Background
Autism spectrum disorders (ASDs) constitute a heterogeneous group of neurodevelopmental disorders characterized by impaired social interaction, disrupted development of communication skills, and repetitive behaviors [
1]. Over an affected individual’s lifetime, costs of care can reach about $3.2 million while the annual cost to society is an estimated $35 billion [
2]. Such burdensome costs combined with new high estimates in prevalence—including numbers as high as 1 in 68 children [
3]—call for a need to understand pathophysiology fully and to develop new treatments. Genetic studies in autism have pinpointed a heterogeneous group of loci and genes, largely emerging from studies of rare and/or de novo genetic variation [
4‐
9]. Common susceptibility variants and inherited variants have been harder to identify in autism [
10‐
13]. Further, some recent twin studies, such as a study by Hallmayer et al., have reported a more moderate genetic heritability than older studies [
14]. These studies suggest a relatively lower concordance for autism between monozygotic twins (approximately 58% concordance) and a higher concordance between dizygotic twins (approximately 20%) as compared to older twin studies on autism (see [
15] for a recent meta-analysis of twin studies on autism). In addition to supporting a strong role for genetics, the results of Hallmayer et al. implicate a shared twin environment, such as the in utero environment, as an additional factor that may play a role in susceptibility to autism.
Transcriptome studies in autism have investigated quantitative differences in gene expression between the mRNA samples extracted from post-mortem tissue from patient brains as compared to control brains [
16‐
19]. One advantage of transcriptome studies is that they may pinpoint genes and molecular processes that are relevant to pathophysiology yet the approach circumvents the need to generate hypotheses about the genetic architecture or the gene-by-environment interactions leading to disease. Gene expression represents the summation between genetic burden and environmental insults or experience. In one of the largest studies to date, gene pathways involving synapses were found to be most enriched among the genes with decreased expression in autism, whereas pathways involving neuroimmune and microglial response were enriched among the genes with increased expression in autism [
17]. Similar findings were noted in a more recent and larger RNA-seq study of autism cerebral cortex [
19]. Interestingly, immune gene alterations had been reported previously in autism as a preliminary finding in a much smaller dataset [
16].
We have conducted an alternative analysis of the transcriptome data using differentially expressed genes from an RNA-seq dataset [
19] and three previously published microarray datasets [
16‐
18]. We discovered that a gene pathway related to mitochondrial function was downregulated in autism cerebral cortex and correlated with a pathway related to synapse function. Recent independent reports have also identified downregulation of genes related to mitochondrial processes in autism post-mortem brain [
20,
21]. These transcriptome data are also concordant with additional multifaceted findings that support a role for mitochondrial dysfunction in autism pathology [
22,
23]. In addition, autism severity may be correlated with abnormalities in biomarkers of mitochondrial function [
22], and further still, a mitochondrial signature has been seen in other neuropsychiatric conditions, such as in schizophrenia [
24]. Overall, our data support a model wherein mitochondrial processes may play an important role in the primary pathophysiology and/or progression of neuropsychiatric diseases.
Discussion
We have conducted a reanalysis of autism and control post-mortem brain gene expression using a recent RNA-seq study [
19] and three other similar gene expression studies [
16‐
18]. We discovered that genes related to mitochondria are significantly downregulated in autism brains relative to control. Abnormalities related to mitochondria have been implicated in autism pathogenesis through several lines of evidence, such as over-representation of mitochondrial disease in ASD patients and elevation of biomarkers of metabolism such as lactate and pyruvate [
22]. Further, genes for select electron transport chain complexes have been shown to be lowly expressed in the cortex of children with autism [
35].
We also observed that this mitochondria pathway gene expression correlated strongly with that of a synapse pathway, suggesting a common pathophysiology. Consistently, Gandal et al. recently described a gene module related to synaptic transmission and mitochondria that was downregulated in both autism and schizophrenia [
36]. Schizophrenia has also previously been shown to have decreased expression of mitochondria-related genes [
24].
In our study, we noted other similarities to schizophrenia, as well. For example, we particularly noted that genes related to inhibitory interneurons were downregulated. In prior studies,
GAD1 and
GAD2 have been shown to be reduced in parietal and cerebellar cortex in autism [
37] and GABA receptor density is reduced in post-mortem autism cerebral cortex [
38]. Similar inhibitory interneuron gene alterations are seen in the cerebral cortex in schizophrenia [
39]. The reason for common downregulation of inhibitory interneuron and mitochondrial genes in autism and schizophrenia is unclear. However, it is noted that both conditions are also associated with gene-by-environment interactions related to the immune system, suggesting a similar pathophysiology [
40]. The immune system’s role is evidenced by each condition’s association with maternal immune activiation during pregnancy [
41,
42], as well as with genetic variation in major histocompatibility complex genes [
43,
44].
Because we have not explored protein or functional analyses, we cannot discern whether these gene expression changes are part of the primary pathology or secondary pathology or both. However, in vitro experiments have shown a close interplay between mitochondria and synapse regulation. For example, Li et al. showed that GTPases that control mitochondrial fission and fusion also regulate synapse plasticity and density [
45]. These researchers further showed that increased neuronal activity increased mitochondrial fission in a neuron while decreased activity increased fusion, suggesting a mitochondrial response to neuronal energy needs. For autism, primary synaptic dysregulation could result in reduced neuronal energy demand and thus mitochondrial activity. Alternatively, several studies, including those that report gene mutations or susceptibility variants in mitochondrial genes [
46,
47], support the notion that primary mitochondrial defects may occur in autism. Regardless, abnormalities in mitochondria are a feature of synaptic gene dysregulation in idiopathic autism and deserve additional study. A pertinent question that results is whether autism symptoms would be responsive to medicines or supplements that are used in treatment of primary mitochondrial disease. This hypothesis has been tested in small studies [
48], but further studies may be warranted, particularly after peripheral biomarkers become available for stratifying patients into groupings that may be more amenable to these treatments.
Several other factors might affect differential expression. To account for possible confounders of the association between autism and gene pathways, we adjusted for RIN, PMI, sex, age, and cortical region in our analyses. However, some variables were not available for multivariate analysis, including those related to treatment, lifestyle, and other technical confounders. However, given that the analyses validated across datasets, the pathway results are robust, and other possible confounders are unlikely to alter interpretation of these pathways’ associations with autism.
Although the magnitudes of the expression changes of each pathway were relatively small, small-magnitude gene expression differences can still have profound effects, a finding seen in other psychiatric conditions, including schizophrenia [
49,
50]. Additionally, because each cortical sample is from a heterogeneous cell population, small changes may also represent dilution of a single cell type’s gene expression changes. In our study, gene expression changes likely reflect neurons, given the observed correlations with genes related to synaptic and axonal function.