Profiling RNA Cargo in Extracellular Vesicles From hiPSC‐Derived Neurons of Alzheimer's Disease Patients

ABSTRACT

Alzheimer's disease (AD) is a major neurodegenerative disorder that affects more than 55 million people, with an incidence that is projected to triple by 2050. Despite continuous advancements in the field, reliable treatment and early detection strategies remain elusive. Extracellular vesicles (EVs) play a major role in cellular communication throughout the body. In this study, we assessed the cargo of neuronal‐specific EVs for their potential as AD biomarkers. We isolated EVs released from iPSC‐derived excitatory glutamatergic neurons generated from eight AD patients (ADiNEVs) and six healthy controls (iNEVs). We performed RNA‐sequencing and identified significant differences in RNA cargo between ADiNEVs and iNEVs. Notably, fewer small nuclear RNAs (snRNAs) were found in ADiNEVs. RNA transcripts significantly more abundant in ADiNEVs included MT‐CO1, PRR32 and IGSF8 messenger RNAs. We also observed fewer XIST long noncoding RNAs and miR‐7‐5p microRNA content in ADiNEVs. These findings suggest that precision medicine approaches, such as characterising the content of EVs from a patient's own cells, could advance early detection and management of AD.

1. Introduction

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that poses a growing public health crisis among the ageing population. Presently, AD affects more than 55 million people worldwide, with cases expected to triple by 2050 (Scheltens et al. 2021; Nichols et al. 2022). In addition to progressive decline in memory, cognition and productive behaviours, neuropathological hallmarks of AD include the presence of extracellular amyloid‐beta (Aβ) aggregates and intracellular neurofibrillary tangles (NFTs) of hyperphosphorylated tau (Mattson 2004; Hampel et al. 2021). AD pathogenesis is not fully understood, but is known to involve complex relationships between various molecular pathways, including neuroinflammation, blood‐brain barrier disruption, metabolism dysregulation and gene‐environment interactions (Mattson 2004; Knox et al. 2022; Preeti et al. 2022; Boyd et al. 2022). ​

The strongest known risk variant implicated in the development of AD is the ɛ4 allele of apolipoprotein E (APOE) (Fortea et al. 2024); however, the presence of APOE4 is insufficient to cause disease. Genome‐wide association studies (GWAS) have consistently implicated more than 35 genes that may contribute to AD pathogenesis (Wightman et al. 2021), including genes linked to familial AD, such as presenilin 1 (PSEN1), presenilin 2 (PSEN2) and amyloid‐beta precursor protein (APP) (Sherrington et al. 1995; Levy‐Lahad et al. 1995). Despite detailed knowledge of AD genetic architecture, current treatment strategies (e.g., lecanemab and donanemab) target only Aβ aggregation and provide limited symptom delay without halting disease progression (McDade et al. 2022; Gueorguieva et al. 2023). As such, ongoing research efforts are being directed towards unravelling the biological mechanisms that underlying AD aetiology, as well as searching for biomarkers that can be used for early screening and detection.

Extracellular vesicles (EVs) are lipid bilayer‐enclosed vesicles released by all cell types. EVs play an instrumental role in cellular communication by transporting lipids, proteins and RNAs from their originating cells (Witwer and Théry 2019; Welsh et al. 2024; Cheng and Hill 2022). In the central nervous system, various cell types (e.g., neurons, microglia, oligodendrocytes and astrocytes) release EVs that regulate critical functions, including myelin sheath maintenance, synaptic plasticity and neuronal firing (Huang et al. 2023; Schnatz et al. 2021). A number of recent studies have also implicated EVs in the pathophysiology of several neurodegenerative disorders, including AD (Jin et al. 2021; Vandendriessche et al. 2020). EVs have been shown to contribute to the progression of AD by spreading neuroinflammatory factors and misfolded proteins such as Aβ and tau throughout the brain (Rajendran et al. 2006; Baker et al. 2016). Differences in EV RNA cargo, with a historical emphasis on miRNA and other small RNAs, have also been observed in patients with AD or mild cognitive impairment (MCI), thus suggesting a potential link between EVs and AD pathogenesis (Gámez‐Valero et al. 2019; Cheng et al. 2015; Huang et al. 2024).

While such differences in EVs have been observed, their precise role and use as prognostic markers for AD remains to be established. EVs in the circulatory system represent a relatively accessible source of biomarkers that could be used to provide early diagnostic information. However, patient blood contains EVs originating from many cell types throughout the body. As such, it is imperative to determine if there are characteristics of neuron‐specific EVs that can be used to (1) distinguish AD patients from those without cognitive impairment, and (2) identify and isolate EVs originating exclusively from neurons. To address these questions, we focused on excitatory glutamatergic neurons, which are vital for synaptic transmission and cognitive function in the human brain. In AD, these neurons are particularly vulnerable to damage, where their dysfunction contributes to synaptic failure and memory decline (Hardingham and Bading 2010). We hypothesized that EVs released from these neurons may reflect disease‐specific alterations relevant to AD pathology.

In this study, the human induced pluripotent stem cells (iPSCs) were derived from patients with AD and healthy individuals and then differentiated into excitatory glutamatergic neurons. The released EVs were separated from the conditioned medium, and small RNA sequencing of EV cargo was performed. We report significant differences in RNA cargo of EVs from AD patient‐derived neurons compared with controls, suggesting that circulating neuronal EVs may be able to provide patient‐specific information regarding AD status.

2. Methods

2.1. Study Participants and Sample Generation

Prior to sample and data collection, we received approval from the Johns Hopkins Institutional Review Committee (IRB‐00038456), as well as informed consent from all participants involved in this study. As reported in our previously published work (Sagar et al. 2023), human iPSCs were generated from the peripheral blood mononuclear cells (PBMCs) or skin fibroblasts of 14 participants and differentiated into excitatory glutamatergic neurons. In this study, we report data from EVs that were isolated from the cell culture‐conditioned medium of these same iPSC‐derived neurons (Figure 1). Participants included six healthy controls and eight individuals diagnosed with AD by evaluation of clinical pathology by a certified physician (Table 1). For full descriptions of iPSC generation and glutamatergic neuron differentiation, refer to our published methodology (Sagar et al. 2023).

FIGURE 1.

Study workflow.

TABLE 1.

Information on samples used in this study.

Identifier	Diagnosis	Age	Sex	iPSC source	Collection site
JHU‐AD‐01	sAD	68	F	PBMC	ADRC/MATC
JHU‐AD‐02	sAD	60	F	PBMC	ADRC/MATC
JHU‐AD‐03	sAD	70	M	PBMC	ADRC/MATC
JHU‐AD‐04	sAD	62	F	PBMC	ADRC/MATC
JHU‐AD‐05	sAD	72	M	PBMC	ADRC/MATC
JHU‐AD‐06	sAD	55	F	PBMC	ADRC/MATC
JHU‐AD‐07	sAD	70	M	PBMC	S‐CitAD
JHU‐AD‐08	sAD	81	M	PBMC	ADRC/MATC
Mean		67.3
JHU‐WT‐01	Control	56	F	Skin Fibroblast	CIMR
JHU‐WT‐03	Control	81	F	PBMC	ADRC/MATC
JHU‐WT‐04	Control	85	F	PBMC	ADRC/MATC
JHU‐WT‐05	Control	56	F	Skin Fibroblast	CIMR
JHU‐WT‐07	Control	88	F	PBMC	ADRC/MATC
JHU‐WT‐08	Control	84	F	PBMC	ADRC/MATC
Mean		75

Note: S‐CitAD clinical trial (ClinicalTrials.gov Identifier: NCT03108846).

Abbreviations: ADRC, Johns Hopkins Alzheimer's Disease Research Center (ADRC); CIMR, Coriell Institute for Medical Research (CIMR); MATC, Johns Hopkins Memory and Alzheimer's Treatment Center (MATC); sAD, sporadic Alzheimer's Disease.

2.2. EV Separation

EVs were enriched from the cell culture medium of iPSC‐derived neurons from AD patients (ADiNEVs) and controls (iNEVs) using size‐exclusion chromatography (SEC) columns and an automated fraction collector (AFC; IZON, Cambridge, MA, USA). On Day 21, 12 mL of supernatant from each neuronal cell culture was collected and centrifuged at 300 × g for 10 min at 4°C. To deplete cells and cellular debris, the resulting supernatant was transferred to a fresh Falcon tube and centrifuged at 2000 × g for 10 min at 4°C. The supernatant from this second spin was then transferred to a fresh Eppendorf tube and stored at −80^°C until EV extraction.

To initiate the EV separation protocol (Huang et al. 2023; Arab et al. 2021; Huang et al. 2020), the supernatant samples were thawed overnight at 4°C and ultrafiltrated through an Amicon Ultra‐15 centrifugal filter unit (Ultracel‐100KD; 15 mL capacity; Merck Millipore, MA, USA) by centrifuging at 2000 × g for 20 min at 4^°C. Filtrates were loaded onto a SEC column (IZON Smart Columns qEVoriginal 70 nm; IZON Science, Cambridge, MA) and eluted with 6 mL of DPBS. After discarding the void volume (2.8 mL), the first 2 mL fractions (EV‐enriched) were collected and concentrated to a final volume of 200 µL by a second ultrafiltration through an Amicon Ultra‐2 centrifugal filter unit, again by centrifuging at 2000 × g for 20 min, at 4^°C. Samples were stored at −80°C in protein LoBind tubes (Cat#022431064, Eppendorf) until further analysis.

2.3. Nano‐Flow Cytometry

The concentration and size distribution of the separated EVs were measured using the Flow NanoAnalyzer (NanoFCM), according to the manufacturer's instructions and previously published protocols (Huang et al. 2023; Arab et al. 2021; Huang et al. 2023). To calibrate the instrument, 250 nm fluorescently labelled silica beads and the NanoFCM Silica Nanospheres Cocktail #1 (Cat# S16M‐Exo) were used. EV samples were diluted in DPBS (from 1:100 to 1:200) and recorded using NanoFCM 2.0 software for 1 min. Particle size was calculated using a calibration curve based on the flow rate and side scatter intensity.

2.4. Transmission Electron Microscopy

EVs (10 µL; 4.0 × 10⁸ particles/mL) were imaged using transmission electron microscopy (TEM), according to previously published protocols (Arab et al. 2021; Huang et al. 2023; Huang et al. 2020). First, samples were adsorbed onto glow‐discharged, carbon‐coated copper grids (EMS CF400‐CU‐UL) for 2 min and then washed 3 times with dropwise application of 1x Tris‐buffered saline. Next, grids were blotted and negatively stained with 2 drops of 1% uranyl acetate (UAT) and tylose water. The grids were blotted and imaged at 60,000× magnification using a Hitachi 7600 TEM set at 80 kV with XR80 charged‐coupled device (AMT imaging, Woburn, MA, USA). Images were analysed using Image J (Schneider et al. 2012), and all particle structures were manually selected for quantification.

2.5. Single‐Particle Interferometric Reflectance Imaging Sensor

EV phenotype was assessed using a single‐particle interferometric reflectance imaging sensor (SP‐IRIS), as previously described (Huang et al. 2023; Arab et al. 2021; Huang et al. 2020) and following the manufacturer's instructions. Briefly, EVs were diluted to a concentration of 4.0 × 10⁸ particles/ml using 1x Incubation Solution (from Leprechaun Exosome Human Tetraspanin Kit 251–1044, Unchained Labs). A 40 µL of each dilution was incubated on the surface of microarray chips (Leprechaun Exosome Human Tetraspanin Kit 251–1044, Unchained Labs) coated with anti‐tetraspanin antibodies against human tetraspanins CD9, CD63 and CD81, and mouse IgG isotype control. Chips were imaged using an ExoView R100 instrument (NanoView Biosciences, Brighton, MA, USA), and data were analysed using NanoViewer (version 3.0.14) software by NanoView.

2.6. RNA Extraction

Total RNA was extracted from ADiNEVs and iNEVs using a modified protocol combining the Qiagen miRNeasy Mini Kit (Cat# 217004, Qiagen) with Zymo‐Spin Columns (Cat# C1003‐50, Zymo Research). Specifically, the Qiagen kit lysis and wash buffers and Zymo columns were used. RNA was eluted in 40 µL of nuclease‐free water, as described in our previous studies (Huang et al. 2023; Huang et al. 2024).

2.7. Small RNA Sequencing and Data Analysis

For each sample, small RNA (sRNA) libraries were generated from 8 µL of RNA using the BioLiqX Small RNA‐seq Kit (HB202A, Heidelberg Biolabs GmbH). The average size and yield of the small RNA library fragments were assessed using a Fragment Bioanalyzer (DNA 1000, 5067–1505; Agilent) and found to be approximately 175–180 bp, aligning with the expected size range for small RNA libraries, including adapter sequences. Multiplexed libraries were pooled together such that the final concentration of each was 0.5 nM and sequenced on a NovaSeq 6000 using the S4 Reagent Kit v1.5 (200 cycles #20028313; Illumina).

Raw reads were first trimmed from poly A‐tailsusing cutadapt (version 3.4) software (Martin 2011). The trimmed and size‐selected (>15 nt) reads were aligned to custom‐curated hg38 reference transcriptomes using Bowtie 1 (version 1.2.2; allowing one mismatch tolerance) (Langmead et al. 2009). First, reads were mapped to RNA species with low sequence complexity and/or high repeat number, including rRNA, tRNA, RN7S, snRNA, snoRNA, scaRNA, vault RNA, RNY and the mitochondrial genome (mtRNA). Reads that did not map to these references were next aligned to a human miRNA reference, then to references for protein‐coding mRNAs and long non‐coding RNAs (lncRNA). The reads that did not map to the above RNAs were aligned with the remaining transcriptome (other ncRNAs). Finally, all reads that did not map to the human transcriptome were aligned to the human genome reference (rest hg38) corresponding to introns and intergenic regions, as we have done previously (Huang et al. 2024). Differentially expressed (DE) sequences were identified by the R package DESeq2 (version 1.16) (Anders and Huber 2010; Love et al. 2014; Roberts and Pachter 2013) with false discovery rate (FDR)‐adjusted p value (Padj) < 0.05.

Content value and comparison value were used to evaluate the amount of small RNA biotypes in the EV sample. Content value is the percentage of a specific RNA type in each sample, calculated by dividing the amount of RNA type y in sample x by the total amount of all RNA types in the same sample x. Comparison value is calculated by dividing the amount of a specific RNA type y in sample x by the sum of that RNA type y across all samples. If we use A (y | x) to denote the amount of RNA y in sample x (x $\epsilon$[1, n], y $\epsilon$ [1, m]), The calculation of content value can be expressed with the equation:

$$\mathrm{Conten}{\mathrm{t}}_{\mathrm{sample}x,\mathrm{RNA}y}=\frac{\mathrm{A}(y|x)}{{\sum }_{y=1}^m\mathrm{A}(y|x)}$$

Similarly, the calculation of the comparison value can be expressed as:

$$\mathrm{Compariso}{\mathrm{n}}_{\mathrm{sample}x,\mathrm{RNA}y}=\frac{\mathrm{A}(y|x)}{{\sum }_{x=1}^n\mathrm{A}(y|x)}$$

2.8. Quantitative Real‐Time Polymerase Chain Reaction

Quantitative real‐time polymerase chain reaction (qRT‐PCR) was performed to validate differentially expressed (DE) transcripts identified by sRNA sequencing analysis. Complementary DNA (cDNA) was synthesised from total RNA by using the High‐Capacity RNA‐to‐cDNA Kit (Thermo Fisher Scientific, Cat. #4387406) according to the manufacturer's protocol. This reverse transcription includes a 2X RT buffer mix containing deoxynucleotide triphosphates (dNTPs), random octamers and oligo(dT)₁₆ primers. 12 ng of cDNA template was added to each reaction, along with SsoAdvanced Universal SYBR Green Supermix (Bio‐Rad, Hercules, CA, USA), 5 µM of each primer (Table 2), and nuclease‐free water. Four ADiNEV and four iNEV samples, all derived from female donors, were randomly selected for validation. Each reaction was conducted using technical triplicates to ensure reproducibility. For normalising the qRT‐PCR data, we selected GAPDH (Padj = 0.631, Log₂FC = 0.447), as it was found to be nonsignificantly different in our sRNA sequencing data between ADiNEVs and iNEVs (Noerholm et al. 2012). The relative fold change in expression between ADiNEVs and iNEVs was calculated using the ΔΔCq method (Bustin et al. 2009). Whilst no universal consensus exists for reference transcripts in EV studies, GAPDH showed relatively stable expression across ADiNEVs and iNEVs in our RNA‐seq dataset (GEO database under accession ID: GSE297197), supporting its use as a normaliser in this context.

TABLE 2.

qRT‐PCR primer sequences.

Gene	Forward primer (5’ – 3’)	Reverse primer (5’ – 3’)	qRT‐PCR: amplicon size
MT‐ND4	CCCTCGTAGTAACAGCCATTCTC	CGACTGTGAGTGCGTTCGTAGT	136bp
HIST1H1A	GGCAAAGAAACCTGCTAAGGCTG	TAAGAGCTGCCAACGACACACC	128bp
IGSF8	CCTCGCCAAAGCCTATGTTCGA	GGTACACTGTGCCTCCTGCTAG	143bp
XIST	GTAGGTGTGCTGATAACCAAGGC	GGGAAAGGAAGATTGAGGGTGG	128bp
miR7‐5p	ACACTCCAGCTGGGTGGAAGACTAGTGATTT	CTCAACTGGTGTCGTGGAGTCGGCAATTCAGTTGAGACAACAAA	110bp
GAPDH	ACTTTGTCAAGCTCATTTCCTGGTATGAC	GGTACTTTATTGATGGTACATGACAAGGTG	280bp

3. Results

3.1. EVs Separated From AD and Control Samples Exhibit Consistent Size and Phenotype

EVs were separated from the cell culture‐conditioned medium of previously characterised iPSC‐derived excitatory neurons (Sagar et al. 2023) by using an SEC‐column‐based method in combination with additional ultrafiltration washes. RNA from EVs were then extracted and sequenced as shown in the study workflow (Figure 1). We then characterised particle count, morphology and marker proteins of EVs derived from ADiNEVs and iNEVs using NanoFCM, TEM and SP‐IRIS. Our results demonstrate successful separation and characterisation of EVs from the cell culture medium of all iPSC‐derived neurons.

NanoFCM revealed identical size distributions for both groups of samples (AD and control), specifically a population that roughly follows a power‐law distribution down to the sensitivity limit of the assay, as implemented (Figure 2a). Additionally, we characterised both the early SEC fractions (first 2 mL, EV‐enriched) and late SEC fractions (subsequent 4 mL, free protein‐enriched) to confirm the consistency of isolation for our workflow (Figure S1). We also confirmed the presence of similar‐sized EVs using TEM and characterised early and late SEC fractions of EVs (Figure 2b, Figure S2). Using SP‐IRIS, we also confirmed the presence of typical EV tetraspanins CD63, CD81 and CD9 in the captured EV population (Figure 2c) and that the mean EV diameter was not significantly different between tetraspanin‐captured ADiNEVs and iNEVs. Our results show higher detection of CD63 and CD81 than CD9, in addition to CD63 and CD81 co‐localisation (Figure 2d).

FIGURE 2.

EV characterisation. (a) Average size distribution of EVs normalized to the peak of the power‐law distribution. (b) Representative negative staining by transmission electron microscopy (TEM) to visualize EVs (scale bar‐100 nm). (c) The presence and colocalization of tetraspanins CD63, CD81, and CD9 on EVs were analyzed by a Single‐Particle Interferometric Reflectance Imaging Sensor (SP‐IRIS). (d) Size profile of tetraspanin‐captured ADiNEVs and iNEVs by SP‐IRIS.

3.2. Analysis of RNA Biotypes Reveals Less snRNA/RNU Content in ADiNEVs

sRNA‐sequencing of RNA cargo in ADiNEVs and iNEVs yielded an average of 121 million and 162 million reads per sample, respectively. We confirmed that various RNA biotypes were detected in all EV samples (Figure 3a,b). Ribosomal RNAs (rRNAs), transfer RNAs (tRNAs) and small nuclear RNAs (snRNAs; RNUs) were the three most abundant RNA biotypes detected in ADiNEVs and iNEVs (Figure 3a). By contrast, vault RNAs (vRNAs) and microRNAs (miRNAs) were the least represented type of RNA cargo (Figure 3a). Amongst all RNA biotypes, only snRNA/RNU content was significantly less abundant in ADiNEVs (p = 0.02) relative to iNEVs (Figure 3b). The lower level of snRNA/RNU was also observed when only ADiNEVs from female donors and control iNEVs were compared (p = 0.0188). snRNAs/RNUs play a critical role in mediating spliceosome function, and there is evidence that RNA spliceosome components can accumulate and form cytoplasmic aggregates in the brains of AD patients (Hales et al. 2014; Cheng et al. 2021). It is also well‐known that splicing abnormalities are linked to AD susceptibility (Raj et al. 2018). Our results demonstrate that evidence of splicing dysregulation in AD brains can be observed in EVs; thus implicating snRNA/RNU content in neuronal EVs as a potential biomarker for AD.

FIGURE 3.

Small RNA biotypes in EV samples. (a) The proportion of different RNA biotypes in EVs. The content value represents the percentage of a specific RNA type in each sample, calculated by dividing the amount of RNA type y in sample x by the total amount of all RNA types in the same sample (x). The comparison value is calculated by dividing the amount of a specific RNA type y in sample x by the sum of that RNA type y across all samples. The darker colour indicates a higher percentage of RNAs observed in the EV cargo of the samples. (b) Comparison of RNA biotype content between ADiNEVs and iNEVs (*p < 0.05,—= not significant). (c) Principal component analysis (PCA) on small RNA profiles of ADiNEVs and iNEVs.

Finally, following small RNA‐sequencing (sRNA‐seq), we performed principal component analysis (PCA) to assess the overall variance in the dataset. PCA analysis reveals partial separation between ADiNEVs and iNEVs, with ADiNEVs clustering more tightly and iNEVs displaying a broader distribution (Figure 3c). This suggests that there are no batch effects, sequencing biases or major sources of variation in sample quality. Overall, these results demonstrate that we were able to isolate EVs and sequence the RNA cargo in ADiNEVs and iNEVs with a high degree of technical accuracy.

3.3. RNA Cargo Profiles Are Significantly Different in ADiNEVs

Examining individual RNAs, Analysis of sRNA‐seq data also revealed a total of 57 (43 less abundant and 14 more abundant) significant DE transcripts (Padj < 0.05) in ADiNEVs compared with iNEVs (Figure 4a; Table S1). Amongst transcripts with the greatest, most significant fold change differences between ADiNEVs and iNEVs were mitochondrially encoded mRNAs, including cytochrome c oxidase subunit‐I (MT‐CO1; Padj = 1.40 × 10⁻⁷, Log₂FC = 7.25) and NADH‐ubiquinone oxidoreductase chain 4 (MT‐ND4; Padj = 1.72 × 10⁻⁶, Log₂FC = 2.72) (Figure 4b).

FIGURE 4.

RNA content differences in ADiNEVs, relative to iNEVs. (a) Volcano plot showing the Log₂ fold change of significantly differentially expressed RNAs (Padj < 0.01) in ADiNEVs compared to iNEVs. (b) Log₂ RNA level of more abundant and less abundant mRNAs in ADiNEVs (Padj < 0.05). (c) Log₂ RNA level of more abundant and less abundant non‐coding RNAs and pseudogenes in ADiNEVs (Padj < 0.01). ***p < 0.001, **p < 0.01, *p < 0.05.

Significantly more abundant ADiNEV cargos also included six additional mRNA transcripts (HIST1H1A, PRR32, IGSF8, RPS4Y1, DDXY3 and EIF1AY) and six pseudogenes or unannotated transcripts (Figure 4b,c; Table S1). HIST1H1A (Padj = 0.0035, Log₂FC = 6.35) encodes Histone H1.1, IGSF8 (Padj = 0.0477, Log₂FC = 1.63) encodes immunoglobulin superfamily member 8 and PRR32 (Padj = 0.0048, Log₂FC = 1.40) encodes Proline‐rich protein 32. Conversely, NUTM2F mRNA (Padj = 0.0412, Log₂FC = −2.58) and XIST lncRNA (Padj = 0.0302, Log2FC = −2.29) were significantly less abundant transcripts in ADiNEVs (Figure 4b,c; Table S1). The other less abundant transcripts in ADiNEVs included six RP11‐derived ncRNAs, two miRNAs and 33 pseudogenes or unannotated transcripts (Table S1).

3.4. Sex‐Stratified RNA Differences in ADiNEVs and iNEVs

RPS4Y1 (Padj = 0.0005, Log₂FC = 2.9388), DDXY3 (Padj = 0.0450, Log₂FC = 2.2779) and EIF1AY (Padj = 0.0475, Log₂FC = 2.6804) are all Y‐chromosome‐linked genes that were significantly more abundant in ADiNEVs (Table S1). Given that the ADiNEV samples were derived from four female and four male participants, but all control iNEV samples were from female participants, we posited that the appearance of these transcripts could be driven by the presence of males in the ADiNEV group. We thus performed an unsupervised clustering heatmap of all 57 DE RNAs (Figure 5a), which demonstrates that ADiNEVs and iNEVs exhibit distinct clustering patterns. Females and males within the ADiNEV group also clustered independently. Based solely on the significantly DE transcripts between ADiNEVs and iNEVs, we plotted male‐ and female‐derived ADiNEVs on a uniform manifold approximation and projection (UMAP) and also observed sex‐specific clustering (Figure 5b). These results provide further evidence that some of the DE transcripts from our original analysis could be driven by the lack of male‐derived control iNEVs.

FIGURE 5.

(a) Heatmap of differentially expressed RNAs between ADiNEVs and iNEVs (Padj < 0.05). (b) Uniform Manifold Approximation and Projection (UMAP) clustering analysis for ADiNEVs from both female and male samples based on DE RNAs (Padj <0.05) (c) Log₂ RNA level of more abundant and less abundant RNAs in female‐derived ADiNEVs. (d) The validation of relative fold change in the expression of DE RNAs was evaluated by qRT‐PCR.

3.5. Consistent AD‐Related RNA Differences in EVs Across Sex‐Stratified and Whole‐Group Comparisons

Therefore, we re‐analysed the RNA cargo differences between female‐only ADiNEVs and iNEVs by removing the four male‐derived ADiNEVs from our analysis. This analysis revealed a total of 396 (218 less abundant and 178 more abundant; Table S2) significant DE transcripts (Padj < 0.05) between female‐only ADiNEVs and female iNEVs. Amongst the 218 transcripts found to be less abundant, eight were precursor microRNAs (pre‐miRNAs), 11 were mature microRNAs (miRNAs), 22 were RP11‐derived transcripts, 153 were unannotated transcripts, 10 were protein‐coding mRNAs and 14 were other non‐coding RNAs (ncRNAs), including the long non‐coding RNA XIST (Padj = 0.0407, Log₂FC = −1.8939; Figure 5c). Of these, 27 DE transcripts were consistently less abundant in the overall group comparison (comparing all ADiNEVs and all iNEVs) and the sex‐stratified analysis (comparing female‐only ADiNEVs and iNEVs). This subset included 20 unannotated transcripts, 4 RP11‐related RNAs, 3 additional ncRNAs and XIST. Notably, XIST has been previously implicated in AD biology (Yue et al. 2020; Chanda and Mukhopadhyay 2020; Yan et al. 2022). Interestingly, the most significantly less abundant RNA in ADiNEVs is miR‐7‐5p (Padj = 0.0001, Log₂FC = −2.3737; Figure 5c), overexpression of which has been seen to directly inhibit XIST (Li et al. 2020). Furthermore, pre‐miR‐7‐1 (Padj = 0.0023, Log₂FC = −2.8231) and pre‐miR‐7‐2 (Padj = 0.0116, Log₂FC = −3.0333) are also among the top 15 most significantly less abundant RNAs in female‐only ADiNEVs (Table S2).

Conversely, analysis of female‐only ADiNEVs and iNEVs revealed 178 significantly more abundant transcripts, including 93 mRNAs, 5 pre‐miRNAs, 16 RP11‐related RNAs, 7 additional ncRNAs and 57 pseudogenes or unannotated transcripts. Interestingly, only 4 DE transcripts were significantly more abundant in both sets of analyses. Amongst these, MT‐CO1 (Padj = 0.00027, Log₂FC = 6.8124), IGSF8 (Padj = 0.0179, Log₂FC = 2.0112) and PRR32 (Padj = 0.0184, Log₂FC = 1.2675) mRNA transcripts (Figure 5c), as well as CTB‐31O20.6 ncRNA (Padj = 0.0486, Log₂FC = 5.0854) (Table S2) remained significantly more abundant in ADiNEVs. Notably, other reports align with our findings, which suggest that MT‐CO1 upregulation is associated with AD (Zhang et al. 2022; Lunnon et al. 2017; Kim et al. 2020). IGSF8 has been implicated as a phosphorylated Tau‐interacting protein present in NFTs (Drummond et al. 2020) and has also been identified as an enriched protein in AD cerebrospinal fluid proteomes (Modeste et al. 2023). PRR32 is an X‐linked gene on Xq25 with links to intellectual disability and neurodegeneration in amyotrophic lateral sclerosis (ALS) (Watanabe et al. 2016; Shtilbans et al. 2011); however, to our knowledge, PRR32 has not been implicated in AD until now.

The entire list of transcripts significantly more abundant in ADiNEVs was submitted to Enrichr (https://maayanlab.cloud/Enrichr/) for Gene Ontology (GO) analysis, and results suggest that enriched RNA cargos are involved in GO biological processes, “Positive Regulation of Necrotic Cell Death (GO:0010940)” (Padj = 0.008903) and “Regulation of Necrotic Cell Death (GO:0010939)” (Padj = 00.01230). However, amongst the significantly less abundant ADiNEV RNA cargo, there was no evidence of significantly enriched biological processes, cellular components or molecular functions.

To further validate a subset of DE mRNAs identified in female ADiNEVs and iNEVs, we performed qRT‐PCR (Figure 5d). Whilst the results supported the enrichment trends observed in our RNA‐seq analysis for selected transcripts, including MT‐ND4 (p = 0.21), HIST1H1A (p = 0.23), IGSF8 (p = 0.09), XIST (p = 0.11) and miR‐7‐5p (p = 0.55) the differences between ADiNEVs and iNEVs did not reach statistical significance.

4. Discussion

A primary goal of AD research is to prevent or slow progressive neurodegeneration before the emergence of symptomatic cognitive decline. However, a lack of reliable biomarkers for early disease detection represents a major impediment to successful AD management. In this study, we aim to identify candidate biomarkers for AD detection by examining the RNA content of EVs from AD patient‐derived neurons.

Previous studies have examined EVs from human brain tissue (Huang et al. 2024; Cheng et al. 2020; Bub et al. 2022). Whilst these studies offer invaluable insights, brain tissues contain EVs released by heterogenous cell types, including microglia, astrocytes, oligodendrocytes and neurons, and they are also not typically accessible by biopsy. To search for characteristics of neuron‐specific EVs that could be used to distinguish AD patients from those without cognitive impairment, we first generated patient and control iPSC lines and differentiated them into homogeneous populations of excitatory glutamatergic neurons.

In this study, we conducted small RNA profiling of EVs isolated from excitatory glutamatergic neurons, which were derived from eight AD patient‐derived and six healthy control‐derived hiPSC lines. We observed differential RNA profiles between EVs from AD‐iPSC‐derived neurons (ADiNEVs) and EVs from control‐iPSC‐derived neurons (iNEVs) and highlighted biological differences between them that may represent new biomarkers for AD.

We report that the content of small nuclear RNAs (snRNAs; RNUs) was significantly lower in all ADiNEVs, as well as in female‐only ADiNEVs, compared with iNEVs. This may provide novel insights into RNA‐mediated molecular mechanisms of AD. snRNAs are small (∼150 nucleotides) molecules that play a critical role in mediating the function of spliceosomal machinery (Valadkhan and Gunawardane 2013). Previous studies have shown that snRNAs/RNUs localize with tau and paired helical filaments (PHFs; the main component of NFTs), and that U1 snRNA overexpression leads to aggregation and splicing deficits in AD brains (Hales et al. 2014; Cheng et al. 2021).

Our findings also reveal that MT‐CO1, IGSF8 and PRR32 mRNAs are significantly more abundant in ADiNEVs. These three mRNAs represent the most likely biomarker candidates in our dataset. Due to limited sample availability, we were unable to include male‐derived iNEV controls in our analyses. Our original DE analysis was able to detect this biological difference between cases and controls, and implicated significant enrichment of Y‐chromosome‐originating transcripts. However, after excluding male ADiNEVs and querying only the four female ADiNEVs against all six female iNEVs, MT‐CO1, IGSF8 and PRR32 remained significantly more abundant in ADiNEVs. This suggests that MT‐CO1, IGSF8 and PRR32 mRNAs are more abundant in EVs originating from neurons in the brains of AD patients, regardless of sex.

To our knowledge, PRR32 has not been previously associated with AD and may represent a novel, neuronal EV‐specific biomarker for AD. PRR32 is an X‐linked gene that has been implicated in other neurological disorders, including ALS, multifocal motor neuropathy (MMN) and moderate to severe intellectual disability (Watanabe et al. 2016; Shtilbans et al. 2011). IGSF8 and MT‐CO1 have been previously implicated in AD pathogenesis; however, IGSF8 is a critical cell‐surface protein that plays a pivotal role in signalling pathways that directly impact synaptic function and neuronal health within AD‐relevant neuroanatomy (Apóstolo et al. 2020). Furthermore, compelling evidence from other studies shows a marked increase in the IGSF8 protein co‐localising with NFTs in AD patients, as well as significant enrichment of IGSF8 in microglial EVs from an AD mouse model (Drummond et al. 2017; Santiago et al. 2021). Previous reports have also observed MT‐CO1 upregulation in the blood and EVs of patients with AD (Zhang et al. 2022; Lunnon et al. 2017; Kim et al. 2020). Mitochondrial dysfunction is a well‐known aspect of AD biology (Ashleigh et al. 2023), and our results demonstrate that biological events central to AD pathology can be detected in patient‐derived EVs.

By contrast, XIST was significantly less abundant in ADiNEVs. Regardless of whether males were included in the analysis, this significant relationship remained consistent. XIST is essential for X‐chromosome inactivation, such that only 1 X‐chromosome is expressed in a given cell. In agreement with our findings, XIST has been identified as a novel target for AD intervention (Yue et al. 2020), due to detectable levels of miRNAs that regulate XIST in plasma‐isolated EVs (Visconte et al. 2023), as well as the connection between X‐chromosome instability and preferential susceptibility to AD amongst females (Chanda and Mukhopadhyay 2020). In fact, XIST may repress neprilysin (NEP), an enzyme involved in Aβ degradation (Yan et al. 2022). The relationship between the depletion of XIST, miR‐7‐5p, pre‐miR‐7‐1 and pre‐miR‐7‐2 suggests a potential mechanistic explanation. XIST has been seen to repress NEP in mice (Yan et al. 2022), and miR‐7 has been seen to repress XIST in human cell lines (Li et al. 2020). The significant depletion of pre‐miR‐7‐1, pre‐mirR‐7‐2 and miR‐7‐5p in ADiNEVs suggests that miR‐7‐5p may be downregulated in AD neurons. With less available miR‐7 to repress XIST, XIST may be more likely to sequester to repress NEP, which would render NEP less capable of Aβ degradation. During this process, dysregulation of miR‐7 in AD neurons may be reflected in ADiNEVs. Furthermore, changes in the packaging of XIST RNA into EVs could support the lower abundance of XIST observed in ADiNEVs. This could lead to the further accumulation of XIST intracellularly, resulting in a decrease in Aβ degradation ability in AD neurons (Squadrito et al. 2014; O'Grady et al. 2022). However, since the RNA profile of ADiNEVs does not directly reflect the RNA profile of the AD neurons from which they were derived, future studies will be focused on addressing these relationships between the AD neurons, ADiNEVs and XIST. Nonetheless, the strength of observable XIST reduction in female participant‐derived ADiNEVs further demonstrates that neuronal EVs from AD patients represent a source of easily accessible, AD‐relevant biomarkers. While our study was not designed to investigate genomic variants, such as sex‐linked autosomal SNPs, we recognise that genetic background could contribute to the transcriptomic differences observed between ADiNEVs and iNEVs, particularly in a sex‐specific manner. Given the known sex‐related differences in AD risk and progression, integrating SNP data with EV transcriptomic profiles in future studies may help clarify the molecular mechanisms underlying these differences.

We also observed 20 unannotated transcripts, 4 RP11‐related RNAs and 4 additional ncRNAs that were significantly dysregulated in ADiNEVs and iNEVs across both sets of analysis. Whilst these sequences are not fully characterised, and their specific roles are unknown, they likely represent aspects of key disease‐relevant pathways. Further investigation into the dysregulation of these RNAs within EVs may uncover new insights into RNA‐mediated molecular mechanisms of AD.

In summary, we observed significant differences in the RNA cargo profiles of EVs isolated from iPSC‐derived excitatory glutamatergic neurons that were generated from individuals with AD, compared with EVs isolated from iPSC‐derived neurons generated from healthy controls. Through bioinformatic analyses, we identify differences in RNA content and DE transcripts in EVs that may play a role in AD pathology. Overall, we show that the RNA cargo of neuronal EVs may reflect biologically pertinent events in the onset and progression of AD, including splicing dysregulation, mitochondrial dysfunction and the clearance of Aβ plaques and NFTs. Therefore, we demonstrate that patient‐derived neuronal EVs represent a reservoir of potential biomarkers that could be used for precision medicine approaches in early AD screening, diagnosis and management.

Author Contributions

Ram Sagar: writing – original draft, methodology, formal analysis, investigation, validation. Yiyao Huang: investigation, writing – original draft, formal analysis. Daiyun Dong: writing – review and editing, validation, formal analysis, software, data curation. Rachel J. Boyd: writing – review and editing, formal analysis, data curation. Waqar Ahmed: writing – review and editing, data curation. Kenneth W. Witwer: writing – review and editing, project administration. Vasiliki Mahairaki: conceptualization, funding acquisition, writing – review and editing, visualization, project administration, supervision, resources.

Conflicts of Interest

K.W.W. is president of the International Society for Extracellular Vesicles; is or has been an advisory board member of B4 RNA, Dart Biosciences, Everly Bio, Interactome Biotherapeutics, Libera, NeuroDex, NovaDip, and ShiftBio; holds stock options with NeuroDex; and privately consults as Kenneth Witwer Consulting.

Supporting information

Supplementary Fig.1: jex270074‐sup‐0001‐figureS1.jpg

Supplementary Fig.2: jex270074‐sup‐0002‐figureS2.jpg

Supplementary Tables: jex270074‐sup‐0003‐tablesS1‐S2.xlsx

Data Availability Statement

Small RNA sequencing data of the iPSC‐neuron derived EVs used in this study are available at the GEO public database under the accession number GSE297197.