miRNA expression in autism spectrum disorders
In this study, we demonstrate the differential expression of 43 miRNA species in LCLs from individuals with ASD relative to controls (Table 1), 16 of which are brain-specific, brain-related, or involved in neural differentiation [59–62]. Although the total number of samples in this study is modest, the use of discordant monozygotic twins and sibling case-controls offers the ability to identify differences in miRNA against the same or closely related genotype, which is an advantage in investigations of epigenetic mechanisms contributing to autism. We have previously used this strategy in first identifying gene expression differences in these same monozygotic twins  and sibling case-controls , and then validated our initial findings with a larger study involving 116 unrelated case-controls . Here, we further utilize the original gene expression data of these same samples to demonstrate that differentially expressed miRNA can account for approximately 36% of the differentially expressed transcripts [21, 40], thus implicating miRNA as a potent regulator of gene expression in ASD. Functional analyses of the putative gene targets that show inverse correlation with the expression of miRNA reveal numerous processes relevant to or associated with ASD that are potentially regulated by the differentially expressed miRNA (Table 2, Figure 4). These processes include embryonic development, synaptic development and function, circadian rhythm signaling, inflammation, androgen metabolism, and digestive functions, mirroring the major findings of our gene expression analyses [21, 40, 77] Significantly, we verify inverse changes in the levels of putative target genes of two of the altered brain-specific miRNAs through the use of anti-miRs (for knockdown) and pre-miRs (for overexpression) (Figure 5).
To date, only two other studies have conducted miRNA expression profiling of autistic individuals. Talebizadeh and colleagues  evaluated the global expression of 470 known human miRNAs using LCLs derived from six autistic individuals and six sex- and age-matched controls by miRNA microarray assays. Of these 470 miRNAs, they found nine that were significantly differentially expressed in the autistic samples. Three of the nine miRNAs were replicated in our study, with similar up-regulation of miR-23a and miR-23b, but down-regulation of miR-132. Although we have no specific explanation for this contrasting result for miR-132, differences between our study and that of Talebizadeh et al.  include our use of related samples (that is, co-twins/siblings) as controls, a custom-printed rather than commercial platform, and the restriction of our study to male subjects. Additional analyses are thus required to further explain the differences in miRNA expression data between these two studies on LCLs.
Abu-Elneel et al.  investigated the expression of 466 human miRNAs in postmortem cerebellar cortex tissue of 13 autistic individuals using multiplex quantitative PCR and found 13 down-regulated and 16 up-regulated miRNAs. Interestingly, the up-regulation of miR-23a and down-regulation of miR-106b reported in the autistic cerebellar cortex were also found in our study using LCLs. Predicted potential target genes of miR-23a were found to be associated with neurological diseases and skeletal and muscular system development and functions, whereas those of miR-106b were associated with neurological diseases, inflammatory diseases, and gastrointestinal diseases (Table 2). These findings support the hypothesis that miRNA dysregulation in peripheral blood cells can reflect at least some miRNA alterations occurring in the brain, thus lending support to the use of LCLs as a surrogate tissue to study miRNA expression in individuals with ASD.
Brain-related miRNAs are differentially expressed in LCLs from ASD patients
Our earlier studies profiling gene expression in LCLs from monozygotic twins and siblings discordant for diagnosis of autism and unrelated autistic case-controls reveal the differential expression of hundreds to thousands of genes [21, 40, 77], suggesting that higher level epigenetic gene regulatory mechanisms are involved in ASD. The present study provides further insight into the post-transcriptional gene regulatory network associated with ASD by identifying differential miRNA expression as one mechanism for the differential gene expression associated with ASD. Interestingly, at least 16 of these miRNAs have been previously reported by Sempere and colleagues  to be brain-specific, brain-enriched, or induced by neuronal differentiation. Krichevsky and colleaques  reported significant changes in the expression of nine miRNAs during brain development; one of these miRNAs (miR-103) was also significantly differentially expressed in our study. Thus, the differential expression of these brain-related miRNAs in LCLs suggests that gene expression differences previously observed in LCLs [21, 40, 77] may reflect similar changes in the brain, possibly due to global or system-wide dysregulation of miRNA expression.
Biological functions associated with the confirmed miRNAs and their target genes
Using miRNA TaqMan qRT-PCR, we confirmed four differentially expressed miRNAs (hsa-miR-219-5p, hsa-miR-139-5p, hsa-miR-29b, and hsa-miR-103) previously reported to be associated with the brain [59–62]. Of the confirmed miRNAs, we observed a significant decrease in brain-specific hsa-miR-219, which is associated with circadian rhythm and N-methyl-D-aspartate (NMDA) glutamate receptor signaling, both of which have been implicated in ASD [72–77, 80, 81]. In particular, Kocerha and colleagues  found that disruption of NMDA receptor signaling resulted in decreased levels of miR-219 in mice. Hypofunction of NMDA receptor signaling has been associated with a number of neurological disorders, including autism [83–85], attention deficit hyperactivity disorder [86, 87], and schizophrenia . One of the putative target genes whose expression was confirmed to be inversely correlated with hsa-miR-219 expression is PLK2 (Figure 4), a serine/threonine kinase expressed in the brain  that participates in regulation of cell cycle progression  and homeostatic plasticity of hippocampal neurons [69, 70]. A recent study found that PLK2 was induced during prolonged epileptiform activity, and was required for the activity-dependent reduction in membrane excitability of pyramidal neurons, suggesting PLK2's role in preventing escalating potentiation and in maintaining synapses in a plastic state . PLK2 induction in hippocampal neurons resulted in weakening of synapses through phosphorylation and degradation of post-synaptic spine-associated Rap GTPase-activating protein (SPAR), a regulator of actin dynamics and dendritic spine morphology [69, 71], leading to loss of mature dendritic spines and synapses [91, 92]. Over-expression of PLK2 in individuals with ASD due to decreased hsa-miR-219 levels as observed in this study (Figure 5, Table 4) may thus lead to global reduction in synaptic strength and neuronal excitability, which could be partially responsible for the synaptic dysfunction implicated in ASD.
Another confirmed brain-specific miRNA differentially expressed in individuals with ASD is hsa-miR-29b. Besides its confirmed target, ID3 (Figure 5), which is involved in regulating the biological clock (see below), other target genes that show expression levels inversely correlated with the over-expression of this miRNA include COL6A2 (Collagen, type VI, alpha 2), CLIC1 (Chloride intracellular channel 1), ARPC5 (Actin related protein 2/3 complex, subunit 5, 16 kDa), and KIF26b (Kinesin family member 26B). Interestingly, a number of mutations in COL6A2 have been observed in muscular disorders, including Bethlem myopathy [93–95] and Ullrich congenital muscular dystrophy [94, 96–98]. Mutation in the COL6A2 gene results in decreased COL6A2 transcript, leading to disruption of collagen formation and stability, which results in decreased muscle strength . A number of motor impairments and muscular disorders, including muscular dystrophy, hypotonia, and muscle weakness, are observed in individuals with ASD [50, 99, 100]. It is therefore interesting to postulate that suppression of COL6A2 as a result of up-regulated hsa-miR-29b may be one of the genetic mechanisms underlying muscular disorders and motor impairments frequently observed in individuals with ASD.
Among brain-enriched miRNAs , hsa-miR-139-5p was selected for confirmation analysis using miRNA TaqMan qRT-PCR assay. Although the precise targets in brain are not known, one of its putative targets (myomegalin or PDE4DIP (Phosphodiesterase 4D interacting protein)) is a homolog of brain-enriched CDK5RAP2 (CDK5 regulatory subunit associated protein 2), a gene that regulates brain size [101–104], which has been shown to be abnormal in ASD [105–119]. Interestingly, this miRNA has been shown to be involved in prion-induced neurodegeneration .
Two of the most up-regulated miRNAs, miR-103 and miR-107 (Table 1), have been reported to be paralogous miRNAs. miR-103 and miR-107 are expressed in many human organs, with the highest concentrations occurring in brain tissue . Furthermore, miR-103 was demonstrated to change during corticogenesis in mice . Although the specific targets of miR-103/107 in brain are unknown, these miRNAs are known to be associated with lipid metabolism , and in fact reside within introns of the pantothenate kinase (PANK) genes, which catalyze the biosynthesis of Coenzyme A, a critical component in fatty acid biosynthesis and oxidation. It should be noted that, while PANK was not found to be among the significantly differentially expressed genes in this study, it was found to be increased in ASD and in the same direction as miR-103/107 in our previous study of a larger cohort of 31 autistic individuals with severe language impairment and 29 controls . Aside from the association of PANK mutations and a neurodegenerative (Hallervorden-Spatz) disease [122, 123], alterations in lipid and fatty acid metabolism are also known to be associated with ASD. Vancassel and colleagues  examined the levels of phospholipid fatty acids in the plasma of individuals with ASD compared to controls with mental retardation and found significant reductions in docosahexaenoic acid (22:6n-3) levels in autistic individuals, resulting in significantly lower levels of total n-3 polyunsaturated fatty acids. The dysregulation of miR-103/7 may therefore contribute to abnormal lipid and fatty acid metabolism in ASD.
miRNAs regulating circadian rhythm are significantly dysregulated in ASD
Recently, dysregulation of circadian rhythm has been considered as a mechanism for impairments in neurological and other functions (for example, sleep, digestive) in ASD [72–77]. In particular, the circadian rhythm (or 'clock') genes have been posited to underlie social timing deficits associated with autism , as well as lead to the sleep disorders frequently observed in ASD [125, 126]. Bourgeron  also proposed an important role for circadian rhythm with respect to regulation of synaptic genes (NLGN3 (Neuroligin 3), NLGN4 (Neuroligin 4), NRXN1 (Neurexin 1), and SHANK3 (SH3 and multiple ankyrin repeat domains 3)), thus affecting susceptibility to ASD. Our large-scale genomic study also found strong support for an association between ASD and circadian rhythm dysfunction . Interestingly, as many as 15 circadian rhythm genes, including AANAT (Arylalkylamine-N-acetyltransferase), BHLBH2 (Class B basic helix-loop-helix protein 2), CRY1 (Cryptochrome 1 (photolyase-like)), NPAS2 (Neuronal PAS domain protein 2), PER1 (Period homolog 1), PER3 (Period homolog 3), and DPYD (Dihydropyrimidine dehydrogenase), were differentially expressed exclusively in the most severe phenotype of ASD, which was characterized by severe language impairment [77, 127]. It is interesting to note that two of the most significantly down-regulated miRNAs (miR-219 and miR-132) in individuals with ASD have been reported to be involved in modulating the master circadian clock located in the suprachiasmatic nucleus [128–131]. Specifically, brain-specific miR-219 was a target of the master circadian regulator CLOCK and BMAL1 (Brain and muscle ARNT-like 1) complex, exhibited robust circadian rhythm expression, and fine-tuned the length of the circadian period in mice [130, 131]. It is relevant, therefore, that we demonstrate that PLK2, which is involved in circadian rhythm signaling, is a target of miR-219 (Figure 5).
Functional analyses of putative target genes using IPA (Table 2) also showed that other miRNAs (hsa-miR-29b and hsa-miR-376a) are significantly associated with circadian rhythm signaling, with hsa-miR-29b targeting the ID3 gene, which might be important for entrainment and operation of the mammalian circadian system through ID3 interaction with CLOCK and BMAL1 . Significantly, we show that hsa-miR-29b pre-miR precursor results in the down-regulation of ID3 transcript. ID3 is also a neuronal target of MeCP2 (Methyl CpG binding protein 2), which is the causative gene for Rett syndrome . Other putative targets of brain-specific hsa-miR-29b are genes known to interact in the regulation of the biological clock, including ARNTL (Aryl hydrocarbon receptor nuclear translocator-like; BMAL1), ATF2 (Activating transcription factor 2), DUSP2 (Dual specificity phosphatase 2), PER1, PER3, and VIP (Vasoactive intestinal peptide). Although only DUSP2 was found to be differentially expressed in the current analysis, it is interesting to note that our recent large-scale gene expression study of LCLs from over 100 unrelated case-controls found significant decreases in PER1 and PER3 transcript levels in individuals with the most severe phenotype of ASD . However, further experimental studies are required to determine whether or not the over-expression of hsa-miR-29b results in the suppression of these two PER genes.
Target genes of miRNAs involved in functions and processes associated with ASD
To obtain more insight into the biological functions regulated by each of the differentially expressed miRNAs, the potential target genes of each miRNA were predicted in silico and uploaded into IPA network prediction software. For most miRNAs, target genes were predicted to be involved in neurological disease and nervous system development and function on the basis of gene enrichment within the dataset (Table 2). This finding suggests that the significantly differentially expressed miRNAs may lead to post-transcriptional dysregulation of target genes that, in turn, leads to the disruption in neurological functions contributing to ASD pathophysiology.
The dysregulation of these specific miRNAs may also potentially impact other physiological functions. Besides the neurological functions, almost half of the differentially expressed miRNAs targeted a number of genes involved in gastrointestinal disorders and hepatic diseases, which have been found in approximately 50% of individuals with ASD [133, 134]. Our findings thus provide a plausible explanation for some of the systemic effects observed in ASD that affect other organs in addition to the nervous system.
Steroid hormones have been suggested to be involved in the etiology or susceptibility to ASD [135, 136]. In particular, previous studies have reported elevated androgen levels in the serum of autistic individuals, including females [135, 136], and we have recently reported changes in genes in LCLs that correlated with increases in testosterone [40, 77]. Androgens and estrogens are known to participate in synaptic plasticity in the brain of rats. Whereas estrogens have been found to take part in synaptic plasticity in the hippocampus of female rats , androgens can modulate that function in both male and female rats . Within this context, it is noteworthy that four of the differentially expressed miRNAs (miR-16, miR-186, miR-25, and miR-195) target genes participating in estrogen receptor signaling. miR-136, which was one of the most down-regulated miRNAs found among all five ASD samples, is also associated with androgen and estrogen metabolism.
miRNAs are known to act through translational repression [23–27]. However, the repressed transcripts are often degraded in P-bodies, ultimately leading to reduced transcript levels for a particular miRNA-repressed gene . This inverse correlation between miRNA and target gene transcript levels is further suggested by the observed inverse correlation between miRNA 'host' genes and the miRNA target transcripts using a novel analysis called HOCTAR (for 'host gene oppositely correlated targets') . Thus, an increase in a particular miRNA is likely to lead to decreased transcript levels of target genes and vice versa. However, inverse correlation of miRNA and target mRNA levels is not necessarily observed. Nevertheless, comparing the miRNA expression data obtained by the present study with data obtained by our previous cDNA microarray analysis of these same samples reveals that the direction of change for roughly 27% of the differentially expressed genes was inversely correlated with that of the respective potentially regulatory miRNAs. Relational gene networks constructed using computational network prediction tools show that the inversely correlated target genes of the significantly differentially expressed miRNAs are linked to autism as well as to co-morbid disorders frequently reported in many autistic individuals (Figure 3). For example, a number of genes in the network are linked to synaptic function, such as regulation of synapse, synaptic plasticity, and synaptic transmission. Synaptic plasticity has been comprehensively described in the context of fragile X syndrome and linked to autism . FMRP (Fragile X mental retardation protein), the key protein missing in fragile X syndrome, is an RNA binding and transport protein that regulates the translation of many other proteins important for synaptic plasticity, including neuroligins 3 and 4 and SHANK, all of which have been previously associated with autism [12, 13, 139, 140] Muscular dystrophy and muscle disease are also known to be among the co-morbid disorders frequently found in autism . Thus, putative target genes of the differentially expressed miRNAs identified in this study can be associated with both neurological as well as co-morbid features of ASD.
Although the major behavioral symptoms of ASD appear to be of neurological origin, the prevalence of gastrointestinal abnormalities, hypotonia, and immune disorders in individuals with ASD have led some researchers to view ASD more as a systems disorder that is a result of gene and environment interactions. Thus, several recent studies, including three from our laboratory [21, 40, 77], have used LCLs as a surrogate experimental model to better understand the pathobiology of ASD as well as to identify peripheral biomarkers of ASD for diagnostic purposes [21, 38, 40, 77, 127, 141, 142]. In particular, our previous study of monozygotic twins discordant for diagnosis or severity of autism revealed differentially expressed genes with known neurological functions of potential relevance to autism . Because identical twins share the same genotype, this study suggested the involvement of epigenetic factors in the regulation of gene expression in ASD. Furthermore, the global scale of the observed changes in gene expression suggested the operation of 'master switches' that can activate or suppress multiple genes at once. Non-coding RNAs, including miRNAs, are potential epigenetic regulators of gene expression and can operate in this fashion [24, 143–146].