Forty five patients who met the 2001 International League for Associations of Rheumatology (ILAR) classification criteria for OJIA (i.e. disease involvement of ≤4 joints in the first 6 months of disease) (2) undergoing arthrocentesis at the Pediatric Rheumatology, IRCCS Gaslini Institute, Genova, SC Pediatric Immuno Rheumatology, IRCCS Ca’ Granda Maggiore Hospital, Milano, and Pediatric Rheumatology and Immunology, Hospital “Vito Fazzi”, Lecce, as part of clinical care were enrolled consecutively in the study from October 2018 through June 2020 and followed up every 3 months for 2 years after disease diagnosis (June 2022). All patients had clinically active disease, with joint effusion, swelling, pain, and stiffness at the time of sampling, and onset of symptoms for no more than 6 months before enrollment. The number of active joints was determined by standard clinical evaluation followed by ultrasound or MRI evaluation. Arthrocentesis was performed at the time of disease diagnosis under local anesthesia or, in case of younger patients or multiple joints, under general anesthesia. Various clinical and laboratory parameters were measured. A previous or current treatment with anti-inflammatory drugs was considered as an exclusion criterion. Patients underwent different therapeutic regimens after diagnosis during the follow-up period. Disease relapse (referred to as occurrence of new flares within the onset joints or other joints), polyarticular extension, and iridocyclitis development were determined prospectively during the longitudinal follow up evaluation. No attempt was made to segregate patients on the basis of disease duration, number of active joints involved at disease onset, or therapeutic regimen during follow-up. Twenty-four age- and gender-matched children undergoing minor orthopedic procedures at the Gaslini Institute were enrolled as a control group. Detailed clinical and laboratory examination of control subjects was carried out to rule out infections, inflammatory, and chronic diseases. The main demographic, clinical, laboratory, and therapeutic characteristics of the study cohorts are reported in Table 1 . The protocol of the study was reviewed and approved by the Ethics Committee of the Region Liguria (Approval 165/2018), the Ethics Committee Milano Area 2 (Approval 639/2019), and the Ethics Committee ASL Lecce (Approval N° 36/2019), and the procedures were carried out according to the approved guidelines and in adherence to the general ethical principles set forth in the Declaration of Helsinki. Written informed consent to participate in the study was obtained from the parents or the patient’s legal guardian prior to sample collection.

Synovial fluid (SF) aspirates were collected from affected joints of thirty OJIA patients (SF_OJIA) at the time of arthrocentesis under vacuum into tubes containing EDTA, as described in Ref (45). Peripheral blood (PB) samples from thirty OJIA patients (PL-OJIA) and twenty-four control children (PL_CTR) were obtained by venipuncture and collected in EDTA tubes. Among them, fifteen PB and SF samples were derived from the same OJIA patients ( Figure 1 ). Specimens were centrifuged at 500 x g for 10 minutes at room temperature (RT) to obtain cell-free SF and PL and stored at -80°C until use.

EV isolation from 500 μl of SF and PL samples was performed using the exoRNeasy Serum/Plasma Midi kit (Qiagen Italia, Milano, Italy), that uses membrane affinity spin columns to efficiently capture intact EVs from small sample volumes, according to the manufacturer instructions, as described (45). Cell-free SF samples were either treated with 2U/ml Hyaluronidase (HYase) (Sigma, Merck Life Science, Milano, Italy) for 30 minutes at 37°C to remove contaminating hyaluronan extracellular matrix (ECM) components. Samples were firstly centrifuged at 500 x g for 10 minutes at RT and then at 16,000 × g for 15 minutes at RT to eliminate cellular debris. PL samples, not undergoing HYase treatment, were directly centrifuged at 16,000 × g for 15 minutes. Supernatants were mixed with one volume of XBP binding buffer and loaded onto the exoEasy spin column to bind EVs to the membrane. After centrifugation at 500 x g for 1 minute at RT, the flow-through was discarded and the column was washed with 2 ml of XWP Washing Buffer and spun at 2,500 ×g for 5 minutes at RT. This step was repeated twice to properly remove any contaminating agent. Then, the column was washed with 3.5 ml of XWP Washing Buffer and spun at 5,000 x g for 5 minutes at RT, to wash-off not specifically bound material. The spin column was transferred to a new collection tube and EVs were lysed directly in the column by addition of 100 μl of RIPA Buffer (Thermo Fisher Scientific, Milano, Italy) and incubated for 5 minutes at 4°C. The column was then centrifuged 500 × g for 5 minutes, and EV-prot lysates were collected.

EV lysates were processed according to the in-StageTip procedure, as described previously (46, 47). 100 ul of EV lysates in RIPA buffer were reduced and alkylated with 10 mM TCEP and 4 mM CAA in thermomixer 10 min at room temperature and at 1000 rpm. Proteins were then isolated by the PAC method (48). Briefly, protein aggregation was induced by addition of 70% CAN, and 200 μg of magnetic beads were added to capture aggregated proteins. Magnetic beads were retained by the magnet and the supernatant was removed. Beads were washed one time with 1 ml acetonitrile, followed by one wash with 1 ml 70% ethanol and one wash with 1 ml isopropanol. Washed beads were resuspended in 100 μl TRIS 25 mM pH 8, and captured proteins were digested O.N. at 37°C with 0.7 μg Trypsin and 0.3 μg LysC. Obtained peptides were desalted in Stage-Tips (49) (normalization to 7 ug peptides) and analyzed by a nano-UHPLC-MS/MS system using an Ultimate 3000 RSLC coupled to an Orbitrap Fusion Tribrid mass spectrometer (Thermo Scientific Instrument). Peptide elution was performed with an EASY spray column (75 μm x 25 cm, 2 μm particle size, Thermo Scientific) at a flow rate of 400 nl/min using a linear gradient of 7-45% solution B (80% ACN and 20% H2O, 5% DMSO, 0.1% FA) in 65 min. MS analysis was performed in DDA mode. Orbitrap detection was used for MS1 measurements at a resolving power of 120K in a range between 375 and 1500 m/z and with an AGC target of 400000, maximum injection time 50 ms. Advanced Peak Determination was enabled for MS1 measurements. MS/MS spectra were acquired in the linear ion trap (rapid scan mode) after higher-energy C-trap dissociation (HCD) at a collision energy of 28% and with an AGC target of 10000, maximum injection time 22 ms. 2 sec. cycle time was performed for data dependent MS/MS analysis, during which precursors with a charge range 2-5 were selected for activation in order of abundance. Quadrupole isolation with a 1.6 m/z isolation window was used, and dynamic exclusion was enabled for 15 s.

Protein raw data were processed with MaxQuant software (50) version 2.0.3.0. A false discovery rate (FDR) of 0.01 was set for identifying proteins, peptides, and peptide-spectrum match (PSM). For peptide identification, a minimum length of 7 amino acids was required. Andromeda engine, incorporated into MaxQuant software, was used to search MS/MS spectra against the Uniprot human database (release UP000005640_9606 July 2021). In the processing the Acetyl (Protein N-Term), Oxidation (M) and Deamidation (NQ) were selected as variable modifications and the fixed modification was Carbamidomethyl (C). Quantification intensities were calculated by the default fast MaxLFQ algorithm with the activated option ‘match between runs’. The resulting protein groups were analyzed using the Perseus software, version 1.6.15.0 (51). Contaminants and decoys were filtered out and LFQ values were subsequently log2 transformed for further statistical analysis. Proteins with at least 30% of valid values in each group were retained for the analysis prior to any relative quantification. Missing values (NA) in retained samples were imputed (52, 53) from a normal distribution using a width of 0.3 and a downshift of 1.8.

Principal Component Analysis (PCA) with default parameters was performed using Perseus algorithm to visualize correlated groups of data. The first two principal components were plotted. Volcanoplot and heatmap representation with unsupervised hierarchical clustering analysis were carried out to visualize proteomic data using the Perseus software. Overlapping and exclusive elements among the lists of EV-prots were defined by Venn diagrams using InteractiVenn web-based tool (54).

Protein-protein interaction (PPI) network analyses were carried out on selected significantly deregulated EV-prot using the Search Tool for the Retrieval of Interacting Genes (STRINGApp in Cyoscape 3.9.1) (55) to construct functional interaction networks among proteins. PPI enrichment p-value was used to assess whether proteins in a group of have more interactions among themselves than what would be expected from a random set of proteins of similar size, drawn from the genome. We set up a required minimum confidence score of 0.4 (medium confidence) and considered significant PPI enrichment p value <0.05.

Gene ontology (GO) enrichment analysis was conducted on deregulated proteins using the ShinyGO web server (version 0.76.1), a large annotation database derived from Ensembl and STRING-db (56). The human proteome served as a reference for the GO enrichment analysis. EV-prot were classified according to the GO biological process collection. An FDR < 0.05 was considered significant, and fold enrichment scores were used to assess the significance of resulting GO terms and pathways enrichment.

Weighted correlation network analysis (WGCNA) was carried out separately on SF_OJIA and PL_OJIA using a specific software R package in the Perseus platform to construct coexpression modules associated with patient clinical parameters, as described in (57, 58), after filtering of 50% of valid values in total for both PL_OJIA and SF_OJIA. The power parameter, selected by a ‘‘Soft-threshold’’ activity set on a scale-free fit index of 0.9, for PL_OJIA was 8, while for SF_OJIA was 18. Both network types were set to signed with a cor correlation function.

The potential confounding effect of age or sex (male vs female) between CTR subjects and OJIA patients was assessed by Mann-Whitney test or Fisher’s exact test, respectively. Normality or lognormality of numerical distributions was assessed by D’agostino & Pearson test. P values lower 0.05 were considered statistically significant.

Two-sample Student’s t-test was used to identify differences in the EV-prot expression levels when the comparison involved samples derived from OJIA patients and CTR subjects (i.e., SF_OJIA vs PL_CTR and PL_OJIA vs PL_CTR). Paired Student’s t-test was applied to determine the significance of EV-prot expression differences between paired PL and SF samples, after further filtering for the minimum 30% of valid values in each of the two groups (e.g., SF_OJIA vs PL_OJIA). For both two-sample Student’s t-test and paired Student’s t-test only EV-prots with log2 fold change higher than 1 or lower than -1 were considered significantly modulated. To reduce the probability of false positive findings deriving from multiple hypothesis testing a permutation-based false discovery rate (FDR) p-value lower than 0.01 was applied.

Two sample Z-test for proportions was used to detect EV-prots whose distribution of missing values between OIJA and CTRL samples resulted statistically significantly different. Proteins with z-test scores lower than -4.0 or higher than 4.0 (p<0.00005) were considered to have statistically significant differences in the distribution of missing values. When PL and SF samples were relative to the same patients, the statistically significant difference in the distribution of missing values was assessed by a matched pair binomial test. Differences with p-value lower than 0.0001 were considered statistically significant. Study sample size (power) calculation was performed by G*Power software version 3.1 (59).

Experiments were carried out to characterize the protein expression profile of EVs released into the SF and PL of new-onset OJIA patients. Forty-five OJIA patients with clinically active disease were enrolled in the study at the time of disease diagnosis and therapeutic arthrocentesis, and SF and PL samples were collected. PL samples were obtained in parallel from twenty-four age- and gender-matched CTR children. The demographic (age and sex) and main clinical features of patients and control subjects, various laboratory parameters and known markers of disease activity such as erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), and anti-nuclear antibodies (ANA), the number and type of active joints at onset, therapeutic regimens administered after diagnosis, disease oligoarticular or polyarticular course, and iridocyclitis development within a 2 years follow-up period are reported in Table 1 . The 2 years follow up was set up because polyarticular extension in OJIA occurs more frequently within this time after initial disease presentation (9). Mean patient age at onset was 5.2 years and female/male ratio was 25 (55,6%)/20 (44,4%), whereas mean control subject age was 7.2 years and female/male ratio was 13 (54.2%)/11 (45.8%). Association between age or sex (male vs female) and subject group (OJIA vs CTR) was investigated to estimate whether demographic features could represent potential confounding factors in the present study. Mann-Whitney test and Fischer’s exact test did not show any significant association between the case and control groups analyzed with regard to age or sex, respectively (p>0.05), thereby excluding a potential confounding effect of these characteristics. Positivity for ANA was observed in nineteen patients (42%). Twenty-three patients (51%) had involvement of 1 joint, whereas sixteen patients (35.5%) had 2 affected joints, and only six patients (13%) had arthritis in more than 2 joints. The most frequently involved joint was the knee (fourty-three patients, 95.5%), whereas all other joints were affected in three or fewer patients (< 7%). The most common therapeutic intervention after diagnosis was intra-articular glucocorticoids (all patients), whereas only a subset of ten patients was given methotrexate (MTX) (22.2%) and two received NSAID (4.4%). Treatment with MTX or other anti-inflammatory drugs was often administered to patients showing disease relapse. At the end of the 2 years follow up, twenty-eight patients (62.2%) had experienced new disease flares, eleven (24.4%) developed polyarticular extension, while thirty-four (75.6%) maintained an oligoarthritis phenotype, and seven patients (15,6%) developed iridocyclitis. The general features of the cohort of patients enrolled in the present work are similar to those described in other previous studies (9). We performed an a priori sample size calculation with G*Power software (effect size d=1.0, α = 0.05, 1-β = 0.95, standard deviation=1.0) to assess a fold change difference of at least 2.0 between OJIA patients and CTR groups. The total sample size resulted of 54 subjects, which is below the total number of subjects (patients and CTR) enrolled in the study, confirming that the study contains enough power to make a reasonable conclusion.

EVs were isolated from collected samples according to our recently set up experimental protocol (45). The characteristics of EVs isolated using this procedure in terms of size, morphology, and marker expression were reported previously (45). The number of EVs isolated from SF and PL ranged between 5.14x10¹⁰-1.22x10¹¹ and 7.06x10¹⁰-1.48x10¹¹ particles/ml, respectively, as quantified by NTA. EV protein cargo was profiled using high-resolution mass spectrometry coupled with liquid chromatography (LC-MS/MS). A flowchart summarizing the main steps of sample processing and data analysis is depicted in Figure 1 . A total of 2444 proteins were identified. Proteins resulting after filtration of the most common contaminants (e.g., albumin, cytoskeletal keratins, collagen-alpha-1, tropomyosin-beta, thrombospondin-1) were 2282. Greater protein abundance was detectable in both SF (average number 1015 ± 31.9) and PL specimens (average number 1013 ± 23.4) from OJIA patients respect to PL from CTR (average number 657 ± 17.7) ( Figure 2A ).

EV-prot expression patterns were then determined in the different groups of specimens. We carried out both EV-prot differential expression and missing value distribution analysis. Differential expression was evaluated by comparing the levels of common EV-prots among groups by the Student’s t test method, applying a filter based on NAs. Only EV-prots with at least 30% of valid values in each group were retained for imputation and analysis, whereas the rest was filtered out (as detailed in the Materials and Methods). The filtration step reduced the total number of proteins to 594. As shown by the PCA reported in Figure 2B , SF_OJIA, PL_OJIA, and PL_CTR samples could be clearly clustered into three well-defined groups based on their EV-prot expression profiles, with substantial homogeneity among samples belonging to the same group. Analysis of the distribution of missing values was investigated on all the 2282 EV-prots to identify those exclusively expressed in each sample group. A matched pair binomial test was applied to analyze paired PL_OJIA vs SF_OJIA samples, whereas the two sample Z-test for proportions was applied to analyze unpaired OJIA and CTR samples.

To identify EV-prots potentially implicated in OJIA joint pathogenesis, we conducted a comparative analysis of the proteome of EVs isolated from SF_OJIA respect to that of paired PL_OJIA samples from newly-diagnosed OJIA patients. Statistically significant differences in EV-prot expression levels between groups were visualized by volcano plot and heat map representation ( Supplementary Figure 1 ). Volcano plot revealed a clear separation between SF and PL samples on the basis of the EV-prot expression profiles (Panel A). Unsupervised hierarchical clustering analysis confirmed the presence of distinct EV-prot patterns in paired SF and PL samples, with substantial homogeneity among samples belonging to the same group (Panel B). To define EV-prots that mostly contributed to the observed differences, we performed a differential expression analysis of the proteomics datasets and identified a total of 235 statistically significantly modulated EV-prots (FDR<0.01) between the two groups of specimens, of which 110 and 125 were specifically up- and downregulated in SF vs paired PL, respectively. The list of modulated EV-prots is reported in Supplementary Table 1 . These results indicate that the protein expression profile of EVs isolated from the joints of new-onset OJIA patients strongly differs from the systemic profile.

Because the observed differences in EV-prot expression levels could be due at least in part to the diverse nature of the fluid analyzed (SF vs PL) and, thus, not be specific to the disease state, we compared the 235 EV-prots identified in SF_OJIA vs PL_OJIA with those resulting from the comparison between SF_OJIA and PL_CTR samples and filtered out proteins commonly differentially expressed in the two comparisons and thus potentially related to the specific fluid characteristics. Volcano plot ( Supplementary Figure 1C ) and heat map ( Supplementary Figure 1D ) clearly clustered SF_OJIA and PL_CTR specimens into two well-defined groups based on their EV-prot expression levels, with a total of 268 proteins (130 up- and 138 down-regulated) identified as significantly differentially expressed (FDR < 0.01) ( Supplementary Table 2 ). Common and exclusive differentially expressed proteins resulting from the comparison of the two datasets were visualized by the Venn diagram and the Vulcano plot depicted in Figures 3A, B . 152 EV-prots were commonly modulated in paired SF vs PL from OJIA patients and SF_OJIA vs PL_CTR (indicated by red and blue circles in the Volcano plot), whereas a restricted subset of 83 EV-prots was specifically modulated upon comparison of SF_and PL_OJIA samples (black circles in the Vulcanoplot), probably representing protein specifically related to the disease. Among them, 31 were up-regulated and 52 down-regulated ( Figures 3A, C ). The complete list of this panel of proteins is reported in Table 2 . These data demonstrate the existence of a SF-associated EV-prot signature with potential value as an early molecular indicator of the joint pathologic state.

Understanding the biological processes in which EV-prots are implicated may help to elucidate disease molecular mechanisms and be instrumental for the identification of new putative biomarkers and therapeutic targets. To help interpret the data biologically and define regulated biological processes in OJIA joints, network and pathway analysis were then carried out on the selected subset of 83 EV-prots found specifically differentially expressed in SF_OJIA vs PL_OJIA samples using STRINGApp and the ShinyGO web server, respectively. Network analysis demonstrated that modulated proteins displayed significantly more interactions than those expected for a random set of proteins of similar size drawn from the genome (PPI enrichment p-value<0.05), indicating their biological connection as a group ( Figures 3D, E ). Pathway analysis showed the significant enrichment (FDR ≤ 0.05) of 547 GO biological processes (283 associated to up- and 263 to down-regulated EV-prots) in SF with respect to PL samples ( Supplementary Table 3 ). A selection of the most significantly enriched terms associated to up- and down-regulated EV-prots is depicted in Figures 3D, E , respectively. Proteins associated with each selected pathway are listed in Supplementary Table 4 . The majority of up-regulated EV-prots was implicated in biological processes related to cartilage/bone organization, such as regulation of extracellular matrix (ECM) disassembly and collagen carbolic processes, chondroitin sulfate metabolic processes, connective tissue and skeletal tissue development, regulation of peptidase activity, cell adhesion, motility and migration,. They include collagen alpha-1(II) chain (COL2A1), chondroitin sulfate proteoglycan 4 (CSPG4), fibroblast activation protein (FAP), nidogen-2 (NID2), vitronectin (VTN), low-density lipoprotein receptor-related protein 1 (LRP1), EGF-containing fibulin-like extracellular matrix protein 1 (EFEMP1), follistatin-like protein 1 (FSTL1), protein S100-A8 (S100A8), lymphocyte function-associated antigen-3 (CD58), macrophage colony-stimulating factor 1 (CSF1), and cluster of differentiation 14 (CD14), Conversely, down-regulated EV-prots were mostly involved in the regulation of actin filament polymerization and cytoskeleton organization, including actin-alpha 1 (ACTA1), actin-related protein 2/3 complex subunits (ARPC1B,2,3,4,5), actin-related protein 2 (ACTR2), profilin-1 (PFN1), tropomyosin alpha-3 chain (TPM3), cofilin-1 (CFL1), adenylyl cyclase-associated protein 1 (CAP1), zyxin (ZYX), and vasodilator-stimulated phosphoprotein (VASP). Many down-regulated proteins were implicated in inflammatory and innate immune processes, such as neutrophil, granulocyte, and myeloid leukocyte degranulation, activation, and -mediated immunity, immune effector process, cell secretion, and exocytosis, e.g. platelet endothelial cell adhesion molecule (PECAM1), bridging integrator 2 (BIN2), Src kinase-associated phosphoprotein 2 (SKAP2), and receptor-type tyrosine-protein phosphatase eta (PTPRJ, CD148).

The distribution of missing values between paired SF and PL samples from OJIA patients was then investigated by two-tailed matched pair binomial test to identify EV-prots expressed exclusively in the joints or in the circulation of OJIA patients and thus obtain further insights into the mechanisms of disease development. P value ≤ 0.0001 was considered statistically significant. This analysis identified subsets of 33 and 28 EV-prots exclusively expressed in SF or PL groups, respectively (p-value<0.0001, Table 3 ). Heatmap visualization of EV-prot expression values showed a clear association of significant EV-prots and SF or PL samples ( Figure 4A ), visually confirming the results of the binomial test. Network analysis established a significant connection among some identified EV-prots (PPI p-value<0.05, Figures 4B, C ). Pathway analysis of these proteins showed the significant enrichment (FDR ≤ 0.05) of 98 GO biological (41 related to EV-prots expressed in SF and 57 in PL) ( Supplementary Table 5 ). A selection of the most significantly enriched terms is depicted in Figures 4B, C , and the list of EV-prots specifically associated with each pathway is shown in Supplementary Table 6 . EV-prots expressed exclusively in SF were involved in several innate immune pathways and cell/matrix interactions, including ECM organization, regulation of dendritic cell (DC) antigen processing/presentation and of neutrophil activation, negative regulation of phagocytosis, macrophage, neutrophil, and granulocyte activation, leukocyte degranulation and -mediated immunity, immune effector processes, and cell adhesion, Among them, collagen alpha-1(V) chain (COL5A1), hyaluronan and proteoglycan link protein 1(HAPLN1), integrin alpha-M (ITGAM), integrin subunit alpha x (ITGAX), protein S100-A10 (S100A10), CD74 (HLA class II gamma chain), and low affinity Ig-gamma Fc region receptor II-b (FCGR2B) are critical for the maintenance and the normal metabolism of joint tissues.

These results characterize the biological processes regulated by EV-prots within the joints of OJIA patients at an early stage of the disease and identify EV-prots potentially implicated in disease pathogenesis, representing new early putative joint-specific biomarkers of OJIA development.

Proteome evaluation was then carried out on the full cohort of PL_OJIA specimens compared to PL_CTR samples to identify EV-prots that could represent early putative diagnostic biomarkers. As shown by the Volcano plot depicted in Figure 5A , patients and control children expressed clearly distinct EV-prot patterns, which was confirmed by unsupervised hierarchical clustering analysis shown in the heat map ( Figure 5B ). We detected significantly altered expression levels of a total of 202 EV-prots between groups, of which 110 were upregulated and 92 downregulated in patient specimens, suggesting specific disease-related changes in the expression levels of circulating EV-prots. The complete list of differentially regulated EV-prots is reported in Supplementary Table 7 .

Network analysis demonstrated that modulated proteins displayed significant functional interactions (PPI enrichment p-value<0.05) ( Figures 5C, D ). Pathway analysis identified 1099 significantly enriched (FDR ≤ 0.05) biological processes (802 associated to up- and 297 to down-regulated EV-prots) ( Supplementary Table 8 ). A selection of the processes with the most significant enrichment score and associated proteins are reported in Figures 5C, D and Supplementary Table 4 . Among them, several are related to inflammation and cytoskeleton organization, including: (i) neutrophil, granulocyte, and myeloid leukocyte degranulation, activation, and -mediated immunity, immune effector processes, cell secretion, exocytosis and actin cytoskeleton organization for up-regulated proteins; (ii) complement activation, ECM organization, regulation of peptidase activity and proteolysis, angiogenesis, cell adhesion, migration, and motility for downregulated proteins. Proteins mostly involved in these processes include several Ras-related proteins (e.g. Rab-11B, RAB8A, RAP1B), Rab GDP dissociation inhibitor beta (GDI2), heat shock protein A8 (HSPA8), calpain-1,2 catalytic subunits (CAPN1), catalase (CAT), glutathione S-transferase omega-1 (GSTO1), and annexin A1 (ANXA1), matrix-remodeling-associated protein 5 (MXRA5), carboxypeptidase B2 (CPB2), respectively.

Interestingly, some of the EV-prots differentially modulated in PL_OJIA vs PL_CTR overlapped with those observed in SF_OJIA vs PL_CTR. Common and exclusive differentially expressed proteins were visualized by the Venn diagram depicted in Figure 5E following comparison of the two datasets. Among common EV-prots, 76 displayed consensual regulation; specifically 38 up-regulation, e.g. HSPB1, HSPA1B, platelet glycoprotein 4 (CD36), proto-oncogene tyrosine-protein kinase Src (SRC), lyn proto-oncogene (LYN), peroxiredoxin-1 (PRDX1), thioredoxin (TXN), glutathione S-transferase P (GSTP1), pyruvate kinase m1/2 (PKM), glyceraldehyde-3 phosphate dehydrogenase (GAPDH), L-lactate dehydrogenase a chain (LDHA), triosephosphate isomerase (TPI1), phosphoglycerate kinase 1 (PGK1), and 38 down-regulation, e.g. complement factor H-related protein 1 (CFHR1), mannose-binding protein C (MBL2), mannan binding lectin serine peptidase 1 (MASP1), collectin-10 (COLEC10), insulin-like growth factor-binding protein 3 (IGFBP3) ( Supplementary Tables 2, 7 underlined proteins), potentially representing disease state indicators measurable at both the systemic and local levels able to discriminate new-onset OJIA patients from healthy children.

Two sample Z-test for proportions was used to detect EV-prots whose distribution of missing values resulted statistically significantly different between the full cohort of OJIA and CTR sample groups. EV-prots with z-scores lower than -4.0 or higher than 4.0 (p<0.00005) were considered significantly exclusively expressed. This analysis showed that a total of 410 and 294 EV-prots were exclusively expressed in SF_OJIA or PL_OJIA, respectively, whereas 90 and 7 were exclusively expressed in PL_CTR as compared to SF_OJIA or PL_OJIA ( Supplementary Table 9 ). Heat map analysis showed EV-prot association with OJIA patients or CTR subjects, visually confirming the results of the z-test ( Figures 6A, C ). Network analysis carried out on the EV-prots expressed exclusively in SF_OJIA ( Figure 6B ) and PL_OJIA ( Figure 6D ) showed significant interactions among these proteins (PPI p-value<0.05). Pathway analysis revealed the significant enrichment (FDR ≤ 0.05) of a total of 821 biological processes related to EV-prots exclusively expressed in SF and 613 to those exclusively expressed in PL, whereas 288 were enriched with proteins expressed in PL_CTR and not in SF ( Supplementary Tables 10, 11 ). In contrast, no biological pathway associated with the 7 proteins expressed in PL_CTR and not in PL_OJIA was found. A selection of the most significantly enriched GO terms and the list of associated proteins are shown in Figures 6B, D and Supplementary Table 6 . Proteins identified in both SF and PL were involved in inflammatory responses, being mostly related to neutrophil, granulocyte, and leukocyte degranulation, activation, and mediated immunity, regulated exocytosis, and immune effector processes. Proteins detectable in SF were also involved in cytoplasmic translation, protein targeting to membrane and endoplasmic reticulum, mRNA and peptide metabolic processes, whereas a significant number of those detectable in PL are critical for cytoskeleton organization being mainly implicated in actin polymerization, Golgi vesicle transport, antigen processing and presentation. CRP, serum amyloid A (SAA1), cartilage intermediate layer protein 1 (CILP), RALB, RAB10, Rho GTPase-activating protein 18 (ARHGAP18), ras-related proteins (RAB1A, RAB5B, RAB14), heat shock 70 kDa protein 4 (HSPA4), plasminogen-activator inhibitor 1 (SERPINE1, B6,9), SLAM family member 5 (CD84), signal transducer and activator of transcription 3 (STAT3), spleen associated tyrosine kinase (SYK), triggering receptor expressed on myeloid cells-like transcript-1 (TREML1), and drebrin 1 (DBN1) were some of the EV-prots identified in PL_OJIA vs PL_CTR samples. As shown by the Venn diagram depicted in Figure 6E , 77 EV-prots were shared by the PL_OJIA and SF_OJIA datasets, among which Ras homolog gene family member a (RhoA), cell division control protein 42 homolog (CDC42), Ras-related protein-A (RALA), RAB5C, RAB35, ITGA5, basigin (BSG, CD147), and leukocyte elastase inhibitor (SERPINB1) ( Supplementary Table 9 , underlined proteins), suggestive of a generalized disease state indicator.

Taken together, these data demonstrate that the expression levels of several EV-prots or exclusively expressed EV-prots subsets can differentiate new-onset OJIA patients from CTR children with potential diagnostic value.

Finally, we aimed at establishing the existence of an association between EV-Prot expression profile of SF-OJIA and PL_OJIA samples collected at disease onset and patient clinical parameters, such as ANA positivity, disease relapse, polyarticular extension, and iridocyclitis development within the 2 years of follow-up. WGCNA was run on the full cohorts of SF- and PL-derived EV-prot datasets characterized in newly-diagnosed OJIA patients to identify biologically significant clusters of proteins (modules) associated to the selected clinical parameters, as described in Ref (60). We reasoned that these EV-prot clusters could potentially represent early indicators of patient outcome. A heatmap was drawn based on the interactive relationships of identified co-expression modules with the OJIA traits ( Figures 7 , 8 ). The different clinical groups were clustered by hierarchical clustering and visualized by a dendrogram on the bases of their relation with different EV-prot modules. For SF_OJIA, WGCNA revealed a total of 9 EV-prot coexpression modules (each distinguished by a specific color) associated with the different clinical parameter considered, whereas for PL_OJIA, we identified a total of 7 EV-prot modules. With regard to SF samples ( Figure 7A ), the brown, yellow and green modules seemed the most interesting, being the ones that best discriminated the two clusters into which the clinical groups are divided. Modules brown and yellow were found significantly correlated to negativity for ANA, oligoarticular course, no iridocyclitis, and no relapse, whereas the green one was highly associated with positivity for ANA, polyarticular extension, iridocyclitis development, and relapse. With regard to PL samples ( Figure 8A ), the most interesting modules seemed to be the blue, which correlates with positivity for ANA, iridocyclitis development, and no relapse, and the turquoise, highly associated to ANA-, oligoarticular course, and no iridocyclitis. The list of proteins belonging to selected modules is shown in Supplementary Table 12 .

PPI network analysis and GO annotation of the most significant co-expression modules were then carried out to assess interconnectivity and explore their functions ( Figures 7B–D , 8B, C ). A large number of proteins within each module was connected as a group. Proteins associated to the most relevant functional pathways are shown in Supplementary Table 13 . In SF ( Figures 7B–D ), the brown module contained EV-prot related to ribonucleoprotein complex assembly, organization and biogenesis, mRNA metabolic processes, and protein translation. The most represented proteins in this pathway included several heterogeneous nuclear ribonucleoproteins (HNRNPs) (e.g. HNRNPA2B1, HNRNPD, HNRNPM, HNRNPH1, and HNRNPA3), high mobility group protein B1 (HMGB1), nucleolin (NCL), and proliferating cell nuclear antigen (PCNA). The yellow module contained proteins involved in the regulation of blood coagulation, ECM organization, regulation of protease activation and proteolysis, such as a few collagen molecules (COL2A1, COL6A1,2,3, COL11A1), tetranectin (CLEC3B), attractin (ATRN), lumican (LUM), osteomodulin (OMD), fibulin-1 (FBLN1), dentin matrix acidic phosphoprotein 1 (DMP1), fibronectin (1FN1), cartilage acidic protein 1 (CRTAC1), transforming growth factor-beta-induced protein (TGFBI), CD109 antigen (CD109), and neuropilin-1 (NRP1). The green module contained proteins with known regulatory effects on inflammatory/immune processes, such as granulocyte, neutrophil, and myeloid cell degranulation, activation and mediated immunity and immune effector processes, including SERPINB1, HSPA1B, immune costimulatory protein b7-h3 (CD276), neutrophil gelatinase-associated lipocalin (LCN2), low affinity Ig-gamma Fc region receptor III-A (FCGR3A), complement components C5 and C6, protein S100-A12, CRP, FAP, matrimetalloproteinase-9 (MMP9), myeloperoxidase (MPO), and secreted phosphoprotein 1 (SPP1).

In the PL ( Figures 8B, C ), biological processes of the blue module were mainly enriched in proteins related to glycosaminoglycan metabolic processes, ECM organization, post translational protein modification, angiogenesis, cell adhesion, migration and motility, whereas the turquoise module revealed proteins involved in the regulation of granulocyte, neutrophil, and myeloid cell degranulation, activation and mediated immunity, exocytosis, actin cytoskeleton organization, response to wounding, Proteins in the blue module include ATRN, FBLN1, CLEC3B, NRP1, TGFB1, CD109, CSF1, macrophage colony-stimulating factor 1 receptor (CSF1R), thrombospondin-2 (THBS2), fibrillin 1 (FBN1), endosialin (CD248), CD166 antigen (ALCAM), CD44 antigen, lymphocyte cytosolic protein 1(LCP1), lipopolysaccharide-binding protein (LBP), and, fibrinogen-like protein 1 (FGL1). Proteins in the turquoise module included SERPINB1, STAT3, SYK, LYN, HSPAIB, HSPB1, CD36, BSG, junctional adhesion molecule C (JAM3), ITGA5, PRDX1, SRC, PKM, aldolase A (ALDOA), solute carrier family 2 (SLC2A3), glucose-6-phosphate isomerase (GPI), disintegrin and metalloproteinase domain-containing protein 10 (ADAM10), P-selectin (SELP), and CD226 antigen (DNAM-1).

We can conclude that a few clusters of EV-prots highly associated with patient clinical parameters can discriminate subgroups of patients with different disease courses at 2 years of follow up, suggesting their potential as early biomarkers of disease outcome.