Conserved and Tissue‐Specific RNA and Protein Cargos of Small Extracellular Vesicles From 120 Tissue Sites Across 34 Organs in a Single Pig

ABSTRACT

Small extracellular vesicles (sEVs) carry diverse cargo essential for intercellular communication, but their organism‐wide characterization across all tissues is still limited. Here, we present a multi‐omics sEV atlas covering the whole‐body of 120 tissue sites, four body fluids and nine gastrointestinal contents from the same pig. We identified 7995 proteins, 337 miRNAs as well as 105,409 transcripts (93,409 full‐length) in sEVs by mass spectrometry, next‐ and third‐generation RNA sequencing. Among these, a conserved cargo set of 252 proteins, 18 miRNAs and 1227 transcripts was defined, including RNA‐binding proteins (ANXA2, ALIX and MVP), mitochondrial proteins (ATP5F1A, COX6C and VDACs), the let‐7 miRNA family and ferritin genes (FTH1/FTL). Notably, sEV biogenesis‐associated ACTG1, CFL1 and ANXA2 were detected in all samples, suggesting their potential as novel sEV protein markers. In contrast, 2638 proteins, 152 miRNAs and 1000 transcripts showed organ‐specific expression patterns, which are likely to underpin organ‐specific biological processes. Moreover, by using this multi‐omic dataset and additional RIP assays, we found that HNRNPU participates in the sorting of hypoxia‐responsive EPAS1 into sEVs. This comprehensive sEV atlas establishes a foundation for future studies in sEV cell biology.

A whole‐body map of sEVs from a single pig, including 133 samples spanning 10 systems, revealed striking tissue‐specific heterogeneity in yield. Importantly, a conserved core cargo set comprising candidate sEV markers (ACTG1, CFL1, and ANXA2), RNA‐binding proteins, the let‐7 miRNA family, and ferritin genes was defined from 105,409 (88.6% full‐length) nanopore transcripts, 337 miRNAs, and 7,995 proteins. Furthermore, HNRNPU regulates the sorting of transcription factor EPAS1 mRNA into sEVs, a finding supported by multi‐omics analysis and RIP assays.

1. Introduction

Extracellular vesicles (EVs), key mediators of intercellular communication, comprise diverse subpopulations, including exosomes, ectosomes, apoptotic bodies and other vesicles (Dixson et al. 2023). Among these, exosomes (diameter <200 nm) represent the most extensively studied subpopulation (Lötvall et al. 2014). They mediate intercellular communication by delivering RNAs, proteins and other biomolecules to recipient cells, thereby regulating gene expression and inducing biological effects (Kalluri and LeBleu 2020). Given the technical challenges in defining their specific biogenesis pathways, and following MISEV2023 guidelines (Welsh et al. 2024), we use the operational term ‘small EVs’ (sEVs) throughout this study rather than ‘exosomes’.

The heterogeneity of sEVs has been widely recognized through studies of body fluids and cell lines (Kalluri and LeBleu 2020; Bu et al. 2026). More recently, isolating sEVs directly from tissues has gained significant attention, as it provides a physiologically relevant in vivo snapshot while exhibiting tissue‐specific heterogeneity (Crescitelli et al. 2025). For instance, ATPase Na/K transporting subunit alpha 3 (ATP1A3) is abundantly expressed in brain‐derived sEVs (You et al. 2023), while proteins such as fibroblast activation protein (FAP) are significantly upregulated in tumour‐derived sEVs (Cvjetkovic et al. 2024). Current research on tissue sEVs is predominantly centred on tumours and the central nervous system (Crescitelli et al. 2025, 2021, 2020). Although recent studies have begun to profile sEV cargo across other major tissues such as heart, liver, and brain, revealing inter‐organ specificity (e.g., ASGR1 for liver sEVs vs. CASQ2 for heart sEVs) (Watanabe et al. 2025; Abdelmohsen et al. 2023) and intra‐organ heterogeneity (e.g., region‐dependent differences in neuronal markers, with NCAM, CD271, and NRCAM more abundant in medulla, cerebellum, and occipital region, respectively) (Huang et al. 2023), comprehensive studies across diverse organ systems remain lacking, and this lack has fundamentally limited our understanding of systemic sEV composition across multiple tissues within an organism, including their underlying conserved patterns and tissue‐specific variations.

sEVs mirror the physiological state of their parent cells (Fowler et al. 2025) and hold great promise for disease diagnosis and therapeutic development (Cheng and Hill 2022). Their cargo form interconnected networks that reveal both the underlying sorting mechanisms (Fabbiano et al. 2020) and the coordinated regulation of post‐delivery biological functions (Blaser et al. 2023). The sEV proteins serve dual roles as functional effectors and vesicle identity markers (e.g., CD9, CD63, TSG101) (Welsh et al. 2024). They also exhibit tissue‐ and cell type‐ specific expression patterns (Garcia‐Martin et al. 2022; Abdelmohsen et al. 2023), where markers such as ATP1A3 (You et al. 2023), and mitochondrial MT‐CO2/COX6c (Jang et al. 2019) facilitate the identification and tracing of tissue‐derived sEVs within biofluids such as blood, which highlights their potential in improving diagnostic predictions for Alzheimer's disease and cancer. However, these markers, identified from a limited set of tissue‐derived sEVs, may lack sufficient specificity in complex body fluids. Critically, current sEV identity markers have been primarily identified from in vitro cultured cells (Kugeratski et al. 2021), their generalizability across tissue‐derived sEVs requires further validation.

RNAs, including miRNAs and mRNAs, represent another class of extensively studied sEV cargo. sEV‐miRNAs mediate post‐transcriptional gene silencing in recipient cells (O'Brien et al. 2020) and serve as promising diagnostic biomarkers (Mori et al. 2019). While body fluid‐derived sEVs permit miRNA detection (Min et al. 2019), diagnostic precision requires tissue‐specific profiling (Barnett et al. 2023). Yet, current systematic tissue‐level insights stem from only five cell lines (Garcia‐Martin et al. 2022), leaving tissue‐level variation largely unexplored. sEV‐encapsulated mRNAs have attracted considerable interest due to their translational potential in recipient cells (Lai et al. 2015). The growing therapeutic applications of sEV‐mRNAs, particularly for targeted therapies (Dong et al. 2023) and CRISPR‐Cas delivery systems (Stranford et al. 2024), underscore the need to characterize the endogenous transport capacity of full‐length mRNAs in native sEVs. While third‐generation sequencing (TGS) has verified full‐length mRNA encapsulation within sEVs (Padilla et al. 2023; Ransom et al. 2024; Yu et al. 2023), preliminary evidence of tissue‐specific differences in this capacity calls for systematic investigation to advance sEV‐mRNA‐based targeted delivery platforms.

A multi‐tissue, multi‐omics sEV atlas is essential for systematically characterizing sEV variation. Pigs serve as a valuable biomedical model for this purpose due to their genomic, physiological and anatomical similarities to humans (Lunney et al. 2021). Here, we established an extensive porcine sEV atlas comprising 133 samples (including 120 tissue sites), which integrates sEV characterization with multi‐omics profiling (proteomics, small RNAs and transcriptomics) using mass spectrometry and next‐generation/third‐generation sequencing. To enable a balanced, systematic comparison across the 133‐sample atlas, we adopted a broadly applicable proteolytic protocol that relies on trypsin's well‐established versatility in primary cell isolation across highly varied tissue types, such as skin (Blanco et al. 2016), brain (Lim and Alvarez‐Buylla 1999), muscle (Miersch et al. 2018) and lung (Barkauskas et al. 2013). Subsequent multi‐omics analysis included: (1) identification of conserved and organ‐specific sEV cargo; (2) systematic comparison of mRNA expression profiles across sequencing platforms; and (3) integrative multi‐omics network analysis of sEV cargo interactions. This dataset provides the most complete characterization of sEV cargo variation to date, establishing a robust foundation for source tracking and clinical applications of sEVs.

2. Materials and Methods

2.1. Animals

Animals for this study were selected from an F12 population, which was derived from a crossbreeding program involving four Chinese indigenous and four Western commercial pig breeds (Yang et al. 2022). Specifically, two 240‐day‐old, healthy full‐sibling pigs with no antibiotic treatment for 1 month prior to slaughter were selected for sample collection based on standard veterinary observation and proper rearing conditions. A total of 133 samples were collected from these two pigs, with 127 obtained from a castrated boar and the remaining six (female reproductive tissues) obtained from a full‐sibling sow. Additionally, fresh liver and lung tissues from three commercial pigs were sourced from a slaughterhouse to evaluate the impact of enzymatic digestion (trypsin vs. a Dispase II/Collagenase II cocktail) on sEV isolation from tissues.

2.2. Tissue Collection and Preprocessing

After slaughter, tissues were rinsed with 1× PBS (E607008, Sangon Biotech), minced into soybean‐sized pieces and flash‐frozen in liquid nitrogen. For trypsin digestion, tissues were thawed at 37°C, rinsed twice with 1× PBS, and cut into 1 mm³ fragments in a sterile petri dish. For each tissue type, three 0.2 g aliquots (totalling ∼0.6 g) were prepared and each digested with 2 mL of 0.25% trypsin (25200‐072, Gibco) at 37°C for 2 h until the suspension became turbid. Digestion was terminated in each aliquot by adding 3 mL of 10% sEV‐depleted DMEM (SH30243.1, Hyclone). The resulting three mixtures were then pooled and centrifuged at 2000 × g for 15 min at 4°C to remove cells and debris.

For Dispase II/Collagenase II cocktail digestion (Watanabe et al. 2025), each 0.2 g tissue aliquot was incubated in 2 mL of HEPES buffer (100 mM HEPES [H4034, Sigma–Aldrich], pH 7.4, 2.5 mM CaCl₂) supplemented with Collagenase II (10 mg/mL, S10054, Yuanye Biotechnology) and Dispase II (10,000 PU/mL, S25046, Yuanye Biotechnology). After 1 h of digestion at 37°C, the reaction was terminated by adding 2 mL of 2 mM EDTA to each tube. Each mixture was passed through a 70 µm cell strainer (352350, Falcon). Following the same procedure as for trypsin digestion, the resulting three filtrates were pooled and centrifuged at 2000 × g for 15 min at 4°C to eliminate residual cellular debris.

2.3. Collection and Preprocessing of Plasma, Urine, Bile and Cerebrospinal Fluid

Whole blood was collected into sterile 50 mL centrifuge tubes containing EDTA·2k (G0280, Solarbio) anticoagulant and kept on ice for 30 min. Two‐step centrifugation (1900 × g for 10 min and 3000 × g for 15 min at 4°C) was performed to obtain the plasma supernatant. For sEV isolation, which was conducted in two batches, 25 mL of plasma was quickly thawed at 37°C before each isolation and diluted with 1× PBS to 100 mL.

The collection and preprocessing of urine, bile and cerebrospinal fluid (CSF) were performed as follows: Mid‐stream urine and urine stored in the bladder were collected in sterile 50 mL centrifuge tubes. In addition, bile and CSF were collected using sterile syringes from the gallbladder and the foramen magnum, respectively. Following 10–20 min of ice storage, the three body fluids were centrifuged at 3000 × g at 4°C (urine and CSF: 15 min; bile: 10 min) to remove cells and cellular debris, resulting in 200 mL of cell‐free urine, 24 mL of cell‐free bile and 8 mL of cell‐free CSF. Prior to sEV isolation, the 24 mL of bile was diluted with 1× PBS to a total volume of 100 mL. Finally, all three cell‐free body fluids were stored at −80°C.

2.4. Collection and Preprocessing of Gastrointestinal Contents

Immediately after slaughter, gastrointestinal contents were collected using sterile spoons and flash‐frozen in liquid nitrogen. Approximately 2 g of gastrointestinal contents were thawed at 37°C and then homogenized in 50 mL of 1× PBS. The mixture was centrifuged at 1000 × g for 10 min at 4°C to remove contaminants including cells, cellular debris and undigested food particles. The supernatant was collected and stored at −80°C.

2.5. Isolation of sEVs by Differential Centrifugation (DC) Combined With Sucrose Cushion Ultracentrifugation (SCUC)

The frozen pre‐processed samples were thawed at 37°C and centrifuged at 2000 × g for 30 min to remove cellular debris, retaining the supernatant. The supernatant was then centrifuged at 10,000 × g for 45 min to remove large EVs and subsequently passing through a 0.45 µm filter. Crude sEVs were pelleted by ultracentrifugation of the filtrate at 100,000 × g for 90 min (total run time). The pellet was resuspended in 1 mL of pre‐cooled 1× PBS, and then gently layered onto a 250 µL cushion of 30% sucrose prepared in deuterium oxide (D807644‐10×0.55 mL, Macklin). After centrifugation at 100,000 ×g for 90 min (total run time), the lower 250 µL of the 30% sucrose cushion was collected, diluted to 3 mL with 1× PBS and centrifuged again at 100,000 × g for 90 min (total run time) to wash the sEVs. Finally, sEVs derived from 0.6 g of tissue were resuspended in 300 µL of 1× PBS, while sEVs isolated from gastrointestinal contents, plasma, urine, bile and CSF were resuspended in 200 µL of 1× PBS. All centrifugation steps were conducted at 4°C and all sEV suspensions were stored at −80°C.

2.6. Nano Flow Cytometry (NanoFCM)

The size distribution and concentration of sEVs were analysed using a Flow NanoAnalyzer (N30E, NanoFCM Inc., China). The system was first calibrated with 200 nm polystyrene beads (2.01 × 10¹⁰ particles/mL), followed by size calibration using four monodisperse silica bead standards (68 nm, 91 nm, 113 nm and 155 nm). For analysis, each sEV sample (10 µL) was diluted with 1× PBS to a final volume of 30 µL to achieve an optimal particle count rate of 2000–8000 particles/min. Data were acquired for 1 min at an operating pressure of 1.0 kPa. Particle concentration and size distribution were analysed using NanoFCM's NF Professional software (v1.08), with background signals from 1× PBS subtracted from all measurements.

2.7. Transmission Electron Microscopy (TEM)

For TEM analysis, 10 µL of sEV suspension was adsorbed onto a copper grid for 1 min, and excess liquid was blotted using filter paper. The sample was then negatively stained with 10 µL of 2% uranyl acetate for 1 min, followed by removal of excess stain. After air‐drying at room temperature, images were acquired using a transmission electron microscope (HT‐7700, Hitachi) operating at 100 kV.

2.8. Detergent Sensitivity Assay

For the membrane integrity assay, 9 µL of sEVs were incubated with Triton X‐100 (9002‐93‐1, Macklin) at a final concentration of 10% on ice for 1 h. The particle concentration was subsequently measured using NanoFCM to determine the proportion of detergent‐labile vesicles.

2.9. Proteinase K Protection Assay

To verify that protein cargo was encapsulated within the sEV lipid membrane, a Proteinase K (PK) protection assay was performed. Intact sEVs were compared against membrane‐disrupted controls, which were pre‐incubated with 0.5% Triton X‐100 on ice for 1 h. Both groups were treated with 1 µg/mL PK (S10085, Yuanye) for 5 min. The reaction was subsequently quenched with 5 mM PMSF (ST507, Beyotime Biotechnology), incubated on ice for 10 min, before further analysis by Western blotting to assess the degradation of internal proteins.

2.10. Protein Extraction and Concentration Determination

For tissue protein extraction, samples were ground into a fine powder using liquid nitrogen in a pre‐cooled mortar. The resulting tissue powder was resuspended in 1× RIPA buffer supplemented with protease inhibitors, including PMSF (ST507, Beyotime Biotechnology) and a Protease Inhibitor Cocktail (HY‐K0011, MCE). The suspension was processed using a tissue homogenizer (SCIENTZ‐48, SCIENTZ) to ensure thorough lysis. The lysates were then centrifuged at 15,000 × g for 15 min at 4°C, and the resulting supernatant was collected. For sEV protein extraction, an sEV suspension (10 µL) was lysed in 5× RIPA buffer (1 M Tris‐HCl [pH 7.4], 5 M NaCl, 100% NP40, sodium deoxycholate, 10% SDS and ddH2O) on ice for 30 min to release the sEV proteins. Protein concentrations for both tissue and sEV samples were determined using a BCA kit (P0011, Beyotime Biotechnology). Briefly, 20 µL of the sample was mixed with BCA working reagent and incubated at 37°C for 30 min. The absorbance was measured at 562 nm using a microplate reader (Varioskan LUX, Thermo), and concentrations were calculated based on a standard curve.

2.11. Protein Detection by Mass Spectrometry and Data Analysis

For proteomic preparation, an aliquot of 20–50 µL of the sEV suspension was used for tryptic digestion. Peptides generated by digestion were analysed using a NanoElute nanoLC system coupled to a timsTOF Pro2 mass spectrometer (Bruker) in 4D label‐free data‐dependent acquisition (DDA) mode. To improve comparability of proteomic profiling across samples, a constant enzyme‐to‐substrate ratio (1:50) was maintained during digestion, and equal amounts (200–300 ng) of peptides were loaded for each LC‐MS/MS run to minimize technical variation during data acquisition.

Following column equilibration with 0.1% formic acid, peptides were separated on a C18 analytical column (25 cm × 75 µm, 1.6 µm; IonOpticks) at 300 nL/min using a binary gradient of 0.1% formic acid (solvent A) and 0.08% formic acid in 80% acetonitrile (solvent B). Mass spectrometry analysis was conducted in positive ion mode for 70 min (m/z range 100–1700) with parallel accumulation‐serial fragmentation (PASEF) detection, employing the following parameters: ion mobility range 0.6–1.6 V·s/cm², capillary voltage 2000 V, drying gas flow 3 L/min at 180°C and collision‐induced dissociation energy 42 eV.

Mass spectrometry data were searched against the Sus scrofa UniProt reference proteome (UP000008227, accessed 27 December 2023) supplemented with potential common laboratory contaminants (e.g., keratins, bovine serum albumin and trypsin) from MaxQuant's built‐in database using MaxQuant (v2.4.11.01) (Cox and Mann 2008). Search parameters included: tryptic digestion with up to two missed cleavages, peptide mass range ≤4600 Da, minimum peptide length 7 amino acids, precursor mass tolerance 10 ppm and fragment mass tolerance 0.005 Da. A 1% false discovery rate threshold was applied at both peptide and protein levels. Post‐translational modifications included variable N‐terminal acetylation and methionine oxidation, with fixed carbamidomethylation of cysteine residues. Protein were quantified using both intensity‐based absolute quantification (iBAQ) and label‐free quantification (LFQ) approaches. Next, the MaxQuant‐generated proteinGroups.txt file was processed in Perseus (v2.0.11) (Tyanova et al. 2016) to remove the potential common laboratory contaminants, reverse hits and proteins identified only by sites. Only proteins containing at least one unique peptide were retained. LFQ intensities were log₂(x + 1) transformed for cross‐tissue comparison.

2.12. Extraction and Concentration Measurement of Total RNA From sEVs

Total RNA was extracted from sEVs using the miRNeasy Mini Kit (217004, Qiagen), according to the manufacturer's instructions. RNA quantification was performed using the Quantus^TM Fluorometer (Promega) with the QuantiFluor RNA System solution (E3310, Promega). RNA integrity was assessed using the Qsep100 automated electrophoresis system (BIOptic) with input amount adjusted based on the measured concentration.

2.13. Small RNA Sequencing and Analysis

Small RNA libraries were prepared using the QIAseq miRNA Library Kit (Qiagen, 331505), which captures small RNA molecules with 3'‐hydroxyl and 5'‐phosphate groups. The final cDNA libraries were sequenced on an Illumina NovaSeq 6000 platform generating 150 bp single‐end reads. Raw sequencing data were quality‐trimmed (Q score ≥ 10, Length ≥ 16 nt) using Cutadapt (v1.11) (Martin 2011), and followed by the removal of contaminants and rRNAs using exceRpt (v5.0.0) (Rozowsky et al. 2019). The filtered reads were then aligned to the Sus scrofa reference genome (v11.1) (Harrison et al. 2024) and a transcriptome database incorporating miRBase 22 (Kozomara et al. 2019), GtRNAdb (Chan and Lowe 2016), RNAcentral (Petrov et al. 2017), Ensembl GTF v109 (Harrison et al. 2024) and CircAtlas3.0 (Wu et al. 2024) via exceRpt (1‐nt mismatch, minimum length 16 nt), enabling quantification of miRNAs, tRNAs, piRNAs, annotated RNAs and circRNAs. For miRNA target prediction, the miRanda v3.3a algorithm was employed to analyze 3' untranslated regions (3'UTRs) from the porcine reference genome, applying stringent parameters including a minimum alignment score of 150, a maximum binding free energy of −10 kcal/mol, and mandatory seed region complementarity (Zhong et al. 2023).

2.14. Next‐Generation Sequencing and Analysis

Poly(A)‐tailed RNA libraries were prepared using the RNALib Single Cell WTA Kit (A5001, Lifeint) following the manufacturer's protocol. Briefly, first‐strand cDNA was synthesized with oligo‐dT primers and reverse transcriptase, followed by sequential steps of second‐strand synthesis, 15–16 cycles of PCR amplification, and magnetic bead‐based purification. The purified cDNA was used to prepare Illumina‐compatible libraries with the Lifeint Transpose DNA Library Prep Kit (A5005), involving transposase‐mediated fragmentation (∼300 bp), 14‐cycle PCR amplification and size selection. Final libraries were sequenced on an Illumina NovaSeq 6000 platform (150 bp paired‐end reads). Raw sequencing data were processed with fastp (v0.23.4) (Chen 2023) for quality control. Clean reads were aligned to the Sus scrofa reference genome (Ensembl release 11.1) using HISAT2 (v2.1.0) (Kim et al. 2019). Gene expression quantification was performed using StringTie (v2.1.5) (Pertea et al. 2015) base on the Sus scrofa annotation (Ensembl release 109) (Harrison et al. 2024).

2.15. Third‐Generation Sequencing and Analysis

The cDNA library was prepared using the cDNA‐PCR Barcoding Kit V14 (SQK‐PCB114.24, Oxford Nanopore Technologies). Total RNA (100–200 ng) was first ligated with polyT‐containing cDNA‐RT adapters followed by adapter digestion. Reverse transcription with strand switching was then performed to generate cDNA. The sample was purified using 0.5× AMPure XP beads (Beckman Coulter) to remove cDNA fragments smaller than 500 bp, followed by 15 cycles of PCR amplification and another round of 0.5× AMPure XP bead purification. Rapid sequencing adapters were subsequently ligated to the cDNA, and the final PCR‐cDNA library was sequenced on a PromethION 48 sequencer (Oxford Nanopore Technologies) using Flow Cell R10.4.1 with five samples multiplexed per flow cell. Base calling was performed on the generated POD5 files using Dorado (v0.7.1) (https://github.com/nanoporetech/dorado) with the dna_r10.4.1_e8.2_400bps_sup@v5.0.0 model at a quality threshold of 7, followed by demultiplexing into barcode‐specific files.

Full‐length reads (FLRs) were identified using Pychopper (v2.7.9) (https://github.com/nanoporetech/pychopper), and then aligned to the S. scrofa 11.1 (Harrison et al. 2024) reference genome using minimap2 (v2.17‐r941) (Li 2018). Gene and transcript quantification was performed using IsoQuant (v3.4.1) (Prjibelski et al. 2023) with the BAM files. Novel transcripts classified as either ‘novel in catalog’ or ‘novel not in catalog’ that were detected in at least two samples were used to generate an enhanced annotation file. This enhanced annotation served as the reference for a second round of quantification. Alternative splicing analysis was conducted using SUPPA (v2.4) (Trincado et al. 2018) to identify seven event types: alternative 3' splicing (A3), alternative 5' splicing (A5), alternative first exon (AF), alternative last exon (AL), retained intron (RI), skipped exon (SE) and mutually exclusive exons (MX). Additional characterization of alternative untranslated regions (UTRs) and subsequent analysis and visualization of splicing events followed established methods by Glinos et al. (Glinos et al. 2022).

2.16. Analysis of Conservation and Organ Specificity of sEV Cargo

sEV cargo quantification was performed using method‐specific normalization: label free quantification (LFQ) for proteins, reads per million mapped reads (RPM) for small RNAs, and ranscripts per million (TPM) for NGS/TGS transcripts. Preliminary filtering was applied to retain cargo components: In a given tissue, proteins with LFQ > 0, small RNAs with RPM > 1, and genes/transcripts with TPM > 1 were retained. To evaluate the organ specificity and conservation of sEV cargo, we generated organ‐level expression matrices by calculating the mean expression values for proteins [log₂(LFQ+1)], miRNAs (RPM) and genes (TPM) across sEV samples from the same organ. The specificity index tau (τ), was employed to quantify the degree of organ‐specific expression (Yanai et al. 2005; Lüleci and Yılmaz 2022), defined as follows:

$$\tau =\frac{{\sum }_i^N\left(1-x_i\right)}{N-1}$$

where $N$ represents the total number of organs and $x_i$ denotes the expression level in organ $i$ normalized by the maximum expression value observed across all organs.

Based on these metrics, sEV cargo was first classified as ‘conserved expression’ if they were present in >95% of organs, with proteins (LFQ > 0), miRNAs (RPM > 10), or genes (TPM > 1). To identify the high‐confidence conserved sEV proteins shown in Figure 2f, we applied a further filtering workflow: proteins were restricted to those detected in all 131 samples and classified as MISEV2023 Category 2 (cytosolic proteins). Among the remaining cargo, ‘organ‐specific expression’ for proteins was defined as tau > 0.85 and LFQ > 0 in at least one organ. The miRNA and gene datasets were further categorized into four expression patterns based on TPM/RPM values and tau thresholds: (1) ‘Null expression’ (expression ≤1 in all organs), (2) ‘Wide‐spread expression’ (tau < 0.85 and expression > 1 in ≥ 1 organ), (3) ‘Weak expression’ (tau ≥ 0.85 and expression ≤10 in all organs) and (4) ‘Specific expression’ (tau ≥ 0.85 and expression >10 in ≥1 organ).

FIGURE 2.

Comprehensive overview of sEV protein cargo, including number, subcellular localization, classification, and abundance. (a) Number of detected proteins per organ. Black dots represent samples from distinct regions within each organ. Boxes are colored by organ. (b) Top 10 GO Cellular Component terms ranked by human −log₁₀(adjusted p value), shown with corresponding enrichment results in the pig database. (c) Distribution of non‐redundant annotated proteins across organs for selected GO Cellular Component terms based on the human database. Left panel: Violin plots display the distribution of protein counts, with adjacent pie charts illustrating the proportion of proteins in each category that overlap with Extracellular vesicles. Red and grey segments represent overlapping and non‐overlapping proteins, respectively. Right panel: Radar chart showing the proportion distribution of these proteins across various organs. Colors of GO terms, violin plots, and data points correspond to the respective cellular components. (d) Overlap of detected proteins with the Vesiclepedia (Chitti et al. 2024) and Exocarta (Keerthikumar et al. 2016) databases. (e) Classification of detected proteins according to UniProt (UniProt Consortium 2025) and the Human Protein Atlas (Uhlén et al. 2015). (f) Expression heatmap of high‐confidence conserved sEV proteins detected across all samples. Selection criteria: conserved across >95% of organs, detected in all 131 samples, and classified as MISEV2023 Category 2 (cytosolic proteins). Column color coding for organs is as in panel a.

To identify tissue‐conserved genes, we downloaded gene expression profiles from RNA‐seq of 226 samples from Karlsson et al. (2022) and generated an expression matrix for 64 tissue types by averaging values across samples from the same tissue. Using these profiles, genes detected in ≥95% of tissues (at least 61 out of 64 tissues) were classified as tissue‐conserved.

2.17. Gene Enrichment Analysis

Gene Ontology (GO) enrichment (human and pig, maxGSSize = 5000) and Kyoto Encyclopaedia of Genes and Genomes (KEGG) pathway (pig only) analyses were performed using clusterProfiler package (v4.12.0) (Yu et al. 2012), with a p value ≤ 0.05 considered as significant.

2.18. Western Blotting (Wb) Analysis

Protein samples were denatured at 95°C for 5 min, separated by 12% sodium dodecyl sulphate‐polyacrylamide gel electrophoresis (SDS‐PAGE) under a constant voltage of 80–100 V, and transferred onto methanol‑activated polyvinylidene difluoride (PVDF) membranes (IPVH00010, Millipore) at a constant current of 300 mA (1 min/kDa) in an ice bath. Membranes were blocked with 5% (w/v) non‑fat milk in Tris‑buffered saline with Tween 20 (TBST) for 1 h at room temperature, followed by overnight incubation at 4°C with primary antibodies (1:1000; see Table 1). After three 10‑min washes with TBST, membranes were incubated with HRP‑conjugated goat anti‑rabbit IgG (1:5000; 31460, Invitrogen) for 1 h at room temperature. After three additional 10‑min washes with TBST, protein bands were visualized using ECL substrate (1810212, CLINX) and captured with a chemiluminescence imaging system (ChemiScope 3000 mini, CLINX).

TABLE 1.

Information of primary antibodies.

Target protein	Catalog No.	Manufacturer	Dilution
CD81	41779	SAB	1:1000
TSG101	28283‐1‐AP	Proteintech
Alix	12422‐1‐AP	Proteintech
Syntenin‐1	ab19903	Abcam
ACTG1	11227‐1‐AP	Proteintech
ANXA2	11256‐1‐AP	Proteintech
CFL1	10960‐1‐AP	Proteintech
NNMT	15123‐1‐AP	Proteintech
SFTPC	10774‐1‐AP	Proteintech
GAPDH	AC002	Abclonal

2.19. Reverse Transcription‐Quantitative Real‐Time PCR (RT‐qPCR)

For both mRNA and miRNA detection, 1 µg of total RNA was reverse transcribed using the All‐In‐One 5X RT MasterMix (G592, abm) under the same conditions: 37°C for 15 min, 60°C for 10 min and 95°C for 3 min. For miRNA reverse transcription, specific stem‐loop RT primers (Table 2) were additionally included. The resulting cDNA was diluted 20‑fold, and quantitative real‐time PCR (qPCR) was performed using a 2× SYBR Green PCR Master Mix (A4004, Lifeint) with 0.2 µM primers (see Table 3) in a total reaction volume of 20 µL containing 8 µL of diluted cDNA. The thermal cycling program consisted of an initial denaturation at 95°C for 10 min, followed by 40 cycles of 95°C for 15 s, 60–62°C for 20 s and 72°C for 25 s, with a final melting curve analysis from 65 to 95°C. Target expression levels were normalized to the exogenous spike‐in cel‐miR‐39 and calculated using the 2^−ΔCt ×10⁶ method.

TABLE 2.

Stem‐loop reverse transcription primers.

Target	Sequence (5'→3')
let‐7a	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACAACTAT
let‐7f‐5p	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACAACTAT
let‐7i‐5p	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACAGCACA
miR‐122‐5p	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACACAAAC
miR‐126‐3p	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCGCATT
cel‐miR‐39	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCAAGCT

TABLE 3.

Quantitative real‐time PCR primers.

Target	Strand	Sequence (5'→3')
let‐7a	Forward	GCCGAGTGAGGTAGTAGGTTGT
let‐7f‐5p	Forward	GCCGAGTGAGGTAGTAGATTGT
let‐7i‐5p	Forward	TGAGGTAGTAGTTTGTGCT
miR‐126‐3p	Forward	GCCGAGTCGTACCGTGAGTAAT
miR‐122‐5p	Forward	GCCGAGTGGAGTGTGACAATGGT
cel‐miR‐39	Forward	AGCCCGTCACCTGGTGTAAATC
miRNA Universal	Reverse	CAGTGCAGGGTCCGAGGTAT
FTH1	Forward	ACTGCCATGGAATTTGCGCT
	Reverse	TTAGCTCTCACTGCTCCCCA
FTL	Forward	TGAGCTCCCAGGTTCGTCAG
	Reverse	AGCCCAGAGAGAGGTAGGTG
ALB	Forward	CCCCAAGTGTCAACTCCAACT
	Reverse	CCGTCTGTTCACCAAGGACT
FGG	Forward	CGTCTGATTCTACCGGGCTC
	Reverse	CCCACACATAGTTGGGCAGT
SFTPC	Forward	CAAACGCCTTCTTGTCGTGG
	Reverse	GCCGCTGGTAGTCATAGACC
SFTPB	Forward	CGGAAGTTCCTGGAGAAAGA
	Reverse	GCTCTGGATGCTCAGGTTTG

2.20. Immunofluorescence (IF) Staining

Lung tissues were fixed in 10% neutral buffered formalin (L25010121, YULU) overnight, with the fixative replaced after 24 h. Following paraffin embedding, sections were deparaffinized using an environmental friendly dewaxing solution (G1128, Servicebio) and rehydrated through a graded ethanol series (100092183, Sinopharm). Heat‐induced antigen retrieval was performed in EDTA buffer (pH 8.0; G1206, Servicebio) for 30 min. Sections were then washed with PBS (G0002, Servicebio) and blocked with 3% BSA (GC305010, Servicebio) for 30 min. Subsequently, the sections were incubated overnight at 4°C with primary antibody against SFTPC (1:200; 10774‐1‐AP, Proteintech), followed by Cy3‐conjugated goat anti‐rabbit IgG (1:300; GB21303, Servicebio) for 50 min in the dark. Nuclei were counterstained with DAPI (G1012, Servicebio) for 10 min. To minimize background interference, sections were treated with an autofluorescence quencher (G1221, Servicebio) for 5 min before being mounted with anti‐fade mounting medium (G1401, Servicebio). Images were scanned with a Pannoramic MIDI digital slider scanner (3DHISTECH).

2.21. Integrated Multi‐Omics Analysis

Spearman correlation coefficients and corresponding p values between multi‐omics datasets were calculated using the Hmisc package in R, with subsequent multiple testing correction via the Benjamini–Hochberg (FDR‐BH) method to generate adjusted p values. Molecular pairs showing significant correlations (r > 0.6, FDR‐adjusted p < 0.05) were selected for co‐expression network construction and interaction prediction. The co‐expression network was visualized using Cytoscape (v3.9.1) (Shannon et al. 2003), while RBP‐RNA interactions were predicted using the catRAPID (v2.0) web server (Armaos et al. 2021). MiRNA‐target interactions were identified through 3'UTR binding analysis with miRanda (v3.3a) (Enright et al. 2003). Enrichment analyses (KEGG and GO) were conducted on conserved proteins and mRNAs, along with target genes of conserved miRNAs via the Metascape online platform (v3.5) (Zhou et al. 2019).

2.22. RNA Immunoprecipitation Sequencing (RIP‐seq) and Data Analysis

RIP‐seq was performed on snap‐frozen pig lung tissues pulverized under liquid nitrogen. Tissue lysates were divided into input (10%), anti‐HNRNPU immunoprecipitation (80%; 16365‐1‐AP, Proteintech) and rabbit IgG control (10%; Cell Signalling Technology) groups. RNA was extracted using TRIzol reagent (15596026, Invitrogen). Strand‐specific libraries were prepared with a UMI‐enabled KC‐Digital Stranded mRNA Library Kit for Illumina (DR08502, Seqhealth) to eliminate PCR duplicates. Size‐selected libraries (200–500 bp) were sequenced on the DNBSEQ‐T7 (MGI Tech Co., Ltd. China) platform with PE150 configuration.

Raw sequencing reads were quality‐controlled and adapter‐trimmed using Trimmomatic (v0.36) (Bolger et al. 2014). PCR duplicates were removed via an in‐house UMI‐based deduplication pipeline, and the resulting tags were aligned to the Sus scrofa genome (Ensembl release 109) using STAR (v2.5.3a) (Dobin et al. 2013) with default parameters. Peaks were called with exomePeak (v3.8) (Meng et al. 2013) and annotated with BEDTools (v2.25.0) (Quinlan and Hall 2010).

2.23. RIP‐qPCR

qPCR validation was performed using cDNA remaining from RIP‐seq library construction. Amplification of the exon‐containing peak_3778 (Chr3: 94175808–94176577, identified by exomePeak) region in EPAS1 transcript was conducted using specific primers (Forward: 5’‐AGTACCGGATGCTTGCGAAG‐3’; Reverse: 5’‐TTGCGGGGGTTGTAGATGAC‐3’) with Taq Pro Universal SYBR qPCR Master Mix Q712‐02(VAZYME, China) on a SLAN‐96S automated medical qPCR analysis system. Reactions contained 10 µL 2× Master Mix, 1 µL cDNA, 0.5 µL each primer (10 µM) and 8 µL nuclease‐free water in 20‐µL volumes. Thermal cycling included: 95°C 3 min; 40 cycles of 95°C 15 s, 60°C 30 s with plate reading; melt curve analysis 60–90°C. Percentage input (%Input) represents the fraction of RNA target recovered relative to the starting input, calculated as:

$$\%\text{Input}=100\times 2^{(A-B)}$$

where A are adjusted input C_t values, calculated as raw C_t —log₂ ⁽²⁰⁾ to correct for total RNA, and B represent the C_t values for the immunoprecipitation or control IgG groups, respectively.

2.24. Statistical Analysis and Visualization

All statistical analyses and graphical visualization in this study, including tau value and correlation calculations, were performed using R (v4.1.2) (Null et al. 2011). Data were presented as mean ± standard deviation (SD). Comparisons between two groups were conducted using unpaired two‐tailed Student's t‐tests, while correlations were assessed using Pearson's or Spearman's coefficients, depending on data distribution. Statistical significance was defined as p < 0.05.

3. Results

3.1. Characterization of Porcine sEVs From 133 Samples

To validate the biological fidelity of the trypsin‐based isolation method, we compared it against a gentle Collagenase II/Dispase II cocktail by assessing sEV size, morphology, membrane integrity and canonical marker expression. Both methods yielded sEVs with comparable size distributions, peaking at approximately 75 nm (Figure S1a–c). The lipid‐bound nature of these particles was confirmed by Triton X‐100 assays on lung sEVs, showing >95% rupture, with a slightly higher proportion observed in the trypsin group (Figure S1d). Furthermore, TEM imaging and Western blotting (WB) confirmed that sEVs from both methods maintained intact cup‐shaped morphologies and expressed canonical markers, including CD81, TSG101 and ALIX (Figure S2a–c). Crucially, Proteinase K protection assays on lung sEVs demonstrated that internal cargo (Syntenin‐1 and ALIX) was effectively shielded by the lipid bilayer (Figure S2d). Altogether, these results establish our optimized trypsin‐based pipeline as a high‐fidelity method that isolates intact sEVs comparable to those obtained via gentler enzymatic digestion.

Utilizing this trypsin‐based pipeline, we isolated sEVs from a total of 133 samples derived from 34 different organs, primarily from a single male donor (with female reproductive organs from a full sibling), using multiple spatial samplings per organ (detailed in Table S1). Since the sEVs were primarily derived from the same individual, this design ensured a uniform genetic and environmental background, enabling a robust comparison of their abundance and cargo profiles across different tissue sites (Figure 1a). Subsequent sEV characterization was performed using NanoFCM, TEM and cargo profiling via multi‐omics approaches (Figure 1b). The size of isolated sEVs exhibited a classic cup‐shaped distribution with a major peak of 75 nm (Figures 1c and S3–S5). Particle numbers ranged from 5.5 × 10^{1 1} to 7.93 × 10⁹ per organ, with the pancreas and adipose contain the highest and lowest particles. The yields of sEV RNA and protein content increased with the number of particles. The highest levels of RNA and protein were found in the spleen (5710.67 ng) and muscle (2496.98 µg), while the lowest were in adipose (122.99 ng) and male reproductive system (104.92 µg) (Figures 1d, S6a and Table S2). Overall, the small intestine, kidney, gland, lung, liver and brain ranked among the top 10 organs for average yields of sEV‐derived particle count, protein content and RNA content (Figure 1e).

FIGURE 1.

Characteristics of small extracellular vesicles (sEVs) derived from whole‐body tissues/body fluids/gastrointestinal contents of pig. (a) Schematic overview of sample collection. A total of 133 samples were collected from 34 organs, with multiple samples per organ taken from distinct anatomical sites. All organs were obtained from a boar, except for the FRS, which was collected from a sow in the same litter. Different colors indicate different organs, and the numbers represent the sample count per organ. (b) Workflow for sEV isolation, characterization, and cargo profiling. sEVs were isolated via differential centrifugation combined with sucrose cushion ultracentrifugation. Characterization was performed using NanoFCM and TEM, while cargo profiling was conducted using mass spectrometry and RNA sequencing. Symbols ①, ②, and ③ denote tissue, body fluid, and gastrointestinal contents, respectively. Numbers in parentheses indicate the number of samples analysed at each step. (c) 2D density heatmap of sEV size and concentration across 34 organs (n = 133 samples) measured by NanoFCM. Color intensity: Color represents relative data density, with red highlighting the most frequent size‐concentration core across all samples. Blue line: Depicts the average concentration trend fitted via local regression (Generalized Additive Model, span = 0.3). The vertical dashed line indicates the peak size at 75 nm. (d) Yields of sEV particles, proteins, and RNAs per unit sample (1 mL or 1 g), normalized from raw data measured by NanoFCM, BCA assay, and Quantus ^TM Fluorometer, respectively. (e) Venn diagram showing overlap of top 10 organs for sEV production based on average yield of particles, proteins, and RNAs from samples within each organ. (f) Relative proportions of the three sEV size bins across all samples. The sEV size bins (EV1: <60 nm; EV2: 60–80 nm; EV3: 80–150 nm) were classified according to the size criteria of Zhang et al. (Zhang et al. 2018), and the proportions of these sEV size bins were calculated from concentration data measured by nano‐flow cytometry from a 10‑µL aliquot of the sEV suspension. (g) Representative TEM images showing the morphology of the three sEV size bins derived from the spinal cord. (h) Pearson correlation heatmap of sEV sEV size bin ratios versus sEV yield. Red, positive correlation; blue, negative correlation. Mucous membrane, MM; Cerebrospinal fluid, CSF; Male reproductive system, MRS; Female reproductive system, FRS; Stomach contents, Contents1; Small intestine contents, Contents2; Large intestine contents, Contents3.

Interestingly, sEV exhibited pronounced heterogeneity in size distribution and were classified into three operational size bins: EV1 (<60 nm), EV2 (60–80 nm) and EV3 (80–150 nm) (Figure 1f, g). Although larger particles (EV2 and EV3) predominated across most tissue sites, we observed notable organ‐specific preferences, including a pronounced enrichment of EV1 in the brain and spinal cord (Figure S6b). Notably, the brain exhibited a 3.3‐fold higher EV1 proportion than other tissue sites (Figure S6c,d), highlighting the distinctive size‐distribution pattern in neural tissues. Further investigation of the relationship between these size bins and their cargo compositions suggested a weak but statistically significant positive correlation between RNA content and EV3 (r = 0.20, p = 0.02) (Figure 1h). In contrast, protein content exhibited no significant correlation with EV1 (p = 0.27), EV2 (p = 0.92) or EV3 (p = 0.72). Collectively, our pan‐tissue sEV characterization data revealed inter‐organ heterogeneity in both secretion capacity and size‐distribution profiles.

3.2. Classification and Cross‐Organ Conservation of sEV Proteins

To provide a systematic resource for investigating sEV‐mediated intercellular communication and biomarker discovery, we mapped the sEV proteome across a comprehensive atlas of 131 porcine samples (Table S1), with bile and CSF excluded due to insufficient protein yield for proteomic profiling. In total, we identified 8716 sEV proteins, with 7995 (≥ 1 unique peptide each) retained for downstream analysis. An average of 3468.5 proteins were detected per organ, with the highest numbers in brain (mean: 5296.8) and spinal cord (mean: 5498.0) (Figure 2a). Enrichment analysis of the 7995 identified sEV proteins revealed a significant overrepresentation of EV‐associated GO Cellular Component (CC) terms (e.g., ‘extracellular exosome’, ‘extracellular vesicle’) in the human database, supporting the high purity of the isolated sEVs (Figure 2b and Table S3). However, these top‐ranked terms showed reduced significance and fewer annotated proteins in the porcine analysis, likely reflecting the less comprehensive porcine GO annotation. Furthermore, we mapped proteins to nine representative human CC categories (Figure 2c, left). The predominance of extracellular vesicle proteins over nuclear and ribosomal ones supports high sEV purity, and overlaps with intracellular compartments (e.g., endoplasmic reticulum) likely reflect shared functional components. Radar chart analysis confirmed that extracellular vesicle term remained predominant across all tissue‐derived sEVs (Figure 2c, right), supporting method reliability despite tissue variation. Among these, 6752 proteins (84%) were known sEV proteins, while 1263 were novel discoveries, expanding the sEV proteome (Figure 2d). More importantly, 3478 RNA‐binding proteins (RBPs), 98 transcription factors (TFs), 3132 membrane proteins, 485 secreted proteins, 3282 receptors and 3513 ligands were identified using UniProt (UniProt Consortium 2025) and the Human Protein Atlas (Uhlén et al. 2015), further demonstrating sEVs' multifunctional roles in intercellular communication, signal transduction and gene regulation (Figure 2e). To obtain correlation profiles across 133 samples with identical background, we performed correlation analysis. The correlation heatmap revealed that intra‐organ correlations were stronger than inter‐organ correlations, supporting the existence of organ‐specific sEV proteins (Figure S7a). Notably, the abundance ratios of different protein types in sEVs exhibited cross‐organ heterogeneity, particularly for RBPs (Figure S7b), suggesting organ‐specific sEV protein loading patterns.

We identified 252 core sEV proteins that were conserved across organs (>95% of samples, Table S4). These proteins includes mitochondrial‐associated proteins (e.g., ATP5F1A, COX6C and VDAC family members), 183 RNA‐binding proteins (e.g., ANXA2, ALIX and MVP) and 20 MISEV2023‐recommended EV markers [e.g., CD63, SDCBP (Syntenin‐1) and PDCD6IP (ALIX); Figure S8]. Importantly, 10 conserved proteins were consistently detected across all 131 samples and classified as Category 2 (cytosolic proteins in EVs) according to the MISEV2023 guidelines (Figure 2f). Based on their functional roles in sEV biogenesis, we recommend ACTG1, CFL1 and ANXA2 as robust candidate markers (Fabbiano et al. 2020; van Niel et al. 2018; Cazzaro et al. 2022; Latham et al. 2013). This recommendation is supported by their confirmed expression in the parent tissues and consistent detection in sEVs, regardless of whether Trypsin or a Dispase II/Collagenase II cocktail is used for enzymatic dissociation (Figure S9a, b).

Additionally, we evaluated the abundance of classic sEV markers (CD9, CD63 and CD81), recently proposed novel markers (Syntenin‐1, ACTB, MSN and RAP1B) and candidate markers identified in this study (ACTG1, CFL1 and ANXA2) (Figure S9c). Among the classical sEV markers, CD63 showed the highest detection rates (tissues: 97%; body fluids: 100%; gastrointestinal contents: 100%), while CD81 exhibited the lowest detection rates (83%, 50% and 63%, respectively). Of the novel markers, MSN (99% detection) showed higher expression than Syntenin‐1 (94%) in tissue‐derived sEVs. In contrast, the candidate markers ACTG1, CFL1 and ANXA2 have equivalent detection rates with classic sEV markers, further validating their utility as robust sEV markers. Notably, possible contaminant lipoproteins (APOA1, APOA2 and APOB) were detected in sEVs, but their abundances were markedly lower than the marker protein CD63, suggesting effective sEV enrichment with lipoprotein depletion (Figure S9d).

3.3. Organ‐Specific sEV Proteins Detectable in Body Fluids

To identify candidate markers for isolating circulating sEV subpopulations and tracing their cellular origins, we used the Yanai tissue specificity index tau (τ) to systematically quantify sEV protein distribution across our 129 samples (comprising 120 tissue sites and nine gastrointestinal contents). Based on a stringent threshold of tau >0.85, we identified 2638 highly organ‐specific sEV proteins, of which 1023 (38.8%) were membrane‐associated (Figure 3a; Table S5). The distribution of these specific proteins varied significantly across anatomical sites, with the number of specific proteins per organ ranging from 3 to 500 (Figure 3b). Among the anatomically distinct, individual organs analysed, the spinal cord (500 proteins), tonsil (295 proteins) and brain (199 proteins) were identified as the top three sources for sEV detection, with representative markers including NSG1, CD8B, NCAM1 and SYT17, respectively (Figure 3c). These organ‐specific expression patterns were further corroborated by the PigGTEx database (Teng et al. 2024) (e.g., brain: NCAM1; heart: MYOZ2; lung: SFTPC; liver: NNMT) (Figures S10 and S11a,b). Consistent with these transcriptomic profiles, WB analysis confirmed the organ‐specific expression of SFTPC and NNMT in pig lung and liver tissues (Figure S11c). Crucially, these markers remained consistently detectable in sEVs regardless of the enzymatic dissociation method used (trypsin vs. Dispase II/Collagenase II cocktail), confirming that our organ‐specific sEV proteins faithfully reflect their tissue origins (Figure S11d). To further resolve their cellular origins, we analysed public pig single‐cell transcriptome expression matrices from lung (Huang and Xiao 2026), liver (Rao et al. 2023) and ileum (Xiao et al. 2026), which mapped SFTPC predominantly to alveolar type 1 (AT1) and type 2 (AT2) cells, and NNMT to hepatocytes, while neither gene was detected in ileal cells (Figure S12a–c). This cellular specificity was further supported by immunofluorescence (IF) staining of SFTPC in porcine lung sections (Figure S12d). Beyond this molecular specificity, the stable proportion of organ‐specific membrane proteins observed across different organ‐derived sEVs (Figure 3b) provides a reliable basis for organ‐specific sEV enrichment.

FIGURE 3.

Identification of organ‐specific sEV proteins and their detection in body fluid‐derived sEVs. (a) Overlap in number of identified organ‐specific and membrane proteins. (b) Number and categorical percentages of organ‐specific proteins. X‐axis labels are consistent between upper and lower plots. Bars in the lower plot are colored by organ. (c) Organ‐specific protein expression heatmap. Color scale represents Z‐score values from row‐scaled log₂ (LFQ + 1). Right‐side labels indicate protein names, with color coding denoting organ specificity. Row and column colors correspond to organs as in panel b. (d) Number of organ‐specific membrane proteins detected in plasma‐ and urine‐derived sEVs. (e) The number and relative proportion of protein abundance for organ‐specific proteins in plasma‐ or urine‐derived sEVs.

Subsequently, the expression of organ‐specific sEV proteins was evaluated in body fluids. We detected 195 and 108 organ‐specific sEV proteins in plasma and urine sEVs, respectively, including 23 overlapping proteins (7 membrane proteins) (Figure 3d). Notably, urine sEVs exhibited the highest number (35 proteins) and relative abundance (38.22%) of kidney‐specific sEV proteins, whereas plasma sEVs showed the greatest number (32 proteins) and relative abundance (16.96%) of spinal cord‐specific sEV proteins (Figure 3e, Table S6). Together, these findings demonstrate that organ‐specific sEV proteins are differentially represented in urine‐ and plasma‐derived sEVs, revealing distinct preferences of body fluid‐isolated sEVs for monitoring organ physiology.

3.4. Conserved and Organ‐Specific miRNA Signatures in sEVs

Small RNA sequencing of sEVs generated 17.94 billion raw reads, with an average of 9.1 million high‐quality clean reads per sample retained after quality control (Table S7). Subsequent composition analysis showed that miRNAs, tRNAs and mRNA fragments constituted an average of 32.32% (±18.78%), 33.68% (±17.41%) and 15.87% (±13.59%) of the total sEV small RNA repertoire, respectively (Figures 4a and S13). A total of 337 mature sEV miRNAs were identified across whole organs, with the top three being the brain (311 miRNAs), lung (310) and small intestine (302) (Figures 4b and S14a). Further analysis revealed that these three organs displayed miRNA expression patterns distinct from those of other organs (Figure S14b–d). This finding aligns with the observation that these three organs ranked among the top 20 in sEV particle, RNA and protein abundance, highlighting their important roles in sEV‐mediated communication (Figure S6a).

FIGURE 4.

Organ‐specific mature miRNA expression profile in sEVs. (a) Composition of small RNA libraries, with bars colored by organ. (b) Number of miRNAs identified in each organ. Bar colors follow the organ‐color scheme in panel a. (c) Distribution of counts and tau scores of five miRNA types categorized by abundance and tau score. Bar and line colors represent different types of miRNAs. (d) Number of samples containing detectable miRNAs and their average abundance. Point colors maintain the same miRNA class‐color mapping as in panel c. (e) Expression heatmap of organ‐specific miRNAs. Right‐side labels indicate miRNA symbols, with color coding denoting organ specificity. Row and column colors correspond to organs as in panel a. (f) Proportional abundance of the top 10 most abundant miRNAs in sEVs derived from four body fluids. (g) The number of organ‐specific miRNAs detected in body fluid‐derived sEVs. (h) Enrichment of GO Biological Process terms for organ‐specific miRNA target genes in body fluid sEVs. Error bars represent standard deviation (SD).

From the total pool of 337 sEV‐miRNAs, 18 were identified as conserved and 152 as organ‐specific (Figure 4c, Table S8). The conserved miRNAs, including the let‐7 family (let‐7a, let‐7e, let‐7f‐5p and let‐7i‐5p), exhibited high abundance (Figures 4d and S15a), and their target genes were enriched in cell proliferation‐related pathways such as MAPK signalling and endoplasmic reticulum protein processing (Figure S15b). On the other hand, organ‐specific miRNAs were mainly derived from the lung (29.6%), glands (16.5%) and spinal cord (14.5%) (Figures 4e and S15c). Representative examples included muscle‐specific miR‐1, lung‐specific miR‐126‐3p, liver‐specific miR‐122‐5p, brain‐specific miR‐149 and pancreas‐specific miR‐216 (Figure 4e). These miRNAs reflect tissue physiology; for instance, liver‐specific miR‐122‐5p targets were enriched in liver homeostasis pathways, including steroid biosynthesis, Wnt signalling and insulin signalling (Figure S15d,e). Specifically, RT‐qPCR was used to quantify the expression of both conserved and organ‐specific miRNAs in liver and lung sEVs (Figure S16a–e). Excluding let‐7f‐5p, these expression profiles remained consistent between the two enzymatic dissociation methods (trypsin vs. Dispase II/Collagenase II cocktail), indicating that the choice of dissociation protocol has minimal impact on sEV‐miRNA recovery. While high biological variance restricted statistical significance in certain groups, the overall expression trends were corroborated by public porcine miRNA‐seq datasets from the PIGOME database (Han et al. 2025) (Figure S16f). Therefore, organ‐specific sEV‐miRNAs may serve as baseline markers of tissue origin, while conserved sEV‐miRNAs likely play fundamental roles in regulating cell proliferation.

Body fluids systemically transport biological signals by carrying sEVs throughout the organism. Analysis of sEVs from plasma, urine, bile and CSF revealed 175, 216, 112 and 90 miRNAs, respectively (Figure S15f). Consequently, each body fluid‐derived sEV population displayed a unique miRNA expression signature, indicative of its systemic exposure (Figure S15g). Five miRNAs (let‐7a, let‐7e, let‐7f‐5p, let‐7i‐5p and miR‐26a) were the most abundant and conserved in all body fluid‐derived sEVs, implicating their potential in maintaining cellular homeostasis (Figure 4f). Besides, the plasma‐, urine‐, bile‐ and CSF‐derived sEVs contained 83, 107, 56 and 44 organ‐specific sEV‐miRNAs respectively (Figure 4g), including 29 shared miRNAs predominantly originating from lung (11), spinal cord (8) and salivary gland (4). Target genes of these 29 miRNAs showed significant enrichment in cellular response to oxygen‐containing compounds, protein transport, regulation of cellular localization and vesicle‐mediated transport, suggesting that organ‐specific sEV‐miRNAs in body fluids may participate in regulating oxidative stress response and vesicle secretion in distal cells (Figure 4h).

3.5. Long‐Read Transcriptome Profiling Reveals Organ‐Specific Complexity and Functional Insights Into sEVs

Transcript profiling, compared to gene expression profiling, provides isoform‐specific resolution that reveals alternative splicing and post‐transcriptional regulation, offering a more precise link to functional protein outcomes. Here, we conducted Oxford Nanopore TGS to uncover the sEV transcriptome, generating 1.63 billion raw reads from 119 samples. After quality control, 1.59 billion reads were retained, including 1.25 billion full‐length reads (FLRs) totalling 1.02 Tb in length (Figure 5a, Table S9). Among these FLRs, 0.40 billion (31.9%) reads mapped unambiguously to single transcripts (‘unique’), of which 0.37 billion (91.5%) exceeded 1 kb in length and 0.30 billion (75.0%) fully spanned all splice sites of transcripts (‘full_splice_match’) (Figure 5a). The remaining FLRs either mapped ambiguously (‘unique_minor_difference,’ ‘inconsistent,’ ‘ambiguous’) or aligned to non‐transcript regions (‘intergenic,’ ‘noninformative,’ ‘unmapped’) (Figures 5a and S17, Table S10). Transcripts classified as ‘novel in catalog’ or ‘novel not in catalog’ but with more than one occurrence were merged into the reference annotation (Ensembl release 109) to generate an enhanced annotation database. The integration analysis identified 96,474 novel transcripts (comprising 35,961 ‘nic: novel in catalog’ and 60,515 ‘nnic: novel not in catalog’), corresponding to 14,789 protein‐coding genes, each of which acquired between 1 and 25 new transcripts (Figure S18a). Among these novel transcripts, 39.1% (15,010) featured alternative first exons, 70.6% (20,618) possessed alternative 5' ends and 36% (9623) represented exon‐skipping events (Figure S18b). While alternative first exon and 5' end usage may partly arise from RNA degradation, exon‐skipping events demonstrated significantly higher protein‐level validation rates in human datasets (Glinos et al. 2022), thus more robustly indicating functional isoform production.

FIGURE 5.

Comprehensive transcriptomic profiles of sEVs. (a) Classification and variation of TGS reads across samples after genome alignment. Left panel: Hierarchical assignment of full‐length reads into IsoQuant categories. Right panel: Box plots showing per‐sample read counts for fragments >1 kb, unique assignments, and total full‐length reads. (b) Comparison of transcriptomic profiles between NGS and TGS. Left panel: Number of genes and transcripts (TPM > 1) detected across different datasets. Top right panel: Pearson correlation of gene expression levels between NGS and TGS across all samples. Bottom right panel: Spearman correlation between NGS and TGS for all genes within each individual sample. (c) Gene and transcript expression levels (TPM) and organ distribution breadth. TPM values are averaged over all genes expressed in a given number of organs. (d) Tissue conservation of sEV‐conserved genes. Left panel: Venn diagram showing the overlap of conserved genes (TPM > 1 in ≥ 95% of samples) among tissue, sEV TGS, and sEV NGS datasets. Right panel: Box plot of expression levels for the 1010 conserved genes detected in both sEVs and tissues. e, Tissue conservation of sEV‐deficient genes. Left panel: Overlap between conserved genes in tissues and sEV‐deficient genes (TPM > 1 in ≤5% of samples) from TGS and NGS datasets. Right panel: Expression levels of the 37 sEV‐deficient genes in sEVs and tissues. (f) Tau value distributions and classification of genes/transcripts. (g) Comparison of expression patterns between NGS‐derived genes and TGS‐derived transcripts. The value in the upper‐left corner represents the number of transcripts under the y‐axis category whose corresponding genes fall under the x‐axis category. In contrast, the value in the bottom‐right corner indicates the number of genes under the x‐axis category with transcripts classified under the y‐axis category. The corresponding genes of the TGS_Trans were used for the cross‐platform comparison.

Thereafter, we retained a total of 105,409 transcripts with TPM > 1, corresponding to 25,399 genes (including 17,427 protein‐coding genes and 7395 lncRNA genes) for subsequent analysis. Importantly, 93,409 transcripts (88.6%) derived from 17,575 genes (69.2%) were supported by full‐splice‐match reads. In detail, 933 to 15,083 protein‐coding genes and 550 to 28,126 transcripts were detected across 119 samples. Overlap analysis with next‐generation sequencing (NGS) data revealed high consistency (88.1% shared genes) and strong correlation (R ² = 0.65, p < 2.2e−16) between the two sequencing technologies (Figure 5b, Table S11). Principal component analysis (PCA) and hierarchical clustering did not resolve distinct tissue groups based on overall gene expression correlation. However, intra‐organ correlations were significantly stronger than inter‐organ correlations (Figure S19a–c). Although most organs showed comparable gene and transcript detection rates across tissues and sequencing platforms, blood vessels, small intestines and livers exhibited greater heterogeneity (Figure S20a). Notably, we found that a small subset of genes accounted for the majority of transcriptional output (Figure S20b), implying a selective mechanism for loading long RNA into sEVs. Overall, these results demonstrate the superior precision of transcript‐level sequencing for characterizing long RNA isoforms and structures in sEVs.

In this study, protein‐coding genes detected by NGS (NGS_Genes) showed a broader organ distribution (>30 organs), while protein‐coding transcripts detected by TGS (TGS_Trans) exhibited more specific expression profiles (<5 organs) (Figure 5c). Interestingly, organ‐specific coding transcripts exhibited low abundance and were predominantly novel transcripts (Figure S21a). In contrast to protein‐coding genes, sEV‐lncRNAs from both NGS and TGS displayed an intermediate organ distribution (∼20 organs) (Figure S21b,c). Moreover, the lncRNA TGS_Trans with narrow tissue distribution (< 5 organs) consisted primarily of novel lncRNA transcripts (Figure S21d). These results indicate that coding transcripts and lncRNA detected by TGS provide more precise and enhanced organ‐specific information than NGS data, offering greater potential for elucidating inter‐organ heterogeneity in sEV transcriptomes.

Further analysis identified 1010 conserved EV tissue genes through an intersection of conserved sEV (4694 NGS_genes, 4660 TGS_genes, 1059 genes correspond to 1227 TGS_Trans) and tissue (9796 genes) transcriptomes. These genes were consistently expressed at higher levels in sEVs than in their tissues of origin and were predominantly enriched in fundamental biological processes such as peptide metabolism, translation, respiratory electron transport and ribonucleoprotein complex biogenesis (Figures 5d and S22a). Among them, the iron transport genes FTH1 and FTL showed the most pronounced upregulation in sEVs (11.3‐ and 22.6‐fold higher than in tissues, respectively), whereas ribosomal protein genes accounted for 40% (8/20) of the most highly expressed sEV genes (Figures 5d and S22b). Notably, the corresponding proteins for these highly expressed genes also exhibited strong cross‐tissue conservation (Figure S22b). All 121 ribosomal protein genes were consistently detected in sEV transcriptomes, though their protein‑level conservation was comparatively lower (Figure S22c). These findings collectively link sEV physiology to core functions including iron transfer and translational machinery. In contrast, intersecting conserved tissue‐genes with sEV‐genes detected in fewer than 5% of organs (1049 NGS_genes, 4666 TGS_genes, 19,722 genes corresponding to 86,366 TGS_Trans) yielded 37 EV‐deficient genes, of which 34 corresponding proteins were undetectable in sEVs from any tissue (Figure 5e). These patterns indicate the existence of a selective mRNA and protein sorting mechanism during sEV biogenesis process.

We subsequently classified sEV genes/transcripts based on expression levels and tau values, identifying 3174 organ‐specific sEV genes and 6891 organ‐specific transcripts, alongside three other expression categories (Figure 5f). Transcripts demonstrated a significantly higher degree of organ specificity than genes, as reflected by their tau value distributions (Figure 5f). This superior resolution was corroborated by cross‐platform comparisons: most NGS‐based organ‐specific sEV genes were either undetectable (2609) or weakly expressed (1136) in TGS data, whereas a majority of TGS‐based organ‐specific transcripts (5751) originated from genes that were widely expressed in NGS data (Figure 5g). Among the 661 genes identified as organ‐specific in both TGS_Trans and NGS_Genes analyses, 422 (63.8%) were consistently assigned to the same organs of origin, with the liver (51 genes), bone marrow (39 genes) and pancreas (34 genes) showing the highest concordance (Figure 5g and S23). The biological relevance of these organ‐specific signatures was supported by functional enrichment analysis, where the pathways associated with organ‐specific sEV genes aligned closely with the physiological functions of their tissues of origin. For instance, liver‐specific genes were linked to intracellular iron homeostasis and coagulation, cardiac‐specific genes to sarcomere organization and striated muscle contraction, and pancreatic‐specific genes to lipid catabolism and glucose regulation (Figure S24). These transcriptomic patterns were further validated using the PIGOME database (Han et al. 2025) and by performing RT‐qPCR on tissue‐derived sEVs. RT‐qPCR confirmed the high expression of conserved genes (FTH1, FTL) and the strict organ‐specificity of ALB/FGG (liver) and SFTPB/SFTPC (lung) all within sEVs, consistent with public RNA‐seq data (Figure S25a–g). This specificity remained robust across different enzymatic protocols, affirming its intrinsic nature. Furthermore, IGV visualization confirmed the presence of intact, full‐length transcripts for these markers in sEVs from their respective organs, while they were undetected in sEVs from other organs (Figure S25h–k). In summary, our results establish that sEV transcripts serve as more precise molecular indicators of tissue origin than sEV genes, revealing a layer of organ‐specific information that is functionally coherent and highly reproducible across complementary sequencing technologies.

3.6. Global Mapping of sEV Cargo Interaction Networks

To investigate the interactions among sEV cargo molecules, we constructed a network based on Spearman correlations. This analysis revealed multiple strong positive correlations (r > 0.6) between pairs of sEV cargos — including RBPs, miRNAs and mRNAs. Notably, many of these correlated pairs are documented interaction partners in RNAInter database (Kang et al. 2022) (Figure 6a). Several RBPs (e.g., PTBP1, TARDBP) and miRNAs (e.g., miR‐30c, miR‐424, miR‐142) exhibited particularly high interaction potential, as indicated by their elevated cumulative correlation coefficients and interaction frequencies (Figure 6b). Notably, 13 RNAInter‐annotated RBPs and miR‐424‐3p showed strong correlations (>0.6) with their respective partner RNAs (Figure 6c). For instance, the presence of a miR‐424‐3p binding site in the UBE2E2 3'UTR, coupled with their positive correlation, suggests that this miRNA may facilitate the sorting of UBE2E2 mRNA into sEVs (Figure 6c, Table S12). Furthermore, we found that three mRNAs (EPAS1, PTPN18 and CNN2) contain motifs matching RBPs whose abundance positively correlated with their mRNA expression (Figure 6c, Table S13). Among these, the transcription factor EPAS1 mRNA was notable for containing three predicted binding motifs for HNRNPU and displayed a positive expression correlation of r = 0.7 with HNRNPU (Figure 6d, e). RIP‐seq analysis confirmed HNRNPU binding to EPAS1 mRNA, and subsequent RIP‐qPCR validation demonstrated a mean enrichment of 12.3‐fold over the IgG control (p < 0.05; n = 3) (Figure S26a,b). Intriguingly, while EPAS1 mRNA (particularly full‐length transcripts in lung‐derived sEVs) was readily detectable (Figure 6f), its encoded protein was not detected in any sEV samples analysed. This disconnect suggests a potential RBP‐mediated, mRNA‐specific intercellular communication pathway independent of protein translation.

FIGURE 6.

Integrated multi‐omics analysis of sEVs. (a) Spearman correlation between pairwise combinations of protein, miRNA, and mRNA datasets. ‘Total’ includes all combinations; ‘RNAInter’ indicates documented interactions from the RNAInter database. (b) Interaction capabilities of RBPs (binding miRNAs and mRNAs) and miRNAs (targeting mRNAs) from RNAInter database. The x‐axis represents cumulative Spearman correlation coefficients between miRNAs/RBPs and their interacting RNAs; dot sizes are proportional to the number of interacting RNAs. (c) Interaction network built from RNAInter‐annotated RBPs and miR‐424‐3p, incorporating RNAs with expression correlation coefficients >0.6. Components predicted to interact by miRanda or catRAPID are highlighted in red. (d) Scatter plot showing the expression levels of HNRNPU and EPAS1. x‐axis and y‐axis represent log₁₀‐transformed abundance of HNRNPU and EPAS1, respectively. Spearman correlation values are indicated in red. (e) Binding motif of HNRNPU to EPAS1 mRNA predicted by catRAPID. (f) Full‐length read counts of EPAS1 in sEVs. (g) Number of overlapping genes at mRNA and protein levels. (h) Gene‐level correlation between mRNAs and their encoded proteins. (i) Organ detection breadth of mRNAs versus their encoded proteins. (j) Organ detection breadth of genes detected exclusively as mRNA or protein. (k) Heatmap of enriched terms for conserved proteins, conserved mRNAs, and conserved miRNA targets analysed by Metascape, colored by −log₁₀ transformed p values. Right‐side colors: orange‐red indicates GO Biological Process terms; light blue indicates KEGG pathways.

Although the relationship between mRNA and its encoded protein within cells is well‐established (Buccitelli and Selbach 2020), their concordance within sEVs remains poorly characterized. In this study, we identified 6842 coding genes with both mRNA and protein detected in sEVs (Figure 6g). The proteins encoded by these genes showed higher cross‐organ expression conservation than their corresponding mRNAs, with mRNA‐protein correlation coefficients ranging from −0.5 to 0.75, indicating distinct sorting mechanisms for these two molecular types (Figure 6h, i). Furthermore, the 897 protein‐exclusive sEV cargo exhibited stronger organ‐specificity than the 15,052 mRNA‐exclusive cargo (Figure 6j), highlighting their potential utility as markers for tracing the organ origin of sEVs. Despite this differential loading, functional enrichment analysis revealed a biological convergence: both conserved proteins/mRNAs and target genes of conserved miRNAs were consistently enriched in fundamental cellular processes, including vesicle‐mediated transport regulation, carboxylic acid metabolism, protein catabolism, organelle protein localization, cytokine response and cellular component biogenesis regulation (Figure 6k). These findings suggest that although sEVs may package mRNAs and proteins through partially independent mechanisms, the resulting cargo networks are functionally coordinated to drive core biological pathways. Collectively, this study constructed a protein‐miRNA‐mRNA cargo network, providing insights into the coordinated interactions among different sEV molecules.

4. Discussion

A comprehensive multi‐omic atlas of tissue‐derived sEVs is essential for systematically understanding the diversity of sEV cargo across organs, elucidating their potential biological functions and cooperative transport mechanisms, and identifying clinical biomarkers. Nevertheless, current sEV studies predominantly rely on cultured cells or biological fluids, which fail to capture local molecular signatures of the tissue microenvironment (Lee et al. 2024). To date, only one proteomic study has systematically characterized sEVs across tumour tissues‐covering merely 18 tumour types from 85 donors (Hoshino et al. 2020). In this study, we employed the pig as a model organism owing to its growing importance in biomedical research (Lunney et al. 2021), establishing the most comprehensive (120 tissue sites) multi‐omic atlas of tissue‐derived small extracellular vesicles (sEVs).

sEV isolation is a critical yet challenging step in sEV research, with no optimal method currently established to date (Welsh et al. 2024). Ultracentrifugation, the most commonly used EV separation method for cell supernatants and body fluids (Royo et al. 2020), may struggle to achieve high‐purity sEVs from tissue (Lee et al. 2024). In comparison, the isolation protocol combining ultracentrifugation with a 30% sucrose cushion enhances both sEV purity and yield while maintaining operational simplicity (Matejovič et al. 2021). Hence, we combined ultracentrifugation with a 30% sucrose cushion to maximize sEV yield for multi‐omic analyses. The sEVs we isolated were consistent in size and morphology with classical sEVs, and expressed canonical markers (e.g., CD63, Syntenin‐1, MSN) (Welsh et al. 2024). Notably, contaminant lipoproteins were co‐isolated with tissue‐derived sEVs in our study, a challenge also reported in recent tissue EV studies (Zhang et al. 2023; Xu et al. 2024; Dong et al. 2020). Emerging evidence suggests that lipoproteins can adsorb onto EV membranes in the extracellular environment and become part of the protein corona (Tóth et al. 2021), which may explain their persistent presence in tissue sEV preparations. Therefore, future validation of key sEV components should employ alternative methods (e.g., iodixanol density gradient centrifugation) to improve reliability.

sEVs exhibits high heterogeneity (Kalluri and LeBleu 2020), which can be reflected in the yield of sEVs. For instance, murine liver sEVs produce 4.5‐fold more particles than cardiac sEVs (Ge et al. 2024). Our study employed standardized protocols (fixed tissue weight/uniform sEV isolation) to confirm this heterogeneity and extended the observations to both protein and RNA levels. Results showed that liver sEVs yielded 1.51‐fold (particles), 2.78‐fold (protein) and 1.29‐fold (RNA) higher levels than cardiac sEVs. Conversely, adipose tissues demonstrated the lowest sEV yield in our study, an observation that contrasts with the high production levels reported for adipocytes (3T3‐L1) (Garcia‐Martin et al. 2022). This discrepancy may physiologically stem from the low adipocyte density per unit adipose tissue mass, while being technically attributable to incomplete tissue dissociation caused by adipose sectioning difficulties. Future studies should optimize tissue processing methods (e.g., frozen sectioning (Yang et al. 2022)) to precisely quantify sEV yields and cargo composition. Notably, the small intestine, kidney, gland, lung, liver and brain consistently ranked among the top 10 high‐yield organs across all three sEV production metrics: particle count, protein content and RNA content. This robust pattern underscores their potential as key hubs in sEV‐mediated intra‐ and inter‐tissue communication networks. Further investigation into their sEV secretion mechanisms may provide critical insights for improving engineered sEV production (Ng et al. 2022). sEVs also exhibit strong conservation. For instance, we identified a panel of organ‐conserved sEV cargo molecules—including mitochondrial‐associated proteins, let‐7 family miRNAs and ferritin mRNAs—that are ubiquitously expressed intracellularly and participate in regulating fundamental cellular processes (Ding et al. 2024; Itokazu et al. 2022; Daou et al. 2022).

CD9, CD63 and CD81 are canonical sEV markers (Jeppesen et al. 2019); however, their expression exhibits heterogeneity across both biofluid‐ and tissue‐derived sEVs, with the complete absence of CD9/CD81 observed in certain cell line‐derived sEVs (Garcia‐Martin et al. 2022; Kugeratski et al. 2021; Hoshino et al. 2020). Consistent with previous findings, we confirmed the heterogeneous expression of these three classical markers, detecting them in 94%, 97% and 83% of tissue‐derived sEVs, respectively. Similarly, the recently identified novel sEV makers (ACTB, MSN, RAP1B and Syntenin‐1) (Kugeratski et al. 2021; Hoshino et al. 2020) showed detection rates of 89%, 99%, 81% and 94%, respectively, in our tissue‐derived sEVs. This reveals a limited applicability of these novel markers except for MSN, which was undetectable in sEVs from only one adipose tissue sample. Instead, we identified three robust markers (ACTG1, CFL1 and ANXA2) with 100% coverage across 133 samples (tissues, body fluids and gastrointestinal contents), whose biomarker potential independently validated in cell line‐derived sEVs (Garcia‐Martin et al. 2022; Kugeratski et al. 2021; Jeppesen et al. 2019). These three proteins play important roles in sEV biogenesis and cargo sorting: ACTG1 mediates the docking and fusion of multivesicular endosomes (van Niel et al. 2018), CFL1 promotes sEV secretion through rapid actin depolymerization (Cazzaro et al. 2022), and ANXA2 facilitates RNA loading (Fabbiano et al. 2020). Nevertheless, their cross‐species consistency and method‐dependent performance warrant further evaluation.

Recent studies utilizing third‐generation sequencing have revealed the potential diversity and functionality of sEV mRNAs (Padilla et al. 2023; Ransom et al. 2024; Yu et al. 2023). Our research further expands the understanding of sEV mRNA diversity and its cross‐tissue ubiquity. Notably, our large‐scale sequencing effort captured 300 million reads (24%) that match all splice junctions of 93,409 transcripts (17,575 genes), establishing the most comprehensive full‐length transcript repertoire of sEVs to date. Notably, although two rounds of short fragment filtering were performed during third‐generation RNA library preparation, only 24% of total reads qualified as full‐splice‐match reads. This finding is consistent with two independent ONT‐based studies (Padilla et al. 2023; Yu et al. 2023), demonstrating that fragmented RNAs indeed constitute the predominant population in sEVs. In contrast, a PacBio‐based study reported 78.5% full‐length transcripts in sEVs (Ransom et al. 2024). This likely reflects size‐selection artifacts rather than true biological features. Furthermore, our study revealed a large number of reads that failed to map to known exonic regions, an observation that warrants further investigation.

In summary, our study provides a foundational multi‐omic atlas of porcine tissue‐derived sEVs and significantly advances our understanding of variation in sEVs across organs, representing a critical pioneering step toward elucidating sEV‐mediated inter‐tissue communication networks in complex organisms. Although our atlas offers a comprehensive view of sEV organ specificity, future studies with larger, multi‐individual cohorts are necessary to validate our findings and assess biological variability. In addition, studies with sufficient sEV materials could incorporate pre‐digestion normalization strategies (e.g., based on sEV protein mass or particle number) as an additional quality‐control layer to further enhance the quantitative robustness of proteomic profiling. Future integration with bulk or single‐cell transcriptomics will further elucidate the cellular origins of sEVs and provide mechanistic insights. This resource can also provide valuable insights into cell‐cell communication, cross‐organ communication and the coordinated transport of sEV cargo in porcine or other research models.

Author Contributions

Naixiang Yu: writing – original draft, formal analysis, investigation, visualization, data curation, validation. Jinxiu Liu: visualization, formal analysis, investigation, writing – original draft. Chenliang Lai: investigation, formal analysis. Siyi Liu: investigation. Jiacheng Wei: investigation. Jiahong Zhong: investigation. Sanya Xiong: investigation. Qingjie Zeng: investigation. Junwu Ma: investigation. Zhou Zhang: writing – review and editing, formal analysis, visualization, writing – original draft, investigation. Zhiyan Zhang: writing – review and editing, project administration, funding acquisition.

Funding

This study was supported by National Key Technology in Agricultural Project (NK2022110601).

Ethics Statement

All experiments involving pigs were approved by the Animal Ethics Committee of Jiangxi Agricultural University (JXAULL‐2021‐37).

Conflicts of Interest

The authors declare no conflicts of interests.

Supporting information

Supporting Information: jev270374‐sup‐0001‐TableS1.xlsx

Supporting Information: jev270374‐sup‐0002‐TableS2.xlsx

Supporting Information: jev270374‐sup‐0003‐TableS3.xlsx

Supporting Information: jev270374‐sup‐0004‐TableS4.xlsx

Supporting Information: jev270374‐sup‐0005‐TableS5.xlsx

Supporting Information: jev270374‐sup‐0006‐TableS6.xlsx

Supporting Information: jev270374‐sup‐0007‐TableS7.xlsx

Supporting Information: jev270374‐sup‐0008‐TableS8.xlsx

Supporting Information: jev270374‐sup‐0009‐TableS9.xlsx

Supporting Information: jev270374‐sup‐0010‐TableS10.xlsx

Supporting Information: jev270374‐sup‐0011‐TableS11.xlsx

Supporting Information: jev270374‐sup‐0012‐TableS12.xlsx

Supporting Information: jev270374‐sup‐0013‐TableS13.xlsx

Supporting Information: jev270374‐sup‐0014‐FigureS1‐S26.docx

Data Availability Statement

The data supporting the findings of this study are available within the manuscript, Supplementary Information or source data. All relevant experimental data have been submitted to the EV‐TRACK knowledgebase (https://www.evtrack.org/, accession number EV260021). The original datasets generated by sequencing presented in the study are deposited in the Genome Sequence Archive (GSA) database at https://ngdc.cncb.ac.cn/gsa/, accession number PRJCA053153. The raw mass spectrometry files and the search/identification files obtained with MaxQuant have been deposited to the ProteomeXchange Consortium through the iProX partner repository at https://www.iprox.cn/ under accession code IPX0014587000. This study did not generate any unique program, and custom code supporting the current study is available at https://github.com/NaixiangYu/PigEVatlas.