Single-nuclear transcriptomics reveals diversity of proximal tubule cell states in a dynamic response to acute kidney injury

Significance

A single acute kidney injury event increases the risk of progression to chronic kidney disease (CKD). Combining single-nucleus RNA sequencing with genetic tracing of injured proximal tubule cells identified a spatially dynamic, evolving injury response following ischemia–reperfusion injury. Failed proximal tubule repair leads to the persistence of a profibrotic, proinflammatory Vcam1⁺/Ccl2⁺ cell type exhibiting a senescence-associated secretory phenotype and a marked transcriptional activation of NF-κB and AP-1 pathway signatures, but no signs of G₂/M cell cycle arrest. Insights from this study can inform strategies to improve renal repair and prevent CKD progression.

Abstract

Acute kidney injury (AKI), commonly caused by ischemia, sepsis, or nephrotoxic insult, is associated with increased mortality and a heightened risk of chronic kidney disease (CKD). AKI results in the dysfunction or death of proximal tubule cells (PTCs), triggering a poorly understood autologous cellular repair program. Defective repair associates with a long-term transition to CKD. We performed a mild-to-moderate ischemia–reperfusion injury (IRI) to model injury responses reflective of kidney injury in a variety of clinical settings, including kidney transplant surgery. Single-nucleus RNA sequencing of genetically labeled injured PTCs at 7-d (“early”) and 28-d (“late”) time points post-IRI identified specific gene and pathway activity in the injury–repair transition. In particular, we identified Vcam1⁺/Ccl2⁺ PTCs at a late injury stage distinguished by marked activation of NF-κB–, TNF-, and AP-1–signaling pathways. This population of PTCs showed features of a senescence-associated secretory phenotype but did not exhibit G₂/M cell cycle arrest, distinct from other reports of maladaptive PTCs following kidney injury. Fate-mapping experiments identified spatially and temporally distinct origins for these cells. At the cortico-medullary boundary (CMB), where injury initiates, the majority of Vcam1⁺/Ccl2⁺ PTCs arose from early replicating PTCs. In contrast, in cortical regions, only a subset of Vcam1⁺/Ccl2⁺ PTCs could be traced to early repairing cells, suggesting late-arising sites of secondary PTC injury. Together, these data indicate even moderate IRI is associated with a lasting injury, which spreads from the CMB to cortical regions. Remaining failed-repair PTCs are likely triggers for chronic disease progression.

Acute kidney injury (AKI) is a major global health problem with an annual incidence of ∼13.3 million cases worldwide (1). Major causes of AKI are ischemia–reperfusion injury (IRI) often related to surgical procedures, sepsis, and nephrotoxic drugs (2). AKI is associated with increased in-hospital mortality, longer hospital stays, and higher health care costs (3). Moreover, AKI can contribute to the development of chronic kidney disease (CKD), a condition currently affecting around 9.1% of the world’s population and responsible for 1.2 million annual deaths (4). Analysis of kidney biopsies in kidney transplant patients has identified similar transcriptional and cellular signatures to those observed following severe IRI resulting in chronic disease progression, suggesting longer-term consequences for even moderate kidney injury (5–7). To date, there is no pharmacological strategy to treat AKI or prevent disease progression from AKI to CKD. A comprehensive understanding of the pathophysiologic mechanisms underlying AKI and driving the transition from AKI to CKD is therefore imperative.

Proximal tubule cells (PTCs) are the predominant cell type in the mammalian kidney. The epithelial proximal tubule is responsible for maintaining electrolyte balance and fluid homeostasis. PTCs exhibit high metabolic activity, are especially susceptible to ischemia, and play a critical role in the pathophysiology of CKD (2). In fact, in response to injury surviving PTCs can activate an intrinsic repair process, reenter mitosis, and regenerate the tubular epithelium restoring kidney architecture and function (8–11). However, renal repair is limited; even if renal function returns to baseline after an acute insult, residual inflammatory and fibrotic processes in the kidney can contribute to the development of CKD (2, 9, 12). Studies in animal models have linked disease progression to PTCs in G₂/M cell cycle arrest, which secrete proinflammatory and profibrotic cytokines (12, 13), but the underlying mechanisms driving the transition from AKI to CKD remain to be delineated.

Single-cell and single-nucleus mRNA sequencing (snRNA-seq) provides an unbiased approach to characterize transcriptional profiles associated with kidney injury and repair. Here, we employed genetic labeling strategies focused on PTCs to specifically study injured PTCs in a moderate IRI model using snRNA-seq. Lineage tracing experiments highlight a cell type with a proinflammatory, profibrotic, senescence-associated secretory phenotype, which originates from PTCs failing to repair at the primary injury site, and at cortical injury sites, likely arising from a spread of injury following the primary IRI event. These data highlight a dynamic injury response identifying cell types, cell pathways, and molecular targets for therapeutic evaluation in the transition of AKI to CKD.

Results and Discussion

In a previous study of the progression from AKI to CKD in a long-term bilateral IRI model, we identified keratin-20 (Krt20) as a marker of injured PTCs (5). To enable the isolation and tracing of Krt20⁺ cells in renal injury, we generated a Krt20^T2A-CE mouse line through homology-directed repair (HDR) following CRISPR-Cas9-cleavage of Krt20 in mouse embryo stem cells and the production of germline chimeras (Fig. 1A). HDR results in the production of a CRE-ERT2 (CE) fusion protein in Krt20-expressing cells through the insertion of a T2A self-cleaving peptide at the carboxyl terminus of Krt20. Crossing Krt20^T2A-CE to Gt(ROSA)26Sor^{tm5(CAG-Sun1/sfGFP)Nat} mice, which express green fluorescent protein (GFP) linked to a nuclear membrane protein in the presence of Cre and are further referred to as INTACT (Isolation of Nuclei TAgged in specific Cell Types) mice (14, 15), enabled tamoxifen-dependent labeling of nuclei in injured PTCs.

Fig. 1.

Krt20 inducible Cre-LoxP system traces injured PTCs in post-IRI kidneys. (A) Schematic diagram of CRE-ERT2 (CE) knocked into Krt20 genomic locus through CRISPR. (B) Experimental setup. (C) Colocalization of GFP reporter with Krt20 and Havcr1 on IRI samples. (Scale bar, 20 μm.) (D) Quantification of GFP⁺ cells per region of interest (ROI) using five ROIs per kidney section (n = 2). (E) Quantification of GFP⁺ cells per ROI in cortex and cortico-medullary boundary (CMB) using 10 cortical and CMB ROIs per kidney section (n = 2). (F) Quantification of colocalization of GFP reporter, Krt20, and Havcr1 per ROI, 10 ROIs per kidney section (n = 2). (G) UMAP plot of the integrated IRI and control single-nuclei RNA-sequencing datasets. (H) Violin plots of marker genes arranged by cell type; cluster numbers are indicated above the plots. In D–F, the gray box corresponds to the middle 50th percentile, the horizontal line to the median, and the whiskers indicate the 1.5 interquartile range. ***P < 0.001.

To model moderate kidney injury as may occur in a variety of non–life-threatening clinical settings such as in the course of kidney transplant surgery, we assayed different ischemia intervals. In double transgenic male mice, an 18-min ischemia interval resulted in a small but significant increase (twofold to threefold) in basal creatinine and blood urea nitrogen (BUN) levels 48 h post-IRI (SI Appendix, Fig. S1 A and B). Creatinine and BUN levels were restored to baseline by 28 d post-IRI. Double transgenic mice subjected to IRI were injected with tamoxifen on days 5 and 6, or days 26 and 27 post-IRI, and kidneys collected at day 7 (designated “early”; n = 2) or day 28 (designated “late”; n = 3) post-IRI, respectively (Fig. 1B). For snRNA-seq, GFP⁺ nuclei were isolated using fluorescence-activated cell sorting (FACS) and snRNA-seq profiles identified through 10× Genomics profiling and NextGen sequencing. As a control dataset, we combined our published snRNA-seq data of FACS-sorted GFP⁺ nuclei where a GFP linked to the histone protein H2B is present in all nephron cells (16), with additional replicate data from this line (one nonsurgery and three sham-operated control mice).

Immunofluorescence studies confirmed successful labeling of Krt20-expressing cells with a labeling efficiency of around 34% (Fig. 1 C and F). The labeling efficiency was consistent in comparisons of data at day 7 and day 28 post-IRI. Most labeled cells were positive for the proximal tubule injury marker Havcr1 (Fig. 1F and SI Appendix, Fig. S1E) (17). While the total number of labeled Krt20-expressing PTCs decreased from day 7 to day 28 post-IRI, the predominant location of these cells shifted from the cortico-medullary boundary (CMB) (zone highlighted in SI Appendix, Fig. S1C) in early injured samples to the outer cortex at 28 d (Fig. 1 D and E and SI Appendix, Fig. S1 D and F).

IRI and control snRNA-seq datasets were integrated using the Seurat R package (18), low-quality profiles filtered from the main dataset, and 24,086 nuclear profiles analyzed by unsupervised clustering. The resulting 13 distinct clusters are presented as two-dimensional uniform manifold approximation and projection (UMAP) plot (Fig. 1G). Differential gene expression analysis identified cluster-specific marker genes and the clusters were annotated using known marker genes of kidney cell types, immune cells, vasculature, and interstitial cells (Fig. 1H and SI Appendix, Fig. S2B). As expected, our datasets strongly enriched for proximal tubule nuclei in both the control and IRI treatments. Small clusters of non-PTC signatures in the control and IRI datasets most likely represent a relatively lenient selection of GFP⁺ cells during FACS sorting. Using known marker genes, we unambiguously identified three dominant clusters as proximal tubule segments. Two other closely related clusters showed strong expression of the proximal tubule-specific injury marker Havcr1 and were identified as injured PTCs (17). One of these injured proximal tubule clusters was distinguished by enriched expression of cell cycle genes, indicating a cycling state. The injured PTC clusters as well as the immune cell cluster were almost exclusively derived from the IRI replicates (SI Appendix, Fig. S2A).

For in-depth characterization of different PTC states, we performed a reclustering analysis of all PTCs (marked in Fig. 1G), which resolved 12 clusters. Data are presented as UMAP plot and tabulated lists of the most differentially expressed genes (Fig. 2A and Dataset S1). Both control and IRI replicates were represented together in 7 of the 12 clusters (0, 2, 3, 6, 7, 8, 9), while the other 5 clusters were mostly composed of IRI replicates, further referred to as IRI clusters (1, 4, 5, 10, 11) (Fig. 2C). Clustering of variable genes into modules of coregulated genes showed a clear distinction in the transcriptional profile between these two groups of clusters (SI Appendix, Fig. S3). Clusters composed of both control and IRI replicates showed higher expression of Hnf4a, a key transcription factor governing proximal tubule differentiation, and its target gene Lrp2, than clusters predominantly derived from IRI samples (Fig. 2B) (19, 20). This is in line with the previously described down-regulation of the normal proximal tubule transcriptional signature upon injury (5, 21, 22). Clusters expressing Lrp2 were further subdivided into distinct proximal tubule segments through segment-specific markers such as Slc5a12 (S1), Slco1a1 (S2), and Slc7a13 (S3) (Fig. 2B).

Fig. 2.

PTCs adopt transcriptionally diverse cell states in response to injury. (A) UMAP plot of the PTC clustering. (B) Dot plot of cluster enriched gene expression. Violin plot of male-specific S3 segment marker Cyp7b1 and female-specific S3 segment marker Slc7a12 across clusters split in IRI and control cells. (C) Stacked bar plot of the composition of clusters by group. (D) GO analysis of biological processes and cellular components (F) across clusters. Top three enriched GO terms per cluster are shown. (E) Gene-regulatory network analysis using single-cell regulatory network inference and clustering (SCENIC) across IRI cells. Heatmap showing regulon activity across clusters. Transcription factors representative for each group are highlighted. See Dataset S3 for a complete list of identified regulons.

Interestingly, several clusters displayed normal S1, S2, and S3 signatures, but enrichment of different proximal tubule marker genes indicating heterogeneity among the PTCs. The S3 cluster 8, for example, showed ectopic activation of Slc7a12, a female-specific S3-marker gene, in IRI replicates (23), while also showing expected expression of characteristic male S3 segment-specific genes like Cyp7b1 (Fig. 2B). The predicted up-regulation of Slc7a12 in male PTCs after IRI was confirmed by RNAscope in situ hybridization studies (Fig. 3A). Some Slc7a12⁺ PTCs were positive for Slc22a7, a marker of both female and male proximal tubule segment S3, while others showed coexpression of the vascular cell adhesion molecule-1 (Vcam1), a known marker of inflammatory renal diseases and AKI (21, 24). These data indicate Slc7a12 expression is up-regulated in injured Vcam1⁺, Slc22a7⁻ PTCs, and suggest that Slc7a12 expression may persist in PTCs that recovered, regaining Slc22a7 and losing Vcam1. Alternatively, Slc7a12 may also be ectopically activated in Slc22a7⁺ cells without a strong injury signature. Analysis of a published snRNA-seq dataset of male kidneys 14 d after unilateral ureteral obstruction through a searchable database (humphreyslab.com/SingleCell/) also showed enriched Slc7a12 expression in PTCs (25). Thus, the activation of a female-like Slc7a12 expression pattern in the male kidney on AKI is not confined to the IRI model.

Fig. 3.

Validation of identified injured PTC states. (A) A combination of RNAscope (Slc22a7 and Slc7a12) and immunofluorescence (Vcam1 and Aqp1) in the same sections shows up-regulation of female-specific gene Slc7a12 in male mice after IRI. (B–D) Immunofluorescence staining of Cdh6, Cdh13, Ki67, Vcam1, Hnf4a, and Lrp2 in proximal tubules validates identified PTC states. (Scale bar: A–D, 20 μm.)

Gene ontology (GO) term analysis of the differentially expressed genes in each proximal tubule cluster showed enrichment of terms related to normal proximal tubule structure and function in the S1 and S2/3 clusters across all GO categories. For example, in the category biological process, the terms “organic anion transport” and “drug transport” were enriched, while in the category cellular component, the terms “brush border membrane” and “apical part of cell” were overrepresented (Fig. 2 D and F and Dataset S2). Interestingly, the S1 cluster 7 and the S2/3 cluster 3 showed enrichment of “respiratory chain,” “mitochondrial inner membrane,” and “cytochrome c oxidase activity” (Dataset S2), possibly indicating increased energy demand in these cells. Spag5, encoding a mitotic spindle protein shown to be up-regulated in response to oxidative stress in cancer (26) and in response to high glucose treatment in human podocytes (27), was overrepresented in these clusters as well as in cluster 8 (Fig. 2B).

Among the IRI clusters we identified two “early injured,” two “early + late injured,” and one “late injured” cluster based on the predominant composition of clusters by IRI replicates (Fig. 2 B and C). The early injured clusters showed enriched expression of Havcr1 and the cadherin encoding genes Cdh6 and Cdh13. In the normal kidney, Cdh6 and Cdh13 demarcate cells making up the descending thin limb of the inner stripe of the outer medulla for cortical and juxtamedullary nephrons, respectively (https://cello.shinyapps.io/kidneycellexplorer/) (23). Immunofluorescence studies confirmed the de novo up-regulation of Cdh6 and Cdh13 in injured PTCs (Fig. 3 B and C). The early injured cluster 10 was distinguished by high expression of genes involved in DNA replication such as Mki67 and Top2a, indicative of a cycling cell state (Figs. 2B and 3D). The restriction of a strong replicative response to early injury clusters is in line with the injury-invoked replicative repair.

Early and late injured clusters 1 and 4 were characterized by enriched expression of Cald1 and Pdgfd, while Kcnip4 was markedly up-regulated in cluster 4 specifically. Cald1 encodes Caldesmon 1, an actin-linked regulatory protein of the smooth muscle and nonmuscle contractile apparatus that is expressed in fibroblasts in the healthy kidney and has been associated with diabetic nephropathy (28, 29). The platelet-derived growth factor D (Pdgfd) is known to play a role in renal fibrosis (30), and Kcnip4, encoding potassium channel-interacting protein 4, has recently been described to be up-regulated in a distinct proinflammatory, profibrotic PTC state that persists weeks after AKI (21). Consistent with this finding, Kcnip4 was expressed in the late injured cluster 11 although at lower levels than in cluster 4. Cluster 11 was distinguished from other clusters by strong expression of Vcam1 and the monocyte chemoattractant protein-1 encoding gene Ccl2. Ccl2 is tightly linked on mouse chromosome 11 with Ccl7 and the predicted protein coding gene model Gm17268. All three transcripts were among the most specific markers of the cluster 11 late injury cell population (Dataset S1). The presence of Ccl2⁺, Vcam1⁺, Krt20⁺ cells in the injured kidney 4 wk after AKI was validated by immunofluorescence and RNAscope studies (see Fig. 5 A and B and SI Appendix, Fig. S4C).

Fig. 5.

Immunofluorescence and RNAscope validate Nfkb1 expression in Vcam1⁺/Ccl2⁺ PTCs. (A) Immunofluorescence staining validates Nfkb1 and Vcam1 in Krt20⁺ cells. (B) Combination of RNAscope in situ hybridization (Nfkb1, Ccl2, GFP) and immunofluorescence staining (Vcam1) shows Nfkb1, Ccl2, Vcam1, and Krt20-INTACT GFP-coexpressing cells. (C and D) Quantification of (C) colocalization of Nfkb1 with Ccl2 in PTCs and of (D) Ccl2, Vcam1 and Nfkb1 in GFP⁺ PTCs using five regions of interest (ROIs) per kidney section (n = 3). The gray box corresponds to the middle 50th percentile, the horizontal line to the median, and the whiskers indicate the 1.5 interquartile range. (E) Combination of RNAscope in situ hybridization (Il34, Dock10) and immunofluorescence staining (Vcam1, Krt20-TDT) shows Il34 and Dock10 expression in Vcam1⁺, Krt20⁺ cells marked by the fluorescent reporter tandem tomato (TDT) in Krt20^T2A-CE/+;R26R^tdTomato/+ mice. (F) Immunofluorescence staining shows Cd45⁺, Nfkb1⁺ inflammatory cells around the Nfkb1⁺ tubules. See SI Appendix, Fig. S4E for validation of Nfkb1 antibody. (G) Box plots of reads per kilobase per million mapped reads (RPKM) values of indicated genes in bulk RNA-seq analysis of human kidney transplants biopsies in a successfully repaired [1] or a transition state [2] toward a chronically injured state [3]. Biopsy samples from living donors (LDs) were included as a representation of normal kidney tissue. Groups were compared by Mann–Whitney U test. Box corresponds to the middle 50th percentile, the horizontal line to the median, and the whiskers indicate the data range. Data adopted from ref. 7. (Scale bar: A, B, E, F, 20 μm.)

GO term analysis indicated a clear functional distinction between normal S1, S2/S3, and IRI clusters (Fig. 2 D and F and Dataset S2). None of the terms related to normal proximal tubule structure and function were enriched in the IRI clusters, while new terms associated with changes in epithelial state, dedifferentiation, migration, and fibrosis were overrepresented (Fig. 2D). Furthermore, cellular components known to play a role in cell adhesion and migration like “actin cytoskeleton” and “adherens junction” were enriched in the IRI clusters (Fig. 2F and Dataset S2). The early proliferating cluster showed very cluster-specific enrichment of cell cycle related GO terms, underlining the confinement of a cycling cell state to the cells in this cluster (Fig. 2D and Dataset S2).

To obtain further insight into the regulatory states of cells in each cluster, we used single-cell regulatory network inference and clustering (SCENIC) to reconstruct gene-regulatory networks (Fig. 2E and Dataset S3) (31). S1, S2, and S3 showed activity of regulons involved in proximal tubule development, including Hnf4a, Hnf1b, and Esrra (19, 20, 32). Activity of these regulons associated with core PTC function was predicted to be down-regulated in the IRI clusters, which is in line with the marked loss of normal PTC gene expression profiles. Interestingly, two of the PTC clusters annotated as S1 and S3, cluster 7 and cluster 3, showed predicted regulon activity of Hsf1, encoding a master transcriptional regulator of heat shock proteins involved in stress and cancer (33) and Hcfc1 (host cell factor C1), a transcriptional regulator that targets cell cycle genes depending on Hsp90 for its stability (34). In addition, cluster 3 shared activity of some regulons with IRI clusters. Clusters 3 and 7 likely represent PTCs with reduced PTC functions, reflected by reduced Hnf4a activity, and a stressed transcriptional state.

Proliferating early injured cells displayed E2f3 regulon activity consistent with the replicating state of these cells. Gene-regulatory network analysis further suggested involvement of the Hippo signaling pathway in early injured cells: up-regulation of the Tead1 regulon, a nuclear effector in an inactive Hippo signaling state associated with cell proliferation. The Hippo signaling pathway is known to respond to altered tension and mechanical constraints in epithelial tissues (35), regulating organ development and growth through the control of cell proliferation and apoptosis. YAP and TAZ, key transcriptional coactivators acting together with Tead DNA binding partners, have been shown to play a role in cystic kidney disease progression and development of renal fibrosis following AKI (36). Consistent with Hippo engagement in early injured cells, differential gene expression analysis showed up-regulation of regulators involved in YAP-inactivation and Hippo signaling (e.g., Sav1, Stk3), as well as effectors of the transcriptional response when Hippo signaling is silenced (e.g., Yap1, Tead1) (37) and well-known YAP-target genes such as Axl and Ctgf. SCENIC also predicted IRI cluster-specific regulon activity of Nfat5, a transcription factor mediating cellular responses to hypertonic stress and hypoxia that has been described to diminish renal injury following IRI by reducing caspase-3–dependent apoptosis (38). Nfat5 expression was enriched in all IRI clusters (Dataset S1).

Late injured cells of cluster 11 showed strongly up-regulated activity of regulons related to the NF-κB (Nfkb1, Nfkb2, Rela, Relb, Bcl3) and activator protein-1 (AP-1; Jun, Junb, Jund, and Fosl2) signaling pathways. Similar gene-regulatory networks have been described in many human cancers and provide further evidence for a proinflammatory state of these cells (39). In addition, activity of the Smad3 regulon, an intracellular signal transducer activated by transforming growth factor-β (TGF-β), suggests these PTCs may also respond to TGF-β signaling that is linked to driving renal fibrosis in mesangial cells and adjacent stromal cell types (40, 41). IRI clusters also displayed regulon activity of Myc, encoding the proto-oncogene c-Myc, which has been implicated in tubular cell apoptosis, and Tfcp2l1, which is associated with renal tubule development (42, 43).

Given the presence of the late injury cluster 11 in the kidney after normalization of kidney function, and an AKI linkage to development of CKD (9), we focused further analysis on this cluster. The most strongly enriched GO term in cluster 11 was “small GTPase mediated signal transduction,” and dedicator of cytokinesis protein 10 (Dock10), a guanine nucleotide exchange factor that can activate Rho GTPases, was among the most up-regulated genes in cluster 11 (Fig. 4C and SI Appendix, Fig. S4A). Dock10 has been described to be necessary for GTPase activation in amoeboid invasion of melanoma cells (44). Interestingly, GO terms like “ameboidal-type cell migration” and “epithelium migration” were also enriched in late injured cluster 11 cells, suggesting a migratory phenotype of these cells (SI Appendix, Fig. S4A).

Fig. 4.

Vcam1⁺/Ccl2⁺ proximal tubule subpopulation shows proinflammatory and profibrotic signature, but no G₂/M cell cycle arrest. (A) Top 20 enriched KEGG pathways in late injured cluster 11. (B) Depiction of the KEGG pathways TNF and NF-κB signaling with the involved genes up-regulated in cluster 11. (C) Violin plots displaying expression of selected features across proximal tubule clusters. The cluster numbers correspond to Fig. 2A. (D) Gene set enrichment analysis of published cell cycle and (E) senescence/senescence-associated secretory phenotype (SASP) gene sets across clusters. See Dataset S4 for details on used gene sets.

Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis indicated a strong up-regulation of TNF and NF-κB signaling: The NF-κB pathway is the primary mediator of TNF signaling (45) and AGE (advanced glycation end-product)/RAGE (AGE receptor) signaling, which has been linked to diabetic nephropathy (Fig. 4 A and B) (46). Differential gene expression analysis revealed up-regulation of the NF-κB complex (Nfkb1, Relb, Rela) as well as of potential activators/mediators (Tnfα, Il-1r, Cd40, Rank), regulators (Malt1), and target genes of the NF-κB signaling pathway (Fig. 4 B and C). Several of the up-regulated NF-κB target genes have been described to play a role in kidney disease: Il34 has been shown to promote macrophage proliferation in the kidney, thus mediating AKI and ensuing fibrosis (47); knockout of the neutrophil chemoattractant Cxcl1 has been described to ameliorate cisplatin-induced AKI (48) and Icam1 deficiency has been shown to be protective in IRI (49). This underlines the importance of late injured Vcam1⁺/Ccl2⁺ PTCs in kidney disease progression.

A secondary validation of predictions from the snRNA-seq analysis by RNAscope in situ hybridization and immunofluorescence analysis confirmed the coexpression of Ccl2 and Nfkb1 as well as the expression of Dock10 and Il34 in injured Krt20⁺, Vcam1⁺ PTCs (Fig. 5 A–E). Approximately 17% of GFP⁺ cells were positive for Ccl2, Nfkb1, and Vcam1, while no cells expressed all four markers in control kidneys (Fig. 5D). Quantification of the percentage of Ccl2, Nfkb1, and Vcam1-copositive cells among Krt20⁺ cells showed similar results (SI Appendix, Fig. S4 C and D). Furthermore, Nfkb1-expressing PTCs were spatially associated with Cd45⁺ immune cells, consolidating the role of these cells in persistent renal inflammation weeks after AKI (Fig. 5F).

To examine the human relevance of data from the moderate mouse AKI model, we assessed the expression of genes characteristic of Vcam1⁺/Ccl2⁺ cells in a previously published bulk RNA-seq dataset from human kidney transplant biopsies (7). This analysis delineated two distinct transcriptional trajectories in response to human AKI (7): successful repair [1] or disease progression via a transition state [2] to chronic injury [3]. Expression of Vcam1, Ccl2, Nfkb1, and Dock10 were all significantly elevated in kidneys in the chronically injured state (Fig. 5G), further supporting the relevance of Vcam1⁺/Ccl2⁺ PTCs in the AKI-to-CKD transition.

Consistent with a prediction of up-regulated AP-1 regulon activity (Fig. 2E), expression of several AP-1 family members was increased in cluster 11 cells (Dataset S1). AP-1 also acts as a proinflammatory effector of both the TNF and the AGE/RAGE signaling pathways and has been implicated in development of organ fibrosis (50, 51). In addition, expression of genes linked to TGF-β signaling and kidney fibrosis like Tgfb2, Tgfbr2, Smad1-3, Col4a1, and Col4a2 was enriched in late injured cells. Intriguingly, prosurvival (Birc2, Birc3, Traf1/2) as well as proapoptotic (Fas, Casp4, Apaf1, Pmaip1) target genes of both NF-κB and p53-mediated signaling were up-regulated in late injured cluster 11 cells (Fig. 4C and Dataset S1). Thus, proapoptotic gene activity may be countered by prosurvival networks maintaining dysfunctional epithelial cells within the nephron weeks after AKI.

Previous studies of long-term injury responses following IRI and other forms of AKI have pointed to PTCs arrested in the G₂/M phase of the cell cycle as a trigger of post-AKI fibrosis (12, 13). To examine the cell cycle state of PTCs in this study, we performed gene set enrichment analysis using diagnostic cell cycle gene sets (52, 53). Enrichment of all gene sets indicated the presence of cycling cells in the cycling early injured cluster. However, no enrichment of G₂/M phase gene sets could be detected in late injured cells (Fig. 4D). Cell cycle analysis using the CellCycleScoring function in Seurat to assign a S-phase or a G₂/M-phase score to every single cell in the dataset yielded the same result (SI Appendix, Fig. S4B). In accordance, immunofluorescence staining with phosphorylated histone 3 at serine 10 (p-H3), a G₂/M-phase marker (13), showed that <0.5% of all Vcam1⁺ cells on day 28 post-IRI expressed p-H3 (SI Appendix, Fig. S4 F and G). These cells were predominantly located in the cortex (SI Appendix, Fig. S4H). The Vcam1⁺/Ccl2⁺ cells in the late injured cluster 11, however, comprised 6% of all Vcam1⁺ cells in the IRI_d28 samples of the snRNA-seq dataset. Together, these data argue against a G₂/M phase arrested cell state in the majority of Vcam1⁺/Ccl2⁺ injured PTCs in this moderate injury model.

Analyses of age-related organ deterioration and tissue fibrosis have suggested that chronically senescent cells with a shared senescence-associated secretory phenotype (SASP) are involved in the development of tissue inflammation and fibrosis in many organ systems (54–57). Since the phenotype of a senescent cell is dynamic and heterogeneous depending on the type of senescence inducer, and on the cell and tissue type, we compared our data with transcriptomic signatures characteristic of the aging kidney and senescent cells and/or SASP, in various cell types subjected to different methods of senescence induction (Fig. 4E) (58–64). The “senescence core signatures” identified by Hernandez-Segura et al. (62) in fibroblasts, keratinocytes, melanocytes and astrocytes and by Casella et al. (63) in fibroblasts and endothelial cells were not enriched in late injured cells. Furthermore, the top 20 genes up-regulated in aging mouse kidneys by single-cell RNA-seq showed no enrichment in any of the clusters (58). However, a list of genes with a concordant increase in mRNA and protein level in the aging kidney (59) showed significant enrichment in three out of five IRI clusters, including cluster 11. Likewise, genes encoding proteins recently shown to be increasingly secreted by renal cortical epithelial cells upon irradiation induced senescence (64) and other SASP gene sets were significantly overrepresented in cluster 11 late injured cells.

To determine whether late injured cells arise from cells initiating replication in response to the primary IRI, we used a Ki67^Cre/ERT2 mouse line (65) crossed to INTACT mice and induced labeling with tamoxifen injection either on days 2 and 3 or on days 5 and 6 post-IRI. Following GFP labeling of replicating cells on days 2 and 3 post-IRI to the kidney at day 28 showed that 89% of Vcam1⁺/Ccl2⁺ cells at the CMB were GFP⁺ (Fig. 6 A and B). In contrast, only 27% of Vcam1⁺/Ccl2⁺ cells in the cortex were GFP⁺ at the day 28 time point. These observations lead to several conclusions. Cells with cluster 11 characteristics are evident at the site of primary injury in the hypersensitive S3 proximal tubule segments. Cells here initiated injury-invoked replication but failed to regenerate a normal PTC phenotype. In addition, cells with an indistinguishable molecular profile in cortical S1 and S2 segments reflect injury sites not associated with early injury-invoked proliferative repair. Furthermore, no cortical increase was observed in replicating cells among the Vcam1⁺/Ccl2⁺ population at 28 d when replicating cells were labeled at 5 and 6 d post-IRI (Fig. 6B). Thus, most cortical Vcam1⁺/Ccl2⁺ cells either originate from cells that were injured during the initial IRI, but did not initiate replication or show obvious injury responses at this time, or from a secondary spread of the injury within the CMB to the cortex. The observed spatial shift of the predominant location of Krt20-INTACT GFP positive cells from the CMB on day 7 to the cortex on day 28 post-IRI (Fig. 1E and SI Appendix, Fig. S1 D and F) supports the latter hypothesis.

Fig. 6.

The temporal and spatial relationship of PTC injury signatures in the mouse kidney. Lineage tracing of proliferating cells using a Ki67^Cre/ERT2;INTACT mouse line with tamoxifen (Tam.) injections either on day 2 and 3 or on day 5 and 6 post-IRI. (A) RNAscope in situ hybridization (Ccl2, Krt20, GFP) combined with immunofluorescence staining (Vcam1) shows colocalization of Ki67-INTACT GFP-positive cells with Vcam1, Krt20, and Ccl2 on day 28 post-IRI when tamoxifen is injected on day 2 and 3 post-IRI. (B) Quantification of GFP reporter in Ccl2⁺/Vcam1⁺ cells in the cortex and the cortico-medullary boundary (CMB) at day 28 post-IRI in mice injected with tamoxifen either on day 2 and 3 (n = 4) or on day 5 and 6 (n = 3) post-IRI. (C) Immunofluorescence staining of proximal tubules with Krt20, Vcam1, Hnf4a, and LTL and quantification (D) of Vcam1⁺ PTCs in uninjured kidneys of 3- and 18-mo-old C57BL/6N mice. The gray box corresponds to the middle 50th percentile, the horizontal line to the median, and the whiskers indicate the 1.5 interquartile range. *P < 0.05. (Scale bar: A and C, 20 μm.)

The extent of renal injury and repair in the absence of an extrinsic trigger of AKI is an open question. Interestingly, a small number of control cells coclustered with injured PTCs (Fig. 2C and SI Appendix, Fig. S2A). Consistent with this finding, Krt20^T2A-CE labeling identified rare Krt20⁺/Vcam1⁺ PTCs in healthy mouse kidneys (age: 3 mo; SI Appendix, Fig. S5). Some of the identified Vcam1⁺ PTCs were Relb⁺, and transcribed Il34 and Dock10, indicating activation of the NF-κB signaling pathway in these cells, as in cell populations enriched in IRI (SI Appendix, Fig. S5). Rare Vcam1⁺ and Vcam1⁺/Havcr1⁺ PTCs have also been identified in normal adult human kidney samples (24, 66). These data suggest rare sporadic injury responses in the mouse and human kidney may mirror those initiated broadly in PTCs following IRI. These rare events are likely to increase with age. Consistent with this prediction, when Vcam1⁺ PTCs were scored in healthy 3- and 18-mo-old C57BL/6N mice, a marked increase of Vcam1⁺ PTCs was observed with age (Fig. 6 C and D).

Conclusions

PTCs play a major role in the regeneration of kidney structure and function following AKI (9). Our study provides high-resolution profiling of diverse injured PTC states, highlighting a role for NF-κB and AP-1 pathway activity in PTCs, and suggesting these regulatory pathways may drive PTCs to a pathological state. Of particular significance, we identified a Vcam1⁺/Ccl2⁺, proinflammatory, profibrotic PTC population, with a SASP that is shared with other organ systems (60–62).

Lineage tracing experiments suggest Ccl2⁺ PTCs originate from failed repair of cells that initiated replication in response to the initial IRI insult, as hypothesized in a recent study (21), as well as from secondary injury sites within the cortex. Limited replicative labeling studies suggest secondary injury may not associate with cell death induced replicative repair, although additional studies are required for a definitive insight. Intriguingly, Ccl2 has been shown to induce activation of NF-κB and AP-1 signaling as well as IL6 and ICAM1 expression in PTCs in vitro (67), and senescent cells are known to induce senescence in neighboring cells by paracrine signaling (68). Thus, autocrine and paracrine signaling by Ccl2⁺ PTCs may contribute to the persistence of Ccl2⁺ PTCs weeks after injury.

Previous studies have described G₂/M phase arrest as a characteristic of maladaptive PTCs after AKI (13). However, in our moderate injury model, neither snRNA-seq analysis nor immunolocalization of p-H3⁺ cells identified G₂/M arrest in direct injury-invoked or secondary injury-associated Vcam1⁺ cells. In fate-mapping experiments, only a minor fraction of cortical Vcam1⁺/Ccl2⁺ cells showed a history of injury-invoked replication, which further supported the absence of G₂/M arrest in the majority of these cells. Additionally, a recent snRNA-seq study in a more severe IRI model (21) showed similar qualitative response features to the injury responses reported here. Thus, G₂/M arrest may not be a cell-intrinsic feature of proinflammatory PTCs remaining after kidney function is restored following AKI.

Senescent cells have emerged as important players in age-related kidney fibrosis and various kidney diseases (54, 68–70), and NF-κB signaling has been identified as a SASP activator (71). Our data demonstrate that Vcam1⁺/Ccl2⁺ PTCs share features with senescent cells including expression of Plaur encoding the urokinase-type plasminogen activator receptor (uPar). The improved outcome from successful removal of uPar⁺ cells through chimeric antigen receptor (CAR) T cell therapy in a liver fibrosis model (57), suggests a possible experimental route to blocking CKD progression following AKI.

Interestingly, Vcam1⁺/Ccl2⁺ cells exhibit an unusual mixed proapoptotic and antiapoptotic phenotype. As observed with cancer cells, despite the abnormal organization and cellular profile, antiapoptotic activities may promote survival of maladaptive PTCs that fail to complete renal repair. Targeting cell-intrinsic antiapoptotic pathways could be an alternative strategy to eliminate Vcam1⁺/Ccl2⁺ cells and determine whether their removal prevents progression to CKD.

Materials and Methods

See SI Appendix for a detailed description of materials and methods.

Mice, Surgical Procedures, and Serum Analysis.

Mouse husbandry, handling, and surgical procedures were performed according to Institutional Animal Care and Use Committee guidelines at the University of Southern California (protocol numbers 11911 and 20894). IRI experiments were carried out on adult male mice (age, 11–19 wk; weight, 24–30 g) as in ref. 5, with a clamping time of 18 min to achieve a moderate injury level (no mortality until day 28 post-IRI). For sham surgery, the same procedure was performed except for clamping of the renal pedicles. Serum was collected for creatinine and/or BUN analysis at 48 h post-IRI and at collection.

Tissue Collection, Immunofluorescence, RNAscope in Situ Hybridization, and Image Analysis.

Tissues were collected after organ perfusion with ice-cold DPBS (HyClone). Tissue sections were prepared and immunofluorescence staining performed as previously described (5). For RNAscope experiments, frozen sections were postfixed for 1–2 h in 4% PFA at 4 °C and then followed the manufacturer’s protocol (ACD 323100-USM) for fixed-frozen tissues using TSA Plus fluorophores (Life Technologies). Dataset S5 lists used antibodies and probes. Images were acquired by confocal microscopy and quantified in randomly selected regions of interest or in sagittal sections as indicated (n ≥ 2 for all image analyses) using Fiji (72). Datasets were compared by two-sided Student’s t test or two-sided Wilcoxon test depending on their distribution. P < 0.05 was considered significant.

Nuclei Isolation and FACS Sorting.

Nuclei were isolated from flash-frozen tissues with lysis buffer, Dounce homogenization, and filtering as in ref. 25 and resuspended in 1% BSA–DPBS supplemented with RNase inhibitor and Hoechst 33342. Hoechst⁺, GFP⁺ singlets were FACS sorted into a BSA–DPBS RNase inhibitor solution.

Single-Nucleus RNA-Seq.

Nine samples were processed for snRNA-seq: two Krt20^T2A-CE;INTACT kidneys collected on day 7 after IRI, three Krt20^T2A-CE;INTACT kidneys collected on day 28 after IRI, and four control kidneys. FACS-sorted GFP⁺ nuclei were counted and processed according to the 10× Chromium Single Cell 3′ protocol (v2 or v3). 10× libraries were sequenced using an Illumina NextSeq 500/550 High Output Kit v2. Alignment of sequencing reads was performed with the Cell Ranger Single-Cell Software Suite 3.0 (10× Genomics) using STAR aligner (73) on the University of Southern California High Performance Cluster. Filtered gene-cell expression matrices, removing cell barcodes not represented in nuclei, were created with the cellranger count function.

snRNA-Seq Data Analysis.

The R packages Seurat v.3.2.2 (18), ggplot2 v.3.3.2, Matrix v.2.3-18, and dplyr v.1.0.2 were used for preprocessing, data analysis, and visualization in R Studio (R version 3.6.3). Seurat was used to filter out genes expressed in less than three nuclei and nuclei with low-quality profiles. Data were normalized and scaled using SCTransform (74) and integrated using integration anchors as described in ref. 18. The number of useful principal components (PCs) was determined using the ElbowPlot function and graph-based clustering performed. Differential gene expression analysis was performed on individual clusters. The subset function was employed to create a new Seurat-object comprising only PTC clusters, on which PC analysis, reclustering, and differential gene expression analysis were performed.

Gene modules were identified using the R package Monocle3 v.0.2.3.0 (75), GO and KEGG pathway enrichment analysis was performed using ClusterProfiler v.3.14.3 (76) and KEGG pathways visualized with GOplot v.1.0.2 (77). Fisher’s exact test with Benjamini–Hochberg correction for multiple testing was employed for gene set enrichment analysis of published cell cycle and senescence gene sets (Dataset S4). The R package SCENIC v.1.1.2-2 was used to perform gene regulatory network analysis on IRI nuclei across all PTC clusters as previously described (31). The expression matrix generated with the SCTransform function in Seurat was used as input.

Analysis of Published Bulk RNA-Seq Dataset.

Data were adopted from ref. 7. Samples were assigned to groups 1 (successfully repaired), 2 (transition), and 3 (chronically injured) according to the pseudotime analysis performed in ref. 7 (see ref. 7, figures 3C and 4A for pseudotime analysis). Groups were compared by Mann–Whitney U test. Box plots were generated using Prism 9.

Data Availability

The single-nuclei RNA sequencing data have been deposited in the Gene Expression Omnibus (GEO) database (accession no. GSE171417). All study data are included in the article and/or supporting information. Previously published data were used for this work (https://doi.org/10.1172/jci.insight.123151; https://doi.org/10.1038/s42255-020-0238-1).