Patients with SLE were enrolled according to the 2019 EULAR/ACR classification criteria, ensuring diagnostic consistency across the study cohort (24). LN was defined as renal involvement in SLE patients, either by a 24-hour urinary protein excretion ≥500 mg or according to the International Society of Nephrology/Renal Pathology Society (ISN/RPS) classification criteria, with histopathological evaluation based on the 2018 revised ISN/RPS classification for LN (25, 26). HCs were age- and sex-matched volunteers with no history of autoimmune or chronic disease, normal laboratory tests, and no structural abnormalities on imaging (chest CT, abdominal and urinary system ultrasound). Individuals with prior severe infection, malignancy, metabolic or autoimmune disorders, or other conditions affecting renal function were excluded. The detailed flowchart of participant enrollment and study design is illustrated in Figure 1.

Between March 2022 and March 2024, 51 SLE patients were enrolled from the Department of Rheumatology and Immunology, Nanjing Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University. Patients were stratified into SLE with LN (n = 29) and SLE without LN (n = 22) subgroups according to renal involvement. Twenty age- and sex-matched HCs were recruited from the hospital’s Physical Examination Center. Demographic characteristics, clinical data, and treatment histories were systematically recorded. After an overnight fast (≥8 h), midstream morning urine samples (~8 mL) were collected under sterile conditions. The samples were centrifuged at 2,000 × g for 30 min at room temperature to remove cellular debris, and the resulting supernatants were aliquoted into 2-mL tubes and stored at −80 °C until further analysis.

The uEVs were isolated using a commercial precipitation-based kit (Total Exosome Isolation Reagent; Thermo Fisher Scientific, USA) according to the manufacturer’s instructions. Briefly, thawed urine samples were centrifuged at 2,000 × g for 30 min at 4 °C to remove residual debris. Subsequently, 800 µL of the clarified supernatant was mixed with an equal volume of the isolation reagent, vortexed thoroughly, and incubated at room temperature for 1 h. The mixture was then centrifuged at 10,000 × g for 1 h at 4 °C, and the resulting uEV pellet was resuspended in sterile phosphate-buffered saline (PBS) and stored at −80 °C until further analysis. The morphology of uEVs was examined by transmission electron microscopy (TEM) following negative staining with 2% phosphotungstic acid. Particle size distribution and concentration were quantified using nanoparticle tracking analysis (NTA; NanoSight, UK), with sample dilutions adjusted to achieve approximately 200 particles per field. The expression of canonical uEV protein markers was verified by Western blotting, performed using standard SDS–PAGE and PVDF membrane transfer procedures.

Metabolites were extracted from uEV suspensions using a protein precipitation procedure. Briefly, 100 µL of each uEV suspension was mixed with 400 µL of ice-cold acetonitrile/methanol (1:1, v/v) containing L-2-chlorophenylalanine (0.02 mg/mL) as an internal standard. The mixture was vortexed for 30 s and subjected to ultrasonic extraction (5 °C, 40 kHz) for 30 min. Samples were then incubated at −20 °C for 30 min to facilitate protein precipitation, followed by centrifugation at 13,000 × g for 15 min at 4 °C. The resulting supernatants were evaporated to dryness under a gentle nitrogen stream, reconstituted in 100 µL of acetonitrile/water (1:1, v/v), sonicated for 5 min, and centrifuged again (13,000 × g, 10 min, 4 °C). The clarified supernatants were transferred to LC–MS vials for subsequent metabolomic analysis. Quality control (QC) samples were prepared by pooling equal aliquots from all study samples and injected periodically (one QC sample per 5 study samples) throughout the analytical sequence to monitor instrument stability and assess analytical reproducibility.

Untargeted metabolomic profiling was performed using an ultrahigh-performance liquid chromatography (UHPLC) system coupled to a Q Exactive™ Exploris 240 mass spectrometer (Thermo Fisher Scientific, USA). Chromatographic separation was achieved on an ACQUITY UPLC HSS T3 column (100 × 2.1 mm, 1.8 μm; Waters, USA) maintained at 40 °C, with an injection volume of 10 μL. The mobile phase consisted of (A) 0.1% formic acid in water/acetonitrile (95:5, v/v) and (B) acetonitrile/isopropanol/water (47.5:47.5:5, v/v/v) containing 0.1% formic acid. Mass spectrometric detection was performed in both positive and negative electrospray ionization (ESI) modes over an m/z range of 70–1050. The optimized ion source parameters were as follows: sheath gas pressure, 60 psi; auxiliary gas pressure, 20 psi; auxiliary gas heater temperature, 350 °C; capillary temperature, 320 °C; and spray voltages of +3.4 kV and −3.0 kV for the respective modes. Data were acquired in data-dependent acquisition (DDA) mode, with full MS and MS/MS resolutions of 60,000 and 15,000, respectively. Higher-energy collisional dissociation (HCD) was applied with stepped collision energies of 20, 40, and 60 eV.

For metabolomics analysis, raw LC–MS/MS data were processed using Progenesis QI software (Waters, USA) for alignment, peak detection, and annotation against HMDB and Metlin databases, achieving metabolite identification at MSI Level 2 confidence. Features detected in >80% of samples were retained, missing values were imputed (group minimum), and signal intensities were normalized to total ion current; variables with >30% RSD in QC samples were removed. Multivariate analyses (PCA and OPLS-DA) were performed using the ropls R package, with model robustness validated by seven-fold cross-validation and a permutation test (n = 200). Differential metabolites were defined by a VIP score ≥ 1.0, fold change (FC) ≥ 1.2 (or ≤ 0.83), and an P-value < 0.05. Pathway enrichment was conducted using the KEGG database.

For clinical and general statistical comparisons, data were analyzed using SPSS 24.0 (IBM Corp., USA). Continuous variables were tested for normality (Shapiro–Wilk test) and presented as mean ± SD (analyzed by Student’s t-test) or median with IQR (analyzed by Mann–Whitney U test). These univariate tests were also applied to the metabolomics data to derive P-values. Categorical variables were compared using the Chi-square or Fisher’s exact test. A two-tailed P-value < 0.05 was considered significant. Data visualization was performed using GraphPad Prism 8.0, GenesCloud (for heatmaps/volcano plots), and BioRender (for schematics).

A total of 51 patients with SLE were enrolled, including 29 patients with LN and 22 patients without renal involvement, alongside 20 age- and sex-matched HCs. All LN patients tested positive for antinuclear antibodies, and 24-hour urinary protein excretion exceeded 500 mg, consistent with established diagnostic criteria for LN. Additional laboratory parameters—including routine hematological tests, renal function indices, and immunological markers—are summarized in Table 1.

The uEVs were isolated and characterized following the Minimal Information for Studies of Extracellular Vesicles (MISEV) 2018 guidelines. Western blot analysis confirmed the enrichment of canonicalEV markers (TSG101, CD63, CD9) and the absence of the endoplasmic reticulum protein calnexin, indicating minimal cellular contamination (Figure 2A). TEM revealed the typical cup-shaped morphology and bilayered membrane structure of EVs (Figure 2B). NTA demonstrated a unimodal size distribution with a mean diameter of approximately 120 nm (Figures 2C, D). Collectively, these results confirm the successful isolation of high-purity uEVs, suitable for subsequent metabolomic profiling.

Untargeted metabolomic profiling of uEVs identified metabolites spanning multiple chemical classes, with lipid-associated species representing the dominant category across all samples (Figure 3A). PCA demonstrated a progressive shift in the global uEV metabolic landscape from HCs to SLE patients without LN and to LN patients, indicating disease-associated metabolic remodeling (Figure 3B). While most LN samples formed a distinct cluster separated from HC, a subset of LN patients clustered closely with HC in the PCA space, showing partial overlap of uEV metabolic profiles between these individuals and healthy subjects. Hierarchical clustering analysis further confirmed this pattern, as several LN samples were grouped together with HC rather than with the main LN cluster based on their metabolite abundance profiles (Figure 3C). This observation suggests that uEV metabolic alterations in LN are heterogeneous and may be influenced by factors such as differences in renal disease activity, treatment exposure, or partial metabolic normalization in certain LN patients, resulting in uEV profiles that more closely resemble those of healthy controls.

Multivariate statistical analyses were performed in SIMCA-P to compare metabolomic profiles between the total SLE cohort and HCs, as well as between the SLE with LN and SLE without LN subgroups. OPLS-DA score plots in mixed ion mode (ESI⁺/ESI^- combined) revealed clear separation between groups, indicating distinct metabolic phenotypes (Figures 4A, B). Model validity was further confirmed by 200-permutation tests, demonstrating robust predictive performance (Figures 4C, D). Differential metabolites were identified based on VIP > 1, fold change (FC) > 1.2, and P < 0.05. A total of 513 differential metabolites were detected between SLE patients and HCs (242 upregulated, 271 downregulated), while 284 differential metabolites were identified between LN and non-renal SLE subgroups (230 upregulated, 54 downregulated) (Figures 4E, F). The analysis successfully identified and validated distinct and reliable differences in the uEV metabolomic profiles among the SLE with LN, SLE without LN, and HC groups.

To further explore the biological functions of the identified differential metabolites, we performed pathway enrichment analysis using the KEGG database. As visualized in the bubble plots (Figures 5A, B), the analysis revealed distinct metabolic signatures associated with SLE and lupus nephritis. Notably, lipid metabolism pathways were prominently dysregulated in both comparisons. Specifically, Sphingolipid metabolism and Glycerophospholipid metabolism emerged as the top-enriched pathways, exhibiting high statistical significance and metabolite counts. Additionally, pathways such as Linoleic acid metabolism and Purine metabolism were also significantly altered, particularly in the stratification of LN patients. These findings suggest that the profound remodeling of lipid and nucleotide metabolism may drive the pathogenesis of SLE and its renal complications. Detailed enrichment results are listed in Supplementary Tables 1, 2.

To differentiate SLE with LN from SLE without LN, we constructed a Random Forest (RF) classifier. We first validated the model structure, where the classification error stabilized with increasing tree numbers, ensuring model stability and mitigating overfitting (Figure 6A). Subsequently, Recursive Feature Elimination with Cross-Validation (RFECV) was employed to optimize the feature set, identifying a panel of 35 priority metabolites (Supplementary Table 3) that yielded the lowest prediction error (Figure 6B). The final model demonstrated exceptional performance, achieving an Area Under the Curve (AUC) of 1.000 (95% CI: 1.000–1.000) with an optimal cut-off value of 0.36 (Figure 6C). Variable importance analysis (Figure 6D) further highlighted PE-NMe (18:1 (9Z)/20:4 (8Z,11Z,14Z,17Z)), Lyso-Gb1, and PC(20:5 (5Z,8Z,11Z,14Z,17Z)/TXB2) as the top discriminators. These findings establish a robust, non-invasive diagnostic tool for LN, underscoring the potential of specific lipid metabolic signatures in disease stratification.

Receiver operating characteristic (ROC) curve analyses were performed to evaluate the diagnostic performance of uEV metabolites for LN. The selected metabolites demonstrated robust discriminatory capacity (Figure 7), with detailed area under the curve (AUC) values provided in Supplementary Table 4. Among them, Glucosylsphingosine (Lyso-Gb1), PE-NMe(18:1(9Z)/20:4(8Z,11Z,14Z,17Z)), and PC(20:5(5Z,8Z,11Z,14Z,17Z)/TXB2) showed the highest diagnostic performance, achieving AUCs of 0.912, 0.906, and 0.897, respectively. To determine whether these uEV metabolites provide added diagnostic value beyond conventional clinical and laboratory markers, their discriminatory performance was systematically compared with established indicators of renal involvement and disease activity, including 24-hour urinary protein excretion, urine albumin-to-creatinine ratio (UACR), and complement levels (C3 and C4). As LN patients were defined and stratified according to 24-hour urinary protein, this parameter exhibited an expected near-perfect discriminatory performance (ROC AUC = 1.00), while UACR also demonstrated excellent performance (ROC AUC = 0.98). In contrast, complement C3 and C4 showed comparatively limited discriminatory ability (Figure 8). Notably, uEV metabolites not only effectively distinguished LN from non-LN patients but were also closely associated with disease activity and renal dysfunction, with these associations largely independent of proteinuria and complement levels. Collectively, these findings suggest that uEV metabolomic signatures capture metabolic and pathophysiological alterations not fully reflected by conventional proteinuria-based markers, thereby providing complementary molecular information for refined assessment and stratification of LN.

To further clarify whether the identified metabolites were primarily associated with disease activity rather than treatment exposure, we performed correlation analyses between the top three metabolites and key clinical and treatment-related variables (Figure 9). The analyses demonstrated that all three metabolites were positively correlated with disease activity indices, including SLEDAI and renal SLEDAI (rSLEDAI), and negatively correlated with estimated glomerular filtration rate (eGFR). In contrast, no significant correlations were observed between these metabolites and treatment-related variables, including corticosteroid use, immunosuppressive agents, or hydroxychloroquine, nor with other clinical parameters such as complement levels, age, or medication status. These results indicate that the top metabolites are more closely linked to disease activity and renal impairment rather than medication exposure, supporting their relevance as disease-associated metabolic signatures.