Integrated multi-omics analyses of synaptosomes revealed synapse-associated novel targets in Alzheimer’s disease

Abstract

Synapse dysfunction is an early event in Alzheimer’s disease (AD) caused by various factors, including amyloid beta, p-tau, inflammation, and aging. However, the precise molecular mechanism underlying synapse dysfunction in AD remains largely unknown. To understand this, we comprehensively analyzed the synaptosomes fraction in post-mortem brain samples from AD patients and cognitively normal individuals. We conducted high-throughput transcriptomic analyses to identify changes in microRNA (miRNA) and mRNA levels in synaptosomes extracted from the brains of unaffected individuals and those with AD. Additionally, we performed mass spectrometry analysis of synaptosomal proteins in the same sample group. These analyses revealed significant differences in the levels of miRNAs, mRNAs, and proteins between the two groups. To gain further insights into the pathways or molecules involved, we employed an integrated omics approach to study the molecular interactions of deregulated synapse miRNAs, mRNAs, and proteins in samples from individuals with AD and the control group, demonstrating the impact of deregulated miRNAs on their target mRNAs and proteins. Furthermore, the DIABLO analysis revealed complex relationships among mRNAs, miRNAs, and proteins that could be key in understanding the pathophysiology of AD. Our study identified novel synapse-associated candidates that could be critical in restoring synapse dysfunction in AD.

Introduction

Alzheimer’s disease (AD) is a neurodegenerative disorder responsible for 60–80% of dementia cases in the United States (Alzheimer’s Facts and Figures, 2023) [1]. A myriad of factors has been associated with AD, including aging, genetics, lifestyle, infections, and injuries [2]. These factors can contribute to neurodegeneration and cognitive decline in patients. The two main pathological hallmarks of AD, amyloid-beta (Aβ) plaques and neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau (p-tau), are believed to be responsible for neural deterioration [3]. Recent studies indicate that one of the earliest contributors to AD is synaptic dysfunction, which subsequently leads to synaptic degeneration [4–7]. Both Aβ oligomers and NFTs have been implicated in synaptic dysfunction by altering multiple synaptic events [4–6, 8].

Synapses are fundamental units that allow neuronal communication throughout the nervous system and with other types of brain cells [9]. Functional synapses are essential for normal cognitive functions and brain activities. Thus, it is not surprising that synaptic dysfunction in AD manifests as cognitive deficiencies. Aβ-mediated synaptic dysfunction is one of the heavily studied mechanisms of AD pathology. In 1998, Aβ oligomers were identified as causing the loss of dendritic spines, which are often the postsynaptic component [10]. This disrupts NMDA-dependent long-term potentiation (LTP) and instead promotes long-term depression (LTD) [11]. Further studies have also detected Aβ in the presynaptic terminals of glutamatergic neurons, where it upregulates the amount of glutamate in the synaptic cleft. This leads to excitotoxicity and desensitization of glutamate receptors, contributing to synaptic dysfunction [11]. Meanwhile, p-tau is present in both presynaptic and postsynaptic terminals, inducing synaptic dysfunction through its own mechanisms. These include impairing presynaptic vesicle release, maturation of dendritic spines, and mitochondrial function in synapses. P-tau can also trigger synaptic phagocytosis by microglia [8]. While it is understood that Aβ plaques, p-tau, aging, and inflammation all contribute to synaptic dysfunction, the exact molecular mechanisms of synaptic dysfunction in AD  remain unknown [5, 12, 13, 14]. Recently, synaptosomes have been employed to investigate synaptic dysfunction at a molecular level in AD and other neurodegenerative diseases. They consist of intact forms of synapses, encompassing all the components involved in synaptic transmission, including presynaptic and postsynaptic membranes, mitochondria, proteins, neurotransmitters, mRNA, microRNA (miRNA), and more [5, 15, 16].

One of the components of synaptic dysfunction that has been gaining interest is miRNAs. MiRNAs are small, non-coding RNAs that regulate gene expression and are widely studied in AD from different perspectives, such as potential biomarkers, therapeutic candidates, and understanding the disease’s pathology [17–22]. Regarding synaptic activity, they have been implicated in modifying synaptic protein expression and transcription factors that lead to synaptic dysfunction in AD [5, 15, 19–22]. Numerous studies have identified specific miRNAs that play crucial roles in the synapses of AD, including miRNA-34a, miRNA-92, miRNA-125b, and more [20, 23–25]. In 2022, our lab also identified miRNA-501-3p, miRNA-502-3p, and miRNA-877-5p as novel synaptosomal miRNAs upregulated with AD progression [5]. Because many different miRNAs are differentially expressed in AD, they have been extensively studied as potential biomarkers [20, 26–30]. While many studies have explored the mechanisms of miRNA-related pathologies in AD, we are still uncertain how specific synaptosomal molecules change in the context of synaptic dysfunction.

The dysregulation of mRNA expression and subsequent protein translation in AD has also been studied. Hundreds of differentially expressed mRNAs have been identified in AD. Among these, genes that impact the function of microglia and astrocytes have been shown to be associated with a clinical diagnosis of AD [31]. Additionally, genes that affect the electron transport chain and protein binding are also involved in AD [32]. While identifying differentially expressed genes is crucial, mRNA expression accounts for only 40% of protein variance in mammals [33]. Therefore, proteomic studies are essential for achieving a more comprehensive understanding of AD pathology. In addition to Aβ and tau, several other synaptic proteins have been identified in AD, including Calsyntenin-1, GluR2 (Glutamate receptor 2), GluR4 (Glutamate receptor 4), and Neurexin-2A [34]. These proteins have been examined for their potential as synaptic biomarkers for AD; however, little is known about their direct contributions to synaptic dysfunction.

To date, no studies have assessed the interplay between synaptic miRNAs, mRNAs, and proteins in the context of synaptic dysfunction in AD. It remains unclear how specific miRNAs, mRNAs, and proteins change in the AD synapse compared to the control synapse. Additionally, it is uncertain whether the deregulation of all three molecular subsets is interconnected and if their expression depends on one another. Therefore, to understand the status of each molecular subset, we conducted transcriptomic and proteomic analyses of miRNAs, mRNAs, and proteins in AD and control synapses. We also employed a multi-omics integrative approach to evaluate how each component changes in AD and to analyze their interactions in order to understand the molecular basis of synaptic dysfunction. The current study provides novel synapse-associated molecular signatures that could serve as potential therapeutic targets and synaptic biomarkers in AD.

Materials and methods

Post-mortem brain samples

Post-mortem brains from AD patients and unaffected controls were obtained from NIH NeuroBioBanks- [1] Human Brain and Spinal Fluid Resource Center, 11301 Wilshire Blvd (127 A), Los Angeles, CA. [2] Brain Endowment Bank, University of Miami, Millar School of Medicine, 1951, NW 7th Avenue Suite 240, Miami, FL. [3] Mount Sinai NIH Brain and Tissue Repository, 130 West Kingsbridge Road Bronx, NY. Brain tissues were dissected from AD patients from Brodmann’s Area 10 of the frontal cortices (n = 27) and age and sex-matched unaffected controls (n = 14). The demographic and clinical details of study specimens are provided in Supplementary Table 1. The study was conducted at the Molecular and Translational Medicine Department, Texas Tech University Health Sciences Center El Paso, and the Institutional Biosafety Committee (IBC protocol # 22008) approved the study protocol for the use of human post-mortem brain tissues obtained from NIH NeuroBioBanks. The NIH NeuroBioBanks mentioned above operated under their institution’s IRB approval, and they obtained written informed consent from the donors [5, 35]. The NIH NeuroBioBanks given us the permission of using post-mortem brains in our research.

Synaptosomes extraction

Synaptosomes were extracted using Syn-PER Reagent, as described in our previous study [5]. Briefly, 50 mg of brain tissue was used from each sample for synaptosomes extraction in 1 mL of Syn-PER Reagent. Tissues were homogenized slowly by Dounce glass homogenization on ice with ~10 slow strokes. Samples were centrifuged at 1400 g for 10 min at 4 °C to remove the leftover tissue debris. After centrifugation, the supernatant was transferred to a new tube. Again, the supernatant (homogenate) was centrifuged at a speed of 15,000 g for 20 min at 4 °C. The supernatant was removed as a cytosolic fraction, and synaptosomes were recovered in pellet form. The isolated synaptosomes were characterized by transmission electron microscopy following our previous study [5]. The representative synaptosome images from AD and controls are shown in Supplementary Information Fig. 1 (SI Fig. 1). Further, synaptosome pellets were processed for RNA and protein extraction.

miRNA- and mRNA-Seq analysis

A brief demographic details of post-mortem brain samples used for synaptosomal miRNA and RNA sequencing analysis are provided in Table 1. Total RNA, including miRNAs, was extracted from the synaptosomes of post-mortem brain samples from individuals with AD and control subjects using a TriZol reagent (Invitrogen; CA, USA) with some modification. RNA quality and purity were assessed using NanoDrop 2000c (Thermo Fisher Scientific; MA, USA). Transcriptomic sequencing for mRNAs and miRNAs was conducted commercially by LC Sciences in Houston, Texas. The flow chart for the miRNA HiSeq analysis is presented in SI Fig. 2. All samples, 27 AD and 14 controls were included in the analysis.

Table 1.

Demographic details of control and AD post-mortem brain samples used for miRNA and mRNA sequencing analysis.

Demographic	Controls (n = 14)	AD (n = 27)	Difference (p-value)
Age (years) (mean ± SD)	80.07 ± 11.04	79.85 ± 8.29	0.9432
Sex (male/female)	7 M; 7 F	14 M; 13 F	1.0
PMI (hrs) (mean ± SD)	12.99 ± 7.54	13.23 ± 6.42	0.9178

Library construction and sequencing

The Poly(A) RNA sequencing library was prepared according to Illumina’s TruSeq stranded mRNA sample preparation protocol. RNA integrity was assessed using the Agilent Technologies 2100 Bioanalyzer. Poly(A) tail-containing mRNAs were purified with oligo(dT) magnetic beads through two rounds of purification. Following purification, the poly(A) RNA was processed for DNA library construction. Quality control analysis and quantification of the sequencing library were conducted with the Agilent Technologies 2100 Bioanalyzer High Sensitivity DNA Chip. Paired-end sequencing was carried out on Illumina’s NovaSeq 6000 sequencing system. All extracted RNA was utilized in the library preparation according to Illumina’s TruSeq small RNA sample preparation protocols (Illumina, San Diego, CA, USA). Single-end sequencing at 50 bp was executed on Illumina’s HiSeq 2500 sequencing system following the manufacturer’s recommended protocols.

Bioinformatics analysis of miRNAs

Raw reads were subjected to an in-house program, ACGT101-miR (LC Sciences, Houston, Texas, USA), to remove adapter dimers, junk, low complexity sequences, and common RNA families (rRNA, tRNA, snRNA, snoRNA), as well as repeats. Subsequently, unique sequences ranging from 18 to 26 nucleotides in length were mapped to specific species precursors in miRBase 22.0 using BLAST search to identify known miRNAs and novel 3p- and 5p-derived miRNAs. Length variation at both the 3’ and 5’ ends and one mismatch within the sequence were permitted in the alignment. The unique sequences mapping to specific species of mature miRNAs in the hairpin arms were classified as known miRNAs. Sequences mapping to the other arm of the known specific species precursor hairpin, opposite the annotated mature miRNA-containing arm, were considered novel 5p- or 3p-derived miRNA candidates. The remaining sequences were mapped to other selected species precursors (excluding specific species) in miRBase 22.0 via BLAST search, and the mapped pre-miRNAs were subsequently BLASTed against the specific species genomes to identify their genomic locations. Unmapped sequences were BLASTed against the specific genomes, and hairpin RNA structures containing sequences were predicted from the flanking 80 nt sequences using RNAfold software (http://rna.tbi.univie.ac.at/cgi-bin/RNAfold.cgi). The criteria for secondary structure prediction included: [1] a maximum of 12 nucleotides in one bulge in the stem, [2] at least 16 base pairs in the stem region of the predicted hairpin, [3] a cutoff for free energy (kCal/mol ≤−15), [4] a total length of hairpin (sum of up and down stems plus the terminal loop) of at least 50, [5] a maximum of 20 nucleotides in the hairpin loop, [6] a maximum of 8 nucleotides in one bulge in the mature region, [7] a maximum of 4 biased errors in one bulge in the mature region, [8] a maximum of 2 biased bulges in the mature region, [9] a maximum of 7 errors in the mature region, [10] at least 12 base pairs in the mature region of the predicted hairpin, and [11] at least 80% of mature sequence in the stem.

Analysis of differentially expressed miRNAs

The differential expression of miRNAs based on normalized deep-sequencing counts was selectively analyzed using the Fisher exact test, Chi-squared 2×2 test, Chi-squared nXn test, Student t test, or ANOVA, depending on the design of the experiment. The significance threshold was set at 0.01 and 0.05 for each test.

The prediction of target genes of miRNAs

To predict the genes targeted by the most abundant miRNAs, two computational target prediction algorithms (TargetScan 8.0 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 Gene Ontology (GO) terms and KEGG pathways of these most abundant miRNAs and mRNA targets were also examined annotated.

mRNA, transcriptome assembly and differential expression

Sequencing and mapping of NGS paired-end reads (150 bp) to the human transcriptome (hg38), with annotation from Ensembl release version 101 (August 2020), were performed using the HISAT2 aligner [36]. SAM files were converted to BAM files, which were then sorted and indexed using Samtools [37]. Read counts were computed from the alignment files with the FeatureCounts program from the subread-2.0.1 package [38], utilizing reference annotation from Ensembl version 101. Counts normalization and differential gene expression analyses were carried out using the DESeq2 R package [39]. The filtering criteria for differentially expressed genes were set to P-adjusted value < 0.05. The data were visualized using R statistical software, incorporating packages including ggplot2, plotly, and pheatmap enhanced Volcano.

Gene ontology and KEGG pathways

Biotype classification and composition analyses were performed using the Ensembl database (REST API) on filtered differentially expressed genes. GO analysis was conducted by querying the bioinformatic Database for Annotation, Visualization, and Integrated Discovery (DAVID) [40]. The following annotation categories were included in the analysis: KEGG Pathways, GO: Biological Processes, GO: Molecular Function, and GO: Cellular Components. The data were visualized in R statistical software (version 4.0.3) using the packages ggplot2 and plotly/heatmap for dot plots and heatmaps respectively.

Protein, mass spectrometry analysis

The LC-MS mass spectrometry analysis of synaptosomal proteins was performed on 5 AD post-mortem brains (Braak stage VI) and 5 control post-mortem brains samples with technical duplication at the Mass Spectrometry Research Facility, University of South Alabama, Alabama. All samples, 5 AD and 5 controls were included in the analysis.

Sample preparation

The synaptosomal proteins from AD and control post-mortem brain tissues were prepared using the SynPer kit from Thermo Fisher, followed by protein digestion. The pellets were stored at −80 °C upon receipt and thawed at room temperature immediately before preparation. To the pellets, 25 μL of 8 M urea was added and mixed at 37 °C for 10 min at 600 rpm. The samples were diluted with 200 μL of 50 mM ammonium bicarbonate (ABC) containing 10 mM tris-carboxyethyl phosphine (TCEP) for reduction. In-solution protein digestion was performed using 4 μL of sequencing-grade modified porcine trypsin (0.8 μg) (Promega, Madison, WI) overnight at 37 °C in a shaker set at 600 rpm. Samples were centrifuged at 16,100 × g for 15 min at 4 °C in the tabletop microcentrifuge. Two hundred microliters of the supernatant were transferred to a snap–top autosampler vial for HPLC isolation.

Mass spectrometry data analysis

In this study, mass spectrometry (MS) data were analyzed using a combination of MaxQuant software for peptide and protein identification and quantification, followed by statistical analysis in R with the Proteus package [41–43]. Initially, the raw MS data files generated by the liquid chromatography-tandem mass spectrometry (LC-MS/MS) experiment were imported into MaxQuant (version 2.6.5). Key parameters in MaxQuant were set to maintain a 1% false discovery rate (FDR) for both peptide and protein identification, ensuring high-confidence results. The search was conducted against a species-specific protein database, with parameters configured for enzyme specificity (e.g., trypsin), a defined number of missed cleavages, and variable and fixed modifications, like oxidation of methionine and carbamidomethylation of cysteine, respectively. Label-free quantification (LFQ) was employed to assess protein abundance across samples without needing stable isotope labeling. The “match between runs” feature was enabled to assist in matching peptide features across different LC-MS/MS runs, thereby increasing the number of identified peptides and enhancing protein quantification consistency across replicates. After running the MaxQuant pipeline, the results, including the proteinGroups.txt file, were exported for further analysis.

Next, the output from MaxQuant was loaded into R for analysis using the Proteus R package. Proteus provides a robust framework for proteomics data analysis, featuring functions for filtering low-quality data, normalization, and statistical testing. Initially, the data were filtered to exclude proteins with poor identification or low coverage, ensuring the analysis focused on high-confidence protein identifications. LFQ intensity values were then normalized to correct for technical variation across samples. Statistical testing for differential protein expression between conditions was conducted using moderated t-tests or other appropriate methods based on the experimental design, with Proteus managing the multiple hypothesis testing correction via the Benjamini-Hochberg procedure to control for false positives. The results included lists of significantly differentially expressed proteins, which were further interpreted using downstream bioinformatics tools for functional enrichment and pathway analysis [41–43].

Western blot analysis

The proteins were extracted from the AD and control post-mortem brains. Briefly, 20 mg of brain tissue was suspended in RIPA buffer (Thermo Scientific; IL, USA) supplemented with protease inhibitors (Thermo Scientific; IL, USA) and disrupted by ultrasonication (Q-Sonica Q125; CT, USA); amplitude 80%, pulse 10 sec on/off, time 30 s. Thereafter, tissue debris was removed by centrifugation (13,000 x g for 20 min). Protein concentration was estimated using the BCA assay (Thermo Scientific; IL, USA). Equal amounts of protein (40 μg per sample) were separated by SDS-PAGE on 10% polyacrylamide gels and transferred onto PVDF membranes (BioRad, Hercules, CA, USA). The membranes were blocked in 5% BSA for 1 h at room temperature. Then, the membranes were incubated overnight at 4 °C with primary antibodies against GPI, UQCRC1, TIMM50, VAT1L, and GAPDH (1:1000 v/v, Proteintech, Rosemont, IL, USA). Details of the antibodies and their dilutions are listed in Supplementary Table 2. After washing the membranes three times with TBS-T buffer at 10-min intervals, they were incubated for 1 h at room temperature with a secondary antibody (rabbit anti-mouse horseradish peroxidase [HRP] 1:10,000). Following three additional washes with TBS-T buffer, proteins were detected using chemiluminescence reagents (Thermo Fisher Scientific; IL, USA) and visualized with an Amersham imager 680 (GE Healthcare Biosciences, Uppsala, Sweden). Protein band intensities were quantified using ImageJ software (1.54 d, Java 1.8.0_345: http://imagej.org) for densitometry analysis, as described in more detail in our earlier study [44], and relative protein expression levels were normalized to GAPDH as loading controls.

Multi-omics integration analysis

Identifying genes with high variance: Identifying genes with high variance across biological replicates involves detecting genes whose expression levels show significant variability among different samples under the same condition. High variance may indicate biologically relevant differences or technical noise impacting gene expression consistency. We performed a comprehensive analysis using the R programming language to identify these genes, utilizing the capabilities of “limma” and “edgeR.” packages.

Data preparation

Creation of DGEList Object: We began by creating a DGEList object from the raw count data, which includes the counts of reads mapped to each gene across all samples. This object is essential for downstream analysis in edgeR. Filtering Lowly Expressed Genes: To ensure the robustness of our analysis, we filtered out genes with low expression levels across samples. This step reduces noise and enhances the detection of genuinely variable genes.

Normalization and transformation

Normalization of Counts: We normalized the raw counts for variations in sequencing depth and RNA composition using the trimmed mean of M-values (TMM) method in edgeR. Normalization ensures that comparisons of gene expression levels across samples are precise. Variance-Stabilizing Transformation: We applied the voom transformation from the limma package to stabilize the variance across the mean expression levels. This transformation is essential for making the data suitable for modeling.

Linear modeling and statistical analysis

Fitting the Linear Model: We fit a linear model to the transformed data using the lmFit function from the limma package. This step estimates gene expression levels and accounts for the experimental design. Computing Statistics with eBayes: We used the eBayes function to calculate empirical Bayes statistics for the linear model coefficients. This method enhances the reliability of statistical inferences by borrowing strength from the ensemble of genes. Extracting Top Genes by P-Value: Finally, we ranked the top genes with high variance based on their P-values, considering those with the lowest P-values as the most significant variables across replicates. Using edgeR and limma in our analysis provided a robust framework for identifying genes with high variance in expression levels across biological replicates. By combining normalization, variance-stabilizing transformation, linear modeling, and empirical Bayes statistics, we ensured the identification of genes exhibiting genuine biological variability, minimizing the impact of technical noise.

Statistical analysis

RNA-Sequencing (miRNA and mRNA): Differential expression analysis was performed using the R package DESeq2. This method applies the Wald test to evaluate whether the log2 fold change in gene expression between control and AD groups is significantly different from zero. The Wald statistic, calculated as the estimated coefficient divided by its standard error, yields p-values that are subsequently adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) method. Genes with an FDR < 0.05 were considered statistically significant. Proteomics: For proteomic data, differential expression analysis was conducted using the limma package in R. Limma fits a linear model to the log-transformed protein abundance values for each protein and applies an empirical Bayes approach to moderate standard errors across proteins. This generates moderated t-statistics, which enhance statistical power and stability, particularly in datasets with small sample sizes. Resulting p-values are adjusted for multiple comparisons using the Benjamini-Hochberg FDR method, and proteins with an FDR < 0.05 were considered statistically significant. Two-way analysis of variance was employed to assess the data across four groups, evaluating synapse protein changes between the controls and AD samples at different Braak stages. Student’s t test was used to evaluate the synapse protein changes between the controls and AD samples at Braak stage VI samples. Statistical parameters were calculated using GraphPad Prism software, version 8.0 (GraphPad, San Diego, CA, USA) (www.graphpad.com). Data was analyzed as standard error of mean (SEM) and P values of less than 0.05 were considered statistically significant.

DIABLO analysis

We conducted a multi-omics integration analysis using the DIABLO (Data Integration Analysis for Biomarker discovery using Latent components) framework, available in the mixOmics package in R programming language [45]. The analysis involved the following steps: Sample Selection: Only five samples per group (AD and HC) had proteomics data, so we included those samples for further analysis of mRNA-seq, miRNA-seq, and proteomics. Data Integration: We combined the preprocessed datasets into a list format suitable for DIABLO analysis. Each omics dataset was represented as a data frame where rows corresponded to samples and columns to features (genes, miRNAs, or proteins). Design Matrix Specification: To facilitate the integration of these datasets, we defined a design matrix that specified the connection strength between each pair of datasets. The diagonal elements of the design matrix were set to zero, while off-diagonal elements were set to 0.1 to allow for moderate integration strength. Model Fitting: We established the number of components at five, enabling the model to capture key latent variables across the datasets. The DIABLO model was fitted to the data using the “block.splsda” function from the mixOmics package. Visualization and Analysis: We visualized the sample projections on the latent components using the “plotIndiv” function, which allowed for examining clustering patterns according to the outcome variable. The contribution of variables to each component was visualized using the “plotVar” function. A heatmap of the integration was generated using the “cimDiablo” function, and a circos plot of the integration was created using the “circosPlot” function. Extraction of Loadings: To identify the key biomarkers, we extracted the loadings of each variable onto the components, selecting components one and two for this analysis. Model Validation: The performance and stability of the DIABLO model were assessed through cross-validation. We employed repeated k-fold cross-validation to ensure the robustness of the selected components and the identified biomarkers.

Results

MiRNA-Seq analysis of synaptosomal miRNA in Alzheimer’s brain

The miRNA-Seq analysis was conducted on synaptosomal RNAs isolated from the cortical area (BA10) of 14 controls and 27 AD post-mortem brains. A total of 133 precursor-miRNAs (pre-miRNA), 1307 mature miRNAs, and 498 novel Potential Candidates (PCs) were found to be expressed in the AD and control synapses (Supplementary Table 3). Among all categories of small RNAs, 1793 molecules, including pre-miRNAs, miRNAs, and PCs, were significantly deregulated in AD versus (vs) control synapses (Supplementary Table 4). SI Fig. 3 shows the complete heat map of all deregulated miRNAs in AD vs control synapses. Next, we filtered the mature miRNA list to include only those with adjusted p-values and a higher fold change. As a result, 455 mature miRNAs were found to be significantly deregulated (p < 0.05; fold change +/− 2) in AD vs control synapses (Supplementary Table 5). Further, to identify the topmost deregulated miRNAs, we narrowed down the miRNA selection criteria with miRNA reads (>10 per sample) and fold change (+/− 6-fold; p < 0.05), and we found 38 miRNAs that were more significantly deregulated in the AD synapse relative to the control synapse. The fifteen miRNAs were upregulated, and 51 miRNAs were found to be downregulated, as shown in the miRNA heatmap (Fig. 1A). These miRNAs were further segregated based on their log2 fold change and p-values, as shown in the volcano plot (Fig. 1B). miRNA-122-5p, miRNA-132-3p, miR-34c-5p and miRNA-212-3p, and miRNA-132-5p showed the highest Log10 p-value and fold change variance. Furthermore, miRNA correlation analysis was performed using the log2 mean values from both AD and control groups. Six miRNAs highlighted in blue displayed a significant correlation regarding their mean value in AD and controls, as shown in the correlation plot (Fig. 1C). Next, we conducted the in-silico bioinformatic analysis of downregulated and upregulated miRNAs in AD synapses. The KEGG/GO enrichment analysis showed that downregulated miRNAs are involved in several brain and neuron-related cellular pathways (Fig. 1D; Supplementary Table 6). Importantly, miRNAs are associated with AD, synapse, learning and memory, and other AD-related pathways. Similarly, the upregulated miRNAs were also involved in AD, synapse assembly, Huntington’s disease (HD), social behavior, and behavioral fear response (Fig. 1E; Supplementary Table 7). These results confirm that the deregulation of synapse miRNAs in AD is detrimental to normal synapse function, brain activity, and AD.

Fig. 1. MiRNA-Seq analysis in AD vs control synapse.

A Heatmap displaying differentially regulated miRNAs: the top upregulated and downregulated miRNAs in AD (n = 27) synaptosomes compared to control (n = 14) synaptosomes. B A volcano plot showing the top deregulated miRNAs’ (log₁₀ p-value). MiRNAs with an FDR < 0.05 were considered statistically significant. C Correlation analysis of the top differentially regulated miRNAs in AD vs control synaptosomes with a significant R-value. D Gene set enrichment analysis of top downregulated miRNAs shows affected biological pathways in human diseases, along with their p-values and number of genes. E Gene set enrichment analysis of top-upregulated miRNAs shows depleted and enriched biological pathways in human diseases, along with their p-values and numbers of genes.

mRNA-Seq analysis of synaptosomal mRNA in Alzheimer’s brain

mRNA or polyA RNA-Seq analysis was conducted on synaptosomal mRNAs in the same 14 controls and 27 AD post-mortem brains. A total of 662 mRNAs were found to be significantly deregulated (p < 0.05; fold-change +/− 2) in AD vs control synapses (Supplementary Table 8). SI Fig. 4 shows the complete heat map of significantly deregulated mRNAs in AD vs control synapses. To identify the topmost deregulated mRNAs, we narrowed down the miRNA selection criteria to mRNA reads (>10 per sample) and a fold change of ±2, which revealed the top 20 significantly deregulated genes (Fig. 2A). The heatmap showed the top ten upregulated genes were- SHANK1, HIVEP3, SLC7A2, SAMD4A, BCL9L, GAREM2, GLIS3, MTSS2, NWD1, and KDM6B. While downregulated genes were- TF, TUBA1A, MAL, EIF5B, SEPTIN4, CNP, S100B, SEMA3B, MOBP and PLP1 (Fig. 2A). The same top gene transcripts were segregated in AD and controls based on their log2 fold change and p-values, as shown in the volcano plot (Fig. 2B). Next, correlation analysis revealed a significant correlation between potential gene targets in AD and controls, as indicated by their mean values in both groups, as shown in the correlation plot (Fig. 2C). Further, we conducted the in-silico bioinformatic analysis to determine the roles of deregulated miRNAs in human disease, biological pathways, and cellular processes (SI Figs. 5 and 6). A KEGG/GO analysis of the upregulated mRNAs shows that most of these genes are involved in AD-related processes: Focal adhesion, Rap1 signaling pathway, Wnt signaling pathway, MAPK signaling pathway, cAMP signaling pathway, HIF-1 signaling pathway, TGF-beta signaling pathway, Hippo signaling pathway, FoxO signaling pathway and Hedgehog signaling pathway (SI Fig. 5). KEGG/GO analysis of the downregulated mRNAs also shows their involvement in AD, HD, Parkinson’s disease (PD), Amyotrophic lateral sclerosis, motor proteins, and pathways of neurodegeneration (SI Fig. 6). The downregulated mRNAs were also found to be involved in neuron-related biological processes, molecular functions, and synapse-related cellular components (SI Fig. 6). Additionally, we conducted the gene set enrichment analysis for the deregulated genes separately in controls and AD. The deregulated mRNA involved in multiple normal cellular processes and biological pathways in healthy controls (SI Fig. 7), while in AD cases, deregulated mRNAs are mostly involved in neurological and synapse-related biological pathways, as shown by KEGG pathway analysis (SI Figure 8). These results confirmed that synaptosome-localized mRNA population levels are significantly altered in AD and associated with multiple neuronal pathways.

Fig. 2. mRNA-Seq analysis in AD vs control synapse.

A Heatmap showing the top differentially regulated genes in AD (n = 27) vs control (n = 14) synaptosomes. B Volcano plot depicting the top differentially regulated genes in AD vs control synaptosomes. Genes with an FDR < 0.05 were considered statistically significant. C Correlation analysis of top genes in AD vs control synaptosomes with significant R-value.

Mass spec analysis of synaptosomal proteins in Alzheimer’s brain

To identify differentially regulated proteins, we performed mass spec analysis on synaptosomal proteins from 5 controls and 5 AD post-mortem brains (Braak stage VI) with two technical replicates. A total of 1066 proteins were identified in the synaptosomal fraction in AD and controls (Supplementary Table 9). Furthermore, proteins were categorized based on their significant fold changes. A total of 152 proteins were determined to be significantly deregulated (p < 0.05; fold change +/−2) in AD vs control synapse (Supplementary Table 10). To pinpoint the most deregulated proteins, we narrowed down the protein selection criteria by using a fold change of (+3/−3) and p-values (p < 0.05). Seventeen proteins were identified as the most significantly upregulated (TIMM50, NONO, VAT1L, MOBP, ENPP6, PPP3R1, RDX, TF, HSPA2, VGF, H1-2, SEPTIN4, NEFM, MOG, RPLP2, INA, and BCL2L13), and sixteen proteins were found to be significantly downregulated (GPI, UQCRC1, DNAJA1, PGAM5, CASKIN1, FGG, CLIC4, VPS50, FGB, EEF1A2, SCRN1, GATD3, PYGM, YWHAE, YWHAG, and UCHL1) as shown in the protein heatmap (Fig. 3A). The top deregulated proteins were further categorized based on their log2 fold change and Log10 p-values, as depicted in the volcano plot (Fig. 3B). Several proteins showed a significant correlation in their Log2 mean values between AD and controls, as illustrated in the correlation plot (Fig. 3C). Additionally, we performed in silico bioinformatic analysis to determine the roles of deregulated proteins in human disease, biological processes, and cellular pathways. The KEGG pathway analysis of upregulated proteins revealed the significant involvement of upregulated synaptosomal proteins in key neurological pathways, including pathways related to neurodegeneration, multiple diseases, PD, Long-term potentiation, and the HIF-1 signaling pathway (Fig. 3D). Similarly, upregulated proteins at synapses also participated in multiple biological pathways, including dopaminergic, cholinergic, and serotonergic synapses (Fig. 3E). These findings confirmed that the levels of synaptosome-localized proteins are significantly altered in AD.

Fig. 3. Mass-spec analyses of differentially regulated proteins in AD vs control synapse.

A The heatmap of the top-up and downregulated proteins in AD (n = 5) vs control (n = 5) synaptosomes includes two technical duplicates. B Volcano plot depicting the top differentially regulated proteins in AD vs control synaptosomes. Proteins with an FDR < 0.05 were considered statistically significant. C Correlation analysis of the expression of the top differentially regulated proteins in AD vs control synaptosomes shows a significant R value. D KEGG pathway analysis of the top upregulated proteins shows significant fold enrichment and protein counts in human diseases and biological pathways. E KEGG pathway analysis of top downregulated proteins shows significant fold enrichment and protein count in human diseases and biological pathways.

Status of top synaptosomal proteins in Alzheimer’s progression

Based on the biological functions of identified proteins, we validated the levels of the top four deregulated proteins – downregulated (GPI and UQCRC1) and upregulated (TIMM50 and VAT1L) in AD progression.

Immunoblotting  analyses were performed for the GPI, UQCRC1, TIMM50, and VAT1L proteins on AD post-mortem brain samples in different Braak stages and control brains (Fig. 4A). The results showed that the levels of two downregulated proteins, GPI (Glucose Phosphate Isomerase) and UQCRC1 (Ubiquinol-Cytochrome C Reductase Core Protein 1), were changed in AD. However, significant downregulation was found in GPI levels with AD severity (Braak Stages V/VI) (Fig. 4B). We did not observe any significant change in the level of TIMM50 (Translocase of Inner Mitochondrial Membrane 50). However, the level of another top-upregulated protein, VAT1L (vesicle amine transport 1 like), significantly increased in AD with increased Braak stages (Fig. 4B). Since immunoblotting was performed using the cortical tissue and not on the isolated synaptosomes, this could explain why all four proteins do not show changes in AD progression. Nonetheless, our results identified two new synapse-associated proteins (GPI and VAT1L) that significantly changed AD progression.

Fig. 4. Immunoblotting analysis of top deregulated proteins in AD post-mortem brain samples.

A Western blots for GPI, UQCRC1, TIMM50, and VAT1L proteins in post-mortem brains from control and AD brains at Braak stages I/II, III/IV, and V/VI. B Densitometry analysis of GPI, UQCRC1, TIMM50, and VAT1L protein blots in control and AD post-mortem brains with different Braak stages. Data are presented as mean ± SEM (n = 4) **P < 0.01, ***P < 0.001 using two-way ANOVA. C Western blots for GPI, UQCRC1, TIMM50, and VAT1L proteins in post-mortem brains from control and AD brains at Braak stage VI. Densitometry analysis of (D) GPI, (E) UQCRC1, (F) TIMM50, and (G) VAT1L protein blots in control and AD post-mortem brains with Braak stage VI. Data are presented as mean ± SEM (n = 4) *P < 0.05, **P < 0.01 using student’s t test.

Additionally, we confirmed the status of these four proteins exclusively in advanced AD stage (Braak stage VI) and control post-mortem brain samples (Fig. 4C). Immunoblotting analysis revealed a significant decrease in the levels of GPI, UQCRC1, and TIMM50 proteins, while VAT1L protein levels showed a notable increase in Braak stage VI Alzheimer’s post-mortem brains (Fig. 4D–G). This finding verified a significant correlation between these proteins and the disease severity.

Molecular interaction of synapse-specific miRNA-mRNA-Protein in Alzheimer’s disease

After conducting analyses specific to each dataset, we examined the molecular interactions of deregulated miRNA, mRNA, and proteins in individuals with AD compared to controls. We found that the fold expression intensity of miRNAs is inversely related to their target mRNA and proteins (Fig. 5). Most miRNA expression is low (green) while their target protein expression is higher (red). However, we did not observe a significant change in the mRNA expression of specific proteins. Nevertheless, the top deregulated miRNAs: miRNA-122-5p, miRNA-132-3p, miRNA-194-5p, miRNA-212-3p, miRNA-34c-5p, miRNA-3200-3p, miRNA-192-5p, miRNA-421, miRNA-132-5p, and miRNA-340-5p could negatively modulate their target mRNA expression and protein translation, as shown in the heatmap of Fig. 5. Furthermore, the gene network map revealed the interaction of each miRNA with its target proteins that were deregulated in AD synaptosomes. Each miRNA candidate interacted with multiple protein targets (Fig. 6). This analysis demonstrated that the top deregulated miRNA, mRNA, and protein expression may be interdependent, and changes in miRNA expression may lead to the deregulation of their target proteins.

Fig. 5. Multi-omics integration analysis of synapse miRNA-mRNA-proteins in AD.

Heat map showing the expression levels of miRNAs, their target mRNA, and proteins in AD synaptosomes.

Fig. 6.

Gene integration analysis of top deregulated miRNAs and their interaction with target proteins in AD synapse.

Multi-omics integration analysis of synaptosomal miRNA, mRNA, and protein

In addition, to better understand the molecular relationship between synapse miRNA, mRNA, and protein in the same individuals (both with AD and controls), we carried out Data Integration Analysis for Biomarker Discovery using Latent components (DIABLO). This is a multi-omics integrative analysis aimed at identifying differences between individuals with AD and controls [45]. For the DIABLO analysis, we selected five control samples and five samples from individuals with severe AD (Braak stage VI).

Circos plot for an integrative framework

The Circos plot, derived from the DIABLO analysis, reveals intricate relationships between different molecular data types - mRNA, miRNA, and proteins - from AD patients and healthy controls (Fig. 7A). By integrating these multi-omics data, the plot highlights significant correlations, illustrating how various molecular features interact within the context of AD. Using the first two components (Comp 1-2) allows for a focused representation of the most critical patterns observed in the data, providing insights into potential biomarkers and underlying biological mechanisms of AD. The plot emphasizes features with strong correlations (absolute value of r ≥ 0.7), differentiating positive and negative correlations through distinct connecting lines. For instance, mRNAs such as BAHCC1 and CAMK2N1, miRNAs like hsa-miRNA-1260a and hsa-miRNA-124-5p, and proteins including EEF2 and USO1 are shown to have significant interactions. These high-correlation features suggest potential regulatory relationships or co-involvement in disease-related pathways. The differential expression of these features between AD patients and healthy controls further underscores their relevance. For example, miRNAs such as hsa-miRNA-1260a may regulate mRNA targets implicated in neuronal function or pathology, while proteins like EEF2 and USO1 might play roles in cellular processes disrupted in AD.

Fig. 7. DIABLO analysis of synapse miRNA-mRNA and proteins in AD.

A Circos plot showing the positive and negative correlations between different molecular data types: mRNA, miRNA, and proteins in AD and control post-mortem brains. B The clustered heatmap shows the expression levels of various molecular features, including mRNAs, miRNAs, and proteins, across samples of AD and control post-mortem brains.

Clustered image map representing the multi-omics molecular signature expression

The heatmap generated from the DIABLO analysis provides a comprehensive visualization of multi-omics data integration for AD (Fig. 7B). This figure illustrates the expression levels of various molecular features, including mRNA, miRNA, and protein, across samples from AD patients and controls. The color gradient, ranging from blue to red, represents the range of expression levels, with blue indicating lower expression and red indicating higher expression. The observed patterns reveal significant molecular differences between AD patients and healthy controls.

In the heatmap, mRNAs such as BAG3, ZHX3, and CAMK2N1, miRNAs like hsa-miRNA-1260a and hsa-miRNA-124-5p, and proteins including EEF2 and USO1 showed varying expression levels between AD and HC samples. These differential expressions suggest their potential role as biomarkers for AD. For instance, upregulated genes and miRNAs in AD samples (indicated by red shades) could be involved in disrupted pathways. Clustering features with similar expression profiles indicates co-regulation or involvement in related biological processes, further emphasizing their relevance in the disease context.

Discussion

AD is a synaptic failure disease; therefore, it is crucial to understand the in-depth molecular biology underlying synapse function and dysfunction in AD. Synapse is a small cargo that retains specific RNAs and proteins that work together for normal synaptic plasticity. So far, multiple genes and proteins have been identified at synapses that play very close roles in neurotransmission and synaptic activity [46]. Deregulation of these synaptic genes leads to synapse dysfunction in AD. Other than mRNA and synaptic proteins, several small RNAs (miRNAs) have also been identified in AD synapses that play critical roles [5, 22, 46]. A homeostatic balance among miRNA, mRNA, and protein expression is necessary for a healthy synapse function. It is well known that AD synaptic dysfunction is initiated by multiple factors such as Aβ and p-tau proteins, aging, inflammation, environmental factors, and other genetic and epigenetic factors. Indeed, the etiology of AD remains unclear and involves the interplay of genetic and environmental factors, including diet and lifestyle [47–49]. Again, synapse dysfunction in AD is largely unknown specially at molecular level. Is it not clear whether the deregulation of specific miRNAs, mRNAs, or proteins causes synaptic dysfunction in AD, or if all three molecular subsets work together and lead to synaptic dysfunction? In the current study, we implemented a multi-omics integrative analysis to understand the status of each molecular subset: miRNAs, mRNAs, and proteins in AD. How each molecular subset changes in AD relative to control situations and what the connections are among each subset.

Changes in synapse miRNAs in AD

We first analyzed changes in the small RNA transcriptome, specifically miRNA levels in the synaptosomal fraction isolated from post-mortem brains of individuals with AD and the control group. We found both novel and previously reported miRNAs in AD. Among the most significantly downregulated miRNAs in AD are miRNA-132 [50], miRNA-34 [51], and miRNA-212-3p [52], which have been extensively studied in AD. Another significantly downregulated miRNA, miRNA-122-5p (more than 6-fold down), is less studied in the context of AD. A recent study has identified the role of miRNA-122-5p in compromised microglial chemotaxis and reduced restrictions of AD pathology [53]. A marked reduction in miRNA-122-5p in the synapse further emphasizes its relevance in AD research. Additionally, gene set enrichment analysis revealed the significant involvement of downregulated miRNAs in AD and various pathways related to AD and brain function. On the other hand, the top upregulated miRNA, miRNA-199a, is involved in AD development by regulating Neuritin expression in APP/PS1 mice [54]. Furthermore, miRNA-451a has recently been explored as a serum biomarker in AD [55], and miRNA-144 is being investigated in AD pathogenesis as a key modulator of ADAM10 protein [56]. However, there is limited information available for miRNA-1247-5p in the context of AD. Additionally, downregulated miRNAs were found to be involved in AD and synapse-related biological pathways.

Changes in synapse mRNAs in AD

Similar to miRNAs, we also investigated changes in the expression of protein-coding genes in the same samples from individuals with AD and controls. As expected, we found several genes whose expression was significantly affected in AD compared to the controls. Most of these deregulated genes had been previously associated with AD, but their functions in relation to AD have been minimally studied. The top-upregulated gene, SHANK1, has been linked to neuropsychiatric disorders and cognitive dysfunction in various neurodegenerative diseases [57, 58]. Variants of the HIVEP gene have been reported to be dysregulated in AD [59]. Similarly, the downregulated gene TF (Transferrin) has been well-studied in AD and is associated with AD risk factors [60]. Another gene, TUBA1A (Tubulin), is also associated with trafficking defects and impaired motor behavior [61]. Our gene set enrichment analysis (SI Fig. 6) revealed the significant involvement of downregulated genes in brain-related biological pathways such as neuronal function, synaptic function, and AD progression. Many of these genes are involved in AD either directly or indirectly; however, further research is required to understand the roles of these genes in relation to AD.

Changes in synapse proteins in AD

In a manner similar to miRNAs and mRNAs, our proteomic analysis revealed interesting proteins that were deregulated in AD compared to controls. These protein targets have barely been investigated for their role in AD. One of the novel proteins downregulated in AD synapses was GPI (Glucose Phosphate Isomerase). Validation analysis also showed the downregulation of GPI with AD Braak stages. This is the first time GPI proteins have been reported in AD. GPI regulates glucose metabolism, and its deficiency causes fetal hemolytic anemia [62]. Since synapse functioning requires regulated glucose metabolism and ATP, it is very important to investigate the role of GPI in synaptic function and AD.

Another downregulated protein was UQCRC1; however, our validation analysis showed significant changes in UQCRC1 level in AD post-mortem brains specially braak stage VI samples. Similarly, the top-upregulated novel proteins were TIMM50, NONO, and VAT1L. Based on their biological functions, we validated the levels of TIMM50 and VAT1L in AD post-mortem brains, showing significant changes in TIMM50 and VAT1L levels in advanced AD stage. VAT1L has been identified in the brain and is mainly associated with cellular functions related to neuronal maintenance, neurotransmission, and Tau pathology [63]. Therefore, further research is needed to understand the role of VAT1L in neuronal dysfunction at AD synapses. The significant deregulation of GPI, UQCRC1, TIMM50 and VAT1L proteins in the AD braak stage VI may be directly linked to substantial synapse loss in advanced AD.

In addition to synaptic and neuronal proteins, our study identified several non-neuronal mRNA and protein signatures that are not specific to synapses (e.g., MOBP, MOG, MAL, ENPP6, SCRN1, RDX, SEPTIN4, PLP1, S100B, and CNP). These molecules are specific to other brain cell types such as astrocytes, microglia, oligodendrocytes, and endothelial cells. Although synapse formation is a complex neurocircuitry mechanism that transmits information from one neuron to another, the architecture of synapses and their proper functioning are systematically supported by other brain cells in addition to neurons [64–70]. It is well known that astrocytes and microglia form the tripartite and quadripartite synapse with neurons, thereby remodeling the structural and functional capabilities of synapses in the human brain [64–70]. Oligodendrocyte precursor cells (OPCs) play a significant role in neural circuit refinement, remodeling, axon interaction, and the establishment of bona fide synapses with neurons and OPCs [71, 72]. Beyond synaptic interaction, oligodendrocytes also contribute to AD progression [73, 74]. Furthermore, endothelial cells have been found to play a role in synaptogenesis and promote the formation of synaptic connectivity [75, 76]. Previous proteome analysis of AD and control synaptosomes also detected some of these non-neuronal and non-synaptic proteins in the synaptosomes fraction [77]. The FGG (fibrinogen protein) is a liver-specific protein, and previous studies have indicated that FGG can accumulate in the cortex of AD brains and is strongly associated with Aβ deposition [78, 79]. This may explain the detection of FGG in the synaptosomes fraction. Considering our data and previous studies, it is possible that, in addition to traditional synaptic proteins, other non-neuronal proteins also play a significant role in normal synaptic function under physiological conditions and in synaptic dysfunction under pathological conditions such as AD. Therefore, these proteins warrant further investigation and could serve as novel targets to mitigate synaptic dysfunction in AD and other neurodegenerative diseases.

Multi-omics integration analysis reveals an intricate relationship between miRNAs, target mRNA, and proteins

The integration of multi-omics data in AD revealed a connection between deregulated miRNAs and their target proteins. One specific downregulated miRNA, miRNA-122-5p, is associated with certain upregulated target proteins in AD. This analysis provides an overview of disrupted miRNA-mRNA-protein interactions in AD and suggests strategies to restore their normal function. Integrating multi-omics data using DIABLO provides a powerful approach to understanding the molecular complexity of AD. By identifying highly correlated and differentially expressed mRNAs, miRNAs, and proteins, researchers can identify potential biomarkers for early diagnosis or therapeutic targets. These molecular features may contribute to the development of AD through various mechanisms, including altered gene expression, disrupted signaling pathways, or impaired protein functions. For example, BAHCC1 and CAMK2N1 may be involved in neuroinflammatory responses or synaptic signaling alterations in AD, while specific miRNAs may modulate these processes by targeting relevant mRNAs.

Furthermore, the DIABLO analysis facilitates a comprehensive understanding of the disease by integrating various types of molecular data. This broader view can lead to the discovery of new interactions and pathways that may be overlooked in studies focusing on a single type of molecular data. As a result, the insights gained from this integrated approach not only improve our understanding of the molecular basis of AD but also open up possibilities for developing therapeutic strategies that target multiple factors.

The grouping of samples into distinct categories (AD and HC) based on their molecular profiles emphasizes the power of integrating multiple types of molecular data in distinguishing between healthy and diseased states. The ability to differentiate AD patients from healthy individuals based on their molecular signatures underlines the diagnostic potential of these biomarkers. Additionally, the identified mRNAs, miRNAs, and proteins offer insights into the underlying biological mechanisms of AD. For example, the increased expression of specific miRNAs in AD samples may indicate their role in regulating genes involved in neurodegeneration, while differentially expressed proteins could point to disrupted cellular functions.

Our multi-omics integration analysis revealed a tightly interconnected network of dysregulated miRNAs, mRNAs, and proteins in AD, highlighting complex regulatory mechanisms that contribute to disease pathogenesis. Notably, altered expression of specific miRNAs, such as the downregulation of miR-122-5p, was associated with upregulation of target mRNAs and proteins, suggesting disrupted post-transcriptional regulation. These molecular signatures were enriched in pathways related to neuroinflammation, synaptic signaling, and neuronal dysfunction—hallmarks of AD. By integrating transcriptomic and proteomic data, we identified coordinated biomolecular modules with potential diagnostic and therapeutic relevance. Although further validation in larger cohorts and functional studies is needed, this systems-level approach offers valuable insights into the molecular underpinnings of AD. It opens new avenues for biomarker discovery and targeted intervention.

Overall, the multi-omics data helped identify key molecular signatures associated with AD. The insights gained from this analysis enhance our understanding of the molecular basis of the disease and pave the way for the development of targeted diagnostic and therapeutic strategies. It is essential to validate these findings in larger and more diverse groups to translate them into clinical applications, ultimately aiming to improve the diagnosis, prognosis, and treatment of AD.

In conclusion, our study identified novel miRNA, mRNA, and protein targets associated to synapses in AD within the same individuals under diseased and normal conditions.

Author contributions

Conceptualization and supervision: SK; experimental performance: SK, ER, AH, BS, MMT, and DR; analysis, interpretation, and validation of data: SK, SG, and ER; writing and original draft preparation: SK, ER and DD; review, editing, and finalization of manuscript: SG, ER, and SK. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Institute on Aging (NIA), National Institutes of Health (NIH), grant number K99AG065645, R00AG065645, R00AG065645-04S1, SARP mini grants TTUHSC EP, The Edward N. & Margaret G. Marsh Foundation, and TTUHSC EP MTM Startup Funds to S.K. S.S.G. is a CPRIT Scholar in Cancer Research. S.S.G. was supported by a First-time faculty recruitment award from the Cancer Prevention and Research Institute of Texas (CPRIT; RR170020). S.S.G. is also partly supported by the NIH 1RO1AI175837-01, Lizanell and Colbert Coldwell Foundation, The Edward N. and Margaret G. Marsh Foundation, The American Cancer Society (RSG-22-170-01-RMC), NIH 1R16GM149497 grants and CPRIT-TREC (RP230420).

Data availability

Data for RNA sequencing is available on NCBI Gene Expression Omnibus (GEO) for the following datasets: mRNA (GSE276756) and miRNA (GSE276898). Datasets for proteomics are available on ProteomeXchange via the PRIDE (EMBL- EBI) database (PXD055784).

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

For AD and control human brain tissues, written informed consent for post-mortem brain donation was obtained from the families of donors through the NIH NeuroBioBank. Prior to their transfer to the TTUHSC El Paso, all samples were de-identified, thereby exempting them from the oversight of the Institutional Review Board (IRB).

Supplementary information

The online version contains supplementary material available at 10.1038/s41380-025-03095-w.