Autism spectrum disorder (ASD) is a group of heterogeneous neurodevelopmental disabilities characterized by an early onset of impaired social function, abnormal stenosis, and repetitive behaviors and interests (Lord et al., 2018). Worldwide ASD prevalence is approximately 0.8–1.7%, and the ratio of ASD in males:females varies between 2:1 and 4:1 (Mattila et al., 2011; Russell et al., 2014). ASD etiology is incompletely understood, but genetic, environmental, and immune factors may play a part in its pathogenesis (McAllister, 2017; Vorstman et al., 2017). Efficacious drugs for the core symptoms of ASD are lacking. Some antiepileptic and psychotropic drugs have been approved for treating the epilepsy and mood abnormalities associated with ASD (Berry-Kravis et al., 2018).

Behavioral and educational interventions remain the most popular treatment measures for ASD. Early behavioral interventions improve the core symptoms of ASD patients, as well as their verbal and cognitive ability and social adaptability (Lai et al., 2014; Lord et al., 2018). Several genes have been associated with ASD, and many metabolites (e.g., amino acids, organic acids, phospholipids, and purines) show significantly different expression in ASD patients (Vorstman et al., 2017; Likhitweerawong et al., 2021). However, due to heterogeneity among ASD patients, the diagnosis of ASD is reliant on evaluation of the patient’s behavior. Therefore, identifying the characteristic molecular markers may contribute to understanding of the pathophysiological mechanism of ASD, thereby increasing the chance of a correct diagnosis and improvement in therapeutic interventions.

A microRNA (miRNA) is a non-coding RNA with a length of ∼22 nucleotides. miRNAs can inhibit the translation of target messenger (m)RNAs by binding to partially complementary sequences within them (Correia de Sousa et al., 2019). The sequence fragments on the target mRNA that can bind to the miRNA are called “miRNA response elements” (MREs) (Salmena et al., 2011). In 2011, Salmena et al. first proposed the concept of “competitive endogenous RNAs” (ceRNAs), which are different long non-coding-RNAs (lncRNAs), mRNAs, circular RNAs (circRNAs), or pseudogenes with identical MREs that can bind to the same miRNA competitively and interact with each other indirectly. Thus, these RNAs compete with and regulate each other’s expression, with miRNA as the core, forming a complex transcriptional regulatory network (Salmena et al., 2011). For example, as the main type of ceRNA, in a lncRNA–miRNA–mRNA network, lncRNA and mRNA interact indirectly by sharing the same target miRNA. In short, lncRNA can downregulate miRNA expression, whereas miRNA can downregulate mRNA expression; thus, lncRNA can upregulate mRNA expression indirectly. Conversely, mRNA can also increase lncRNA expression. This action helps to coordinate the proliferation, differentiation, and apoptosis of cells (Schmitz et al., 2016). Multiple studies have shown that cancer, ischemic stroke, Parkinson’s disease, and other diseases are related to an abnormal ceRNA network comprising lncRNA–miRNA–mRNA (Long et al., 2019; Zhang et al., 2020; Berti et al., 2021; Li et al., 2021). However, the specific role of a ceRNA network in ASD has not been elucidated.

We aimed to screen the potential diagnostic markers and therapeutic targets of ASD from the perspective of a ceRNA network. We downloaded ASD-related datasets from the Gene Expression Omnibus (GEO) database to screen differentially expressed mRNAs, lncRNAs, and miRNAs in ASD patients to construct a ceRNA network. Based on analyses of differences in the infiltration of immune cells between samples from ASD patients and healthy volunteers (HVs), and combining analyses of the ceRNA network, diagnostic marker genes were obtained by least absolute shrinkage and selection operator (LASSO) regression analyses. Finally, we validated the diagnostic model through validation datasets and samples of peripheral blood from ASD patients (Figure 1).

Since the first report of autism in 1943 by Kanner, understanding of this disease has changed dramatically. Initially, ASD was thought to be caused by inappropriate parenting practices. However, it was discovered later that ASD is a developmental disorder of the central nervous system (CNS), and individuals who carry rare de novo genetic mutations are at a greater risk of developing ASD. Furthermore, it was not until the 1980s that, due to the development of epigenetics, ASD was recognized to be the result of a combination of environmental stimuli in a certain genetic background.

Sequencing has been the most promising strategy to reveal the pathogenesis and diagnosis of ASD (Stefanski et al., 2021). Through an exome-sequencing study of >10,000 ASD patients, Satterstrom et al. found most ASD-associated genes to show differential expression in the early stages of brain development, and that these genes primarily regulated gene transcription, neuronal development, and physiological function (Satterstrom et al., 2020). By taking a single-cell genomics approach, Velmeshev et al. found that differentially expressed genes in ASD patients were concentrated in excitatory neurons and microglia in the cerebral cortex, and were functionally enriched in neuronal projections and adhesions. Furthermore, differential expression of specific genes in cortical–cortical projection neurons correlated with the severity of the clinical manifestations of ASD (Velmeshev et al., 2019). Through whole-exome sequencing, Liu et al. found that the function of the genes associated with gut microbes in ASD patients was mainly enriched in the innate immune response, protein glycosylation, and retrograde axonal transport, which emphasized the role of the brain–gut axis and neuroimmunity in ASD development (Liu et al., 2021). However, even though >1,000 mutations, chromosomal abnormalities, and copy-number variants have been associated with ASD development, each type of variant cannot be used to explain most cases due to the heterogeneity of ASD.

An approach using bioinformatics analysis can be helpful in this scenario. A weighted gene co-expression network was constructed to analyze the genes known to be related to ASD, and to find intersections of these risk genes to discover novel diagnostic and therapeutic options (Parikshak et al., 2015). We attempted to obtain an ASD-related ceRNA regulatory network through bioinformatics analysis. Furthermore, we combined our ceRNA network with differential analyses of immune-cell infiltration to identify ASD biomarkers.

miRNAs are at the core of a ceRNA network. There have been many reports of abnormal expression of multiple miRNAs in the brain tissue and peripheral blood of ASD patients (Xiong et al., 2020). Given that one miRNA can simultaneously regulate expression of multiple genes, and that each gene is also regulated by multiple miRNAs, an “miRNA imbalance” may explain the heterogeneity among ASD patients. Moreover, miRNAs are epigenetic regulatory molecules, so changes in their expression can help unravel the link between heredity and the environment (Masini et al., 2020).

We constructed a ceRNA network comprising 49 lncRNAs, 30 miRNAs, and 236 mRNAs by processing the differentially expressed miRNAs, mRNAs, and lncRNAs obtained from targeting predictions in databases. Through analyses of functional enrichment, we discovered that the top-10 biological processes and molecular functions involved in the genes in this network were mainly related to mitochondrial-energy metabolism, cellular communication, transcriptional regulation, and glial-cell fate, whereas the main signaling pathways involved were associated with energy metabolism and neurodegeneration. These findings are consistent with the underlying pathological changes in ASD: examinations using functional magnetic resonance have demonstrated abnormal activation/inhibition of several brain regions involved in social interaction and information integration in the brains of ASD patients. Moreover, neuropathological examinations have revealed extensive abnormalities in synaptic development/structure and brain function in ASD patients. The brain is a region of high energy metabolism whose neurons are highly dependent on mitochondrial energy metabolism to support neurogenesis, signaling, and synaptic remodeling (Hollis et al., 2017). Furthermore, transcriptional regulation and glial-cell fate can influence neuronal activity and synaptic development, while themselves being highly dependent on mitochondrial energy supply. Hence, mitochondrial dysfunction and abnormal energy metabolism may be a potential common mechanism in ASD pathogenesis.

Studies have demonstrated immune dysregulation, inflammation, and endogenous production of autoantibodies due to abnormalities of innate and adaptive immune systems in ASD patients (Meltzer and Van de Water, 2017). Autoantibodies against neuronal antigens have been detected in children with ASD and their mothers. Moreover, use of immunoglobulin-G from the mothers of mice and rhesus monkeys with ASD has been shown to induce ASD-like symptoms in their offspring. Those findings suggest that the humoral immune system is critical for ASD pathogenesis (Gupta et al., 2010).

We discovered infiltration of fewer iT_regs as well as Th1, Th2, NKT, and CD8⁻ T cells in ASD patients than in HVs. Infiltration of more CD4−initial, central-memory, effector-memory, monocytes, and CD4⁻ T cells was noted, which suggested that cellular immunity was also associated with ASD development. Mostafa et al. showed a reduction in the number of CD4⁺CD25^high T_regs in ASD children (Mostafa et al., 2010). Abdallah et al. reported a reduction in cytokine secretion by Th1 and Th2 cells in children with ASD (Abdallah et al., 2012). Enstrom et al. found a lower responsiveness by NK cells in children with ASD (Enstrom et al., 2009) and an increase in monocyte response (Enstrom et al., 2010). Interestingly, some scholars have reached different conclusions. Gupta et al. reported a reduction in the number of CD8⁺CCR7⁺ CD45RA CD8⁺ central-memory T cells in ASD children (Gupta et al., 2010). Ashwood et al. suggested that increased responses from Th1 cells were associated with the core symptoms of ASD (Ashwood et al., 2011). We noted increased infiltration of Th17 cells (which have been reported to be associated with ASD (Choi et al., 2016)) but the difference was not significant. Thus, even though a heterogeneous pattern of changes in the immune cells of patients with ASD is apparent (which may be related to differences in samples and selection of different HVs), there is little doubt that cellular immunity participates in ASD development, and that clinical application of immunomodulators (e.g., pro-immunoglobulin and IL-6 antagonists) improves (at least in part) ASD symptoms in children.

Given the relevance of the immune system in ASD, we used an established ceRNA network to synthesize the results of an analysis of immune cell infiltration, and we obtained five diagnostic genes using the LASSO algorithm: ABCA1, DGCR2, GFOD1, GLRX, and SEC16A. Analyses of ROC curves revealed that the diagnostic model based on a combination of these five diagnostic genes had an AUC of 0.923. This value is higher than the AUC (0.910) of the three-methylation-markers diagnostic model identified using the same algorithm (Zhang et al., 2022), the AUC (0.910) of the five-gut-bacterial-markers diagnostic model identified by a metagenomic analysis combined with a random forest algorithm (Wan et al., 2022), and the AUC (0.860) of the autism-risk-index diagnostic method based on eye-tracking measures (Frazier et al., 2018), but slightly lower than the AUC (0.947) of the DarkASDNet diagnostic model based on 3D-fMRI (Ahammed et al., 2021). Finally, we confirmed the expression of these five genes in collected serum samples and the validation dataset. The results showed that the five genes had differential expression in ASD patients, which was consistent with the trend predicted by the diagnostic model.

ABCA1 is a transmembrane protein that maintains cellular cholesterol homeostasis and affects the transport of free cholesterol and phospholipids (Brousseau, 2003). Within the nervous system, ABCA1 is expressed mainly in neurons and astrocytes. ABCA1 is associated with apolipoprotein 1-mediated transport of cholesterol and phospholipids, which is involved in the development of Alzheimer’s disease and Parkinson’s disease. ABCA1 is also associated with cardiolipin-driven mitochondrial dysfunction (Jacobo-Albavera et al., 2021). Approximately 23% of the body’s cholesterol is found in the CNS, and it is an important raw material for the growth and differentiation of neurons. Moreover, metabolic dysregulation characterized by reduced phospholipid levels in the brain and plasma has been implicated in ASD pathogenesis (Alfawaz et al., 2018).

DGCR2 is associated with susceptibility to schizophrenia, and is expressed throughout brain development. DGCR2 encodes an activation-dependent adhesion protein involved in regulating the migration and localization of neurons during early corticogenesis (Molinard-Chenu and Dayer, 2018). Belangero et al. found that certain mutations in DGCR2 may also be associated with cortical thickness (Belangero et al., 2019). Abnormal projections of cortex–cortical circuits (Chow et al., 2011) and abnormal cortical growth during infancy (Hazlett et al., 2017) are also present in the brains of ASD patients, which suggests that DGCR2 may be involved in ASD development. However, the exact mechanism must be investigated.

GLRX encodes GLRX, a crucial member of the dithiol-disulfide oxidoreductase family, which is involved in regulation of the transcription factors involved in antioxidative stress and synthesis of control DNA by regulating the activity of ribonucleotide reductase (Chang et al., 2020). Studies have demonstrated increased oxidative stress in the brain of children with ASD, increased levels of glutathione disulfide, and decreased levels of glutathione/glutathione disulfide (reduced/oxidized glutathione) in peripheral blood (Chen et al., 2021). Bowers and colleagues found, through a family-lineage study, that mutations in GLRX (a critical component of the glutathione-GLRX reduction system) resulted in a fourfold increase in ASD risk (Bowers et al., 2011).

Little research has been done on GFOD1, but it shows high expression in the early development of zebrafish brains. GFOD1 may be associated with neurons that produce gamma-aminobutyric acid (Lechermeier et al., 2020). Increased expression of GFOD1 has been noted in the hippocampal tissue of patients with temporal-lobe epilepsy and patients with attention deficit hyperactivity disorder (Lasky-Su et al., 2008; Dixit et al., 2016). Trent et al. found increased expression of GFOD1 in a study of X^Y*O mice (a genetically defective mouse associated with attention deficit hyperactivity disorder and ASD) (Trent et al., 2014), but the exact function of the GFOD1 protein is not known.

SEC16A encodes an endoplasmic-reticulum exit-site scaffolding protein, which is an important component of coat protein complex II (COP II) vesicles and mediates the transport of substances from the endoplasmic reticulum to the Golgi apparatus (Hughes and Stephens, 2010). COPII is involved in metabotropic glutamate-mediated long-range inhibition (Pick et al., 2017), and defective long-term depression of metabotropic glutamate receptors is associated with ASD development (Li et al., 2015). COPII is also involved in the development of Crohn’s disease and obligatory spondylitis (Nakamura et al., 2021). Hence, SEC16A may be involved in ASD development through the immune system.

Rapid advancement of sequencing technology has led to great progress in screening for the diagnostic markers of ASD, but it had two main limitations in our study. First, the sequencing results for mRNAs, lncRNAs, and miRNAs arose from three datasets and belonged to different patients, which brings uncertainty to the diagnosis of ASD. Second, the results of bioinformatics analysis must be validated by clinical and in vitro studies, but the sample size of our validation dataset was small.

We constructed an ASD-associated ceRNA regulatory network. This network will provide a new perspective for explaining the biological processes of ASD. Furthermore, we identified five immune cell-associated potential molecular marker genes of ASD, which could be novel targets for therapy and immunotherapy strategies against ASD.