Autism spectrum disorder is a heterogeneous disorder typified by deficits in social communication and restricted, repetitive patterns of behavior (18). Children with ASD are often adherent to daily routines, preoccupied with specific topics, and over sensitive to sensory inputs such as sounds or textures (19). In most cases, ASD symptoms are first recognized in early childhood. The average age of ASD diagnosis is 4 years (20). Rates of ASD are increasing. Recent reports from the Centers for Disease Control and Prevention estimate the prevalence of ASD to be 1:68 children (1:45 among males) (21). This increase may be the result of multiple factors including public awareness, changes in diagnostic criteria, and environmental influences. However, genetics also plays an important role in ASD (22).

A large number of familial studies demonstrate that ASD is a heritable disorder (23, 24), with estimated genetic contributions accounting for >50–60% of ASD risk (25). Indeed, a large number of genetic variants and chromosomal abnormalities are linked to ASD (26). These variants tend to be highly penetrant but rare. However, in some cases, common variants with low penetrance have been reported (27). Neither explains the incidence of ASD in the general population. Genetic epidemiologic studies have also shown that ASD is not a single disease but a constellation of symptoms involving multiple gene networks (28). Posttranscriptional mechanisms, such as miRNA, that broadly influence gene expression without altering the DNA code represent one means of altering entire gene networks (29). Increasingly, investigators have turned toward these mechanisms to explain the dysregulation of neurodevelopmental pathways that occurs in ASD (30, 31).

MicroRNAs are short (18–25 nucleotide), non-coding RNA molecules (32) that influence gene expression and numerous cellular processes, including proliferation, differentiation, and apoptosis (33). Over 2500 mature miRNAs (and 1800 precursors) have been identified in humans (34). Many of these are evolutionarily conserved (35). The genes for miRNAs are found in inter- or intra-genic regions (36). They are transcribed into precursor hairpin structures (pri-miRNAs) by RNA polymerase-II and trimmed into pre-miRNA by the RNase Drosha. After leaving the nucleus, Dicer liberates the pre-miRNA into a single-stranded molecule that can be bound within the miRNA-induced silencing complex (miRISC) (37). The targeting of the mature miRISC to individual mRNAs is based on the “seed sequences” that reside at the 5′ end of the miRNA and are complementary to the 3′ regions of mRNAs. Each seed sequence aligns with hundreds of mRNAs, and there are several miRNAs with shared seed sequences (38). Once bound to a target mRNA, the miRISC complex reduces the efficiency of gene expression by repressing protein translation or promoting degradation of the mRNA transcript (37).

MicroRNAs regulate approximately two-thirds of human mRNAs (39). Each miRNA has hundreds of potential mRNA targets, giving it the ability to modulate entire gene networks. In addition, miRNAs can circulate within exosomes, microvesicles, or RNA-binding proteins such as high-density lipoproteins, and thereby travel extracellularly to alter gene expression in distant tissues. This renders them exceedingly stable and easily measured in serum, plasma, saliva, urine, and other biofluids (40). However, miRNAs are most ubiquitous in the CNS, which expresses an estimated 70% of all miRNAs (41). Notably, expression of brain miRNAs changes throughout childhood and varies across brain regions (42). Neurons and glia are abundant sources of miRNAs, which can be readily transported across the blood–brain barrier (BBB). Moreover, miRNAs in neurons also help compartmentalize or localize mRNA expression and translation within specific subcellular regions such as axons and dendrites.

MicroRNAs play important roles in neurogenesis, synaptogenesis, and neuronal migration (43, 44). A major means by which these roles are manifest is the effect that neuronal miRNAs have on spatial localization or compartmentalization of protein translation in different neuronal subregions, such as axons, dendrites, and synapses (45, 46). Recent data support the relevance of these processes for neurodevelopmental disorders. For example, in mice with fragile X-associated tremor/ataxia syndrome (FXTAS), the fragile X mental retardation (FMR1) transcript is targeted by miR-221, miR-101, and miR-129-5p (47). Synaptosomal preparations from FXTAS mice demonstrate dysregulation of transcripts in learning and social interaction along with reduced levels of miR-221. Other examples include studies of disruption of miRNA regulatory genes such as Dicer or DGCR8, which produce alterations in synaptic plasticity within the prefrontal cortex (48). The gene for DGCR8 is located at 22q11.21 and a well-characterized hemi-deletion of this region (referred to as velocardiofacial syndrome) produces a complex phenotype along with ASD traits in approximately one-third of subjects (49). Moreover, mice with DGCR8 haploinsufficiency have reduced dendrite complexity, fewer neuronal spines, and decreased neurogenesis (50). Complementing these data, dysregulation of miRNA expression has been described in a number of developmental, neuropsychiatric, and neurodegenerative disorders including schizophrenia, Alzheimer’s disease, and Parkinson’s disease (36).

Four studies examining postmortem brain tissue of human subjects with ASD have identified 91 miRNAs with differential expression compared with typically developing controls. The exact pathophysiologic role of these miRNAs in ASD remains unclear; however, many are implicated in neurogenesis and synaptogenesis. A study of cerebellar cortex in 13 adults with ASD (8) identified 28 miRNAs with differential expression and 7 of these targeted the SHANK3 transcript. SHANK3 is involved in regulation of synaptic density and has copy number variants (CNVs) and point mutations in nearly 1% of ASD cases (51). A study by Wu et al. (52) examined cerebellar cortex in 28 ASD subjects. The authors found miR-21-3p was upregulated in ASD subjects and demonstrated that its overexpression in human neuroprogenitor cells could repress multiple M16 hub genes, including DLGAP1 (a scaffold protein that interacts with SHANK3). Thus, SHANK3 represents one pathophysiologic target with an established role in ASD that appears to be targeted by multiple miRNAs identified in multiple studies of ASD brain tissue.

This review synthesizes the findings of 12 studies, which compared miRNA expression in human cases of ASD to miRNA profiles in healthy controls (Table 1). Four of these studies involved brain tissue (8, 52–54), three involved peripheral blood (12–14), one examined saliva (15), one examined olfactory precursor cells (16), and three employed cultured lymphoblasts (9–11).

The 12 studies together identified 219 miRNAs with potential implications in ASD (Table 2A), including 185 unique and 34 overlapping miRNAs (15.5%). Twenty-seven overlapping miRNAs changed in the same direction (Table 2B) in ≥2 studies, and 3 changed in the same direction in ≥3 studies (miR-23a, miR-146a, and miR-106b). Notably, 14 of the 27 miRNAs shared seed sequences with other ASD-associated miRNAs, suggesting that the pathophysiology of miRNA dysregulation in ASD might be controlled, at least in part, at the seed level. There is no single miRNA identified in all the 12 studies. This may be a function of ASD heterogeneity, miRNA tissue specificity, or advances in RNA quantification, which have led to varying alignment techniques and identification of new miRNAs.

Of the 27 overlapping miRNAs, 22 show differential regulation in studies of the CNS. The remaining five are expressed in the developing brain (42, 53). These 27 were predicted to target 1245 mRNAs (Table S1 in Supplementary Material) with target scores ≥93 (55).¹ Notably, 86 of these mRNAs overlapped with the 519 human ASD candidate genes with supportive evidence existing in the Simons Foundation Autism Research Initiative (SFARI) database,² representing a 6.6-fold enrichment for ASD-associated genes compared to chance (odds ratio: 7.7, 95th CI = 6.0–9.9, z = 16.2, p < 0.0001).

We next examined the evidence for functional enrichment of specific biologic processes using the Database for Annotation, Visualization, and Integrated Discovery (DAVID). This revealed 225 total cluster mappings (56), with the top nodes including a KEGG pathway involved in neurotrophin signaling [n = 25 genes, false discovery rate (FDR) = 0.0018]. In fact, this node had the largest fold change (FC) (3.0-fold) among all detected clusters (Table S2 in Supplementary Material). DAVID enrichment analysis also identified 662 (55.9%) of the target mRNAs as expressed in brain tissue, with additional enrichment of genes expressed in epithelium (n = 268), fetal kidney (n = 35), placenta (n = 275), amygdala (n = 61), and T-cells (n = 37) (Table S3 in Supplementary Material).

To further examine functional relevance of the most probable mRNA targets, the Ingenuity Pathway Analysis (IPA) databases for “Canonical Pathways” and “Diseases and Biological Function” were used. These analyses were performed using a restricted set of 293 mRNAs that were predicted to be targets of the 27 overlapping miRNAs with miRDB target scores ≥95. The results indicated enrichment of 20 Canonical Pathways (FDR < 0.05), including two related to neurotrophin signaling (Table S4 in Supplementary Material). Moreover, the “Diseases and Biological Function” analysis yielded a list of 500 networks for the 293 mRNA targets, including 93 (18.6%) involved in brain function, of which 35 survived FDR correction (Table S5 in Supplementary Material). These represented several functions, including brain development, neurogenesis, neural and glial differentiation, synaptic development and function, and cognitive and motor function. The diseases identified among the brain-related networks included schizophrenia, abnormal brain morphology and trigeminal nerve morphology, abnormal posture, cognitive impairment, and seizure disorders.

Upstream Regulatory analysis was performed in IPA considering only experimentally observed and high-confidence miRNA–mRNA interactions, which reduced our list of 27 miRNAs to 20. Notably, 199 mRNAs were targeted by ≥2 miRNAs, and 55 genes were targeted by ≥5 of the 20 miRNAs. The significance of this was determined using a Fisher’s exact test, which confirmed enrichment for 49 mRNAs (Table S6 in Supplementary Material). Protein–protein interaction (PPI) networks were also explored in the 199 target genes using the STRINGv10 database,³ which yielded a significant PPI network containing 137 edges and several nodes involved in brain function (Figure S1 in Supplementary Material). Among this network was an intriguing interaction between neuronal cell adhesion molecule (NRCAM; involved in directional signaling of axonal cone growth and implicated in ASD) (57), semaphorin 3a (SEMA3A; a chemoattractive agent involved in axon guidance) (58), and brain-derived neurotrophic factor (BDNF; essential for synaptogenesis in the developing hippocampus and altered in the periphery of ASD subjects) (59).

MicroRNA expression is a dynamic process, changing throughout development and across brain regions (43). A study of non-coding RNA in the superior temporal gyrus of ASD subjects demonstrated changes in regional miRNA profiles that could contribute to ASD (60). Specifically, while miRNA patterns in the superior temporal sulcus (a region relevant to social interaction) and primary auditory cortex changed throughout development in the typical brain, many of the changes were blunted in subjects with ASD. Four of the individual miRNAs identified (miR-93-3p, miR-103, miR-132, and miR-320) were among the 219 miRNAs altered in other studies of ASD, with 3 of the 4 changed in multiple studies (Table 2).

The molecular mechanisms underlying miRNA dysregulation in ASD are still being explored. One explanation is that autistic miRNA patterns may represent a permanent response to an early environmental insult, such as perinatal hypoxia (61). In this scenario, miRNA regulation could be disrupted by a layer of epigenetic machinery such as DNA methylation. In a study of 12 subjects with ASD, expression of miR-142 was increased in Brodmann’s area 10 compared with healthy controls – a change accompanied by hypomethylation of the miR-142 promoter region (53).

Another possible mechanism is that individual miRNAs sequences are altered in children with ASD. Pooled analysis of common and rare single nucleotide polymorphisms (SNPs) in 449 cases of ASD using whole exome sequencing has identified 2 miRNA clusters: miR-133b/miR-206 and mir-17/miR/18a/miR-19a/miR20a/miR-19b-1/miR-92a (62). Four of these individual components have disrupted expression levels in children with ASD, and miR-92a is downregulated in multiple studies (9, 14).

Yet another explanation could be that the location of specific miRNAs at CNV loci leads to their dysregulation. Of the 41 CNV loci associated with ASD, >10% contain miRNAs (63). Three of the 10 “hub miRNAs” (i.e., miRNAs at CNV loci with the greatest gene network connectivity) have been previously identified in studies of ASD (miR-34a, miR-548, and miR-195) and target multiple genes involved in neurodevelopment. Alterations in miR-195 have been reported in serum (13, 14) and lymphoblasts (10) of ASD individuals. Furthermore, a key gene involved in miRNA synthesis (Dicer) was identified as a common target for many of the hub miRNAs, suggesting that some CNVs in ASD might cause extensive dysregulation of the entire miRNA machinery.

Although the brain is considered the primary pathological target in ASD, alterations in miRNA expression have been noted in a number of other body sites (13, 15). This may explain (in part) why children with ASD experience medical comorbidities outside the CNS, including gastrointestinal issues such as esophageal reflux and constipation (64). They also have altered oral-motor function (speech apraxia) and sensory processing (65, 66). Given these comorbidities, it should come as no surprise that oral miRNA patterns are altered in children with ASD. In addition, the proximity of the oropharynx to the CNS and its abundant sensory and motor nerve innervation (involving branches of the glossopharyngeal, facial, vagus, and trigeminal nerves) could provide an ample source of miRNA (67, 68).

Alterations in miRNA profiles of peripheral blood sites may be explained by circulating miRNA that has crossed the BBB (especially considering that ASD is not associated with hematologic disorders). In peripheral sources such as blood, brain-related miRNA may be stable and abundant as a result of exosomal transport (13, 14). Thus, while biofluids such as serum and saliva may not represent optimal models for studying CNS pathophysiology, they may represent ideal reservoirs for ASD biomarkers.

The potential of miRNAs as biomarkers in ASD has been explored in studies of serum and saliva. In a study of serum miRNAs from 55 children with ASD, the authors identified three miRNAs (miR-130-3p, miR-181b-5p, and miR-320a) with an area under the curve (AUC) >0.85 (13). In our own more recent study, involving 24 children with ASD, a set of 14 salivary miRNAs showed more than 95% accuracy (AUC = 0.92) at differentiating control and ASD subjects (15). This study also showed that expression patterns of individual salivary miRNAs were significantly correlated with several measures of adaptive behavior. This highlights how the utility of miRNA extends beyond simple ASD diagnosis and may 1 day be used to predict ASD phenotype and severity.