Single-Nucleus RNA Sequencing Identifies New Classes of Proximal Tubular Epithelial Cells in Kidney Fibrosis

Significance Statement

Proximal tubular cells (PTCs) are numerically the predominant constituent of the kidney and are central to regeneration versus organ fibrosis following injury. However, variations in their phenotype are not well characterized. Single-nuclear RNA sequencing revealed phenotypes of PTCs in normal mouse kidney and changes in kidneys undergoing regeneration and fibrosis following aristolochic acid exposure. Five abundant and four rare PTC phenotypes were found, with abundant clusters mapped to different tubular segments and rare phenotypes mapped to proliferative, dedifferentiated, and fibrosis-associated phenotypes. These data identify unrecognized heterogeneity in PTC phenotypes and reveal novel PTCs associated with kidney fibrosis.

Visual Abstract

Abstract

Background

Proximal tubular cells (PTCs) are the most abundant cell type in the kidney. PTCs are central to normal kidney function and to regeneration versus organ fibrosis following injury. This study used single-nucleus RNA sequencing (snRNAseq) to describe the phenotype of PTCs in renal fibrosis.

Methods

Kidneys were harvested from naïve mice and from mice with renal fibrosis induced by chronic aristolochic acid administration. Nuclei were isolated using Nuclei EZ Lysis buffer. Libraries were prepared on the 10× platform, and snRNAseq was completed using the Illumina NextSeq 550 System. Genome mapping was carried out with high-performance computing.

Results

A total of 23,885 nuclei were analyzed. PTCs were found in five abundant clusters, mapping to S1, S1–S2, S2, S2-cortical S3, and medullary S3 segments. Additional cell clusters (“new PTC clusters”) were at low abundance in normal kidney and in increased number in kidneys undergoing regeneration/fibrosis following injury. These clusters exhibited clear molecular phenotypes, permitting labeling as proliferating, New-PT1, New-PT2, and (present only following injury) New-PT3. Each cluster exhibited a unique gene expression signature, including multiple genes previously associated with renal injury response and fibrosis progression. Comprehensive pathway analyses revealed metabolic reprogramming, enrichment of cellular communication and cell motility, and various immune activations in new PTC clusters. In ligand-receptor analysis, new PTC clusters promoted fibrotic signaling to fibroblasts and inflammatory activation to macrophages.

Conclusions

These data identify unrecognized PTC phenotype heterogeneity and reveal novel PTCs associated with kidney fibrosis.

The kidneys excrete waste and maintain homeostasis by blood filtration and subsequent selective reabsorption. The core functional unit of the kidney is the nephron, of which about 1 million are present in each human kidney. The nephron extends from the glomerulus, which is the site of blood filtration, to the collecting duct system through which nephrons converge to drain urine to the lower urinary tract. Ultrafiltrate is processed to urine via active transport, which underpins reabsorption of water, electrolytes, glucose, amino acids, and other constituents, and is balanced by secretion to achieve homeostasis. Highly specialized, sequential processing of urine occurs in discrete nephron segments, reflecting underlying cellular specialization.¹

The proximal tubules are the site for the majority of the resorptive activity, comprising active transport of an estimated 150 L/d of solute-rich fluid in a healthy human.² The scale of proximal tubular resorptive activity makes the kidney a highly metabolically active organ, second only in oxygen consumption to the brain.³ Proximal tubular cells (PTCs) are highly susceptible to ischemic and toxic injury, and their responses are central to progression and recovery in AKI and chronic kidney injury of diverse etiology.⁴ PTCs play a core role in dictating renal recovery and fibrosis responses following injury and in CKD.⁵

PTCs are a preeminent contributor to the overall cellular composition of the kidney.⁶ Single-cell RNA sequencing is a powerful technique in discriminating cellular characteristics in the kidney and elsewhere. Surprisingly little advance in knowledge about PTC phenotypes has so far come from single-cell RNA sequencing, where these cells have typically been reported as a single cluster, often numerically much larger than other clusters analyzed in detail.^6,7

We hypothesized that there may be underappreciated heterogeneity in PTCs, and we used a clinically relevant mouse model of chronic aristolochic acid nephropathy (AAN) to investigate the responses of PTCs during kidney fibrosis. We have delineated PTC subtypes found in normal mouse kidney and further describe novel PTC phenotypes associated with kidney fibrosis.

Methods

Mouse Model of Chronic AAN

Eight 8- to 9-week-old C57BL/6 male mice bred by Charles River Laboratories were involved in the experiment. Animals were randomly allocated to control versus aristolochic acid (AA) injection. Chronic AAN was induced in four mice by intraperitoneal injection of AA (2.5 mg/kg body wt; Sigma A5512) twice weekly for 2 weeks (Figure 1A). Inflammation and tissue fibrosis developed after repetitive injuries, followed by tissue remodeling and fibrosis in the subsequent 2 weeks. Another four mice that did not receive intervention were used as naïve controls. Mice were housed with free access to chow and tap water on a 12-hour day/night cycle in a specific pathogen-free environment. All mice were euthanized 4 weeks after the first injection of the AAN group. We perfused chilled PBS (1×) via the left ventricle before kidney harvest. Both kidneys were harvested and processed: one for single-nucleus RNA sequencing (snRNAseq) and one for histopathology. Serum creatinine measurement was performed using the enzymatic mouse creatinine assay kit (Crystal Chem). Masson trichrome stain for the formalin-fixed, paraffin-embedded sections was performed for each mouse to confirm healthy or fibrotic status. Experiments were performed in line with institutional and United Kingdom Home Office guidelines under the authority of an appropriate project license.

Figure 1.

Descriptive analysis of the mouse model of fibrosis induced by repeated injection of Aristolochic Acid (AA). (A) Workflow of the animal model used. AA (2.5 mg/kg) was administered intraperitoneally (ip) on four occasions on days 0, 3, 7, and 10. (B) Significant difference of serum creatinine at the end of the experiment (n=7 in each group). (C) Masson trichrome stain of kidneys from a healthy mouse (left panels; naïve) and a mouse with chronic AAN (right panels; AAN) taken at day 28. Cytoplasm is stained red, collagen is stained blue, and nuclei are in dark brown, which helps to identify renal fibrosis. Significant fibrotic changes developed in the mouse kidney due to AAN. (D–F) IHC stain of Ki-67, α-smooth muscle actin (α-SMA), and HAVCR1 (Kim-1) in naïve and AAN kidneys. The micrographs shown are representative of four naïve mice and four AAN mice. (G and H) Quantification of collagen (cyan signal) in Masson trichrome stain and α-SMA signal were used to confirm fibrosis in the AAN model. (I) Nuclear Ki-67 DAB signal was used to quantify proliferating cells as a percentage of all hematoxylin-stained cells. Scale bars = 50 μm.

Isolation of Nuclei

One-quarter of a harvested kidney from each mouse was minced into <2-mm pieces and transferred to a Dounce tissue grinder containing 2 ml of lysis buffer (Nuclei EZ Lysis buffer; Sigma NUC101) supplemented with protease inhibitor (Sigma 5892970001) and RNase inhibitors (Promega N2615 and Life Technologies AM2696). Kidneys were homogenized and transferred into 50-ml tubes containing 2 ml of lysis buffer, incubated for 7 minutes on ice, filtered through a 40-μm cell strainer, and centrifuged at 500×g for 5 minutes at 4°C. The pellets were resuspended in 4 ml lysis buffer, incubated for another 7 minutes on ice, and then, centrifuged again at 500×g for 5 minutes at 4°C. The pellets were then resuspended in 4 ml wash and resuspension buffer (1× PBS, 1.0% BSA, and RNase Inhibitor; Sigma 3335399001) and filtered through a 20-μm cell strainer. Samples were then processed immediately using the 10× Genomics single-cell library preparation protocol.

Library Preparation and RNA Sequencing

Library preparation was performed using Chromium Single Cell 3′ Reagent Kits v3 (10× Genomics), and cDNA quality was evaluated by fragment analysis (5200 Fragment Analyzer System; Agilent). RNA sequencing was carried out using the Illumina NextSeq 550 System.

Bioinformatics Analysis

The sequencing data were processed using the zUMIs pipeline (version 2.3.0).⁸ The pipeline was used to first discard reads with low-quality barcodes and UMIs and then, map reads to the mouse reference assembly (Mus_musculus.GRCm38.95). The barcode-gene matrix generated by zUMIs was analyzed using the R package, Seurat (version 3.1.3).^9,10

In Seurat, cells for individual samples were retained if they contained ≥400 genes and genes identified in three or more nuclei. After merging the four naïve kidneys and four AAN kidneys, cells were filtered again to remove nuclei expressing ≤400 genes or ≥7500 genes or with mitochondrial gene expression ≥10%. The feature counts were normalized with scale.factor = 10,000. The top 2000 variable genes were identified and scaled. The principal component analysis (PCA) result of the scaled data was obtained. The number of principal components included in the downstream analysis was determined by identifying the knee point of the elbow plot generated after running the JackStraw procedure. FindNeighbors and FindClusters (resolution 0.8 as default) functions were applied on the basis of previously identified principal components to identify clusters of nuclei. The t-distributed stochastic neighbor embedding plot was used to visualize clustering results of the naïve and AAN datasets. DoubletFinder (version 2.0.2) was then used to exclude potential doublets.¹¹ After doublet removal, the naïve and AAN datasets were integrated (dims = 1:50). The integrated dataset was then scaled and processed with PCA, FindNeighbors, and FindClusters (resolution = 3.0). Final clustering results were visualized using Uniform Manifold Approximation and Projection (UMAP). For differential gene expression (DGE) analysis, we used the FindMarker command (the Wilcox method as the default). Significance was defined as a gene with an adjusted P value =0.05, a ≥0.25 average log fold difference between the two groups of cells, and presence detected in at least 10% of cells in either of the two populations. P value adjustment was performed using Bonferroni correction on the basis of the total number of genes in the dataset. Harmony (version 1.0), which is another batch integration method, was used to validate the pipeline of cell clustering and showed consistent results with Seurat (not shown).^12,13

The cell cycle analysis used CellCycleScoring to identify cells in the G2/M and S status. Cells with G2M.Score > 0.15 and G2M.Score > S.Score were assigned as G2M status. Cells with S.Score > 0.15 and S.Score > G2M.Score were assigned as G2M status. Cell with G2M.Score < 0.15 and S.Score < 0.15 were assigned as G1/G0 phase.

For trajectory analysis, we used the R package Monocle3 (version 0.2.2.0).¹⁴ The metadata from the integrated Seurat object and the top 2000 variable genes from the integrated assay were loaded to Monocle3. The analysis involved all PTC clusters and the proliferative PTCs identified in proliferative cells using default parameters and removing clusters where low gene numbers were detected (clusters 0 and 1) (Supplemental Figure 1). The PTCs were reclustered by Monocle3, and then, the trajectory analysis was performed using the learn_graph function. The RNA velocity of each cell was calculated by velocyto.py and velocyto.R (version 0.6) using the spliced and unspliced RNA counts provided by the 10× cellranger package.¹⁵ Pseudotime analysis was performed on the basis of the RNA velocity result.

We performed combined analysis of snRNAseq results using this AAN-induced CKD dataset and the murine ischemic reperfusion injury–induced AKI dataset published by Kirita et al.¹⁶ The two datasets were merged, and the top 2000 variable genes from the combined dataset were obtained for PCA. We used harmony as the batch integration method. The clustering result of all cells from both datasets was shown by UMAP. For PTC analysis, we analyzed PTCs from both datasets using the Seurat integrate data function.

Pathway Analysis

We conducted gene set enrichment analysis to understand pathways of the new PT clusters by using an R package, WebGestaltR (version 0.4.3).¹⁷ We evaluated the pathway enrichment in four major functional databases: KEGG, Panther, Reactome, and WikiPathways. The recommended false discovery rate cutoff of 0.25 was used (https://www.gsea-msigdb.org/gsea/index.jsp).

Ligand-Receptor Analysis

The analysis was on the basis of a well-reviewed dataset with 2557 ligand-receptor pairs from a published study.¹⁸ PTs S1–S3 were combined with normal PT, and fibroblast-1 and -2 were combined with fibroblasts in the ligand-receptor analysis. Ligands and receptors with an average fold change ≥0.25 in the DGE analysis were paired. Pairing results of ligand genes of New-PT1, New-PT2, New-PT3, and receptor genes of fibroblasts, immune cells, and normal PT were shown using the crossproducts of ligand-receptor gene expression on a heat map. We also show individual ligand-receptor pairs with the ligands from the New-PT clusters and receptors from fibroblasts, immune cells, and normal PTCs using a circular visualization tool, circlize (version 0.4.8), in R.¹⁹

Immunohistochemistry and Immunofluorescence Staining

Formalin-fixed, paraffin-embedded kidney sections from the AAN and the control group were used for immunohistochemistry(IHC)/immunofluorescence stain. Antigen retrieval at 120°C for 20 minutes was carried out for rehydrated kidney reactions. For IHC stains, the UltraVision LP HRP Polymer enhancer system was applied (Thermo Scientific 12624007) for rabbit/mouse primary antibodies. Sections were blocked with 3% hydrogen peroxide followed where necessary by mouse-on-mouse block (Vector MKB-2213–1) , and then Ultra V block. The sections were then stained with primary antibodies for Ki-67 (Abcam ab15580) and α-smooth muscle actin (Invitrogen MA5–11547) followed by primary antibody enhancer, HRP Polymer, DAB, and hematoxylin counterstain. For Havcr1 IHC stain, sections were blocked with 3% hydrogen peroxide, avidin/biotin block (Vector SP-2001), and 10% donkey serum, and they were stained with goat anti-HAVCR1 primary antibody (R&D AF3689). The sections were then stained with biotinylated donkey anti-goat secondary antibody (Abcam ab6884), VECTASTAIN ABC-HRP enhancer kit (Vector PK-4000), DAB, and hematoxylin counterstain. For immunofluorescence stains, sections were incubated with mouse on mouse block where needed and with 10% goat serum. Primary antibodies included anti-SLC4A4 (Invitrogen PA5–57344), anti-VCAM1(Invitrogen MA5–11447), anti-FODX1(LifeSpan BioSciences LS-B9155-LSP), anti-Akap12 (Abcam ab49849), anti-WT1 (Sigma MAB4234), anti-NCAM1 (Abcam ab220360), anti–Tenascin C (Abcam ab108930), anti-P21 (novusbio NBP2–29463), and anti-HAVCR1. For secondary antibodies, we used goat anti-mouse or goat anti-rabbit Alexa Fluor 488– or 594–conjugated antibodies (Invitrogen). After primary and secondary antibody staining, the sections were stained with Hoechst 33342. The TrueVIEW autofluorescence quenching kit (Vector SP-8400–15) was used to diminish autofluorescence. Immunostained tissue slides were visualized and digitized using a confocal laser scanning microscope (Carl Zeiss LSM800).

Quantitative Image Analysis

Immunostained tissue slides were visualized and digitized using an Olympus DP27 5MP color camera attached to a Leica DMLA microscope or by a confocal laser scanning microscope (Carl Zeiss LSM800). Images were analyzed with the ZEN2012 software (Zeiss), and quantification was performed with Qupath software.²⁰ A pixel classifier was used to detect the collagen (cyan) stain, and thresholds were determined to express the positive stain as a percentage of total area (in more than ten fields of view per animal, n=3 per group). Quantification of the α-smooth muscle actin DAB signal was used to confirm fibrosis in the AAN model. The nuclear Ki-67 DAB signal was used to quantify proliferating cells as a percentage of all hematoxylin-stained cells using QuPath’s cell detection algorithm.

Results

Sample Processing and Quality Control

We isolated kidney nuclei immediately following terminal anesthesia and tissue harvesting. Rapid processing of tissue to nuclei was used to minimize artifacts arising during processing that may have limited discrimination of PTC subclusters in previous reports of single-cell sequencing from the kidney.^6,21 The AAN group had significantly higher serum creatinine (Figure 1B). Histologic evidence confirmed the presence of renal fibrosis in mice exposed to AA and its absence in control animals (Figure 1, C–F). Compared with healthy kidneys, AAN kidneys demonstrated a significant increase in collagen deposition (cyan signal in Masson trichrome stain) and α-smooth muscle actin DAB signal, indicating renal fibrosis (Figure 1, G and H). The AAN kidneys also had a higher percentage of Ki-67–positive cells (Figure 1I). Nuclei of kidneys acquired from mice exposed to AA to induce kidney fibrosis (AAN mice) and naïve controls were processed using the 10× platform. R packages Seurat and DoubletFinder were used for data integration, quality control, and doublet removal (Supplemental Table 1).^9–11 Downstream analysis included a total of 23,885 nuclei.

Nuclei Clustering and Cell-Type Identification

The clustering result was evaluated using UMAP (Figure 2A, Supplemental Figure 1). The number of nuclei comprising each cluster and the distribution on the UMAP of nuclei from each individual mouse are shown in Supplemental Figure 2 and Supplemental Table 2. Nuclei from all mice contributed to every cluster, except for that subsequently labeled “New-PT3,” which was seen in four of four AAN-treated mice and zero of four naïve controls.

Figure 2.

Clustering and cell-type identification of 23,885 nuclei using combined datasets from four naïve mice and four AAN mice. (A) UMAP plot of the combined dataset is shown by splitting conditions. We identified all major cell types in the kidney and four new classes of cells, labeled as proliferative cell and New–proximal tubule (PT) clusters 1–3. (B) The dot plot shows the expression levels and the percentages of gene expression of the canonical genes in each distinct cell type. (C) A feature plot of regional-specific genes shows the expression of Cyp2e1 (purple) in cortical PTCs and Cyp7b1 (green) in medullary PTCs. (D) Expression of canonical genes of PCs in the outer inner medullary collecting duct (OMCD) and the inner medullary collecting duct (IMCD) shows that the PCs from the two different regions were well clustered. ATL, ascending thing limb; CNT, connecting tubule; DCT1/DCT2, distal convoluted tubule 1/2; DTL, descending thin limb; IC-A, intercalated cell, type A; IC-B, intercalated cell, type B; S1/S2/S3, segment 1/2/3 of proximal tubule.

We next used canonical markers of kidney cell populations to identify major cell types in the kidney: podocyte (Nphs1), endothelial cells (Flt1), mesangial cells (Igfbp5), juxtaglomerular (JG) cells (Ren1), PTCs (Slc34a1), proliferative cells (Top2a and Mki67), descending thin limb (Aqp1), ascending thing limb (Clcnka), thick ascending limb (TAL; Slc12a1 and Umod), distal convoluted tubule 1 (Slc12a3) and distal convoluted tubule 2 (Slc12a3 and Slc8a1), connecting tubule (Slc8a1), principal cell (PC)-outer medullary collecting duct and inner medullary collecting duct (Aqp2), intercalated cells type A (Atp6v1b1 and Slc4a1) and type B (Atp6v1b1 and Slc26a4), transitional epithelium (Upk1b), immune cells (Ptprc), and fibroblasts (Pdgfrb and Cfh). Figure 2B shows the expression level and the percentage of expression of the canonical genes per cluster. Regional-specific genes of mouse kidney identified by Ransick et al.²² were used to localize clusters in cortical-medullary and outer-inner medullary regions. Figure 2C shows regional expression of Cyp2e1 in cortical PTCs and Cyp7b1 in medullary PTCs. The expression of canonical genes for PC-outer medullary collecting duct and PC-inner medullary collecting duct shows that the PCs from the two different regions were well clustered (Figure 2D).

Clarifying the Pdgfrb+ Clusters

Some functionally distinct renal cell types share lineage and have many common features in their gene expression profiles, which have complicated their recovery in discrete clusters in recent landmark studies.^21,23 These include mesangial cells, JG cells, and fibroblasts, which differentiate from Foxd1+ cortical stromal cells and acquire similar genetic features after profibrotic stimulation.^24–26 Four clusters closely proximal to one another on UMAP showed expression of the shared markers for mesangial cells, fibroblasts, and JG cells: Cfh, Fhl2, and Pdgfrb (Supplemental Figure 3). Mesangial cells were identified from expression of the mesangial-restricted Igfbp5 and Itga8 (http://www.proteinatlas.org),^27,28 whereas JG cells were the only cluster expressing Ren1. Two renal fibroblast clusters were identified, of which the Fibroblast-1 cell number increased >200% in fibrotic kidney. We provisionally identified Fibroblast-1 as myofibroblast containing on the basis of Meis1 expression.²⁹ Our dataset is limited in this regard by low Acta2 detection (Supplemental Figure 3B), and further characterization of renal fibroblast populations may benefit from supplementary technical approaches.

Analysis of Proliferative Cells and Cell Cycle

One cluster expressed markers of proliferative cells not seen in other clusters, notably including Top2a and Mki67. Uniquely, this cluster comprised four distinct and separated subclusters of nuclei. These four subclusters were located in close proximity to the endothelial, PTC, TAL, and fibroblast clusters (Supplemental Figure 4A). Consistent with this cluster comprising proliferating cells from the adjacent clusters, the four subclusters also expressed canonical markers for the cell type (e.g., Flt1-endothelial, Slc34a1-PTC, Slc12a1-TAL, and Pdgfrb-fibroblast) (Supplemental Figure 4B). Cell cycle analysis identified cells in G2/M and S phases, localized mainly to the proliferative and immune cell clusters (Supplemental Figure 4C). G2/M arrest of PTCs contributes to fibrogenesis after kidney injury,³⁰ and cell numbers of G2M- and S-phase PTCs increased in the AAN kidneys.

Analysis of PTC Subclusters

The proximal tubule is divided into segments S1, S2, and S3 on the basis of microscopic characteristics. The S1 segment is the longest and comprises PTCs with extensive apical microvilli, basolateral infoldings, cytoplasmic complexities, numerous long mitochondria, and a prominent endocytic region.³¹ These features are present but less evident in the S2 segment, which demonstrates a gradual transition from the S1 segments and additionally displays more numerous peroxisomes and larger secondary lysozymes. The S3 segment is more distinct, comprising simple cuboidal cells without the above features.

Five clusters were identified corresponding to proximal tubular segments S1–S3: “S1,” “S1–S2,” “S2,” “S2-cortical S3,” and “medullary S3.” These PTC clusters were labeled according to their expression of genes localized to PTC segments in previous studies. Slc34a1 encodes Na⁺-P_i cotransporter 2a, which is localized in fully differentiated PTCs along all segments except for medullary S3.^22,32,33 Slc5a2 encodes a low-affinity/high-capacity Na⁺/glucose cotransporter, sodium/glucose cotransporter 2, which is responsible for 80%–90% of the glucose reabsorption in S1.³⁴ Slc22a6 encodes organic anion transporter 1 in S2, which plays a critical role in drug and xenobiotic elimination and has been linked to AAN.^33,35–38 Slc13a3 encodes sodium-dependent dicarboxylate transporter in S2, which is responsible for transport of succinate and other Krebs cycle intermediates.³⁹ Slc7a13, also known as Agt1, is an amino acid transporter localized to the apical membrane of the S3 segment and is considered an S3-canonical gene.^33,40 The expression levels of segment-specific solute transporter–related genes were reviewed to validate established PTC categories. Relevant genes were identified in the first instance from the study from the laboratory of Knepper and coworkers,¹ in which bulk deep sequencing was applied to microdissected renal tubules to identify nephron segment–specific transcriptomes. Additional studies are indicated for validation of key anchor genes below. These included markers for S1 (Slc5a2 and Slc5a12), S2 (Slc22a6, Ca4, and Slc13a3), S3 (Slc5a10, Slc7a13, and Atp11a), medullary S3 [Cyp7b1, Slc6a13, and Slc34a1(−)], and pan-PTC markers (Slc34a1, Lrp2, and Slc4a4) (Figure 3A, Supplemental Figure 5A).^1,22,41–43 The differences in genes enriched in each PT cluster supported their attribution to specific PT segments (Supplemental Figure 6).

Figure 3.

Gene expression profiles of PTC clusters. (A) The dot plot shows the expression levels and the percentages of gene expression in each cluster of the canonical genes in the normal and new classes of PTCs. (B–D) The volcano plots show the significant genes in differentially expressed gene analysis of the three New-PT clusters by comparing the RNA profiles of one cluster with all other clusters of the dataset. Significance was defined as a gene with an adjusted P value =0.05, a ≥0.25 average log fold difference between the two groups of cells, and the presence detected in at least 10% of cells in either of the two populations.

Identification of New PTC Clusters

Four additional clusters were located in close proximity to clusters S1, S1–S2, S2, S2-cortical S3, and medullary S3 on UMAP. First is the largest component of the distributed proliferative cluster. This part of the proliferative cluster, labeled proliferative PTCs, expressed proximal tubular markers and proliferative marker genes, including Ki67 and Cdca3, that Park et al.⁶ highlighted as identifying a novel cell type in normal mouse kidney and that Wu et al.²¹ identified as proliferative PT in the unilateral ureter obstruction model of renal fibrosis. The other three clusters showed strong signals for genes upregulated in tubules following kidney injury in previous bulk sequencing and other historical approaches (Vcam1, Havcr1, and Akap12), demonstrated expression of canonical PTC genes, and were labeled as New-PT clusters. Clusters New-PT1 and New-PT2 increased in abundance in kidneys undergoing fibrosis following injury, in which circumstances the “New-PT3” was also found. The discrete gene expression signatures of New-PT1–3 clusters included genes associated with renal injury response and fibrosis progression (Figure 3A, Supplemental Figure 5A). DGE analysis comparing each individual cluster with all other clusters of the whole dataset detected 786 differentially expressed genes in New-PT1, 637 in New-PT2, and 1318 in New-PT3 (Supplemental Table 3). Analysis of adjusted P value and average log fold change of differentially expressed genes in New-PT1–3 clusters demonstrated that each had a discrete identifying signature (Figure 3, B–D).

Characteristics of the New Classes of PTCs

Trajectory analysis was performed to infer paths of cell-state transitions within the PTC clusters. Trajectory analysis showed a continuous change in RNA profile in normal PTCs, ordered along the anatomic axis from S1 to S3 tubular segments (Figure 4A). RNA velocity analysis was then used, in which the RNA processing activity evident in the transcriptome of each cell is used to evaluate transcriptional reprogramming and to predict the future state of cells. Represented graphically by an arrow, the velocity of each cell indicates its rate and direction of transcriptional change. RNA velocity analysis showed that the New PT1–3 and the proliferative PTCs exhibited strong directional change toward other states, whereas the normal PT clusters exhibited stable transcriptional profiles, concordant with mature cell phenotypes (Figure 4B). The data further indicated that New-PT1 is an intermediate cell type, from which cells may differentiate in two directions, “New-PT1–New-PT2–proliferative PT-normal PTCs” and “New-PT1–New-PT3” (Figure 4B). Pseudotemporal ordering was also performed (Figure 4C), with time 0 set at New-PT1 on the basis of the results of the RNA velocity analysis. An ordered progression of cell states in pseudotime was seen, from New-PT states through anatomically distinct tubular segments (Figure 4C).

Figure 4.

Trajectory, RNA velocity, and pseudotime analysis of the PTCs. (A) Trajectory analysis shows the dynamic process of the PTCs. RNA expression profiles change continuously along the anatomic axis in normal PTCs (PT-S1 to PT-S3). (B) RNA velocity analysis shows that New-PT1 is an intermediate cell type that may differentiate in two directions: “NewPT1–New-PT2–proliferative PT-normal PTCs and “New-PT1–New-PT3.” (C) PTC pseudotime analysis.

Genetic profiles suggested that the cells comprising these new PTC classes were dedifferentiated. Expression of Kim-1, Vcam1, CD44, Anax3, Akap12, and Ncam1 has been reported in dedifferentiated PTCs in the literature.^21,44,45 Current evidence suggests that, following kidney injury, surviving differentiated PTCs can transform to dedifferentiated PTCs and then undergo proliferation and redifferentiation to restore normal proximal tubular morphology and function.³² We identified clusters labeled New-PT1 and New-PT2 as well as proliferative PTCs in normal kidney, suggesting that this dedifferentiation-proliferation-differentiation process occurs in normal circumstances.

Trajectory and RNA velocity analyses suggested that the New-PT1 represented an intermediate cell type bridging normal PTCs, New-PT2, New-PT3, and proliferative-PT. New-PT1 expressed injury markers, including Vcam1, Pdgfb, Pdgfd, Bmp6, Il34, Itgb6, and Itgav. New-PT1 also partially preserved typical PTC markers, Slc4a4 and Slc5a10. Some cells in New-PT1 also expressed Havcr1; this subset of Havcr1-positive New-PT1 cells was only identified in AAN mice (Supplemental Figure 5B). The New-PT1 cluster was labeled as “dedifferentiated intermediate PTCs.” New-PT2 cells expressed genes characteristic of the developing kidney and reactivated during tubular regeneration, including Ncam1, Tnc, Tgfbr3, Foxd1, and Wt1, and were labeled as “dedifferentiated regenerating PTCs.” New-PT3 cells were detected only in AAN mice, indicating a PTC phenotype specific to progressive kidney fibrosis. Havcr1, also known as Kim1, was a prominent marker of New-PT3, which also had prominent expression of Chd2. Apart from genes related to kidney injury and fibrosis, this cluster expressed Ckd6 and Cdkn1a (also known as p21), associated with cell cycle and also with cell death. The senescence-associated secretory phenotype (SASP) pertains to senescent cells with cell cycle arrest that remain metabolically active and release senescence-associated proteins.^46,47 SASP-related genes, including Cdkn1a (P21), Cdkn2b (P15), Tp53, Tgfb1, Serpine1 (Pai1), Ccl2 (Mcp1), Cxc1, and Ccn2, were enriched in New-PT3 (Supplemental Figure 7). This cluster was identified as “dedifferentiated senescent PTC.”

Microscopic validation for New-PT markers was performed (Figure 5). Occasional PTCs exhibiting VCAM-1– and SLC4A4-positive staining were identified as New-PT1 and were present in normal and AAN kidneys (Figure 5, A and B). HAVCR1 and P21 were selected as New-PT3 markers. Consistent with the snRNAseq dataset, these were not detected in normal kidney but were seen in a proportion of PTCs in AAN kidneys (Figure 5, C and D), in keeping with the presence of the New-PT3 cluster only following injury. Microscopic evaluation of genes selected as New-PT2 markers identified not only PTC but also, glomerular parietal epithelial cell staining. Therefore, an extended set of markers, comprising FOXD1, AKAP12, WT1, TNC, and NCAM1, was used to validate the New PT-2 cluster. Tubular staining for each marker was evident, with coexpression in rare tubular cells discernible through merged signal in normal mouse kidney and in increased numbers of cells in kidneys from the AAN group (Figure 5, E, G, I, and K). Coexpression of markers was also evident through merged signal in parietal epithelial cells, also consistent with the presence of parietal epithelial cells in the New-PT2 cluster (Figure 5, F, H, J, and L). As expected from its known expression pattern and consistent with the snRNAseq data, WT1 was also detected in podocytes (Figure 5, J and L). Similarly, TNC staining pattern was consistent with its presence in mesangial cells, in addition to its strong interstitial staining in the fibrotic kidney (Supplemental Figure 8).

Figure 5.

Immunofluorescence staining of the New-PT clusters. Costaining of sections from naïve and AAN kidneys for (A and B) SLCA4 and VCAM1, (C and D) HAVCR1 and P21, (E–H) Foxd1 and Akap12, and (I–L) Foxd1 and WT1. (F, H, J, and L) Foxd1, Akap12, and WT1 were detected in glomeruli as well as the tubulointerstitium. Arrowheads indicate cells exhibiting dual positivity (merged signal). Scale bars = 20 μm.

For further validation, we performed combined analysis of snRNAseq results using this AAN-induced CKD dataset and the murine ischemia-reperfusion injury–induced AKI dataset published by Kirita et al.,¹⁶ Results of cell clustering and cell-type identification were comparable (Supplemental Figure 9A), with significantly fewer injured and severely injured PTCs in CKD (Supplemental Figure 9B). UMAP of PTCs also addresses the relationship of new PT clusters identified in the two studies. The New PT clusters identified in this paper cluster in the combined analysis with the failed repair PTs identified in the dataset of Kirita et al.,¹⁶ whereas the proliferative PTs cluster with the repairing PTs. The severely injured PT cluster identified by Kirita et al.¹⁶ in their ischemic AKI model was not found in our dataset.

Pathway Analysis of the New Classes of PTCs

Pathway enrichment analysis was carried out for a further understanding of the molecular interaction network of New-PT1–3. Functional and signaling pathway analyses in KEGG showed a general reduction of metabolic pathways among the three New-PT clusters (Figure 6) (https://www.kegg.jp/kegg/). Environmental information processing was enriched, including signal transduction (KEGG category 3.2) and signaling molecules and interaction (KEGG category 3.3) pathways, except for the AMPK signal pathway, which acts as a sensor of cellular energy status. Consistent with categorization as “dedifferentiated senescent PTCs,” New-PT3 showed enrichment of cell growth and death (KEGG category 4.2), including cell cycle, apoptosis, p53 signaling pathway, and cellular senescence. The cellular community (KEGG category 4.3) and cell motility (KEGG category 4.3) pathways were highly evident in the three new classes of PTCs, corresponding to their dedifferentiated phenotype. Immune system activation pathways were also enriched in each New-PT cluster. A table summarizing the pathway analysis results on the basis of KEGG, Panther, Reactome, and WikiPathways databases is provided as Supplemental Table 4.

Figure 6.

Pathway analysis using the KEGG databas identifies principal moelcular networks activated in each PT cluster. Results shows metabolic reprogramming, enrichment of cellular communication and cell motility, and various immune activations in New-PT clusters. The enrichment was calculated by comparing the RNA profiles of one New-PT cluster with all other PTC clusters. Pathways with a false discovery rate <0.25 are listed.

Intercellular Crosstalk between the New Classes of PTCs and Other Cell Types

Ligands and receptors detected in clusters under analysis with a positive average fold enrichment were paired to elucidate intercellular signal transduction networks. For the purposes of this analysis, PTs S1–S3 were combined with “normal PT,” and fibroblast-1 and -2 were combined with “fibroblast” (Figure 7A). We summarized ligand-receptor pairs for ligands from the New-PT clusters and receptors from fibroblast, immune cell, and normal PT clusters (Figure 7, B–D, Supplemental Figure 10, Supplemental Table 5). The New-PT clusters expressed multiple profibrotic signals to fibroblasts, and analysis of the interaction of the New-PTs with immune cells showed several ligand-receptor pairs associated with macrophage activation. New-PTs also expressed signals to EGF receptors, expressed abundantly by normal PTCs. In comparison, normal PTs demonstrated low expression of ligands when compared with New-PTs. The strongest ligand signal from normal PTs was for netrin, a secreted laminin-related protein that is linked to suppression of inflammatory and injury signals in the kidney. The expression levels of ligands and receptors in each cluster are shown in Supplemental Figure 11.

Figure 7.

Ligand-receptor analysis identifies ligands expressed in PT clusters twinned with cognate receptors expressed in other cell clusters. (A) PTs S1–S3 are combined and labelled as "normal PT", and fibroblast-1 and -2 are combined and labelled as "fibroblast" for ligand-receptor analysis. A heat map shows the crossproduct of ligand gene expression from normal PT, New-PT1, New-PT2, New-PT3, and receptor gene expression from (B) fibroblast, (C) immune cells, and (D) normal PT. ATL, ascending thing limb; CNT, connecting tubule; DCT1/DCT2, distal convoluted tubule 1/2; DTL, descending thin limb; IC-A, intercalated cell, type A; IC-B, intercalated cell, type B; IMCD, inner medullary collecting duct; OMCD, outer inner medullary collecting duct; S1/S2/S3, segment 1/2/3 of proximal tubule.

Discussion

Here, we have characterized the cellular composition of the adult mouse kidney, comparing healthy animals with those recovering from toxic proximal tubule injury caused by AA. Our experimental approach benefitted from refinements introduced by other investigators, demonstrating the benefits of rapidly processing unsorted whole-kidney nuclear preparations prior to transcriptomic profiling. We were thus able to delineate major populations of cells that have proved challenging to resolve in previous studies, including mesangial cells, fibroblasts, and JG cells. Furthermore, we identified proliferating cells of distinct lineages.

PTCs make a predominant contribution to the wet weight of normal kidney, are highly metabolically active, and play central roles in kidney recovery versus fibrosis following injury. Existing single-cell analyses from the kidney tissue have grouped PTCs in a single large cluster, often numerically dwarfing all other clusters presented. Here, we have identified five clusters of PTCs abundant in normal kidney tissue, each mapping to proximal tubule segments on the basis of canonical marker expression. These comprised clusters with expression profiles consistent only with a single tubular segment (S1, S2, or medullary S3) as well as those enriched in genes found across neighboring segments (S1–S2 and S2-cortical S3). These data uncover significant complexity in the PTC phenotype and provide an expression map of abundantly expressed genes within the major cell phenotypes at single-cell resolution.

Kidneys from animals treated with recurrent doses of AA to induce renal injury and subsequent fibrosis contained an increased representation of proliferating cells, immune cells, and fibroblasts.⁴⁸ We further identified three new PTC clusters more prominent in kidneys undergoing fibrosis. The first of these, New-PT1, displayed an expression profile intermediate between canonical PTC clusters and clusters New-PT2 and New-PT3, and trajectory analysis further suggested that this cluster may represent PTCs in transition between canonical and these rarer phenotypes. Intriguingly, clusters New-PT2 and New-PT3 demonstrated enriched expression of a panel of genes expressed in proximal tubules following injury. Rather than the diffuse expression in PTCs suggested by prior bulk analyses, however, our data reveal restricted expression of specific markers by cluster. Cluster New-PT2 expressed multiple genes associated with tubular regeneration following injury and was labeled dedifferentiated regenerating PTC on this basis. Within these regeneration-associated genes, Ncam1 is an early nephron progenitor marker that is also seen in proximal tubules after injury, and it may contribute to recovery of PTC function.^45,49 Tnc protects against kidney injury and promotes tubular regeneration.⁵⁰ Tgfbr3 attenuates TGF-β signaling through processes including glycosaminoglycan modifications of the types I and II TGF-β receptors.⁵¹ New-PT2 also demonstrated enriched expression of Foxd1 and Wt1, genes reactivated during tubular regeneration processes. Cluster New-PT3 exhibited unique enrichment for Havcr1, a transcript that is nearly undetectable in normal kidney but occurs promptly after AKI. Havcr1 expression may be upregulated chronically after kidney injury, and its persistent expression leads to renal fibrosis.^52,53 New-PT3 cells also expressed other genes linked to fibrotic responses in tubular cells, namely Cdh2 (N-Cadherin), which is associated with fibroblast growth factor signaling and cell invasiveness, and several genes linked to the SASP, including Cdkn1a (P21), Cdkn2b (P15), Tp53, Tgfb1, Serpine1 (Pai1), Ccl2 (Mcp1), Cxc1, and Ccn2. New-PT3s were accordingly labeled dedifferentiated senescent PTCs.

At microscopy, combined expression of selected genes was confirmed in rare PTCs in normal kidney for New-PT1 and New-PT2 markers and at increased number for New-PT1–3, providing validation of identified expression phenotypes. New-PT2 markers were also identified in glomerular parietal epithelial cells, also consistent with the presence of parietal epithelial cells in the New-PT2 cluster. Shared expression of identifying markers has previously been noted for scattered tubular cells and for a proportion of parietal epithelial cells,⁴⁴ and as a consequence, the potential for a proportion of parietal epithelial cells to represent tubule-committed progenitor cells has been suggested.⁵⁴ Combined analysis with the data of Kirita et al.¹⁶ provided further support for discrete, minority PT phenotypes, most notably postinjury. Alignment of the New-PT3 subset of cells identified in this work with the failed repair PTs identified in the work of Kirita et al.¹⁶ supports their identification of this as a distinct and important phenotype in fibrosis following renal injury. Definitive establishment of the contribution of this and the other identified PTC phenotypes in renal recovery versus fibrosis after injury will be an important future direction for research.

In summary, these data identify principle cellular phenotypes existing in the proximal tubule of the kidney. They further uncover PTC clusters with discrete, fibrosis-associated phenotypes delineated by unique expression profiles of disease-associated markers.

Disclosures

T. Bowen reports research funding from BBI Group, Sun Chemical, and UCB and other interests/relationships as a Kidney Research UK Research Grants Committee member, Chair of the Renal Scientists Committee for UK Renal Association, and a council member (elected renal scientist) for UK Renal Association. P.R. Taylor reports scientific advisor or membership as Editor for Frontiers in Immunology and funding from the UK Dementia Research Institute. All remaining authors have nothing to disclose.

Funding

C.-T. Liao is funded by Kidney Research UK/MedImmune postdoctoral fellowship PDF_006_20151127. P.R. Taylor is funded by Wellcome Trust Investigator award 107964/Z/15/Z.

Data Sharing Statement

RNA sequencing data and the annotated barcode-gene matrices are available from the ArrayExpress database (accession code no. E-MTAB-9390).

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2020081143/-/DCSupplemental.

Supplemental Table 1. Summary of genome mapping results.

Supplemental Table 2. Nuclei clustering results.

Supplemental Table 3. Results of differentially expressed gene analysis.

Supplemental Table 4. Pathway analysis.

Supplemental Table 5. Ligand-receptor interaction.

Supplementary Figure 1. Uniform manifold approximation and projection of the combined datasets.

Supplementary Figure 2. UMAP plots of the combined dataset.

Supplementary Figure 3. Characteristics of the Pfgfbr+ clusters.

Supplementary Figure 4. Cell cycle analysis.

Supplementary Figure 5. Dot plot of PT-specific gene expression.

Supplementary Figure 6. Dot plot of the cluster-enriched gene expression of the nine PT clusters.

Supplementary Figure 7. Expression of senescence-associated genes.

Supplementary Figure 8. Immunofluorescence staining for the New-PT2 markers.

Supplementary Figure 9. Clustering results of combined single-nucleus RNA sequencing datasets of the murine ischemic reperfusion injury–induced AKI published by Kirita et al. and the AAN-induced CKD.

Supplementary Figure 10. Ligand-receptor interaction of the new PT clusters and their adjacent cell types.

Supplementary Figure 11. Expression of ligands and receptors in each cell type.