synaptic connectivity, neurotransmitter, neuron projection and vesicles, and chromatin remodeling pathways (Ayhan & Konopka, Reference Ayhan and Konopka2019; Gordon et al., Reference Gordon, Forsingdal, Klewe, Nielsen, Didriksen, Werge and Geschwind2019; Voineagu et al., Reference Voineagu, Wang, Johnston, Lowe, Tian, Horvath and Geschwind2011). Recently, an integrated genomic study also identified from autism brain tissue a component of upregulated immune processes associated with hypomethylation (Ramaswami et al., Reference Ramaswami, Won, Gandal, Haney, Wang, Wong and Geschwind2020). These reported pathways are in strong accordance with numerous independent autism studies that integrated genetic data with brain transcriptomes (Courchesne, Gazestani, & Lewis, Reference Courchesne, Gazestani and Lewis2020; Uddin et al., Reference Uddin, Tammimies, Pellecchia, Alipanahi, Hu, Wang and Scherer2014; Yuen et al., Reference Yuen, Merico, Bookman, Howe, Thiruvahindrapuram, Patel and Scherer2017). A large-scale analysis of brain transcriptome from individuals with autism identified allele-specific expressions of genes that are often found to be impacted by pathogenic de novo mutations (Lee et al., Reference Lee, Kang, Gandal, Eskin and Geschwind2019a). The majority of the studies are in consensus that genes that are highly active during prenatal brain development are enriched for clinically relevant mutations in autism (Turner et al., Reference Turner, Coe, Dickel, Hoekzema, Nelson, Zody and Eichler2017; Uddin et al., Reference Uddin, Tammimies, Pellecchia, Alipanahi, Hu, Wang and Scherer2014; Yuen et al., Reference Yuen, Merico, Bookman, Howe, Thiruvahindrapuram, Patel and Scherer2017). Recently, a large number (4635) of expression quantitative trait loci were identified that were enriched in prenatal brain-specific regulatory regions comprised of genes with distinct transcriptome modules that are associated with autism (Walker et al., Reference Walker, Ramaswami, Hartl, Mancuso, Gandal, de la Torre-Ubieta and Geschwind2019).

Fig. 4. Most commonly reported three pathways (Ayhan & Konopka, Reference Ayhan and Konopka2019; Gordon et al., Reference Gordon, Forsingdal, Klewe, Nielsen, Didriksen, Werge and Geschwind2019; Voineagu et al., Reference Voineagu, Wang, Johnston, Lowe, Tian, Horvath and Geschwind2011) associated with autism. (a) The synaptic connectivity and neurotransmitter pathway involves genes (yellow rectangular box) within presynaptic and postsynaptic neurons. Neurotransmitter transport through numerous receptors is an essential function of this pathway; (b) the chromatin remodeling pathway involves binding of remodeling complexes that initiate the repositioning (move, eject, or restructure) of nucleosomes that potentially can disrupt gene regulation; and (c) the neural projection pathway [adapted from Greig, Woodworth, Galazo, Padmanabhan, & Macklis (Reference Greig, Woodworth, Galazo, Padmanabhan and Macklis2013)] involves the projection of neural dendrite into distant regions and the migration of neuronal cells through ventricular (VZ) and subventricular zones (SVZ) into the different cortical layers (I-VI).

Clinical and therapeutic implications

In some, but not all, best practice clinical guidelines, genetic tests such as fragile X testing, chromosomal microarray, and karyotype testing are part of the standard medical assessment in a diagnostic evaluation of autism to identify potentially etiologically relevant rare genetic variants (Barton et al., Reference Barton, Tabor, Starks, Garrison, Laurino and Burke2018). The guidelines vary with respect to whether genetic testing is recommended for all people with autism, or based on particular risk factors, such as ID, seizures, or dysmorphic features. The DSM-5 diagnosis of autism includes a specifier for associated genetic conditions (APA, 2013). Although genetic test results may not usually have consequences for treatment changes, the results could inform recurrence risk and provide families with access to information about symptoms and prognosis. In the future, gene therapy, CRISPR/Cas9, and genome editing technologies may lead to the gene-specific design of precision medicine for rare syndromic forms of autism (Benger, Kinali, & Mazarakis, Reference Benger, Kinali and Mazarakis2018; Gori et al., Reference Gori, Hsu, Maeder, Shen, Welstead and Bumcrot2015).

Given that a substantial proportion of the genetic liability to autism is estimated to be explained by the cumulative effect of a large number of common SNPs, polygenic scores have gained traction as potential biomarkers. However, the predictive ability of polygenic scores from the largest autism GWAS to date is too low to be clinically useful. The odds ratio when comparing the top and bottom polygenic score decile groups is only 2.80 (95% CI 2.53–3.10) (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019). Additionally, polygenic scores based on the samples of European ancestry do not translate well in populations with diverse ancestry (Palk, Dalvie, de Vries, Martin, & Stein, Reference Palk, Dalvie, de Vries, Martin and Stein2019).

Genetic testing can in the future become useful for informing screening or triaging for diagnostic assessments or identifying who may be more likely to respond to which type of intervention (Wray et al., Reference Wray, Lin, Austin, McGrath, Hickie, Murray and Visscher2021). Genetics may also help identify individuals with autism who are at a high risk of developing co-occurring physical and mental health conditions or likely to benefit from treatments of such conditions. A top research priority for autistic people and their families is addressing co-occurring mental health problems (Autistica, 2016), which may sometimes be the primary treatment need as opposed to autism per se. Genomics may also be helpful to repurpose existing treatments and better identify promising treatments. There are active clinical trials to repurpose drugs in autism (Hong & Erickson, Reference Hong and Erickson2019). Moreover, genetics can be used to identify social and environmental mediating and moderating factors (Pingault et al., Reference Pingault, O'Reilly, Schoeler, Ploubidis, Rijsdijk and Dudbridge2018), which could inform interventions to improve the lives of autistic people.

Notably, there are important ethical challenges related to clinical translation of advances in genetics, including concerns about discriminatory use, eugenics concerning prenatal genetic testing, and challenges in interpretation and feedback (Palk et al., Reference Palk, Dalvie, de Vries, Martin and Stein2019). People with autism and their families are key stakeholders in genetic studies of autism and essential to include in discussions of how genetic testing should be used.