In this study, we included 11 patients with newly diagnosed pMN, all of whom had circulating autoantibodies against phospholipase A2 receptor 1 (PLA2R1). None of the patients had a history of other diseases or treatment of steroids and immunosuppressive drugs in the past year. One patient refused to undergo kidney biopsy. Based on the KDIGO 2021 guideline for the management of glomerular diseases [10], the patients were categorized into three groups: low-risk group (3 patients), moderate-risk group (4 patients) and high-risk group (4 patients).

Peripheral venous blood samples were collected from patients before treatment. PBMCs were isolated using density gradient centrifugation with Ficoll-Paque from 5 milliliter fresh anticoagulated blood. The isolated cells were resuspended in cryopreservation medium [900 µL of fetal bovine serum supplemented with 100 µL of dimethyl sulfoxide (DMSO)] for long-term storage liquid nitrogen.

Cryopreserved PBMCs from all patients were thawed, and cell viability was assessed, consistently exceeding 90%. Single-cell suspensions containing 10,000 to 20,000 cells were loaded for library construction using the Chromium Single Cell 3^′ Library (10$\times$ Genomics, Inc., Pleasanton, CA, USA) following the manufacturer’s protocol. The quality of the purified libraries was evaluated, and sequencing was performed on an Illumina NovaSeq X Plus platform (Illumina, Inc., San Diego, CA, USA) using 150-bp paired-end reads.

We performed single-cell RNA sequencing (scRNA-seq) on PBMCs from 11 patients with pMN and integrated a published scRNA-seq dataset from 5 healthy donors as a control group [11]. Samples were stratified into four groups correspondingly: normal (healthy donors), low (low-risk patients), moderate (moderate-risk patients), and high (high-risk patients). Raw sequencing data were processed using the Cell Ranger Software Suite 7.1.0 (10$\times$ Genomics, Inc., Pleasanton, CA, USA), aligning reads to the human reference genome (refdata-gex-GRCh38-2020-A). Filtered count matrices were imported into Seurat 4.3 (https://satijalab.org/seurat/) for downstream analysis. Quality control criteria included: 500–5000 detected genes per cell, total unique molecular identifier (UMI) count $\le$20,000, and mitochondrial gene expression $\le$10%. Doublets were identified and removed using DoubletFinder 2.0.3 (https://github.com/chris-mcginnis-ucsf/DoubletFinder). The cleaned data were normalized and scaled using Seurat’s built-in functions.

To integrate cells from different individuals and risk levels, we employed Harmony 1.2 (https://github.com/immunogenomics/harmony) to correct batch effects and create a shared embedding [12, 13, 14, 15]. Dimensionality reduction and clustering were performed using 2000 highly variable genes and 30 principal components. Marker genes for each cluster were identified using Seurat’s FindAllMarkers function with default parameters. Initial cell type annotation was performed using CellTypist 1.6.3 (https://www.celltypist.org/) with the ‘Immune_All_Low.pkl’ model [16], followed by manual refinement based on known marker genes from the literature.

To explore the differences between disease and control groups, we identified differentially expressed genes (DEGs) for each cell type using Seurat’s FindMarkers function, comparing patients (different risks) with healthy donors. Genes with $|$log2FC$|$ $\ge$0.5 and adjusted p-value $\le$ 0.05 were considered significant DEGs. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the clusterProfiler 4.6.2 (https://github.com/YuLab-SMU/clusterProfiler) package with default parameters to elucidate the biological processes and pathways associated with the identified DEGs.

To assess the expression of specific gene sets, we utilized Seurat’s AddModuleScore function to calculate signature scores for each cell [13]. Comparisons of signature scores among different disease and control groups were performed using analysis of variance (ANOVA) and the Kruskal-Wallis test to identify statistically significant differences.

To predict cell-cell communication, we employed CellChat 1.6.1 (https://github.com/sqjin/CellChat) to analyze ligand-receptor pairs [17]. Interaction strength was calculated for individual ligand-receptor pairs between each pair of cell types and aggregated to determine the overall interaction strength of signaling pathways. Comparative analyses were performed between disease groups (low, medium, and high risks) and the control group to identify alterations in cell-cell communication associated with disease progression. Significant signaling pathways were calculated and ranked based on differences in the overall information flow within the inferred networks between disease and control groups.

All statistical analyses were performed using R 4.2.3 (https://www.r-project.org/). Differential expression testing was conducted using FindMarkers with the Wilcoxon Rank Sum test, and p-values were adjusted by the Bonferroni correction method. Enrichment analyses were performed using the clusterProfiler package with the permutation test, and p-values were adjusted by the Benjamini-Hochberg method. To assess the statistical significance of signature score changes observed in a given cell subtype among different groups, we employed Bonferroni’s test for multiple comparison. To perform Pearson correlation analysis between candidate genes in high-risk patients and estimated glomerular filtration rate (eGFR), a linear regression model was applied to estimate the slope of the regression line and its 95% confidence interval. The data used for correlation analysis were from our own study and a public database Nephroseq v5 (https://nephroseq.org). An adjusted p-value 0.05 was considered statistically significant and annotated above the box plot as follows: p 0.05 *, p 0.01 **, p 0.001 ***, and p 0.0001 ****.

Apart from the kidney biopsy tissue, peripheral blood mononuclear cells are also important for dissecting the immunological mechanism of pMN. Using the 10$\times$Genomics scRNA-seq method, we obtained our own dataset from 11 patients (Fig. 1A), comprising 3 low-risk, 4 moderate-risk and 4 high-risk individuals. Anti-PLA2R1 antibodies were positive in all patients (Table 1). We compiled our data and a published one with 5 healthy individuals, which served as a negative control group (Supplementary Table 1). After stringent quality control and filtration, a total of 178,185 single-cell transcript data of PBMCs were analyzed. After correcting batch effects between the two datasets, we integrated the data to perform unsupervised graph-based clustering. Five major lineages (T cells, B cells, NK cells, Monocytes and cDC2) were identified depending on a referenced automatically annotation (Fig. 1B,D) [16]. In addition, other kinds of immune cells were not used to perform analysis due to their low abundance, including innate lymphoid cell 3 (ILC3), natural killer T cells (NKT), immature lymphoid cells and platelets and so on (Fig. 1B). The published data contained the same types of cells as ours. Cell annotations were verified independently by two investigators to ensure accuracy. No significant differences were observed in the overall distribution of cell populations across the four groups (Fig. 1C).

We conducted GO enrichment analysis on the five kinds of immune cells, comparing patients with pMN to healthy donors. Top up-regulated and down-regulated signaling pathways were showed in individual type cells. Immune-response related pathways were notably more active in patients across all cell types (Fig. 1E). The average capacity of antigen presentation was reduced in B cells from patients (Fig. 1E).

In order to discovery more positive results, we decided to divided specific type of immune cells into several subtypes. Based on reference annotations, individual cell lineages were categorized as shown in Fig. 2. CD4⁺ T cells were divided into three subgroups, including naive T (Tn) and central memory T (Tcm), effector T (Te) and effector memory T (Tem) and regulatory T cells (Treg). CD8⁺ T cells were classified into three subgroups, including Tn and Tcm, Tem and Temra (CD45RA⁺), Tem and resident memory T cells (Trm) (Fig. 2A–C). B cells included naive (Bn), activated B (Ba), non-switched memory (Bnsm) and switched memory B cells (Bsm) (Fig. 2D–F). NK cells included CD56^+⁣/- NK cells (or CD56^dimCD16^high) and CD56⁺ NK cells (or CD56^brightCD16^low) (Fig. 2G–I). Monocytes consisted of two subgroups, namely classical (CD14⁺CD16^–) and non-classical monocytes (CD14^–CD16⁺) (Fig. 2J–L). Of note, classical dendritic cell 2 (cDC2) and MAIT did not subdivide further in our dataset (Fig. 1B).

As we know, circulating immunoglobulin G4 (IgG4) are thought to bind to deposited autoantigens and form the subepithelial complex. To investigate the potential correlation between immunoglobulins and disease severity, we analyzed the data of B cells. From naive to memory B cells, expression levels of IGHD and IGHM gradually decreased, confirming that our data was qualified (Fig. 3A). Consistent with IgG4 deposition in glomerular biopsies, IGHG4 expression in memory B cells was higher in patients than in healthy donors (Fig. 3D). Similar trend was observed for IGKC (Fig. 3F). Intriguingly, transcript levels of IGHG3 in memory B cells were gradually upregulated from low-risk to high-risk patients (Fig. 3A,B). Whereas expression levels of IGHA1 and IGHG2 in activated and memory B cells were lower in patients (Fig. 3C,E). Together, these results suggest that IgG3 might participate into the progression of pMN with antibodies against PLA2R1 [18].

Complement activation plays a central role in the pathogenesis of pMN. Subepithelial deposition of circulating immune complexes promotes the formation of the terminal complement component C5b-9, which subsequently damaged the glomerular basement membrane (GBM) [18]. To determine whether the complement system was active in peripheral immune cells, we analyzed complement-related markers in B cells. We found that complement receptor 1 (CR1) expression in B cells was higher in patients compared to controls (Fig. 3G). Similar trend was showed in patients with regard to C1S, C2 and C5. To provide a more accurate assessment, we developed a score using a panel of biomarkers up-regulated in patients, including C1S, C2, C5, CR1, CD46, CD55 and ITGAX. Scores for memory B cells and monocytes were significantly higher in patients than in healthy donors (Fig. 3H). These findings indicate that abnormal activation of complement system was present in patients with pMN, potentially contributing to disease progression.

Chronic inflammation is a prominent feature of many autoimmune diseases [19], including pMN. Numerous proinflammatory pathways, involving interleukins, interferons, tumor necrosis factors, chemokines, growth factors and their corresponding receptors, may contribute to the development of pMN. Nevertheless, that the relationship between proinflammatory responses and pMN progression remains poorly understood. To address this issue, we investigated the expression levels of these inflammatory factors across different types of immune cells.

With regard to interleukin signalings, we found that the levels of interleukin 1B (IL1B) and IL15 in monocytes and cDC2 were significantly elevated in patients compared to the control group (Fig. 4A–C). Whereas IL16 expression in lymphoid cells was significantly decreased in patients [20], similar results were showed about IL32 and IL18 in T and cDC2 cells, respectively (Fig. 4D, Supplementary Fig. 1A,B). Regarding interleukin receptors, we found that IL2RG and IL6ST in lymphoid cells were significantly higher in patients with pMN (Fig. 4E). The similar trends were observed in monocytes and cDC2 for IL6R and IL10RB. Multiple scoring metrics further confirmed the findings (Fig. 4F).

In terms of interferon signaling, expression levels of interferon gamma (IFNG), IFNG antisense RNA1 (IFNG-AS1) and interferon alpha and beta receptor subunit 2 (IFNAR2) in specific type cells were significantly higher in patients (Fig. 4G–I, Supplementary Fig. 1C–F).

Regarding the tumor necrosis factor superfamily, we noted that TNFRSF13C, TNFAIP3, TNFAIP8 and TNFSF8 in T cells were significantly upregulated in patients, as confirmed by specific score systems (Fig. 4J–N, Supplementary Fig. 1G,I). In contrast, expression levels of TNFSF13B and TNFRSF14 in lymphoid cells were downregulated in patients (Fig. 4J,L,M, Supplementary Fig. 1H).

Additionally, several chemokines in monocytes were higher in patients compared to healthy donors, including CCL3L1, CXCL8 (IL8) and CXCL10 (Fig. 4O–R, Supplementary Fig. 1J), with a similar trend for CXCR4 in lymphoid cells (Supplementary Fig. 1K). In contrast, expression levels of CX3CR1 and CCR7 in specific kinds of cells were relatively lower in patients.

Growth factors also play an essential role in inflammatory responses. Scores of growth-factor family members (TGFBR1, TGFBR3, IGF2R and PDGFD) were significantly higher in patients with pMN, with a similar trend in high-risk patients compared to low-risk group (Fig. 4S,T). Conversely, expression levels of some factors (MIF, FGF23 and PDGFA) significantly decreased in patients than in the control group.

Taken together, these data indicate that a range of proinflammatory pathways is activated, contributing to the development of pMN with anti-PLA2R1 antibodies.

It is well established that proinflammatory responses are often activated by pattern recognition receptors (PRRs) signalings [21], which includes toll-like receptors (TLRs) [22], NOD-like receptors (NLRs) [23], RNA sensors (RIG-I like receptors, RLRs) [24], DNA sensors (cyclic GMP-AMP synthase (cGAS) receptors, cGLRs) and so on [25]. To investigate whether PRR pathways participate into the development of pMN, we examined their expression in patients compared to healthy donors. Notably, many of the PRRs in specific type of cells were elevated in patients, including NLRs (NLRP1, NLRP3 and NLRC5) (Fig. 5D–F), RNA sensors (DDX58, IFIH1, DDX9 and DDX36) (Fig. 5G,H), DNA sensors (cGAS and IFI16) (Fig. 5I–L) and TLR1 (Fig. 5A,B). In contrast, some were relatively lower in patients, such as TLR10 (Fig. 5A,C) and TMEM173 (Supplementary Fig. 1L).

Given the downstream effects of PRR signalings, we next examined the activation of transcript factors (TFs)-related pathways that are often implicated in immune response pathways. Interestingly, we found that several TF-related pathways were upregulated in patients than in healthy donors, including MAPKs, AP1 (FOS, JUN, JUNB and JUND), janus kinase (JAK) and signal transducer and activator of transcription (STAT) members (JAK1, JAK2 and STAT3), nuclear factor kappa B family (NFKB1, RELB and NFKBIA), forkhead domain family (FOXO1 and FOXO3), TRAF3, HIF1A and IRF1 (Supplementary Fig. 2) [26, 27, 28].

Together, these results suggest that a broad range of innate immune signaling pathways in PBMCs may play a role in the pathogenesis of pMN with anti-PLA2R1 antibodies.

According to the data of different expression genes (DEGs) between patients and healthy donors, we found several interesting biological processes enriched in patients, which might affect the function of immune cells. Cellular senescence plays a key role in chronic diseases, including autoimmune diseases, metabolic diseases and tumors [29]. We found that a senescence-related gene panel was significantly enriched in patients than in the control group (Supplementary Fig. 3A). Statistically, integrated score of cellular senescent genes was higher in immune cells of patients (Supplementary Fig. 3B–D). In terms of circadian rhythm and chromatin remodeling, similar trends were showed in patients (Supplementary Fig. 3E–K) [30, 31, 32]. Altogether, these functional abnormalities of immune cells in pMN need to be explored comprehensively, which would offer more potential therapy targets for patients with pMN.

Although there are some biomarkers used for distinguishing high-risk patients from others, such as proteinuria, serum albumin and eGFR [10], it is limited that our understanding of the immune pathogenesis in the development of high-risk pMN. To address this issue, we analyzed the DEGs between high-risk and low-risk patients.

Intriguingly, genes of defense response to viruses in T, NK cells and monocytes were significantly upregulated in the high-risk group compared to the low-risk group (Fig. 6A,B,G,H,J,K) [33, 34]. Similarly, interferon-mediated signaling pathway in B cells was more active in high-risk group (Fig. 6D,E). In cDC2 cells, genes related to antigen presentation were significantly increased in high-risk patients (Fig. 6M,N), as confirmed by scores of upregulated gene sets (Fig. 6C,F,I,L,O). Of note, all of the scores were also effective for distinguishing high-risk patients from moderate-risk population.

In order to develop a score system for identifying high-risk patients in the overall population, we selected upregulated genes in high-risk patients compared to low-risk individuals in most types of immune cells, including SLC35F1, AC105402.3, STAT1, EPSTI1, MX1 and EIF2AK2 (Supplementary Fig. 4A,B). The gene TRIM22 was excluded due to its high expression in healthy donors (Supplementary Fig. 4A). Importantly, this score system was also effective in distinguishing high-risk patients from moderate-risk individuals.

To further distinguishing moderate-risk patients from high-risk and low-risk populations, we selected genes gradually upregulated from low-risk to high-risk patients in most types of immune cells to develop different score systems (Supplementary Fig. 4C,E–G). Notably, SLC35F1 in B cells was a valuable biomarker for distinguishing the three risk groups (Supplementary Fig. 4D) [35, 36].

To confirm the predictable value of upregulated biomarkers in high-risk patients compared to low-risk population, we performed correlation analysis between these biomarkers and eGFR in newly-diagnosed patients with pMN. As shown in Table 2 and Supplementary Fig. 5, most effective biomarkers were negatively correlated with eGFR. Of note, upregulation of SLC35F1 and BTS2 in most types of PBMCs could effectively predict poor kidney function. Both GBP5 and KLF6 also had the predictable role in more than one type of immune cells. Conversely, only two genes (LRMDA and MYO1F) were positively associated with eGFR in cDC2 cells.

Given the small sample size of our study, we attempted to figure out whether the screened biomarkers enriched in high-risk patients had a robust capability of predicting declined eGFR in the development of nephrotic syndrome and other chronic kidney diseases (Nephroseq v5 database). Indeed, the predictable values of most biomarkers were confirmed in a larger population (Supplementary Table 2). Despite the validated results were calculated based on kidney tissues, this fact suggests that biomarkers in blood cells have a similar effect on risk stratification.

In addition, we assessed the predictable values of prior selected biomarkers upregulated in total patients with pMN compared to healthy donors. Similarly, most of them were negatively correlated with declined eGFR in patients with nephrotic syndrome or other chronic kidney diseases (Supplementary Table 3).

Altogether, we identified a series of effective biomarkers used for distinguishing high-risk patients from low-risk individuals with pMN and for predicting declined eGFR of patients.

Prediction of intercellular communication networks is also necessary for figuring out the pathogenic mechanism of pMN. We analyzed potential cell-cell interactions by examining co-expression patterns of curated ligand-receptor pairs (Fig. 7A and Supplementary Fig. 7).

With regard to the relative information flow, we found that some interaction pairs increased significantly in disease groups, such as cell chemokines (CCL) and cytokines (TGFB and IFN-II) (Fig. 7A, Supplementary Fig. 7A,B,D). And many up-regulated pairs correlated with functional changes in cell adhesion, such as CD6, CD226, ALCAM, NECTIN and NCAM (Fig. 7A) [37]. TIGIT-related pairs may facilitate the inhibition of monocytes by regulatory T cells, suggesting a protective mechanism that could prevent excessive monocyte-driven inflammation (Fig. 7A, Supplementary Fig. 7C) [38].

Conversely, information flow of some interaction pairs decreased remarkably in patients, such as human leukocyte antigen (HLA) molecules (major histocompatibility complex, class I and II) (Fig. 7A–C,E,F). All of the HLA-I molecules from PBMCs were decreased in patients compared to healthy donors (Supplementary Fig. 6A–C, Fig. 7H). Similar trends were showed in B cells for most of the HLA-II molecules (Fig. 7I,J). However, with regard to memory B cells, expression levels of HLA-DQA1 and HLA-DRB1 were significantly higher in patients (Fig. 7J–L). Overall, these findings suggest that memory B cells may play a role in the pathogenic mechanisms of pMN though HLA-DQA1 or HLA-DRB1 [39, 40].

In addition, we found the putative CLEC2 (C-type lectin domain family 2) and KLRB1 (killer cell lectin like receptor B1) pairs were relatively activated in patients (Fig. 7A,D,G). Indeed, the expression levels of CLEC2B, CLEC2C (CD69) and CLEC2D in most cells were significantly upregulated in patients compared to healthy donors (Supplementary Fig. 6H) [41], as confirmed in CLEC score system (Supplementary Fig. 6I). KLRB1 was regarded as an inhibited receptors in CD56^+⁣/- NK cells [42]. CLEC2B was expressed mainly in monocytes, T and NK cells, CLEC2C and CLEC2D were expressed prominently in T, B and NK cells (Supplementary Fig. 6H). We speculated that B and T cells might inhibit cytotoxic NK cells to prevent further kidney injury. This hypothesis was supported by decreased expression of granzyme M (GZMM) in NK (CD56+/-) cells (Supplementary Fig. 6C–G). With regard to perforin 1 (PRF1) and GZMB, similar expression patterns were showed in NK (CD56+/-), Tem and Temra (CD8+) cells (Supplementary Fig. 6C). Furthermore, we found that index of cytotoxicity was significantly lower in patients than in healthy donors using our developed score (including GZMA, GZMB, GZMM, PRF1) (Supplementary Fig. 6G). Abnormal expression of Fc receptors in classical monocytes and cDC2 cells, such as FCAR and FCGR2B (Supplementary Fig. 6J-L), which may further disturb the antibody-dependent cell-mediated cytotoxicity [43, 44]. Altogether, these finding suggests that the presence of a compensatory mechanism could mitigate glomerular injury.