Protein A- or G-coated magnetic beads were purchased from Millipore Sigma (Saint Louis, MO). The native human C1Q was purchased from Abcam (Boston, MA, USA). Papain solution was purchased from Millipore Sigma, MO. The CD14, CD34, CSTA, UBC, and BST1 antibodies were purchased from R&D Systems (Minneapolis, MN, USA), the P3H1, RGS12, and GUK1 monoclonal antibodies were purchased from Santa Cruz Biotechnology (Dallas, TX, USA), and the MEF2C and PHACTR4 monoclonal antibodies were purchased from Cell Signaling Technologies (Danvers, MA, USA) and Abcam, respectively. The anti-human IgG antibody was purchased from Jackson ImmunoResearch (West Grove, PA).

Serum samples from lupus nephritis and healthy controls were collected at the University of Texas, Southwestern Medical Center at Dallas. The samples were aliquoted and stored at -80°C. All human subject-related procedures were performed following the institutionally approved IRB protocols, and all consents were obtained before sample collection. The detailed demographics and clinical information are summarized in Table 1. Active lupus nephritis (LN-active) is defined as Systemic Lupus Erythematosus Disease Activity Index (SLEDAI) greater than 4 and the renal domain of SLEDAI (rSLEDAI) greater than 0. Inactive lupus nephritis (LN-inactive) is defined as SLEDAI less than 4 and rSLEDAI equal to 0. In the immunoproteomic screening study, serum samples from 3 LN patients (SLEDAI = 4, 19, and 20, respectively) or 3 healthy controls were pooled, respectively, for further experiments.

40 µl of Protein A- or G-coated magnetic beads was gently mixed and washed with 0.1% PBST. 100 µl of 1:100 PBS-diluted pooled serum samples from LN or healthy controls was then mixed and incubated with the Protein A/G magnetic beads at room temperature for 30 min with gentle shaking. The immune complex bound beads were then washed three times with 50 µl PBS. A magnetic stand was used to hold the beads, while the supernatant was removed after washing. The washed beads were suspended in 50 µl of 0.1 mg/ml papain solution (0.04 M EDTA, 0.04 M L-cysteine) and incubated at 37°C for 1 h. The magnetic stand was used to remove the beads, and the resultant supernatant was transferred to a new tube, and 50 µl of 0.06 M iodoacetamide was added to terminate the reaction of papain digestion, followed by gel purification and mass-spectrometry analysis.

100 µl of 200 µg/ml native human C1q was diluted into PBS, coated onto a 96-well microplate (Thermo Fisher, Waltham, MA), and incubated overnight at room temperature. The microplate was washed three times with 150 µl PBST and blocked, and then 100 µl of 1:100 PBS-diluted pooled serum samples was added into the wells and incubated for 2 h at room temperature. The aqueous portion was removed, and the microplate was washed three times with 150 μl PBST. A 50-µl 0.1-mg/ml papain solution was added to each well and incubated at 37°C for 1 h. Then, 50 µl of 0.06 M iodoacetamide was added into the well to terminate the papain digestion, followed by gel purification and mass spectrometry analysis.

Supernatants from the Protein A/G- or C1q-captured immune complex samples were purified by a brief running (≤10 min) of SDS-PAGE. The gel band containing samples was subjected to in-gel digestion (19). An aliquot of the tryptic digest (in 2% acetonitrile/0.1% formic acid in water) was fractionated through liquid chromatography and then analyzed by an Orbitrap Fusion™ Tribrid™ mass spectrometer (Thermo Scientific™) interfaced with a Dionex UltiMate 3000 Binary RSLCnano System. Peptides were separated onto an Acclaim™ PepMap™ C18 column at a flow rate of 300 nl/min. Gradient conditions were 3%–22% B for 40 min; 22%–35% B for 10 min; 35%–90% B for 10 min; and 90% B held for 10 min (solvent A, 0.1% formic acid in water; solvent B, 0.1% formic acid in acetonitrile). The peptides were analyzed using a data-dependent acquisition method; Orbitrap Fusion was operated with the measurement of FTMS1 at resolutions of 120,000 FWHM, scan range of 350–1,500 m/z, AGC target 2E5, and maximum injection time of 50 ms. During a maximum 3-s cycle time, the ITMS2 spectra were collected at a rapid scan rate mode, with HCD NCE 35, 1.6-m/z isolation window, AGC target 1E4, maximum injection time of 35 ms, and dynamic exclusion which was employed for 35 s. The raw data files were processed using Thermo Scientific™ Proteome Discoverer™ software version 1.4, and spectra were searched against the UniProt Homo sapiens database using the Sequest search engine v2.3.02 (Matrix Sciences, Chicago, IL, USA) run on an in-house server. Search results were trimmed to a 1% FDR for stringency and 5% for relaxed conditions using Percolator. For the trypsin digestion, up to two missed cleavages were allowed. MS tolerance was set at 10 ppm; MS/MS tolerance at 0.6 Da. Carbamidomethylation on cysteine residues was used as fixed modification; oxidation of methionine and phosphorylation of serine, threonine, and tyrosine were set as variable modifications. The sum of scores of individual peptides for identified proteins was used to identify proteins from the immune complex.

A total of 77 candidate proteins with protein accession numbers were converted into gene names with DAVID (https://david.ncifcrf.gov) and UniProt (https://www.uniprot.org) within Homo sapiens species. The functional enrichment gene-set analysis for three GO (Gene Ontology) sub-ontologies was performed with ClusteProfiler 4.0 (20) with a p-value cutoff of 0.05 and the “Benjamini–Hochberg” p-Adjust-value method. The “Disease class,” “Up tissue,” and “Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathway” annotations were retrieved from DAVID (21).

Six publicly available SLE or LN databases (Supplementary Table 3) were downloaded to cross-validate the autoantigens discovered using immunoproteomics in the present study. In the 6 databases, 3 major genomic technologies were used, including single-cell RNA Sequencing (scRNA Seq), total RNA sequencing (RNA-Seq), and gene expression microarray (RNA-Array), to eliminate bias generated by single technology or sample type. The differentially expressed genes (DEGs) were defined as adjusted p-values (p-adjust) smaller than 0.05 between SLE/LN and healthy controls. In the two scRNA databases, Seurat V4 was used to generate an expression matrix, PCA dimension reduction, and cell cluster as described previously (22). Each cluster cell type was identified with reference-based package SingleR (23) and verified with canonical markers. Then, the DEGs were discovered by DEsingle (24). For two RNA databases, processed count files were directly obtained from GEO, and the DEGs were discovered by DESeq2 (25). The databases from the two RNA-Array were downloaded and standardized, and the DEGs were screened using the GEOquery and Limma (26, 27).

ICx in the serum of LN and controls was measured using an in-house-developed ELISA kit. Briefly, to capture antigen-specific ICx, commercial monoclonal antibodies (tested by Western blot) against each autoantigen were coated onto the Immulon 2 HB flat-bottom microtiter plates (Thermo Fisher, MA) overnight at 4°C and then blocked. Serum samples were diluted into 1:100 in reagent diluent, added into the wells, and incubated for 2 h at room temperature. Anti-human IgG antibody (Jackson ImmunoResearch, PA) was diluted into 1:20,000 and added into the wells and incubated for 1.5 h at room temperature followed by color development. The ELISA signal was read by an Epoch plate reader (BioTek, Winooski, VT) at 450 nm, and the blank was subtracted from the OD450 readings of the samples.

All data were analyzed, plotted, or visualized with R 4.1.0 language or ggplot2 (28) package unless stated otherwise. Biomarker-level group-wise comparisons of statistical significance (p-values) were analyzed using the Wilcoxon test. Principal component analysis (PCA) was conducted to transform the serum levels of ICx into uncorrelated principal components, and only the first two components were used for the dot plot. Sensitivity (true positive ratio), 1-specificity (false positive ratio), and area under the curve (AUC) values were determined by receiver-operating characteristic (ROC) analysis using a pROC package (29). The least absolute shrinkage and selection operator (LASSO) (30) was used to evaluate and select candidate serum ICx as the best panel of biomarkers to distinguish LN from HC or LN-active from LN-inactive with the largest value of lambda under 3-fold cross-validation. The correlation between serum ICx levels and clinical parameters was determined by Spearman’s correlation coefficient.

To discover novel ICx that are differentially expressed in SLE or LN, we designed two strategies to capture ICx, as illustrated in Figure 1, showing Protein A/G magnetic beads or C1q-based capture of immunoglobulins or the antigen–antibody complex. Pooled serum samples from 3 LN patients or 3 health controls were used to capture and enrich ICx. As shown in Supplementary Figure 1, SDS-PAGE was used to visualize the immunocaptured products. After gel purification and in-gel digestion, the peptides were analyzed using liquid chromatography and tandem mass spectrometry (LC-MS/MS). In total, 239 and 61 immunoglobulin or immunoglobulin-binding proteins were discovered via Protein A/G and C1q, respectively. Based on the cumulative protein score of each protein, we ended up with identifying 52 (Protein A/G method) and 27 (C1Q method) proteins which were only detectable in the serum of LN but not healthy controls by mass spectrometry, as shown in Supplementary Table 1 and Supplementary Table 2. Among the 77 unique proteins identified, TTR and KRT14 were found in both Protein A/G and C1q procedures. Gene Ontology (GO) enrichment analyses were performed to investigate the enrichment function of these proteins that are expressed in the LN patients, as shown in Figure 2. In the biological process (Figure 2A), the protein functions of complement activation and B cell activation were significantly enriched which indicated that the 77 proteins may be involved in forming an immune complex during disease development and contribute to the pathogenesis of SLE or LN. The most significant cellular component (CC), immunoglobulin complex, represents 16 immunoglobulins which consist of 11 IgH, 3 IgK, and 2 IgL. Several IgH were found associated with B cell receptor (BCR) analysis of immune-mediated diseases, such as IGHV4-34 and IGHV4-59, which are highly expressed in SLE (31).

Given that the 77 proteins captured through immunoprecipitation were only detectable in LN but not healthy controls using the immunoproteomic approach, we assumed they may be functionally relevant to SLE or LN. To confirm this, we performed a bioinformatics analysis to determine which proteins may be involved in the disease course of SLE or LN and may potentially serve as a disease biomarker. As shown in Supplementary Table 1 and Supplementary Table 2, among the 77 proteins, 32 of them are highly abundant proteins including keratins, histones, complement system proteins, and albumin and immunoglobulin family proteins which have already been discussed extensively in previous studies (32–36). Therefore, they were eliminated from further analysis in the current study. For the remaining 45 proteins, we performed the following bioinformatics analysis: Database for Annotation, Visualization and Integrated Discovery (DAVID), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Ontology Biological Processes (GO-BP), to determine the relevance of the candidate autoantigens (AAgs) to SLE or LN.

In DAVID analysis, we collected 5 features of DAVID functional annotation as described previously (21). As SLE is aberrant in immune function and causes end-organ damage, particularly renal damage (37), the “renal” or “immune” disease-related proteins or the proteins expressed in SLE-affected tissues (such as skin, muscle, bone, joints, kidney, spleen, and brain) were considered “positive” and are indicated in red in Figure 3. In KEGG analysis, the proteins involved in “Systemic lupus erythematosus - Homo sapiens (human)” KEGG pathway (https://www.genome.jp/entry/pathway+hsa05322) and its eight related pathways (hsa04060, hsa04514, hsa04610, hsa04612, hsa04630, hsa04660, hsa04662, and hsa04670) were considered “positive” and are indicated in red in Figure 3. In GO-BP analysis, the “immune response” and “Inflammatory” annotated proteins were considered “positive” and are indicated in red in Figure 3. The “Cumulative Protein Score” is the sum of the peptide scores for proteins in LN patients, which was normalized to be in the range of 0~5.

Next, we used 6 published gene expression databases of SLE or LN as detailed in Supplementary Table 3, to determine if the genes that encode the autoantigens were differentially expressed in SLE/LN compared to healthy controls. The 6 databases were derived based on data from single-cell RNA sequencing, total RNA sequencing, or gene expression microarray on human SLE/LN genomes from various cell or tissue types (38–43). Differentially expressed genes (DEGs) were defined as adjusted p-values (p-adjust) less than 0.05 between SLE/LN and healthy controls. As shown in Figure 3, the filled color represents the log (fold-change) values (p.adjust < 0.05) as indicated in the scale bar.

Two general criteria were used to determine which candidate proteins (antigens) would be selected to move forward for the next phase of validation studies: 1) the protein received at least one “positive” in the comprehensive functional annotations (DAVID/KEGG/GO, Figure 3) and 2) the protein-coding genes are “upregulated” in two genomic databases (Supplementary Table 3) or highly upregulated in one genomic database. As a result, 10 candidate proteins satisfied both criteria 1 and criteria 2 and were selected for validation.

A total of 10 candidate AAgs including BST1, CD14, CD34, CSTA, FN1, MEF2C, P3H1, PHACTR4, RGS12, and UBC were selected for validation. Given the fact that the immune-capture process was designed to capture either immunoglobulins or the antigen–antibody complex with Protein A/G- or C1Q-based methods, the proteins/antigens identified by mass spectrometry were presumably the “bound-form” and originated from the antigen–antibody complex. Therefore, validating serum ICx or bound-form antigens is more reasonable than validating the free-form antigens. We then used monoclonal antibodies against each specific AAg to coat the ELISA plate, and the corresponding ICx were captured and detected on the plate using a secondary antibody, anti-human IgG conjugated with horseradish peroxidase (HRP). In the validation study, we used an independent cohort of 62 LN patients and 21 healthy controls. Detailed patient demographics and clinical characteristics information are presented in Table 1. The LN patients were divided into LN-active and LN-inactive based on SLEDAI and rSLEDAI index, as detailed in Methods.

As shown in Figure 4A, the BST1, PHACTR4, RGS12, and UBC-specific ICx were significantly upregulated in LN-inactive patients, compared to the healthy controls. Furthermore, BST1, CD34, RGS12, and UBC-specific ICx were significantly elevated in LN-inactive compared to LN-active patients (Figure 4A). Surprisingly, P3H1 immune complex levels were significantly downregulated in LN-inactive patients, compared to the healthy controls. Also, FN1- and MEF2C-specific ICx were decreased in LN-active compared to healthy controls. This inconsistency with the screening data may be due to the heterogeneity of SLE and the relatively smaller sample size in the screening cohort. Also, some of the commercial monoclonal antibodies may not be optimal (e.g., affinity or epitopes) in maximally capturing serum ICx by ELISA. In addition, it is conceivable that serum CD34 and MEF2C ICx levels may respond to drug treatment, because they were significantly lower in the LN-active group compared to the LN-inactive group following drug treatment (Supplementary Figure 2). However, no significant responses to drug treatment were observed in the other ICx as well as in anti-dsDNA autoantibodies (Supplementary Figure 2).

To uncover the distribution patterns of the 10 ICx in an unbiased manner, dimension reduction principal component analysis (PCA) was performed using the ELISA validation data. The percentage of variance for the first and second principal components (PCs) were 48.32% and 23.16%, respectively, which were used to map both LN and HC samples (Figure 4B). The HC samples were clearly separated from LN patients implying the distinction of their ICx expression patterns. The discriminative capabilities of 10 ICx for LN vs. HC were evaluated with ROC analysis (Figure 4C). P3H1, PHACTR4, and RGS12 ICx outperform other markers with AUC values greater than 0.8. Furthermore, when individual ICx were combined into a biomarker panel with the LASSO model, the discriminative ability was significantly improved (Figure 4D). In the LN vs. HC group, the AUC value was 0.99, where the greatest positive and negative contributing variables were PHACTR4 and FN1, respectively. In the LN-active vs. LN-inactive group, 4 positive (CD34, MEF2C, RGS12, and UBC) ICx and 2 negative (CD14 and CSTA) ICx contributing variables were identified using the LASSO model. When combined into a panel, these biomarkers can discriminate LN-active from LN-inactive with an AUC value of 0.85. In comparison, when using clinical parameters to discriminate LN-active from LN-inactive, the AUC values were 0.81 for renal chronicity index (CI), 0.62 for renal activity index (AI), 0.58 for white blood cell counts (WBC), and 0.91 for proteinuria. Hence, the LASSO-derived serum biomarker panel (CD14, CD34, CSTA, MEF2C, RGS12, and UBC ICx) outperformed the renal pathology indices such as AI and CI.

Next, we determined if serum ICx were associated with clinical and pathological parameters and if they had a diagnostic value in reflecting renal pathological change without the need of renal biopsy. As shown in the correlation heatmap in Figure 5, most serum ICx exhibited a significant negative correlation with SLEDAI, rSLEDAI, WBC, and proteinuria. Interestingly, serum ICx levels exhibited a negative correlation with AI but a positive correlation with CI, as well as the individual components of AI and CI. This is consistent with the fact that serum ICx was lower in LN-active compared to LN-inactive patients. Besides, the longitudinal studies indicated that BST1, CSTA, and UBC serum ICx levels changed over two time points, which could track with changes in SLEDAI and/or rSLEDAI (Supplementary Figure 3).

To determine if these AAg-encoding genes are differentially expressed in various cell types in LN, we examined the gene expression of the AAgs at a single-cell resolution using a database of scRNA sequencing of PBMC from 8 SLE adult patients and 5 healthy controls (42). As shown in Figure 6A, 8 different cell types were identified from 82,748 cells and each cell population in the SLE and HC groups is displayed in Figures 6A, B. CD14, BST1, CSTA, and MEF2C were found to have an enriched gene expression in CD14+ or CD16+ monocytes, and MEF2C was also highly expressed in SLE B cells compared to healthy controls. Furthermore, the significance test using Wilcoxon comparison between SLE and HC in different cell types is presented in Figure 6C. P3H1, PHACTR4, and UBC were ubiquitously expressed in all cell types (Figure 6D). In the CD4+ T cell cluster, P3H1 and UBC were found upregulated, whereas MEF2C was downregulated in the SLE group, compared to healthy controls. In monocytes, CD14 were highly expressed in CD16+ cells, and MEF2C and FN1 were upregulated in CD14+ cells in SLE compared to healthy controls. UBC exhibited overall the strongest expression across all cell types, and it was significantly upregulated in CD4+ T cells, CD8+ T cells, and natural killer (NK) cells in SLE compared to HC. However, it was significantly downregulated in dendritic cells (DCs) in SLE, compared to HC. The gene expression profiles of these AAgs were also examined in 1,496 cells from LN renal tissues using the scRNA Seq database (details are shown in Supplementary Figure 4). The results demonstrate that MEF2C were highly expressed in leukocytes, endothelial cells, and mesangial cells; FN1 exhibited an overall higher gene expression in renal cells compared to PBMCs; UBC was also found highly expressed across all cell types in the kidney, which is consistent with the PBMC data.

In summary, we found that PHACTR4, RGS12, UBC, CSTA, and BST1 are concordantly elevated in both our ICx assays and the public gene expression database, as detailed in Figure 4 and Supplementary Table 3.