This study was conducted at the National Institute for Infectious Diseases Lazzaro Spallanzani in Rome, Italy. The study was approved by the institutional review board (approval number: 9/2020), and written informed consent was obtained from patients. 17 severe and 25 mild hospitalized COVID-19 patients were selected according to the WHO categorization [https://www.who.int/news-room/fact-sheets/detail/coronavirus-disease-(covid-19)] with a laboratory-confirmed diagnosis of SARS-CoV-2 infection. 23 healthy donors (HD) were enrolled as the control group. Quantitative proteomic analysis and functionality assays was performed on EVs purified from severe (N = 8) and mild COVID-19 (N = 12) patients compared to those of the HD (N = 5) Further details are in the Results section.

The following commercial chemiluminescence microparticle antibody assays (CMIA): the SARS-CoV-2 specific anti-N and the anti-S/RBD tests [AdviseDx SARS-CoV-2 IgG II and SARS-CoV-2 IgG II Quant, respectively, ARCHITECT^® (Chicago, IL, USA) i2000sr Abbott Diagnostics, Chicago, IL, USA] were used, according to the manufacturer’s instructions. Index > 1.4 and binding antibody units (BAU)/mL>7.1 are considered positive, respectively. Viral RNA was extracted using the QIAsymphony (QIAGEN, Hilden, Germany), and real-time reverse transcription polymerase chain reaction (RT-PCR) targeting the E and RdRp viral genes was employed to assess the presence of SARS-CoV-2 RNA (25). Confirmation of diagnosis was performed by in-house RT-PCR targeting the viral membrane protein (M) gene, followed by Sanger sequencing (327 bp). Follow-up of the infection course was then performed using only the E gene real-time RT-PCR. Other respiratory tract infections were investigated using a multiplex nucleic acid test (QIAstat-Dx Respiratory Panel, QIAGEN, Hilden, Germany).

Plasma samples were obtained by centrifuging heparin-anticoagulated peripheral blood for 10 minutes at 1800 rpm, then immediately stored at -80 °C until analysis. Peripheral blood mononuclear cells (PBMCS) from HD were isolated using gradient centrifugation (Lympholyte, Cedarlane, Canada), counted with Trypan blue, and prepared for further experiments.

Inflammatory cytokines (IL-8, Il-1β, IL-6, TNF-α) in plasma samples were quantified using an automated multiplex immunoassay on the Ella instrument (Protein Simple/Biotechne, San Jose, CA, USA). Ella is an advanced benchtop automated ELISA system created to provide precise, consistent data without requiring any manual procedures. The Ella platform’s Simple Plex™ assays, featuring a distinctive microfluidic design, are fully validated, extremely sensitive, and capable of measuring up to 8 analytes at once. The detection limit is 0.19 pg/mL for IL-8; IL-1β: 0.064 pg/mL; IL-6: 0.11 pg/mL; TNF-α: 0.03 pg/mL.

A rapid nickel-based approach was used to isolate plasma EVs. In particular, this method exploits nickel cations’ capacity to capture EVs and uses chelating agents to elute non-aggregated vesicles in a physiological pH solution, preserving their integrity and functional capacities (26). Briefly, 1 ml of plasma sample was centrifuged at 2800 g for 10’ to remove potential cellular contaminants and diluted 1:3 in PBS 1X. After 30’ incubation with nickel beads in shaking at RT, the EVs were eluted in a suitable elution volume by shaking for 10’ at 28 °C. The EV characterization in terms of size and concentration was performed using Nanoparticle Tracking Analysis (Nanosight). The results were analyzed using NanoSight NTA3.4 software.

Phenotype and frequency of EVs were evaluated by multiparametric flow cytometry using the Dye Integer EV Detection kit (Becton-Dickinson, San Jose, USA), which contains a lipophilic cationic dye that diffuses into EVs based on their membrane potential, and FITC-conjugated phalloidin, which binds to the cytoskeletal protein actin in vesicles with damaged membranes. Absolute cell counts were obtained using a TruCount absolute counting tube (Becton-Dickinson, San Jose, USA). The staining mix was prepared in PBS 1X (Corning) by first adding the lipophilic cationic dye and phalloidin, followed by the antibody cocktail consisting of anti-CD45 BV510, anti-CD41a PE, and anti-CD31 PE-Cy7. The antibodies and fluorochromes used are described in Supplementary Figure 1. Before use, the staining mix was centrifuged at 16,000x g for 15 minutes to remove particles. Samples’ plasma was processed according to the kit’s procedure. Briefly, 50 µl of filtered PBS 1X (Corning) was loaded into a trucount tube without perturbing the bead pellet. After, 50 µl of plasma was added using reverse pipetting and 95 µl of staining mix. After 45 minutes of incubation in the dark at room temperature, the tubes were supplemented with 1 mL of filtered 1X PBS, and the cells were acquired using a BD FACSLyric (Becton-Dickinson, San Jose, USA). The cells were then analyzed using BD FACSuite (Becton-Dickinson, San Jose, USA). Gating Strategy Analysis: A dot plot was used to display phalloidin, and a lipophilic cationic dye (CLP) was generated to create a gate in the phalloidin-negative area, thereby excluding vesicles with damaged membranes (Supplementary Figure 2). This gate was then applied to a CLP/FSC dot plot to discriminate EV events from the background. The enriched EV area was identified as phalloidin-negative, CLP-positive events. This EV area was further analyzed on a CD45/CD41 dot plot, where CD45+ events were considered leukocyte-derived EVs. A gate excluding CD45+ events was subsequently plotted on a CD31/CD41 dot plot to assess other EV phenotypic populations. CD31+/CD41+ events were identified as platelet-derived EVs, while CD31+/CD41- events were classified as endothelium-derived EVs.

For quantitative mass spectrometry analysis, 2,29 E + 09 particles (NTA-determined) per sample were used. The purified EVs were lysed using RIPA buffer and analyzed by a Q Exactive Plus Orbitrap LC-MS/MS system, as described in (27). Briefly, after overnight trypsin digestion (0.6 μg/sample) at 37 °C, the peptide mixture was acidified with 0.5% trifluoroacetic acid and fractionated using the High pH Reversed-Phase Peptide Fractionation Kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. Peptides in each fraction (8/sample) in 2.5% acetonitrile, 0.1% TFA, and 0.1% formic acid solution were then analyzed by a UltiMate 3000 RSLCnano-LC system (Thermo Fisher Scientific) connected online via a nano-ESI source to a Q Exactive plus™ Hybrid Quadrupole-Orbitrap™ Mass Spectrometer (Thermo Fisher Scientific). Each sample was run twice as a technical replicate. Proteins were automatically identified using MaxQuant (v. 1.6.17.0). Tandem mass spectra were searched against the Homo sapiens dataset of UniprotKB database. Quantitative comparisons among different samples were performed using the label-free quantification (LFQ) algorithm calculated by MaxQuant. Perseus software (version 1.6.15.0) (28) was used to analyze LC–MS data. Protein intensities obtained from MaxQuant LFQ values were log2-transformed and grouped to compare healthy donors (HD) with COVID-19 patients, stratified by disease severity into mild and severe. Proteins quantified in at least one group with ≥70% coverage were retained, and missing values were imputed using a normal distribution (width = 0.3, downshift = 1.3). The complete data matrix was subjected to Principal Component Analysis (PCA), and the first two components were extracted (FDR < 0.05). For volcano plots, 250 randomized permutation tests based on a two-sided t-test (FDR < 0.05, S0 = 0.1) were applied to identify differentially abundant proteins (DAPs) between mild and severe COVID-19. Hierarchical clustering was performed after Z-score normalization, followed by ANOVA and Tukey’s HSD post hoc tests (FDR < 0.05). Clusters were generated using Euclidean distance, yielding 12 clusters across HD, mild, and severe groups. DAPs between mild and severe cases were further analyzed for pathway enrichment using the STRING app in Cytoscape, querying Reactome, KEGG, and WikiPathways databases (FDR < 0.05, redundancy = 0.4). Enrichment results were visualized as a bubble plot in GraphPad Prism 9.0, where bubble color reflected FDR values and bubble size indicated the number of DAPs associated with each pathway term.

To assess the effect of the EVs obtained from sera of COVID-19 patients and HD on the functionality of Vδ2 T cells, we stimulated PBMCs from 11 HD with a Vδ2-specific antigen (phosphoantigen, PhAg, IPH1101, 3 µM, Innate Pharma, France) for 18–20 hours at 37 °C and 0.05% of CO₂. The resting or PhAg-stimulated PBMC were then treated with EVs from COVID-19 patients and HD [1.7 × 10^9 EVs on 150,000 T γδ (1.14 × 10^9 EVs/cell)] for 18–20 hours. After 1 hour, Brefeldin A (Serva, Euroclone, Italy) was added at 10 µg/mL to block cytokine secretion. At the end of stimulation, PBMCs were stained with a cocktail of monoclonal antibodies: Vd2-FITC, CD3-Percp-Cy5.5, and HLA-DR-ECD. Briefly, PBMCs were stained with mAbs for 20 minutes at 4 °C, then washed with PBS once and fixed with 1% paraformaldehyde (Sigma, Italy). After the fixation, cells were washed and stained for TNF-α (PE) for 15 minutes at RT. Cells were washed and acquired on the DXFlex cytometer (Beckman Coulter). The gating strategy is described in Supplementary Figure 3. Data were analyzed with DxFlex software 2.0.2.18 (Beckman Coulter).

The non-parametric Mann-Whitney test was used in this study. The statistical analyses were performed using the GraphPad Prism software 8.0. For proteomic data, all the statistical analyses were performed using Perseus software as described above.

In this study, 17 severe and 25 mild COVID-19 patients hospitalized at the National Institute for Infectious Diseases Lazzaro Spallanzani with a laboratory-confirmed diagnosis of SARS-CoV-2 infection were selected, according to the WHO categorization. The 76.4% (13/17) of severe COVID-19 subjects were hospitalized in the Intensive Care Unit (ICU). Among the cohort of severe subjects, 8 ICU hospitalized-subjects died. 23 healthy donors (HD) were enrolled as the control group. Table 1 describes the clinical and biological features of the study cohort, showing age media, gender, hospitalization status, and mortality.

The sample cohorts enrolled in this study were first analyzed for plasmatic inflammatory cytokine detection by ELISA. As expected, severe patients showed higher levels of pro-inflammatory cytokines [IL-6 (p<0.0001), IL-8 (p<0.0001), and TNF-α (p<0.0001)] with respect to both HD and mild patients [IL-6 (p<0.0001), IL-8 (p<0.0001)]. (Figure 1A). EVs extracted from plasma samples of severe and mild patients, as well as of HD, were isolated through a nickel-based approach. To analyze the cellular origin of the EVs, we compared the expression of key markers of platelets (CD31+/CD41+), leukocytes (CD45+/CD41-), and endothelium (CD31+CD41-) among severe, mild COVID-19, and HD. Data in Figure 1B, show a higher frequency of platelet-derived EVs in severe subjects than in HD (p<0.0006) and in mild subjects than in HD (p<0.003). No differences were observed between endothelial and leukocyte EVs. EVs among the three groups. Additionally, both severe and mild COVID-19 patients showed a higher number of EVs-platelet-derived compared to the EVs-endothelial-derived (severe p<0.0001; mild p<0.0006) and to the Leukocytes-derived EVs (severe p<0.0001 and mild p<0.0006).

To characterize EVs in term of nanoparticle size, concentration and proteomic cargo, a nanoparticle tracking analysis following by quantitative proteomic analysis was performed on EVs purified from severe (N = 8) and mild COVID-19 (N = 12) patients compared to those of the HD (N = 5). The clinical and biological features of this sub-cohort are reported in Table 2.

As shown in Figure 2A, the number of nanoparticles in severe COVID-19 samples was significantly lower compared to HD and mild COVID-19 EV samples; on the other hand, nanoparticle mean size was higher in mild and severe COVID-19 samples with respect to HD (Figure 2B).

A total of 348 proteins were identified by MaxQuant software (Supplementary Table 1). Among them, in accordance with the “minimal experimental requirements for definition of EVs and their functions,” several proteins that properly assess EVs’ features were identified (29). In particular, we identified EVs transmembrane proteins associated with the plasma membrane and/or endosomes (LAMP2, ITGA2B), cytosolic EVs proteins (HSPA8, GAPDH), and secreted proteins recovered with EVs also or “corona “proteins (FN1, COL14A1, LGALS3B) (Figure 2C). Furthermore, also gene ontology analysis revealed that the most enriched cellular compartments in EVs extracts analyzed are extracellular space, blood microparticle and extracellular vesicle, as expected (Figure 2D). Next, to identify differentially abundant proteins (DAPs) across three sample datasets, a computational analysis was conducted. Principal component analysis (PCA) revealed three separated cluster groups corresponding to healthy donors, mild, and severe COVID-19 samples (Figure 3A). To highlight statistically significant DAPs between severe and mild COVID-19 EVs samples, a Volcano plot analysis was applied. In particular, 68 DAPs were increased and 125 DAPs decreased in severe COVID-19 EVs samples with respect to mild COVID-19 EVs (Figure 3B; Supplementary Table 2). Among them several proteins involved in complement activation (C4, C5, C6, C9), coagulation cascades (SERPINF2, FN1) and SARS-CoV-2 signaling pathway (SAA1, SAA2, CRP, FGA) were identified. Indeed, bubble plot analysis based on DAPs detected by Volcano plot reveals Innate Immune System, platelet degranulation and network map of SARS-COV-2 signaling pathway as the most significant enriched pathways modulated between mild and severe patients (Figure 3C). All the DAPs belonging to each enriched pathway are listed in Supplementary Table 3. Taken together, these proteomic data provide evidence that circulating EVs derived from severe and mild COVID-19 patients differ in protein composition and their specific cargoes, suggesting a role for EVs in innate immune system regulation during COVID-19 disease.

Given that the innate immune system emerged as one of the most significantly enriched pathways in severe COVID-19, we focused on Vδ2 T cells, a subpopulation able to produce large amounts of pro-inflammatory mediators. To investigate the immunomodulatory effect of EVs obtained from COVID-19 patients on the Vδ2 T cells response, PhAg-stimulated PBMC from HD were incubated with EVs derived from severe (n=8) and mild (n=10) COVID-19 subjects, and from healthy controls (n=5) also subjected to proteomic analysis. PhAg-stimulated PBMCs without EVs were used as a control.

At first, we analyzed the modulation of the HLA-DR on Vδ2 T cells. As shown in Figure 4A, EVs from severe COVID-19 patients induced a higher frequency of Vδ2 T cells expressing HLA-DR compared to EVs from mild (p<0.0001) or EVs-untreated control (p<0.05). Moreover, we analyzed the effect of EVs on Vδ2 T cell functionality. As shown in Figures 4A, B, EVs from severe COVID-19 patients induced a higher frequency of TNF-α-producing Vδ2 T cells compared with EVs from mild patients (p < 0.0002) and with EV-untreated cultures (p < 0.0001). Moreover, the treatment with EVs from mild patients induced a significant decrease compared EVs-untreated control (p<0.03). In contrast, no significant effects of EVs derived from the sera of HD were observed.