Endothelial cell-released CD93 Contributes to Podocyte Injury in Idiopathic Nephrotic Syndrome

Abstract

Idiopathic Nephrotic Syndrome (INS) is considered a podocyte disease triggered by immune-derived factors. Endothelial activation occurs, but whether activated endothelium contributes to podocyte injury is unknown. We tested the hypothesis that CD93, a protein primarily expressed in endothelium, is a contributory etiologic factor of podocyte injury. We studied 460 patients with INS and 150 with other podocytopathies. CD93 was analyzed in kidney tissue, urine, and serum samples. We performed cell culture studies to investigate the source of soluble CD93 and its effects on cultured human podocytes. We tested the efficacy of CD93 blockade in vitro and in vivo and investigated the relationship between soluble CD93 and clinical outcomes in human INS. CD93 was highly expressed by glomerular endothelial cells (GEnCs) in human INS, and INS sera stimulated cultured human GEnCs to release CD93. Mechanistically, soluble CD93 mediated podocyte activation via β1 integrin/FAK signaling in cultured human podocytes. CD93 blockade mitigated activation of cultured human podocytes and albumin permeability in human GEnCs-podocyte co-cultures as well as albuminuria, glomerulosclerosis, and podocyte loss in two models of nephrotic syndrome (PodTgfbr1 mice and Adriamycin). Cd93 knockout mice showed less proteinuria and glomerulosclerosis, compared to controls, following Adriamycin injection. In patients with INS, soluble CD93 was high in urine in approximately 90% and 50% of patients in relapse and remission, respectively. High urinary CD93 associated with faster decline in kidney function and slower response to immunosuppression. Soluble and glomerular CD93 was also elevated in other podocytopathies. We conclude that soluble CD93 contributes to podocyte injury.

One Sentence Summary:

Soluble CD93 contributes to podocyte injury and CD93 blockade emerges as a candidate therapeutic in INS.

INTRODUCTION

Idiopathic Nephrotic Syndrome (INS) encompasses a group of kidney diseases characterized by heavy proteinuria, podocyte injury, and variable clinical outcomes (1–4). It is considered an immune cell-driven podocyte disorder in that circulating autoantibodies specifically target podocyte proteins (5–8). Upon injury, podocytes exhibit non-specific molecular changes involving several proteins that modulate the podocyte’s cytoskeleton resulting in podocyte shape changes namely foot process effacement (FPE) (9). Yet, the precise mechanisms of podocyte injury are incompletely understood, and predictive markers of disease are limited (10).

Along with podocytes, the glomerular filtration barrier (GFB) consists of the glomerular basement membrane (GBM) and glomerular endothelial cells (GEnCs). Evidence that the endothelium is activated in INS is being increasingly appreciated (11–16). Ye et al. recently identified autoantibodies to the endothelium in 90% of patients with INS, and these correlated with circulating endothelial markers (16). We also showed that INS sera in relapse directly activate human GEnCs (13), suggesting that endothelial alterations are not simply a secondary phenomenon in these patients. Endothelial activation contributes to the pathogenesis and progression of certain proteinuric kidney diseases (17, 18); however, its importance in INS remains untested.

Upon activation, human GEnCs may release factors associated with different biological effects (19). A key feature of INS is podocyte focal adhesion kinase (FAK) activation, which mediates podocyte injury, FPE and proteinuria in vivo (20, 21). This led us to search for candidate endothelial proteins involved in FAK activation. Cluster of Differentiation 93 (CD93), a ~120-kDa transmembrane glycoprotein primarily expressed by endothelium (22), modulates endothelial cell adhesion via β1 integrin/FAK signaling (23–25). In inflammatory states, CD93 can be released from the cell surface as a soluble and biologically active form (26, 27). Here, we combined human, cell, and animal studies to test the hypothesis that activated GEnCs release soluble CD93 which contributes to podocyte activation in INS.

RESULTS

GEnCs are a source of soluble CD93 in INS

In quiescent endothelium, CD93 expression is low, but it increases upon endothelial activation by various stimuli (28). We first analyzed RNA sequencing (RNA-seq) data from micro-dissected human glomeruli to determine the expression of CD93 in patients with INS (n=149 INS, n=53 controls; NEPTUNE cohort (29), see “Study Participants” in the Materials and Methods). CD93 mRNA expression was higher in patients with INS than in controls independently of the degree of proteinuria (Fig. 1A) or histological diagnosis (fig. S1A) including minimal change disease (MCD) and focal segmental glomerulosclerosis (FSGS). Analysis of available single-cell RNA-seq data (30), confirmed endothelial cells as the primary cell type expressing CD93 in the kidney of patients with INS, and expression was higher in GEnCs from patients with MCD than those with FSGS (Fig. 1B). To further validate this, we performed immunostaining of kidney tissue from a separate cohort of 34 children and adults with MCD and FSGS in relapse and from 9 subjects without glomerular disease (controls). CD93 expression was higher in glomeruli from patients with INS than in controls and showed a non-significant trend toward higher expression in MCD compared to FSGS (Fig. 1C and fig. S1B–D). CD93 colocalized with an endothelial-specific lectin (31), thereby documenting its endothelial origin in the INS glomerulus (Fig. 1C).

Fig. 1. GEnCs Are a Source of Soluble CD93 in INS.

(A) Data represents the log² CD93 mRNA expression in micro-dissected glomeruli from patients with MCD/FSGS and controls (n=149 and n=53, one-way ANOVA with Tukey’s test). Doted lines within violin plots represents median, 1^st and 3^rd quartiles. (B) Data represents CD93 mRNA expression by single-cell analysis of kidney tissue in patients with FSGS/MCD (n=8 FSGS, n=2 MCD) (top panel) and in patients with FSGS/MCD and “high” TNF signaling (n=1 FSGS, n=4 MCD)(bottom panel) (30). C) Data represents expression of CD93 (green) and endothelial marker ULEX (red) in kidney tissue from patients with MCD/FSGS and controls (n=24/7 and n=9) (left panel). Images captured at 40x, scale bars, 30 μm. Additional images are shown in figure. S1B. Right panel shows CD93 expression quantification. One-way ANOVA with Tukey’s test was performed. Data presented as mean±SD. (D) Data represents CD93 expression in GEnCs’ lysates following incubation with healthy and INS sera using a CD93 antibody raised against its extracellular domain (441 – 581) (left panel). Right panel shows quantification of CD93 expression normalized to GAPDH (n=6 and n=7, unpaired t test). Data presented as mean±SD. (E) Data represents CD93 levels, measured by ELISA, in supernatants from GEnCs previously treated with healthy and INS sera (n=6 per group, unpaired t test). Data presented as mean±SD. N number of participants, C Healthy control, INS idiopathic nephrotic syndrome, UPCR Urinary protein-to-creatinine ratio (g/g), MCD Minimal change disease, FSGS Focal segmental glomerulosclerosis, ATL Ascending thin limb, DCT, Distal convoluted tubule, DTL Descending thin limb, EC Endothelial cell, FIB Fibroblast, PC/IC Principal and intercalated cells, POD Podocytes, PT Proximal tubule, TAL Thick ascending limb, GEnCs Glomerular endothelial cells, SD standard deviation. **p≤0.01, ***p≤0.001, and ****p≤0.0001.

Because INS sera in relapse directly stimulates GEnCs (13), we performed cell culture studies to test whether activated GEnCs are a source of soluble CD93. Human GEnCs were cultured with sera from patients with INS in relapse and healthy subjects. Cells exposed to disease sera, compared to controls’, demonstrated lower CD93 protein expression, using an antibody against its extracellular domain, in GEnCs lysates (Fig. 1D). In contrast, CD93 levels were higher in cell supernatants from GEnCs previously exposed to disease sera compared to those exposed to healthy sera (Fig. 1E), supporting that INS sera triggered the release of CD93 from GEnCs. Next, we tested whether INS sera could also trigger CD93 release in other types of endothelial cells. Because human umbilical endothelial cells (HUVECs) are widely utilized to study endothelial biology, we cultured HUVECs with sera from the same healthy subjects and patients with INS in relapse. Contrary to that observed with GEnCs, we found no differences in CD93 expression in HUVECs lysates exposed to healthy and disease sera (fig. S2).

Soluble CD93 activates cultured human podocytes

In vitro GEnCs neighbor podocytes, and both cell types release autocrine and paracrine factors critical to maintain a healthy GFB (32, 33). It is well established that direct injury to endothelial cells or podocytes can alter the neighboring cells (34, 35). Because GEnCs showed high CD93 expression in human with INS, and human GEnCs release soluble CD93, we next tested whether soluble CD93 could contribute to human podocyte activation via β1 integrin signaling. By co-immunoprecipitation studies, we demonstrated that soluble CD93 bound to β1 integrin in cultured human podocytes (Fig. 2A). Next, we found that recombinant CD93 caused human podocyte FAK phosphorylation, a surrogate of β1 integrin activation (36, 37), and this was mitigated by a β1 integrin neutralizing antibody (38) (Fig. 2B). Additionally, we showed that recombinant CD93, but not CD93 or β1 integrin antibodies, accelerated cell migration in cultured human podocytes. This effect was blunted when CD93 or β1 integrin antibodies were added into the cell culture media (Fig. 2C). These data support that soluble CD93 activated cultured human podocytes in a β1 integrin dependent-manner.

Fig. 2. Soluble CD93 Mediates Podocytes Injury.

(A) Data represents co-immunoprecipitation of CD93 and β1 integrin in cultured human podocytes exposed to rCD93 or INS serum in relapse (n=1). Omission of CD93 antibody (−) represents the internal control for each experiment. (B) Data represents phosphorylated FAK expression in human podocytes treated with rCD93, without and with pre-incubation of the β1 integrin blocking antibody Mab13(38) (upper panel). Bottom panel represents the quantification of phosphorylated FAK, from upper panel, normalized to GAPDH. Quantification includes 3 time points (0, 3 and 6 hours) from 4 independent experiments (one-way ANOVA with Tukey’s test). Data presented as mean±SD. (C) Left panel shows images, following wounding, of cultured human podocytes (unstimulated podocytes, and podocytes exposed to CD93 or β1 integrin antibodies without or with the addition of rCD93). Images captured at 40x, scale bars, 200 μm. Right panel shows the percentage of cell culture plate area covered by human podocytes following wounding (one-way ANOVA with Tukey’s test). Data presented as mean±SD. IP immunoprecipitation, rCD93 recombinant CD93, ab antibody, h hours, RL relapse, SD standard deviation. *p≤0.05, ****p≤0.0001.

CD93 blockade prevents injury in cultured human podocytes

In cultured human podocytes, we found that sera from patients with INS in relapse, compared to healthy individuals’, caused FAK activation; and this was blunted by adding a CD93 antibody to patients’ sera (Fig. 3A). To test the effect of CD93 blockade in albumin permeability, we used a well-established culture system, involving human GEnCs and podocytes, called glomerulus on-a-chip (39, 40). This approach recapitulates key features of the human glomerular filtration barrier (39). Sera from nephrotic patients, compared to controls’, increased albumin permeability in vitro (Fig. 3B), which is a surrogate of albuminuria in humans (39). This was blunted by adding a neutralizing CD93 antibody to patients’ sera (Fig. 3B).

Fig. 3. CD93 Blockade Prevents Podocyte Activation In vitro.

(A) Data represents phosphorylated FAK expression in human podocytes treated with serum from healthy subjects and patients with INS with and without addition of CD93 antibody to culture media (left panel). Right panel shows the quantification of phosphorylated FAK , from left panel, normalized to GAPDH (n=3 per group, one-way ANOVA with Šidák multiple comparison test). Data presented as mean±SD. (B) Data represents albumin leakage in the human GEnC-podocyte co-culture on-a-chip system following exposure to sera from healthy (n=4) and INS subjects (n=5) with or without neutralizing CD93 antibody (one-way ANOVA with Šidák multiple comparison test). Data presented as mean±SD. pFAK Phosphorylated FAK, INS idiopathic nephrotic syndrome, ab antibody. SD standard deviation. *p≤0.05. **p≤0.01, and ****p≤0.0001.

CD93 blockade mitigates albuminuria and podocyte injury in vivo

To assess causality between CD93 and albuminuria in vivo, we first tested the effect of a CD93 neutralizing antibody in the PodTgfbr1 mouse model of INS (41–43). In this model, doxycycline (DOX) induction results in TGFβR1 signaling in podocytes causing an increase in albuminuria and endothelial injury by day 4, FPE by day 7, and global glomerulosclerosis with tubulointerstitial fibrosis by day 14, as previously described (41). This model is clinically relevant to INS as transforming growth factor-β (TGF-β) signaling plays a role in the disease pathogenesis (44). As expected, DOX-induced mice exhibited increasing levels of albuminuria (Fig. 4A) (41). These mice demonstrated a non-significant rising urinary CD93 levels by day 4, coinciding with albuminuria onset, and peaked between day 7-14 (Fig. 4B). By day 14, mice injected with CD93 neutralizing antibody demonstrated less albuminuria and glomerulosclerosis, but higher podocyte count than those not treated with CD93 neutralizing antibody (Fig. 4C to E). To further validate the effect of CD93 blockade, we studied the Adriamycin (ADR) mouse model as this is widely accepted to study mechanisms of podocyte injury and albuminuria in INS (45). Following ADR injection, mice injected with CD93 blocking antibody showed less albuminuria, urinary CD93 levels, glomerulosclerosis and podocyte loss than control mice (fig. S3). To further assess for a causal link between CD93 and albuminuria, we used a Cd93 global knockout (KO) mouse model, and ADR to induce podocyte injury. As expected, Cd93 KO mice lacked CD93 in the kidney (fig. S4) and was also undetectable as soluble form in serum and urine by ELISA. Compared to controls, Cd93 KO mice showed similar degree of albuminuria at baseline, but less albuminuria and glomerulosclerosis following ADR injection. Control mice also showed an increase in urinary CD93 after injury (Fig. 5A to C).

Fig. 4. CD93 Blockade Mitigates Albuminuria, Glomerulosclerosis and Podocyte Loss In vivo.

(A) Data represents serial measurements of albuminuria, adjusted to urine creatinine, in PodTgfbr1 mice (n=5-11 mice per time point, [day 0, 2 female mice; day 4, 1 female, day 7, 4 females, and day14, 3 females; the rest of the mice were male]). One-way ANOVA with Tukey’s test for multiple comparisons was performed. Data represents mean±sem. (B) Data represents urinary CD93, adjusted to urine creatinine, from PodTgfbr1 mice (n=4-6 mice per time point [1 female per time point, rest of mice were male]). One-way ANOVA was performed with Tukey’s multiple comparisons test. Data represents mean±sem. (C) Data shows albuminuria, adjusted to urine creatinine, in PodTgfbr1 mice treated with or without CD93 blocking antibody (n=4-6 mice per time point [1 female per group, rest of mice were males]). One-way ANOVA was performed with Tukey’s multiple comparisons test. Data represents mean±sem. (D) Data shows PAS staining of glomeruli from mouse kidney tissue (n=3-6 mice per group [CTRL, 3 female mice; DOX, 3 female mice, DOX/CD93 ab, 1 female; rest were male mice], one-way ANOVA with Tukey’s multiple comparisons test). Data represents mean±sem. Images captured at 40x. Scale bars, 10 um. (E) Left panel shows triple-immunofluorescence staining including podocyte markers synaptopodin (red) and WT1 nuclei (green) as well as DAPI (blue) in PodTgfbr1 mice. Arrows depict nuclear WT1 and DAPI in podocytes (upper panel). Images captured at 20x. Scale bars, 10 um. Right panel shows podocyte quantification, from left panel, including >35 glomerular profiles/mouse, 4-6 mice per time point (DOX and DOX/CD93 ab included 3 female mice; rest were males). One-way ANOVA with Tukey’s multiple comparisons was performed. Data presented as mean± sem. PAS Periodic acid–Schiff, DOX doxycycline, ab antibody. SEM standard error of the mean. *p≤0.05, **p≤0.01, ***p≤0.001 and ****p≤0.0001.

Fig. 5. CD93 KO Mice Are Protected From Adriamycin-Induced Albuminuria and Glomerulosclerosis.

(A) Data represents albuminuria in Cd93 KO and control mice at baseline (day 0) and day#6 following single injection of ADR (n=9 male and 5 female mice per time point, except for control/ADR group that included 4 females, one-way Anova with Tukey’s multiple comparison test). Data presented as mean±sem. (B) Urinary CD93, adjusted to urine creatinine, in control mice at baseline and following ADR injection (n=6 male and 3 female mice per time point, unpaired t test). Data presented as mean±sem. (C) Data shows PAS staining of glomeruli from mouse kidney tissue (n=6 mice per group [day#0, 3 and 2 females; day#6, 1 female mouse per group, rest of mice were male], one-way ANOVA with Tukey’s multiple comparisons). Images captured at 20x. Scale bars, 25 um. Data presented as mean±sem. KO Knock-out, ADR Adriamycin. PAS Periodic acid–Schiff. SEM standard error of the mean. *p≤0.05, **p≤0.01, ***p≤0.001, and ****p≤0.0001.

High urinary CD93 is associated with unfavorable kidney outcomes in human INS

Because INS sera stimulated GEnCs to release CD93, we investigated the clinical value of soluble CD93. By ELISA, we first measured soluble CD93 in urine and serum from children with unbiopsied INS or MCD and control subjects (Discovery cohort, see “Study Participants” in Materials and Methods, table S1 and S2). Approximately 85% of patients in relapse and approximately 60% of patients in remission showed higher urinary CD93 levels (approximately 15 and 3-fold increase, respectively) than healthy controls (fig. S5A). Next, we measured urinary CD93 in 228 children and adults from the NEPTUNE cohort study (table S3) to assess validity, generalizability, and clinical value (29). Approximately 95% and 50% patients in relapse and remission, respectively, had higher urinary CD93 levels (approximately 30-fold and approximately 3-fold increase, respectively) than controls; and levels were higher in patients with FSGS than in those with MCD (Fig. 6A and fig. S5B). Additionally, higher urinary CD93 levels were associated with higher risk for kidney function loss and slower time to achieve clinical remission (Fig. 6B and C, table S4). Although the group of patients studied in remission was relatively small, those with higher urinary CD93 levels showed a trend toward higher risk to develop proteinuria over time (Fig. 6D, table S4). Urinary CD93 also showed a positive correlation with proteinuria and was inversely correlated with kidney function (fig. S5C and D, respectively). Additionally, urinary CD93 levels were approximately 2-fold higher in patients during relapse without immunosuppression compared to those on any immunosuppressive therapy (fig. S5E), but no differences were found in patients studied during remission (fig. S5F).

Fig. 6. Soluble CD93 is Associated with Unfavorable Kidney Outcomes in INS.

(A) Data shows urinary CD93, adjusted to urine creatinine, in healthy subjects and patients with INS (unbiopsied, MCD, FSGS) from the NEPTUNE study. Two patients were classified as in remission, 2 as active disease and 2 as relapse based on dipstick (negative/trace, 2+ and 4+ respectively) and not UPCR. Dotted line represents 95th percentile cut-off value for the healthy control group. Data presented as median±IQR. One-way ANOVA with Dunnett’s test was performed. (B-D) Data represents Kaplan-Meier curves for the relationship between urinary CD93 and clinical outcomes. Colored lines represents the proportion of patients with the studied outcomes according to urinary CD93 quartiles, unless stated otherwise. Numbers shown withing the graph represent the number of patients in each quartile. Kaplan Meier curves show unadjusted data. Adjusted Cox-proportional hazard models are shown in table S4. (B) Kaplan-Meier curve shows relationship between urinary CD93 and time to composite end-stage kidney disease or 40% decline in eGFR in patients with INS regardless of disease state (HR 1.27, CI [1.00, 1.61], p=0.04). (C) Kaplan Meier shows the relation between urinary CD93 and time to complete remission (HR 0.78, CI [0.65, 0.94], p=0.009). (D) Kaplan Meier shows the relationship between urinary CD93 and time to develop proteinuria (HR 1.33, CI [0.94, 1.89], p=0.10). Colored lines represent outcomes according to urinary CD93 levels either above or below to the median for patients in remission. M median value of urinary CD93, N number of participants, MCD minimal change disease, FSGS focal segmental glomerulosclerosis, HC healthy control, INS Idiopathic nephrotic syndrome, UPCR Urine protein-to-creatinine ratio, Prop. Proportion, IQR interquartile range, eGFR estimated glomerular filtration, HR hazard ratio, CI confidence interval. *p≤0.05, ****p≤0.0001.

Serum CD93 levels were higher in approximately 55% of patients with INS in relapse, including NEPTUNE and discovery cohorts, compared to those in remission or healthy controls (fig. S6A). Although approximately 20% of patients in remission had higher levels than controls, there were no statistical differences between these groups (fig. S6A) nor between patients with MCD and FSGS (fig. S6B). Although serum CD93 correlated with proteinuria and urinary CD93 (fig. S6C and D); there was no association with kidney function or immunosuppression use, time to remission, time to proteinuria or kidney disease progression (fig. S6E to J). Together, our findings identify urinary, not serum, CD93 as a molecule associated with unfavorable kidney outcomes in INS.

CD93 involvement is present across podocytopathies

Podocytopathies have distinct etiologies (autoimmune, inflammatory, genetic) but often share signaling pathways involving key proteins that ultimately modulate the podocyte cytoskeleton such as FAK and SRC (20, 46–49). Likewise, endothelial injury is common in podocytopathies (50–54), so we evaluated the specificity of CD93 across podocyte diseases. We measured soluble CD93 in urine (n=150) and serum (n=115) from patients with nephrotic-range proteinuria or nephrotic syndrome due to autoimmunity (membranous nephropathy, Ig A nephropathy, lupus nephritis), metabolic (diabetic kidney disease), and genetic condition (podocin mutation) (table S5). Most patients with severe proteinuria, regardless of the underlying kidney disease, showed urinary CD93 levels 10 to 50-fold higher than controls (Fig. 7A). Patients with podocytopathies, except for Ig A nephropathy, also showed higher serum CD93 levels than controls (Fig. 7B). Analysis of an RNA-seq dataset of micro-dissected glomeruli showed high CD93 expression in patients with IgA nephropathy, lupus nephritis, membranous nephropathy, and additionally, in the kidney tubulointerstitial compartment in patients with lupus nephritis (Fig. 7C). Analysis of single-cell RNA-seq data was available for patients with diabetic kidney disease and confirmed endothelial cells as the primary source of CD93 in the kidney (Fig. 7D to F, and fig. S7A and B) (55, 56). Thus, our data suggest that soluble and glomerular CD93 are also involved across podocytopathies.

Fig. 7. CD93 Involvement Across Podocytopathies.

(A) Data shows soluble CD93 levels measured in urine and in (B) serum samples, respectively, from healthy subjects (n=55 urine and n=50 serum samples) and patients with membranous nephropathy (MN, n=20 for both sample types), IgA nephropathy (IgAN, n=24 and 23), lupus nephritis in relapse (LN-NS, n=40 and 32) and diabetic kidney disease (DKD, n=20 for both sample types), and genetic nephrotic syndrome (n=26 and 20). Dotted line represents the 95th percentile cut-off value for the healthy control group. Data presented as median±IQR (Kruskal-Wallis test with Bonferroni correction was performed) for A and presented as mean±SD (one-way ANOVA with Tukey’s multiple comparisons test was performed) for B. (C) Data represents CD93 mRNA expression (fold change) in human kidney tissue according to anatomical compartment (glomerular versus tubulointerstitial compartment) and histological diagnosis. Number of participants were: IgAN-glomerular (27, with 21 controls), IgAN-interstitial (25, with 31 controls), LN-glomerular (32, with 14 controls), LN-interstitial (32, with 15 controls), MN-glomerular (62, with 8 controls), MN-interstitial (64, with 10 controls). t-test analysis was performed for each podocytopathy. Data obtained from www.nephroseq.org. (D-F) Analysis of publicly available single-cell RNA-seq from human kidney tissue in patients with DKD (55, 56). (D) Uniform Manifold Approximation and Projection (UMAP) plot showing CD93 enrichment across endothelial cell subsets in single-cell RNA-seq datasets from healthy controls and DKD. (E) Data shows the expressed percent and average CD93 expression in kidney endothelial cells from healthy controls and patients with DKD (n=5 and n=9, respectively). (F) Data shows CD93 differential expression from pooled endothelial cells from DKD (5 donors, 626 endothelial cells) and controls (9 donors, 1995 endothelial cells). Wilcoxon rank-sum test was performed. IQR interquartile range. *p≤0.05, **p≤0.01, ***p≤0.001 and ****p≤0.0001.

DISCUSSION

INS is viewed exclusively as immune-driven podocyte-specific disease and is characterized by variable clinical outcomes (1–9). Despite recent breakthroughs in the field(6–8), the precise mechanisms of podocyte injury remain unclear, and predictive markers of disease are needed (10). In this study, we identified soluble CD93 as an endothelial-released protein that contributes to podocyte activation. Our findings offer insights into the pathobiology of INS by identifying soluble CD93 as a candidate therapeutic target to mitigate podocyte injury.

Our study provides evidence that GEnCs play a role in the pathogenesis of INS. We demonstrated that CD93 is highly expressed by GEnCs in INS and is released from GEnCs upon stimulation with patient’s sera. Our experimental studies unveil a mechanistic pathway linking soluble CD93 with podocyte activation via β1 integrin/FAK signaling; and further demonstrated a causal link between soluble CD93 and albuminuria in vivo. Consistently, CD93 blockade mitigated albuminuria, glomerulosclerosis and podocyte loss, and CD93 KO mice were protected from ADR-induced albuminuria. Cd93 KO mice did not have increased albuminuria at baseline, suggesting that CD93 may not be necessary to maintain a healthy glomerular filtration barrier. DOX-induced PodTgfbr1 mice exhibited high urinary CD93 levels coinciding with the occurrence of endothelial injury and albuminuria but prior to the development of podocyte FPE described in this model (41). Thus, soluble CD93 may serve as an early marker of endothelial injury which could help identify patients who may benefit of CD93 blockade. Our findings are relevant given that current therapies for INS consist of non-targeted immunosuppressive drugs that are still used arbitrarily, and associate with variable efficacy and important side effects (57).

INS is widely viewed as podocytopathy (9). Our study provides evidence to support that, in addition to podocytes (6–8), GEnCs are also a cellular target in INS. We showed that INS sera stimulated GEnCs to release CD93 indicating that GEnC alternations in INS are not simply a secondary phenomenon. This aligns with the recent identification of anti-endothelium autoantibodies and elevated endothelial markers in circulation and glomeruli from patients with INS (11–13, 16). By combining single-cell RNA-seq and immunostaining, we documented the endothelial origin of CD93 in INS, and additionally, we showed that CD93 detected in urine (approximately 100-140 kilodaltons) matched the size of the CD93 released from cultured GEnCs. Although soluble CD93 can be released by some immune cells, predominantly monocytes (26), transcriptome analysis of peripheral blood mononuclear cells from patients with INS did not show differences in CD93 expression during active disease and following steroid treatment (58). Additionally, inflammatory cells are usually absent in the kidney tissue of patients with INS, particularly MCD (59). In contrast, inflammatory cells are plausible additional sources of soluble CD93 in patients with inflammatory podocytopathies who exhibit high levels of urinary CD163, a marker of monocyte activation and surrogate of kidney inflammation (60–64). Altogether, our findings identify GEnCs as important sources of soluble CD93 in INS, with immune cells as potential additional sources in inflammatory podocytopathies.

A histological hallmark of INS is podocyte injury with effacement of their foot processes (9). This is a non-specific phenotypic response of podocytes to diverse insults (immune, inflammatory, metabolic, genetic) which provides limited clinical or biological insights (15, 65), thereby indicating the urgency to identify additional markers. Our study, one of the largest involving patients with nephrotic syndrome, identified urinary CD93 as a candidate marker associated with unfavorable kidney outcomes in INS. One striking finding was the non-significant trend toward higher risk to develop future proteinuria in patients in remission with high urinary CD93. This finding suggests that the absence of proteinuria may not indicate that the disease process is fully curtailed and that the presence of CD93 in urine is not simply the consequence of the leaky glomerular filtration barrier. In fact, some patients with nephrotic range proteinuria showed normal urinary CD93 levels. In serum, CD93 levels were not associated with clinical outcomes in INS contrary to that reported in patients with type 2 diabetes mellitus (T2DM), vasculitis or asthma. This disparate results may be related to distinct cellular sources of soluble CD93 across diseases. Thus, the systemic endothelium and circulating monocytes may be important sources of soluble CD93 in certain conditions (T2DM, vasculitis, asthma) (66–68). In contrast, patients with INS show direct activation of GEnCs with high CD93 expression and release into the urine, and modestly into circulation. These findings indicate a primary glomerular origin of CD93 in INS. INS sera did not trigger the release of CD93 from HUVECs as opposed to GEnCs, suggesting the presence of circulating factors with affinity for GEnCs in patients with INS.

Our study has several limitations. Our clinical studies are limited to a single determination of CD93. Given the relapsing nature of INS, serial measurements of soluble CD93 may provide additional insights into the value of CD93 as predictive biomarker. But, before considering urinary CD93 as predictive biomarker, additional clinical studies involving several cohorts of patients with INS and other forms of nephrotic syndrome are needed. Our study lacks paired kidney/soluble CD93 data, so future studies are needed to determine whether urinary CD93 may be a surrogate of glomerular CD93 expression. Given the association of urinary CD93 with proteinuria, the latter may be a confounding factor in some patients. Future studies are also needed to investigate whether soluble CD93 may activate additional signaling pathways relevant for podocyte biology and to identify other candidate endothelial-released factors.

Collectively, our data suggests that CD93 upregulation in GEnCs represents a shared adaptive response to diverse insults to maintain endothelial integrity (23, 69, 70), which could explain its involvement across podocytopathies. Its release as soluble form likely reflects the severity of the injury and disease activity as reported in some conditions (66, 67). Consistently, patients with FSGS showed markedly higher urinary CD93 levels and a non-significant trend toward less CD93 expression in GEnCs than patients with MCD. In summary, soluble CD93 contributes to podocyte injury and emerges as candidate therapeutic target in INS.

MATERIALS AND METHODS

Study approval.

Human studies were approved by the Institutional Review Boards of all participating Institutions: Children’s Hospital Colorado (United States of America, USA) #13-2700, #16-1752, #152211, Rocky Mountain Kidney Center (USA) #13-2700, Nationwide Children’s Hospital (USA) # STUDY00003177, The Ohio State University (USA) #2000H0055, Hospital Universitario Central de Asturias (Spain) #221/19, Hospital Niño Jesus (Spain) #R-0011/20, Hospital Universitario Santa Lucia (Spain) #EO-39/21, Hospital San Juan de Dios (Spain) # PIC-138-19, and Kansai University (Japan) #2020084. We also obtained de-identified human samples via material transfer agreements from three different consortiums: 1) the Nephrotic Syndrome Study Network (NEPTUNE), a USA-based observational study that recruits patients with INS(29), 2) LEARNS consortium (Netherlands)(71), that involves a double-blind, placebo-controlled randomized controlled trial on the efficacy of levamisole in children with a first episode of INS, and 3) the PodoNet registry, that is focused on clinical, genetic and experimental research into hereditary diseases of the podocyte(72). Archived human kidney tissue was obtained from the Children’s Hospital Colorado and Hospital Arnau de Vilanova (Spain) #ID-RTF065. Written informed consents, and assents if appropriate, were obtained from parents of participants and participants before samples were collected in accordance with the Declaration of Helsinki.

Animal protocols and procedures were approved by IACUC at The Icahn School of Medicine at Mount Sinai (PodTgfbr1 and BALB/c-ADR models) and Nationwide Children’s Hospital (Cd93 KO mouse).

Study design.

Human studies include children and adults, male and female, with clinical diagnosis or biopsy-proven nephrotic syndrome as described in the section “Study participants”. The number of patients included in the study reflects the patients from whom a serum or urine sample was available from each participant center. For patients from the NEPTUNE cohort study, we only obtained one urine or serum sample, collected at 4 month of enrollment, to ensure feasibility. No patient was excluded from the study, and investigators were not blinded during data analysis. For mice studies, male and female were included to recapitulate human studies. No prior power calculations for sample size were performed. Sample sizes (n=4-14) experiments were based on previous work. The number of mice, including male and female, are shown in the corresponding figure legends. Each mouse was considered a biological replicate. When possible, investigators were blinded during acquisition of biosamples and analysis of microscopy images. However, in some instances, there were obvious differences in animals’ state of health between control and experimental group preventing complete blinding. All other experiments were not conducted blinded.

Study Participants.

Demographic characteristics are shown in table S1 (discovery cohort), table S2 (control participants), table S3 (NEPTUNE cohort study) and table S5 (other podocytopathies) in the Supplementary Materials. INS cohort: This includes 460 children and adults. All adults had biopsy-proven MCD or FSGS whereas most children were diagnosed according to clinical criteria reflecting worldwide clinical practice and current clinical guidelines. Discovery cohort: 128 patients, all children, were recruited from 8 centers across the world (Spain −4-, USA −3-, Japan −1-) and from 1 international consortium (LEARNS)(71). A single time urine and serum samples were obtained. Additionally, we obtained archived human kidney tissue from 34 patients with INS of whom 27 had MCD (20 adults, 7 children) and 7 FSGS (6 adults, 1 child), and from 9 patients without glomerular disease who served as “controls” (7 adults and 2 children) from Children’s Hospital Colorado. Longitudinal cohort (NEPTUNE): To study the clinical value of soluble CD93 in INS, we additionally studied 298 patients recruited by NEPTUNE(29). This is a prospective, observational study involving 38 sites across the United States. Specifically, we evaluated the relationship between soluble CD93 with; kidney disease progression including time to end-stage kidney disease (defined as initiation of dialysis, receipt of kidney transplant or estimated glomerular filtration rate [eGFR] less than 15 mL/min per 1.73m2) or 40% loss in eGFR, among all participants; time to remission among those patients with active disease; and time to develop proteinuria over time (“future proteinuria”) among those studied at time of disease remission (no proteinuria). In this cohort, 23 patients had a clinical diagnosis of INS, 127 had MCD and 148 had FSGS. We obtained a single urine and serum sample from 228 and 108 patients respectively. For analysis purposes, patients without biopsy (all children) were grouped with patients with MCD. Additionally, data on CD93 mRNA expression from micro-dissected kidney biopsy tissue was available in 153 of the 298 patients and from 53 additional controls. The latter included 8 living donors from the European Renal cDNA Bank, and 45 subjects who underwent total nephrectomy (University of Michigan) and were enrolled in the PRECISE study(73, 74). Other podocytopathies: This includes 150 patients with different types of podocytopathies including IgA nephropathy (n=24), membranous nephropathy (n=20), lupus nephritis (n=60), diabetic kidney disease (n=20) and genetic nephrotic syndrome (podocin mutation, n=26) from whom urine and serum samples were obtained once. In this group, individuals were recruited from The Ohio State University except for those with genetic nephrotic syndrome who were recruited by the international research consortium PodoNet(72).

Nephrotic range proteinuria or relapse were defined by urinary protein/creatinine ratio >2 g/g or 3+ or greater by urine dipstick, remission when <0.3 g/g or negative/trace dipstick; and active disease when ranged 0.3-2 g/g(57) (*8.84, correction factor to mg/mmol).

Because of constraints on patients’ samples, the number of urine or serum measurements available varied as reflected in the respective figures.

Control Participants:

This includes 52 children and 43 adults without history of glomerular or kidney disease from whom we obtain 50 serum (26 children and 24 adults) and 55 urine samples (36 children and 19 adults). Samples from all adults and from 24 children were obtained from the healthy donor biobank at Precision for Medicine (https://www.precisionformedicine.com). The remaining pediatric samples were collected from children evaluated at the Pediatric Nephrology Clinic at Children’s Hospital Colorado due to remote episode of urinary tract infection, kidney stone, extraglomerular hematuria, rhabdomyolysis, transient hypokalemia, spurious hyperkalemia, or congenital single kidney.

Sample collection and measurement of CD93

Urine samples from NEPTUNE consortium were collected with a protease inhibitor (Sigma-Aldrich# P1860, concentration 1.25 μL/mL) whereas rest of the urine samples were collected without any additives. Otherwise, human biosamples and cell supernatants were processed using standard protocols and stored at −80°C until testing. ELISA determination: Quantification of CD93 in blood (dilution 1:100), and cell supernatants (dilution 1:3) was performed using a CD93 ELISA Kit (R&D #DCD930, RRID:AB_2629490) according to manufacturer instructions. Blood samples were serum except in 25 patients and 5 control individuals in whom only plasma was available. The same ELISA kit was used to measure CD93 in undiluted urine (normalized to urine creatinine) after validation by western blotting (fig. S8). All ELISA measurements were run in duplicate. Immunofluorescence of human kidney tissue and cell culture methods are provided in the supplementary “Material and methods”.

Animal models

Mice were housed at 20–22 °C in individually ventilated cages, humidity controlled (55%) with free access to food and water, and a 12-hour light/dark cycle. Male and female mice were 8–11-week-old except for PodTgfbr1 mice (6–8-week-old). Mice were genotyped by polymerase chain reaction (PCR) for the presence of the corresponding transgenes (fig. S4 and table S6). PodTgfbr1 mouse: To generate double-transgenic (DT) mice for podocyte-specific, inducible TbrI(AAD) expression, we crossed DOX-inducible tet-O-TbrI(AAD) with NPHS2-rtTA mice, as previous described (41, 43). Single-transgenic (ST) were used as controls. For induction of TGFβR1 signaling, mice were fed chow containing doxycycline (DOX; 2000ppm) in regular AIN-76A pellets (Research Diets). BALB/c ADR mouse model: Mice were injected with a single dose of ADR (10 mg/kg; Sigma-Aldrich, St Louis, MO) or saline (control mice) via tail vein, and euthanized 6 days after injection. From day 1 to 5, mice received 2 daily 2-ml intraperitoneal (i.p.) injections of an isotonic solution (69.4 mM glucose, 77 mM NaCl) to prevent weight loss. CD93 blockade: A subgroup of PodTgfbr1 and BALB/C mice received a well-characterized CD93 neutralizing antibody (42) (200 mcg/mouse, i.p.) on day 0, 4, 8, and 12 and on day 0 and 2, respectively. Cd93 KO mouse: Cd93 KO (Cd93^−/−) mice were obtained from Dr. Anna Dimberg (Uppsala University) (24). Cd93 KO mice have a C57BL/6 background. Because C57BL/6 mice require a high ADR dose (25mg/kg) to induce albuminuria, we administered Cd93 KO, male and female, mice a high ADR dose via retroorbital injection (75). Control mice received saline via retroorbital injection. To prevent dehydration, mice received 0.5 ml of dialysate solution (Gambrosol Trio, Gambro AB) twice a day via i.p. injection from day 1 to 5, and mice were euthanized on day 6. Urine measurements: Urine albumin-creatinine ratio (uACR) was determined using mouse-specific albumin ELISA and creatinine companion kits (Ethos Biosciences, cat#1011 and #1012 respectively); and urinary CD93 was measured with commercial ELISA (ab#233620) and adjusted to urine creatinine.

Statistical analysis

Data were expressed as mean±standard deviation (SD), mean±standard error of the mean (SEM), and median±interquartile range (IQR) as appropriate. Categorical variables are presented as percentages. Differences among two groups were examined using unpaired two-tailed Mann-Whitney U test, unpaired t test and paired t test as appropriate. Comparisons of more than two groups were performed using one-way ANOVA with multiple comparisons as applicable. In urine, CD93 was analyzed as a log-transformed variable. Time to event analyses were performed using cox-proportional hazards models and Kaplan-Meier plots for visualization. Three different analyses were performed: time to remission (i.e., urinary protein-to-creatinine ratio [UPCR]<0.3g/g) among those not in complete remission at the time of CD93 assessment; Time to kidney disease progression (i.e., reaching end-stage kidney disease or 40% reduction in eGFR); and time to return of proteinuria (UPCR>0.3g/g) among those in complete remission at time of CD93 assessment. Analyses were also repeated for the following cohorts: FSGS+MCD, FSGS only, and MCD only. For each cohort-outcome, we fit a multivariable cox-proportional hazards models with the following covariates: log-uCD93/creatinine, histology (FSGS vs. MCD—variable not included for FSGS-only and MCD-only analyses), log-UPCR, interstitial fibrosis (%), and APOL1 (high risk vs. not high risk). Backwards selection was performed by removing non-significant variables from the model in descending order of p-value to preserve precision. Multiple imputation based on fully conditional specification was used to handle missing data. Spearman’s correlation was used to study associations. Quantification of fluorescence was performed using ImageJ. Analyses were performed using GraphPad Prism (version 9, GraphPad Software) and STATA, v12.1.

Data and materials availability:

All data associated with this study are present in the paper or supplementary materials. The RNA seq data discussed in this publication were previously published by other investigators and deposited in NCBI’s Gene Expression Omnibus. GEO Series accession numbers are provided in the supplementary materials as reference. PodTgfbr1 and Cd93−/− mice may be available through a material transfer agreement upon contact with the corresponding author.