Esm-1 mediates transcriptional polarization associated with diabetic kidney disease

Abstract

Background Esm-1, endothelial cell-specific molecule-1, is a susceptibility gene for diabetic kidney disease (DKD) and is a cytokine- and glucose-regulated, secreted proteoglycan, that is notably expressed in kidney and attenuates inflammation and albuminuria. Esm1 has restricted expression at the vascular tip during development but little is known about its expression pattern in mature tissues, and its precise effects in diabetes.

Methods We utilized publicly available single-cell RNA sequencing data to explore the characteristics of Esm1 expression in 27,786 renal endothelial cells obtained from four adult human and three mouse databases. We validated our findings using bulk transcriptome data from an additional 20 healthy subjects and 41 patients with DKD and using RNAscope. Using correlation matrices, we relate Esm1 expression to the glomerular transcriptome and evaluated these matrices with systemic over-expression of Esm-1.

Results In both mice and humans, Esm1 is expressed in a subset of all renal endothelial cell types and represents a minority of glomerular endothelial cells. In patients, Esm1(+) cells exhibit a highly conserved enrichment for blood vessel development genes. With diabetes, these cells are fewer in number and profoundly shift expression to reflect chemotaxis pathways. Analysis of these gene sets highlight candidate genes such as Igfbp5 for cross talk between cell types. We also find that diabetes induces correlations in the expression of large clusters of genes, within cell type-enriched transcripts. Esm1 significantly correlates with a majority genes within these clusters, delineating a glomerular transcriptional polarization reflected by the magnitude of Esm1 deficiency. In diabetic mice, these gene clusters link Esm1 expression to albuminuria, and over-expression of Esm-1 reverses the expression pattern in many of these genes.

Conclusions A comprehensive analysis of single cell and bulk transcriptomes demonstrates that diabetes correlates with lower Esm1 expression and with changes in the functional characterization of Esm1(+) cells. Esm1 is both a marker for glomerular transcriptional polarization, and a mediator that re-orients the transcriptional program in DKD.

Introduction

Diabetic kidney disease (DKD) is a common complication of diabetes mellitus, with a prevalence of ∼15% and is associated with poor long-term outcomes^1–3. Chronic inflammation is a major contributor to kidney disease in diabetes^4–12. Macrophage infiltration in the kidney correlates with the development of DKD^13–15, while constitutive deletion of lymphocytes, cell adhesion and chemoattractant mediators alleviate the severity of DKD in mice^{7–9, 16}. Thus, reducing inflammation may be a viable strategy to improve renal outcomes in patients with diabetes. However, there are no available kidney-enriched immune modulators.

Endothelial cell-specific molecule-1 (Esm-1), is a potential kidney-enriched immune modulator. Esm-1 is an endothelial-secreted circulating proteoglycan, notably expressed by the human glomerulus^{17, 18}. Esm-1 was initially described as an inhibitor of the interaction between the leukocyte integrin LFA-1, and its endothelial ligand ICAM-1¹⁹, thus explaining its likely anti-inflammatory role in acute and chronic inflammatory conditions²⁰. Esm-1 may attenuate inflammatory processes in DKD, as suggested by the higher Esm1 expression levels observed in the glomeruli of mice resistant to albuminuria and DKD¹⁸. Our group has addressed the contribution of Esm-1 to the development of renal inflammation associated with DKD by using complementary gain- and loss-of-functional approaches. We found that in diabetic, Esm-1 deficiency mice, add back of Esm-1 induces protection from podocyte loss and albuminuria²¹. Several studies have also outlined a role for Esm-1 as a mediator of angiogenesis with a restricted expression to a minority of cells at the vascular tip^22–25.

The precise anatomical and functional characteristics of Esm1(+) endothelial cells have not been described, and little is known about the role of this cell population in the mature kidney or in the development of DKD. We hypothesize that Esm1 delineates a distinct subset of specialized cells in the glomerular vasculature, and that loss of Esm1 expression in diabetes may also determine the transcriptional changes occurring in endothelial and neighboring glomerular cell types to mediate phenotypic changes in DKD.

In the present study, using single-cell RNA-sequencing, microarray data, and Esm-1 over-expression in an experimental model of DKD, we investigate the abundance and functional specialization of the Esm1(+) cell population in human and mouse mature kidney, we explore the influence of diabetes on its characteristics, and finally assess the effects of Esm-1 on the transcriptional and phenotypical features of DKD.

Methods

Bioinformatic analysis

Data collection and cluster identification

We downloaded gene raw counts or normalized gene expression matrix for each single-cell library from the supplemental material in Barry et al, 2019²⁶, GEO databases (GEO Accession No. GSE129798; GEO Accession No. GSE131882; GEO Accession No. GSE140989), from the REC Dehydration database (https://endotheliomics.shinyapps.io/rec_dehydration/) or from the Human Kidney Cell Atlas (https://cellgeni.cog.sanger.ac.uk/BenKidney_v2.1/Mature_non_PT_parenchyma_v2.1.h5ad, mature human kidneys) (Table S1). For the glomerulus wide-genome expression profiling, we downloaded normalized processed data from GEO (GEO Accession No. GSE96804) (Table S2).

For the REC (Renal Endothelial Cell) Dehydration database (Database M1²⁷), we downloaded normalized processed data filtered for high-quality cells from control mice. We then identified glomerular, cortical and medullary renal endothelial cells (gRECs, cRECs and mRECs) according to cluster annotations provided with the database. For the database from Barry et al, (Database M2²⁶), we downloaded the raw counts of the scRNA-seq data, and following the criteria from the original publication, we analyzed cells that had between 200 and 2500 detected genes for downstream analyses to filter for high-quality single-cells²⁶. We selected adult kidney cells according to cluster annotations provided with the database and normalized the expression matrix using NormalizeData with a scale set at 10,000 in the Seurat R package. We identified gRECs, cRECs and mRECs according to the expression of their top 50 marker genes on t-SNE projections using the Multigene expression tool in BD Genomics Dataview (version 1.2.2). For GSE129798 (Database M3²⁸), we downloaded normalized processed data filtered for high-quality cells. We selected RECs according to the expression of the top 50 marker genes of RECs on t-SNE projections using the Multigene expression tool in BD Genomics Dataview. From the genesets we obtained the top 50 marker genes of RECs from the supplementary material in Ransick et al, 2019²⁸. We identified cells from the Z2 and Z3 anatomical regions as mRECs. For RECs of the Z1 anatomical region, we identified gRECs and cRECs according to the expression of their top 50 marker genes on t-SNE projections using the Multigene expression tool in BD Genomics Dataview. We obtained the top 50 marker genes of gRECs and cRECs from the supplementary material in Dumas et al, 2020²⁷.

For GSE131882, we downloaded index read count expression matrix from control (Database H1²⁹) and diabetic subjects (Database H4²⁹) and analyzed genes expressed in >3 nuclei and nuclei with at least 500 genes to filter data for high-quality cells, following the criteria from the original publication²⁹. We did not retain data from Control 2 and Diabetes 2 subjects because of a high rate of nuclei not passing quality filtering. We then normalized the expression matrix using NormalizeData with a scale set at 10,000 in the Seurat R package (version 3.1.3) and restricted downstream analysis to RECs according to the expression of the canonical REC marker genes PECAM1 and FLT1 on t-SNE projections in BD Genomics Dataview. We then identified the various clusters of RECs (glomerular capillaries, afferent and efferent arterioles / peritubular capillaries, ascending and descending vasa recta and other ECs) according to the expression of the top 50 marker genes on t-SNE projections using the Multigene expression tool in BD Genomics Dataview. We labeled glomerular capillary cells as the gRECs subset, and every other EC as the cRECs subset. We obtained the top 50 marker genes of the REC clusters from the supplementary material in Lake et al, 2019³⁰. For the Human Kidney Cell Atlas database (Database H2³¹), we downloaded normalized processed data filtered for high-quality cells. We selected RECs for downstream analysis and identified the various clusters (glomerular endothelium, peritubular capillary endothelium, ascending and descending vasa recta endothelium) according to the supplementary material in Stewart et al, 2019³¹. We labeled glomerular endothelium cells as gRECs, peritubular capillary endothelium as cRECs and ascending and descending vasa recta endothelium as mRECs. For GSE140989 (Database H3³²), we downloaded normalized processed data filtered for high-quality cells. We selected RECs for downstream analysis and identified the various clusters (glomerular capillaries, peritubular capillaries, and arterioles afferent and efferent) according to cells annotations provided by the authors of the original publication³². We then normalized the expression matrix using NormalizeData with a scale set at 10,000 in the Seurat R package to adjust the normalization scale to that of other databases. Genesets of the top 50 marker genes used for the identification of each cluster in single-cell databases are described in Table S3^{27, 30}. For GSE96804, we downloaded normalized processed data of glomerulus expression profiling on the Affymetrix microarray platform. For each subject, we obtained memberships to control (Database H5³³) or diabetic (Database H6³³) groups from the annotations provided with the database.

The main characteristics of each database are shown in Tables S4 (single-cell) and S5 (gene expression profiling).

t-SNE projections

We processed normalized data in BD Genomics Dataview to generate t-SNE projections. We used the select variable genes function to reduce the number of genes to a subset of most variable genes. We calculated the dispersion of log-transformed gene expression data for each gene and divided the genes into a number of bins based on mean gene expression per cell. In each bin, we calculated the z-score of the dispersion of each gene and retained genes with a z-score exceeding the dispersion threshold. We filtered out cells which had no molecule counts for the retained set of genes. We then calculated t-SNE coordinates on log-transformed data with a pseudo-count of 1 added using the bh-t-SNE function to construct a two-dimensional representation of the data (see Table S6 for parameter settings for each database).

Single-cell expression analysis

We performed expression analyses on counts matrices scaled by total counts for each cell, multiplied by 10,000 and transformed to log space. We visualized the expression of Esm1 and canonical genes of REC clusters on t-SNE-projections using the Gene expression in BD Genomics Dataview. We assessed the expression of canonical genes for various REC clusters in Esm1(+) cells through heatmap analysis. We produced heatmaps using the ggplot2 R package (version 3.3.0) on the basis of average log-transformed expression of the top 50 available marker genes (excluding Esm-1) according to previously published gene sets (see Table S3 for the detailed list of gene sets). We used the ggplot2 R package for visualization of log-transformed gene expression through violin plots.

Differential expression analysis

We used the differential expression function in BD Genomics Dataview to perform a negative binomial test on linear counts between Esm1(+) and Esm1(-) cells for all genes in each database.

Trajectory inference

We used the STREAM package in Python³⁴ with default parameters to place cells onto pseudotime trajectories. We used a non-parametric local regression method (LOESS) to fit the relationship between mean and standard deviation values and selected the genes above the curve that diverged significantly as variable genes³⁵. We performed dimensionality reduction using the Modified Locally Linear Embedding (MLLE) method³⁶. We imported vasculature compartments as input data and subsequently visualized each cell membership onto linear pseudotime plots along with log-transformed expression of Esm-1. We subsequently compared scaled expression levels by Mann Whitney U tests.

Total RNA expression analysis

For the gene expression profile analysis of the human glomerulus (Database H6), we displayed the distributions of gene expression as boxplots using the ggplot2 R package and compared gene expressions between groups using the limma R package implementation in the Geo2R web tool.

For the gene expression profile analysis in mice with STZ-induced diabetes, we used the DEseq2 R package to process the data. We performed a shifted logarithm transformation of counts normalized on size factors to generate a log transformed count matrix for correlations analysis. We used the rlog function to generate a regularized log transformed matrix for Differential expression analysis.

Correlations analysis

We used the corrplot function from the corrplot R package to display correlation matrices between gene sets. When indicated, we performed hierarchical clustering of genes order using the hclust method to identify clusters of correlated genes. We used the ggplot2 R package to display heatmaps of correlations.

We compared the magnitude of the correlation patterns between the H5 and H6 databases by retaining the number of correlations with an arbitrarily fixed absolute r-value greater than or equal to 0.8 among all tested gene combinations.

For the untargeted correlation analyses in database H6, we retained genes with a q-value < 0.05 for differential expression in endothelial, mesangial and podocytes clusters from the supplementary material in Lake et al³⁰ and in dendritic cells and macrophages from the supplementary material in Stewart et al³¹.

For the untargeted correlation analyses of the gene expression profiling in streptozotocin-induced DKD in mice, we retained genes with a q-value < 0.05 for differential expression in endothelial clusters from the supplementary material in Park et al³⁷ and Ransick et al²⁸, in podocytes and macrophages clusters from the supplementary material in Park et al³⁷ and in mesangial clusters from Karaiskos et al³⁸.

For selected genes of interest, we calculated and displayed linear regressions with 95% CI on scatter plots using the lm method in the ggplot2 R package. For analysis of the IGFBP5 interactome, we identified candidate genes using the Biogrid 3.5.185 website.

Gene and ontology enrichment analysis

We used the online software Metascape (http://metascape.org/) tool³⁹ to perform single and multiple gene sets overlap analysis and identify enriched gene annotations across databases. We ran this analysis with default parameters to infer a hierarchical clusterization of significant ontology terms into a tree based on Kappa-statistical similarities among gene memberships. We selected the term with the best q-value within each cluster as its representative term. We used the 50 genes with top positive log fold-changes in counts between Esm1(+) and Esm1(-) cells and with a q-value of less than 0.05 as a multi-list input for Metascape enrichment analysis comparing databases H1, H2, H3, M1, M2, and M3. We used all differentially expressed genes in Esm1(+) cells with a q-value of less than 0.05 as single inputs for Metascape enrichment analysis comparing databases H1 and H4. For Metascape enrichment analysis of database H6, we used the genes respectively included in the negative and positive Esm1 correlation clusters and with a q-value < 0.05 as a multi-list input. For Metascape analysis of the gene expression profiling of mice with STZ-induced DKD, we used the whole sets of genes belonging to the indicated clusters as single inputs (see Table S7 for gene lists used in Metascape analysis).

Gene lists overlap

We used the UpSetR Shiny web application (https://gehlenborglab.shinyapps.io/upsetr/)⁴⁰ with standard parameters to visualize overlapping between lists of genes.

Human samples

We obtained human biopsy specimens from neo-neoplastic tissue from nephrectomy specimens from a biorepository of patients with and without diabetes mellitus. We matched representative non-diabetic controls and diabetic samples for age and sex at the time of nephrectomy (Table S8).

Animal models

Induction of diabetes and over-expression of circulating Esm1

We induced diabetes in eight week-old male DKD-susceptible (DBA/2) mice and used hydrodynamic tail-vein injection to over-express Esm1 as detailed elsewhere⁴¹. Data from this model obtained in diabetic mice without Esm1 over-expression (Database M4) and with Esm1 overexpression (Database M5) are accessible from Geobank with the accession number GSE175449.

Quantification of Albuminuria

We collected urine from mice in singly housed in metabolic cages (Tecniplast, Italy) as detailed elsewhere²¹.

RNA extraction for RNAseq analysis

We extracted kidney glomeruli from control, diabetic, and diabetic mice with Esm-1 over-expression for parallel analysis, as detailed elsewhere⁴¹. We isolated glomerular RNA by the Zymoresearch RNA preparation micro kit (Zymoresearch, Irvine, CA). We removed genomic DNA by RNase-free DNase I (Fermentas, Waltham, MA). We then reversed transcribed RNA into cDNA by using Improm II reverse transcriptase (Promega, Madison, WI) and random hexamer primers (Genelink, Elmsford, NY). We submitted cDNA samples to CD Genomics (New York, NY) for DNA quality control, library preparation, and sequencing (Illumina Novaseq PE150 sequencing, 20M read pairs).

In situ hybridization via RNAscope

We studied localization of target mRNA using the RNAscope Multiplex Fluorescent v2 kit (Advanced Cell Diagnostics, Newark, CA), according to the manufacturer’s instructions. We detected hybridization signals on 5-mm formalin-fixed paraffin-embedded transverse kidney sections using the TSA Plus fluorophores Opal 520 and Opal 690, with respective dilutions at 1:750 and 1:1500 (Akoya Biosciences, Waltham, MA). We mounted slices with ProLong Gold Antifade Mountant (Thermo Fisher Scientific, Waltham, MA) and viewed with a Leica DM5000 (Leica Microsystems, South San Francisco, CA). We included tissue from Esm1 knockout mice²¹ and dapBcontrol probes as negative controls. Availability of probes limited our analysis. Emcn, and Flt1 were the best glomerular capillary marker genes for whom an available probe could be used in mouse and human sections, respectively. Genesets of the top 50 marker genes used for the identification of cell subpopulations are described in Table S3^{27, 30}.

Study Approvals

The Stanford University School of Medicine Institutional Review Board (IRB) approved the collection and storage of human samples and clinical data from the Stanford Pathology Biobank, including a waiver of informed consent. The Stanford Research Compliance Office Administrative Panel for the Protection of Human Subjects – IRB approved the use of stored human biopsy samples. The Stanford Institutional Animal Care and Use Committee approved the experiments in mice.

Statistical analysis

Details regarding the number of samples for each statistical comparison are provided in Table S4. For single-cell analysis, statistical significance was assessed with Chi2 square test using the chisq.test function in R when comparing categorical data. We performed binomial regressions on linear counts using the differential expression function in BD Genomics Dataview to compare single-cell gene expression levels. For the glomerulus gene expression profiling analysis, unpaired t-test were performed using the GEO2R web implementation of the limma R package with default settings. We run Spearman tests using the corr.test function from Psych R package to assess the correlation between two continuous variables. A bilateral p-value < 0.05 was considered statistically significant for all tests. When indicated, we calculated q-values using the q-value function from the q-value R package (version 2.18.0) with default parameters.

Results

Esm1 is restricted to defined subsets of renal endothelial cells in mice and humans

To understand the features associated with the variation of Esm1 expression in DKD, we first explored the characteristics of Esm1 expression in healthy kidney vasculature. In mouse kidney, we first compared the expression of Esm1 with canonical genes of vasculature compartments. Esm1 is expressed in 12% to 18% of gRECs, 42% to 77% of cRECs, and 14% to 38% of mRECs (Figure 1A and Table S4). Subsequently, t-SNE projections show predominant expression of Esm1 in regions enriched for Igfp3, Igfbp5 or Plpp3, three markers of cRECs²⁷ (Figures S1A-C). Congruent with this observation, Esm1(+) cells express higher levels of cRECs markers Igfbp3, Igfbp5 and Plpp3 (Figure S1D). Heatmap analysis of the top 50 marker genes of vasculature compartments confirms the preferential expression of cRECs markers (Igfbp3, Igfbp5, AW112010, H2-K1, Rsad2, Kdr, Gpihbp1) in Esm1(+) cells (Figure 1B). We performed localization of Esm1(+) cells by RNAscope. We observe preferential distribution of Esm1(+) cells in peritubular capillaries. We are also able to identify modest expression of Esm1 in glomeruli (Figure 1C). However, further analysis reveals heterogeneity in transcriptomic signatures associated with the expression of Esm1 within each compartment. Indeed, marker genes of afferent arterioles (Sat1, Tmsb4x, Fabp4, Bst2, Ly6e, H2-K1) and from the terminal portion of the afferent arterioles associated with the juxtaglomerular apparatus (Tmem167b, Dll4, Cd300lg, Ifi44) are co-expressed with Esm1 in gRECs. In contrast, cells with glomerular capillary markers were not enriched for Esm1 expression (Figure S1E). Furthermore, in cRECs, Esm1 is mostly co-expressed with the top markers of cortical capillaries (Ly6c1, Itm2b, Il10rb, Igfbp3, Igfbp5, Nrp1, Cyp26b1), as well as markers of capillary angiogenic (Gpihbp1, Fabp4, Sparc, Cd300lg) and interferon markers (Ifi44, Bst2, Rsad2) (Figure S1F). Similar to cRECs, most capillary angiogenic (Gpihbp1, Sparcl1, Fabp4, Sparc) and interferon markers (Rsad2, Ifi3m, B2m) are enriched in Esm1(+) mRECs (Figure S1G).

Figure 1. Esm1 is expressed in various vasculature subsets including glomerular endothelial cells in mice.

Data are from mice single-cell databases M1, M2, M3 for (A) and (B), and from C57BL/6 mouse kidney sections for (C). (A) t-SNE plots of RECs, are color-coded for the expression of Esm1 (upper panel) and for vasculature compartments (lower panel). In the upper panel, no expression (grey), low (light purple) and high (dark purple) expression of Esm1 are shown. In the lower panel, glomerular (g, green), cortical (c, blue), and medullary (m, yellow) are shown. (B) Heatmap of the relative fold-change in log-transformed expression-level of top 50 marker genes of mouse kidney vasculature compartments in Esm1(+) vs. Esm1(-) RECs. For heatmaps, blue is lower expression in Esm1(+) cells, red is higher expression in Esm1(+) cells, white is same expression in Esm1(+) and Esm1(-) cells. (C) RNAscope for Esm1 (red) and Emcn (green) in the renal cortex. White arrows show glomerular expression of Esm1. REC, renal epithelial cells.

We then explored the characteristics of Esm1(+) cells in human kidney by comparing Esm1 expression with that of canonical markers genes in vasculature compartments. Esm1 is expressed in 23% to 51% of gRECs, vs. 13% to 33% of cRECs and 8% of mRECs (Figure 2A and Table S4). Accordingly, Emcn and Syne1, two glomerular capillaries markers, track with Esm1 expression on t-SNE projections (Figure S2A) and are found at higher levels of expression in Esm1(+) cells in databases H2 and H3 (Figure S2F). Furthermore, we find enrichment of the top 50 glomerular capillaries markers (EMCN, SYNE1, PTPRB, LDB2, FLT1, HSPA1A) in Esm1(+) cells (Figure 2B). In human renal cortex, we observe that Esm1 is mostly expressed in glomeruli, though modest expression is found in extra-glomerular cRECs (Figure 2C). These data show that glomeruli are the predominant source of renal Esm1 expression in humans, in contrast to the mouse kidney where this compartment accounts for a minority of Esm1 expression.

Figure 2. Esm1 is mainly expressed in glomerular endothelial cells in humans.

Data are from human single-cell databases H1, H2, H3 for (A) and (B), and from human kidney sections for (C). (A) t-SNE plots of RECs, are color-coded for the expression of Esm1 (upper panel) and for vasculature compartments (lower panel). In the upper panel, no expression (grey), low (light purple) and high (dark purple) expression of Esm1 are shown. In the lower panel, glomerular (g, green), cortical (c, blue), and medullary (m, yellow) are shown. (B) Heatmap of the relative fold-change in log-transformed expression-level of top 50 marker genes of human kidney vasculature compartments in Esm1(+) vs. Esm1(-) RECs. For heatmaps, blue is lower expression in Esm1(+) cells, red is higher expression in Esm1(+) cells, and white is similar expression in Esm1(+) and Esm1(-) cells. (C) RNAscope for Esm1 (red) and Flt1 (green) in the human renal cortex. White arrows show glomerular expression of Esm1. REC, renal epithelial cells.

Genes co-expressed with Esm1 differ across species, but are part of similar pathways

Esm1 is not enriched in the same vascular compartments in mice and humans. Thus, we aimed to assess whether subsets of cells expressing Esm1 are defined by common shared functions. Thus, we compared the top 50 marker genes in Esm1(+) (vs. Esm1(-)) cells in mouse and human kidney samples. We find that 13 of the Esm1 co-expressed genes are shared between the three mouse gene lists, and 39 genes are found in common in at least two mouse gene lists. No co-expressed genes are shared between all three human gene lists; however, 12 genes are found in common in at least two human gene lists. Finally, 16 genes are shared across one mouse gene list and one human gene list. By contrast, gene ontologies are consistent across databases and species (Figure 3A). Four ontologies clusters, i.e., blood vessel development, epithelial cell proliferation, cellular response to growth factor stimulus, and morphogenesis of an epithelium, are significantly enriched in all mouse and human databases (detailed genes lists are shown in Table S7). Among these ontologies, blood vessel development is the most significantly enriched term, with 22% to 38% of the hits from each gene list falling into that term (enrichment score: 4.2 to 9.0; common q-value = 2.51*10^-12) (Figure 3B). These four ontologies clusters are located in the same region on the gene ontology network, with a high similarity in ontology terms, as reflected by the number and thickness of edges linking each cluster to another (Figure 3C). We found several cluster nodes to be enriched for all human and mouse databases in each cluster node (Figure 3D) (see Supplementary File 1 for interactive visualization of detailed enriched ontologies networks in Cytoscape). These results underline the biological processes shared by Esm1(+) cells, across multiple databases and species.

Figure 3. Esm1(+) cells share common functions across species.

(A) Circos plot showing genes and annotations overlaps between gene lists expressed in Esm1(+) cells from each of the three mouse (M1-M3) and human (H1-H3) databases. Outer arcs represent the identity of each gene list. We represented each gene from a gene list as a spot on an inner arc. Genes that appear in multiple lists (black) and genes that are unique to the indicated list (white) are noted under each bar. Grey lines link the same gene that are shared by multiple gene lists. Black lines link the different genes falling into the same ontology term if statistically significantly enriched and with size no larger than 100. (B) Heatmaps of enrichment of ontology clusters, showing the hypergeometric q-values, enrichment scores and percentages of genes falling into each term for the top 20 most significantly enriched ontology cluster in each gene list. We performed a hierarchical clustering of significant ontology terms into a tree based on Kappa-statistical similarities among their gene memberships. We selected the term with the best q-value within each cluster as its representative term and display it in the heatmaps. Grey cells represent the non-significantly enriched ontology clusters in each gene list. The percentage of genes (green), enrichment score (blue), and q-value (orange) are shown for each ontology. (C) Network of enriched ontology terms. Each circle node represents a term, with the node’s size being proportional to the number of input genes falling into that term, and its color represents its ontology cluster identity, defined as the term with the best q-value within each cluster. Terms with a similarity score > 0.3 are linked by an edge, with the thickness of the edge proportional to the similarity score. The network is visualized with Cytoscape (v3.1.2) with “force-directed” layout and with edge bundled. (D) Network of enriched ontology terms with nodes displayed as pies. Each pie sector is proportional to the number of hits originated from a gene list. Color code for pie sector represents a gene list.

In diabetes, Esm1(+) cells undergo changes in relative Esm1 expression, abundance, and identity

To better understand the relationship between glomerular expression of Esm1 and diabetes, we quantified the glomerular expression of Esm1 in DKD-sensitive vs. resistant mice. Consistent with prior studies¹⁸, we observe lower glomerular intensity of Esm1 in the DBA/2 DKD-sensitive, compared to the C57BL/6 DKD-resistant strain (average MFI: 0.09 vs. 0.14, p < 0.05) (Figure S3A). We also find lower intensity of Esm1 from glomeruli of diabetic mice, compared to non-diabetic controls (average MFI: 0.03 vs. 0.09, p < 0.05) (Figure S3B). To verify these results in humans, we measured glomerular Esm1 mRNA in 20 healthy controls and 41 patients with diabetes and find significantly lower expression with diabetes (log fold-change = −2.46, p = 4.06*10^-9) (Figure 4A). In a separate cohort, we then quantified glomerular Esm1 staining. Similar to results from mice, we observe reduced expression of human Esm1 in diabetic patients, compared to healthy controls (average MFI: 0.01 vs. 0.21; p < 0.05) (Figure 4B).

Figure 4. Glomerular expression of Esm1 is decreased in diabetes and is associated with changes in specialization of Esm1(+) cells.

(A) Total RNA glomerular expression of Esm1 in 20 control and 41 diabetic subjects. Boxplots represent median [IQR]. (B) Quantification by RNAscope of human Esm1 glomerular expression. (C) Violin plots showing the single-cell expression profiles of Esm1 in human kidney vasculature compartments in control and diabetic subjects. Percentages of cells expressing Esm1 are shown below violin plots. (D) Heatmap of the relative fold-change in log-transformed single-cell expression-level of top 50 marker genes of human kidney vasculature compartments in Esm1(+) vs. Esm1(-) RECs. (E) Visualization of inferred branching points, pseudotime trajectories, and expression of Esm1 in diabetic and control subjects. We colored each zone according to cell cluster annotations (upper panel) or based on the expression of Esm1 (lower panel). We annotated branching points and extremities arbitrarily. At a given pseudotime, the width of each branch is proportional to the total number of cells. In the lower panel, magnitude of Esm1 expression is indicated by the scale provided. Data are from human genome expression profile databases H5 and H6 for (A), from 5µm human kidney sections for (B), and from human single-cell databases H1 and H4 for (C), (D) and (E). REC, renal epithelial cells. * p < 0.05 as indicated. ^a p < 0.05 vs. All RECs Control. ^b p < 0.05 vs. gRECs Diabetes. ^c p < 0.05 vs. cRECs Control. ^d p < 0.05 vs. gRECs Control.

To further characterize the transcriptional adaptation of human Esm1(+) cells to diabetes, we investigated the characteristics of Esm1 single-cell expression in healthy controls and diabetic patients from two databases. Compared to cRECs, per-cell Esm1 expression is higher and is detected in a greater percentage of cells in gRECs, both from controls (log fold-change = 0.09, p = 5.82*10^-4; 23% vs. 13%, p = 0.005) and diabetic patients (log fold-change = 0.11, p = 2.98*10^-12; 17% vs. 3%, p = 1.86*10^-6) (Figure 4C). We find significant decreases in expression of Esm1 (log fold-change = −0.079; p = 6.55*10^-25) and % of Esm1(+) cells (10% vs. 17%; p = 0.004) from patients with diabetes, compared to healthy controls. Canonical markers of glomerular capillaries are also highly expressed in the Esm1-enriched region on t-SNE projections (Figure S4A), and in Esm1(+) cells (Figure S4B). Furthermore, Esm1(+) cells express most of the top 50 marker genes of glomerular capillaries (EMCN, NRG3, SYNE1, LDB2, FLT1, GRB10), by contrast with marker genes of other vasculature compartments (Figure 4D).

Additionally, we analyzed a pseudotime projection to infer the trajectory of cell specialization in the kidney vasculature. In diabetes, we observe most of Esm1 expression at the extremities of the S0S1 and S0S4 diverging branches, delineating two subsets of specialized cells, mostly corresponding to glomerular capillaries (Figure 4E, Figure S5A and Table S9). In control subjects, Esm1 is expressed in the S0S1 and S1S3 branches, two regions showing mixed enrichment for glomerular and cortical endothelial cells. Importantly, in contrast to the distribution observed in diabetes, Esm1 is also expressed in the transition region connecting these two branches (Figure 4E, Figure S5B and Table S10). Taken together, these results demonstrate a reduction and restriction of Esm1 expression to specialized subsets of glomerular endothelial cells in diabetes, compared with controls.

Diabsetes is also associated with a shift in the transcriptomic signature of Esm1(+) cells

To assess whether the lower abundance of Esm1(+) cells in diabetes is associated with a shift in function, we characterized the transcriptomic signature of these cells. Esm1 expression is associated with a significant enrichment of 89 genes in healthy controls and 181 genes in patients with diabetes, with only seven genes overlapping between these two conditions (Figure 5A). While vascular development represents the ontology with the greatest number of Esm1(+) cell-enriched genes in controls, with 21 genes (enrichment score = 13; q-value = 1.4*10^-6), chemotaxis has the most number of enriched genes in patients with diabetes, with 20 genes (enrichment score = 4.5, q-value = 1.26*10^-5). The ‘positive regulation of vascular development’ ontology is not significantly enriched in Esm1(+) cells in diabetes (Figure 5B). Furthermore, we observe lower expression for most of the vascular development related genes in patients with diabetes, mostly driven by a lower expression in Esm1(+) than higher expression in Esm1(-) cells (Figure 5C). Conversely, in diabetes, we observe an increased enrichment in Esm1(+) cells for most genes in the chemotaxis ontology, due to higher expression in Esm1(+) cells and lower expression in Esm1(-) cells (Figure 5D). Taken together, these data underline a shift in the transcriptional program of Esm1(+) cells with diabetes.

Figure 5. Diabetes is associated with a shift in the transcriptomic signature of Esm1(+) cells.

(A) Venn diagram of the overexpressed genes in Esm1(+) cells in diabetic and control subjects. (B) Heatmaps of enrichment of ontology clusters, showing the hypergeometric q-values, enrichment scores and percentages of genes falling into each term for the top 20 most significantly enriched ontology cluster in each gene list. We performed a hierarchical clustering of significant ontology terms into a tree based on Kappa-statistical similarities among their gene memberships. We selected the term with the best q-value within each cluster as its representative term and display it on the graph. We ordered the ontology clusters by bar length representing the percentage of genes falling into each term. Enrichment scores are displayed as circle size and q-values as red gradient as indicated. (C) Heatmaps of the relative fold-change in Esm1(+) vs. Esm1(-) RECs (left) and in diabetes vs. control (right) in log-transformed expression-level of genes differentially expressed in Esm1(+) cells from control subjects falling into the Vascular development ontology cluster. (D) Heatmap of the relative fold-change in Esm1(+) vs. Esm1(-) RECs (left) and in diabetes vs. control (right) in log-transformed expression-level of genes differentially expressed in Esm1(+) cells from diabetic subjects falling into the chemotaxis ontology cluster. For heatmaps, relative expression in Esm1(+) vs. Esm1(-) cells is shown in controls (left panel) and diabetes (right panel) as indicated. All data are from human single-cell databases H1 (control) and H4 (diabetes). RECs, renal endothelial cells.

Esm1 single-cell co-transcriptome delineates glomerular tran in diabetes

We next assessed whether changes in the transcriptomic profile of Esm1(+) cells observed at the single-cell level could be confirmed in bulk transcriptomes. To that end, we analyzed a series of 20 healthy controls and 41 patients with diabetes for correlations between glomerular expression levels of Esm1 and genes associated with the transcriptomic shift in Esm1(+) cells from our single-cell analysis. This analysis reveals that in diabetes compared with healthy controls, transcript levels within these gene clusters are significantly more correlated with each other and with Esm1. We observe a significant and direct correlation of glomerular Esm1 with 14/21 vascular development genes. Accordingly, expression levels of these genes decrease along with that of Esm1 in patients with diabetes. By contrast, these genes have a more stable glomerular expression level in healthy controls, regardless of Esm1 expression, resulting in few correlations with Esm1 (Figures 6A and S6A). In patients with diabetes, we observe a significant and direct correlation with Esm1 for 7/8 genes that have a high single-cell co-expression with Esm1, which signifies that these chemotaxis genes have lower expression in diabetes, similar to Esm1. We also find a significant inverse correlation with Esm1 for 6/11 genes with a low single-cell co-expression with Esm1, i.e., these genes have higher expression in diabetes, opposed to the decreased expression of Esm1. In contrast, in healthy controls, we find minimal correlation between expression of chemotaxis genes and Esm1 (Figures 6B and S6B).

Figure 6. Human single cell diabetes-induced transcriptional shift in Esm1(+) cells observed in bulk transcriptomes.

(A) Matrix of correlations of Esm1 with genes differentially expressed in Esm1(+) cells from control subjects falling into the Vascular development ontology cluster. We ordered genes from higher to lower expression log fold-change in Esm1(+) vs. Esm1(-) cells in control subjects. (B) Matrix of correlations of Esm1 with genes differentially expressed in Esm1(+) cells from diabetic subjects falling into the chemotaxis ontology cluster. We ordered genes from higher to lower expression log fold-change in Esm1(+) vs. Esm1(-) cells in diabetic subjects. (C) Matrix of correlations of Esm1 with genes of the Igfbp5 interactome. We ordered genes following hierarchical clustering of correlations. For all correlation matrices, we displayed significant correlations as direct (red circles) or indirect (blue circles), with greater size representing more significant correlations. All data are from human gene expression profile databases H5 (control) and H6 (diabetes).

Potential cross-talk of Esm1(+) cells across glomeruli

To assess for cross-talk between Esm1(+) cells and other glomerular cell types, we examined gene expression among secreted genes. Of the chemotaxis genes that have higher expression in Esm1(+) cells from patients with diabetes, we further characterized IGFBP5. IGFBP-5 is an endothelial cell-expressed secreted protein, and we next analyzed for correlations between glomerular Esm1 and the IGFBP5 interactome in neighboring cells. We identified 21 genes coding for proteins with evidence of physical interaction with IGFBP-5. Among these, seven are found in a cluster of genes that all show a significant inverse correlation with Esm1 in diabetes, i.e., the expression of these genes is higher in diabetic patients with lower Esm1 glomerular expression. This cluster notably encompasses six genes from the extracellular matrix organization ontology, and 5/6 are abundantly expressed in neighboring mesangial cells (Figures 6C and S6C). Consistent with global changes in chemotaxis in diabetes, Esm1, through IGFBP5, links a subpopulation of endothelial cells with up-regulation of extracellular matrix genes enriched in mesangial cells.

Decreased Esm1 expression correlates with the level of transcriptional polarization in diabetes in individual glomerular cell types

To further study the relationship of Esm1 with different glomerular cell types, we measured correlations of Esm1 with genes enriched in specific cell types. For expression patterns of endothelial-enriched genes³⁰, we find stronger correlation patterns in diabetes, with 253/22578 correlations identified with an absolute r-value ≥ 0.8, vs. 51/22578 correlations with an absolute r-value ≥ 0.8 in control subjects (p<0.05). We identify two clusters of genes in diabetes with direct and inverse correlations with Esm1, respectively, and these gene clusters are absent in healthy controls (Figure 7A). Among these clusters, we find a significant, direct correlation for 55 genes and a significant, inverse correlation for 36 genes from correlation clusters, respectively (Table S11). Blood vessel morphogenesis has the most number of genes directly correlated with Esm1, consistently with our previous findings of vascular development, encompassing 30/55 hits from this gene list (enrichment score = 11; q-value = 10^-13). Nop56p-associated pre-rRNA complex is the only ontology cluster exclusively enriched with genes inversely correlated with Esm1, with 13/36 genes from this list (enrichment score = 45; q-value = 1.58*10^-10). In addition, three ontology clusters are enriched for genes for both types of correlations, including RAGE signaling and extracellular matrix organization related pathways, suggesting that reduced Esm1 expression in diabetes is associated with a more balanced effect on those pathways (Figure 7B). We then compared the expression levels of genes falling into the Nop56p-associated pre-rRNA complex and blood vessel morphogenesis between healthy controls and patients with diabetes and find that direct and inverse correlations with Esm1 expression are observed in diabetes. Therefore, genes that inversely correlate with Esm1 in diabetes exhibit either higher (e.g., COL4A2: log fold-change = 0.51, q-value = 1.08*10^-3), similar (e.g., IGFBP7: log fold-change = 0.06, q-value = 0.4) or lower (e.g., RPS16: log fold-change = −0.29, q-value = 3.36*10^-6) expression compared to the control group. Similarly, genes that directly correlate with Esm1 in diabetes show either higher (e.g., DOCK4: log fold-change = 0.35, q-value = 1.27*10^-2), similar (e.g., PIP4K2A: log fold-change = −0.15, q-value = 0.32) or lower (e.g., TGFBR2: log fold-change = −0.96, q-value = 4.77*10^-5) expression compared to the control group (Figure 7C).

Figure 7. Esm1 depletion is a sensor of the transcriptomic polarization of endothelial-enriched genes in diabetes.

(A) Matrix of correlations of endothelial-enriched genes in healthy controls and patients with diabetes. We ordered genes by hierarchical clustering of correlations based on the color key as indicated. (B) Heatmaps of enrichment of ontology clusters, showing the hypergeometric q-values, enrichment scores and percentages of genes falling into each term for the top 20 most significantly enriched ontology cluster in each gene list among endothelial-enriched genes in patients with diabetes. We performed a hierarchical clustering of significant ontology terms into a tree based on Kappa-statistical similarities among their gene memberships. We selected the term with the best q-value within each cluster as its representative term and display it in the heatmaps. Grey cells represent the non-significantly enriched ontology clusters in each gene list. The percentage of genes (green), enrichment score (blue), and q-value (orange) are shown for each ontology. (C) Expression levels in control and diabetes of genes correlated to Esm1 in diabetes, displayed vs. Esm1 (scatter plots, left panels) and alone (boxplots, right panels). For each group, linear regression curve with 95%CI and median [IQR] are displayed on scatter plots and boxplots, respectively. Inverse (Inv) and direct (Dir) correlations with Esm1 expression in diabetes are shown. All data are from human gene expression profile databases H5 (control) and H6 (diabetes). * p < 0.05 as indicated.

We next studied expression patterns of genes enriched in neighboring cell types, i.e., mesangial cells³⁰, podocytes³⁰, and immune cells³¹. In mesangial-enriched genes, we find stronger correlation patterns in diabetes, with 105/9591 correlations identified with an absolute r-value ≥ 0.8, vs. 27/9591 correlations with an absolute r-value ≥ 0.8 in control subjects (p<0.05). Further, we find two clusters of genes in diabetes with direct and inverse correlations with Esm1 (Figure 8A). Among these, there is a significant correlation for 31 and 16 genes of the direct and inverse correlation clusters, respectively (Table S11).

Figure 8. Esm1 depletion is a sensor of the transcriptomic polarization of neighboring glomerular cell type specific-enriched genes in diabetes.

Matrix of correlations of glomerular cell type specific-enriched genes in healthy controls and patients with diabetes (A, C, E, G) and heatmaps of enrichment of ontology clusters for select ontology clusters significantly enriched in at least one gene list of correlation among cell type-specific enriched genes in patients with diabetes (B, D, F, H). Mesangial cells (A, B), podocytes (C, D), dendritic cells (E, F), and macrophages (G, H) are shown. For all matrices of correlations, genes are ordered following hierarchical clustering of correlations. Red is for direct correlations, blue is for inverse correlations, and white indicates no correlation. For all heatmaps we show the hypergeometric q-values, enrichment scores and percentages of genes falling into each term. We performed a hierarchical clustering of significant ontology terms into a tree based on Kappa-statistical similarities among their gene memberships. We selected the term with the best q-value within each cluster as its representative term and display it in the heatmaps. Grey cells represent the non-significantly enriched ontology clusters in each gene list. The percentage of genes (green), enrichment score (blue), and q-value (orange) are shown for each ontology. Inverse (Inv) and direct (Dir) correlations with Esm1 expression in diabetes are shown. All data are from human gene expression profile databases H5 (control) and H6 (diabetes).

Regulation of cell adhesion is the ontology cluster with the most number of genes directly correlated with Esm1 (15/31 genes; enrichment score = 8.7; q-value = 2.51*10^-6), while cellular response to growth factor stimulus encompasses the most number of genes inversely correlated with Esm1 (10/16 genes; enrichment score = 11; q-value = 6.31*10^-5). Furthermore, we find three ontology clusters enriched for genes from both lists (focal adhesion, transmembrane receptor protein tyrosine kinase signaling and urogenital system development) (Figure 8B). In podocytes-enriched genes, we find stronger correlation patterns in diabetes, with 5166/21736 correlations identified with an absolute r-value ≥ 0.8, vs. 269/21736 correlations with an absolute r-value ≥ 0.8 in control subjects (p<0.05). In diabetes, we only identify one cluster of genes in diabetes with direct correlations with Esm1 (Figure 8C), among which 180 are statistically significant (Table S11). Blood vessel development is the ontology cluster with the most number of genes directly correlated with Esm1 (48/180 genes; enrichment score = 3.8; q-value = 7.94*10^-8). (Figure 8D). In dendritic cells and macrophages-enriched genes, we find stronger correlation patterns in diabetes, with 198/1225 and 218/1275 correlations identified with an absolute r-value ≥ 0.8, vs. 7/1225 and 6/1225 correlations with an absolute r-value ≥ 0.8 in control subjects (p<0.05), respectively. Dendritic cells and macrophages-enriched genes each have a cluster of indirect correlation with Esm1 in diabetes (Figures 8E and 8G), including 34 and 30 significant correlations, respectively (Table S11). Staphylococcus aureus infection has the most number of hits from dendritic cell-enriched genes (17/34 genes, enrichment score = 76; q-value 1.58*10^-10) (Figures 8F) while complement and coagulation cascades encompass the greatest number of hits from macrophage-enriched genes (15/30 genes; enrichment score = 61, q-value = 10.31*10^-10) (Figures 8H, intersections between lists of genes enriched in human glomerular compartments are shown in Figure S7). Taken together, across glomerular cell type-specific genes, we observe a consistent polarization in the transcriptomic profile in diabetes, and that these clusters correlate closely with changes in Esm1 expression.

Over-expression of Esm-1 modulates the glomerular transcriptional shift associated with albuminuria in DKD

To understand the significance of glomerular transcriptional modifications in DKD, we studied the expression profile of genes enriched in glomerular compartments in a murine model of DKD with reduction in albuminuria (Database M4) and with over-expression of Esm1 (Database M5)⁴¹. In 234 of 300 endothelial-enriched genes^{28, 37}, we observe correlations between Esm1 expression and albuminuria (κ = 0.53; p = 2.29*10^-29) (Figure 9A and Table S12), i.e., genes that are directly correlated with Esm1 show an inverse correlation with albuminuria, and those showing an inverse correlation with Esm1 directly correlate with albuminuria. Furthermore, for 206 of these 234 correlation genes, systemic over-expression of Esm-1 reverses the expression profile (κ = 0.25; p = 3.4*10^-6) (Figure 9A and Table S12). Many of the genes that directly correlate with albuminuria and are down-regulated by Esm-1, are from a vasculature development ontology (Figure 9A, Tables S7 and S12). Conversely, we find another cluster of genes that inversely correlate with albuminuria and are up-regulated by Esm-1 are related to blood vessel morphogenesis (Figure 9A, Tables S7 and S12). We find similar patterns of genes with an inverse correlation between glomerular Esm1 and albuminuria that are enriched in different glomerular cell types: mesangial cells (35/49 genes; κ = 0.5; p = 4.1*10^-7), podocytes (210/271 genes; κ = 0.49; p = 3.33*10^-23) and macrophages (55/81 genes; κ = 0.4; p = 7.7*10^-6) (Figures 9B-D and Table S12). Moreover, a large subset of these genes are directly regulated by Esm-1: mesangial (33/49 genes; κ = 0.37; p = 7.7*10^-4) and podocytes-enriched genes (188/271 genes; κ = 0.21; p = 1.3*10^-4) but not in macrophage-enriched genes (38/81 genes; κ = 0.04; p = 0.64) (Figures 9B-D and Table S12). Subsequently, we identify clusters of genes with inverse correlations with Esm1, direct correlations with albuminuria and that are downregulated with Esm-1 over-expression, mainly enriched for the regulation of vascular smooth muscle cell proliferation in mesangial-enriched genes, for actin filament-based process and ECM organization in podocytes enriched-genes and for the Staphylococcus aureus infection pathway in macrophages-enriched genes (Figure 9B-D, Tables S7 and S12). On the other hand, we find clusters of genes with direct correlations with Esm1, inverse correlations with albuminuria and that are upregulated with Esm-1 over-expression, mainly enriched for the regulation of vascular smooth muscle contraction in mesangial-enriched genes, for the regulation of cellular response to growth factor stimulus in podocytes enriched-genes and for myeloid leukocyte differentiation in macrophages-enriched genes (Figures 9B-D, Tables S7 and S12, intersections between lists of genes enriched in mouse glomerular compartments are shown in Figure S9). Taken together, over-expression experiments with Esm-1 demonstrate that for a large subset of glomerular cell type-specific genes that relate to albuminuria, Esm-1 determines expression of these genes.

Figure 9. Systemic overexpression of mouse Esm-1 regulates the expression of genes correlating with albuminuria in mice with DKD.

Matrix of correlations of glomerular cell type-specific-enriched genes, heatmaps of correlations with Esm1, urine albumin-to-creatinine ratio (UACR), glomerular macrophages (GM) and heatmap of expression log fold-change in mice with mouse Esm-1 systemic overexpression vs. no overexpression. Data for endothelial (A), mesangial (B), podocytes, and (C), and macrophages (D) are shown as indicated. For correlations matrices and heatmaps, red is for direct correlations, blue is for inverse correlations, and white indicates no correlation. For expression log fold-change heatmaps, blue is lower expression in mouse Esm-1 overexpression, red is higher expression in mouse Esm-1 overexpression, and white indicates no significant differences in expression with over-expression of mouse Esm-1. N = 4 in mice with DKD and no overexpression of mouse Esm-1. N = 3 in mice with DKD and overexpression of mouse Esm-1. All data are from DBA/2 mice with streptozotocin-induced DKD (Databases M4 and M5, see Zheng, et al., for details,⁴¹).

Discussion

We show that Esm1(+) cells segregate as a distinct subset in each vasculature compartment, thus defining a specialized population of cells. Combined with the variety of the vascular compartments enriched for Esm1, intra- and inter-species, our data suggest that the transcriptomic signature of Esm1(+) cells may not be solely explained by their spatial localization. Our results highlight a common set of functions in mature kidney shared by this subpopulation, including but not limited to angiogenesis, thus designating Esm1(+) cells as playing a role in the tissue proliferation process. These results are consistent with previous observations showing the critical role of Esm1 in vasculogenesis, with high expression at the vascular tip^21–24. A common set of functions based on distinct co-expressed genes has been described before⁴², but has not previously been associated with Esm1, endothelial cell subtypes, or kidney-specific genes. Esm1(+) renal endothelial cells from patients with diabetes exhibit lower expression of a majority of genes involved in vasculature development. Taken together, these results may reflect an impaired ability to regenerate injured renal endothelium in diabetes due to various potential causes including: (1) a decrease in the pool of cells specialized in the tissue and endothelial repair process; and/or (2) a loss of function of this particular cellular subset. However, further studies are needed to assess if these impaired pathways are a cause or a consequence of fewer Esm1(+) cells.

On the other hand, we observe a distinct shift in the role of Esm1(+) cells from patients with diabetes. This shift, characterized by upregulation of chemotaxis-related genes, despite occurring in a minority population of endothelial cells, is reflected in the glomerular transcriptome by a lower expression of chemotaxis-related genes that are highly correlated with Esm1 expression in subjects with lower expression levels of Esm1. Esm-1 transcription is up-regulated by several pro-inflammatory mediators¹⁷. Thus, one may propose that Esm1 expression persists in a subset of cells involved in the inflammatory process. Although their cell numbers are lower with diabetes, higher expression of chemotaxis-related genes in Esm1(+) cells compared to other glomerular endothelial cells, may indicate a role for these cells to provide a homing signal for endothelial progenitor cells to attenuate glomerulosclerosis and renal fibrosis⁴³. In addition, this shift in the pattern of Esm1(+) cell transcription can be linked to the inhibitory role of Esm1 in the leukocyte recruitment process^{18, 19, 44, 45} in specific disease states. Therefore, the higher expression of Esm1 in cells involved in the chemotactic process may reflect a potential mechanism for regulating vascular inflammation, thus supporting the hypothesis of its renoprotective effect in the diabetic kidney.

More generally, our results highlight the transcriptional polarization which occurs in the glomerular vasculature in diabetes. We observe organized alignments in the variations of gene expression that are not found in healthy controls. Interestingly, despite a loss of Esm1(+) cells in diabetes and lower expression of Esm1, Esm1 appears to be a marker of this transcriptional polarization, as most variations in gene expression indirectly correlate with the intensity of glomerular Esm1. This transcriptional polarization is observed not only in glomerular endothelial cells, but also in neighboring glomerular cell types, thus raising the possibility of cross-talk between Esm1(+) cells and other glomerular cell populations in DKD. This finding is consistent with other studies showing evidence of intra-glomerular cell cross-talk^{29, 46–49}. Given the ontologies aligning with Esm1 expression in our study, we speculate that this communication network could provide feedback that may enable cells to tune their signaling activity to regulate specialized functions, including vascular growth and chemotaxis. Future studies on glomerular cross-talk will explore autocrine and paracrine effects of secreted factors by Esm1(+) cells, including IGFBP-5.

Interestingly, our study highlights a correlation between Esm1 and IGFBP5 downregulation in diabetes. Given that IGFBP-5 is secreted, we investigated possible associations with the transcriptomic profile of the IGFBP5 interactome in neighboring glomerular cells. Based on this analysis, we find in patients with DKD with lower Esm1 expression a consistent upregulation of THBS1, SPP1, ADAM12, FHL2, LTBP1 and SERPINE1^50–53, which are all involved in the regulation of IGFBP5 in the extracellular matrix. Except for SPP1, all these genes have an enriched mesangial expression²⁹, thus highlighting that variations in the Esm1(+) cell transcriptome might affect gene expression in neighboring glomerular cells. Interestingly, when analyzing the expression of endothelial genes correlating with Esm1 expression, extracellular matrix pathways are found among the top enriched ontologies.

Another novel finding from our study is the relationship between the glomerular transcriptomic polarization and albuminuria. To the best of our knowledge, our study is the first to identify a systematic orientation of the glomerular-enriched transcriptome in which expression of genes align in parallel with the severity of albuminuria and with correlation to Esm1 expression in DKD. We therefore determined next whether Esm1 is a marker or mediator of this polarization. We identify subsets of these genes and pathways correlated to albuminuria in which gene expression is reversed by systemic over-expression of Esm-1. Importantly, we find that Esm-1 attenuates transcription of podocyte-enriched genes that are related to the extracellular matrix and are associated with high levels of albuminuria. This latter result is consistent with published data showing the protective effect of Esm-1 against podocyte loss⁴¹ and provides novel insights regarding potential molecular targets to prevent the development of DKD. How Esm-1 acts as a global mediator of glomerular gene expression (e.g., through epigenetic modifications, chemotaxis, etc.) will be an area for future study. A limitation of our study is that tools are still needed to tune the number of Esm1(+) cells or glomerular (vs. systemic) Esm1 expression to assess the contribution of local Esm-1 to modulate global gene expression, podocyte number, and albuminuria.

The role of inflammation has been widely described in the development of DKD^{15, 54–58}, and several studies outline a change in circulating Esm-1 concentrations in diabetes^{59, 60}. Furthermore, Esm-1 is known as an anti-inflammatory proteoglycan^{18, 19, 44}, and DKD susceptibility in mice has been linked to a deficiency in Esm1 expression in glomerular endothelial cells^{18, 41}. Esm-1 may significantly alter innate immunity genes as over-expression reduces albuminuria and alters expression of genes related to the interferon pathway⁴¹. In this study, over-expression of Esm-1 significantly altered expression of 18 interferon-related genes. Interestingly, 16 of these 18 genes were available in the bulk-RNA human database used in our study³³, and 8 of these 16 interferon-related genes significantly correlate with Esm1 expression, consistent with the findings reported by Zheng et al. Glomerular Esm1 inversely correlated with glomerular macrophage infiltration; however, over-expression of Esm-1 did not reduce glomerular or tubulointerstitial macrophage infiltration⁴¹. Consistent with these findings, we also did not observe that Esm-1 over-expression appreciably reversed expression of genes that clustered with macrophage infiltration. Whether Esm-1 regulates a more specific subset of macrophages or whether only local Esm-1 modulates leukocyte infiltration is another area for further study.

In summary, our study shows that Esm1 demonstrates glomerular expression in mice and humans. Esm1 expression delineates a specialized subpopulation of mature endothelial cells, with a transcriptomic signature enriched for vascular and tissue development processes that changes with diabetes. Diabetes is associated with a fall in glomerular Esm1 expression, with a transcriptional polarization characterized by variations of subsets of genes and pathways enriched in glomerular endothelial cells and in neighboring glomerular cells, tightly correlating with Esm1 glomerular expression. Expression of genes enriched in glomerular compartments correlate with Esm1 and albuminuria and can be rescued by systemic over-expression of Esm-1. From publicly available single cell and bulk transcriptomes, and through experimental validation by over-expression of Esm-1 in DKD-susceptible mice, our results generate novel mechanistic hypotheses about the role of Esm1, the subset of Esm1(+) cells, its role in glomerular crosstalk, and its role as sensor of transcriptional polarization in DKD.

Supplementary Materials

Supplementary Figure Legends

Figure S1. Expression of Esm1 compared to canonical genes of vasculature compartments in mouse kidney.

(A-C) t-SNE plots of RECs, color-coded for the expression of Esm1 (purple) and canonical genes of glomerular (g) (A), cortical (c) (B), and medullary (m) RECs (C). Magnitude of gene expression is indicated by the scale provided. (D) Violin plots showing the expression profiles of canonical genes of mouse kidney vasculature compartments in Esm1(+) and Esm1(-) RECs. (E-G) Heatmap of the relative fold-change in log-transformed expression-level of top 50 marker genes of subclusters in Esm1(+) vs. Esm1(-) glomerular (g) (E), cortical (c) (F), and medullary (m) RECs (G). All data are from single-cell mice databases M1, M2 and M3. RECs, renal endothelial cells.

Figure S2. Expression of Esm1 compared to canonical genes of vasculature compartments in human kidney.

(A-E) t-SNE plots of RECs, color-coded for the expression of Esm1 (purple) and canonical genes of afferent and efferent arterioles / peritubular capillaries (A), glomerular (B), and ascending vasa recta (C), descending vasa recta (D), and other endothelial cells (E). Magnitude of gene expression is indicated by the scale provided. (F) Violin plots showing the expression profiles of canonical genes of human kidney vasculature compartments in Esm1(+) and Esm1(-) RECs. All data are from single-cell human databases H1, H2 and H3. RECs, renal endothelial cells.

Figure S3. Relationship between glomerular expression of Esm1 and susceptibility to diabetic kidney disease in mice.

(A) Quantification by RNAscope of Esm1 glomerular expression in DBA/2 mice, compared to C57BL/6 mice. (B) Quantification by RNAscope of Esm1 glomerular expression in DBA/2 mice with STZ-induced diabetic kidney disease, compared to controls. * p < 0.05, as indicated.

Figure S4. Expression of Esm1 in patients with diabetes.

(A) t-SNE plots of RECs, color-coded for the expression of Esm1 and canonical genes of human kidney vasculature compartments from patients with diabetes. Grey is no expression, light color is low expression, dark color is high expression. (B) Violin plots showing the expression profiles of canonical genes of human kidney vasculature compartments in Esm1(+) and Esm1(-) RECs from patients with diabetes. All data are from single-cell human databases H1 (control) and H4 (diabetes). RECs, renal endothelial cells.

Figure S5. Glomerular RNA expression of genes co-expressed with Esm1 at the single-cell level and from the Igfbp5 interactome.

Z-scores of differential expressions of the top 15 differentially expressed genes on diverging branches according to trajectory inference analysis from patients with diabetes (A) and healthy controls (B). All data are from gene expression profile human databases H1 (control, purple) and H4 (diabetes, green).

Figure S6. Glomerular RNA expression of genes co-expressed with Esm1 at the single-cell level and from the Igfbp5 interactome.

(A) Expression level of genes differentially expressed in Esm1(+) cells from healthy controls falling into the vascular development ontology cluster. We ordered genes from highest to lowest single cell expression fold-change in Esm1(+) vs. Esm1(-) cells. (B) Expression level of genes differentially expressed in Esm1(+) cells from healthy controls falling into the Chemotaxis ontology cluster. We ordered genes from highest to lowest single cell expression fold-change in Esm1(+) vs. Esm1(-) cells. (C) Expression level of genes from the Igfbp5 interactome falling into the extracellular matrix organization ontology. For each group, we displayed gene expression on the y-axis vs. Esm1 expression on the x-axis and linear regression curves with 95% CI are shown. (E) Gene expression enriched in endothelial cells. (M) Gene expression enriched in mesangial cells. All data are from gene expression profile human databases H5 (control, purple) and H6 (diabetes, green).

Figure S7. Overlaps between gene lists used in the human gene expression profile analysis.

(A) Intersections between cell type-enriched whole gene lists used in the gene expression profile analysis. (B) Intersections between lists of cell type-enriched genes falling into the cluster of direct correlations and showing a significant correlation with Esm1. (C) Intersections between lists of cell type-enriched genes falling into the cluster of inverse correlations and showing a significant correlation with Esm1. All data are from gene expression profile the human database H6 (diabetes).

Figure S8. Overlaps between gene lists used for the gene expression profile analysis in DBA/2 mice with streptozotocin-induced DKD.

(A) Intersections between compartment-enriched whole gene lists used in the gene expression profile analysis. (B) Intersections between lists of compartment-enriched genes falling into the cluster of direct correlations with Esm1 and inverse correlations with urine albumin-to-creatinine ratio (UACR). (C) Intersections between lists of compartment-enriched genes falling into the cluster of inverse correlations with Esm1 and direct correlations with UACR. All data are from DBA/2 mice with streptozotocin-induced DKD (Database M4).

Author Contributions

A.G., N.K., and V.B. designed the study; A.G., X.Z. and V.B. analyzed the data; A.G. and V.B. made the figures; A.G., X.Z. and V.B. drafted and revised the paper; all authors have seen and approved the final version of the manuscript.

Financial Disclosures

The authors have no relevant conflicts of interest or financial disclosures.

Abbreviations

AEA

afferent and efferent arterioles

AVR

ascending vasa recta

cRECs

cortical renal endothelial cells

DKD

diabetic kidney disease

DVR

descending vasa recta

Esm-1

endothelial cell-specific molecule-1

gRECs

glomerular renal endothelial cells

mRECs

medullary renal endothelial cells

PTC

peritubular capillaries

Acknowledgments

A.G. was funded by a Fulbright Hauts-de-France grant and by the France – Stanford Center for Interdisciplinary studies. X.Z. was funded in part by a Larry L. Hillblom Foundation Postdoctoral Fellowship award and the Holmgren Family Foundation. V.B. was funded, in part, by the NIDDK (R01 DK091565).

The authors thank Drs. Rajasree Menon and Matthias Kretzler for sharing of data information.