A human-mouse atlas of intrarenal myeloid cells identifies conserved disease-associated macrophages in lupus nephritis

Abstract

Monocytes and macrophages in patients with lupus nephritis exhibit altered behavior compared to healthy kidneys. How to optimally use mouse models to develop treatments targeting these cells is poorly understood. This study compared intrarenal myeloid cells in four mouse models and 155 lupus nephritis patients using single-cell profiling, spatial transcriptomics, and functional studies. Across mouse models, monocyte and macrophage subsets consistently expanded or contracted in disease. A subset of murine classical monocytes expanded in disease; these cells, expressed Cd9, Spp1, Ctsd, Cd63, Apoe, and Trem2, genes associated with tissue injury in other organs and that play roles in inflammation, lipid metabolism, and tissue repair. Resident macrophages expressed similar genes in clinical disease. In humans, we identified analogous disease-associated monocytes and macrophages that were associated with kidney histological subtypes and disease progression, sharing gene expression and localizing to similar kidney microenvironments as in mice. This cross-species analysis supports the use of mouse functional studies for understanding human lupus nephritis.

eTOC summary:

This study analyzes the intrarenal myeloid compartment in four lupus nephritis mouse models and 155 humans using single-cell, spatial transcriptomics, and functional studies. It identifies shared myeloid subsets linked to disease and tissue damage, supporting mouse models’ relevance for human lupus nephritis interventions.

INTRODUCTION

Systemic lupus erythematosus (SLE) affects more than 200,000 Americans, with over 40% developing lupus nephritis and 8-10% progressing to renal failure over a 20-year span(Petri et al., 2021; Izmirly et al., 2021; Deonaraine et al., 2021). This autoimmune condition arises from the breakdown of immune tolerance, leading to the production of autoantibodies targeting nuclear material. Intrarenal deposition of immunoglobulin G (IgG) and complement triggers inflammation and pathological changes in the glomeruli and tubulointerstitium, resulting in tissue injury(Davidson, 2016; Maria and Davidson, 2020). Deposition of immune complexes in the subendothelial regions of the glomerulus leads to recruitment of immune cells from the circulation and proliferative (Class III and IV) glomerulonephritis, whereas recruitment of immune complexes to the subendothelial region causes membranous (Class V) glomerulonephritis characterized by podocyte injury and proteinuria but less glomerular inflammation. Both forms of glomerulonephritis can be associated with tubulointerstitial injury and fibrosis that lead to end stage renal disease(Davidson et al., 2013).

Myeloid cell infiltration has been linked to kidney injury and fibrosis in patients with both proliferative and membranous lupus nephritis(Davidson, 2021). Through single-cell RNA sequencing (scRNA-seq), we previously identified five novel myeloid subsets in kidney biopsies from 23 lupus nephritis patients and 10 healthy donors(Arazi et al., 2019a). Subsequent scRNA-seq analyses of kidney biopsies from 155 lupus nephritis patients(Deonaraine et al., 2021) offer a more comprehensive view of this immune compartment, described in an upcoming resource study. However, functional studies on small numbers of cells isolated from human kidney biopsies remain challenging.

Mouse models of lupus have been instrumental in elucidating the molecular and cellular mechanisms underlying lupus and lupus nephritis(Davidson, 2021; Du et al., 2015; Richard and Gilkeson, 2018) including myeloid cell depletion that resulted in reduced disease severity(Lenda et al., 2003); (Triantafyllopoulou et al., 2010). These models, characterized by antinuclear autoantibodies and immune complex-mediated glomerulonephritis reflect various aspects of human lupus nephritis, with diverse histopathological changes resulting in kidney injury(Davidson, 2016); (Du et al., 2015); (Richard and Gilkeson, 2018). To leverage these models effectively, it is crucial to identify molecular and cellular features relevant to human disease. Here, we characterized intrarenal myeloid cells in four lupus mouse models that encompass diverse pathogenic mechanisms and anti-nuclear antibody subtypes(Du et al., 2015); (Richard and Gilkeson, 2018) and compared them with patient intrarenal myeloid cell states. Two models, NZW/BXSB and Sle1.Yaa, feature a translocation from the X to the Y chromosome that confers an extra copy of Tlr7 in males, resulting in either early and severe or later and indolent proliferative lupus nephritis, respectively. When combined with other predisposing alleles, the additional Tlr7 copy plays a pivotal role in driving murine lupus nephritis, by inducing the production of anti-chromatin and anti-RNA antibodies in a B cell intrinsic manner, with severe proliferative lupus nephritis characterized by renal recruitment of patrolling myeloid cells(Carlin et al., 2013); (Stamatiades et al., 2016). In non-autoimmune mice 4-8 fold overexpression of Tlr7 is sufficient to drive lupus nephritis and in humans, single nucleotide polymorphisms (SNPs) associated with Tlr7 increase the risk of SLE and lupus nephritis(Wang et al., 2014). Recent work demonstrated that TLR7 gain-of-function mutations contribute to lupus pathogenesis in humans(Brown et al., 2022).

We utilized two additional spontaneous lupus nephritis strains characterized by lymphoid cell expansion but with normal Tlr7 expression levels. NZB/W mice harbor genetic loci predisposing to SLE, shared with SLE patients, resulting in intrinsic defects in lymphocyte tolerance in females(Morel and Wakeland, 1998). This strain manifests the production of anti-double-stranded DNA (dsDNA) antibodies and proliferative lupus nephritis histologically resembling human proliferative lupus nephritis(Sahu et al., 2014; Bethunaickan et al., 2014; Dall’Era et al., 2015). MRL/lpr mice feature Fas deletion in the permissive MRL genetic background, resulting in B cell and CD4/CD8 double negative (DN) T cell expansion and the production of multiple autoantibodies that drive severe proliferative and interstitial lupus nephritis. This model exhibits kidney-infiltrating DN T cells and T cells producing IFNγ or IL-17, consistent with findings in human lupus nephritis with tubulointerstitial inflammation(Li et al., 2020; Sharma et al., 2021; Arazi et al., 2019b; Crispín et al., 2008). Single cell RNA sequencing of renal myeloid cells from this strain revealed several subsets of infiltrating and resident myeloid cells including an altered resident macrophage state and expansion of an infiltrating subset with a signature of FcGR mediated phagocytosis in the MRL/lpr strain compared with the non-autoimmune MRL/MpJ control(Richoz et al., 2022).

Because LN can be highly heterogeneous, we compared the molecular and spatial landscape of the myeloid compartment in the kidneys of these four mouse models with renal biopsy samples from 155 lupus nephritis patients and 30 healthy controls. By aligning intrarenal myeloid cell transcriptional profiles across mouse and human lupus, we identified similar myeloid subsets between species and revealed comparable myeloid states associated with disease, including multiple subtypes of monocytes, macrophages, and dendritic cells. These findings provide insights into specific mouse cell subsets that mirror human lupus nephritis, paving the way for functional studies in this field.

RESULTS

Mice and humans share comparable intrarenal myeloid subsets and gene programs in lupus kidneys.

We isolated viable total hematopoietic (CD45+) or myeloid (CD11b+ or CD11c+) cells from fresh murine kidneys or peripheral blood using magnetic beads and FACS sorting from four lupus mouse models pre- or post- clinical lupus nephritis, defined by proteinuria of ≤30 mg/dl or ≥300 mg/dl for >2 weeks respectively(Sahu et al., 2014) (Fig. 1A). Fixed proteinuria was chosen as a clinical marker due to its association with global renal histologic and transcriptional changes in murine lupus nephritis(Bethunaickan et al., 2014) and poor clinical outcomes in patients(Dall’Era et al., 2015). Additionally, we processed frozen kidney biopsies collected by the AMP-RA/SLE network from 30 healthy kidney donors or 155 adult patients (≥18 years old) diagnosed with systemic lupus erythematosus (SLE) according to the European League Against Rheumatism and the American College of Rheumatology criteria(Aringer et al., 2019), with a urine:protein creatinine ratio >0.5 and a clinical indication for a percutaneous kidney biopsy (Fig. 2A). We then utilized the 10x single-cell RNA-seq platform to measure single-cell transcriptomes and applied bioinformatic filters to recover 20,299 murine and 23,819 human intrarenal myeloid cells for further analysis (Fig. S1A, Tables 1, 2).

Figure 1. Intrarenal myeloid states from 4 lupus mouse strains are revealed by single cell RNA sequencing.

(A) Single cell suspensions were prepared from kidneys harvested from 4 lupus mouse strains exhibiting pre-disease or established clinical nephritis. (B) Dot plot of markers to distinguish monocytes, macrophages, dendritic cells, and T and B-cells. The color scheme reflects z-score values, ranging from −1 (blue) to 2.3 (red). (C) Heatmap depicting scaled expression of the top 20 discriminative markers for myeloid clusters in Figure 1B. Colors correspond to z-score values, from −2.2 (blue) to 2.2 (red). (D) Individual UMAP plots depict intrarenal myeloid cells from the integrated analysis of pre-disease and lupus nephritis for each strain, with a total of 2,993 cells in each UMAP to enable comparison.

Figure 2. Comparing intrarenal myeloid states from lupus mouse strains and patients by single cell RNA sequencing.

(A) Single cell suspensions were prepared from biopsies collected from 155 patients and 30 controls (bottom). (B, C) UMAP of integrated single-cell analysis of myeloid cells colored by their respective cellular identities. (D) Scatter plots depict cluster-specific marker genes from mice (y-axis) compared to humans (x-axis), showing their expression levels (log-fold change) relative to other renal myeloid cells. Depicted to the right of each scatter plot are KEGG database annotations of the conserved genes (each with adjusted p-value <0.01). (E) Pairwise comparison of PAGA scores (transcriptional relatedness) of intrarenal myeloid clusters in mice (D) and humans (E). PAGA scores quantify the connectivity between clusters: higher scores indicate stronger connectivity or more likely cellular transitions, while lower scores suggest weaker connectivity or fewer transitions. The scores are calculated based on the proportion of edges connecting cells from different clusters in the original high-dimensional cell-cell graph.

We identified intrarenal murine myeloid cell types across four lupus mouse models by partitioning cells into 10 clusters of monocytes, macrophages, and dendritic cells, based on their differentially expressed markers (Fig. 1B-D; Table 3). To identify human intrarenal myeloid counterparts, we utilized Harmony(Korsunsky et al., 2019) to remove batch effects and to integrate and embed the mouse and human single-cell datasets into a common low-dimensional space (Fig. 2B, C; S1B). We assigned mouse cluster identities to human cells using k-nearest neighbor label transfer in the Harmony embedding and ran uniform manifold approximation and projection (UMAP) to visualize the embedding. We assessed the integration of our interspecies single-cell datasets by computing confidence scores for each human cell following label transfer (Fig. S1C, D). We observed the highest scores for DC1 and pDCs, while MDSCs showed lower scores.

To examine transcriptional overlap among equivalent myeloid cell types in mice and humans, we plotted differentially expressed gene (DEG) log-fold changes for each cell type from both species and conducted gene set enrichment analysis (Fig. 2D). NC1 monocytes from both species expressed genes that regulate the actin cytoskeleton, transendothelial migration, and FcGR mediated phagocytosis. C1 monocytes expressed gene sets associated with transendothelial migration, cell adhesion, glycolysis and HIF-1 signaling; C2 monocytes shared gene programs for oxidative phosphorylation, phagocytosis, and cholesterol homeostasis. Resident macrophages shared gene programs for phagocytosis and antigen processing and presentation. This analysis uncovered shared gene programs within matching intrarenal myeloid cell types involved in lupus nephritis. We further compared transcriptional relationships between equivalent clusters in both species using partition-based graph abstraction (PAGA), a method employed to reconstruct lineage relationships among cell states(Wolf et al., 2019). This revealed similar transcriptional strengths between equivalent clusters (Fig. 2E, F). Together, these analyses indicate that our single-cell integration accurately identifies matching intrarenal myeloid cell types in both humans and mice.

We compared the frequencies of intrarenal myeloid subsets in mice and humans prior to and after the onset of clinical lupus nephritis (Fig. 3A-D). Single-cell datasets revealed an increased frequency of non-classical and classical 2 monocytes and a concurrent decrease in resident macrophage frequency in mice with clinical nephritis compared to those without nephritis (Fig. 3B). In humans, however, the most striking difference was an increased frequency of classical 2 monocytes in LN compared with healthy biopsies (Fig. 3D). To more precisely quantify myeloid subset proportions, we devised a discriminative 16-color flow cytometry panel using scRNA-seq subset-specific markers to monitor prevalent myeloid subsets in whole mouse kidneys of pre-disease and nephritic mice (Figs. 3E, F, S2 A-C; Table 4). As a percentage of total live kidney cells, all macrophage subsets increased in frequency in nephritic compared with prenephritic mice (Fig. 3F), consistent with our prior work showing that all renal monocytes and macrophages expand in disease(Sahu et al., 2014), but to different degrees with C2, along with other monocytes (NC1 and NC2), being disproportionately more frequent during lupus-related renal inflammation and tissue injury (Fig. 3F). We then treated nephritic NZB/W mice with the standard immunosuppressants, mycophenolate mofetil and prednisone at the onset of fixed proteinuria and examined kidneys from mice in complete remission after 8 weeks (Fig. 3F). Myeloid expansion reversed, indicating that our findings were related to lupus-related inflammation rather than age.

Figure 3. Disease-Associated Cellular Shifts In Intrarenal Myeloid Cells In Mice And Humans.

UMAP of renal myeloid cells from all mouse strains (A) or humans (C) colored by disease status. Normalized bar plots of myeloid subset frequencies pre-disease or with clinical nephritis for each lupus mouse strain (B) and major histologic classes in patients with >25 myeloid cells per patient (D) from scRNA-seq droplet proportions. (E) tSNE of flow cytometry of intrarenal myeloid subsets before and after onset of clinical nephritis. NZB/W (n=3) plots contain 16,500 intrarenal myeloid cells from 3 pre-nephritic and 3 lupus nephritic mice, and both B6.Sle1Yaa plots contain 24,000 intrarenal myeloid cells from 3 pre-nephritic and 4 lupus nephritic mice. Cell are from the CD45+ CD11b+ XCR1- Ly6G- gate (Fig. S2). Ly6G+ neutrophils and Xcr1+ cDC1 cells were excluded for visualization. (F) Percentage of the indicated cell type relative to all live NZB/W kidney cells in pre-disease, lupus nephritis, or medication-induced remission as measured by flow cytometry. Each dot = 1 mouse. (Kruskal-Wallis ANOVA with Dunn’s correction for multiple comparisons: ns: p > 0.05; * p < 0.05; **: p < 0.01; ***: p < 0.001; ****: p < 0.0001).

Intrarenal resident macrophages undergo state transitions with the progression of murine lupus nephritis.

We identified a major intrarenal mouse cluster as resident macrophages (RMs) in single cell RNA seq and flow, marked by established resident macrophage indicators such as Adgre1 (F4/80), Cd81, Cd74, and C1qa(Zimmerman et al., 2019) (Fig. 1B,C; Fig. S2A, panel ix; Table 3). As expected, RMs were nearly undetectable in blood by scRNA-seq (Fig. 4A, B) or flow cytometry (Fig. S2A panel ix compared to S2C panel ix) and they showed weak transcriptional relationships with classical and non-classical monocytes in PAGA analysis (Fig. 2E), suggesting they did not originate from these cells. RMs demonstrated enrichment in MHC class II antigen presentation, cellular debris clearance via complement (C1qa, C1qb, C1qc), and Fc gamma receptor-mediated phagocytosis (Fcgr1, Fcgr2b, Fcgr3, Axl) compared to monocytes (Fig. 1C, Table 3).

Figure 4. Pre and Post-Nephritic Murine Resident Macrophages Shift Phenotypes in Advanced Disease.

(A, B). Co-clustering of myeloid cells collected from kidneys (light blue) and matched blood (red) from NZB/W (A) or NZW/BXSB (B) mice with lupus nephritis. (C) Wright-Giemsa stain of murine renal RMs collected pre-disease and in lupus nephritis from NZB/W mice. Scale bar indicates 10 microns. (D) Summary plots of Cd11b, F4/80, FcgR1, Fcgr4 mean fluorescence intensity (MFI) (*: p < 0.05) showing surface marker expression shifts in RMs after the onset of murine Disease. (E) Volcano plot of bulk RNASeq of F4/80hi/CD11bhi/CD11clo resident macrophages (RMs) isolated from perfused kidneys of 16wk pre-nephritic and 30-40w nephritic (>300mg proteinuria for > 2 weeks) NZB/W mice. Red dots represent individual genes that are significant, with an adjusted p-value (−log10) greater than 1.3 and a log2 fold-change exceeding 0.5. (F) Phagocytosis of APC-labeled ova-anti-ova immune complexes in pre and post-nephritic resident macrophages (CD11b⁺ CD11c⁻ Ly6C⁻ CD81⁺ F4/80⁺). Non-parametric t-test for 2-group comparisons. (G-I fibroblast-RM co-culture studies). (G) Gating strategy for Thy1+ fibroblasts in the co-cultures. (H, I) Percent FAP+ fibroblasts and FAP MFI in the co-cultures. Fibroblasts from the co-cultures with pre-nephritic RMs have an intermediate phenotype. (J-S) Metabolic studies. Indicated macrophages or monocytes from pre-disease or nephritic NZB/W and Sle1.Yaa mice were stained with ATP biosensor (J-N), mitotracker green (O, P) or TMRM (Q-S). (K-M) ATP MFI from the indicated myeloid subsets. Each dot represents a single mouse. Pre-disease (black) and nephritis (red) samples are shown together as there was no difference between them. In both strains, C2 and RM subsets have higher ATP MFI than C1, NC1 and NC2 subsets. (O) Mitotracker green staining of all classical vs all non-classical monocytes and RMs from NZB/W mice. Kruskal Wallis ANOVA with Dunn’s correction for multiple comparisons. (R, S) TMRM staining of all classical monocytes (R) and all non-classical monocytes and RMs (S) from pre-disease (black) and lupus nephritis (red) NZB/W kidneys. For comparisons in this figure, we used Mann Whitney non-parametric t-test for 2-group comparisons and Kruskal Wallis ANOVA with Dunn’s correction for multiple comparisons: * p<0.05; **p<0.01; ***p<0.001.

RMs changed their phenotype with disease progression.

Morphologic analysis revealed that RMs isolated from nephritic mice were larger and contained more vacuoles than those of pre-disease mice (Fig. 4C). Flow analysis revealed that nephritic RMs clustered separately from those of pre-nephritic mice (Fig. 3E), exhibiting higher expression of Adgre1 and Itgam, and lower expression of CD14 and Fcgr1 (Fig. 4D). We examined whether these changes in surface marker expression and morphology were associated with gene expression changes by comparing bulk sequencing of RMs isolated from NZB/W kidneys before and after disease onset (Fig. 4E, Table 5). Pre-nephritic RMs exhibited expression of Mmp12, Mmp13 and a set of inflammatory mediators (Ccl3, Ccl4, Ccl5, Ccl12, Ccl22, Ccl23, Cxcl2, Cxcl16), whereas nephritic RMs expressed an injury associated gene program that has been observed across different tissues and species (Apoe, Cd63, Ctsd, Lgmn, Pltp, Trem2)(Keren-Shaul et al., 2017; Fabre et al., 2023; Jaitin et al., 2019; Subramanian et al., 2024; Houben et al., 2017; Oelschlaegel et al., 2020), complement related factors (C3, C4B, C1RL) and a different set of inflammatory mediators (Ccl7, Ccl8, Ccl9, Cxcl13, Ifng, Il10, S100A4). Our analysis of DEGs and cytokine/chemokine expression from single-cell pre- and post-nephritic RMs supports these results (Fig. S3 see “RM”; Table 6). We note that the fewer genes identified in our single-cell RNA sequencing data may be attributed to inherent differences between bulk and single-cell technologies. We next assessed several RM functions in vitro, focusing on intrarenal activities relevant to lupus nephritis. Compared to pre-nephritic RMs, nephritic RMs exhibited a significant deficit in immune complex uptake (Fig. 4F) (which are known to accumulate in kidneys and drive inflammation(Flores-Mendoza et al., 2018)) without a deficit in uptake of either non-opsonized latex beads or aggregated protein (not shown). Additionally, when co-cultured with a fibroblast cell line, only nephritic RMs induced FAPa expression (Fig. 4G-I), a protease that is observed in fibrosis with known roles in remodeling of the extracellular matrix(Croft et al., 2019). To assess energy utilization, we measured ATP production (Fig. 4J-N), mitochondrial mass (Fig. 4O, P), and mitochondrial membrane potential (Fig. 4Q-S). RMs exhibited the highest ATP content and membrane potential that remained unchanged in pre and post-nephritic kidneys, indicating an alternatively activated state that was unaffected by disease progression.

To gain insight into the cellular subsets of RMs that could be driving these functions, we sub-clustered our RM single cell data into six RM subclusters and measured their frequency pre- and post-lupus nephritis (Fig. 5A, S4A; Table 7). Across all strains, RM1 frequency decreased (p<0.05), while RM0 increased in advanced disease (p<0.05) (Fig. 5B). RM1 exhibited expression of chemokines (Ccl3, Ccl4, Cxcl2) and the anti-fibrotic gene Mmp13(Ren et al., 2019) (Fig. 5C, Table 8), while RM0 expressed genes associated with a conserved injury response observed in different tissues and species (Apoe, Cd63, Ctsd, Lgmn, Pltp, Trem2)(Keren-Shaul et al., 2017; Fabre et al., 2023; Jaitin et al., 2019; Subramanian et al., 2024; Houben et al., 2017; Oelschlaegel et al., 2020). These single cell analyses were consistent with findings from bulk RNA sequencing analysis (Fig. 4E, Table 5). RM1 and RM0 displayed high transcriptional similarity by PAGA (Fig. S4B), suggesting a cellular transition. We applied SCENIC(Aibar et al., 2017; Van de Sande et al., 2020) to our single cell data to infer transcription factor (TF) activity and co-regulated gene modules driving this hypothesized differentiation of RM1 to RM0 (Fig. 5D). RM1 TFs known for regulation of cellular identity (Cebpa(Kim et al., 2024), Jun, Jund, Fos, Fosb, Irf8(Tamura et al., 2000); (Kurotaki et al., 2015)), inflammation (Creb5, Fos, Fosb, Nfkb2(Mussbacher et al., 2023)), and fibrosis (Mmp13) were decreased in RM0, whereas M2-polarization (Maf(Liu et al., 2020)) activity was increased (Fig. 5D, S5A, B). We conclude that disease-associated RM0 resident macrophages may derive from a homeostatic RM1 subset via a distinct set of TFs.

Figure 5. Murine Resident Macrophages are Composed of Subsets that Shift in Advanced Disease and Correspond to Human Clusters.

(A) Normalized bar plots comparing frequencies of mouse intrarenal RM subsets before and after onset of clinical nephritis, measured by scRNA-seq subcluster proportions in each strain. (B) Mean frequency of RM1 or RM0 subsets from all strains. Each point represents 2-4 mice of the same strain. To compare groups, we used Mann Whitney non-parametric t-test: * p<0.05. (C) Volcano plot of differentially expressed genes between RM1 and RM0 subsets from single cell RNA-seq. Red dots represent individual genes that are significant, with an adjusted p-value (−log10) greater than 1.3 and a log2 fold-change exceeding 0.25. (D) Regulon activity in RM1 or RM0 subsets, with positive Log2 Fold Change (Log2 FC) values indicating higher RM0 activity. (E-G) Mouse cell-types mapped in the human data set. (E) UMAPs visualizing RM1 and RM0 mouse cluster identities mapped in human myeloid single-cell data (top) using k-nearest neighbor label transfer or mouse (bottom). (F) Scatter plot comparing relative expression of RM1 and RM0 marker genes between mice (y-axis) and humans (x-axis). Conserved gene expression for RM1 is in the lower left quadrant, and for RM0 in the upper right quadrant. (G) Conserved gene sets enriched in RM1 versus RM0 identified by KEGG analysis.

Other RM clusters were less prevalent compared to RM0 and RM1 and did not change in frequency in advanced disease (Fig. 5A). RM2 expressed genes associated with RAGE-dependent apoptosis (S100a6)(Xia et al., 2017), phagocytosis promotion (Sirpb1c)(Hayashi et al., 2004), inhibition of kidney monocyte infiltration (Ccl6), and fibrosis prevention (Ccl9)(Hemmers et al., 2022); RM3 expressed interferon-stimulated genes, chemokines, and the Cd40 receptor (Fig. S4A, Table 7). RM4, although infrequent, differentially expressed Hpgd, known for metabolizing pro-PGE2 to an anti-inflammatory molecule(Ho et al., 2022), Stab1 which facilitates tissue repair by limiting fibrosis(Rantakari et al., 2016), and Gas6, which is anti-inflammatory during acute inflammation and pro-fibrotic in chronic conditions(Bellan et al., 2019). RM5, also rare, was enriched for Spp1, marking pro-fibrotic macrophages (Fig. S4A, Table 7)(Hoeft et al., 2023). We observed a small number of proliferating resident macrophages before and after clinical nephritis, potentially contributing to resident macrophage dynamics (Fig. S4A).

Human resident macrophages divide into RM1 and RM0 subclusters resembling those in mice, and their frequencies correlate with patient disease scores.

Using Harmony, we transferred mouse RM subcluster identities to human cells in our single-cell datasets to explore their relevance to human disease severity. Our focus centered on RM1 and RM0 due to their prominence in both species. Human RM1s co-clustered with a subset of the larger RM parent cluster observed in healthy controls and patients with lupus nephritis (Figs. 5E). RM0 displayed a similar clustering solution but also co-clustered with a subset of classical 2 monocytes (Fig. 5E), a population enriched in patients with active lupus nephritis but less prevalent in healthy controls and other kidney diseases (described below). PAGA analysis revealed a robust transcriptional relationship between human RM1 and RM0 (Fig. S4B), consistent with a transition between these subsets, analogous to observations in mice. Comparing differentially expressed genes between equivalent mouse and human RM1 and RM0 subsets revealed shared transcriptional changes during the hypothesized differentiation (Fig. 5F, Table 9). RM1 in both species expressed cytokines and chemokines (IL1B, CCL3, CCL31L1, CCL4, CCL4L2), whereas RM0 expressed injury-associated genes (CD9, SPP1, TREM2) and genes implicated in lipid and cholesterol metabolism and lysosomal digestion (Fig. 5F, G). Using SCENIC, we identified shared TFs and co-expressed gene modules in mouse and human resident macrophage subsets (Fig. S5A, B, E). RM1 TFs in both humans and mice are involved in macrophage polarization (JUNB, JUND, FOSB) and inflammation (ATF3). In contrast, RM0 TFs are associated with M2-polarization (MAF) and cellular differentiation (FLI1), while the roles of other TFs in macrophages remain less elucidated (SP2 and KLF16). In summary, our analyses uncovered transcriptional changes and potential TF regulators during the hypothesized transition from RM1 to RM0 resident macrophages in both mice and humans during the progression of lupus nephritis.

To discern the clinical relevance of RM subsets, we assessed their frequencies in human kidneys of the various histologic ISN/RPS classes (i.e. class III/IV biopsies were characterized by glomerular cellular hyperproliferation and immune complex deposition in the subendothelium, and class V biopsies were characterized by immune complex deposition at subepithelial sites) and disease severity determined by the activity and chronicity scores, reflecting the extent of tissue injury (Bajema et al., 2018). RM1 displayed a modest decrease in frequency with increasing lupus nephritis activity when all classes were considered together, but lacked correlation with chronicity (Fig. 6A). In contrast, the frequency of RM0 positively correlated with disease chronicity across all classes, a metric associated with progression to end-stage renal disease(Hsieh et al., 2011) (Fig. 6A). We conclude that RM0 was the predominant subset of resident macrophages in human kidneys affected by lupus nephritis and high chronicity regardless of disease class and exhibited an injury-associated gene signature, resembling its counterpart in mice.

Figure 6. Analogous Human Macrophage And Monocyte Subsets Correlated With Disease Indices In Patients With Lupus Nephritis.

(A) Scatter plots showing RM1 and RM0 frequencies relative to other myeloid subsets vs the histologic activity (left) or chronicity (right) index. (B) Boxplots of C2 relative to other myeloid subsets in patients grouped by histologic classes of lupus nephritis, CKD, AKI, or healthy controls. Kruskal-Wallis test: * p<0.05; ** p<0.01; ***p<0.001; ****p < 0.0001. (C) Scatter plots showing C2 frequency relative to other myeloid subsets for each histologic class as a function of histologic activity (top) or chronicity (bottom) index.

A classical 2 monocyte subset emerged in mice with clinical nephritis.

We identified two subtypes of murine classical monocytes, C1 and C2 (Fig. 1B-D). C1 monocytes (Ly6Chi/Ccr2hi) were present in both the blood and kidneys of pre-disease and clinical nephritic mice (Fig. 4A, B; Fig. 7A; Fig. S2A panel viii compared S2C panel viii). In contrast, C2 monocytes were scarce in the blood but expanded in lupus nephritis-affected kidneys as detected by flow cytometry and exhibited increased frequency across all strains in scRNA-seq (p<0.01) (Fig. 7B). Although murine C1 and C2 were transcriptionally closely related according to PAGA analysis (Fig. 2E), C2 monocytes could be distinguished from C1 monocytes in flow cytometry by their expression of Clec4n and lower expression of Ly6c2 and Ccr2 (Fig. S2A panel viii, S2B). Thus, a simple hypothesis suggests that C1 infiltrates from the blood and differentiates into C2 in the kidney. Relative to C1, C2 monocytes were enriched in genes involved in the uptake and degradation of extracellular molecules, lipid metabolism, and complement, and expressed known markers of injury-induced monocytes such as Apoe, Cd63, Cd9, Ctsa, Ctsd, Fabp5, Lipa, Lgmn, Lpl, Plin2, and Trem2 (Fig. 7C, Table 10), which participate in tissue repair and regulation of fibrosis and inflammation in multiple organs(Keren-Shaul et al., 2017; Fabre et al., 2023; Jaitin et al., 2019; Subramanian et al., 2024; Houben et al., 2017; Oelschlaegel et al., 2020). To better understand the hypothesized transition of C1 to C2, we utilized SCENIC to infer active TFs by measuring co-regulated gene modules (Fig. 7D, S5C, D). C1 TFs were known to regulate inflammation (Arid3a, Fos, Irf7, Rfx5, Klf2, Klf4, Irf1) and to promote an M2-like state (Cebpg, Cebpb), suggesting a mixed cellular phenotype. In contrast, C2 TFs were known to promote lipid metabolism (Pparg), phagocytic functions (Mitf), tissue repair (Maf), and to regulate inflammation (Etv5).

Figure 7. Murine Classical 2 Monocytes Expand In Lupus Nephritis And Map To Human Clusters.

(A-E) Mouse. (A) C1 (orange) or C2 (pink) subsets as a percent of Cd11b+ in flow cytometry from NZB/W (left) or B6.Sle1Yaa (right) kidneys (light blue) and matched blood (red) before and after the onset of clinical nephritis; 5 - 8 mice per condition. (B) The frequency of C1 or C2 during pre-disease (minus sign) or clinical nephritis (plus sign) by scRNA-seq droplet proportions. Each data point represents cells pooled from 2-4 mice of the same strain. (C) Volcano plot depicting differentially expressed genes between C1 and C2 subsets from single cell RNA-seq data. Red dots represent individual genes that are significant, with an adjusted p-value (−log10) greater than 1.3 and a log2 fold-change exceeding 0.25. (D) Regulon activity in C1 and C2 subsets, with positive Log2 Fold Change (Log2 FC) values indicating higher C2 activity. (E) Phagocytosis of APC-labeled ova-anti-ova immune complexes. (F-H) Mouse cell-types mapped in the human data set. (E) UMAPs visualizing mouse C1 and C2 cluster identities in human myeloid single-cell data using k-nearest neighbor label transfer or mouse (bottom). (G) Scatter plot comparing relative expression of C1 and C2 marker genes between mice (y-axis) and humans (x-axis). Conserved gene expression for C1 is in the lower left quadrant, and for C2 in the upper right quadrant. (H) Conserved gene sets enriched in C1 versus C2 identified by KEGG analysis. For comparisons in this figure, we used Mann Whitney non-parametric t-test for 2-group comparisons and Kruskal Wallis ANOVA with Dunn’s correction for multiple comparisons: * p<0.05; **p<0.01; ***p<0.001.

Human monocytes resemble those in mice, and their frequencies correlate with histologic class and disease score, and are present in urine.

We identified the equivalent human C1 and C2 clusters using Harmony (Fig. 7F). Relative to C1, human C2 monocytes exhibited distinct expression patterns of CD9, SPP1, APOE, FABP5, GPNMB, and TREM2 (Fig. 7G, Tables 11, 12), which are known markers associated with tissue repair, fibrosis, and inflammation in macrophages across diverse organs in both humans and mice(Keren-Shaul et al., 2017; Fabre et al., 2023; Jaitin et al., 2019; Subramanian et al., 2024; Houben et al., 2017; Oelschlaegel et al., 2020). To explore the potential transition of human C2 from C1, paralleling our observations in mice (Fig. 2E), we conducted PAGA analysis on human myeloid clusters identified by their mouse gene signatures. This analysis revealed notable transcriptional similarity between C1 and C2 (Fig. 2F), suggesting that C2 may originate from infiltrating C1 monocytes within the kidneys in humans as well. Using SCENIC, we identified shared active TFs in both mouse and human C1 and C2 subsets (Fig. S5F). C1 TFs, known for regulating immune responses (KLF2, KLF4, ATF3, FOSB), exhibited diminished activity in C2, whereas C2 TFs, responsible for lipid metabolism (PPARG), phagocytosis (MITF), M2-polarization (MAF), and anti-inflammatory processes (ETV5), displayed elevated activity (Fig. S5C, D). Kegg pathway analysis indicated activity of the glycolytic pathway in C1 and of oxidative phosphorylation in C2 (Fig. 7H).

To assess whether these gene expression differences between C1 and C2 were associated with functional differences, we characterized mouse renal subsets given their overlapping transcriptional profiles with humans and because they could be easily obtained. Because there were few C2 macrophages in prenephritic mice we used macrophages high for Trem2 surface expression in prenephritic and nephritic mice to compare IC uptake (Fig. 7E) (Trem2 is part of the injury associated gene program). Trem2hi macrophages from nephritic mice showed increased uptake of ICs compared to C1s collected from matched mice with a trend towards increased uptake compared to pre-nephritic C2s (p = 0.065), although IC uptake was significantly lower compared to RMs (Fig. 4F). To evaluate energy usage, we analyzed ATP generation (Fig. 4J-N), mitochondrial content (Fig. 4O, P), and mitochondrial membrane potential (Fig. 4Q-S). While there was no difference in mitochondrial mass between C1 and C2, C2 exhibited higher membrane potential and ATP content, similar to RMs, indicating an alternatively activated state in nephritic kidneys. In sum, our analysis suggests that human C2s originate from C1 precursors or NC1, but does not exclude other potential sources. Focused mouse studies of comparable cell types indicated that these C2 cells adopt an alternatively activated phagocytic state, potentially contributing to increased IC uptake.

To assess the clinical relevance of C2, we measured its frequency across histologic classes and disease severity. Our findings revealed its enrichment in class IV kidney biopsies and, to a lesser extent, in class III biopsies, regardless of the presence of class V lesions. In contrast, C2 frequency was lower in pure class V, acute or chronic hypertensive or diabetic kidney disease, or control (Fig. 6B). Notably, C2 frequency showed a positive correlation with lupus nephritis activity across all histologic classes, but not with chronicity (Fig. 6C).

Non-classical monocytes comprise two subtypes and were associated with disease in mice but not humans.

Two intrarenal mouse clusters, NC1 and NC2, were identified as non-classical monocytes based on distinct gene expression patterns, characterized by low Ly6c2 and Ccr2 levels, and elevated Nr4a1, Cebpb, and Itgal expression profiles(Kuriakose et al., 2019); (Tamura et al., 2017) (Fig. 1B, C, Tables 3, 13). Both clusters expanded in the blood and kidneys of mice with clinical lupus nephritis (Fig. 8A), with statistical significance observed for both NZB/W and Sle1.Yaa. NC1 displayed a strong correlation with murine C2 and C1 via PAGA analysis (Fig. 2E), consistent with the established transition from classical to non-classical monocytes(Mildner et al., 2017). By contrast, NC2 demonstrated comparable relations to NC1, C2, and resident macrophages, raising questions regarding its origin. A cell type akin to NC2 was identified in the spleens of Sle1.Yaa mice(Akama-Garren and Carroll, 2022), yet its functional role remains unclear. Comparative analysis between NC1 and NC2, revealed that NC1 cells were enriched for gene sets associated with phagocytosis, oxidative phosphorylation, and C-type lectin receptor signaling, while NC2 exhibited heightened expression of programs linked to cell adhesion and TGF-beta signaling, Fc receptors (Fcgr4, Fcrl5), and Nr1h3, a transcription factor known to suppress inflammatory genes and enhance clearance of apoptotic cells(Joseph et al., 2003); (A-Gonzalez et al., 2009) (Fig. 8B, C). Because NC2 was identified as a rare population in humans, we focused on NC1 transcriptional shifts in pre and post-nephritic disease. KEGG analysis revealed that nephritic NC1 up-regulated genes associated with antigen presentation (H2-Aa, H2-Ab1, H2-Eb1, H2-DMa, H2-DMb1), complement components (C1qa, C1qb, C1qc), and various cytokines and chemokines (Ccl9, Ccr2, Ccl12, Ccr5, Cxcl14, Ccl7, Ccl5, Pf4, Ccl6, Cxcl13, Cxcr4, Ccl8, Il10rb, Ccr1, Ccl2, Ccl4, Cx3cr1) (Fig. S3B, C; Table 14). Nephritic NC1 downregulated genes associated with transendothelial migration (Itga4, Itgal, Cybb, Itgb1, Itgb2, Actg1) and TNF signaling (Tnfrsf1b, Bcl3, Traf1, Socs3, Tnf, Nfkb1). Our analysis suggests that pre-nephritic NC1 cells infiltrate from the blood and initiate an inflammatory response.

Figure 8. Cellular and Transcriptional Characterization of Murine Non-Classical Monocytes (NC1 and NC2) and Their Negative Association with Disease Activity and Chronicity in Lupus Nephritis.

(A) Percentage of Cd11b+ cells identified in flow cytometry as NC1 or NC2 from the indicated compartment from NZB/W or B6.Sle1.Yaa mice pre-disease or with lupus nephritis. Cells were collected from CD45 enriched kidney cell suspensions (light blue) and matched blood (red). 5 - 8 mice per condition. *p < 0.05; **p< 0.01; ***p<0.001; ****p<0.0001). (B) Volcano plot depicting differentially expressed genes between NC1 and NC2 subsets. Red dots represent individual genes that are significant, with an adjusted p-value (−log10) greater than 1.3 and a log2 fold-change exceeding 0.25. (C) Gene sets enriched in NC1 or NC2 were identified using KEGG analysis and selected genes are listed. (E) Scatter plots showing NC1 frequency relative to other myeloid subsets for all histologic classes vs. the histologic activity (A) or chronicity (B) index.

The equivalent human NC1 cluster demonstrated a strong correlation with C2 and C1, akin to mice, according to PAGA analysis (Fig. 2E, F). However, a species-specific distinction was noted: human renal NC1 frequency dropped with disease compared to mice, where it was increased in mice (compare trends for NC1 in Fig. 8E to Fig. 3B), reflecting the robust expansion of this NC1 in the blood during active disease in mice (Fig. 8A). In contrast, in humans, renal NC1 frequency negatively correlated with both activity and chronicity tissue damage indices across all histologic classes (Fig. 8E, F) suggesting it is less expanded than the other renal myeloid subsets in humans.

Dendritic and myeloid-derived suppressor cells exhibit relatively small changes with disease progression.

We delineated four discrete populations of intrarenal dendritic cells, which exhibited low prevalence both before and after kidney injury across the four lupus mouse models (Fig. 3B). The most prevalent population, cDC2, demonstrated transcriptional parallels with previously characterized cDC2b cells(Brown et al., 2019) (Fig. 1C). Identification of cDC1s was based on the expression of Clec9a, Xcr1, and Tlr3, and they exhibited functional gene sets associated with antigen processing and MHC class II presentation. Ccr7+ DCs were likely representative of mreg-DCs(Maier et al., 2020) and expressed genes typical of tissue migratory cDCs, potentially facilitating transit to proximal lymph nodes(Miller et al., 2012). Plasmacytoid DCs, characterized by the expression of Siglech, Tcf4, and Il7r(Dress et al., 2019), were among the least prevalent populations across all four strains (Fig. 3B). Additionally, a small cluster expressing genes associated with granulocytic myeloid-derived suppressor cells was identified(Veglia et al., 2021); (Youn et al., 2012) (Fig. 1B-D). Notably, all DC subsets exhibited relatively minor transcriptional changes from early to late disease stages, consistent with the short renal half-life of these cells(Bethunaickan et al., 2011). Of note, 3 isoforms of CD209 were downregulated in nephritic mice consistent with exposure to inflammatory cytokines in advanced disease(Kämmerer et al., 2003; Relloso et al., 2002) (Fig. S3A, Table 6).

In-situ localization of monocytes and macrophages in mouse kidneys.

Our primary findings demonstrate the association of RM subsets and C2 monocytes with lupus nephritis disease severity in both murine and human subjects. To elucidate the spatial distribution of these and other myeloid subsets within the kidney, and to gain insight into their functional localization, we employed Xenium spatial transcriptomics to map their positions in formalin-fixed paraffin-embedded kidney sections (Fig. 9). Custom probe sets were designed targeting myeloid cluster-specific markers derived from our scRNA-seq datasets, in conjunction with the manufacturer’s multi-tissue panel (mouse panel = 467 genes, human panel = 477 genes) (Tables 15, 16, 17).

Figure 9. In Situ Localization Of Myeloid States In Human And Mouse Kidney Sections.

(A-J) This series presents images of kidney sections and myeloid states within renal compartments from mice (A-D) and humans (H-JK) with lupus nephritis. All panels are stained with hematoxylin and eosin (H&E), highlighting various subsets of monocytes and macrophages (C2, C1, NC1, NC2, RM) as well as dendritic cells (Ccr7, DC1, DC2, pDC). The Activity Index (AI) and Chronicity Index (CI) are noted for humans. Scale bars indicate measurements of 100 microns. (E-M) These panels quantify myeloid subsets by compartment in mice with lupus nephritis (E. NZB/W, F Sle1.Yaa) and pre-nephritic control kidneys (G), as well as in patients with lupus nephritis and controls. The left panel displays the proportion of each indicated cell type within renal compartments across all samples. The right side presents the -log of the q-value, which indicates either depletion (blue) or enrichment (red) of the respective cell type for each renal compartment. Asterisks mark q-value <0.05. (N) Areas of renal compartments that were analyzed in the spatial assessment. Abbreviations: TI = tubulointerstitial; Periglom. = Periglomerular. Definitions: “Glomerulus, inside” = area within annotated glomerular border. “Glomerulus, border” = area that includes −/+ 20 microns from annotated border; “Periglomerular inflammation” = area adjacent to at least 25% of annotated glomerular border containing >50% immune cells. “Tubulointerstitial inflammatory aggregate” = non-adjacent area outside of glomerular border with >10 contiguous immune cells.

Murine resident macrophages were mainly found in the tubulointerstitium, forming rings around nephritic glomeruli and dispersing throughout the interstitium in mice (Figs. 9A-G), which aligns with previous studies(Richoz et al., 2022); (Schiffer et al., 2008); (Bethunaickan et al., 2012). Similarly, human RMs were also predominantly localized in the tubulointerstitium (Figs. 9H-M). While human RMs visually appeared to form rings around nephritic glomeruli in some cases (Fig. 9H.ii) this finding was not statistically significant (Fig. 9L). We hypothesize that these periglomerular RMs represent a distinct human subset that requires further characterization. In humans, C2 monocytes exhibited the highest density inside nephritic glomeruli (Fig. 9L, M), while in nephritic mice, they were found inside glomeruli in NZB/W and also within periglomerular infiltrates (Fig. 9E, F). Additionally, C2 monocytes were enriched within immune aggregates in Sle1.Yaa tubulointerstitial inflammation, suggesting strain-specific differences in C2 localization (Fig. 9E, F). Both NC1 and NC2 were primarily restricted to glomeruli and the glomerular border in control murine kidneys and in advanced disease, with rare occurrences within dense infiltrates and minimal presence in the tubulointerstitium and medulla. Human DC subsets were found in the TI, localizing to the parenchyma. In mice, DC subsets localized to the parenchyma, including inflammatory aggregates in the interstitium and periglomerular regions. Our data underscore that transcriptionally analogous myeloid subsets occupy comparable renal regions in lupus nephritis-affected kidneys across species but with some differences between strains.

Detecting myeloid subsets in urine from patients with active LN and identifying a C2 gene signature for a potential biomarker.

Previously, we showed that myeloid cells were the most frequent immune cell in urine from patients with active lupus nephritis(Arazi et al., 2019a). Here, we assessed which myeloid states based on our updated analysis of 155 patients were present in the urine of these unmatched patients from our prior work. We restricted our analysis to urinary myeloid populations and assigned urine cells to the myeloid kidney cluster from our data set with the highest gene module score, reflecting the average of the top 10 DEGs in the myeloid cluster. Our analysis showed that urine samples had a higher frequency of cells expressing a C2 gene signature, and to a lesser extent, RMs (Fig. 10A). Finally, to identify the most effective genes for classifying C2, we performed a Receiver Operating Characteristic (ROC) analysis. Notably, FABP5, CTSD, CSTB, VIM, CTSB, and SPP1 were among the highest Area Under the Curve (AUC) values (>= 0.80) (Fig. 10B), suggesting they would be relatively strong predictors for identifying C2 that could facilitate further applications in diagnostics and therapeutics. In contrast, top genes for RM0 (C1QC, C1QA, A2M, C1QB) and RM1 (CD74, ITM2B) were below 0.70, indicating a more limited efficacy in distinguishing between the RM subtypes.

Figure 10. Identification of C2- and RM-Like Cells in Urine from Patients with Active Lupus Nephritis and AUC Analysis of Top Genes for C2 Detection In SItu.

(A) the relative frequency of human myeloid cell clusters, specifically C2- and RM-like cells, identified in urine samples from patients with active lupus nephritis using mouse gene signatures. A total of 577 high-quality cells, obtained from 8 patients, were included in the analysis following stringent filtering criteria(Arazi et al., 2019a). (B) ROC curve analysis demonstrating the performance of top AUC markers in differentiating human C2 myeloid cells from other myeloid cell types. The curves illustrate the sensitivity and specificity of each marker, with the area under the curve (AUC) providing a quantitative measure of classification accuracy. (C) Schematic representation depicting disease associated changes in monocytes and macrophages in human LN. (C, top i,ii). At high activity levels, CD14 and CD16 monocytes infiltrate the glomerulus, where they transform into Classical 2 Disease-Associated Macrophages in response to local factors related to glomerular injury. These macrophages may display increased immune complex uptake and an alternatively activated phenotype. (C, bottom, iii, iv). At high chronicity levels, tubulointerstitial RM1 converts to RM0 Disease-Associated Macrophages that may acquire the capacity to activate fibroblasts, but exhibit diminished immune complex uptake. Both glomerular (C2) and tubular (RM1) Disease-Associated Macrophages express an injury-associated gene program composed of Trem2, Spp1, Gpnmb, and Cd9 among other genes (see text).

DISCUSSION

This is the most extensive study to date of intrarenal myeloid states in lupus nephritis. This atlas is based on 4 mouse models of lupus nephritis with different genetic backgrounds and pathogenic drivers at preclinical and clinical stages of disease and 155 adult patients with diverse ethnic/racial backgrounds with different stages and histologic classes of disease. It encompasses cellular definitions that are mapped between mice and humans, localization to renal microenvironments, inferences about the origin of cellular states, and functional insights into the predominant disease associated macrophage (D-Mac) populations across species: classical 2 (C2) monocytes and resident macrophage (RM0) states. D-Macs expressed a conserved set of injury-associated genes as well as programs for debris clearance, lipid metabolism, and tissue repair that correlate with advanced disease, pointing to conserved functional pathways associated with clinically relevant kidney damage in humans. This atlas can be used as a resource and a starting point for mechanistic and in vivo studies.

C2 monocytes in humans were strongly associated with proliferative lupus nephritis (the most severe form of disease) and expanded with disease progression in kidneys from all 4 lupus mouse models. These cells were enriched inside glomeruli in humans and at the glomerular border and in periglomerular areas in both humans and mice(Chen et al., 2020). We hypothesize that C2 monocytes differentiate from infiltrating C1 and/or NC1 monocytes (Fig. 10C), based on our PAGA analysis and their low representation in the blood in mice, while recognizing that PAGA analysis only provides a starting hypothesis but does not prove a lineage relationship between any cell types. Differentiation of C2 may be induced by distinct types of kidney injury that are characteristic of class III/IV glomeruli, like hyper-proliferative glomerular cells, Fc receptor mediated activation by glomerular immune complexes (deposited between the vascular endothelium and the basement membrane)(Bajema et al., 2018); (Bergtold et al., 2006), and exposure to DAMPs from injured cells or circulating inflammatory mediators. These extracellular signals could promote tissue infiltration(Akhmanova et al., 2022) and/or provide differentiation cues that include a switch in metabolic profile. Interstitial signals may also induce C2 differentiation. These signals could include tissue hypoxia (inducing HIF1a expression), a consequence of glomerular injury mediated by either proliferative or membranous glomerular disease. Of note, cells similar to C2 monocytes have been derived in vitro from human CD14+ monocytes in response to combination GM-CSF, IL-17A, and TGF-B1(Fabre et al., 2023), or to pathogenic lipids(Do et al., 2022), and after 72 hours of culture (possibly from apoptotic cells in the cultures)(Buonomo et al., 2022). Transcription factors and regulatory gene networks expressed in both species (ATF3, ETV5, KLF2 , KLF4, FOSB, PPARG, MITF, MAF) may regulate the transition to C2. The unique cytokine profile expressed by these cells in mice suggests that C2 may attract specific immune cell types to nephritic periglomerular and glomerular sites. Thus, the C2 phenotype in LN represents a conserved injury response that can be induced by unique factors from the glomerulus or the interstitium.

C2 monocytes in both species expressed functional gene sets for phagocytosis, lysosomal functions, and cholesterol metabolism as well as an injury-associated gene set (CD9, SPP1, APOE, FABP5, GPNMB, PLA2G7 and TREM2). Monocytes with similar functional pathways and genes were previously reported in association with fibrosis and tissue inflammation in models of injury in multiple organs (but not in a predictive model of macrophage activation derived using acute infectious stimuli and applied to 12 mouse tissues(Sanin et al., 2022)). C2-like macrophages localized to inflamed and scarred regions in cirrhotic livers (scar associated macrophages - SAMs)(Ramachandran et al., 2019), were found in injured lung(Morse et al., 2019), atherosclerotic arteries(Cochain et al., 2018) and inflamed brain (damage associated macrophages - DAMs)(Keren-Shaul et al., 2017), and arose in adipose tissue from mice fed a high fat diet (lipid associated macrophages - LAMs)(Jaitin et al., 2019). Our data suggest that we have also identified the renal counterpart of these cells in humans and mice. Similar kidney TREM2+ macrophages have recently been identified in kidneys and appear to play a protective role. Deficiency of Trem2 in Sle1.Yaa led to elevated levels of periglomerular αSMA and disrupted the normal architecture of periglomerular capillaries(Biniaris-Georgallis et al., 2024). Similarly, in a mouse model of diabetes, Trem2-deficient mice fed a high-fat diet developed worsening kidney damage, suggesting a protective role(Ayshwarya Subramanian et al., 2021). Trem2-deficient SAMs in the liver were pro-fibrotic and pro-inflammatory(Hendrikx et al., 2022) and Trem2-deficient mice failed to generate LAMs, resulting in defective adipocyte clearance and obesity(Jaitin et al., 2019). Finally, studies in cancer demonstrated that loss of Trem2 leads to enhanced anti-tumor immunity(Molgora et al., 2020; Binnewies et al., 2021). Spp1-deficiency promoted liver fibrosis in NASH(Han et al., 2023) and Fabp5-deficiency induced M2 polarization(Hou et al., 2022) through alterations in fatty acid metabolism. Gpnmb reduced macrophage inflammatory functions in obese mice(Prabata et al., 2021). The role of Pla2g7 is less clear since it has been associated with both pro- and anti-inflammatory macrophage functions(Buonomo et al., 2022); (Spadaro et al., 2022). Collectively, these injury-associated genes appear to modulate fibrosis and inflammation in tissue- and context-specific manners. In the human kidney C2 cluster, expression of the injury-associated genes is heterogeneous, suggesting the presence of subclusters that could differentially modulate repair and fibrosis(Richoz et al., 2022; Fabre et al., 2023). Acquisition of additional gene programs such the expression of inflammatory chemokines as observed in C2 and RM from nephritic mice (Fig. S3) may differentiate those subsets that mediate fibrosis rather than repair(King et al., 2024).

Our finding that C2 monocytes were more abundant in biopsies with class III and IV lesions compared to class V and other forms of kidney disease may have clinical implications. First, because C2 are the major immune and myeloid subset in the urine of LN patients(Arazi et al., 2019a), C2 may be useful to distinguish between patients with class III/IV lupus nephritis from pure class V and other forms of acute and chronic kidney disease. Second, the predominance of C2 cells in urine suggests that monitoring urine C2 levels could be a viable strategy for assessing disease classification or the treatment response of LN patients with class III and IV lesions. To enable future studies on this topic, we identified several genes that are strong predictors for identifying C2 (several of which were part of the injury-associated program). Proteomic analyses of urine from patients in the AMP cohort revealed that the decline of urinary macrophage markers CD163 and CD206 outperformed the clinical standard (proteinuria) at predicting 1-year response to therapy(Fava et al., 2024). Further investigations will be required to establish the relationship and the specificity of urine C2 cells and associated proteins to disease activity and therapeutic responses.

Mouse and human kidney resident macrophages (RM) are typically positioned in the tubular interstitium between the tubules and peritubular capillaries, where they monitor trans-endothelial transport and may initiate an inflammatory response shortly after ingesting circulating immune complexes(Stamatiades et al., 2016), particularly those containing endosomal TLR ligands. Previous spatial analyses in a small number of pediatric LN patients positioned infiltrating CD16+ and C2-like macrophages in the glomeruli(Danaher et al., 2024). Our findings confirm and extend this data by showing the positions of C2, NC1 and NC2 macrophages in glomeruli (C2, NC1 and NC2) and/or periglomerular (C2) lesions, while C1 macrophages, surprisingly, are located in the interstitium. The absence of C1 macrophages in the glomeruli may reflect their transition to other phenotypes once they enter the inflammatory environment. Our data reveal that, in contrast to C2, the human RM parent cluster frequency did not differ across histologic classes of lupus nephritis. Instead, the RM phenotype changed with disease progression. In mice, as the disease advanced, the ability of RMs to uptake ICs diminished while they retained the ability to uptake aggregated protein and acquired the capacity to induce fibroblast FAPa expression, a protease associated with fibrosis(Fitzgerald and Weiner, 2020). Furthermore, murine RMs in advanced disease expressed injury-associated genes and a different cytokine set from the prenephritic state. These disease-related changes suggest a diminished role in IC clearance, alongside emerging functions in tissue remodeling and a shift in the profile of immune cells located in extraglomerular sites. Our subclustering analysis revealed that the RM parent cluster was dominated by RM1 and RM0 sub clusters in both species. In mice, RM1 frequency shifted toward RM0 with disease progression, likely accounting for the functional differences. We observed a similar shift in humans that was positively associated with the NIH chronicity index that reflects inflammatory and fibrotic changes in the tubulointerstitium and glomeruli and portends poor clinical outcomes(Hsieh et al., 2011) (Fig. 10C). We acknowledge that RM1/RM0 frequencies were not significantly different in lupus nephritis patients and controls. This is likely due to low numbers of total myeloid cells, as well as low numbers of infiltrating myeloid cells in most healthy controls in our cohort, or modulation of the RM0 phenotype with treatment(Bethunaickan et al., 2011).

RM0 could originate in response to ingestion of immune complexes, tissue debris, urine proteinuria/particulates(He et al., 2024) or exposure to DAMPs released as a result of chronic local or systemic inflammation. RM0 could also derive from local cellular proliferation of the small RM population in mice that expressed Mki67(Jenkins et al., 2011; Murphy and Tall, 2014) but we did not observe this in humans. Blood precursors could also enter inflamed kidneys and differentiate to resident macrophages during kidney injury as reported in parabiosis studies(Salei et al., 2020). RM0 expressed gene programs for lysosomal functions, cholesterol metabolism and several genes from the C2 injury-associated gene set. Thus, similar, although not identical, functional gene programs and injury-associated genes were present in both resident macrophages and infiltrating C2 monocytes in lupus kidneys in both species. This is not without precedent, as prior work identified resident microglia in murine and human Alzheimer’s(Keren-Shaul et al., 2017) and infiltrating monocytes in human liver fibrosis(Ramachandran et al., 2019) that express similar genes.

Resident macrophage (RM1) and infiltrating monocyte (C1) (states that were found before the onset of nephritis in mice or associated with mild lupus nephritis in humans) shared a set of transcription factors with overlapping active regulatory gene modules (JUNB, JUND, ATF3, FOSB, CREB5). Activities of these TFs were reduced in RM0 and C2 cell states while MAF increased. Interestingly, both RM0 and C2 share functional gene sets across species that are related to lipid metabolism and phagocytosis. Together, our data highlight populations of resident macrophages and infiltrating monocytes that converge on a similar cellular state in severe lupus nephritis, but localize to extraglomerular or intraglomerular sites respectively.

Non-classical (NC) 1 monocytes are known to patrol endothelial lumens in glomeruli(Carlin et al., 2013). In our human cohort, NCs were present in kidney biopsies from healthy controls and those with low activity or low chronicity. Interestingly, relative NC1 frequency decreased with increasing activity and chronicity indices, possibly due to differentiation into classical 2 monocytes based on our PAGA analyses. In contrast, mouse intrarenal NC1s increased with disease progression; this likely reflects their robust peripheral expansion, a feature of active disease that is not observed in humans(Barrera García et al., 2016). NC1 from both species expressed genes that may contribute to lupus nephritis pathogenesis. ITGAL promotes homeostatic patrolling behavior on endothelium, CX3CR1 interacts with endothelial CX3CL1 in the presence of a nucleic acid danger signal, and CCL3 that may recruit tissue-damaging neutrophils in both species. In knockout studies, these genes were critical for non-classical monocytes to accumulate in the glomeruli where exposure to a TLR7 agonist increases their endothelial retention and triggers an inflammatory cascade that damages and facilitates removal of endothelial cells(Carlin et al., 2013); (Kuriakose et al., 2019); (Cros et al., 2010). Thus, conserved gene expression suggests that mechanisms first identified in mouse NC likely impact human lupus nephritis. Changes in NC1 gene expression with advanced disease reflect fixed recruitment with a decrease in adhesion molecules and chemokine signaling pathways and acquisition of antigen presentation and pro-inflammatory profiles. We also identified in mice (but not humans) a distinct NC2 monocyte that localized to glomeruli characterized by expression of Itgal and Nr4a1 like NC1 but with additional pro- and anti-inflammatory gene sets. The role of NC2 cells is still unknown.

There are several notable limitations to our study.

First, mapping mouse cell identities to human cells could be imperfect (as we observed in the confidence scores we calculated for each cell), and there may be additional subsets of monocytes and macrophages in humans that we did not explore here. Second, our analysis of cell subset relationships (using PAGA) infers differentiation pathways but only lineage studies can prove clonal relationships. Third, while we identify conserved disease-associated myeloid subsets, we have not yet developed the tools needed to study the impact of individual subsets in lupus nephritis disease in mice. Fourth, our study included single biopsies from each patient regardless of treatment. Therefore, we could not compare timepoints in the same patient nor isolate the impact of different treatments across patients. This can be done in the mouse models. Lastly, human biopsies were primarily collected from the kidney cortex while mouse data were generated from the whole kidney. Nevertheless, our analyses identified comparable intrarenal myeloid subsets with conserved alterations in classical monocytes and resident macrophages associated with advanced disease across species.

By comparing intrarenal myeloid populations in mice and humans with lupus nephritis we identified congruent myeloid cell types and pathways, including convergent programs for debris clearance, lipid metabolism, and tissue repair in resident macrophages and classical 2 monocytes that were associated with advanced disease in both species. In both species, RMs localized to the tubulointerstitium plus the periglomerular areas, while C2 cells were found near injured glomeruli (and within glomeruli in humans), suggesting that the cells play a similar role in the lupus renal microenvironment across species. We also identified important interspecies differences, including the abundance of NC1 in nephritic murine kidneys, which is decreased in humans, and the presence of NC2 in murine kidneys. Our findings provide a basis for testing mechanistic hypotheses derived from human tissues in mouse models, and a map to develop tools that manipulate these subsets and their genes to study the role of human risk variants in their function, determine their lineage relationships and discover their impact on disease and response to therapies.

MATERIALS AND METHODS

Mouse colonies, myeloid cell isolation, and single cell genomic library generation

Mice.

Male mice from Tlr7-overexpression strains bearing the Yaa allele (B6.Sle1.Yaa and NZW/BXSB) and female mice from lymphoid-expansion strains (NZB/W and MRL/lpr) were monitored clinically according to established protocols(Schiffer et al., 2003; Mihara et al., 2000). Briefly, urine samples were collected every other week and tested for proteinuria using dipstick analysis (Multistick; Fisher Scientific), while blood urea nitrogen (BUN) levels were assessed via blood sampling. Mice were considered to have established clinical nephritis when they exhibited fixed proteinuria >300mg/dl for 2 weeks. Mice with terminal renal failure (elevated BUN levels exceeding 30 mg/dl) were excluded. For remission studies, NZB/W mice with new onset clinical nephritis (proteinuria >300mg/dl on 2 occasions >24 hours apart) were treated with mycophenolic acid (200mg/kg for 4 weeks followed by 70mg/kg for 4 weeks) and prednisone (1mg/kg for 4 weeks followed by 0.5mg/kg for 4 weeks) given in the chow and then euthanized. Remission was defined as proteinuria ≤30mg/dl for at least 2 weeks by the end of the 8 week experiment. All experimental procedures were conducted in accordance with the guidelines approved by the Institutional Animal Care and Use Committee (IACUC) of the Feinstein Institutes for Medical Research or Broad Institute of Harvard and MIT.

Murine kidney dissociation and single cell suspension.

Under terminal anesthesia, mice were perfused by gentle intracardiac injection of 10 ml prewarmed (37°C) PBS 1x and kidneys were harvested. To obtain single cell suspensions, kidneys were then incubated for 12 minutes in DMEM containing 1 mg/ml Collagenase D (Roche), 100U/ml DNAse I (Sigma), 0.25mg/ml Liberase (Invitrogen) at 37°C. Tissues were gently dissociated using a pipette and then passed through a 70 μm cell strainer (BD). This single cell suspension was finally enriched for myeloid cells or total CD45+ cells using Mouse anti-CD11b and anti-CD11c or anti-CD45 MicroBeads (Miltenyi) and LS magnetic Columns (Miltenyi) for flow cytometry or FACS sorting.

Murine blood processing.

Prior to termination, whole blood from mice was collected in heparinized tubes and a 10:1 ratio of 1X Pharmalyse buffer (BD) was added for 10 minutes at room temperature. PBMCs in lysed blood were then washed twice with sterile PBS and placed in FACS buffer (PBS containing 2% FCS) for downstream applications.

Flow cytometric analysis and sorting.

For flow cytometric analysis, PBMCs or isolated kidney cells were blocked with normal rat serum and incubated with fluorescent antibodies in Brilliant stain buffer (BD). Mitochondrial stains were performed as previously described(Raparia et al., 2023). Table 2 shows a list of fluorescent antibodies and cell markers used for these experiments. Data was analyzed using FloJo. For 10x, PBMCs and isolated kidney cells were blocked with Fc block (BD) then incubated with PE-anti-CD45 and live dead stain. Cells were sorted based on size to exclude doublets and cellular debris, and live-dead stain to exclude dead cells followed by positive selection for CD45+ cells. Sorting was performed on BD FACS Aria II.

Murine single cell RNA-seq library preparation and sequencing.

Single cell suspensions of myeloid or immune cells from the kidneys or blood from 2 to 4 age- and strain-matched mice with preclinical or clinical nephritis were pooled and washed in sterile PBS. Up to 10,000 live cells were loaded into separate 10x channels for single cell RNAseq on the Chromium platform (10x Genomics). DNA amplification and library construction were carried out according to the manufacturer’s instructions. The purified libraries were quantified and sequenced according to manufacturer’s guidelines using Nexteq or Novaseq S1 platforms (Illumina).

Human kidney biopsy collection, dissociation, and single cell genomic library generation

Biopsy collection.

As part of the Accelerating Medicines Partnership (AMP) RA/SLE Phase 2 consortium patients >16 years old undergoing a clinically indicated kidney biopsy to evaluate proteinuria (urine protein to creatinine ratio >0.5) were enrolled if they met sufficient criteria of systemic lupus erythematosus diagnosis based on the revised American College of Rheumatology or the Systemic Lupus Erythematosus International Collaborating Clinics classification criteria(Hochberg, 1997); (Petri et al., 2012). Patients were excluded if they had a history of kidney transplant, rituximab within 6 months of biopsy, or were pregnant. All patients provided written informed consent. As part of the Kidney Precision Medicine Precision Medicine Project, patients >18 years old undergoing a voluntary kidney biopsy were included if they met the criteria for acute or chronic injury with a history of hypertension and/or type 2 diabetes as previously described(Lake et al., 2023).

Histologic scoring of human lupus nephritis kidney sections.

Scoring was performed centrally by two board-certified pathologists (JH and DD). The NIH “Activity” index represents ongoing kidney damage that is scored by the extent of endocapillary hypercellularity, karyorrhexis/neutrophils, fibrinoid necrosis, subendothelial deposits, luminal thrombus, cellular and fibrocellular crescent formation, and interstitial inflammation(Bajema et al., 2018). “Chronicity” represents kidney damage that is often irreversible and is scored by the extent of sclerosis in the glomeruli, fibrotic crescents, tubular atrophy, and interstitial fibrosis.

Human kidney biopsy dissociation and single cell suspension.

Human kidney biopsies were cryopreserved, thawed and dissociated as described(Arazi et al., 2019a), with modifications. Briefly, samples were dissociated enzymatically with 0.5mg/mL Liberase TL (Roche) in DMEM/F12 (Corning) for 12 minutes at 37C, and then mechanically with a cell strainer pestle (CELLTREAT) on a 70 um strainer (Miltenyi). Cells were washed with cold RPMI (ThermoFisher) supplemented with 10 mM HEPES and 0.04% BSA (Millipore Sigma) and resuspended in 50uL of the same medium. The samples were then filtered through a 40 um strainer (Pluriselect), and after counting by Trypan exclusion, up to 10,000 cells were loaded onto Chromium microfluidic chip for single cell transcriptomics profiling using the 3’ V3 kit (10x Genomics). Sequencing libraries were produced according to the manufacturer’s instructions. Prior to sequencing, DNA fragments derived from mitochondrial transcripts were digested using Depletion of Abundant Sequences by Hybridization (DASH)(Gu et al., 2016), with modifications to optimize for 3’ 10x libraries. Following DASH, libraries were sequenced on a Novaseq S4 (Illumina) according to the manufacturer’s guidelines.

Bioinformatic Analysis

Processing of mouse scRNA-seq data.

Single cell suspensions of intrarenal or peripheral myeloid or immune cells from 2 to 4 age- and strain-matched mice with preclinical or clinical nephritis were prepared using 10x 3’ or 5’ chemistry. In total, the mouse data set included 20,299 myeloid cells after quality control (described below). Fastq files were aligned to the mm10 genome reference with Cellranger v6.0.1 All downstream analysis was performed using R v4.1 and Seurat v.4. The B6.Sle1.Yaa and NZBW strains were each multiplexed with four samples and required demultiplexing with Seurat’s HTODemux function. In each case, cells assigned to the one blood sample were removed, leaving only cells from three kidney samples. Cells classified as inter-sample doublets were also removed. We next performed the following analysis steps separately on all four mouse strains: (i) Ambient RNA was estimated and removed using the R package Soupx v.1.6. (ii) Cells with <500 or >5,000 genes or >5% mitochondrial RNA expression were removed. (iii) A standard Seurat pipeline was used with default parameters for these steps: NormalizeData, FindVariableFeatures, ScaleData, RunPCA, FindNeighbors, FindClusters and FindAllMarkers. (iv) Clusters were manually inspected for various markers and quality metrics, and several clusters representing low quality, non-myeloid and proliferating cells and doublets were removed. Seurat objects from each of the four strains were then merged into one object, to which the standard Seurat pipeline was applied with the following modifications:

1. Rather than calculating variable features across the four strains, which would overemphasize differences between strains, we used this method:

   Step 1: Calculate variable features on each strain subset separately, and rather than returning the top 2000, return all.

   Step 2: Rank each of the four lists from most variable (rank = 1) to least variable (rank = # of features).

   Step 3: For each feature, calculate the average of the 4 strain ranks.

   Step 4: Take the top 2001 features with lowest average rank.

2. After performing PCA, we performed batch correction with the R package Harmony v0.1.1(Korsunsky et al., 2019), harmonizing on strain with theta=2. Subsequent steps (FindNeighbors and RunUMAP) used Harmony components instead of principal components.

Finally, differentially expressed genes were identified using Seurat’s FindAllMarkers and the resulting clusters were manually annotated based on differentially expressed canonical markers as described in the main text.

Processing of human scRNA-seq data.

11 samples were prepared using 10x v2 chemistry, and the rest used 10x v3 chemistry. In total, the human dataset included 23,819 myeloid cells. Single-cell RNA-sequencing data was aligned using the 10x Genomics Cell Ranger pipeline with the GRCh38 reference transcriptome. Cells with at least 500 detected genes, 1000 UMIs and less than 3% reads mapped to mitochondrial genes (following DASH) were kept for subsequent analysis. Scrublet(Wolock et al., 2019) was used to remove suspected doublets, such that the expected doublet rate was calculated based on the number of cells loaded, and parameter values were set to min_counts=2, min_cells=3, min_gene_variability_pctl=90, and n_prin_comps=20.

Translating mouse gene symbols to human gene symbols.

To integrate the mouse and human expression data, we converted the mouse gene symbols to human gene symbols by mapping between mouse and human gene orthologs. We obtained the ortholog mapping using the ‘biomaRt’ R package (v2.46.3), mapping mgi_symbol from the mmusculus_gene_ensembl database to hgnc_symbol from the hsapiens_gene_ensembl database, using the dec2021.archive version of Ensembl. We added additional ortholog pairs from HomoloGene (https://ftp.ncbi.nih.gov/pub/HomoloGene/build68/) to obtain a total of 19,055 ortholog pairs (representing D=17,798 unique human genes and d=18,230 unique mouse genes). We represented this mapping as a matrix with human genes as rows, mouse genes as columns, and values in {0,1} denoting whether a mouse gene maps to a human gene. The vast majority of mouse genes (97% = 17,701/18,230) had one-to-one mappings. To handle the one-to-many mappings, we normalized each column to sum to one to create a count-preserving probabilistic map from mouse to human genes M ∈ R^D×d. We then applied the mapping matrix to the original mouse expression matrix (U_mouse) to obtain the “humanized” mouse expression: U_humanized= MU_mouse. In this way, the UMI counts for a given mouse gene were uniformly distributed among the matching human symbols for one-to-many mappings. For any human orthologs that were missing in the mouse expression data, we filled in the expression with zeroes.

Mouse and human integration and label transfer.

Starting from the human expression matrix and humanized mouse expression matrix, we performed log(CP10K+1)-normalization on each cell. We subset by the set of 16,957 genes overlapping between the two species and concatenated the cells from both species into a combined matrix. We calculated the top 1,500 variable genes within each species separately using the variance-stabilizing transform method in Seurat. After merging the two variable gene sets, we removed ribosomal, mitochondrial, and cell cycle (g2m.genes and s.genes in Seurat v4.1.0) genes, resulting in 2,345 variable genes used for subsequent dimensionality reduction. We scaled the data and performed PCA using the ‘Seurat’ R package (v4.1.0), calculating the top 20 PCs. We then removed the effect of species, sample, and technology (10x version) using Harmony (v0.1.0) with theta = 2.5, 0.5, and 0.5, respectively. This resulted in an integrated low-dimensional embedding of cells from both species. We ran uniform manifold approximation and projection (UMAP) using the uwot R package (v0.1.11) to visualize the embedding. To transfer the cell cluster labels defined in the mouse dataset to the human cells, we predicted the mouse label for each human cell by taking the majority vote of the 5 closest mouse cell neighbors in the Harmony embedding based on Euclidean distance (with ties broken at random), using the ‘class’ (v7.3-20) R package.

Comparing differentially expressed genes across species.

To compare differentially expressed genes (DEGs) for each cluster across species, we used the ‘wilcoxauc’ function from the presto (v1.0.0) R package to calculate DEGs within each species separately (each cluster relative to cells from all other clusters). Human clusters were defined using the label transfer from mouse cells. We then compared the logFC of each DEG in each mouse cluster to the logFC of the same gene in each human cluster, limiting the comparison to significant DEGs in mice (Benjamini-Hochberg adjusted P-value < 0.01).

Inferred differentiation of myeloid clusters.

To identify transcriptional relatedness between clusters and infer differentiation in our scRNA-seq data, we used Partition-based Graph Abstraction (PAGA) based on the k-NN graph above(Wolf et al., 2019). To move from Seurat to the Anndata format supported in the Scanpy and Scvelo Python libraries, the SeuratDisk package in R was used to convert Seurat objects to the h5ad compressed format. Metadata labels are not carried over during this conversion process, so all labels were exported and loaded separately. A diffusion map and a diffusion pseudotime was generated for each dataset for the PAGA analysis. For each analysis, the classical 1 myeloid subset was used as the starting cluster. All connections between clusters in the PAGA plots were given a connectivity score between 0 and 1. UMAP overlay and circular graphs were generated using these connectivity scores.

Inference of regulatory gene networks.

To discover gene regulatory networks underpinning transcriptional changes in selected cellular subsets, we employed the SCENIC (Single-Cell rEgulatory Network Inference) method(Aibar et al., 2017) and the recent pySCENIC implementation(Van de Sande et al., 2020), to compute “regulons”, or pairs of transcription factors and their putative target genes, using single cell RNAseq data. The packages prune the regulon gene list based on putative binding motif enrichment, and evaluate the regulon activity score (RAS) with the “AUCell” function. Because our data sets are relatively large for these packages, we modified the computational pipeline to process subsets of 2000 genes in each iteration, followed by a post-hoc joining of pairs of transcription factors and target genes from all subsets.

Conservation assessment of mouse-human gene regulatory networks.

We used the Harmony package as above to map mouse cell identities to human cells. Only human cells that were mapped from mouse C1 and C2, or RM1 and RM0, were used in our analysis. First, we inferred regulons from patient scRNA-seq data. Next, we assessed conservation of regulons as follows: For a mouse-inferred regulon R_m, associated with transcription factor TF_Rm and target genes set G_Rm, let G_h be the human homolog of TF_Rm. If G_h was found to be a transcription factor in a human-derived regulon R_h with target genes set G_Rh, we set M as the minimum of |G_Rm| and |G_Rh|. We count only the top M target genes of R_m and R_h and define a shared gene as a mouse gene G_m ∈ G_Rm having human homolog gene G_h ∈ G_Rh. The number of shared genes within the regulon N = || {G_m | G_m ∈ G_Rm and homolog(G_m) ∈ G_Rh} ||, and the rate of shared genes within the mouse regulon genes P = N / M are provided as a measure of mouse-human conservation. A small number of modules exhibited opposing trends in mouse and human. For example, modules were shared between mouse C1 and human C2 (CREB3, MXI1) and mouse C2 and human C1 (IRF1). Such modules may arise due to imperfect matching of mouse and human cells, such that regulatory gene modules were computed on human cells that do not fully resemble or encapsulate their C1/C2 mouse counterparts. This may be further exacerbated by transitional states that exist between C1-C2 and their misassignment to either group. Alternatively, these modules may represent legitimate cross-species differences and necessitate further investigation.

Analysis of urine immune cells.

To determine which myeloid states identified in our analysis of 155 patients were present in the urine of unmatched patients from our previous study(Arazi et al., 2019a), we focused specifically on urinary myeloid populations. We classified urine cells after each one based on the highest gene module score(Hao et al., 2024), reflecting the average of the top 10 DEGs in the reference myeloid cluster.

Cellular experiments

Co-culture.

NZW GFP+ mice were generated by backcrossing for >11 generations and used to generate GFP+ NZB/W females. NZB/W GFP+ Mouse Embryonic Fibroblasts (MEFs) were generated and cryopreserved using established methods(Qiu et al., 2016). 40,000 MEFs at passage 3 were plated in a 96-well tissue culture-coated plate at day −1. At day zero, CD11b+ Ly6c− F480+ CD81+ resident macrophages from nephritic and pre-nephritic F1 NZB/W kidneys were isolated by FACS sorting and co-cultured with the MEFs at a ratio of 1:1 in 5% FBS complete media containing 5% FBS. After 48 hours MEFs and resident macrophages were harvested and analyzed via flow cytometry.

BMDM.

Bone Marrow Derived Macrophages (BMDM) were made according to published protocols(Toda et al., 2021). Bone marrow cells from pre-nephritic NZB/W F1 mice were harvested from the femur, fibula and tibia and plated in 6 well plates in complete media containing 10% FBS and 10ng/mL of GM-CSF for 7 days, refreshing the media at day 4. At day 7, the media was replaced with 10ng/mL of GM-CSF plus cytokines (M0 – none; M1- 100ng/mL LPS, 20ng/mL IFNγ; M2- 30ng/mL IL-4, 30ng/mL IL-13; M3a- 50ng/mL IL-17a; IFNa-50ng/mL IFNa) for 6 days; the media was refreshed after 3 days. Six days after stimulation (day 13), BMDMs were harvested and used for downstream assays and flow cytometry analysis.

Phagocytic capacity of macrophages and monocytes.

All ex-vivo phagocytosis assays were performed in a 96-well plate using a single-cell suspension resuspended in warm DMEM 10%, incubated at 37 °C for the appropriate times. They were then washed and stained for flow cytometry analysis. Large ICs were prepared, as described previously(Richoz et al., 2022) by incubating AF647-OVA (1 mg/mL in PBS) with rabbit polyclonal anti-OVA antiserum (3.7 mg/mL) at a 1:10 molecular weight ratio for 60 minutes in a 37°C water bath and were then washed twice (10,000 rpm for 2 minutes with discarding of the supernatant) and resuspended in PBS prior to use. 1mM AF488 labeled heat aggregated albumin in PBS was prepared by incubating protein at 59°C for 45 minutes. 1.0 uM AF488 latex beads were washed with PBS twice at 10,000 rpm for 2 minutes, with discarding of the supernatant, prior to use. Fluorescent green-yellow 1.0uM Latex beads (Polysciences) were prepared by washing x 2 in PBS and then diluted 1:10 prior to use.

Bulk RNA sequencing for resident macrophages before and after disease

Perfused kidneys from 16 week prenephritic and >30 week nephritic NZB/W mice were processed as above and myeloid cells were enriched using anti-CD11b magnetic beads. Resident macrophages (Ly6G-/B220-/CD3-/CD11b+/CD11c-/F4/80hi) were isolated by flow sorting. RNA was extracted using an RNeasy mini kit (Qiagen) and RNA quality was assessed by using an Agilent 2100 Bioanalyzer (Agilent Technologies). Library preparation and RNA-seq were performed by Novogene using an Illumina HiSeq 4000, which generated 23–56 million reads per sample. Raw RNA-seq reads were subjected to quality checking and trimming to remove adaptor sequences, contamination and low-quality reads. Good quality reads were aligned to reference databases including mouse genome (build 10 mm), exon, splicing junction segment and contamination databases including ribosome and mitochondrial sequences using Burrows Wheeler alignment algorithm(Li and Durbin, 2009). After filtering, the reads that were uniquely aligned to the exon and splicing-junction segments with a maximum of 2 mismatches for each transcript were then counted for each corresponding transcript. The genes with a read count of less than 100 across all samples were excluded. The read counts of remaining genes were log2 transformed and normalized at an equal global median value.

In situ localization of cellular states identified by spatial transcriptomics

Tissue Collection and Processing.

Human kidney biopsies were obtained from patients diagnosed with lupus nephritis and healthy controls from the Brigham and Women’s Hospital, while murine kidneys were harvested from lupus mouse models pre-disease or with lupus nephritis. Tissue sections were prepared according to the manufacturer’s instructions for Xenium spatial transcriptomics. Custom probe sets targeting myeloid cluster-specific genes, based on scRNA-seq data sets, were designed for spatial transcriptomic analysis with Xenium. These probe sets were utilized in combination with the manufacturer’s multi-tissue panel, which includes a comprehensive set of genes relevant to the study. Tissue sections were subjected to probe hybridization, allowing for the capture of spatially resolved gene expression profiles directly from the tissue. Following probe hybridization, tissue sections were imaged using the Xenium imaging platform to capture spatially resolved gene expression signals for a single data set. The same sections were then stained with hematoxylin and eosin. High-resolution images were obtained, allowing for the visualization of gene expression patterns across the tissue sections aligned with H&E.

Spatially resolved gene expression data were processed and analyzed using Seurat (version 5.0) and Xenium Explorer (version 3.1).

Cells with the lowest 10% of UMIs and gene counts were removed initially. Data normalization and feature selection were performed using the ‘NormalizeData’ and ‘FindVariableFeatures’ functions in Seurat. Xenium datasets from mouse or human kidney sections were combined using the ‘merge’ function. Subsequent steps included scaling and dimensional reduction with ‘ScaleData’, ‘RunPCA’, ‘RunHarmony’ (version 1.2.0), and ‘RunUMAP’. Clustering was carried out using ‘FindNeighbors’ and ‘FindClusters’, followed by manual inspection to remove non-myeloid cells, proliferating cells, and doublets. The merged Xenium datasets were integrated with human or mouse single-cell RNA-seq reference datasets using the ‘FindIntegrationAnchors’ and ‘FindTransferAnchors’ functions in Seurat. This integration enabled the identification of cells in the Xenium (query) dataset based on reference datasets while retaining their spatial coordinates. Seurat generates a prediction score to quantify the likelihood that a given cell in the query dataset corresponds to a specific cell type in the reference dataset, based on the similarity between the expression profile of the query cell and the reference cell types. Scores range from 0.0 to 1.0, with higher scores indicating a stronger match and greater confidence in the assigned label. To localize Xenium myeloid subsets with high correspondence to the single-cell reference, we mapped cells with prediction scores > 0.75 in situ using the Xenium Explorer (version 3.0) software. Renal compartments were annotated by a board-certified renal pathologist (RS).

Enrichment analysis of cell clusters in kidney regions

For each cell cluster, we calculated the fraction of cells in each region of the kidney (inside glomeruli, bordering glomeruli etc). To check if there was a statistically significant enrichment or depletion of any cluster in any of the kidney regions, we first calculated the fraction of myeloid cells in each region, regardless of cluster. We then used a Binomial test to calculate the p-value of having the observed number of cells in each region for each cluster, under the null hypothesis that the cluster cells have no preference to be in or to avoid the region, i.e. that they follow the distribution of myeloid cells across each region, regardless of cluster. P-values were then corrected for multiple comparisons using the Benjamini-Hochberg method for controlling the FDR. In addition, given the relatively small number of patients, and to avoid cases where one patient was driving the observed statistically significant adjusted p-values, we repeated the above calculation for each patient separately, and reported only cases with FDR ≤ 0.05 for all patients taken together, and in addition FDR ≤ 0.05 for at least two patients when calculated separately.

Study approval (human experimental guidelines approval statement)

The protocols involving human participants received approval from the institutional review boards (IRBs) at each participating location, and all individuals provided written informed consent. Patient enrollment took place at multiple institutions, including Johns Hopkins University, New York University, Albert Einstein College of Medicine, University of Rochester Medical Center, Northwell Health, University of California, San Francisco, Medical University of South Carolina, University of California, San Diego, Cedars-Sinai Medical Center, University of Michigan, University of Texas at El Paso, and University of California, Los Angeles. For the healthy control group, IRB approval was secured from the University of Cincinnati and the Oklahoma Medical Research Foundation. Following informed consent, controls were recruited through the University of Cincinnati. The Oklahoma Rheumatic Disease Research Cores Center was responsible for sample storage, with efforts made to match samples based on sex, race, ethnicity, and age.

Supplementary Material

Table 9

Table 9 DEG analysis of RM subclusters from mice and humans

Table 10

Table 10 DEG analysis of mouse C1 vs C2

Table 11

Table 11 DEG analysis of all analogous myeloid clusters from mice and humans

Table 12

Table 12 DEGs of analogous C1vsC2 from mice and human

Table 13

Table 13 DEG analysis of mouse NC1 vs NC2

Table 14

Table 14 KEGG analysis mouse NC1 and NC2

Table 15

Table 15 Xenium gene set for mouse

Table 16

Table 16 Xenium gene set for human

Table 17

Table 17 Xenium markers to distinguish mouse myeloid subsets

Table 1

Table 1 Intrarenal myeloid cell counts in mice

Table 2

Table 2 Intrarenal myeloid cell counts in humans

Table 3

Table 3 DEG analysis of mouse clusters from single cell RNA-seq

Table 4

Table 4 Antibodies used for flow cytometry

Table 5

Table 5 DEG analysis of bulk RNA seq analysis of pre- and nephritic RMs isolated from NZBWs

Table 6

Table 6 DEGs before and after disease onset in myeloid clusters from scRNA-seq

Table 7

Table 7 DEG analysis of mouse RM subclusters

Table 8

Table 8 DEG analysis of mouse RM1vsRM0

Supp Figures 1-5

Fig. S1. Bioinformatic Quality Metrics of Intrarenal Myeloid Cells from Mice and Human Single Cell Data Sets and Integration. (A) Violin plots and UMAPs depicting the percentage of mitochondrial reads, genes, or UMIs per cell in the mouse (20,299 cells) and human (23,819) data sets. (B) Harmonized UMAP embedding of myeloid cells from lupus mice and patients. (C) Human cells assigned mouse identities based on the 5 nearest neighbors in UMAP space (left) or colored by the number of concordant neighbors (right). (D) Table representation of C in which the fraction of each human cell labeled with a mouse cluster identity (y-axis), binned by the number of concordant neighbors (top row). Average concordance for all cells from each cluster (rightmost column).

Fig. S2. FACS Gating Strategy For Sorting Murine Myeloid Subsets Based On scRNA-Seq In Figure 3. Using beads we isolated CD11b+ and CD11c+ cells from kidney cell suspensions collected from pre-disease mice or with clinical nephritis. (A) We gated for live single cells in the lymphocyte/myeloid gate. cDC1 and neutrophils were excluded using anti-XCR1 and anti-LyG respectively. Remaining cells were analyzed in the CD45+/CD11b+ gate (R5). cDC2 were identified as CD11chi/CD26+. Ly6Chi and Ly6Cint cells (R7) were subsetted using antibodies to Dectin2 and anti-CCR2. Classical 1 monocytes were defined as Ly6Chi/CCR2hi/Dectinlo and Classical 2 monocytes were defined as Ly6Cint/Dectin+/CCR2lo. Populations that were Dectin2lo/CCR2lo (DN) and Dectinhi/CCR2hi (DP) were also distinguished in the R7 gate. Ly6Clo cells (R8) were then used to gate resident macrophages (F4/80hi/CD81hi). Ly6Clo non-classical 2 cells were defined as Itgalhi/Fcgr4hi/Fcrl5hi and non-classical 1 cells were defined as Itgalhi/Fcgr4hi/Fcrl5lo. A small population of unidentified cells remained that comprised <5% of all myeloid cells. (B) Expression histograms of antibody surface markers (x-axis) from intrarenal myeloid subsets shown from A (above). (C) Gating scheme for peripheral myeloid populations collected from pre-disease mice (top) or lupus nephritis (bottom) using a similar isolation and gating strategy as A.

Fig. S3. Differential Gene Expression Before and After the Onset of Murine Disease in Myeloid Clusters from Single-Cell RNA Sequencing. (A) Volcano plots illustrating the differential expression of genes associated with the indicated cluster in Fig. 1B, shown before (left) and after (right) the onset of lupus nephritis. Plots incorporate data from all mouse strains. Red indicates log2 fold change > 0.5, while blue indicates < 0.5. Only points with −log10 p-values > 1.3 (or < 0.05) are displayed. Clusters for pDC and Ccr7 are excluded due to having fewer than 5 DEGs with p-value cut-offs, and genes Jchain and Xist were removed from the plots. Corresponding gene lists are in Table 15. (B) Heatmaps displaying the scaled gene expression levels (y-axis) of (left) cytokines and (right) chemokines in the indicated cluster, compared before and after the onset of disease. See table 15 for p-values for each DEG.

Fig. S4. Differential Gene Expression Of Resident Macrophage Subclusters From Single Cell Rna Sequencing Data And Comparison To Human. (A) Heatmap showing the scaled expression of top 20 discriminative genes for each resident macrophage subcluster. Color scheme is based on z-score, from −2 (blue) to 2 (red). (B) PAGA transcriptional relatedness of resident macrophage subclusters between mice (left) and humans (right). (C) Scatter plots of cluster-specific marker genes from humans (y-axis) to mice (x-axis), displaying their expression levels relative to other resident macrophage subsets.

Fig. S5. Top Genes Comprising Regulons From Mouse RM1, RM0, C1, Or C2 And Comparisons Between Mice And Humans. Comparing regulon activity in (A) RM1 vs. (B) RM0 or (C) C1 vs. (D) C2. (E, F) Overlap of mouse and human regulons enriched in (E) RM1 (LFC>0) and RM0 (LFC<0) or (F) C1 and C2 subsets. Each row displays mouse and human transcription factors, mouse LFC, and human LFC, along with conservation metrics (the number of conserved genes, the ratio of conserved genes in mice compared to humans), as described in Methods.

Fig. S1: Shows quality metrics and integration of mouse and human single-cell myeloid data, including visualizations of data quality, harmonized UMAP embeddings, and cell identity assignments across species. Fig. S2: Depicts detailed FACS gating strategy for isolating murine myeloid subsets based on scRNA-Seq results, including gating schemes for both kidney and peripheral myeloid populations in pre-disease and lupus nephritis conditions. Fig. S3: Shows differential gene expression analysis of myeloid clusters before and after the onset of murine lupus nephritis, presented through volcano plots and heatmaps focusing on cytokines and chemokines. Fig. S4: Shows analysis of resident macrophage subclusters, including gene expression heatmaps, transcriptional relatedness visualizations (PAGA), and cross-species comparisons of cluster-specific marker genes between mice and humans. Fig. S5: Shows regulon analysis in mouse resident macrophage and classical monocyte subsets, with comparisons between mice and humans, including regulon activity, overlap, and conservation metrics.

Non-standard abbreviation list:

SLE

Systemic lupus erythematosus

LN

lupus nephritis

dsDNA

double-stranded DNA

PAGA

partition-based graph abstraction

UMAP

uniform manifold approximation and projection

DEG

differentially expressed gene

Data availability

The murine single cell RNA seq data underlying Figures 1, 2, 3, 5, 7 are openly available in GEO accession: GSE302065

The human single cell RNA seq data underlying Figures 2, 3, 5, 6, 7, 8, 10 are in whole or in part based on data obtained from the ARK Portal (http://arkportal.synapse.org). The Accelerating Medicines Partnership^® RA/SLE kidney scRNA-seq data used for this publication are available under the DOI: https://doi.org/10.7303/syn68564337.1. The ARK Portal hosts data generated by a network of research teams working collaboratively to deepen the understanding of Arthritis and Autoimmune and Related Diseases. It was established by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) and includes data from the Accelerating Medicines Partnership^® (AMP^®) RA/SLE program. The specific data used in this publication is available as a controlled-access dataset. Researchers seeking to use these data must submit:

1. A detailed intended data use statement

2. A completed and signed data use certificate

These access requirements ensure responsible and ethical data sharing within the research community. Instructions for access are available at https://help.arkportal.org/help/data-use-certificate#DataUse&Acknowledgement-Acknowledgement.