A Single-Cell Transcriptome Atlas of the Mouse Glomerulus

Abstract

Background Three different cell types constitute the glomerular filter: mesangial cells, endothelial cells, and podocytes. However, to what extent cellular heterogeneity exists within healthy glomerular cell populations remains unknown.

Methods We used nanodroplet-based highly parallel transcriptional profiling to characterize the cellular content of purified wild-type mouse glomeruli.

Results Unsupervised clustering of nearly 13,000 single-cell transcriptomes identified the three known glomerular cell types. We provide a comprehensive online atlas of gene expression in glomerular cells that can be queried and visualized using an interactive and freely available database. Novel marker genes for all glomerular cell types were identified and supported by immunohistochemistry images obtained from the Human Protein Atlas. Subclustering of endothelial cells revealed a subset of endothelium that expressed marker genes related to endothelial proliferation. By comparison, the podocyte population appeared more homogeneous but contained three smaller, previously unknown subpopulations.

Conclusions Our study comprehensively characterized gene expression in individual glomerular cells and sets the stage for the dissection of glomerular function at the single-cell level in health and disease.

Glomeruli are the key functional units of the kidney filtration apparatus. Within each glomerulus, a capillary tuft is structurally maintained by mesangial cells and provides a three-layered filtration barrier consisting of endothelial cells, the glomerular basement membrane, and podocytes.^1,2 Although these three cell types within the glomerular tuft have long been established, it is as yet unknown whether individual cells within the glomerulus respond to cues to which they are physiologically exposed. Such cues include changing pressure gradients along the capillaries and mechanical strain on mesangial cells, which may differ depending on cell location relative to the glomerular vascular pole.³ Because BP adaptation and mechanoadaptation of glomerular cells are key determinants of kidney function and dysregulated in kidney disease, we tested whether glomerular cell type subsets can be identified by single-cell RNA sequencing in wild-type glomeruli. This technique allows for high-throughput transcriptome profiling of individual cells and is particularly suitable for identifying novel cell types as well as subsets and novel marker genes of known cell populations.^4–6

Methods

Glomerular isolation and preparation of single-cell suspensions were carried out as described⁷ on 8-week-old wild-type CD1 male mice. Flow-sorted cells were dehydrated in methanol,⁸ stored and shipped at −70°C, and rehydrated for highly parallel single-cell transcriptome profiling by Drop-seq.^4,8 This method predominantly detects 3′ ends of polyadenylated mRNA as well as long noncoding RNA molecules. Single-cell data were processed, and genes were quantified with Drop-seq tools v. 1.12⁴ and further analyzed with “dropbead”⁸ and Seurat.⁵ Marker gene identification was carried out with Seurat function “FindAllMarkers”⁵ and visual inspection of violin plots as well as images from the Human Protein Atlas.⁹ Immunofluorescence staining was carried out on glomeruli of Nphs2-Cre/mTmG reporter mice¹⁰ using affinity-purified rabbit antibodies. Images were obtained using confocal microscopy. Animal experiments were approved by the Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen (LANUV NRW, AZ 2013.A 375). Statistical methods were used as indicated.

Raw and processed datasets are available from the Gene Expression Omnibus repository (GSE111107). The interactive online database is available at https://shiny.mdc-berlin.de/mgsca/.

Results

Figure 1A shows the study design. We isolated glomeruli by magnetic bead perfusion followed by magnetic separation and rigorous washing (Supplemental Figure 1A),¹¹ generated single-cell suspensions by enzymatic digestion, and performed highly parallel single-cell RNA sequencing using the Drop-seq method.⁴ A total of 14,722 cells expressing >250 genes and 350 transcripts (defined as unique molecular identifiers) were obtained from four independent biologic replicates (eight mice pooled per replicate). Median numbers of genes and transcripts detected were similar (Supplemental Figure 1B, Supplemental Table 1). Drop-seq works on the basis of Poisson-distributed limiting dilution, and thus, it generates cell doublets (estimated at approximately 10% at the cell concentration used).^4,8,12 To obtain high-quality single-cell data, we used a previously developed algorithm to score cell type–specific marker genes¹² (Supplemental Table 2) and removed 1768 probable doublets. The final dataset contained 12,954 cells, with a median of 630 genes and 950 unique molecular identifiers per cell at a sequencing depth of approximately 9400 aligned reads per cell (Supplemental Figure 1B, Supplemental Table 1).

Figure 1.

Single-cell RNA-sequencing identifies the relevant cell populations in purified glomeruli. (A) Study design and workflow. (B) The plot shows a two-dimensional representation (tSNE: t-Distributed Stochastic Neighbor Embedding) of global relationships among approximately 13,000 single cells expressing >350 transcripts (unique molecular identifiers). Putative cell doublets were removed by scoring cell type–exclusive markers (Supplemental Material). Five clusters became apparent that correspond to known cell types present in glomeruli (12% endothelium [n=1556], 2% mesangium [n=216], and 80% podocytes [n=10,325]) or contaminating cells from kidney tissue (6% tubules [n=828] and 0.2% immune cells [n=29]). Regarding parietal cells, none of the published marker genes were detected in any of the clusters. Cell types were identified by assessing the top most variable genes in each cluster (Supplemental Table 2). (C) Distribution and relative expression of established marker genes (violin plots) for endothelium, mesangium, podocytes, and contaminating tubular and immune cells.(D) Expression of marker genes colored on the basis of normalized expression levels (gray, low; red, high).

As shown in Figure 1B, unsupervised clustering of the remaining 12,954 single cells identified five major cell clusters. On the basis of marker genes, three of these clusters corresponded to known glomerular cell populations: podocytes (80%), mesangial cells (2%), and endothelial cells (12%). The other two clusters corresponded to tubular cells (6%) and a small group of immune cells (0.2%). Glomerular cell type clusters showed specific expression of established marker genes (Figure 1, C and D, Supplemental Table 3), and all replicates contributed to the observed cell clusters (Supplemental Figure 1C). Hierarchical clustering of aggregated reads from all replicates and cell types indicated high correlations according to cell type and independent of the replicate (Supplemental Figure 1D). To control for effects of single-cell preparation, we compared single-cell RNAseq data with bulk polyA-RNAseq libraries prepared from glomeruli before and after dissociation into single cells (bulk1 and bulk2, respectively) (Figure 1A). Although the single-cell data showed good correlations with both bulk mRNAseq datasets (Supplemental Figure 1E), it became apparent that single-cell dissociation affected cell type abundance (Supplemental Figure 1, F and G), explaining an over-representation of podocytes relative to endothelial and mesangial cells (Figure 1B). (Tubules were not affected.) We also compared aggregated reads from our cell type–specific clusters with published mRNAseq datasets obtained from sorted cell populations on the basis of glomerular cell lineage tracing experiments^7,13 (Supplemental Figure 1H). Although samples correlated best by sequencing method, pairwise correlations by cell type supported our cell type assignments.

We continued by characterizing glomerular cell types in more detail and aimed to identify novel cell-specific markers by assessing highly variable genes between clusters.⁴ Established cell-specific marker genes for endothelium, mesangium, and podocytes as well as genes described as relevant to the respective cell type in the literature (Supplemental Methods has details) were comprehensively reproduced as specifically expressed, validating our unsupervised clustering (Figure 2A, Supplemental Figure 2).^7,14,15 Importantly, expression of several previously reported key podocyte genes did not seem to be exclusive to podocytes, a finding bearing important implications for future studies on the function of such genes in kidneys. Examples for such genes include Podxl (for which previously described endothelial expression was confirmed¹⁶), Actn4, and Itgb1 (Figure 2, Supplemental Figure 2). Consequently, we aimed to identify novel cell-specific markers for all three glomerular cell types (Figure 2B). A large proportion of these markers was corroborated on the protein level by immunostaining images obtained from the Human Protein Atlas (Figure 2B).⁹ Novel markers represent a wide variety of molecular functions, including the transcription factor Meis2 identified as specific to endothelial cells and disease genes, such as Pde3a (the gene mutated in autosomal dominant hypertension with brachydactyly, which was identified as specific to mesangial cells), as well as the E3-ubiquitin-ligase Wsb2, a novel podocyte marker. Taken together, we provide a detailed and comprehensive characterization of glomerular cell types at the transcriptome level, including established and novel markers.

Figure 2.

Single-cell transcriptomics reveal novel molecular markers specific to glomerular cell types. (A and B) Distribution and relative expression of individual highly variable genes (violin plots) in endothelium, mesangium, and podocytes. (A) Established markers (bold) and markers identified as relevant to the cell type in the literature (italics). (B) New marker genes identified in this study. (Left panel) Distribution and relative expression (violin plots). (Right panel) Immunohistochemistry images from the Human Protein Atlas (HPA) confirm that marker proteins are expressed in human glomeruli in a histologic pattern as predicted from single-cell transcriptional analysis in mouse glomeruli. Image areas shown (500×500 pixels =200 μm²) correspond to glomeruli taken from larger HPA images.

Thus far, glomerular gene expression has been examined almost exclusively in cell populations rather than single cells. Therefore, it has remained unclear whether cell type heterogeneity exists within the three glomerular cell types. To approach this longstanding question, we focused on subclustering the two larger clusters containing most cells—podocytes and endothelial cells. The latter showed five distinct subclusters (Figure 3A, Supplemental Table 3). Subcluster 4 was identified as residual cell doublets due to expression of high levels of podocyte-specific markers (Figure 3B), and it was excluded from further analyses. The remaining four subclusters showed equal representation from all replicates (Supplemental Figure 3, A and B). Expression of key genes distinguished these subclusters as illustrated in Figure 3, B and C and Supplemental Figure 3C. Interestingly, subcluster 3 was defined by marker genes implicated in key endothelial molecular responses to physiologic and pathologic cues, such as endothelium to pericyte crosstalk (Jag1),^17–19 regulation of angiogenesis (Fbln5),²⁰ endothelial activation (Cxcl1),²¹ and response to complement activation (Cldn5).²² Ehd3, a marker suggested to be specific to glomerular endothelium,^23,24 was unevenly expressed, with enrichment in subclusters 0 and 2 and lower expression in subclusters 1 and 3; this raised the possibility that a fraction of Ehd3-negative cells in the endothelial pool originates from other parts of the kidney. To obtain better functional understanding of the endothelial subclusters, we performed pathway and gene set overdispersion analysis.²⁵ Four gene sets were identified that characterized the subclusters to varying extent with terms relating to “cell adhesion,” “cell maturation,” “stress response,” and “cell proliferation” (Figure 3D, Supplemental Figure 3E, Supplemental Table 4), suggesting that the endothelial subclusters might represent different states of endothelial cell biology between homeostasis and activation. Interestingly, antibody staining in the Human Protein Atlas for some proteins encoded by the subcluster determining genes, such as Fbln2, Hspa1b, S100a4, and Thbd (Figure 3B), was consistent with a nonuniform expression pattern of these genes in human glomerular endothelial cells (Supplemental Figure 3D). Whether this state depends on individual cell localization within a healthy capillary tuft or simply reflects localization in other parts of the kidney requires further investigation.

Figure 3.

Subclustering reveals the presence of endothelial subpopulations. (A) Two-dimensional representation of a subclustering analysis of endothelial cells. Five subclusters (0–4) became apparent. (B) Distribution and relative expression of individual highly variable genes (violin plots) in the different clusters. Cluster 4 corresponds to residual cell doublets as indicated by the expression of podocyte-specific markers (Nphs2 and Cdkn1c). Doublets were excluded from further analysis. (C) Expression of markers colored on the basis of normalized expression levels. Upper panels correspond to the subcluster tSNE (t-Distributed Stochastic Neighbor Embedding) plot as shown in A, and lower panels correspond to the tSNE plot of the whole dataset as shown in Figure 1B. (D) Pathway and gene set overdispersion analysis.³³ The heat map indicates four endothelial subclusters (0, red; 1, green; 2, blue; 3, violet) that show distinct, over-represented gene activation patterns (Supplemental Figure 3E). Corresponding gene clusters are listed in Supplemental Table 4.

Subcluster analysis in podocytes yielded seven subpopulations defined by more subtle gene expression differences (Supplemental Figure 4A, upper panel). In this context, cluster 4 showed an extensive stress response gene expression signature (Supplemental Figure 4A, lower panel). Tissue dissociation–induced changes in gene expression can explain this observation.²⁶ Accordingly, we detected an increased expression of stress response genes in bulk mRNAseq libraries obtained from dissociated glomerular cells compared with whole glomeruli (Supplemental Figure 4, B and C). Reclustering of the podocytes after regressing out stress response genes identified six subclusters (Figure 4A, Supplemental Table 3) with equal representation of all biologic replicates (Supplemental Figure 5, A and B). Three small subclusters (3–5) were identified robustly and independent of the stress response signature (Figure 4A). Marker gene analysis identified only a handful of genes, including Cald1 and Lars2, as well as transcripts coding for mouse-specific microRNAs (Gm10801 and Gm10800) and noncoding RNAs (Gm15564 and Gm23935) (Figure 4, B and C, Supplemental Figure 5C), whereas the remaining three larger clusters did not yield specific markers. Given the small number of coding transcripts among subcluster markers, pathway and gene set overdispersion analysis did not yield significant results (Supplemental Figure 6).

Figure 4.

Subclustering reveals a limited heterogeneity of podocytes. (A) Two-dimensional representation of a subclustering analysis of podocytes (tSNE: t-Distributed Stochastic Neighbor Embedding) after correction for tissue dissociation–induced stress response gene expression (Supplemental Material). Six podocyte subclusters (0–5) became apparent. (B) Distribution and relative expression of individual highly variable genes (violin plots) in subcluster 4. (C) Expression of markers in subcluster 4 (corresponding to B). Expression colored is on the basis of normalized expression levels (gray, low; red, high). (D) Laser-scanning confocal microscopy of isolated glomeruli from kidneys of transgenic Nphs2-Cre×mT/mG double-fluorescent reporter mice.³⁵ Podocytes are genetically marked by Cre-dependent membrane-targeted green fluorescent protein [GFP] (green) fluorescence, whereas nonpodocyte cell types remain membrane-targeted Tomato (red) positive. (Row 1) Whole glomeruli; yellow squares in podocyte staining (green) indicate areas for magnifications as shown below. Cald1 antibody staining (upper square) and IgG control (lower square). (Rows 2 and 3) Insets from whole glomeruli as indicated. Yellow arrowheads point to GFP-positive podocytes that are Cald1 negative (row 2) or unstained by IgG control (row 3). (Rows 4 and 5) Yellow arrowheads point to a GFP-positive, Cald1-positive podocyte. Magnified areas are 22×22 μm². Scale bars: 10 μm.

To corroborate podocyte subcluster markers on the protein level, immunofluorescence staining was carried out for Cald1 and Lars2 on glomeruli obtained from reporter mice, in which podocytes are marked by green fluorescent protein.¹⁰ Colocalization of Cald1 and Lars2 with green fluorescent protein occurred only in a subset of podocytes (Figure 4D, Supplemental Figure 5D), suggesting that heterogeneity among podocytes in healthy glomeruli might exist. Cald1 is a calmodulin and actin binding protein that has been shown to be glucocorticoid responsive in podocytes,²⁷ and it is hypothesized to play a role in the development of diabetic nephropathy.^28,29 Lars2—a mitochondrial Leucyl transfer RNA synthetase—is mutated in Perrault syndrome.³⁰ Although Perrault syndrome is primarily a neurologic disorder, mutations of mitochondrial Leu-transfer RNA are the basis of both Mitochondrial encephalopathy, lactic acidosis, and stroke-like episodes syndrome (MELAS) syndrome and hereditary FSGS,^31,32 again pointing toward a role in podocytes.

Discussion

Our study also highlights a number of important caveats. First, we observed an effect of the single-cell dissociation procedure. A comparison of our single-cell data with bulk transcriptomes revealed an apparent over-representation of podocytes relative to endothelial and mesangial cells as well as a stress response signature in one of the podocyte subclusters. As shown above, the latter kind of artifact can be corrected computationally. Second, although the vast majority of cells sequenced were clearly glomerular, arguing for high purity of isolated glomeruli, an extraglomerular origin for a fraction of endothelial cells is possible. Third, we examined male mice of one strain at an age when glomeruli are still enlarging. Thus, the glomerular subpopulations observed may not necessarily be stable in mice of different ages, sexes, or strains.

In summary, our study comprehensively characterizes gene expression in individual glomerular cells. We identified novel marker genes for all glomerular cell types and found evidence for transcriptional heterogeneity among endothelial cells and podocytes. Earlier publications using single-cell RNA transcriptomics on glomerular cells were limited by focusing exclusively on a single cell type and sequencing small numbers of cells (20 podocytes and 14 mesangial cells, respectively).^33,34 In contrast, our approach has exploited the potential of highly parallel single-cell profiling for profiling a large number of cells in an unbiased way.^33,34 As a resource, we provide an extensive single-cell sequencing dataset of the mouse glomerular transcriptome that can be freely accessed and interrogated online (https://shiny.mdc-berlin.de/mgsca/). Our study thus paves the way for future investigations addressing the response of individual glomerular cells to disease states—as illustrated by a pilot study on kidney biopsy specimens from patients with lupus nephritis.³⁵ The glomerulus is a key system, the understanding of which will greatly benefit from improved single-cell RNA sequencing protocols.

Disclosures

None.