Inflammation-related proteomics of extracellular vesicles as novel biomarkers for systemic lupus erythematosus revealed by proximity extension assay

Abstract

Background

Systemic lupus erythematosus (SLE) is a complex autoimmune disease characterized by dysregulated inflammatory response lacking reliable diagnosis biomarkers and therapy targets. Extracellular vesicles (EVs)-derived cargo as biomarkers and mediators of SLE have garnered significant attention, however, quantitative inflammatory protein profile of SLE EVs remain uncovered.

Objective

Current study focuses on exploring the inflammatory protein landscape of SLE serum EVs via quantitative proximity extension assay (PEA) and evaluates their diagnostic utility for SLE and lupus nephritis (LN).

Methods

In this cross-sectional study, we first utilized PEA to profile inflammatory proteins derived from serum EVs in 101 individuals, including 70 SLE patients and 33 healthy controls (HCs). Subsequently, candidate EV proteins identified from this analysis were subsequently validated via ELISA in an independent cohort comprising 54 SLE patients and 58 HCs. Furthermore, machine-learning classification was utilized to generate prediction models for SLE diagnosis and LN discrimination. Finally, correlation analysis was applied to evaluate the association between EV-derived inflammatory proteins and clinical parameters.

Results

In the sEV PEA discovery cohort, a total of 49 significantly dysregulated inflammatory proteins with 43 elevated proteins were identified in serum EVs from SLE patients. Two precision prediction models were generated using the random Forest algorithm (RF) for SLE identification and LN discrimination, achieving AUCs of 0.999 and 0.793, respectively. Multiple EV proteins such as CCL23, IL-18R1, SCF and CSF-1 showed a significant correlation with SLE severity parameters including SLEDAI, eGFR and UACR. Furthermore, representative EV proteins including IL-18R1, CCL23 and IFN-γ were further tested in the sEV ELISA validation cohort including 54 SLE patients and 58 HCs.

Conclusions

The present study identified a unique pattern EV-derived inflammatory proteins in patients with SLE, which could serve as novel biomarkers for SLE diagnosis and disease monitoring.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13075-025-03681-x.

Introduction

Systemic lupus erythematosus (SLE) is a complex, heterogeneous, and debilitating autoimmune disease characterized by the production of autoantibodies and the deposition of immune complexes [1]. Both innate and adaptive immune dysregulation contribute to its pathogenesis, leading to complement activation, excessive cytokine release, and subsequent inflammation and organ damage, particularly lupus nephritis (LN) [1, 2]. The diverse clinical manifestations and variable therapeutic responses observed in SLE reflect its biological heterogeneity, and our current understanding of its underlying pathogenic mechanisms remains incomplete. Although classification criteria were originally developed for research purposes rather than for definitive diagnosis, they are still widely used in clinical practice due to the absence of validated diagnostic biomarkers for SLE [3, 4].

Extracellular vesicles (EVs) are nanoscale, lipid bilayer-enclosed structures released by cells, functioning as critical mediators of intercellular and interorgan communication through the transfer of diverse bioactive cargos, including proteins, lipids, nucleic acids, and metabolites [5–7]. Their secretion is triggered by various physiological and pathological stimuli, such as complement activation and inflammation, and they have emerged as key players in the pathogenesis of autoimmune diseases [8–12]. In SLE, EVs derived from mesenchymal stem cells (MSCs), T cells or plasma B cells have been shown to regulate inflammatory responses, including the modulation of IFN-γ and TGF-β signaling pathways, thereby contributing to the pathogenesis of SLE [13–17]. Importantly, the ability of EVs to reflect dynamic changes in disease progression makes them promising candidates for disease monitoring, diagnosis, and prognostic evaluation across a wide range of conditions, including SLE [18–20]. Both our studies and those of others have demonstrated that EV-derived microRNAs (miRNAs) and tRNA-derived small noncoding RNAs (tsRNAs) were dysregulated in SLE and LN patients. Notably, the expression levels of these small RNAs are closely associated with clinical parameters, highlighting their potential utility as diagnostic biomarkers for SLE, although the underlying regulatory mechanisms remain to be fully elucidated [21–25].

Although inflammatory responses and associated factors have been implicated in the pathogenesis of SLE and are known to contribute to most of the disease’s organ manifestations, the specific inflammatory protein profile derived from EVs in SLE remains to be fully elucidated [26]. Additionally, a major challenge lies in the limited sensitivity and specificity of conventional methods for detecting EV-derived inflammatory proteins, which has hindered their discovery and constrained their application as reliable biomarkers for SLE. The proximity extension assay (PEA) offers a powerful solution by employing pairs of oligonucleotide-labeled antibodies that bind to proximate epitopes on target proteins. Upon binding, the oligonucleotides hybridize to form a unique DNA sequence, which is then amplified and quantified via real-time PCR [27]. This technique enables the simultaneous detection of low-abundance proteins with high specificity, minimal cross-reactivity, and reduced inter-assay variability, making it particularly well-suited for profiling protein cargo in EVs derived from complex biological fluids [28–31].

The aims of the present cross-sectional study were to investigate the composition of circulating EVs in SLE, with a focus on identifying potential differences in EV-derived inflammatory proteins that may reflect distinct underlying pathogenic mechanisms.

Method

Patients enrollment and study approval

Between 2023 and 2024, patients with SLE, meeting the diagnostic criteria of the American College of Rheumatology (ACR) [4], were enrolled in the study together with age- and sex-matched healthy controls with no previous history of autoimmune diseases. The SLE Disease Activity Index 2000 (SLEDAI-2 K) was assessed and documented for all patients at the time of blood collection. Disease activity was classified as low (SLEDAI-2 K ≤ 6), moderate (7 ≤ SLEDAI-2 K ≤ 12), or high (SLEDAI-2 K >12) [32]. LN was diagnosed and classified based on the classification criteria of the International Society of Nephrology/Renal Pathology Society (ISN/RPS) and evaluated according to the 2018 revision of the ISN/RPS classification for LN [33, 34]. The clinical characteristics of all participants are summarized in Table 1. This study was approved by the ethics committee of our hospital, and all participant-related procedures were conducted in compliance with the Institutional Review Board (IRB) protocols of Drum Tower Hospital, affiliated with Nanjing University Medical School (NO. 2023-568−01). All participants provided written informed consent prior to inclusion in the study.

Serum samples preparation and the isolation of EVs

Peripheral blood samples from SLE patients and healthy controls were collected using the same standard protocols. Specifically, approximately 3 mL of venous blood was drawn into vacuum tubes containing separation gel and centrifuged at 3,000 × g for 15 min at room temperature. The resulting supernatant, designated as serum, was carefully aspirated and stored at −80 °C until further analysis, ensuring that no additional freeze–thaw cycles occurred prior to extracellular vesicle isolation. Sequential centrifugation followed by ultracentrifugation (UC) was used to isolate EVs as previously described with minor modifications [28, 35]. All serum samples were thawed on ice and spun at 3,000 × rpm for 15 min at 4℃ to remove residual cell contamination and debris. Supernatant was diluted 1:3 with EV-free PBS (ice-cold) followed by centrifugation at 12,000 × g for 1 h at 4℃ to remove large particles. The supernatant was filtered through a 0.22 μm filter and ultracentrifuged at 120,000 × g for 70 min at 4℃. The EVs pellet was resuspended and washed with PBS followed by a repeated ultracentrifugation. Finally, the EVs were re-suspended in 100 µL PBS and subjected to subsequent analyses without undergoing a freeze-thaw cycle. For size-exclusion chromatography (SEC) following previously published protocol[36]. Briefly, 1 mL supernatant obtained after centrifugation at 12,000 × g for 1 h was loaded onto a Sepharose CL-6B column (AG0044, BESTCHROM). For fractionation, 1 mL PBS was added for each fraction and the eluate of the first 1 mL elution was counted as fraction 1. Fractions 6–7, enriched in EVs were pooled and concentrated using ultrafiltration (UFC5010, Millipore) for subsequent ELISA analysis.

EVs characterization and analyses

1. Nanoparticle tracking analysis (NTA).

   EVs were diluted in EV-free PBS (1:10,000) and gently mixed before being injected into the sample cell. The autofocus sensitivity and detection channel were set to 80 and EV mode, respectively. The correct field of view was selected to generate EVs detection reports. Particle concentration was calculated by the instrument based on the dilution factor input utilizing ZetaView (version 8.05.14 SP7).

2. EVs protein quantification and western blot analysis (WB).

   The resuspended EVs were lysed and protein was isolated by adding equal volume of RIPA lysis buffer supplemented with 100 mM PMSF and PI according to the manufacturer’s instructions. The Pierce BCA protein assay (ThermoScientific) was used for protein quantification of EVs. About 40 µg of EV protein was loaded in 10% resolving polyacrylamide gel and the transfer was performed at 300 mA for 70 min using PVDF immobilon membrane. Membranes were blocked by 5% skimmed milk or 2.5% BSA followed by incubation with multiple primary antibodies overnight at 4℃. Primary antibodies against CD9 (sc-13118), CD63 (sc-5275, Santa Cruz), CD81 (A4863, Abclonal) and TSG101 (sc-7964, Santa Cruz), APOE (18254-1-AP, Proteintech) and GM130 (A24834, Abclonal) were diluted at 1:1,000 or 1:2,000 in 5% skimmed milk or BSA (TBST). After washing with PBST for 1 h at room temperature, membranes were incubated for 1 h at room temperature with secondary antibodies according to the primary antibodies followed by another 0.5 h TBST washing. Two secondary antibodies were used namely HRP conjugated mouse anti-rabbit (sc-2357, Santa Cruz) or goat anti-mouse secondary antibodies (ZB2305, ZSGB-BIO). Blots were developed by the SuperSignal West Pico PLUS ECL substrate (34580, Thermo Scientific) and recorded on a Tanon 5200 chemiluminescence image analysis system.

3. Transmission electron microscopy (TEM).

   Approximately 20 µL of EVs fully resuspended in PBS were adsorbed onto the copper grid with carbon film for 3–5 min. Excess liquid was removed with filter paper followed by 2% phosphotungstic acid staining for 2 min. The images were finally acquired using a 120 kV transmission electron microscopy (HT7800 RuliTEM, hitachi).

Proteomic analysis via PEA

Proteomics of EVs samples (non-renal SLE: n = 35; LN: n = 35; healthy controls: n = 33) were measured using the Olink^® Target 96 inflammation panel (95302) (Olink Proteomics AB, Uppsala, Sweden), which enabled 92 inflammation-related biomarkers to be analyzed simultaneously according to the manufacturer’s instructions and previous reports [28, 37]. The detailed list of proteins can be found on the Olink website (https://olink.com/products-services/target/inflammation/). PEA-based protein analysis utilized a pair of unique oligonucleotide labelled antibodies which recognized and bound to the proximate location of the protein. Closely proximate oligonucleotides will undergo complementary hybridization, amplification, detection and quantified by microfluidic real-time PCR (Signature Q100). One microliter of each EVs sample was loaded, accompanied by four internal controls for quality control. The Ct value was transformed to normalized protein expression (NPX) values expressed on a log2-scale, with a high value representing higher protein expression. The limit of detection (LOD) was not used as the screening criteria. Samples which passed the quality control were all included in the subsequent biological difference testing across groups according to the company recommendation and a previous study [28].

Quantitative analysis of extracellular vesicles using ELISA

Three representative EV-associated inflammatory proteins were selected for ELISA validation in independent cohorts (SLE: n = 54; healthy controls: n = 58). Commercial ELISA kits were used to measure EV-derived IL-18R1 (Thermo Fisher Scientific, EH263RB), CCL23 (Abcam, ab100611), and IFN-γ (Thermo Fisher Scientific, KHC4021) according to the manufacturers’ protocols. For total EV protein detection, EVs were mixed with an equal volume of RIPA lysis buffer and incubated on ice for 30 min. The resulting EV lysate was then diluted to the appropriate loading volume. For EV surface protein detection, EVs were mixed with PBS without lysis, followed by the addition of specific antibodies and subsequent washing steps in accordance with the manufacturers’ instructions. Absorbance at 450 nm was measured using a SpectraMax iD3 microplate reader (Molecular Devices, USA). Undetectable values were assigned an arbitrary value equal to one-fifth of the sensitivity limit specified in the protocol.

Bioinformatics analysis

1. Ingenuity Pathway Analysis.

   Canonical pathway analysis was performed using QIAGEN Ingenuity Pathway Analysis (https://www.qiagenbioinformatics.com/products/ingenuity-pathway-analysis/) based on the differentially expressed inflammatory EV proteins. The right-tailed Fisher’s exact test was employed to calculate the P-value, and to determine the probability that the association between the genes in the dataset and z-score was calculated, to indicate the likelihood of activation or inhibition of that pathway.

2. Random Forest classification.

   Machine learning analyses were performed using the R caret package and “ranger” method (version 6.0.94), with mtry parameters found using a grid search approach between every combination of min.node.size values of 5 and 15. mtry was between 0.5, 0.7 and 1.2 times the square root of the number of variables. Parameters giving the best accuracy were kept, as calculated by a 5-fold cross-validation repeated 5 times. The “PROC” package were used for random Forest ROC analyses (version 1.18.5).

3. Clinical parameters correlation analysis.

   Correlation analysis was performed using the rcorr method from the Hmisc R package (version 5.1.1). The statistical test employed was “Spearman’s rank” and the correlation graph was plotted using the corrplot R package (version 0.92).

4. GEO database analysis.

   Gene expression data were obtained from the GEO database (accession number: GSE32591, GSE97263, GSE97264, GSE226147 and GSE235660). The datasets include samples from non-renal SLE, LN patients and healthy controls. Log2-scaled microarray gene expression data and log2 TPM (Transcripts Per Million) values for the RNASeq were used as measures of gene expression levels, respectively. Gene expression difference between groups was evaluated using the Wilcoxon rank-sum test. Statistical significance was defined as a P-value < 0.05. Visualizations were created using the “ggpubr” package (version 0.6.0) in R (version 4.4.1).

Statistical analysis

Normalized Protein Expression (NPX), an arbitrary unit on the log2 scale, was employed to assess the abundance of present proteins, and the differentially expressed NPX were obtained by using the Wilcoxon method with a P-value cutoff of 0.05. PCA and UMAP dimension reduction analyses were performed based on Olink-detected protein expression using the R packages FactoMineR and umap respectively. Experimental data are given as the mean ± SEM in histograms or the median and interquartile range (IQR) in boxplots. Normality tests of the data were examined prior to performing the t-tests or ANOVA analyses. Log normal distribution was used when the outcomes did not satisfy the normality assumption. Student’s unpaired t-test for 2-group comparison and one-way ANOVA for the comparisons of over two groups was used when the normality assumption was met. The differences across each group were assessed using ANOVA followed by post hoc comparisons using Bonferroni’s multiple comparisons test. When equal SDs assumption and nearly equal sample size were not satisfied, Welch’s t-test or Games-Howell’s test were used for 2-group comparison or over two groups comparisons, respectively. We used GraphPad Prism 8.0 (GraphPad Software) for statistical analyze. A P-value < 0.05 was considered significant.

Results

Characterization of the SLE serum-derived EVs pattern

Serum EVs were obtained from 35 non-renal SLE patients, 35 LN patients and 33 healthy controls with matched sex and age to identify SLE and LN-related EVs molecular alterations based on the Olink DNA-coupled multiplex PEA targeted protein detection platform (Table 1). Serum EVs isolated from all SLE patients as well as HCs were compared for basic properties such as particle number and shape. NTA revealed an increased serum EV particle concentrations in non-renal SLE patients compared to HCs, which was similar with previous findings (Fig. 1A) [38, 39]. Nonetheless, no significant difference was observed between the LN and the HCs, or between LN and non-renal SLE group. When SLE groups were further separated, we found that only EVs from the female non-renal SLE group showed an increase in the EV particle concentration vs. HCs while there was no significant change across the rest of the groups (Figure S1A). EVs protein content was also measured and the EVs particle/protein ratio was calculated as shown in Fig. 1B and Figure S1B, which present the same variation trend with EVs particle concentration, though only the non-renal SLE group revealed a statistically significant elevation (Fig. 1B). Size distribution was characterized by NTA which showed the particle number elevation from SLE group mainly within the 101–150 nm size range (Fig. 1D and Figure S1C). Intriguingly, we observed that the EV particle number was significantly elevated in female compared with male SLE patients, regardless of renal involvement (Figure S1C). TEM analysis verified the particle shape and approximately 100 nm size of SLE EVs (Fig. 1D). The relatively larger particle size identified in NTA may have resulted from the presence of EVs aggregation, which could be recognized as larger-size particles (Figure S1D). Western blot verified the presence of commonly used markers including CD9, CD63, TSG101 and CD81 (Fig. 1E). Negative markers for EVs including APOE and GM130 were also tested. A weak ApoE signal was detected, which suggested that the EV isolation procedure introduced minimal residual lipoproteins [28, 40], while another negative marker, GM130 was absent in EVs.

Fig. 1.

Characterization of serum EVs obtained from patients with systemic lupus erythematosus (SLE) and healthy controls (HC) A Nanoparticle tracking analysis (NTA) showing EVs particle numbers in the SLE (LN; n = 35, non-renal SLE; n = 35) and HC (n = 33) groups. Data are presented as the number of EVs from 1 mL of serum. A significant difference was determined by Games-Howell's multiple comparisons test. * = P < 0.05; ** = P < 0.01. B Particle number of EVs represented by the number to protein ratio. A significant difference was determined by Games-Howell's multiple comparisons test. ** = P < 0.01. C Particle size distribution of SLE EVs by nanoparticle tracking analysis in SLE (LN; n = 35, non-renal SLE; n = 35) and HC (n = 33) groups. D Representative images captured by transmission electron microscopy (TEM) showing the shape and size of EVs from the SLE and HC groups. E. Western blot analysis of positive and negative markers of EVs from non-renal SLE, LN and HC groups. n = 2 for each group

PEA based characterization of SLE EV proteins for inflammatory biomarkers

We used the Olink platform based on PEA to detect and quantify the protein level of SLE EVs. A total of 92 inflammatory related proteins were evaluated and sample-to sample heterogeneity was first tested without molecule selection by two different dimensionality reduction methods (Figure S2). While PCA exhibited a certain separation across distinct groups (Figure S2A), well resolved SLE patients from healthy controls was achieved by UMAP (Figure S2B), superiority of which was also observed and suggested by published studies [41, 42].

Next, the expression pattern of SLE EV proteins was evaluated. As shown in Fig. 2A, a total of 49 EV proteins were differentially expressed at a statistical significance of P < 0.05 in SLE patients (SLE) compared with HCs, in which proinflammatory markers dominated the dysregulated SLE EV proteins. The Olink strategy found that CD40 and CD8A were two of the overexpressed proteins in SLE EVs. Interestingly, in a previous study, the same trend on CD40 and CD8 was also found by using MACS-Plex assays in T cell-derived EVs from SLE patients, suggested the possibility that at least a portion of the EV protein changes in SLE was contributed by T cell-derived EVs [15]. The expression level of all tested inflammatory proteins in EVs are presented in Figure S3 and listed in Table S1.

Fig. 2.

Proximity extension analysis (PEA) of serum EVs inflammatory proteins in SLE patients A Heatmap showing the normalized expression level of 49 EV protein enriched in the SLE group, with a significant difference in the SLE group, clustered using Euclidean distance. B Ingenuity Pathway Analysis canonical pathways analysis based on 49 altered EV proteins revealing the top 15 potentially activated pathways in the SLE group compared with HCs. The right y-axis and the blue broken line represent the number of genes count for each representative canonical pathway. C Volcano plot showing differentially expressed inflammatory EV in the SLE group vs. HCs. Circle symbols indicate P < 0.05 and adjust P > 0.05, whereas diamond symbols indicate adjust P < 0.05. Red and blue colored symbols represent upregulated and downregulated proteins, respectively. D The expression level of 16 selected differentially altered EVs proteins in SLE (n = 70) and HC (n = 33) presented as NPX values, which are expressed in a log2 scale. * = P < 0.05; ** = P < 0.01; *** = P < 0.001; **** = P < 0.0001. Any significant differences were calculated using the Mann Whitney test

Ingenuity Pathway Analysis was employed to identify potentially related pathways based on aberrant SLE EV proteins. A total of 15 top pathways, including pathogen-induced cytokine storm signaling pathway, IL-10 signaling and IL-17signaling, are presented in Fig. 2B. Association of these pathways activation with the disease course has been previously reported in human and animal models of autoimmune diseases including SLE [43]. Volcano plots were created to show in detail differentially expressed EVs proteins (Fig. 2C). The top 10 upregulated and 6 downregulated, with an FDR < 0.05, are listed. To gain insight into differences in EV proteins for each patient, the expression levels of all 16 dysregulated EVs proteins in SLE were used for comparation (Fig. 2D). Upregulated inflammatory proteins identified in SLE EVs were CSF-1, EN-RAGE (S100A12), IL-18, MCP-1 and CCL4 and have been measured at elevated levels in the serum and/or urine of SLE patients [44–47]. However, these dysregulated EV proteins in SLE were not significantly changed between LN and non-renal SLE (Figure S4A), although higher level of urinary TWEAK was reported in LN patients compared with non-renal SLE patients [48]. Moreover, two EV proteins IL-10RA and SCF were found dysregulated between LN and non-renal SLE (Figure S4B-C). The combined observations thus revealed the unique pattern of EV proteins in SLE.

Validation of inflammatory EV proteins in an independent cohort by ELISA

To validate the findings of PEA analysis, top-regulated EV-derived proteins revealed by PEA including CCL23, EN-RAGE, IL-18R1, and IFN-γ (P < 10–¹⁰, fold change > 2.5) were tested by ELISA. Since the signal of EN-RAGE were indistinguishable in the preliminary experiments (data not shown), CCL23, IL-18R1, and IFN-γ were selected for subsequent validation analyses. First, SLE EVs under RIPA lysis or untreated naive EVs were evaluated and compared with these proteins. Theoretically, because proteins located inside EVs could not be captured by antibody, untreated naive EVs will hinder its ELISA detection and will show a dominant signal under RIPA lysis. We observed that IFN-γ and CCL23 levels were not changed when RIPA lysis was performed on EVs before ELISA testing, while IL-18R1 exhibited a slight but significant elevation (fold change < 1.5). The observation suggested that these proteins were mainly located on the surface of EVs with a small amount of IL-18R1 inside the EVs (Figure S5). To obtain the overall EV protein expression levels, EV lysis was performed followed by ELISA. Similar to Olink results, sEV-derived IL-18R1, CCL23 and IFN-γ showed significant upregulation in SLE patients compared to HCs (Fig. 3A-B), with no significant difference between LN and non-renal SLE (Figure S6A). To evaluate further the diagnosis value of these EV proteins, receiver operating characteristic (ROC) curves were created. The AUC values for IL-18R1 and IFN-γ were 0.873 and 0.841, respectively, which were higher than CCL23 (AUC 0.716). A combination of these EV proteins showed better discrimination performance (AUC 0.917) between SLE patients and HCs (Fig. 3C). For independent validation, EVs were additionally purified using size-exclusion chromatography (SEC) in a randomly selected subset of samples. Nanoparticle tracking analysis showed that the particle concentration peaked in fraction 7 (Figure S7A). The identity and purity of the SEC-isolated EVs were verified by TEM and Western blot analysis (Figure S7B-C). ELISA analysis showed a consistent elevation of IL-18R1 in SLE sEVs isolated by both UC and SEC (Figure S7D), and protein levels displayed a strong correlation between the two isolation methods, indicating high reproducibility of EV-derived protein across EV purification strategies (Figure S7E). Collectively, unique SLE EV proteins were confirmed by ELISA, which exhibited superior diagnostic performance in SLE patients, and also revealed the robust detection of EV proteins by the Olink method.

Fig. 3.

Validation of IL-18R1, CCL23 and IFN-γ levels in EVs using ELISA in an independent cohort A Heatmap showing the expression level of IL-18R1, CCL23 and IFN-γ detected by ELISA in SLE patients (n = 38) and HC individuals (n = 32). B The expression levels of EV-derived IL-18R1, CCL23 and IFN-γ in SLE patients represented by the protein to EV total protein ratio. Any significant difference was determined using an unpaired t-test. ** = P < 0.01; **** = P < 0.0001. C. ROC curve showing the discriminatory abilities of EV-derived IL-18R1, CCL23 and IFN-γ in distinguishing SLE. AUC = area under the curve; 95% CI = 95% confidence interval

Analysis of inflammatory EV proteins for SLE and LN discrimination

To explore further the potential inflammatory EV proteins that could serve as biomarkers for SLE diagnosis and classification, the RF algorithm was employed with five-fold cross-validation. Diagnostic models were generated to discriminate SLE from HC individuals as well as LN from non-renal SLE patients (Fig. 4). The top 20 EV protein features in SLE vs. HC groups and LN vs. non-renal SLE groups, respectively, were listed based on their rank of overall importance in the model (Fig. 4A). The correlation between the top 20 EV proteins corresponding to each group was analyzed to show the potential protein-protein relationship in EVs (Fig. 4B). In SLE patients, EV-derived CXCL11 and CSF-1 were positively associated with CXCL9, IL-18 and MCP-1, while they also had a positive correlation with MMP-10 and SCF in LN patients. In addition, a strong association was found among FGF-21, TGF-α, SIRT2 and IL-15RA in LN vs. non-renal SLE groups. The model was further optimized by modulating the number of features and evaluating performance with regard to accuracy, followed by the selection of the classifier with the best performance (Fig. 4C). After feature selection, the confusion matrix illustrated that the model yielded 90% specificity, 100% sensitivity, 100% positive predicted value (PPV) and 91.7% negative predicted value (NPV) in discriminating SLE from HC individuals (Fig. 4D). The ROC curve achieved a considerable AUC of 0.999 (Fig. 4E). Similar diagnosis performance was achieved when the model was trained for the separation of HC individuals with non-renal SLE or LN, respectively (Figure S8). Furthermore, another RF model was also trained for the diagnosis of LN from non-renal SLE patients that generated 74.3% specificity, 68.6% sensitivity, 72.7% PPV and 70.3% NPV, with a 0.793 AUC of the ROC (Fig. 4D, E). Taken together, the RF diagnostic model showed that serum EVs-derived inflammatory proteins could be potentially valuable as a liquid biopsy test for SLE and LN detection and classification.

Fig. 4.

Identification and evaluation of SLE and LN signatures in serum EVs A Top 20 features selected by RF analysis ranked by overall importance and in the discrimination for SLE vs. HC (left panel) and LN vs. non-renal SLE (right panel). B Heatmap showing the correlation of the top 20 EV proteins selected by RF in SLE vs. HCs (left panel) and LN vs. non-renal SLE (right panel). Spearman’s correlation analysis was utilized and proteins were clustered according to Euclidean distance. C Line chart represented the accuracy change in the model training process of the top 20 features in SLE vs. HC (top panel) and LN vs. non-renal SLE (bottom panel). D-E ROC using the RF classifier to identify SLE from HCs (top panel) and LN from non-renal SLE patients (bottom panel)

Correlation of Olink-derived inflammatory EV proteins with clinical parameters and SLE progression

To examine the correlation of Olink-derived inflammatory EV proteins with clinical parameters, we conducted Spearman’s correlation analysis on significantly changed inflammatory EVs proteins by sex, SLEDAI and clinical parameters related to SLE (Fig. 5 A). SLE EV-derived FGF-19, SCF and TWEAK exhibited a significant correlation with the SLEDAI score, in which higher levels of FGF-19 and SCF showed positive correlation while subjects with higher TWEAK levels had lower SLEDAI scores (Fig. 5A-B). Given the fact that high disease activity status (SLE-H) often represents severe disease outcomes, such as tissue damage and mortality as well as distinct management of SLΕ drugs, treatment strategy, we further evaluated the expression pattern of inflammatory EV proteins between SLE-H (SLEDAI > 12), and SLE-low and SLE-moderate (SLEDAI ≤ 12). Volcano plot were created to show the significantly different EV proteins in SLE patients with high disease activity (Fig. 5 C). A total of 14 proteins were found to be elevated in the SLE-H group (Figure S9). The expression level of FGF-19 was not only positively correlated with the SLEDAI score but also was closely linked with high activity SLE, revealing it could uniformity reflect disease activity status. In addition, three proteins in the ELISA validation were evaluated. The elevated expression of EV-derived IL-18R1 and CCL23 in the SLE-H group, was consistently observed in both PEA screen cohort and the ELISA validation cohort, whereas IFN-γ showing no difference (Figure S6B and Figure S9). Additionally, EV-derived HGF, IL-8, TNSF14, uPA and TWEAK showed positive correlations with both complement components 3 and 4 (C3, C4), while IL-10RA and CCL19 had negative correlations. Relatively low expression levels of EV TWEAK in SLE patients linked with a low C3/C4 level and SLE SLEDAI scores, indicated the potential involvement of specific EV proteins with component system activation in SLE (Fig. 5A-B). Collectively, the Olink data suggested that EV-derived inflammatory proteins linked with key clinical parameters of SLE could serve as potential indicators of the disease severity, a finding that was also partly validated by ELISA.

Fig. 5.

Correlation of EV inflammatory proteins from SLE patients with clinical parameters A Heatmap showing the correlation between Olink-derived expression levels of EV proteins with sex, SLEDAI, complement components 3 (C3, mg/dL), complement components 4 (C4, mg/dL), anti-dsDNA antibody (anti-dsDNA), anti-Smith antibody (anti-Smith), estimated glomerular filtration rate (eGFR, mL/(min × 1.73 m²)) and urinary albumin to creatinine ratio (UACR, mg/mg). The red ellipses indicate a positive correlation while the blue ellipses indicate a negative correlation. The color intensity represents the correlation strength. Significant correlation was determined by Spearman’s correlation analysis. * = P < 0.05; ** = P < 0.01; *** = P < 0.001. B Scatter diagram showing the correlation of SLEDAI with EV-derived FGF-19 (left panel), SCF (middle panel) and TWEAK (right panel). A positive correlation is presented as pink while a negative correlation is shown in blue. Coefficient values and P-values were determined by Spearman’s correlation analysis. C Volcano plot showing differentially expressed inflammatory EV proteins in SLE-H (SLEDAI >12), and SLE-low and SLE-moderate (SLEDAI ≤ 12). Pink colored symbols represent upregulated proteins. D-E Scatter diagram showing the correlation of EV-derived SCF, IL-18R1, CCL23, IL-7 and CSF-1 with eGFR and UACR. Coefficient values and P-values were determined by Spearman’s correlation analysis

Correlation of Olink-derived inflammatory EV proteins with renal damage and LN involvement

Next, the role of inflammatory EV proteins was evaluated in SLE for renal involvement. The urinary albumin to creatinine ratio (UACR) and estimated glomerular filtration rate (eGFR) were utilized to represent kidney function impairment in SLE patients, whereas a low level of eGFR and high value of UACR indicated more severe renal dysfunction. A total of 14 EV proteins were identified correlating with eGFR or UACR, among which 13 EV proteins showed positive correlation with renal damage, while only 4E-BP1 showed a negative correlation (Fig. 5A). Five EV proteins SCF, CXCL11, 4E-BP1, IL-10RA and CSF-1 were also identified as key by the aforementioned RF model in discriminating LN from non-renal SLE patients (Fig. 4A, Figure S10). Patients with a higher level of SCF appeared to have low eGFR as well as high UACR, indicating heavier kidney impairment (Fig. 5D), which could partly explain its ability in LN and non-renal SLE dipartition. Conversely, high levels of EV IL-10RA were found in non-renal SLE compared with LN patients and associated with a low level of eGFR (Fig. 5E, Figure S4B-C). CCL23 and IL-7, two EV proteins that are linked with a high disease activity status of SLE (Fig. 5C), also exhibited a positive correlation with UACR. EV CSF-1 had a strong positive correlation with UACR. Interestingly, serum-derived CSF-1 was validated as correlating with clinical kidney disease activity and reflecting kidney histopathology, which suggested the possibility that a component of the previously found CSF-1 elevation in serum contributed to EV CSF-1 [44]. A combined evaluation confirmed the close association and underlying contribution of Olink-derived inflammatory EV in renal damage monitoring and LN.

Discovery of the cellular origins of inflammatory EV proteins in SLE

Next, we investigated the specific cell types that may be responsible for releasing inflammatory EV proteins that upregulates SLE as well as LN. Transcriptome analysis of diverse cell types was employed including CD4⁺T cells, CD8⁺T cells, naive B cells, antibody-secreting cell populations (ASC), platelets from SLE patients and HCs obtained from datasets GSE97263, GSE97264, GSE226147, GSE235660 and GSE32591 (Figure S11-12). The identified Inflammatory EV proteins from SLE were found to be expressed and dysregulated in different cell types. IFN-γ, CSF-1 and PD-L1 which were elevated in SLE EVs, were consistently elevated in CD8⁺ T cells. In antibody-secreting B cells, IFN-γ, CSF-1 and MCP-1 were upregulated in SLE patients (Figure S11A-B). Platelet-derived IL-18R1 was highly expressed in SLE while MCP-1 and PD-L1 were elevated in both platelets and CD4⁺ T cells obtained from SLE patients (Figure S11C-D). Additionally, LN progression-related EV proteins were further investigated to determine their tissue origins. EV-derived CSF-1 was significantly elevated in kidney tubulointerstitium and slightly decreased in the glomeruli (FC < 1.1) of LN patients (Figure S12). EV-derived IL-10RA was upregulated in both kidney glomeruli and tubulointerstitium, which was opposite to the observation that it was highly expressed in EV from LN patients (Figure S4C). Practically no change in EV-derived SCF in kidney glomeruli and tubulointerstitium was found though it was negatively correlated with LN progression (Fig. 5D, S4C). The combined results indicated the specific blood cell that contributed to the SLE inflammatory EV proteins and a more complicated tissue origin of LN EV proteins.

Discussion

SLE is a complex autoimmune disease characterized by multisystem involvement. The heterogeneous nature of its symptoms, coupled with a limited understanding of its pathogenesis and pathology, poses significant challenges for early diagnosis and effective monitoring of disease progression [49, 50]. The abnormally stimulated immune response is a hallmark of SLE, leading to cytokine upregulation, complement activation, nuclear components and immune complexes production [51, 52]. EVs, encapsulating diverse cargoes, including cytokines and chemokines, play key roles in many inflammation-associated pathological processes [18, 53, 54]. Serum EVs and their inflammation cargo in SLE have not been fully recognized, partly due to the dearth of highly acuate detection and quantification methods. Therefore, in the present study, Olink, a novel approach based on PEA was utilized, which enabled protein-oligonucleotide signal conversion and subsequent protein quantification by RT-PCR. Ninety-two inflammatory proteins from serum-derived EVs of SLE patients were analyzed, revealing distinct inflammatory EV patterns specific to SLE and LN compared to HCs, as well as differences between SLE patients with and without renal involvement. To the best of our knowledge, this is the first study to apply the PEA for characterizing the inflammatory proteomics of SLE-derived EVs and in assessing their diagnostic value in distinguishing SLE and LN.

Extracellular vesicles along with their cargoes have been demonstrated to be involved in multiple pathogenesis of many autoimmune diseases including SLE [8, 55]. Several studies have indicated the dynamic change and biological roles of EVs released by different cell types in SLE animal models or patients. A study of isolated serum EVs from 19 SLE patients, using the precipitation method, revealed the enrichment of content in the SLE group [38]. Our study characterized the amount of SLE serum EVs in an enlarged sample size and observed a similar elevation of EVs in both non-renal SLE and SLE patients, while applying ultracentrifugation, a more robust and widely used extraction method. Interestingly, we observed a sex-related difference in EV numbers, where the particle number was significantly higher in female compared with male SLE patients. Considering that approximately 90% of SLE patients are women, this finding increases the possibility that sex hormones, especially estrogen, may influence EV biogenesis and release, contributing to sex-specific immune responses in SLE. Recent studies have shown that estrogen can actively regulate EV secretion and cargo composition, supporting this hypothesis[57, 58].This observation highlights the potential role of EVs in mediating sex-dependent immune dysregulation and suggests potential as sex-specific disease indicator.

MSCs-derived EVs were shown to inhibit M1 macrophage polarization and proliferation of cytotoxic T cells and elicited their effect in disease remission [58–60]. T cell derived EVs could facilitate chronic immune activation and excessive cytokine production. EVs released by T cells from SLE patients can carry overexpressed eosinophil cationic protein (ECP) and bactericidal/permeability-promoting protein (BPI), leading to multi-tissue inflammation. In the present study, not only ECP but also many other immune markers were identified with elevating trends including CD40 and CD8A [15, 39]. Interestingly, these two proteins were consistently overexpressed in the Olink strategy-derived EVs profile. These results confirmed the uniformity characteristics of the basic properties of SLE EVs and also indirectly suggested the validity and the reliability of the Olink strategy in EV proteins detection and discovery.

We identified that SLE EVs enriched in inflammatory proteins including CCL23, CSF-1, IL-18R1, IFN-γ, EN-RAGE, IL-18, CCL3, MCP-1, PD-L1 and CCL4 as the top upregulated EV proteins. Previous studies that utilized the Olink platform for SLE serum protein detection revealed circulating inflammatory proteins such as SIRT2, IL-18 and CASP8 that were significantly upregulated [61]. Using the same Olink panel, a phase II randomized, placebo-controlled study identified serum PD-L1, IL-6, CCL19, IL-12β etc. as being reduced after baricitinib treatment, thus indicating their relatively high expression level in baseline SLE patients [62]. In our study which focused on EV-derived inflammatory proteins, some but not all proteins were found to be dysregulated, exhibiting diverse variation trends. These findings, along with the results, suggest that while inflammatory proteins are dysregulated in the serum of SLE patients, serum EVs displayed a distinct expression pattern compared to common circulating cytokines. Unlike soluble cytokines that act transiently through membrane receptors, EV-derived cytokines are protected by lipid membranes, which prolong their stability and circulation time. EVs can carry multiple cytokines and target specific tissues depending on their cellular origin and activation state, suggesting a unique, sustained role in immune regulation underlying SLE [63].

PD-L1, a protein binding with PD-1 on T cells and mediating immune checkpoint blocking, was reported to be present on the surface of EVs released from tumor cells where it suppressed CD8⁺ T cell function [64]. Our data suggested that elevation of PD-L1 may act in an EV-mediated mechanism to participate in the pathological B and T cell interaction, leading to substantial immune-tolerance disturbance in SLE [65, 66]. Although the typical and dominant type I IFN signature has been widely observed in SLE, emerging evidence suggests that dysregulated IFN-γ (type II IFN) may occur in the early as well as active stages of SLE and could serve as an ideal SLE therapy targe [67–69]. IL-18R1 is involved with IL-18-mediated IFN-γ synthesis and increased IL-18R showed a positive correlation with SLE activity, suggesting its role as an upstream regulator of the IFN-signaling pathway in SLE [70, 71]. Intriguingly, with regard to serum EVs in SLE, we found a marked elevation of both EV-derived IL-18R1 and IFN-γ. In addition, EV-derived IL-18R1 exhibited a higher expression level in SLE patients with severe disease status. Therefore, this evidence indicates the possibility that activation of the IL-18 pathway contributes to the type II IFN signature, generating its proinflammatory pathogenic role and its early diagnostic value for SLE.

A recent study found a high expression level of serum CCL23 and CSF-1 in SLE patients with a heavier coronary atherosclerosis burden [72]. Expression and correlation analysis revealed that SLE EV-derived CCL23 was elevated in SLE with a higher SLEDA and UACR. Of note, CSF-1 in serum and urine was upregulated in SLE patients with LN and linked with clinical LN activity and its severity [44]. A strong correlation of EV-derived CSF-1 with UACR was also found in the Olink data of SLE patients. EV-derived IL-10RA and SCF exerted converse changes in LN patients when IL-10RA was downregulated. The IL-10 signal pathway is anti-inflammatory and enrichment or deficiency of IL-10 receptor including IL-10RA elicited T cell suppression or ab auto-inflammatory effect, respectively [73, 74]. We speculate that EV-derived IL-10RA also acts as a protection role in SLE patients suppression of which contributes to renal inflammation and LN.

Accumulated evidence suggests EVs have considerable diagnostic potential for SLE and LN. Studies from our group and others have revealed a unique expression pattern of EV-derived miRNAs [75] and tsRNAs [76], while little is known about the diagnostic value of SLE EV-derived inflammatory proteins. We performed a two-step validation and diagnostic efficacy evaluation including independent ELISA confirmation of representative EV proteins followed by machine-learning based EV biomarkers screening. Three proteins of interest namely IL-18R1, IFN-γ and CCL23 were chosen as the representative protein set due to their relatively high fold change as well as their underlying role in SLE monitoring. ELISA also revealed the elevation of three proteins in SLE and higher levels in SLE with high disease activity. Therefore, on the one hand, by utilizing a more general protein detection method, we confirmed the robust performance of the Olink strategy, and on the other hand, three proteins combination have already provided significant diagnostic efficacy for SLE in terms of an AUC of 0.917. Machine-learning based feature selection approaches have been applied to multiple fields represented by biomarker searching and disease discrimination with improved accuracy and efficiency [20]. In current study, we employed the RF algorithm that performed well in noise and overfitting for SLE identification and LN discrimination. The prediction model generated by RF demonstrated impressive diagnostic efficacy for SLE patients from HC, with PPV and NPV of 95.9% and 100%, respectively (ROC AUC 0.999). In the sense of discriminating LN from non-renal SLE, the model provided moderate separation of PPV and NPV at 72.7% and 70.3%, respectively. Consequently, nonlinear analysis was used to construct prediction models comprising more EV protein signatures and achieved reliable diagnostic performance.

Several limitations of the present study should be noted. First, both the sample size and the spectrum of included autoimmune diseases should be expanded in future studies to better confirm the existence of these EV-derived inflammatory proteins in other autoimmune or inflammatory conditions. In addition, a longitudinal cohort with medication treatments would be beneficial for assessing dynamic changes in EV-derived inflammatory proteins over the course of SLE. Second, a preselected Olink Target 96 Inflammation panel was used, which may have excluded other relevant inflammatory proteins not covered by the current detection list. Therefore, a higher-throughput Olink panel is needed to provide a more comprehensive profile of EV-derived inflammatory proteins. Also, due to the antibody-based strategy was applied by both Olink and ELISA which may involve potential antibody cross-reactivity, future studies incorporating peptide recognition based-mass spectrometry will be valuable to provide another validation. Finally, further evidence is required to validate the specific cellular origins of the identified EV-derived proteins. Additionally, detailed exploration using animal models will be necessary to investigate the potential role of these EV-derived proteins in the inflammation-associated pathological processes driving SLE.

Conclusions

This study is the first to identify a unique pattern of EV-derived inflammatory proteins in SLE using the PEA, highlighting its promising diagnostic value for SLE detection and LN discrimination.

Authors’ contributions

Shoubin Zhan: Investigation, Methodology, Writing - Original Draft, Funding acquisition. Zhongyu Wang: Investigation, Visualization, Resources. Ye Xu: Methodology, Formal analysis, Validation. Shengkai Zhou: Investigation. Minghui Ge: Methodology. Yunjie Song: Visualization. Yi Zhu: Investigation. Huan Dou: Supervision, Project administration. Han Shen: Supervision, Conceptualization. Ping Yang: Conceptualization, Writing - Review & Editing, Funding acquisition.

Funding

This study was supported by the National Natural Science Foundation of China (Grant No. 82202600), the Nanjing Drum Tower Hospital Clinical Research Special Fund Project (No. 2024-LCYJ-MS-11 and NO. 2023-JCYJ-QP-25) and Youth Science Fund of the Chen Xiaoping Science and Technology Development Foundation of Hubei Province (CXPJTH124001-2414).

Data availability

Data from the study will be available upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the ethics committee of our hospital, and all participant-related procedures were conducted in compliance with the Institutional Review Board (IRB) protocols of Drum Tower Hospital, affiliated with Nanjing University Medical School (NO. 2023-568-01).

All participants provided written informed consent prior to inclusion in the study.

Conflict of interest

The authors declare no conflicts of interest.

Competing interests

The authors declare no competing interests.