Research Article

Single-cell transcriptomics of the mouse kidney reveals potential cellular targets of kidney disease

See allHide authors and affiliations

Science  18 May 2018:
Vol. 360, Issue 6390, pp. 758-763
DOI: 10.1126/science.aar2131

Touring the kidney, cell by cell

Our kidneys play a critical role in keeping us healthy, a fact of which we are reminded several times each day. This organ's cellular complexity has hindered progress in understanding the mechanisms underlying chronic kidney disease, which affects 10% of the world's population. Using single-cell transcriptional profiling, Park et al. produced a comprehensive cell atlas of the healthy mouse kidney (see the Perspective by Humphreys). An unexpected cell type in the collecting duct appears to be a transitional state between two known cell types. The transition from one cell type to the other is regulated by the Notch signaling pathway and is associated with metabolic acidosis. The authors also find that genetically distinct kidney diseases with common clinical features share common cellular origins.

Science, this issue p. 758; see also p. 709


Our understanding of kidney disease pathogenesis is limited by an incomplete molecular characterization of the cell types responsible for the organ’s multiple homeostatic functions. To help fill this knowledge gap, we characterized 57,979 cells from healthy mouse kidneys by using unbiased single-cell RNA sequencing. On the basis of gene expression patterns, we infer that inherited kidney diseases that arise from distinct genetic mutations but share the same phenotypic manifestation originate from the same differentiated cell type. We also found that the collecting duct in kidneys of adult mice generates a spectrum of cell types through a newly identified transitional cell. Computational cell trajectory analysis and in vivo lineage tracing revealed that intercalated cells and principal cells undergo transitions mediated by the Notch signaling pathway. In mouse and human kidney disease, these transitions were shifted toward a principal cell fate and were associated with metabolic acidosis.

The kidney is a highly complex organ that performs many diverse functions that are essential for health. It removes nitrogen, water, and other waste products from the blood. It controls blood electrolytes and acid-base balance, and it secretes hormones that regulate blood composition and blood pressure. The kidney consists of several functionally and anatomically discrete segments. The glomerulus is a specialized group of capillaries that filters the blood and produces the primary filtrate of water and solutes such as sodium, potassium glucose, and bicarbonate. The proximal tubules then reabsorb the majority of the water and electrolytes, whereas solutes such as uric acid, organic anions, potassium, and protons are secreted into the filtrate. The loop of Henle is primarily involved in solute concentration. The distal tubule and the collecting duct are segments where highly regulated solute transport occurs. Thus, each segment is critical for maintaining electrolyte and water homeostasis.

In the past, kidney cells have been annotated on the basis of their function, their anatomical location, or the expression of a small number of marker genes (1), yet these classification systems do not fully overlap. An emerging technology called single-cell transcriptional profiling allows investigators to monitor global gene regulation in thousands of individual cells in a single experiment (2, 3). In principle, this technology could answer central questions in kidney biology and disease pathogenesis because it has the potential to provide four distinct types of information.

First, unbiased single-cell clustering can redefine kidney cell types on the basis of only their global transcriptome patterns (4). Such analyses have already been applied to other organs (2, 57) and even to whole multicellular organisms (8, 9). These experiments have identified previously unrecognized cells and have cataloged marker genes for previously defined cells, indicating that this approach has the potential to redefine kidney cell types.

Second, single-cell analysis may help dissect the mechanisms underlying common kidney diseases (10, 11). In general, kidney pathologies have been grouped together by their temporal patterns (acute or chronic) or by their target structures (glomerular versus tubular), which has obscured the underlying biology. Previously obtained bulk transcriptome profiles have generated readouts only for predominant cell populations such as the proximal tubular cells (12). Kidney segment–specific RNA-sequencing analysis of the rat kidney has provided useful resources (13), but single-cell analysis can potentially further exploit cell type–specific changes and identify previously unrecognized cell types during disease modulation, independent of preconceived cellular definitions.

Third, single-cell analysis may be able to identify fluctuating states of the same cell type. It is generally believed that terminally differentiated cells have limited plasticity. Most cell plasticity in adults has been observed in the context of differentiation of progenitor cells, best described in the blood and intestine (14). Such cellular transitions have also been documented during the development of the collecting duct (1517). For example, subtypes of intercalated cells (ICs) can change their functional polarity. In addition, stem cell–like populations originating from the principal cell types (PCs) may persist in the adult collecting duct and respond to external stimuli (18, 19), but definition of these plastic cells is lacking.

Fourth, current models of kidney disease cannot distinguish primary cell autonomous responses from secondary cell nonautonomous responses. Single cell–specific gene expression profiles, in contrast, may help identify the readout of disease-associated gene mutations in each cell.

Single-cell profiling and unbiased clustering of mouse kidney cells

We first cataloged mouse kidney cell types in an unbiased manner by using droplet-based single-cell RNA sequencing (20). We isolated and sequenced a total of 57,979 cells from whole kidney cell suspensions derived from seven healthy male mice (one kidney per mouse). Using stringent quality controls (20), we further analyzed 43,745 cells. Clustering analysis identified 16 distinct cell clusters consisting of as few as 24 cells to as many as 26,482 cells per cluster (the clusters were restricted to a minimum of 20 cells) (Fig. 1A).

Fig. 1 Cell diversity in mouse kidney cells delineated by single-cell transcriptomic analysis.

(A) Unsupervised clustering demonstrates 16 distinct cell types shown in a t-distributed stochastic neighbor embedding (tSNE) map (center). Left panels are subclusters of clusters 1, 3, and 7. Percentages of assigned cell types are summarized in the right panel. Endo, containing endothelial, vascular, and descending loop of Henle; Podo, podocyte; PT, proximal tubule; LOH, ascending loop of Henle; DCT, distal convoluted tubule; CD-PC, collecting duct principal cell; CD-IC, collecting duct intercalated cell; CD-Trans, collecting duct transitional cell; Fib, fibroblast; Macro, macrophage; Neutro, neutrophil; lymph, lymphocyte; NK, natural killer cell. (B and C) Violin plots showing the expression levels of representative marker genes across the 16 main clusters. The y axis shows the log-scale normalized read count. (C) Cluster 1 [from (A), left] separates into endothelial cells (Endo), pericytes and vascular smooth muscle cells (Peri), and descending loop of Henle (DLH) cells. Cluster 3 (proximal tubules) separates into S1, S2, and S3 segments or proximal convoluted tubules (PCT) and proximal straight tubules (PST). In cluster 7, intercalated cells (ICs) separate into types A and B.

We next performed several important quality-control analyses to validate our map. First, we ensured that cells from the seven kidneys were distributed evenly in all 16 clusters and that each cluster contained cells from more than four experiments (fig. S1). Next, we examined the effect of mitochondrial gene content (fig. S2A). The clustering of cells was not affected by mitochondrial gene content (fig. S2, B to E). Furthermore, genes whose expression positively correlated with mitochondrially encoded proteins were associated with solute transport (which requires abundant energy) rather than with cellular stress responses (fig. S3). This indicates that the increased mitochondrial gene count was inherent to specific (proximal and distal tubule) cell types in the kidney. In addition, by testing different clustering methods, we found that most methods identified similar cell groups (fig. S4), expressing the same group of marker genes with limited variations in cell separation. Last, we showed that decreasing the cell number from 40,000 to 10,000, 3000, or 1000 cells (fig. S5A) was associated with increasing uncertainty in cell cluster identification and the loss of rarer cell types (fig. S5, B and C).

Classification of kidney cells based on cell type–specific marker genes

To define the identity of each cell cluster, we generated cluster-specific marker genes by performing differential gene expression analysis (Fig. 1B, fig. S6, and table S1) (20). In many cases, the unbiased cluster identifier was a known cell type–specific marker, such as Kdr (encoding vascular endothelial growth factor receptor 2) for endothelial cells, Nphs1 (nephrin) and Nphs2 (podocin) for podocytes, Slc12a1 (Na-K-2Cl cotransporter) for the ascending loop of Henle, and Slc12a3 (thiazide-sensitive sodium chloride cotransporter) for the distal convoluted tubule (Fig. 1B). Immune cells and endothelial cell clusters separated from epithelial cells, but the ureteric bud– (clusters 6 to 8) and metanephric mesenchyme–derived (clusters 2 to 5) epithelial clusters were more closely aligned (Fig. 1A). Although some of the markers were already known, we identified a large number of additional markers, including Cdkn1c and Bcam for podocytes (Fig. 1B and fig. S6B). Further analysis identified eight subclusters within clusters 1, 3, and 7 (Fig. 1C and tables S2 and S3). Cluster 1 separated into endothelial cells; pericyte, vascular smooth muscle, and mesangial–like cells; and descending loop of Henle (DLH) cells. Cluster 3 (proximal tubules) separated into S1, S2, and S3 segments or proximal convoluted and straight segments (fig. S7). ICs (cluster 7) separated into types A and B.

To reliably assign a specific cell type to each cell cluster, we first correlated our gene expression results with bulk RNA-sequencing data from microdissected rat kidney segments (fig. S8) and microarray data on human immune cell types (fig. S9). To further validate our clustering analysis, we used Nphs2CremT/mG, SclCremT/mG, and Cdh16CremT/mG mice as reporter lines to mark podocytes, endothelium, and tubule cells with green fluorescent protein (GFP). The GFP expression in these models confirmed the proposed cell identity of our cell clusters (fig. S10). Altogether, our single-cell transcriptome atlas provides a molecular definition of 18 previously defined kidney and immune cell types, as well as three newly defined cell types.

Mendelian disease genes show cell type specificity

We next tested the hypothesis that hereditary kidney diseases that are characterized by the same phenotypic manifestations originate from the same cell type. We also explored whether the functions of specific cell types in the mouse kidney could be inferred from the expression pattern of human genes whose loss of function results in kidney disease. We found that the mouse homologs of 21 of 29 genes that have been associated with monogenic inheritance of proteinuria in humans were expressed in only one cell type—namely, the podocyte of the glomerulus (Fig. 2A and fig. S11). Although earlier studies have implicated defects in endothelial cells and proximal tubules in the development of proteinuria, and functional and structural changes in these cell types can be seen in patients with proteinuria, our results unequivocally show that podocyte dysfunction is the principal reason for proteinuria (21). As another example, we found that the mouse homologs of genes associated with renal tubule acidosis (RTA) in humans were expressed only by ICs of the collecting duct, confirming the major role of these cells in acid-base homeostasis (Fig. 2A). Furthermore, mouse homologs of genes that have been implicated in blood pressure regulation through analysis of human Mendelian diseases, such as Wnk4, Wnk1, Klh3, and Slc12a3, were expressed specifically in the distal convoluted tubule, whereas Nr3c2, Scnn1b, Scnn1g, and Hsd11b2 were specifically expressed by PCs of the collecting duct (Fig. 2A and figs. S11 and S13).

Fig. 2 Discrete human disease phenotypes are due to mutations in single specific cell types.

Single cell–type specific average expression of human (A) monogenic disease genes and (B) complex-trait genes identified by genome-wide association studies. Mean expression values of the genes were calculated in each cluster. The color scheme is based on z-score distribution; the map only shows genes with maximum z-scores > 2. In the heatmap, each row represents one gene, and each column is a single cell type (defined in Fig. 1). The full list of cell types and genes is shown in figs. S11 and S12.

Following the same logic, we annotated the expression of putative complex-trait disease genes that have been associated with blood pressure, chronic kidney disease (CKD) and serum metabolite levels, nephrolithiasis (e.g., Slc34a1), and RTA (e.g., Atp6v1b1) (Fig. 2B and figs. S12 and S13) (2224). We found that most genes implicated in these traits were expressed only in a single cell type, such as collecting duct cells (RTA) or proximal tubule cells (nephrolithiasis). The expression of genes associated with plasma metabolite levels—such as Slc17a3 (uric acid), Slc51a (bile acid), and Slc16a9 (carnitine) (2224)—and CKD showed strong enrichment for proximal tubule–specific expression, whereas blood pressure–associated genes were mostly expressed in collecting duct cells. Thus, our single-cell transcriptomic analysis highlights specific cells responsible for specific kidney-related disorders, as well as the critical functions of these cells.

Identification of a previously unrecognized cell type in the collecting duct

The collecting duct of the kidney differs from all other kidney epithelia because it originates from the ureteric bud and not from the metanephric mesenchyme. This compartment is composed of at least three distinct cell types: the PCs, which are responsible for sodium, water reabsorption, and potassium secretion, and the type A and B ICs, which are responsible for acid and alkali secretion, respectively. We identified the genes encoding aquaporin 2 (Aqp2) and H+-ATPase (H+-dependent adenosine triphosphatase) subunit (Atp6v1g3) as the key marker genes for clusters 6 and 7, defining these clusters as PCs and ICs (Figs. 1 and 3, A and B).

Fig. 3 Identification of a transitional cell type and a conversion process in the kidney collecting duct.

(A) The expression levels of marker genes across the 16 clusters. The y axis shows the log-scale normalized read count. (B) Gene expression levels in PCs (Aqp2), ICs (Atp6v1g3), and transitional cells (Syt7), demonstrated by a tSNE plot. (C) Representative immunofluorescence images of AQP2 (PC marker), ATP6V1B1 (IC marker), and DAPI (4′,6-diamidino-2-phenylindole) in the kidney collecting duct. The arrow indicates the transitional PC-IC cell type expressing AQP2 and ATP6V1B1. (D) Heatmap showing the expression levels of differentially expressed genes in collecting duct cell types. The color scheme is based on z-score distribution. (E) Venn diagram showing the overlaps of differentially expressed genes between PCs, ICs, and the newly identified cell type. (F) Immunofluorescence staining for PARM1 (transitional cell–specific) and AQP2 (upper panels) or ATP6V1B1 (lower panels) in the kidney collecting duct. “Double-positive” cells are shown by the arrows. (G) Ordering single cells along a cell conversion trajectory using Monocle. Three collecting duct cell clusters were used for ordering and plotted in low-dimensional space with different colors. The tSNE plots next to the trajectory map show differentially expressed genes in the corresponding cell lineages. (H) Aqp2CremT/mG mouse model used for lineage tracing of AQP2-positive cells (left) and immunofluorescence staining for GFP, ATP6V1B1, and AQP2 (right). The far-right panel shows the quantification of GFP-positive cells (mean ± SD; n = 3). AQP2-driven GFP (white) is found in PCs (red and white) and ICs (green and white). (I) Atp6CremT/mG mouse model used for lineage tracing of ATP6ase-positive cells (left) and immunofluorescence staining for GFP, ATP6V1B1, and AQP2 in Atp6CremT/mG mice (right). ATP6V1B1-driven GFP (white) is found in PCs (red and white), ICs (green and white), and transitional cells (red, green, and white).

Unexpectedly, our single-cell profiling identified a third cell cluster. This cell cluster (cluster 8) expressed markers of both ICs and PCs (“double-positive cells”; Fig. 3, A and B) and additional cell type–specific markers. We performed double immunofluorescence staining and in situ hybridization with probes for Aqp2 and Atp6v1b1 (Fig. 3C) and cell type–specific markers such as Parm1 and Sec23b (Fig. 3, D to F, and fig. S14) to validate the existence of this cell type.

To further investigate this unexpected cell type, we used the Monocle toolkit to perform cell trajectory analysis using pseudotime reconstitution of clusters 6 to 8 (20). We found that the newly identified cells were located between PCs and ICs, suggesting that cluster 8 is a transitional cell type (Fig. 3G). Transitional cells showed low expression levels of stress response genes and cell cycle genes, and these cells were present in all batches of our kidney isolates (figs. S15 and S16), excluding the possibility that they were injured cells, a proliferating subtype of collecting duct cells, or an artifact. Furthermore, cell trajectory analysis separated ICs into types A and B and PCs into their subtypes (PCs in the collecting duct and connecting tubule), as previously identified (Fig. 3G and fig. S17) (7, 25). These results indicate that the collecting duct contains not only PCs and ICs but a third distinct, transitional cell type; this raises the possibility that ICs and PCs represent two ends of a spectrum of cellular phenotypes and that they may undergo cellular transitions.

Fluorescent lineage tracing confirms plasticity of collecting duct cells

We next examined whether transitional cells could be identified by conventional in vivo lineage tracing and whether they match our computational characterization. We generated mice that carry a lineage tag in differentiated PCs (Aqp2CremT/mG) or in differentiated ICs (Atp6CremT/mG) (Fig. 3, H and I). We performed triple immunofluorescence labeling in these animals by staining for GFP (all cells of a specific marker origin), AQP2 (PCs), and ATP6V1B1 (ICs). As expected, we found that most of the GFP-positive cells were also AQP2-positive in the Aqp2CremT/mG mice. A subset of the GFP-positive cells expressed ATP6V1B1, an IC marker, but not AQP2. A smaller subset was double-positive for ATP6V1B1 and AQP2. Among the Aqp2CremT/mG GFP-positive cells, 61.6% were AQP2-positive, 29.2% were ATP6V1B1-positive, and 9.2% were double-positive for AQP2 and ATP6V1B1 (Fig. 3H). Similar analyses were performed with the Atp6CremT/mG lineage, which showed that double-positive (AQP2 and ATP6V1B1–positive) transitional cells and ATP6V1B1-negative true PCs can originate from ATP6V1B1-positive ICs (Fig. 3I).

To determine whether cell proliferation might be responsible for this cell plasticity, we calculated the expression levels of cell cycle–regulated genes in the single-cell transcriptome and in cell trajectory maps. We found that only clusters 9 and 16 (newly identified cell types 1 and 2), not any of the collecting duct clusters, expressed high levels of the cell cycle genes (fig. S18). This suggests that cluster 8 is likely to be a transitional cell population and not a proliferating progenitor cell. Thus, in vivo lineage tracing analysis confirmed transitions of PCs and ICs not only during development (1517) but also in the adult collecting duct through a newly identified transitional cell type.

Collecting duct cell plasticity, driven by Notch signaling, results in abnormal cell populations in CKD

For further analysis of the plasticity of collecting duct cells, we identified genes whose expression levels change during transitions of PCs and ICs (fig. S19, A and B) (20). PCs showed enriched expression of genes associated with cell adhesion, water homeostasis, and salt transport, whereas ICs showed enriched expression of genes associated with ATP hydrolysis and synthesis, coupled proton transport, and oxidation-reduction processes (fig. S19C). The gene expression patterns revealed that the Notch signaling pathway was activated during the transition of ICs to PCs. Notch regulates the cellular identity of neighboring cells by the expression of either Notch ligands or Notch receptors. Alternating expression of ligands and receptors creates a signal-sending cell (Notch-off) and a signal-receiving cell (Notch-on). Genes encoding Notch ligands, such as Jag1, were highly expressed by ICs, whereas their expression levels were low in PCs (Fig. 4A). In contrast, PCs showed high expression levels of Notch2 receptor and its transcriptional target Hes1 (which encodes a transcription factor), suggesting that PCs are the Notch signal–receiving cells in the collecting duct. Immunofluorescence studies confirmed exclusive expression of the Notch ligand JAG1 in the ICs (Fig. 4B).

Fig. 4 The IC-to-PC transition is driven by Notch ligand and receptor expression.

(A) Transcriptional profiles demonstrating the spectrum of expression of Notch genes in the collecting duct. Cells are ordered in pseudotime, and color represents expression levels. (B) Double immunofluorescence staining for AQP2 (red) and JAG1 (green) in the kidney collecting duct. (C) Generation of mice with inducible expression of Notch (ICN1) in kidney tubules (left). Dox, doxycycline. Excess AQP2-positive cells and reciprocally decreased ATP6V1B1-positive cells are found in Pax8rtTA/NICD tubules (mean ± SD; n = 3) (right). *P < 0.01. (D) In silico deconvolution of mouse kidney bulk RNA profiling data. Wild-type and Pax8rtTA/NICD samples were used for analysis. (E) Immunofluorescence quantification of cells labeled with AQP2 and ATP6V1B1 in control mice and a mouse model of CKD induced by folic acid (FA) (mean ± SD; n = 3). *P < 0.01. (F) In silico deconvolution of mouse kidney bulk RNA profiling. Control and kidney samples from FA-injected mice were used for analysis. (G) Immunofluorescence staining for AQP2 and ATP6V1B1 in control, Pax8rtTA/NICD, and FA-induced mouse model collecting ducts. AQP2-positive cells are abundant in the latter two and, conversely, ATP6V1B1-positive cells disappear. (H) In silico deconvolution of bulk RNA profiling data derived from kidney biopsy samples of patients with CKD (n = 91). The histological fibrosis scores and HES1 expression levels for the corresponding samples are also shown. (I) Total serum bicarbonate levels in control mice and in mice with FA-induced kidney fibrosis (mean ± SD; n = 5).

To examine whether Notch signaling drives the IC-to-PC cell transition, we generated Pax8rtTA/NICD mice, which enable inducible transgenic expression of the conserved Notch intracellular domain portion of the receptor, specifically in differentiated kidney tubule cells (Fig. 4C). This experimental model allowed us to study only the IC-to-PC transitions occurring in adult mice, as opposed to those occurring during embryogenesis (15). We found that Notch expression disrupted cellular patterning. The number of cells expressing the PC marker AQP2 was increased, whereas the number of cells expressing the IC marker ATP6V1B1 and the type A IC marker ADGRF5 was reduced in parallel (Fig. 4, C and G, and fig. S20). The Notch-mediated transition appeared nearly complete, given that the cells also expressed multiple PC markers, including AQP3 and HSD11B2 (figs. S21 and S22). Last, in silico deconvolution analysis of bulk RNA-profiling data, examining marker gene expression, was performed for control and Pax8rtTA/NICD mice (20) to estimate the proportion of ICs and PCs in the collecting duct. The data were consistent with the results of our lineage tracing experiments (Fig. 4D). Collectively, these data indicate that Notch receptor expression and signaling are sufficient to drive the IC-to-PC transition in the collecting duct of the adult kidney.

Because increased Notch expression has been reported in patients with and animal models of kidney disease (26, 27), we examined whether disease states disrupt the relative numbers of PCs and ICs. In a mouse model of CKD induced by folic acid (FA), which shows structural and functional damage resembling that seen in human CKD, we found a loss of the typical alternating patterns of ICs and PCs. We observed an increase in AQP2-positive cells and a decrease in ATP6V1B1-positive cells (as well as ADGRF5-positive type A ICs) compared with untreated mice (Fig. 4, E and G, and fig. S20). Computational cell deconvolution analysis of bulk RNA sequencing and analysis of marker gene expression in control and FA-induced kidney disease models yielded data consistent with a shift from IC to PC fate (Fig. 4F and fig. S23). Using cell markers identified in mice, we performed computational deconvolution analysis of kidney biopsy samples from 91 patients with hypertensive and diabetic CKD (fig. S24). Again, we found that in comparison with healthy samples, the diseased tissue samples showed a higher ratio of PCs to ICs (Fig. 4H and fig. S25), consistent with increased Notch signaling and HES1 expression (that is indicative of active Notch signaling) in these samples. The shift toward PCs did not correlate with increased expression of cell proliferation–associated genes in PCs or with increased expression of cell death–associated genes in ICs (fig. S25).

Last, we analyzed whether the increased IC-to-PC transition that we observed in the mouse model of CKD (Fig. 4C) and in kidney biopsy samples from patients with CKD (Fig. 4H) had a functional consequence. ICs are uniquely associated with acid secretion in the kidney because they express H+-transporting genes. Conversely, mutations in genes encoding proton pumps such as ATP6V1B1 cause metabolic acidosis, an accumulation of acid in many compartments of the body (fig. S19C and Fig. 2A). We found that total blood CO2 levels (composite measure of serum bicarbonate and partial pressure of CO2) were significantly reduced in the FA-induced kidney disease mouse model, consistent with metabolic acidosis (Fig. 4I). Together, these data show that (i) the IC-to-PC transition is mediated by Notch ligand (IC) and receptor (PC) expression and (ii) a shift toward the PC fate is the likely cause of metabolic acidosis in mouse models of and patients with CKD.


Efforts to describe the cell types that make up the kidney date back to the invention of the microscope. Over the past century, a kidney cell annotation has been developed that is based on the organ’s functions of transporting water and different types of salts. Here, we provide a molecular definition of cell types in the mouse kidney obtained by single-cell RNA sequencing of 57,979 cells. At this resolution, we distinguished 21 major cell types defined by quantitative gene expression; these cells included almost all previously described cell types, newly defined transitional cells in the collecting duct, and two additional undefined cells (clusters 9 and 16; Fig. 1). Our work complements previous efforts that have applied this technology to the kidney. Single-cell sequencing has been used to study fetal mouse kidneys and sorted kidney segments (7, 28, 29), and a recent study identified kidney cell composition changes in patients with lupus nephritis (30).

Our kidney cell atlas provides insight into kidney function and disease pathogenesis. It demonstrates that the expression of monogenic kidney disease genes is restricted to a single cell type. Therefore, most genetic diseases of the kidney can be traced to single cell types. In this light, each cell type appears to make a nonredundant contribution to a specific type of kidney disease. In contrast, previous transcriptomic studies identified changes in multiple cell types, using aggregated data from different kidney diseases. Last, it appears that the single-cell transcriptomics data can be used to infer cell type–specific function in vivo at the organismal level.

The atlas also highlights the role of the collecting duct system of the kidney in health and disease. The expression of genes harboring mutations associated with human disorders such as metabolic acidosis, CKD, and high blood pressure is specifically localized to this kidney segment. One of the most striking results of our analysis of the mouse kidney was the identification of an unexpected cell type related to the well-known ICs and PCs. Computational and lineage tracing analyses indicated that cells of this type are most likely transitional cells and that the number and patterning of ICs and PCs are controlled by Notch signaling in adult mice. This finding suggests that the Notch pathway may play a role in CKD. We speculate that the transition between PCs and ICs is a constitutive process that is activated in disease conditions, because our cell trajectory analysis demonstrates that the transitions occur at low frequency in healthy mice. PCs may be irreplaceable, considering that they are responsible for sodium and water balance and are involved in the regulation of serum potassium levels (31). Elevated serum potassium can cause fatal cardiac conduction abnormalities in patients with chronic kidney failure. On the other hand, acid accumulation owing to the loss of ICs can be partially compensated by regulating the respiratory rate, and, as a result, the organism can maintain near-normal serum pH. Perhaps this rationale explains the preservation or even expansion of PCs, unlike ICs, in disease states.

We have generated a comprehensive cell atlas of the mouse kidney, identified cell type–specific markers along with previously unrecognized cell types, and uncovered unexpected cell plasticity. This information will enhance our understanding of normal kidney function and disease development.

Supplementary Materials

Materials and Methods

Figs. S1 to S25

Tables S1 to S3

References (3239)

References and Notes

  1. See supplementary materials.
Acknowledgments: Funding: Work in the Susztak laboratory is supported by NIH NIDDK R01 DK076077, DK087635, DK105821, and DP3 DK108220. J.P. is supported by American Diabetes Association Training grant #1-17-PDF-036. M.W. and J.B. are supported by NIH 1U54DK104309-01, NIH 2R01DK073462, UG3 DK114926-01, and a Columbia Precision Medicine Pilot Award. Author contributions: J.P. performed computational analysis with assistance from C.Q. and M.L., R.S. generated sequencing data with assistance from J.P. and A.K., R.S. and M.W. performed experiments with assistance from S.H., K.S. designed the research, and K.S., J.P, and J.B. wrote the paper. Competing interests: The authors declare no competing interests. Data and materials availability: Processed and raw data can be downloaded from NCBI GEO (accession number GSE107585).

Stay Connected to Science

Navigate This Article