Whole-transcriptome analysis of serum neuron-derived exosomes reveals specic signature for autism spectrum disorder in Chinese children

Autism spectrum disorder (ASD) is a prevalent developmental disorder that appears in early childhood and manifests repetitive behavioral and social decits. Reported biomarkers for ASD cannot directly and specically reect the abnormality of brain neurons which are preferentially affected and correlated with clinical severity of the disease. Neuron-derived exosomes (NDEs), which carry proteins, lipids and nucleic acids can reect a variety of brain diseases. However, little research has been reported about NDEs in children with ASD. Methods SLC17A6, HTR3A, OSTC, TMEM165, C12orf49, PC-5p-139289_26, and hsa-miR-193a-5p, changed signicantly in the ASD children. Whole-transcriptome analysis of serum NDEs reveals specic signature for ASD in Chinese children, which could serve as predictive biomarkers and provide information for understanding the molecular mechanisms of ASD. Hopefully, our results may provide reference for future diagnostic and therapeutic management of the disease.

In the past decade, searching for genetic biomarkers has been a hot spot in ASD research. Numerous related genes have been reported, including NRXN1, SHANK3, SHANK2, MECP2, SNC2A, CHD8, DYRKIA, POG2, GRIN2B, KATNAL2, NLGN3, NLGN4, CNTN4, CDH10, CDH9, and SEMA5A [4][5][6]. Unfortunately, only 10-38% of ASD cases have been reported with known genetic de cits [7,8]. In recent years, blood/serum biomarkers have drawn much attention due to their accessibility, low cost and rapid detection. In our previous study, we identi ed four candidate peptides as biomarkers, namely SerpinA5-A, ApoC1-A, FABP1-A and PF4-A [9], and α2-3-linked sialylation of apolipoprotein D (APOD) [10] as potential biomarkers for ASD. A recent study also showed that SLC25A12, LIMK1, and RARS might serve as potential blood protein biomarkers for ASD [11]. However, the biomarkers that have been reported cannot speci cally re ect the abnormality of brain neurons which are preferentially affected in autism and the dysregulation of speci c genes in neurons correlated with clinical severity [12,13].
Exosomes are small, single-membrane, secreted extracellular vesicles of ∼30 to ∼200 nm in diameter that have the same topology as the cell and are enriched in selected proteins, lipids, nucleic acids and glycoconjugates. In the central nervous system, almost all types of cells secrete exosomes, which mediate neuron-glial cell communication, promote neuronal repair and growth, and promote the progression of glioblastoma and neurological diseases [14]. Exosomes are very stable in terms of constitution and protect the "biological cargo" they carry from degradation and denaturation in the extracellular environment. Compared with biological uids, such as cerebrospinal uid, blood or urine, exosomes can provide more reliable and accurate biomarkers for neurological diseases [14][15][16]. More importantly, they can cross the blood-brain barrier and have low immunogenicity. Overall, they are a promising source for biomarkers and ideal vehicles for drug delivery, which might be widely used in the diagnosis and treatment of neurological diseases [16]. Currently, it is found that secreted extracellular vesicles increase in the serum of children with ASD and contain IL-1β that stimulates secretion of human microglia cells [17]. Mesenchymal stem cell-derived exosomes can improve autism-like behavior in BTBR mice and may be a cell-free therapeutic tool for ASD [18,19]. These ndings uncover important roles of exosomes, suggesting the necessity of characterizing the detailed molecular status of brain-derived exosomes in ASD.
Recent studies have shown that the surface of exosomes derived from neurons carries neural cell adhesion molecule L1 (L1CAM) that can be utilized to isolate the exosomes from serum/plasma; proteins in neuron-derived exosomes (NDEs) can re ect brain injury, progression from acute mild traumatic brain injury to chronic traumatic brain disease, cognitive dysfunction caused by HIV infection, and neurological abnormalities such as Alzheimer's disease [20,21]. However, the expression of proteins or RNAs in NDEs of children with ASD is rarely reported.
It is well known that exosomal miRNAs can be used as potential diagnostic and prognostic biomarkers as well as therapeutic tools for a variety of neuropsychiatric diseases, such as dementia, Alzheimer's disease, depression, and schizophrenia [22]. In light of this, the present study was conducted to examine the expression of RNAs in NDEs from children with ASD and to reveal the possible mechanisms of the disease. We collected serum samples from 100 ASD children and 60 age-matched typically developed (TD) children, and pooled the samples into groups (n = 20). NDEs of the pooled sera in each group were isolated using L1CAM antibody mediated immunosorbent assay [20,21] and were characterized by nanoparticle tracking analysis, transmission electron microscopy and western blot. Whole-transcriptome of the NDEs was analyzed by lncRNA microarray and RNA-Sequencing. RNAs expressed differently in NDEs from ASD sera as versus those from TD sera were screened, analyzed, and validated. In brief, a total of 1418 mRNAs, 1745 lncRNAs and 11 miRNAs were found differentially expressed. Most of these RNAs were down-regulated in ASD and enriched in neuron-related and glycan-related networks associated with ASD. Levels of some potential markers were found signi cantly changed in ASD.

Study Approval
Approval for this research was obtained from the Ethics Committee of Xi'an Jiaotong University (Xi'an, China). All parents of the participants signed written informed consent. The experiments were carried out in accordance with the ethical guidelines of the Declaration of Helsinki.

Subjects
The study enrolled 100 children with ASD (between 2.5 and 6 years of age; 90 males) and 60 agematched TD children (54 males) as control. The ASD children were recruited from Xi'an Children's Hospital, Xi'an, China. The healthy children were recruited from the same region to minimize the in uence of different environments. The ASD children were examined by a developmental behavioral pediatrician and a pediatric neurologist or psychiatrist. All the consultants agreed on the diagnosis of ASD according to DSM-V criteria. Children with tuberous sclerosis complex, Rett syndrome, Prader Willi syndrome, Angelman syndrome, or Fragile X syndrome were excluded. All the participants were screened via a parental interview for current and past physical illnesses. Those who had any type of infection or disease within two weeks before the time of examination were excluded. ASD was evaluated with the autism diagnostic observation schedule (Table 1).
Collection and preparation of serum samples Venous blood samples were collected by a pediatric nurse. The blood was allowed to clot at room temperature for 30 min, and the clot was then removed by centrifuging at 1,500×g for 10 minutes. The resulting supernatant is immediately transferred to a clean polypropylene tube, and EDTA-free inhibitor cocktail (Halt protease inhibitor; Thermo Scienti c Pierce Protein Research Products, Rockford, IL, USA) was added at a concentration of 10 μ L/mL serum. The obtained serum was aliquoted into small portions and was immediately frozen on dry ice and stored at −80 °C. To tolerate individual variation, 25 μL of each serum sample was collected and every 20 samples were pooled into one subgroup. Altogether, we got 5 ASD subgroups and 3 TD subgroups (n=20). To avoid bias caused by gender difference, proportion of males in each subgroup was the same (90%). The remaining serum in each sample was maintained for further individual validation.
Isolation of serum neuron-derived exosomes (NDEs) NDEs in the serum samples were isolated as described previously [20,21] with minor modi cations.
Brie y, 0.5 ml of serum was incubated with 0.15 ml of thromboplastin-D (Fisher Scienti c, Inc., Hanover Park, IL) at room temperature for 60 min. Then 0.5 ml of calcium-and magnesium-free Dulbecco's balanced salt solution (DBS−2) with protease inhibitor cocktail (Roche Applied Sciences, Inc., Indianapolis, IN) and phosphatase inhibitor cocktail (Pierce Halt, Thermo Scienti c, Inc., Rockford, IL) was added and the mixture was centrifuged at 1,500×g for 20 min. The supernatant was mixed with 252 μl of ExoQuick exosome precipitation solution (EXOQ; System Biosciences, Inc., Mountainview, CA), and incubated for 1 hr at 4°C. The resultant exosome suspension was centrifuged at 1,500×g for 30 min at 4°C and the pellet was re-suspended in 150 μl of DBS−2 with the inhibitor cocktails before immunochemical enrichment of NDEs. Each sample received 100 μL of 3% BSA (1:3.33 dilution of Blocker BSA 10% solution in DBS−2 [Thermo Scienti c, Inc.]) and was incubated with 1 μg of mouse antihuman CD171 (L1CAM neural adhesion protein) biotinylated antibody (clone 5G3, eBioscience, San Diego, CA) for 1 hr at 4°C. Following that, 25 μl of streptavidin-agarose resin (Thermo Scienti c, Inc.) plus 50 μL of 3% BSA was added and the sample was incubated at 4°C for 30 min. After centrifugation at 200×g for 10 min at 4°C and removal of the supernate, 3% BSA was added again, and centrifugation and supernatant removal were repeated. Each pellet was suspended in 50 μl of 0.05 M glycine-HCl (pH 3.0), incubated at 4°C for 10 min, and re-centrifuged at 4,000×g for 10 min at 4°C. The obtained supernatant was transferred to a new Eppendorf tube containing 5 μL of 1 M Tris-HCl (pH 8.0) and stored at −80°C.

Nanoparticle tracking analysis (NTA)
NDEs suspension at a concentration between 1 x 10 7 /ml and 1 x 10 9 /ml was examined using a Nanosight NS300 (NanoSight Ltd., Amesbury, UK), equipped with a 405 nm laser to determine the size and quantity of particles isolated.

Transmission electron microscopy (TEM)
NDEs solution (20-40 μl) was placed on a copper mesh, post-negatively stained with 2% phosphotungstic acid for 10 min, and then dried for 2 min under incandescent light. The copper mesh was observed and photographed under a transmission electron microscope (H-7650 Hitachi microscope; Hitachi, Tokyo, Japan).

Extraction of total RNA in NDEs
Total RNA in NDEs was isolated using the Exosomal RNA isolation kit (Norgen Biotek, 58000) according to the manufacturer's instructions. Brie y, 200 μl of the transferred supernatant containing puri ed NDEs was incubated with 300 uL Lysis Buffer A and 37.5 uL Lysis Additive B at room temperature for 10 min, following which 500 uL of 96-100% Ethanol was added to the mixture and mixed well via 10-second vortexing. Then, 500 uL of the mixture was transferred into a Mini Spin column and centrifuged at 3,000×g for 1 min, and the remaining mixutre was transfered and centrifuged by repeating the steps. After that, 600 uL Wash Solution A was applied and the column was centrifuged at 3,300×g for 30 seconds twice.The spin column was then moved to a fresh 1.7 mL Elution tube, and 50 uL Elution Solution A was added. Finally, centrifugation was performed at 400×g for 1 min and 5,800×g for 2 min to obtain total RNA.
Human lncRNA microarray and data analysis Total RNA was puri ed using a RNeasy Mini Kit (Qiagen, Germany) and was checked for a RIN number to inspect RNA integration with an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, US). LC Biotech Human lncRNA Microarray 4×180 K (Agilent Technologies; Santa Clara, CA) was utilized to detect the expression of mRNAs and lncRNAs in NDEs. The microarray slide contains 26,083 mRNA probes and 1,05,135 lncRNA probes, and lncRNA sequencing data are available from Gencode, UCSC, Ensembl, Refseq, LNCIpedia, NONCODE, LNcRNA Disease, Ernas, NRED and other databases. Ampli cation of cRNA, uorescent labeling and hybridization of the microarray were performed by following the protocol of Agilent Technologies. Brie y, equal amount of RNA from each subgroup was reversely transcribed into cDNA, which was then labeled with Cy3 (GE Healthcare; Biosciences, Piscataway, NJ, USA) and hybridized with the microarray slide. After that, the slide was scanned on the Agilent Microarray Scanner G5761A (Agilent Technologies). Data were extracted with Feature Extraction software 12.0.3.1 (Agilent Technologies), and raw data were normalized by Quantile algorithm. Genes with p value<0.05 and a fold change of at least 2 were selected for further analysis. GO/KEGG pathway enrichment analyses of the target genes were performed using Fisher's exact test. The function of lncRNAs was predicted by analyzing the functional annotations of mRNAs that were highly co-expressed with lncRNAs.
Small RNA library construction, sequencing and data processing Approximately 1 ug total RNA was used to prepare small RNA library according to the protocol of TruSeq Small RNA Sample Prep Kit (Illumina, San Diego, USA). Single-end sequencing (36 bp) was performed with an Illumina Hiseq2500 at LC-BIO (Hangzhou, China). Brie y, the raw reads were subjected to the Illumina pipeline lter (Solexa 0.3), and the dataset was further processed with an in-house program, ACGT101-miR (LC Sciences, Houston, Texas, USA), to remove adapter dimers, junk, low complexity, common RNA families (rRNA, tRNA, snRNA, snoRNA) and repeats. Subsequently, unique sequences with a length of 18~26 nucleotides were mapped to Homo species precursors in miRBase 20.0 by BLAST search to identify known miRNAs and novel 3p-and 5p-derived miRNAs. The hairpin RNA structures containing the sequences were predicated from the ank 80 snt sequences using RNAfold software (http://rna.tbi.univie.ac. at/cgi-bin/RNAfold.cgi). miRNA differential expression based on normalized deep-sequencing counts was analyzed using the Fisher exact test and Student t test, and the signi cance threshold was set to be 0.01 or 0.05. To predict the genes targeted by most abundant miRNAs, two computational target prediction algorithms (TargetScan 50 and miRanda 3.3a ) were used to identify miRNA binding sites. Finally, the data predicted by both algorithms were combined and the overlaps were calculated. The GO terms and KEGG pathways of these most abundant miRNAs, miRNA targets were also annotated.
Quantitative real-time PCR All primers were designed and synthesized by Takara (TakaraBiotechnology, Dalian, China). To avoid false-positive ampli cation of contaminating genomic DNA in the mRNA samples, all the primers spanning different exons were designed (Table 2). For mRNA, cDNA was synthesized using a PrimeScript RT reagent kit (Takara Biotechnology Co, Ltd, Dalian, China). Quantitative real-time PCR (qRT-PCR) was performed using the IQ5 real-time PCR detection system, and GADPH was taken as a control. Relative quanti cation of mRNA expression levels was performed using SYBR Premix Ex Taq II on an FTC-3000TM System (Funglyn Biotech Inc., Toronto, Canada). For miRNA cDNA was synthesized by servicebio RT First strand cDNA Synthesis Kit (Servicebio, Wuhan, China), qRT-PCR was carried out using the SYBR Premix Ex Taq TM II (TaKaRa) and U6 was taken as a control. PCR conditions consisted of a 5 min preincubation at 95•C, followed by 40 repeats of 95°C for 10 s and 60°C for 20 s. All samples were run in triplicate and the average values were calculated. The relative levels of mRNAs EDNRA, SLC17A6, HTR3A, OSTC and TMEM165, as well as miRNAs PC-5p-139289_26 and hsa-miR-193a-5p were calculated using the 2 −ΔΔCt method.

Statistics
All statistical analyses were performed using the software SPSS (version 17) and group statistics are presented as mean ± SD. The t-test for independent variables was used to examine the inter-group differences, and a signi cance level of 0.05 was adopted.

Characterization of serum NDEs
Basic characteristics of the participants were shown in Table 1. NDEs in the pooled sera in the ASD group (including 5 subgroups) and the TD group (including 3 bsugroups) were isolated using L1CAM antibody mediated immunoadsorption (Fig. 1). Nanoparticle tracking analysis showed a higher NDE concentration in the ASD group (2.04 ± 4.35 × 10 10 /ml) than in the TD group (1.20 ± 3.28 ×10 10 /ml). The average particle size of NDEs was 61.50 ± 20.71 nm in the ASD group and 62.07 ± 20.75 nm in the TD group, showing no signi cant difference ( Fig. 2A). Under TEM, both groups of NDEs presented a "saucer" like structure (Fig. 2B). Meanwhile, compared with serum, the obtained NDEs solution was rich in exosomalspeci c marker proteins CD63 and CD81 (Fig. 2C). L1CAM expression was higher in NDEs than in total serum exosomes (Fig. 2D).

Differential expression and bioinformatic analysis of mRNAs in serum NDEs of ASD
Based on lncRNA microarray detection and original data normalization, air-wise Pearson's correlation coe cients of all RNAs among subgroups were shown in Fig. 3A. The coe cients between the biological replicates (subgroups) within each group were obviously higher than those between the two groups ( Fig. 3A). mRNAs and lncRNAs with at least twofold differential expression and a P value of less than 0.05 were subjected to further examination. This resulted in 167 up-regulated and 1251 down-regulated mRNAs in ASD serum NDEs ( Fig. 3B and Supplemental Table 1). Hierarchical clustering analysis (HCA) of these 1418 differentially expressed mRNAs (DEmRs) showed similar expression pro les among the biological replicates within each group but differential pro les between the two groups ( Fig. 3C). Principal component analysis (PCA) of the DEmRs showed that the two groups were separated and the biological replicates in the TD group clustered more closely than those in the ASD group (Fig. 3D). To characterize the distribution of genes for DEmRs on chromosomes and to reveal the susceptible chromosomes, the genes on each chromosome were counted and the ratio of the number of such genes to the total number of genes present on the chromosome (data from Human Genome Resources at NCBI, GRCh37) was calculated. As a result, chromosomes 1 and 2 had the largest number of genes for DEmRs, while chromosomes 21 and Y had the least number of such genes. However, genes with the maximum ratio were on chromosomes 17 and 5, and those with the minimum ratio were on chromosomes 13 and Y (Fig. 3E). GO annotation of DEmRs showed that their products were mainly distributed in cytoplasm, nucleus and plasma membrane; were able to bind with proteins, metal ions and DNA; and were involved mainly in DNA-dependent transcription, small molecule metabolism, transcriptional regulation and other biological processes (Fig. 3F). KEGG analysis revealed that DEmRs participated in mainly three types of processes: (1) the signal transduction processes, such as MAPK signaling pathway, calcium signaling pathway, PI3K-Akt signaling pathway and cAMP signaling pathway; (2) neuron-related pathways such as neuroactive reception-ligand interaction, axon guidance and synaptic vesicle circulation; and (3) glycosylation related pathways such as N-glycan biosynthesis, endoplasmic reticulum protein processing, sugar binding and glycosaminoglycan degradation (Fig. 3G).
Differential expression and functional prediction of lncRNAs in serum NDEs of ASD According to the results of lncRNA microarray, 239 lncRNAs were signi cantly up-regulated and 1506 lncRNAs were signi cantly down-regulated in ASD serum NDEs (Fig. 4A and Supplemental Table 2). HCA of these 1745 differentially expressed lncRNAs (DElnRs) showed similar expression pro les among the biological replicates within each group but differential pro les between the two groups (Fig. 4B). PCA of the DElnRs separated the subgroups into TD and ASD groups as their natural grouping (Fig. 4C). To characterize the distribution of DElnRs on chromosomes and to reveal the susceptible chromosomes, the DElnRs on each chromosome were counted and the ratio of the number of DElnRs to the length of that chromosome was calculated (data from Human Genome Resources at NCBI, GRCh37; the unit of length is Mbp). It was found that chromosomes 1 and 2 had the largest number of DElnRs and chromosomes 21 and Y had the least number of DElnRs; whereas, chromosomes 19 and 17 had the maximum ratio and chromosomes X and Y had the minimum ratio (Fig. 4D). Genes with a distance of less than 100kb from lncRNA were regarded as the target genes for cis-acting. As a result, 382 DElnRs were predicted to be positively or negatively correlated with their target genes (R > 0.8) (Supplemental Table 3). Of these genes, 107 were also DEmRs ( Fig. 4E and supplemental Table 3). Double-omic analysis (https://www.omicstudio.cn/tool) of the 107 pairs of DElnR-DEmR revealed that 81.3% pairs were positively correlated and commonly down-regulated in ASD (Fig. 4F). GO annotation showed that these 107 target genes were mainly cytoskeleton-related proteins (Fig. 4G) with functions such as auxiliary transport protein activity, protein binding and translation regulation (Fig. 4H), and were involved in protein metabolism, transport, and cell growth and/or maintenance (Fig. 4I).

Expression of miRNA in serum NDEs
A total of 4310 mature microRNAs (miRNAs) were examined in the serum NDEs, among which 150 were present in all subgroups (Supplemental Table 4). The miRNA with the highest concentration across all the subgroups was hsa-miR-21-5p_R + 1, accounting for 6.9% in TD and 8.5% in ASD in terms of the normalized read counts (Fig. 7A). Unpaired two-tailed Student's t test identi ed 10 miRNA sequences with signi cant differences (p < 0.05) between ASD and TD (Fig. 7B). The sequence GATTTCTTCCCAGTGCTCTGA was aligned to two pre-miRNAs and was given two names: mmu-mir-6240-p3_1ss8GT and mmu-mir-6240-p5_1ss8GT. Of these 11 miRNA, the one with the biggest variation was PC-3p-38497_124, which was remarkably up-regulated (fold change = 20.32, p = 0.029) in ASD relative to TD (Fig. 7B). The miRNA with the most signi cant change was PC-5p-139289_26, which was absent in the TD subgroups (p = 0.0056) (Fig. 7B). Two other miRNAs (PC-3p-275123_15 and PC-5p-149427_24) were up-regulated in ASD, and another seven miRNAs (e.g., hsa-miR-193a-5p and mmu-mir-6240-p3_1ss8GT) were relatively down-regulated (Fig. 7B). HCA of these differentially expressed miRNAs (DEmiRs) revealed similar expression pro les among the biological replicates within each group but differential pro les between the ASD and TD groups (Fig. 7C). PCA showed that the biological replicates in the TD group clustered more closely (Fig. 7D).

Discussion
At present, the diagnosis of ASD is still based on symptom evaluation, as the underlying pathological mechanism is still unclear. There are no blood-based diagnostic tools or approved drugs for ASD. Research that identi es reliable biological markers of disease status and symptomology in ASD is therefore urgently needed. Neurobiological systems critical to social functioning are arguably the most promising biological sources for ASD biomarkers and therapeutic targets. However, existing methods for brain detection have mostly relied on autopsy or animal models, which are limited because of poor timeliness and species differences. Most cells in the nervous system, including neurons, astrocytes, oligodendrocytes and microglia, secrete exosomes under normal or pathological conditions. Exosomes can re ect the host cell proteins and nucleic acids at the time of secretion and can diffuse across the blood brain barrier into the periphery. Neuron-derived exosomes (NDEs) can be captured by antibodies directed against the cell surface proteins embedded in the vesicle membrane [20,21]. Although investigation of NDEs is relatively novel, attractive evidence from other elds suggests that such investigation can afford insight into the pathological mechanisms and processes associated with Alzheimer's Disease and depressive disorder [25,26]. In the present study, serum NDEs from ASD as well as TD children were collected and characterized scrupulously. By a whole-transcriptome analysis, we screened 1418 mRNAs, 1745 lnRNAs and 11 miRNAs differentially expressed in ASD as against TD. This was validated by examining several candidate RNAs in individual samples. These validated RNAs might be potential biomarkers for ASD diagnosis, which could re ect important molecular events in neurons timely and noninvasively.
Thus far, a putative speech and language region at chromosome 7q31-q33 seems most strongly linked to autism. Cytogenetic abnormalities at the 15q11-q13 locus are fairly frequent in people with autism, and a "chromosome 15 phenotype" is described in individuals with chromosome 15 duplications [27]. Some candidate genes are considered located at chromosomes 7q22-q33 and 15q11-q13 [28], and 21 genes in chromosomal 8p region are identi ed as most likely to contribute to neuropsychiatric disorders and neurodegenerative disorders [29]. Variant alleles of the serotonin transporter gene (5-HTT) on chromosomes 17q11-q12 are more frequent in individuals with autism than in healthy people [28]. In addition, many mutations on NLGN4X, an X-linked cell adhesion molecule, result in ASD [30]. In the present study, chromosome 17 was the commonly and mostly enriched chromosome for both DEmRs and DElnRs in ASD. A high portion of the DEmRs on chromosome 17 participated in cell communication and signal transduction, which are essential for synapse formation and neurotransmitter release.
Abnormal expression of such mRNAs implies the abnormality of these functions in ASD.
Brain-derived exosomes carry and release multiple molecules related to neuronal function and neurotransmission in the brain, which is bene cial for the reciprocal communication between neural cells (e.g., neuron − glia interactions), synaptic plasticity, neuronal development, and neuroimmune communication. In the present study, of the mRNAs differentially expressed in serum NDEs from ASD and TP children, 104 DEmRs (7.3%) were annotated to be related to neuroactive ligand-receptor interaction, pathways of neurodegeneration, glutamatergic synapse, axon guidance, synaptic vesicle cycle, dendrite, neuron projection development, neuron migration and apoptotic process. Most (81.7%) of these neuronrelated mRNAs were down-regulated in ASD. As demonstrated in the pathway of neuroactive ligandreceptor interaction (Fig. 3H), 5 receptors (e.g., EDNRA) were up-regulated and 19 (e.g., HTR3A) were down-regulated in ASD. A previous study reported that neuropeptide receptor gene expression was lower in children with autism and the lower neuropeptide receptor gene expression predicted greater social impairments and greater stereotyped behaviors [31]. In the present study, 5-hydroxytryptamine receptor 3A (HTR3A) signi cantly decreased in the ASD serum NDEs. HTR3A is one of the receptors for 5hydroxytryptamine (serotonin), a biogenic hormone that also functions as a neurotransmitter and a mitogen. Ample evidence suggests that levels of serotonin and serotonin transporter (SERT) increase signi cantly in autistic children than in gender and age-matched controls [32,33]. It thus can be hypothesized that increase of serotonin and SERT may be a kind of cell self-help that compensates for the loss of receptors, but it needs to be experimentally con rmed in the future. Another speci c signature is the decreased expression of vesicular glutamate transporter 2 (SLC17A6) in the ASD serum NDEs.
Receptors for glutamate (Glu), GRIK5, GRIK2 and GRIA4 were also down-regulated. Glu acts as an excitatory neurotransmitter at many synapses in the central nervous system. SLC17A6 mediates the uptake of Glu into synaptic vesicles at presynaptic nerve terminals of excitatory neural cells. The postsynaptic actions of Glu are mediated by a variety of receptors expressed on postsynaptic cell membrane. Emerging evidence suggests that imbalance between excitatory (Glu-mediated) and inhibitory (GABA-mediated) neurotransmission may be a common pathophysiological mechanism in ASD [34,35]. These studies, together with the ndings in the present study, suggest that reduction in the expression of Glu transporter and receptors might be the main reason for the abnormalities of Glu-mediated neurotransmission and hence a therapeutic target in ASD.
Glycans and their conjugates (glycoproteins, proteoglycans and glycolipids) are major constituents of the neural cell membrane and extracellular matrix (ECM). Glycans and glycoconjugates participate in nearly every biological process in the developing brain. A potential link between ASD and changes in glycosylation was rst observed in patients with congenital glycosylation disorders (CDGs) [36]. Recent advances in genome sequencing have identi ed many genetic variants that occur in genes encoding glycosylated proteins (proteoglycans or glycoproteins) or enzymes involved in glycosylation (glycosyltransferases and sulfotransferases) [37,38]. However, it remains unknown whether "glycogene" variants cause changes in glycosylation and whether they contribute to the etiology and pathogenesis of ASDs. In the present study, we analyzed the whole transcriptome of serum NDEs in ASD to screen potential biomarkers and explore the important molecular events in brain neurons of ASD children. Our results showed that a total of 54 DEmRs (3.8%) were glycogenes, and most of them (90.7%) were downregulated in ASD. The 54 DEmRs mainly participated in carbohydrate metabolic process, protein N-linked glycosylation, carbohydrate binding, glycolysis, glycosaminoglycan metabolic process and glycolipid metabolic process. Thereinto, OSTC, MAN1B1 and MGAT5, translating to key enzymes for N-linked glycosylation, were signi cantly down-regulated in ASD. In our previous study, we found a signi cant decrease of STL binding glycans or glycoproteins that contain trimers and tetramers of GlcNAc (core structure of N-glycans) in ASD versus in TD (fold change = 0.54, p = 0.0057) [10]. In all, no matter at the gene level, the transcription level, or the level of translation and post-translation modi cation, abnormalities of glycosylation and carbohydrate metabolism might be an important molecular mechanism of ASD. Moreover, the decrease of receptors and transporters of neurotransmitters may be related with the decrease of glycogenes as most of the receptors and transporters are highly glycosylated. OSTC is a subunit of the oligosaccharyl transferase (OST) complex that catalyzes the initial transfer of a de ned glycan (Glc3Man9GlcNAc2 in eukaryotes) from the lipid carrier dolicholpyrophosphate to an asparagine residue within an Asn-X-Ser/Thr consensus motif in nascent polypeptide chains. In the present study, expression of OSTC signi cantly decreased in ASD serum NDEs, suggesting it as a candidate biomarker for ASD diagnosis.
Recent studies have shown that abnormal expression of miRNAs could be invovled in the underlying pathogensis of ASD. miRNAs are small noncoding mRNAs that regulate gene expression and are often linked to biological processes and implicated in neurodevelopment. A dozen of miRNAs, such as miRNA-125b and miRNA-132, have been observed to regulate the expression of ASD risk genes, act differently on the morphology of the spine and synaptic plasticity in brain neurons, and participate in ASD etiopathogenesis [39]. However, compared with mRNA and lncRNA, fewer miRNAs were found differentially expressed in ASD serum NDEs in the present study. Among 11 DEmiRs, PC-5p-139289_26 was signi cantly up-regulated and hsa-miR-193a-5p was signi cantly down-regulated in ASD, and both of them had the largest number of predicted targets that were differentially expressed in ASD, indicating that these two miRNAs might play important roles in ASD. These targets were mostly involved in glutathione synthesis and recycling and mannosyltransferase activity, which are closely correlated with synthesis of Glu and glycans invovled in the neuron-and glycan-related networks in ASD. However, the relationships between miRNAs and their target genes have not yet been veri ed.

Limitations
This study might have some limitations that merit consideration. Firstly, we did not examine the correlation between the expression of candidate biomarkers and disease severity. This would be addressed in our future research. Secondly, we collected only one blood sample per participant (due to the invasive nature of venipuncture, particularly in children), which limited our ability to assess withinindividual consistency of our biological measures. Thirdly, some of our participants were not medicationfree. Even though their medications were stable (for at least four weeks) before blood collection, it is possible that our results might be in uenced by the medication status. Fourthly, our samples were mainly from a single hospital. Although it is one of the few famous hospitals in the northwest of China that treat ASD children from ve neighboring provinces, most of its patients are still from the local regions. Further research involving participants from multiple areas would be a great addition to the present study.

Conclusions
In short, 1418 mRNAs, 1745 lncRNAs and 11 miRNAs were identi ed differentially expressed in the serum NDEs from the ASD children versus from the TD children. Most of these RNAs were down-regulated and involved in neuron-related and glycan-related networks implicated in ASD. Levels of potential markers, including EDNRA, SLC17A6, HTR3A, OSTC, TMEM165, C12orf49, PC-5p-139289_26, and hsa-miR-193a-5p, changed signi cantly in the ASD children. Whole-transcriptome analysis of serum NDEs reveals speci c signature for ASD in Chinese children, which could serve as predictive biomarkers and provide information for understanding the molecular mechanisms of ASD. Hopefully, our results may provide reference for future diagnostic and therapeutic management of the disease.   Figure 1 Schematic ow diagram of the integrated strategy used herein.