Transcriptomic Signatures of Mitochondrial Dysfunction in Autism: Integrated mRNA and microRNA Profiling

1. Introduction

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition defined by abnormalities in social communication along with restricted and repetitive behaviors and interests [1]. ASD affects one in 31 children in the United States (US) and has a four times higher prevalence in males than in females [2]. ASD is defined by behavioral features, but ASD demonstrate brain-based and/or systematic physiological abnormalities, highlighting the biological underpinnings [3]. These biological abnormalities have been leveraged to develop objective biomarkers, although no diagnostic biomarker is mature enough to utilize in clinic practice [4].

Many individuals with ASD manifest systemic metabolic and mitochondrial disorders, immune system abnormalities and oxidative stress [5]. Metabolic disorders are particularly important since many of these disorders can be mitigated with safe, well tolerated treatments [6,7]. Understanding these underlying disorders can assist in the development of objective biomarkers to assist with diagnosis, identify subgroups and screen for the best candidates for the most effective treatment [5].

Disorders of mitochondrial function may affect a substantial number of individuals with ASD [8]. However, the interpretation of these mitochondrial disorders is complicated as most are non-classical in nature. Indeed, a meta-analysis found that the overall prevalence of classical mitochondrial disease in ASD is 5% but about 30% or more of those with ASD manifest biomarkers of abnormal mitochondrial function [9]. Perhaps more striking, some studies find that 80% of immune cells from children with ASD have electron transport chain (ETC) complex deficits [10,11]. Furthermore, ASD has been linked to increased ETC complex activity [12,13,14,15] and mitochondrial respiration [16,17,18,19,20,21,22,23], whereas classic mitochondrial disease is usually defined by a significant reduction in ETC activity and respiration. Thus, individuals with ASD appear to have unique abnormalities in mitochondrial metabolism which has recently been linked to

Using the Seahorse 96 XF high-throughput respiratory analyzer our group has studied mitochondrial respiration in control and ASD lymphoblastic cell lines (LCLs). It was found that the LCLs from about one-third of children with ASD manifested elevated respiratory rates (about 200% of controls) and were more vulnerable to physiological stress. An important respiratory parameter, reserve capacity (RC), a measure of mitochondrial health, quickly plummeted in these cells when challenged with physiological stress as compared to other cell lines [20]. These physiologically vulnerable cell lines are called AD-A (ASD with abnormal mitochondrial function). Mitochondria from the remaining ASD LCLs were essential equivalent to controls and then are called AD-N (ASD with normal mitochondrial function). We have replicated these results in LCLs in seven additional studies [17,18,20,21,22,23,24] and have demonstrated that this pattern of respiratory abnormalities is also manifested in fresh in peripheral blood mononuclear cells (PBMCs) from individuals with ASD, especially children with neurodevelopmental regression [25,26].

The modulator mechanisms in the AD-A LCLs have been investigated. In one study we found the AD-A LCLs failed to upregulate PINK1, MFN2, SIRT1, SIRT3 DNM1L, HIF1z and PGC1a as compared to the AD-N LCLs. We also found that this increase in mitochondria respiration could be mitigated by rapamycin, implicating the mTOR pathway. To obtain further insight into the molecular mechanisms which drive the mitochondrial phenotype in the AD-A LCLs we use RNAseq mRNA and miRNA profiling in the LCLs and examine whether the differentially expressed (DE) genes are related to the mTOR or cellular and mitochondrial stress pathways previously identified.

2. Materials and Methods

2.1. Materials

RPMI 1640 culture media, penicillin/streptomycin, fetal bovine serum (FBS), phos-phate buffered saline (PBS), BCA Protein Assay Kit were all obtained from Thermo Fisher Scientific (Waltham, MA, USA). XF DMEM and XF-PS 96-well plates were obtained from Agilent Technologies (Santa Clara, CA, USA). The RNeasy mini kit was obtained from Qiagen (Hilden, Germany) and the High Capacity cDNA Reverse Transcription Kit and Power SYBR Green PCR Master Mix from Applied Biosystems (Waltham MA, USA). Poly-D-Lysine, 2,3-dimethoxy-1,4-napthoquinone (DMNQ), meta-phosphoric acid and all other chemicals were obtained from Sigma-Aldrich (St. Louis, MO, USA).

2.2. Cell Lines and Culture

LCLs were obtained from the Autism Genetic Resource Exchange (AGRE; Los Angeles, CA, USA). Details of the LCLs are presented in Table 1. Age was not significantly difference across groups as measured by paired t-test. LCLs were maintained in RPMI 1640 culture medium with 15% FBS and 1% penicillin/streptomycin in a humidified incubator at 37 °C with 5% CO2. All ASD LCLs were linked to the results of the gold-standard Autism Diagnostic Observation Schedule (ADOS) assessments of the children from which the LCLs were derived. These LCLs were used in a previous study examining bioenergetics [18].

2.3. Library Construction and Sequencing

Total RNA was extracted using Trizol reagent (Invitrogen, CA, USA), and sent to LC Sciences (Houston, Texas, USA) on dry ice. RNA integrity and quality were determined by Bioanalyzer 2100 (Agilent, CA, USA) with RIN number >7.0. Approximately, 1 ug of total RNA were used to prepare small RNA library according to protocol of TruSeq Small RNA Sample Prep Kits (Illumina, San Diego, USA). Single-end sequencing of 50bp was per-formed on an Illumina Hiseq 4000 at LC Sciences following the vendor’s recommended protocol [27]. Approximately 10 ug of total RNA was subjected to isolated Poly (A) mRNA with poly-T oligo attached magnetic beads (ThermoFisher, Carlsbad, CA, USA). Following purification, the poly(A)- or poly(A)+ RNA fractions were fragmented into small pieces using divalent cations under an elevated temperature. The cleaved RNA fragments were reverse transcribed to create the final cDNA library in accordance with the protocol for the mRNA-seq sample preparation kit (Illumina, San Diego, CA, USA) and the average insert size for the paired-end libraries was 300 bp (50 bp). Finally, paired-end sequencing was performed on an Illumina Hiseq 4000 following the vendor’s recommended protocol.

2.4. Bioinformatics Analysis

For miRNA Raw reads were processed with ACGT101-miR (LC Sciences, Houston, Texas, USA) to remove adapter dimers, junk, low complexity reads, common RNA families (rRNA, tRNA, snRNA, and snoRNA) and repeats. Unique sequences with lengths of 18–26 nucleotides were mapped to specific-species precursors obtained from the miRBase 21.0 by a BLAST search performed to identify known miRNAs and novel 3p- and 5p-derived miRNAs. The remaining sequences were aligned against the miRbase (Release 21) (https:// www.miRbase.org/) miRNA database, and perfectly matched sequences were considered to be conserved Homo sapiens miRNAs [27].

For mRNA, Raw data (raw reads) of fastq format were firstl processed through in-house perl scripts. In this step, clean data (clean reads) were obtained by removing reads containing adapter, reads containing ploy-N, and low quality reads from the raw data. At the same time, Q20, Q30 and GC content the clean data were calculated. All the downstream analyses were based on clean data with high quality. To quantify gene expression levels, featureCounts v1.5.0-p3 was used to count the number of reads mapped to each gene. The expected number of Fragments Per Kilobase of transcript sequence per Million base pairs sequenced (FPKM) of each gene was calculated based on the length of the gene and reads count. Differential expression analysis of two conditions/groups (two biological replicates per condition) was performed using the DESeq2 R package (1.20.0). DESeq2 provide statistical routines for determining differential expression in digital gene expression data using a model based on the negative binomial distribution. The P value of mRNA differential analysis was corrected for the false discovery rate (FDR), and genes with a FDR < 0.05 were regarded as DEGs. The resulting P-values were adjusted using the Benjamini and Hochberg’s approach for controlling the false discovery rate. A threshold of p<= 0.05 was used in all analysis.

2.5. Target Prediction and Enrichment Analysis

To determine if specific genes were targeted by miRNAs, miRWalk2.0 was used with a threshold of 1.0 [28]. In contrast, the miRWalk2.0 platform performs a robust meta-analytic integration of predicted miRNA–mRNA interactions by aggregating data from 13 independent target prediction algorithms. This multi-source consensus framework enhances the stringency of target identification and minimizes false-positive rates by leveraging algorithmic complementarity and cross-validation across diverse computational methodologies [29,30]. Finally, the DEGs and DEMs (p < 0.05) were further selected to test for functional enrichment analysis using IPA software.

2.6. Relationship to Genes Previous Identifed Differentiating AD-A and AD-N

Several studies have identifed gene expression differences in LCLs between the AD_A and AD-N groups, These include UCP2, mTOR, AMPK, PINK1, PTEN, MFN2, PTEN, AKT1, SOD2, SIRT1, SIRT3, DNM1L, HIF1A, PPARGC1A. In our previous studies comparing AD-N and AD-A we found that the mTOR pathway way higly invoved in modulation of respiration in the AD-A groups. Thus, we exmained the relation of miRNAs to core mTOR genes, including MTOR, RPTOR, RICTOR, mST8, DEPTOR, PRAS40, mSIN1 as well as upstream regulators AMPK (PRKAA1 , PRKAA2, PRKAB1, PRKAB2, PRKAG1, PRKAG2, PRKAG3), LKB1, AKT (AKT1, SKT2, AKT3), PTEN, RHEB, TSC1, TSC2, PI3K (PIK3CA, PIK3CB, PIK3CD, PIK3R1, PIK3R2, PIK3R3). We also examine the genes idenfitied and verifed in our previous miRNA manuscrupt differentaition ASD and siblings controls. These include AUTS2, FMR1, IL27, FOXP1, NTN1, NCAM2, GABRA4. In our previous study we also found that Cam Kinase II was differentially expressed and correlateed with ASD severity, so we examined CAMK2A, CAMK2B, CAMKK2.

Figure 1. Differential expression of mRNAs in LCLs groups. Volcano plot of differentially expressed between AD-A and AD-N. Red dots are up-regulated, Blue dots are down-regulated; Grey dots no change; To determine significant genes (red and blue color dots), the p-value cut-off was set to 0.05 and fold change of 1.0.

Figure 1. Differential expression of mRNAs in LCLs groups. Volcano plot of differentially expressed between AD-A and AD-N. Red dots are up-regulated, Blue dots are down-regulated; Grey dots no change; To determine significant genes (red and blue color dots), the p-value cut-off was set to 0.05 and fold change of 1.0.

3. Results

3.1. Identification of Differentially Expressed Genes (DEGs)

Table 2 summarizes 24 differentially expressed genes (DEGs) between the AD-A and AD-N groups, identified through RNA-seq analysis using a significance threshold of p < 0.05 and an absolute log₂ fold change > 1.0. Among these, 14 transcripts were significantly downregulated (i.e., elevated in AD-A relative to AD-N), while 10 exhibited upregulation (i.e., elevated in AD-N).Genes upregulated in the AD-N cohort include mitochondrial ribosomal protein MRPL41 and the nuclear RNA processing gene RPP25L, suggesting perturbed mRNA maturation dynamics. Additionally, transcripts linked to immunological function (e.g., HLA-DQA2, HEXIM1, PNMA1) and oncogenic pathways(HEXIM1, FRAT2, PNMA1, HMGA1) were enriched.

Conversely, the AD-A group demonstrated upregulation of several transcripts implicated in tumorigenesis (CAB39), intracellular signaling (CAB39), autoimmunity (IFIT3), and mTOR signaling (DEPTOR). Genes involved in transcriptional regulation (TCEAL8, DDX21, CHD3, HIST2H2BE) and phospholipid metabolism (LYPLA1) were also significantly elevated. Notably, none of these DEGs are curated within the SFARI Gene database, indicating a potential divergence from established autism-related gene sets.

Table 2. Top differentially expressed mRNAs between AD-A and AD-N LCLs.

Up Regulated in AD-N (LCLs from Children with Autism with Normal Mitochondrial Function). Bolded an italic are known biomarkers for ASD.
mRNA	log2(FC)	-log10(p)	Gene Name
MRPL41	2.04	1.73	Mitochondrial ribosomal protein l41
HLA-DQA2	1.49	1.56	Major histocompatibility complex, class ii, dq alpha-2
GAPDHP1	1.46	3.03	Glyceraldehyde-3-phosphate dehydrogenase pseudogene 1
RPP25L	1.44	1.72	Ribonuclease P/MRP Subunit p25
SNRPFP2	1.43	1.31	Small nuclear ribonucleoprotein polypeptide F pseudogene 2
HEXIM1	1.26	1.55	Hexamethylene bis acetamide-inducible protein 1
FRAT2	1.17	3.08	Frequently rearranged in advanced t-cell lymphomas 2
FAM110A	1.14	2.18	Family with sequence similarity 110
PNMA1	1.11	1.29	Paraneoplastic antigen ma1
HMGA1	1.02	1.67	High mobility group at-hook 1
Down Regulated in AD-N (LCLs from Children with Autism with Normal Mitochondrial Function)
miRNA ID	log2(FC)	-log10(p)	Gene Name
CTD-2287O16.1	-3.01	1.61	Pseudogene
CAB39	-1.85	1.45	Calcium-binding protein 39
RP11-603J24.17	-1.76	2.02	Antisense gene
IFIT3	-1.53	1.30	Interferon-induced protein with tetratricopeptide repeats 3
CECR1	-1.27	3.31	Adenosine deaminase 2
DEPTOR	-1.15	1.65	Dep domain-containing protein 6
F13A1	-1.13	1.45	Factor xiii, a1 subunit
TCEAL8	-1.13	1.53	Transcription elongation factor a like 8
LYPLA1	-1.12	2.31	Lysophospholipase i
IQUB	-1.11	1.65	Iq motif- and ubiquitin domain-containing protein
FAM103A2P	-1.07	1.98	RNA guanine-7 methyltransferase activating subunit like
DDX21	-1.04	2.52	Dexd-BOX HELICASE 21
CHD3	-1.04	1.55	Chromodomain helicase dna-binding protein 3
HIST2H2BE	-1.01	1.28	Histone gene cluster 2, h2b histone family, member e

Table 2. Top differentially expressed mRNAs between AD-A and AD-N LCLs.

Up Regulated in AD-N (LCLs from Children with Autism with Normal Mitochondrial Function). Bolded an italic are known biomarkers for ASD.
mRNA	log2(FC)	-log10(p)	Gene Name
MRPL41	2.04	1.73	Mitochondrial ribosomal protein l41
HLA-DQA2	1.49	1.56	Major histocompatibility complex, class ii, dq alpha-2
GAPDHP1	1.46	3.03	Glyceraldehyde-3-phosphate dehydrogenase pseudogene 1
RPP25L	1.44	1.72	Ribonuclease P/MRP Subunit p25
SNRPFP2	1.43	1.31	Small nuclear ribonucleoprotein polypeptide F pseudogene 2
HEXIM1	1.26	1.55	Hexamethylene bis acetamide-inducible protein 1
FRAT2	1.17	3.08	Frequently rearranged in advanced t-cell lymphomas 2
FAM110A	1.14	2.18	Family with sequence similarity 110
PNMA1	1.11	1.29	Paraneoplastic antigen ma1
HMGA1	1.02	1.67	High mobility group at-hook 1
Down Regulated in AD-N (LCLs from Children with Autism with Normal Mitochondrial Function)
miRNA ID	log2(FC)	-log10(p)	Gene Name
CTD-2287O16.1	-3.01	1.61	Pseudogene
CAB39	-1.85	1.45	Calcium-binding protein 39
RP11-603J24.17	-1.76	2.02	Antisense gene
IFIT3	-1.53	1.30	Interferon-induced protein with tetratricopeptide repeats 3
CECR1	-1.27	3.31	Adenosine deaminase 2
DEPTOR	-1.15	1.65	Dep domain-containing protein 6
F13A1	-1.13	1.45	Factor xiii, a1 subunit
TCEAL8	-1.13	1.53	Transcription elongation factor a like 8
LYPLA1	-1.12	2.31	Lysophospholipase i
IQUB	-1.11	1.65	Iq motif- and ubiquitin domain-containing protein
FAM103A2P	-1.07	1.98	RNA guanine-7 methyltransferase activating subunit like
DDX21	-1.04	2.52	Dexd-BOX HELICASE 21
CHD3	-1.04	1.55	Chromodomain helicase dna-binding protein 3
HIST2H2BE	-1.01	1.28	Histone gene cluster 2, h2b histone family, member e

GO enchainment analysis (Figure 2) several processes with high rich factor ?? and statistical significance including 75K snRNA binding, adenosine receptor binding and catabolic processes, palmitoyl-protein hydrolase and protein depalmitoylation activity and positive regulator of signal transduction of p53. Significantly inhibited pathways were of peptide crops-linking, nitric oxide metabolic process and mitochondrial larghe ribnosomal subunit.

To elucidate the biological relevance of differentially expressed genes, we examined their association with Gene Ontology (GO) biological processes, focusing on those reaching statistical significance (p < 0.05), as summarized in Table 3. Among the genes enriched in the AD-N subgroup, HEXIM1 and MRPL41 emerged as consistently implicated in distinct molecular pathways.. HEXIM1 was associated with cyclin-dependent protein serine/threonine kinase activity and positive regulation of p53 class mediators while MRPL41 was related to gene translation in the mitochondria. CECR1, DDX21, F13A1, LYPLA1 were consistently related to processes increased in the AD-A group. CECR1 GO processes all points to adenosine catabolism and salvage, DDX21 all point to RNA metabolism, F13A1 all point to healing processes and LYPLA1 GO processes point to lipoprotein modifications and membrane signaling dynamics. FRAT2, HEXIM1, MRPL41, CECR1, DDX21, RPP25L had genes with both increased and decreased differential expression between groups.

Figure 2. Enrichment analysis of the differentially expressed mRNAs by IPA software. The top 20 enriched GO functional processes of the target genes of differentially expressed mRNAs. Red=activated; Blue= Inhibited.

Figure 2. Enrichment analysis of the differentially expressed mRNAs by IPA software. The top 20 enriched GO functional processes of the target genes of differentially expressed mRNAs. Red=activated; Blue= Inhibited.

Figure 3. Differential expression of miRNA in LCLs groups by small RNA-seq analysis (miRNA seq). Volcano plot of differentially expressed between AD-A and AD-N. Red dots are up-regulated, Blue dots are down-regulated; Grey dots no change; To determine significant genes (red and blue color dots), the p-value cut-off was set to 0.05 and fold change of 1.0.

Figure 3. Differential expression of miRNA in LCLs groups by small RNA-seq analysis (miRNA seq). Volcano plot of differentially expressed between AD-A and AD-N. Red dots are up-regulated, Blue dots are down-regulated; Grey dots no change; To determine significant genes (red and blue color dots), the p-value cut-off was set to 0.05 and fold change of 1.0.

3.2. Identification of Differentially Expressed miRNAs (DEMs)

Annotation of total and unique miRNA species across AD-A, AD-N, and control samples revealed comparable distribution profiles, as illustrated by the pie charts in Supplementary Figures S1A and S1B. Analysis of read length distributions across these groups demonstrated a pronounced enrichment around 22 nucleotides, with the majority of high-confidence reads falling within the 21–24 nt range (Supplementary Figure S2), consistent with canonical miRNA length signatures.

Differential expression analysis, visualized through volcano plots (Figure 1A), identified a subset of both annotated and novel miRNAs meeting stringent thresholds of |log₂ fold change| and p ≤ 0.05. A comparative analysis between AD-N and AD-A samples revealed 19 significantly dysregulated miRNAs, including 18 upregulated and 1 downregulated transcripts out of 525 detected miRNA species.

These differentially expressed miRNAs, along with their fold-change values and statistical significance, are detailed in Table 4. Hierarchical clustering and heatmap visualization (Supplementary Figure S3) illustrate expression dynamics across samples, using a two-color gradient to depict relative abundance—ranging from red (upregulation) to green (downregulation).Integrative analysis revealed that a subset of dysregulated miRNAs—namely hsa-miR-1273h-3p, hsa-miR-197-3p, hsa-miR-199a-5p, hsa-miR-204-5p, hsa-miR-874-5p, hsa-miR-100, hsa-miR-941 and hsa-miR-769-5p. —are predicted to target mRNAs previously identified as differentially expressed, suggesting coordinated post-transcriptional regulatory mechanisms driving disease-specific transcriptomic remodeling.

The graph in Figure 4 represents the target genes of the differentially dysregulated miRNAs enriched in GO terms, where the color of the circle indicates the statistical significance expressed in log10 values and the size of the circle the number of target genes involved. KEGG pathway analysis showed that the target genes were notably enriched in the pathways involving transport, transferase activity, DNA transcription, nucleotide and metal ion binding, and ATP binding.

To assess whether the differentially expressed miRNAs converge on common molecular pathways, we utilized miRPath v.3 to perform integrative pathway analysis based on the union of miRNAs targeting both experimentally identified mRNAs and computationally predicted gene targets [31]. KEGG pathway enrichment revealed eight distinct functional categories, while Gene Ontology (GO) analysis identified seven enriched biological process categories. As illustrated in Figure 5, ECM-receptor interaction emerged as the most prominent and broadly targeted pathway, with the largest subset of associated miRNAs, suggesting a central role for extracellular matrix signaling in the regulatory network implicated in ASD subtypes.

4. Discussion

We report the identification of two biologically distinct subgroups of autism spectrum disorder (ASD) based on mitochondrial functional phenotypes in lymphoblastoid cell lines (LCLs): AD-A (ASD with aberrant mitochondrial function characterized by elevated respiration and stress sensitivity) and AD-N (ASD with normative mitochondrial function). To elucidate molecular mechanisms underlying these phenotypes, we conducted integrated transcriptomic profiling—encompassing both mRNA and miRNA sequencing—followed by differential expression analysis, functional enrichment, and pathway interrogation, with a particular focus on mTOR and mitochondrial stress-related genes previously implicated in these ASD subtypes.

RNA-seq analysis revealed 24 mRNAs differentially expressed between AD-A and AD-N groups (p < 0.05, |log₂FC| > 1), with 14 transcripts upregulated in AD-A and 10 in AD-N. Upregulated transcripts in AD-N included MRPL41 and RPP25L, implicated in mitochondrial and nuclear RNA processing, respectively, as well as immune-related and oncogenic transcripts (HLA-DQA2, HEXIM1, PNMA1). In contrast, AD-A LCLs exhibited increased expression of genes associated with intracellular signaling (CAB39), autoimmune processes (IFIT3), and mTOR pathway regulation (DEPTOR), as well as genes involved in transcriptional control (TCEAL8, CHD3, DDX21) and lipid metabolism (LYPLA1). Gene Ontology (GO) enrichment analysis highlighted perturbations in snRNA binding, purinergic signaling, p53 regulation, and depalmitoylation processes..

Out of 525 detected miRNAs, 18 were upregulated and 1 was downregulated in AD-N as compared to AD-A. Several differentially expressed miRNAs, including hsa-miR-1273h-3p, hsa-miR-197-3p, hsa-miR-199a-5p, hsa-miR-204-5p, hsa-miR-874-5p, hsa-miR-100-5p, hsa-miR-941, and hsa-miR-769-5p, target both the genes identified in the mRNA analysis and those previously implicated in ASD, such as the mTOR pathway and Cam Kinase II. Pathway enrichment analysis revealed overrepresentation of pathways involved in transport, transferase activity, DNA transcription, nucleotide/metal ion binding, and ATP binding. The ECM receptor interaction pathway was notably enriched.

The results show a miRNA-mRNA network potentially regulating mitochondrial function in ASD, particularly the mTOR, PI3K/AKT, and autophagy pathways previously associated with mitochondrial regulation, immune response, and neuronal health. The study also connects some of these molecular changes to genes previously differentiated between ASD subtypes and affected by ASD severity.

Of particular note, the mechanistic target of rapamycin (mTOR) emerged as a central node in the transcriptomic network. mTOR exists in two functionally distinct complexes: mTORC1, which governs protein synthesis, mitochondrial metabolism, and autophagy, and mTORC2, which modulates cytoskeletal dynamics and cell survival. Hyperactivation of mTORC1—a phenomenon documented in a subset of ASD individuals—impairs mitophagy, elevates mitochondrial ROS, synaptic pruning and disrupts ATP homeostasis, culminating in synaptic and dendritic pathologies linked to core ASD features [21,28,30,31].

CaMKII also has a critical role in post-synaptic neuronal function where it modulates long-term potentiation and NMDA-dependent synaptic plasticity [32,33]. CaMKII has an important role in regulation of NMDA receptors, resulting neuronal excitability through glutamate transmission [34]. This may be significant as dysregulation of long-term potentiation [35] and glutamate regulation [36] has been implicated in psychiatric disorders. Mutations in CaMKIIα and CaMKIIβ are association with intellectual disability [37] and ASD behaviors [38].

Our previous study examining miRNA which distinguished ASD LCLs from controls identified two candidate miRNAs, hsa-miR-181a-5p and hsa-miR-320a [39], which are highly expressed in brain and spinal cord [40] and they have been identified as dysregulated in previous ASD studies [41,42,43]. Interestingly, both miR-181a-5p [44,45] and miR-320a [46] target AKT3, a key regulator of the PI3K-AKT-mTOR [47] and PTEN/Akt/TGF-β1 signaling pathway [27], two pathways which overlap the pathways identified in this study and are known to be involved in ASD.

5. Conclusions

This study underscores the molecular heterogeneity of ASD by identifying a distinct ASD subgroup characterized by abnormal mitochondrial function (AD-A) and unique miRNA–mRNA expression profiles. The enrichment of transcripts involved in mTOR signaling, cellular stress responses, and mitochondrial regulation highlights convergent pathways that may underpin the metabolic phenotype observed in this subset. These findings not only provide mechanistic insights into ASD pathophysiology but also point to actionable molecular signatures with potential utility as biomarkers for stratification. Importantly, the delineation of ASD subtypes based on mitochondrial and transcriptional dysregulation offers a critical step toward precision diagnostics and paves the way for pathway-targeted therapies tailored to metabolically vulnerable individuals with ASD.