NEPHROLOGY

Single-cell transcriptomics of themouse kidney reveals potentialcellular targets of kidney diseaseJihwan Park,1* Rojesh Shrestha,1* Chengxiang Qiu,1 Ayano Kondo,1 Shizheng Huang,1

Max Werth,2 Mingyao Li,3 Jonathan Barasch,2 Katalin Suszták1†

Our understanding of kidney disease pathogenesis is limited by an incomplete molecularcharacterization 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 kidneysby using unbiased single-cell RNA sequencing. On the basis of gene expression patterns, weinfer that inherited kidney diseases that arise from distinct genetic mutations but sharethe same phenotypic manifestation originate from the same differentiated cell type.We alsofound that the collecting duct in kidneys of adult mice generates a spectrum of cell typesthrough a newly identified transitional cell. Computational cell trajectory analysis and in vivolineage tracing revealed that intercalated cells and principal cells undergo transitions mediatedby the Notch signaling pathway. In mouse and human kidney disease, these transitions wereshifted toward a principal cell fate and were associated with metabolic acidosis.

The kidney is a highly complex organ thatperforms many diverse functions that areessential for health. It removes nitrogen,water, and other waste products from theblood. It controls blood electrolytes and

acid-base balance, and it secretes hormones thatregulate blood composition and blood pressure.The kidney consists of several functionally andanatomically discrete segments. The glomerulusis a specialized group of capillaries that filters theblood and produces the primary filtrate of waterand solutes such as sodium, potassium glucose,and bicarbonate. The proximal tubules then re-absorb themajority of the water and electrolytes,whereas solutes such as uric acid, organic anions,potassium, and protons are secreted into the fil-trate. The loop of Henle is primarily involved insolute concentration. The distal tubule and thecollecting duct are segments where highly regu-lated solute transport occurs. Thus, each segmentis critical for maintaining electrolyte and waterhomeostasis.In the past, kidney cells have been annotated

on the basis of their function, their anatomicallocation, or the expression of a small number ofmarker genes (1), yet these classification systemsdo not fully overlap. An emerging technologycalled single-cell transcriptional profiling allowsinvestigators to monitor global gene regulationin thousands of individual cells in a single ex-periment (2, 3). In principle, this technologycould answer central questions in kidney biology

and disease pathogenesis because it has the po-tential to provide fourdistinct types of information.First, unbiased single-cell clustering can rede-

fine kidney cell types on the basis of only theirglobal transcriptome patterns (4). Such analyseshave already been applied to other organs (2, 5–7)and even to whole multicellular organisms (8, 9).These experiments have identified previously un-recognized cells and have catalogedmarker genesfor previously defined cells, indicating that thisapproach has the potential to redefine kidneycell types.Second, single-cell analysis may help dissect

the mechanisms underlying common kidneydiseases (10, 11). In general, kidney pathologieshave been grouped together by their temporalpatterns (acute or chronic) or by their targetstructures (glomerular versus tubular), whichhas obscured the underlying biology. Previouslyobtained bulk transcriptome profiles have gen-erated readouts only for predominant cell pop-ulations such as the proximal tubular cells (12).Kidney segment–specific RNA-sequencing anal-ysis of the rat kidney has provided useful re-sources (13), but single-cell analysis can potentiallyfurther exploit cell type–specific changes andidentify previously unrecognized cell types dur-ing disease modulation, independent of precon-ceived cellular definitions.Third, single-cell analysis may be able to iden-

tify fluctuating states of the same cell type. It isgenerally believed that terminally differentiatedcells have limited plasticity.Most cell plasticity inadults has been observed in the context of dif-ferentiation of progenitor cells, best described inthe blood and intestine (14). Such cellular tran-sitions have also been documented during thedevelopment of the collecting duct (15–17). Forexample, subtypes of intercalated cells (ICs) canchange their functional polarity. In addition,stem cell–like populations originating from the

principal cell types (PCs) may persist in the adultcollecting duct and respond to external stimuli(18, 19), but definition of these plastic cells islacking.Fourth, current models of kidney disease can-

not distinguish primary cell autonomous re-sponses from secondary cell nonautonomousresponses. Single cell–specific gene expressionprofiles, in contrast, may help identify the read-out of disease-associated gene mutations ineach cell.

Single-cell profiling and unbiasedclustering of mouse kidney cells

We first cataloged mouse kidney cell types inan unbiased manner by using droplet-basedsingle-cell RNA sequencing (20). We isolatedand sequenced a total of 57,979 cells from wholekidney cell suspensions derived from seven healthymale mice (one kidney per mouse). Using strin-gent quality controls (20), we further analyzed43,745 cells. Clustering analysis identified 16 dis-tinct cell clusters consisting of as few as 24 cellsto as many as 26,482 cells per cluster (the clus-ters were restricted to a minimum of 20 cells)(Fig. 1A).We next performed several important quality-

control analyses to validate our map. First, weensured that cells from the seven kidneys weredistributed evenly in all 16 clusters and that eachcluster contained cells from more than four ex-periments (fig. S1). Next, we examined the effectof mitochondrial gene content (fig. S2A). Theclustering of cells was not affected by mitochon-drial gene content (fig. S2, B to E). Furthermore,genes whose expression positively correlatedwith mitochondrially encoded proteins wereassociated with solute transport (which requiresabundant energy) rather than with cellular stressresponses (fig. S3). This indicates that the increasedmitochondrial gene count was inherent to spe-cific (proximal and distal tubule) cell types in thekidney. In addition, by testing different clus-tering methods, we found that most methodsidentified similar cell groups (fig. S4), express-ing the same group of marker genes with lim-ited variations in cell separation. Last, we showedthat decreasing the cell number from 40,000 to10,000, 3000, or 1000 cells (fig. S5A) was asso-ciated with increasing uncertainty in cell clus-ter identification and the loss of rarer cell types(fig. S5, B and C).

Classification of kidney cells based oncell type–specific marker genes

To define the identity of each cell cluster, wegenerated cluster-specific marker genes byperforming differential gene expression analysis(Fig. 1B, fig. S6, and table S1) (20). In many cases,the unbiased cluster identifier was a known celltype–specific marker, such as Kdr (encodingvascular endothelial growth factor receptor 2) forendothelial cells, Nphs1 (nephrin) and Nphs2(podocin) for podocytes, Slc12a1 (Na-K-2Cl co-transporter) for the ascending loop of Henle,and Slc12a3 (thiazide-sensitive sodium chlo-ride cotransporter) for the distal convoluted

RESEARCH

Park et al., Science 360, 758–763 (2018) 18 May 2018 1 of 6

1Renal Electrolyte and Hypertension Division, Department ofMedicine and Genetics, University of Pennsylvania, Philadelphia,PA 19104, USA. 2Renal Division, Department of Medicine,Columbia University, New York, NY 10032, USA. 3Department ofBiostatistics, Epidemiology and Informatics, University ofPennsylvania, Philadelphia, PA 19104, USA.*These authors contributed equally to this work.†Corresponding author. Email: ksusztak@pennmedicine.upenn.edu

on March 4, 2020

http://science.sciencem

ag.org/D

ownloaded from

tubule (Fig. 1B). Immune cells and endothelialcell clusters separated from epithelial cells, butthe ureteric bud– (clusters 6 to 8) and meta-nephric mesenchyme–derived (clusters 2 to 5)epithelial clusters were more closely aligned(Fig. 1A). Although some of the markers werealready known, we identified a large number ofadditional markers, including Cdkn1c and Bcamfor podocytes (Fig. 1B and fig. S6B). Furtheranalysis identified eight subclusters within clus-ters 1, 3, and 7 (Fig. 1C and tables S2 and S3).Cluster 1 separated into endothelial cells; peri-cyte, vascular smooth muscle, and mesangial–likecells; and descending loop of Henle (DLH) cells.Cluster 3 (proximal tubules) separated into S1,S2, and S3 segments or proximal convoluted andstraight segments (fig. S7). ICs (cluster 7) sepa-rated into types A and B.To reliably assign a specific cell type to each

cell cluster, we first correlated our gene expres-sion results with bulk RNA-sequencing data frommicrodissected rat kidney segments (fig. S8) andmicroarray data on human immune cell types (fig.S9). To further validate our clustering analysis,we used Nphs2CremT/mG, SclCremT/mG, andCdh16CremT/mGmice as reporter lines to markpodocytes, endothelium, and tubule cells withgreen fluorescent protein (GFP). The GFP ex-pression in thesemodels confirmed the proposedcell identity of our cell clusters (fig. S10). Alto-gether, our single-cell transcriptome atlas providesa molecular definition of 18 previously definedkidney and immune cell types, as well as threenewly defined cell types.

Mendelian disease genes show celltype specificity

We next tested the hypothesis that hereditarykidney diseases that are characterized by thesame phenotypic manifestations originate fromthe same cell type. We also explored whether thefunctions of specific cell types in the mouse kid-ney could be inferred from the expression patternof human genes whose loss of function results inkidney disease.We found that themouse homologsof 21 of 29 genes that have been associated withmonogenic inheritance of proteinuria in hu-manswere expressed in only one cell type—namely,the podocyte of the glomerulus (Fig. 2A and fig.S11). Although earlier studies have implicateddefects in endothelial cells and proximal tu-bules in the development of proteinuria, andfunctional and structural changes in these celltypes can be seen in patients with proteinuria,our results unequivocally show that podocytedysfunction is the principal reason for protein-uria (21). As another example, we found thatthe mouse homologs of genes associated withrenal tubule acidosis (RTA) in humans wereexpressed only by ICs of the collecting duct,confirming the major role of these cells in acid-base homeostasis (Fig. 2A). Furthermore, mousehomologs of genes that have been implicatedin blood pressure regulation through analysisof human Mendelian diseases, such as Wnk4,Wnk1, Klh3, and Slc12a3, were expressed spe-cifically in the distal convoluted tubule, whereas

Nr3c2, Scnn1b, Scnn1g, and Hsd11b2 were spe-cifically expressed by PCs of the collecting duct(Fig. 2A and figs. S11 and S13).Following the same logic, we annotated the

expression of putative complex-trait disease

genes that have been associated with blood pres-sure, chronic kidney disease (CKD) and serummetabolite levels, nephrolithiasis (e.g., Slc34a1),and RTA (e.g., Atp6v1b1) (Fig. 2B and figs. S12and S13) (22–24). We found that most genes

Park et al., Science 360, 758–763 (2018) 18 May 2018 2 of 6

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 neighborembedding (tSNE) map (center). Left panels are subclusters of clusters 1, 3, and 7. Percentages of assignedcell types are summarized in the right panel. Endo, containing endothelial, vascular, and descending loopof Henle; Podo, podocyte; PT, proximal tubule; LOH, ascending loop of Henle; DCT, distal convoluted tubule;CD-PC,collectingduct principal cell; CD-IC, collectingduct intercalated cell; CD-Trans, collectingduct transitionalcell; Fib, fibroblast;Macro,macrophage;Neutro, neutrophil; lymph, lymphocyte;NK, natural killercell. (BandC)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 endothelialcells (Endo), pericytes and vascular smooth muscle cells (Peri), and descending loop of Henle (DLH)cells.Cluster3 (proximal tubules) separates intoS1,S2, andS3segmentsorproximal convoluted tubules (PCT)and proximal straight tubules (PST). In cluster 7, intercalated cells (ICs) separate into types A and B.

RESEARCH | RESEARCH ARTICLEon M

arch 4, 2020

http://science.sciencemag.org/

Dow

nloaded from

implicated in these traits were expressed onlyin a single cell type, such as collecting duct cells(RTA) or proximal tubule cells (nephrolithia-sis). The expression of genes associated withplasma metabolite levels—such as Slc17a3 (uricacid), Slc51a (bile acid), and Slc16a9 (carnitine)(22–24)—and CKD showed strong enrichmentfor proximal tubule–specific expression, where-as blood pressure–associated genes were most-ly expressed in collecting duct cells. Thus, oursingle-cell transcriptomic analysis highlights

specific cells responsible for specific kidney-related disorders, as well as the critical func-tions of these cells.

Identification of a previouslyunrecognized cell type in thecollecting duct

The collecting duct of the kidney differs fromall other kidney epithelia because it originatesfrom the ureteric bud and not from the meta-nephric mesenchyme. This compartment is com-

posed of at least three distinct cell types: the PCs,which are responsible for sodium, water reabsorp-tion, and potassium secretion, and the type Aand B ICs, which are responsible for acid andalkali secretion, respectively. We identified thegenes encoding aquaporin 2 (Aqp2) and H+-ATPase (H+-dependent adenosine triphosphatase)subunit (Atp6v1g3) as the key marker genes forclusters 6 and 7, defining these clusters as PCsand ICs (Figs. 1 and 3, A and B).Unexpectedly, our single-cell profiling iden-

tified a third cell cluster. This cell cluster (clus-ter 8) expressed markers of both ICs and PCs(“double-positive cells”; Fig. 3, A and B) andadditional cell type–specific markers. We per-formed double immunofluorescence stainingand in situ hybridization with probes for Aqp2and Atp6v1b1 (Fig. 3C) and cell type–specificmarkers such as Parm1 and Sec23b (Fig. 3, D toF, and fig. S14) to validate the existence of thiscell type.To further investigate this unexpected cell type,

we used the Monocle toolkit to perform cell tra-jectory analysis using pseudotime reconstitutionof clusters 6 to 8 (20). We found that the newlyidentified cells were located between PCs andICs, suggesting that cluster 8 is a transitional celltype (Fig. 3G). Transitional cells showed low ex-pression levels of stress response genes and cellcycle genes, and these cells were present in allbatches of our kidney isolates (figs. S15 and S16),excluding the possibility that they were injuredcells, a proliferating subtype of collecting ductcells, or an artifact. Furthermore, cell trajectoryanalysis separated ICs into types A and B andPCs into their subtypes (PCs in the collectingduct and connecting tubule), as previouslyidentified (Fig. 3G and fig. S17) (7, 25). Theseresults indicate that the collecting duct containsnot only PCs and ICs but a third distinct, tran-sitional cell type; this raises the possibility thatICs and PCs represent two ends of a spectrum ofcellular phenotypes and that they may undergocellular transitions.

Fluorescent lineage tracing confirmsplasticity of collecting duct cells

We next examinedwhether transitional cells couldbe identified by conventional in vivo lineage tracingand whether they match our computational char-acterization.Wegeneratedmice that carry a lineagetag in differentiated PCs (Aqp2CremT/mG) or indifferentiated ICs (Atp6CremT/mG) (Fig. 3, H andI). We performed triple immunofluorescence la-beling in these animals by staining for GFP (allcells of a specificmarker origin), AQP2 (PCs), andATP6V1B1 (ICs). As expected, we found thatmostof the GFP-positive cells were also AQP2-positivein the Aqp2CremT/mGmice. A subset of the GFP-positive cells expressed ATP6V1B1, an IC marker,butnotAQP2.A smaller subsetwasdouble-positiveforATP6V1B1 andAQP2. Among theAqp2CremT/mGGFP-positive cells, 61.6% were AQP2-positive, 29.2%were ATP6V1B1-positive, and 9.2% were double-positive for AQP2 and ATP6V1B1 (Fig. 3H). Similaranalyses were performed with the Atp6CremT/mGlineage, which showed that double-positive (AQP2

Park et al., Science 360, 758–763 (2018) 18 May 2018 3 of 6

Fig. 2. Discrete human disease phenotypes are due to mutations in single specific celltypes. Single cell–type specific average expression of human (A) monogenic disease genesand (B) complex-trait genes identified by genome-wide association studies. Mean expressionvalues of the genes were calculated in each cluster. The color scheme is based on z-scoredistribution; the map only shows genes with maximum z-scores > 2. In the heatmap, each rowrepresents one gene, and each column is a single cell type (defined in Fig. 1). The full list of celltypes and genes is shown in figs. S11 and S12.

RESEARCH | RESEARCH ARTICLEon M

arch 4, 2020

http://science.sciencemag.org/

Dow

nloaded from

and ATP6V1B1–positive) transitional cells andATP6V1B1-negative true PCs can originate fromATP6V1B1-positive ICs (Fig. 3I).To determine whether cell proliferation might

be responsible for this cell plasticity, we calcu-lated the expression levels of cell cycle–regulatedgenes in the single-cell transcriptome and in celltrajectory maps. We found that only clusters 9and 16 (newly identified cell types 1 and 2), notany of the collecting duct clusters, expressed highlevels of the cell cycle genes (fig. S18). This sug-gests that cluster 8 is likely to be a transitionalcell population and not a proliferating progeni-tor cell. Thus, in vivo lineage tracing analysisconfirmed transitions of PCs and ICs not onlyduring development (15–17) but also in the adult

collecting duct through a newly identified tran-sitional cell type.

Collecting duct cell plasticity, driven byNotch signaling, results in abnormal cellpopulations in CKD

For further analysis of the plasticity of collectingduct cells, we identified genes whose expressionlevels change during transitions of PCs and ICs(fig. S19, A and B) (20). PCs showed enrichedexpression of genes associated with cell adhesion,water homeostasis, and salt transport, whereasICs showed enriched expression of genes asso-ciated with ATP hydrolysis and synthesis, coupledproton transport, and oxidation-reduction pro-cesses (fig. S19C). The gene expression patterns

revealed that the Notch signaling pathway wasactivated during the transition of ICs to PCs.Notch regulates the cellular identity of neigh-boring cells by the expression of either Notchligands or Notch receptors. Alternating expres-sion of ligands and receptors creates a signal-sending cell (Notch-off) and a signal-receivingcell (Notch-on). Genes encoding Notch lig-ands, such as Jag1, were highly expressed byICs, whereas their expression levels were lowin PCs (Fig. 4A). In contrast, PCs showed highexpression levels of Notch2 receptor and itstranscriptional target Hes1 (which encodes atranscription factor), suggesting that PCs arethe Notch signal–receiving cells in the collect-ing duct. Immunofluorescence studies confirmed

Park et al., Science 360, 758–763 (2018) 18 May 2018 4 of 6

Fig. 3. Identification of a transitional cell typeand a conversion process in the kidneycollecting duct. (A) The expression levels ofmarker genes across the 16 clusters. The y axisshows 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) Representativeimmunofluorescence images of AQP2(PC marker), ATP6V1B1 (IC marker), andDAPI (4′,6-diamidino-2-phenylindole) in thekidney collecting duct. The arrow indicates thetransitional PC-IC cell type expressing AQP2and ATP6V1B1. (D) Heatmap showing theexpression levels of differentially expressedgenes in collecting duct cell types. The colorscheme is based on z-score distribution.(E) Venn diagram showing the overlaps ofdifferentially expressed genes between PCs, ICs,and the newly identified cell type. (F) Immuno-fluorescence staining for PARM1 (transitionalcell–specific) and AQP2 (upper panels) orATP6V1B1 (lower panels) in the kidney collectingduct. “Double-positive” cells are shown by thearrows. (G) Ordering single cells along a cellconversion trajectory using Monocle. Threecollecting duct cell clusters were used forordering and plotted in low-dimensionalspace with different colors. The tSNE plotsnext to the trajectory map show differentiallyexpressed genes in the corresponding celllineages. (H) Aqp2CremT/mG mouse modelused for lineage tracing of AQP2-positivecells (left) and immunofluorescence stainingfor GFP, ATP6V1B1, and AQP2 (right). Thefar-right panel shows the quantification of GFP-positive cells (mean ± SD; n = 3). AQP2-drivenGFP (white) is found in PCs (red and white)and ICs (green and white). (I) Atp6CremT/mGmouse model used for lineage tracing ofATP6ase-positive cells (left) and immuno-fluorescence staining for GFP, ATP6V1B1,and AQP2 in Atp6CremT/mG mice (right).ATP6V1B1-driven GFP (white) is foundin PCs (red and white), ICs (green and white),and transitional cells (red, green, and white).

RESEARCH | RESEARCH ARTICLEon M

arch 4, 2020

http://science.sciencemag.org/

Dow

nloaded from

exclusive expression of the Notch ligand JAG1in the ICs (Fig. 4B).To examine whether Notch signaling drives the

IC-to-PC cell transition, we generated Pax8rtTA/NICD mice, which enable inducible transgenicexpression of the conserved Notch intracellulardomain portion of the receptor, specifically indifferentiated kidney tubule cells (Fig. 4C). Thisexperimental model allowed us to study onlythe IC-to-PC transitions occurring in adult mice,as opposed to those occurring during embryo-genesis (15). We found that Notch expressiondisrupted cellular patterning. The number ofcells expressing the PC marker AQP2 was in-creased, whereas the number of cells express-ing the IC marker ATP6V1B1 and the type A ICmarker ADGRF5 was reduced in parallel (Fig. 4,C and G, and fig. S20). The Notch-mediatedtransition appeared nearly complete, given thatthe cells also expressed multiple PC markers,including AQP3 and HSD11B2 (figs. S21 and

S22). Last, in silico deconvolution analysis ofbulk RNA-profiling data, examining markergene expression, was performed for control andPax8rtTA/NICD mice (20) to estimate the pro-portion of ICs and PCs in the collecting duct.The data were consistent with the results of ourlineage tracing experiments (Fig. 4D). Collect-ively, these data indicate that Notch receptorexpression and signaling are sufficient to drivethe IC-to-PC transition in the collecting duct ofthe adult kidney.Because increased Notch expression has been

reported in patients with and animal models ofkidney disease (26, 27), we examined whetherdisease states disrupt the relative numbers of PCsand ICs. In a mouse model of CKD induced byfolic acid (FA), which shows structural and func-tional damage resembling that seen in humanCKD, we found a loss of the typical alternatingpatterns of ICs and PCs.We observed an increaseinAQP2-positive cells and a decrease in ATP6V1B1-

positive cells (as well as ADGRF5-positive type AICs) compared with untreated mice (Fig. 4, Eand G, and fig. S20). Computational cell decon-volution analysis of bulk RNA sequencing andanalysis of marker gene expression in controland FA-induced kidney disease models yieldeddata consistent with a shift from IC to PC fate(Fig. 4F and fig. S23). Using cell markers identi-fied in mice, we performed computational decon-volution analysis of kidney biopsy samples from91 patients with hypertensive and diabetic CKD(fig. S24). Again, we found that in comparisonwith healthy samples, the diseased tissue sam-ples showed a higher ratio of PCs to ICs (Fig. 4Hand fig. S25), consistent with increased Notchsignaling andHES1 expression (that is indicativeof active Notch signaling) in these samples. Theshift toward PCs did not correlate with increasedexpression of cell proliferation–associated genesin PCs or with increased expression of cell death–associated genes in ICs (fig. S25).

Park et al., Science 360, 758–763 (2018) 18 May 2018 5 of 6

Fig. 4. The IC-to-PC transition is driven byNotch ligand and receptor expression.(A) Transcriptional profiles demonstrating thespectrum of expression of Notch genes inthe collecting duct. Cells are ordered inpseudotime, and color represents expressionlevels. (B) Double immunofluorescence stainingfor AQP2 (red) and JAG1 (green) in the kidneycollecting duct. (C) Generation of mice withinducible expression of Notch (ICN1) in kidneytubules (left). Dox, doxycycline. ExcessAQP2-positive cells and reciprocally decreasedATP6V1B1-positive cells are found in Pax8rtTA/NICD tubules (mean ± SD; n = 3) (right).*P < 0.01. (D) In silico deconvolution of mousekidney bulk RNA profiling data. Wild-typeand Pax8rtTA/NICD samples were used foranalysis. (E) Immunofluorescence quantifica-tion of cells labeled with AQP2 and ATP6V1B1in control mice and a mouse model of CKDinduced by folic acid (FA) (mean ± SD; n = 3).*P < 0.01. (F) In silico deconvolution of mousekidney bulk RNA profiling. Control and kidneysamples from FA-injected mice were usedfor analysis. (G) Immunofluorescence stainingfor AQP2 and ATP6V1B1 in control, Pax8rtTA/NICD, and FA-induced mouse model collectingducts. AQP2-positive cells are abundant inthe latter two and, conversely, ATP6V1B1-positive cells disappear. (H) In silicodeconvolution of bulk RNA profiling dataderived from kidney biopsy samples ofpatients with CKD (n = 91). The histologicalfibrosis scores and HES1 expression levelsfor the corresponding samples are also shown.(I) Total serum bicarbonate levels in controlmice and in mice with FA-induced kidneyfibrosis (mean ± SD; n = 5).

RESEARCH | RESEARCH ARTICLEon M

arch 4, 2020

http://science.sciencemag.org/

Dow

nloaded from

Last, we analyzedwhether the increased IC-to-PCtransition that we observed in the mouse modelof CKD (Fig. 4C) and in kidney biopsy samplesfrom patients with CKD (Fig. 4H) had a func-tional consequence. ICs are uniquely associatedwith acid secretion in the kidney because theyexpress H+-transporting genes. Conversely, muta-tions in genes encoding proton pumps such asATP6V1B1 cause metabolic acidosis, an accumu-lation of acid in many compartments of the body(fig. S19C and Fig. 2A). We found that total bloodCO2 levels (composite measure of serum bicar-bonate and partial pressure of CO2) were signif-icantly reduced in the FA-induced kidney diseasemouse model, consistent with metabolic acidosis(Fig. 4I). Together, these data show that (i) theIC-to-PC transition is mediated by Notch ligand(IC) and receptor (PC) expression and (ii) a shifttoward the PC fate is the likely cause of meta-bolic acidosis in mouse models of and patientswith CKD.

Discussion

Efforts to describe the cell types thatmake up thekidney date back to the invention of the micro-scope. Over the past century, a kidney cell annota-tionhasbeendeveloped that is basedon theorgan’sfunctions of transporting water and different typesof salts. Here, we provide a molecular definitionof cell types in the mouse kidney obtained bysingle-cell RNA sequencing of 57,979 cells. At thisresolution, we distinguished 21 major cell typesdefined by quantitative gene expression; thesecells included almost all previously described celltypes, newly defined transitional cells in the col-lecting duct, and two additional undefined cells(clusters 9 and 16; Fig. 1). Our work complementsprevious efforts that have applied this technologyto the kidney. Single-cell sequencing has been usedto study fetal mouse kidneys and sorted kidneysegments (7, 28, 29), and a recent study identifiedkidney cell composition changes in patients withlupus nephritis (30).Our kidney cell atlas provides insight into

kidney function and disease pathogenesis. It dem-onstrates that the expression of monogenic kid-ney disease genes is restricted to a single celltype. Therefore,most genetic diseases of the kidneycan be traced to single cell types. In this light,each cell type appears to make a nonredundantcontribution to a specific type of kidney disease.In contrast, previous transcriptomic studies iden-tifiedchanges inmultiple cell types, usingaggregateddata from different kidney diseases. Last, it ap-

pears that the single-cell transcriptomics data canbe used to infer cell type–specific function in vivoat the organismal level.The atlas also highlights the role of the col-

lecting duct system of the kidney in health anddisease. The expression of genes harboring mu-tations associated with human disorders suchas metabolic acidosis, CKD, and high bloodpressure is specifically localized to this kidneysegment. One of the most striking results of ouranalysis of the mouse kidney was the identifica-tion of an unexpected cell type related to the well-known ICs and PCs. Computational and lineagetracing analyses indicated that cells of this typeare most likely transitional cells and that the num-ber and patterning of ICs and PCs are controlledby Notch signaling in adult mice. This findingsuggests that the Notch pathway may play a rolein CKD. We speculate that the transition be-tween PCs and ICs is a constitutive process thatis activated in disease conditions, because our celltrajectory analysis demonstrates that the transi-tions occur at low frequency in healthy mice. PCsmay be irreplaceable, considering that they areresponsible for sodium and water balance andare involved in the regulation of serum potassiumlevels (31). Elevated serum potassium can causefatal cardiac conduction abnormalities in patientswith chronic kidney failure. On the other hand,acid accumulation owing to the loss of ICs can bepartially compensatedby regulating the respiratoryrate, and, as a result, the organism canmaintainnear-normal serum pH. Perhaps this rationaleexplains the preservation or even expansion ofPCs, unlike ICs, in disease states.We have generated a comprehensive cell

atlas of the mouse kidney, identified cell type–specific markers along with previously unre-cognized cell types, and uncovered unexpectedcell plasticity. This information will enhanceour understanding of normal kidney functionand disease development.

REFERENCES AND NOTES

1. W. Kriz, L. Bankir; The Renal Commission of the InternationalUnion of Physiological Sciences (IUPS), Kidney Int. 33, 1–7(1988).

2. E. Z. Macosko et al., Cell 161, 1202–1214 (2015).3. G. X. Zheng et al., Nat. Commun. 8, 14049 (2017).4. O. Rozenblatt-Rosen, M. J. T. Stubbington, A. Regev,

S. A. Teichmann, Nature 550, 451–453 (2017).5. M. J. T. Stubbington, O. Rozenblatt-Rosen, A. Regev,

S. A. Teichmann, Science 358, 58–63 (2017).6. E. Lein, L. E. Borm, S. Linnarsson, Science 358, 64–69

(2017).

7. L. Chen et al., Proc. Natl. Acad. Sci. U.S.A. 114, E9989–E9998(2017).

8. N. Karaiskos et al., Science 358, 194–199 (2017).9. J. Cao et al., Science 357, 661–667 (2017).10. F. Hildebrandt, Lancet 375, 1287–1295 (2010).11. L. F. Menezes, G. G. Germino, Wiley Interdiscip. Rev. Syst. Biol.

Med. 7, 39–52 (2015).12. H. M. Kang et al., Nat. Med. 21, 37–46 (2015).13. J. W. Lee, C. L. Chou, M. A. Knepper, J. Am. Soc. Nephrol. 26,

2669–2677 (2015).14. M. Chang-Panesso, B. D. Humphreys, Nat. Rev. Nephrol. 13,

39–46 (2017).15. M. Werth et al., eLife 6, e24265 (2017).16. H. Wu et al., J. Am. Soc. Nephrol. 24, 243–252 (2013).17. H. W. Jeong et al., J. Clin. Invest. 119, 3290–3300

(2009).18. J. Li et al., J. Am. Soc. Nephrol. 26, 81–94 (2015).19. G. J. Schwartz, J. Barasch, Q. Al-Awqati, Nature 318, 368–371

(1985).20. See supplementary materials.21. L. Malaga-Dieguez, K. Susztak, J. Clin. Invest. 123, 4996–4999

(2013).22. C. Pattaro et al., Nat. Commun. 7, 10023 (2016).23. H. R. Warren et al., Nat. Genet. 49, 403–415 (2017).24. S. Y. Shin et al., Nat. Genet. 46, 543–550 (2014).25. P. A. Welling, Annu. Rev. Physiol. 78, 415–435 (2016).26. B. Bielesz et al., J. Clin. Invest. 120, 4040–4054

(2010).27. M. T. Sweetwyne et al., Diabetes 64, 4099–4111 (2015).28. Y. Lu, Y. Ye, Q. Yang, S. Shi, Kidney Int. 92, 504–513

(2017).29. M. Adam, A. S. Potter, S. S. Potter, Development 144,

3625–3632 (2017).30. E. Der et al., JCI Insight 2, e93009 (2017).31. P. A. Welling, Semin. Nephrol. 33, 215–228 (2013).

ACKNOWLEDGMENTS

Funding: Work in the Susztak laboratory is supported byNIH NIDDK R01 DK076077, DK087635, DK105821, andDP3 DK108220. J.P. is supported by American DiabetesAssociation Training grant #1-17-PDF-036. M.W. and J.B. aresupported by NIH 1U54DK104309-01, NIH 2R01DK073462,UG3 DK114926-01, and a Columbia Precision Medicine PilotAward. Author contributions: J.P. performed computationalanalysis with assistance from C.Q. and M.L., R.S. generatedsequencing 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. wrotethe paper. Competing interests: The authors declare nocompeting interests. Data and materials availability:Processed and raw data can be downloaded from NCBI GEO(accession number GSE107585).

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/360/6390/758/suppl/DC1Materials and MethodsFigs. S1 to S25Tables S1 to S3References (32–39)

13 October 2017; accepted 27 March 2018Published online 5 April 201810.1126/science.aar2131

Park et al., Science 360, 758–763 (2018) 18 May 2018 6 of 6

RESEARCH | RESEARCH ARTICLEon M

arch 4, 2020

http://science.sciencemag.org/

Dow

nloaded from

diseaseSingle-cell transcriptomics of the mouse kidney reveals potential cellular targets of kidney

and Katalin SusztákJihwan Park, Rojesh Shrestha, Chengxiang Qiu, Ayano Kondo, Shizheng Huang, Max Werth, Mingyao Li, Jonathan Barasch

originally published online April 5, 2018DOI: 10.1126/science.aar2131 (6390), 758-763.360Science 

, this issue p. 758; see also p. 709Sciencethat genetically distinct kidney diseases with common clinical features share common cellular origins.the other is regulated by the Notch signaling pathway and is associated with metabolic acidosis. The authors also find the collecting duct appears to be a transitional state between two known cell types. The transition from one cell type tocomprehensive cell atlas of the healthy mouse kidney (see the Perspective by Humphreys). An unexpected cell type in

produced aet al.disease, which affects 10% of the world's population. Using single-cell transcriptional profiling, Park This organ's cellular complexity has hindered progress in understanding the mechanisms underlying chronic kidney

Our kidneys play a critical role in keeping us healthy, a fact of which we are reminded several times each day.Touring the kidney, cell by cell

ARTICLE TOOLS http://science.sciencemag.org/content/360/6390/758

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2018/04/04/science.aar2131.DC1

CONTENTRELATED http://science.sciencemag.org/content/sci/360/6390/709.full

REFERENCES

http://science.sciencemag.org/content/360/6390/758#BIBLThis article cites 38 articles, 12 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science. No claim to original U.S. Government WorksCopyright © 2018 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on March 4, 2020

http://science.sciencem

ag.org/D

ownloaded from