Clustering and identification of cell types

We performed single-cell transcriptomics on a human fetal kidney from w16 of gestation, equivalent to 14 weeks of development (Fig 1B). After data pruning and stringent removal of cells affected by stress (S2 Fig, Methods), 6,602 cells were retained for further analysis. Clusters of cells were identified by hierarchical clustering after k-nearest neighbor smoothing [12]. We assigned cell types to these clusters by expression of marker genes from the literature on mouse kidney development. The studies that linked the genes of this literature set to particular cell types are referenced in S1 Table. After merging similar clusters (S3 Fig, Methods), we obtained 22 cell types (S4 Fig) and visualized the single-cell transcriptomes in a two-dimensional t-distributed stochastic neighbor embedding (tSNE) map [13] (Fig 1C).

The mean expression levels of the literature set genes showed clear differences between cell types (Fig 2A). NPCs, which were distributed over four distinct clusters (NPCa–d), were marked by the established markers SIX2 (Fig 1C), CITED1, MEOX1, and EYA1. Expression of these progenitor markers was highest in NPCa, which we hence considered “bona fide” self-renewing NPCs. NPCb showed lower levels of CITED1 and SALL1 and higher levels of GDNF and HES1 compared to the other NPC clusters. HES1, a transcription factor downstream of NOTCH signaling, is important for further renal cell differentiation. Compared to the other NPC clusters, NPCc showed higher expression of CRABP2, which is related to retinoic acid signaling [14]. NPCd exhibited low OSR1, CITED1, and MEOX1 expression and increased levels of LEF1, a known indicator of NPC induction towards differentiation. Compared to the other NPC subtypes, NPCd were also marked by a larger fraction of cells in G2/M-phase of the cell cycle (Fig 2B and 2C) and a higher expression of proliferation markers (Fig 2D), which indicated faster proliferation. We will discuss the relationship between the various NPC clusters in more detail below (see Heterogeneity in the nephrogenic niche).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Known markers elucidated the cell types corresponding to each cluster.

(A) Heatmap of literature set gene expression in the 22 identified cell types. Expression was Freeman-Tukey transformed, averaged over all cells in a cluster and standardized gene-wise. (B) tSNE map of all cells with color indicating the G2/M score (calculated by the Cyclone tool [15]). This score reflects the likelihood that a cell is in G2/M phase. (C) G2/M scores from panel B averaged over the cells in each cell type. (D) Mean expression of proliferation markers [16] (Z-scores) per cell type. The numerical data underlying this figure can be found in S1 Data. CnT, connecting tubule; DTLH, distal tubule/loop of Henle; End, endothelial cells; ErPrT, early proximal tubule; G2/M, cell cycle phase G2/M; ICa, interstitial cells a; ICb, interstitial cells b; IPC, interstitial progenitor cells; Leu, leukocytes; Mes, mesangial cells; NPCa, nephron progenitor cells a; NPCb, nephron progenitor cells b; NPCc, nephron progenior cells c; NPCd, nephron progenitor cells d; Pod, podocyte; Prolif, proliferating cells; PTA, pretubular aggregate; RVCSBa, renal vesicle/comma-shaped body a; RVCSBb, renal vesicle/comma-shaped body b; scRNAseq, single cell RNA sequencing; SSBm/d, s-shaped body medial/distal; SSBpod, s-shaped body podocyte precursor cells; SSBpr, s-shaped body proximal precursor cells; tSNE, t-distributed stochastic neighbor embedding; UBCD, ureteric bud/collecting duct.

https://doi.org/10.1371/journal.pbio.3000152.g002

Nephrogenesis continues with the creation of the pre-tubular aggregate (PTA) cells, which in turn develops into the RV and CSB. In our data, PTA cells were identified based on high expression of LHX1 (Fig 1C), JAG1, WNT4, and CCND1. Because RV and CSB are mainly distinguishable by morphology, cells belonging to these two structures were grouped in our analysis (RVCSB). RVCSB cells were marked by the same genes as PTA cells, but they appeared to proliferate less (Fig 2B–2D). Furthermore, they expressed markers reflecting more advanced regional patterning, which allowed us to discriminate between two subtypes (a and b). RVCSBa had a higher expression of genes that were recently associated with the distal RV (SFRP2, DLL1, LHX1), whereas RVCSBb expressed genes that indicate the proximal RV (CDH6, FOXC2, MAFB, CLDN1, WT1).

The next step in development is the formation of the SSB. In our data set, this structure was represented by three clusters, named according to the part of tubule and glomerular epithelium they are known to give rise to—SSBpr, proximal tubule; SSBm/d, medial/distal; SSBpod, podocytes. SSBpr were identified in our data by markers of the early proximal tubule (ErPrT) (such as HNF1A, CDH6, AMN), as well as low expression of SLC3A1, LRP2, and SLC13A1, which are known to be found in more mature proximal tubule cells. Therefore, SSBpr were likely precursors of the ErPrT cells, which expressed higher levels of early proximal markers together with CLDN2 (Fig 1C), ANPEP, and SLC34A1. Another cluster (SSBm/d) accounted for the precursor cells of the loop of Henle and the distal tubule in the SSB. This cluster could be identified by the presence of IRX1, IRX2, SIM2, SOX9, POU3F3, and HNF1B, together with low expression of PAPPA2 and MAL and the absence of CDH6 and HNF1A. Cell types that are known to develop from the SSBm/d were found together in one cluster (distal tubule/loop of Henle [DTLH]). This cluster showed high expression of the distal markers MAL, CLCN5, SLC12A3, and POU3F3, which are specific to the distal tubule, as well as SLC12A1, PAPPA2, and UMOD, which are found in the loop of Henle. Finally, cells that likely gave rise to podocytes, SSBpod, clustered separately. These cells showed high expression of MAFB and FOXC2, both transcription factors necessary for the development of podocyte identity, and low levels of the mature podocytes markers CLIC5 (Fig 1C), PTPRO, NPHS1, and NPHS2. This cluster also showed the highest expression of OLFM3, previously identified as a specific marker of podocyte precursors residing in the visceral part of the proximal segment of the SSB. In contrast to SSBpod, podocytes (Pods) showed higher expression of mature podocyte markers and lower levels of MAFB. Differences between SSBpod and Pods will be studied in more detail below (see Podocyte development). Because of the high similarity in gene expression between SSB and capillary loop stage, we could not exclude that the SSB clusters also contained cells from the capillary loop stage.

Cells of the connecting tubule (CnT), which connects the distal tubule to the collecting duct, could also be identified in the data. They shared markers with the collecting duct (such as ALDH1A1, TACSTD2, and CDH1), distal tubule (SOX9, POU3F3), and UB (RET, KRT8, KRT18, MMP7). Cells of the UB and collecting duct (UBCD) were strongly marked by well-known genes like AQP2 (Fig 1C), CALB1, KRT8, KRT18, RET, and GATA2, found in the collecting duct as well as the stalk and tip of the UB.

The developing nephrons are surrounded by interstitial tissue, a separate lineage that originates in interstitial progenitor cells (IPCs). We identified IPCs by coexpression of FOXD1 and GDNF. These cells also expressed lower levels of markers known to be found in more mature cells like PDGFRA for ICs or PDGFRB (Fig 1C) and ACTA2 for mesangial cells. We identified two subtypes of ICs (a and b), which were similar in their marker gene profile. Compared to IPCs, they lacked FOXD1 and expressed less (ICa) or no (ICb) GDNF. ICa showed high levels of FGF7, which has been localized to the renal fibroblasts or stroma surrounding the ureter and the collecting system. ICa also showed high levels of TPM2 and ACTA2, markers of smooth muscle-like cells. ICb, on the other hand, expressed genes like DCN (Fig 1C), DES, SERPINE2, and COL3A1, which are known to mark cortical stromal cells. Endothelial cells were identified by markers such as KDR and TEK, whereas leukocytes showed many specifically expressed genes, such as CD37 or CD48. Finally, one cluster of cells (Prolif) had a higher expression of proliferation markers compared to most other cell types (Fig 2D) but lacked discernible cell type markers.

Developmental flow

The literature-based analysis of the found clusters seemed to suggest that cells cluster by developmental progression (e.g., NPCs versus PTA cells), as well as location (e.g., RVCSBa, distal, versus RVCSBb, proximal). Because the interpretation of clusters is sometimes based on genes that are expressed in multiple developmental stages, we wanted to retrieve the developmental flow with an independent method. We used Monocle 2 [17] to learn a graph that represents the developmental hierarchy of the cell types from the PTA on (Fig 3A). Subsequently, cells were placed on a pseudotime scale rooted in the PTA (Fig 3B and 3C). This analysis showed that PTA cells were followed by RVCSBa and RVCSBb, the SSB clusters, and finally the clusters identified as more mature types (DTLH, ErPrT, Pod). Therefore, the clustering was strongly driven by developmental progression. RVCSBa cells were distributed over a fairly broad period of pseudotime and already occurred before branch point 1, which separates proximal from distal cell fates (Fig 3A). This might indicate that some of these cells preceded the RVCSBb, whereas others were primed to develop into distal fates. RVCSBb cells, however, only appeared after branch point 1, which confirmed that they were likely progenitors of proximal cell fates. On three separate branches, SSBm/d preceded DTLH, SSBpr preceded ErPrT, and SSBpod preceded Pods, which confirmed the identity of the SSB clusters. The temporal relationship of the NPC subtypes will be discussed in detail below (see Heterogeneity in the nephrogenic niche).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Pseudotime analysis clarified the developmental relationship of the cell clusters.

(A) Two-dimensional embedding (with the DDRTree algorithm [18]) of all w16 kidney cells, calculated by Monocle 2 [17]. The graph learned by the algorithm is shown as a black line. Colors and labels indicate cell types. (B) Same embedding as in panel A. Color indicates pseudotime calculated by Monocle 2. (C) Box plots of cell type distribution over pseudotime. The numerical data underlying this figure can be found in S1 Data. DTLH, distal tubule/loop of Henle; ErPrT, early proximal tubule; Pod, podocyte; PTA, pretubular aggregate; RVCSBa, renal vesicle/comma-shaped body a; RVCSBb, renal vesicle/comma-shaped body b; SSBm/d, s-shaped body medial/distal; SSBpod, s-shaped body podocyte precursor cells; SSBpr, s-shaped body proximal precursor cells; w16, week 16.

https://doi.org/10.1371/journal.pbio.3000152.g003

Comparison with existing single-cell transcriptomics data

To further confirm the interpretation of the cell clusters, we wanted to compare our data with an existing single-cell transcriptomics study of a w17 fetal kidney by Lindström and colleagues [19]. To that end, we first corrected for batch effects, using a method based on matching mutual nearest neighbors in the two data sets [20]. After correction, the two data sets showed a large degree of overlap (S5A Fig). This allowed us to use the cell types found by Lindström and colleagues to classify the cell clusters found here, using a k-nearest neighbors approach (see Methods). NPCa–c were also classified as NPC by Lindström and colleagues, whereas NPCb were considered “primed NPC,” which supports the notion that NPCb were primed to differentiate. The NPCd cluster was classified as “proliferating cells.” This classification is in agreement with our observation that NPCd seemed to proliferate more than other NPC subtypes (Fig 2B–2D). Because NPCd expressed low levels of NPC markers (such as SIX2 and CITED1), these cells were likely in a transition state between NPCs and PTA cells. Whereas the majority of PTA cells identified here were considered “PTA/RV I” by Lindström and colleagues, RVCSBa cells were spread over multiple cell types. This spread was likely due to the fact that transitory cell types are transcriptionally similar, and their clustering is therefore less robust. Nevertheless, the “PTA/RV II” cluster received most of the RVCSBa cells. RVCSBb cells were called “podocyte precursors” in the Lindström data set, whereas SSBpod as well as Pods were classified as “podocytes.” In our data set, RVCSBb directly preceded SSBpod (Fig 3A), so they could indeed be considered podocyte progenitors. Below, we will show that SSBpod did form a cell state separate from Pods and should not be grouped with them (see Podocyte development). In agreement with our analysis, the majority of SSBpr were classified as “proximal precursor” or “proximal tubule,” and all ErPrT were considered “proximal tubule” by Lindström and colleagues. CnT and DTLH were both classified as “distal/loop of Henle (LOH) precursor.” The fact that two cell types in the study by Lindström and colleagues (“podocytes” and “distal/LOH precursor”) were split in multiple subclusters in our study likely reflects differences in sample preparation. Whereas Lindström and colleagues preferentially released single cells from the nephrogenic niche, here, the whole kidney was used. Consequently, the Lindström data set has a finer resolution of NPCs and early, proliferating cell types, whereas our data set allowed us to resolve more mature cell types. The two data sets therefore complement each other.

Marker identification

To confirm the inferred cell types and also identify novel markers, we pursued two complementary strategies. First, we determined a set of marker genes based on their usefulness as classifiers for individual cell types: for each gene, the performance of a binary classifier was evaluated by the area under the receiver operating characteristic (AUROC) and combined with expression level filtering (see Methods). This resulted in 88 marker genes (S6A Fig, marker set, S3 Table). Only 11 of these markers overlapped with the 89 genes in the literature set (S6C Fig). To our knowledge, many of the remaining markers had not been associated with kidney development in previous studies. As an independent approach, we used the KeyGenes algorithm [21] to identify classifier genes among the 500 most highly variable genes (HVGs), using two-thirds of all cells as a training set. Based on the classifier genes determined by KeyGenes, we next predicted the cell types of the remaining one-third of the cells (test set). Cell types could be predicted with an average certainty (id score) of 0.59; 24% of the cells in the test set obtained an id score higher than 0.8. Of the 95 classifier genes (S6B Fig, KeyGenes set, S3 Table), 24 were the same as in the marker set, and 14 were common with the literature set (S6C Fig).

Because the interpretation of the found cell clusters was largely based on markers identified in mouse development, we were wondering whether the new markers identified here were informative for the classification of cell types in the mouse kidney. Using a scRNA-seq measurement of cells from a whole P1 mouse kidney [22], we plotted the expression of the newly identified marker genes in single cells (S7 Fig). In many cases, markers that were found to label a particular cell type in the human fetal kidney were coexpressed in the same subset of mouse cells. A few markers, however, were either ubiquitously expressed or almost completely absent. This might be due to interspecies differences.

Comparison of different developmental ages

By establishing the identity of cell clusters at w16, we obtained a snapshot of cell type diversity in the fetal kidney. To explore whether the identified expression patterns change dynamically throughout development, we analyzed four additional samples from different developmental ages (w9, w11, w13, and w18), which together contained 11,359 usable cells. Using, again, batch correction based on mutual nearest neighbors [20], we visualized all samples in a common tSNE map (Fig 4, S8 Fig). Overall, gene expression in the different samples was largely overlapping for the majority of cell types. For example, proximal tubules cells (ErPrT) appeared at the same positions in the tSNE map in all samples (Fig 4B). The position of Pods, however, shifted systematically across different ages, which corresponds to a continuing change in expression pattern (Fig 4C). This observation might suggest that Pods further matured in terms of their expression pattern after being specified.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Comparison of different developmental ages suggested continued expression changes in podocytes.

(A) tSNE map combining all five samples (w9, w11, w13, w16, w18). Samples were corrected for batch effects by matching mutual nearest neighbors [20]. Cells in the w9, w11, w13, and w18 samples were classified by comparing to the w16 sample using a k-nearest neighbors-based approach (see Methods). (B) tSNE map of all ages restricted to ErPrT. Labels and colors indicate ages. Six outlier cells were omitted from this plot to improve visualization. (C) tSNE map of all ages restricted to Pods. Labels and colors indicate ages. The numerical data underlying this figure can be found in S1 Data. CnT, connecting tubule; DTLH, distal tubule/loop of Henle; End, endothelial cells; ErPrT, early proximal tubule; ICa, interstitial cells a; ICb, interstitial cells b; IPC, interstitial progenitor cell; Leu, leukocyte; Mes, mesangial cells; NPCa, nephron progenitor cells a; NPCb, nephron progenitor cells b; NPCc, nephron progenitor cells c; NPCd, nephron progenitor cells d; Pod, podocyte; Prolif, proliferating cells; PTA, pretubular aggregate; RVCSBa, renal vesicle/comma-shaped body a; RVCSBb, renal vesicle/comma-shaped body b; scRNAseq, single-cell RNA sequencing; SSBm/d, s-shaped body medial/distal; SSBpod, s-shaped body podocyte precursor cells; SSBpr, s-shaped body proximal precursor cells; tSNE, t-distributed stochastic neighbor embedding; tx, transcript; UBCD, ureteric bud/collecting duct; w16, week 16.

https://doi.org/10.1371/journal.pbio.3000152.g004

Differential expression analysis of Pods of different ages revealed 109 differentially expressed genes (fold change > 2 in any comparison, false discovery rate (FDR) < 0.05, S4 Table). Functional annotation analysis of these genes showed significant enrichment of two gene ontology (GO) terms—“proteinaceous extracellular matrix” (adjusted p-value = 1.9 × 10⁻³, including SPON2, BGN, COL1A2, and CTGF) and “extracellular exosomes” (adjusted p-value = 1.4 × 10⁻³, including NPNT, S100A10, ANXA1, and EPCAM). Some of the differentially expressed genes have been shown to be important for kidney development. For example, NPNT and DCN showed increasing expression from w11 to w18. Knockout of the extracellular matrix protein NPNT in mice decreases the invasion of the UB and causes agenesis or hypoplasia [23]. NPNT was further shown to be expressed in the glomerular basement membrane and to be necessary for podocyte adhesion in mice [24]. Ablation of this gene in mice causes podocyte effacement. As in the case of NPNT, DCN has been reported to be part of the glomerular basement membrane proteins [25]. This gene appeared strongly up-regulated in podocytes between w11 and w13 or w18 (fold changes of 3.25 and 4.6, respectively). The increase of NPNT and DCN expression over time in our data set could reflect an increase in adhesion between podocytes and glomerular basement membrane. Pods further showed significant differential expressions of genes related to stress, like HSPA1A and HSPA1B or NFKB genes (NFKB2, NFKBIA, and REL), with the highest levels at w18. This might suggest that dissociation-related stress increases with age for podocytes, maybe related to stronger adhesion of the cells, or that stress-related genes have another, physiological role in development.

We would like to emphasize that the observed gene expression changes with age should be considered with caution because they might be related to the differences in genotype between the samples. A much larger number of samples would be necessary to rule out such interindividual differences as a cause.

Having established the identity of the cell clusters, we next wanted to demonstrate how the data set can be used to explore different aspects of kidney development. We specifically focused on the nephrogenic niche, which showed pronounced heterogeneity, and the development of podocytes, which progressed via a distinct, intermediate cell state (SSBpod).

Heterogeneity in the nephrogenic niche

The formation of the nephron epithelium starts with the NPCs that differentiate and form the PTA, RV, and CSB. Studies in the mouse suggest that cells in the NPC compartment are not biased towards a particular lineage and patterning is first detectable in the PTA [26]. Nevertheless, the w16 scRNA-seq data indicated the presence of several nephron progenitor subpopulations, NPCa–d. To clarify the temporal relationship of these clusters, we employed Monocle 2 again to arrange them together with the PTA cells on a pseudotime scale (Fig 5A). NPCa clearly preceded NPCb and c, which seemed to appear around the same pseudotime. NPCd cells followed NPCb and c and preceded PTA. This analysis suggested that NPCa are the “bona fide” NPCs and give rise to NPCb and c. NPCd, which were likely more proliferative than the other NPCs (Fig 2B–2D), seemed to be an intermediate state between (slowly cycling) NPCa–c and the PTA.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. The nephrogenic niche exhibited a complex spatial organization.

(A) Pseudotime analysis of the nephrogenic niche (NPC) and the PTA. Two-dimensional DDRTree [18] embedding and the learned graph (shown as a black line) were calculated with Monocle 2 [17]. Labels and colors indicate cell types. (B) Schematic sketch of the CM indicating the distance d from the UB to the edge of the CM (solid arrow) and the relative distance s along the UB (dashed arrow), in which 0 and 1 represent the top and bottom of the CM, respectively. (C) Representative image of SIX2 and CITED1 immunostaining in a w15 human fetal kidney. Dashed lines in the insets indicate the outline of the nuclei, based on DAPI signal. Arrows in the inset point to cells in which CITED1 is concentrated in the nucleus. Scale bar = 50 μm. (D) Quantification of SIX2 and CITED1 immunostaining with respect to the distance d from UB or distance s along the UB; see panel A. Error bars indicate the SEM calculated over all evaluated profiles (n = 24). (E) Representative image of HSPA1A, NR4A1, and CKS2 immunostaining in a w15 human fetal kidney. Scale bar = 20 μm. (F) Representative image of UNCX and CITED1 immunostaining. Arrowheads indicate the presence of immunostaining signal. Scale bar = 100 μm. The numerical data underlying this figure can be found in S1 Data. CM, cap mesenchyme; NPC, nephron progenitor cell; PTA, pretubular aggregate; SEM, standard error of the mean; UB, ureteric bud; w15, week 15.

https://doi.org/10.1371/journal.pbio.3000152.g005

To localize the NPC clusters in the tissue, we made use of the fact that they expressed various levels of CITED1 and SIX2 (Fig 2A): whereas NPCa and NPCc exhibited roughly similar levels of these markers, NPCb and NPCd had lost CITED1 almost completely, while retaining some SIX2 expression. In an immunostaining of a w15 kidney, CITED1 and SIX2 appeared overlapping in a subset of cells (Fig 5B–5D). Quantification of the fluorescence signal (Methods) revealed clear differences between their expression patterns. Whereas SIX2 expression was approximately constant throughout the CM, CITED1 expression decreased, relative to SIX2, with increasing (radial) distance from the UB (Fig 5D). A marked drop of CITED1 was visible between 10 and 20 μm from the UB, which approximately corresponds to the first layer of cells. To exclude that the observed difference between SIX2 and CITED1 expression was due to the different fluorophores on the secondary antibodies, we repeated the experiment with swapped fluorophores. This measurement produced a very similar expression gradient (S9A Fig). To exclude that the observed effect was influenced by PTA found in the CM towards the stalk of the UB, the analysis was repeated, taking only the 20% of CM cells closest to the edge of the cortex into account. A similar expression gradient was observed (S9C Fig). This result implies the existence of a CITED1 low/SIX2 high subpopulation of cells, which are not in contact with the UB. Secondly, we observed that CITED1 decreased relative to SIX2 towards the interface with the PTA and the stalk of the UB (Fig 5D). A similar observation was made when the experiment was repeated with swapped fluorophores (S9B Fig). Taken together, these results suggested that NPCa and NPCc were located closer to the surface of the UB and closer to the tip of the UB compared to the other NPC subtypes. Additionally, we also observed differences in subcellular localization of CITED1 protein within the CITED1 high compartment. Whereas for the majority of cells CITED1 was found in the cytoplasm, in several cells it was concentrated in the nucleus (right inset in Fig 5C). In contrast, SIX2 was always found restricted to the nucleus (left inset in Fig 5C). This observation might indicate that CITED1 was only active in a small population of cells, which would constitute another layer of cell–cell heterogeneity.

In addition to the observed heterogeneity in CITED1 and SIX2, differences between the NPC clusters could also be gleaned from the set of novel markers (marker set, S3 Table).

TMEM100 and RSPO3 specifically marked NPCa. RSPO3 is an activator of the canonical WNT signaling pathway [27], suggesting a role of WNT either in NPC self-renewal or UB branching morphogenesis. Notably, all markers of NPCb (CACYBP, MRPL18, ZFAND2A, DNAJB1) were related to the stress response in some form. The markers of NPCc (CRABP2, HAS2, MDK, HOXC6), which were also expressed in the other NPC types, are all either targets of retinoic acid or binding it [14,28,29]. MDK has been shown to be expressed in the CM of the developing rat kidney, and its neutralization reduced the number of formed nephrons in vitro [30,31]. Finally, the NPCd markers CENPF, HMGB2, CCNB2, and NUSAP1 all have a role in cell cycle regulation or proliferation [16]. HMGB2 was recently implied in the activation of quiescent adult neural stem cells [32].

The observation that markers of NPCb were related to the stress response seemed to suggest that this cluster was created as an artefact of cell dissociation [33], despite our best efforts to remove stressed cells (see Methods). On the other hand, the vast majority of NPCb cells were classified as “primed NPC II” in the Lindström data set (S5 Fig). The fact that NPCb cells were only detected in the w16 and w18 kidneys is consistent with single-cell dissociation becoming increasingly difficult with fetal age, or alternatively, with a progenitor cell aging phenomenon. To explore the differences between NPCb and the other NPC clusters further, we immunostained HSPA1A and NR4A1, both known stress-response genes, in w15 kidney sections (Fig 5E). HSPA1A was identified as a marker of NPCb (S9D Fig, S3 Table), whereas NR4A1 was expressed in multiple NPC clusters but highest in NPCb (S9D Fig). Furthermore, NR4A1 was also identified in the study by Adam and colleagues [22] to be up-regulated in response to elevated temperatures during enzymatic dissociation. HSPA1A and NR4A1 were both observable in the nephrogenic niche at the level of the stalk of the UB and at the transition to the PTA or RV. Additionally, we studied the expression of EGR1, another stress-related gene that marked NPCb, with smFISH (S9E Fig). EGR1 was mainly found toward the stalk of the UB and in a few cells around the tip of the UB, whereas SIX2 and CITED1 transcripts were visible throughout the CM (S9F Fig). Because results obtained in fixed tissue sections are not confounded by dissociation-related artifacts, these immunostainings and smFISH measurements supported the existence of NPCb cells in the fetal kidney.

The fourth NPC cluster, NPCd, was clearly distinguished by proliferation markers (Fig 2B–2D). To locate NPCd in the tissue, we immunostained the cell cycle regulator and NPCd marker CKS2 (S2G Fig). CKS2 signal could be observed around the stalk of the UB and in RVs (Fig 5E). This result supported the interpretation that NPCd were a proliferating transitory state between NPCa–c and PTA.

Given the crucial role of the nephrogenic niche in the development of nephrons, it is likely that misexpression or mutation of genes that are specifically expressed in NPCs affect kidney function. Mining a database of genome wide association studies (GWAS) revealed that genes that were differentially expressed in NPCs were significantly enriched for association with kidney disease (p = 1.7 × 10⁻³, one-sided Fisher’s exact test). No enrichment was found for lung diseases (p = 0.21, one-sided Fisher’s exact test) (see Methods, S10 Fig, S4 Table). Unsurprisingly, several of the disease-associated genes are known regulators of kidney development, such as SALL1, SOX11, and HAS2 [34–37]. The other identified genes had not been previously associated with kidney development. For example, DDX1, which was differentially expressed in NPCs as well as SSBpod, is an RNA helicase that promotes micro-RNA maturation [38]. UNCX, which was broadly expressed in all NPC clusters, is a homeobox transcription factor involved in somitogenesis and neurogenesis [39] and has also been found to be up-regulated in the induced mouse nephrogenic mesenchyme in culture [40]. It was recently associated with renal-function–related traits [41] as well as glomerular filtration rate [42–44]. In our data, the expression profile of UNCX was similar to that of CITED1 (S9H Fig). Immunostaining of UNCX confirmed the scRNA-seq results and showed expression of UNCX in the nephrogenic zone, as marked by CITED1 (Fig 5F and S9I Fig). These findings suggested UNCX as a novel potential regulator of early nephrogenesis.

Podocyte development

Another cell type of high relevance for kidney function is the podocyte. This cell type is critical for filtration and is implied in several forms of kidney disease [45]. Podocytes wrap around the glomerular basement membrane (capillary bed) inside Bowman’s capsule (Fig 6A). Clustering (Fig 1C) and pseudotime analysis (Fig 3) of the w16 kidney data set had indicated that development into Pods occurs via a distinct intermediate state that we dubbed SSBpod here. This cell state was likely related or even identical to previously discovered podocyte precursors [19,46]. In the Lindström data set [19], SSBpod and Pods were both classified as “podocytes,” and the RVCSBb were considered “podocyte precursors” (S5B Fig). To show that SSBpod cells were indeed a localizable cell state distinct from Pods, we further investigated their expression pattern (Fig 6B), focusing on known literature markers (literature set) and the marker set (S3 Table). Compared to RVCSBb, SSBpod showed higher expression of MAFB and FOXC2, which are necessary for the determination of podocyte identity [47,48]. On the other hand, compared to Pods, they exhibited lower expression of genes typically associated with more mature podocytes, like CLIC5, PODXL, and PTPRO. Filtration function-related genes like NPHS1, NPHS2, and PTPRO were expressed at intermediate levels in SSBpod compared to RVCSBb, where they were absent, and Pods, where they are highly expressed. A similar pattern could be observed for genes associated with podocyte polarization or structural organization as well as pedicel growth and patterning. Finally, Pods showed the expression of genes that negatively regulate the cell cycle and support long term survival, consistent with their postmitotic nature [49]. In contrast, SSBpod specifically expressed ORC4, which has a function in DNA replication. However, proliferation markers were lowly expressed in both SSBpod and Pods (Fig 2D), which suggested low proliferative potential in both cell types. In contrast to NPCs, association with kidney disease was not significantly enriched among genes differentially expressed in SSBpod (p-value = 0.1, one-sided Fisher’s exact test). One of the disease-associated genes was OLFM3, which has been associated with glomerular filtration rate [50]. OLFM3, a secreted glycoprotein, has a known function in brain and retina development [51] and has been identified as a marker for podocyte precursors in two independent studies [19,46]. In our data set, it was specifically expressed in SSBpod (Fig 6C) and was a marker for this cell type in the marker set and KeyGenes set (S3 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Podocytes developed via a precursor state localized in the visceral proximal SSB.

(A) Schematic of development from the SSB to the glomerulus. Regional patterning is shown in color—proximal (purple), medial (green), distal (orange). (B) Expression heat map of the subset of marker set genes that are markers for SSBpod and Pods. Expression was Freeman-Tukey transformed, averaged over all cells in a cluster and standardized gene-wise. (C) Two-dimensional tSNE map showing the expression of OLFM3. Expression is indicated by color; expression values of 1 are plotted in gray. (D) Typical images of MAFB, PODXL, and ACTA2 immunostaining in SSB and glomeruli. w15 female kidney. Scale bar = 50 μm. (E) Representative images of smFISH of OLFM3, MAFB, and CLIC5 in SSBpod and Pods. w15 female kidney. Scale bar = 10 μm. (F) Box plots of smFISH signal densities in SSB (n = 10), capillary loop (n = 4), and glomeruli (n = 8), for OLFM3, MAFB and CLIC5 (* adjusted p < 0.05, ** adjusted p < 0.0005). w15 female kidney. (G) Volcano plot of differential gene expression between SSBpod and Pod. L2FC Pod over SSBpod versus −log10 (adjusted p-value). Genes with an adjusted p < 0.05 and L2FC > 1 were considered significant (colored data points). Genes with an AP-1 binding site are shown in red. The numerical data underlying this figure can be found in S1 Data. CnT, connecting tubule; DE, differentially expressed; DTLH, distal tubule/loop of Henle; End, endothelial cells; ErPrT, early proximal tubule; ICa, interstitial cells a; ICb, interstitial cells b; IPC, interstitial progenitor cell; Leu, leukocyte; L2FC, log2 fold change; Mes, mesangial cells; NPCa, nephron progenitor cells a; NPCb, nephron progenitor cells b; NPCc, nephron progenitor cells c; NPCd, nephron progenitor cells d; Pod, podocyte; Prolif, proliferating cells; PTA, pretubular aggregate; RVCSBa, renal vesicle/comma-shaped body a; RVCSBb, renal vesicle/comma-shaped body b; smFISH, single molecule fluorescence in situ hybridization; SSBm/d, s-shaped body medial/distal; SSBpod, s-shaped body podocyte precursor cells; SSBpr, s-shaped body proximal precursor cells; tSNE, t-distributed stochastic neighbor embedding; tx, transcript; UBCD, ureteric bud/collecting duct; w15, week 15.

https://doi.org/10.1371/journal.pbio.3000152.g006

In order to localize SSBpod, Pods and mesangial cells in situ, we immunostained w15 kidney sections with antibodies for MAFB, PODXL, and ACTA2 (Fig 6D). As expected, PODXL and MAFB were found in Pods at the capillary loop stage and in more mature glomeruli. MAFB staining extended to the proximal segment of the SSB, which indicated that SSBpod may be part of this structure. To locate the SSBpod cells more precisely, we performed smFISH on CLIC5 and MAFB, expressed both in Pods and SSBpod (Fig 6E). We observed a subpopulation of MAFB+/CLIC5− cells outside the glomeruli, which we identified as the SSBpod. These cells could be found predominantly in the visceral part of the proximal segment of the SSB but also at the capillary loop stage. This result supported the notion that SSBpod were transient cells that preceded (mature) Pods. Having localized the SSBpod, we next wanted to confirm OLFM3 as a marker of this cell type. smFISH of OLFM3, MAFB, and CLIC5 showed OLFM3 to be coexpressed with MAFB but absent in cells that were positive for CLIC5 (Fig 6E), a marker that persists in podocytes in the adult kidney. Quantification of the density of smFISH signals (Fig 6F) showed that OLFM3 was absent in glomeruli but could be detected in the subpopulation we identified as SSBpod (MAFB+/CLIC5−). In summary, these results supported OLFM3 as a robust marker of podocyte precursors.

Finally, we were wondering whether our data set would also allow us to identify candidate mechanisms that drive development from SSBpod to Pods. Differential expression analysis revealed 228 genes that had a significant, bigger than 2-fold changes between SSBpod and Pods (Fig 6G, S4 Table). Among these we found factors belonging to multiple signaling pathways, such as FGF1, VEGFA, HES1, and EGF1. Vegfa and Fgf1 are known to have a homeostatic function in podocytes [52–54], whereas Hes1, a target of the NOTCH signaling pathway, seems to be necessary for the synthesis of extracellular matrix proteins in these cells [55]. Binding sites for the transcription factor AP-1 were strongly enriched in this set of genes (145 out of 228 genes, adjusted p-value = 1.3 × 10⁻⁵). AP-1 would therefore be an interesting target for perturbation studies in mouse models.

All in all, the results presented here complement other, recent, single-cell transcriptomics studies of the fetal kidney. We demonstrated how the data can be interrogated to find expression patterns that will improve our understanding of human kidney development.

Heterogeneity in the nephrogenic niche

Building an organ during development requires the careful balance between two fundamental processes—growth and the creation of structure. In many organs, these two functions are reconciled by self-renewing progenitor cells that can be induced to differentiate. In kidney development, NPCs give rise to the epithelium of the nephron, the functional unit of the kidney. To balance growth with patterning, self-renewal and differentiation of NPCs have to be tightly controlled. It is well established that the niche of the NPC plays an important role in this control, but the precise mechanisms are not well understood. In particular, it is not clear how the position and movement of NPCs in the niche might impact the induction towards differentiation.

Heterogeneity in the nephrogenic niche was brought to light first by Mugford and colleagues in 2009 [26] and has been confirmed by multiple recent studies [10,19,46,56,57]. Mugford and colleagues used in situ hybridization to study the localization of transcriptional regulators in E15.5 mouse kidney. Three distinct compartments were defined in the CM—inner capping mesenchyme (which lies closest to the cleft of the UB), outer capping mesenchyme (at the tip of the UB), and induced mesenchyme (at the level of the stalk of the UB). Although all compartments expressed Six2, only inner and outer capping mesenchyme expressed Cited1. The induced CM was distinguished by Wnt pathway activity, as evidenced by Wnt4 expression. Several recent scRNA-seq studies confirmed heterogeneity in the nephrogenic niche. Brunskill and colleagues studied an E12.5 mouse kidney and found two subpopulations in the CM, which they classified as uninduced (Six2 positive, Cited1 positive) and induced (Six2 positive, Cited1 negative) [56]. Among the hundreds of genes that were differentially expressed between these two populations, they found genes related to the Wnt signaling pathway as well as protein vesicular trafficking and degradation. Wang and colleagues also found two subclusters in the CM of the human fetal kidney [57]. They interpreted one subcluster as the self-renewing compartment due to higher expression of markers for cell division. The other subcluster, which showed gene expression related to NOTCH signaling (HES1, HEY1), was considered induced. Two studies by Lindström and colleagues [10,19] also explored NPC heterogeneity. The first study [10] identified four NPC clusters (self-renewing, primed, differentiating, and proliferating), whereas the second [19] revealed four clusters of NPCs (I–IV), two clusters of primed NPCs (I–II), as well as several clusters of proliferating cells.

In the data set presented here, we identified four clusters of NPCs. Among these, NPCa are most likely the self-renewing compartment. In agreement with the studies by Lindström and colleagues [10,19], they expressed the highest levels of CITED1 and TMEM100 compared with the other NPCs. Furthermore, they preceded all other NPC clusters in pseudotime analysis. NPCb showed expression of several genes that modulate NOTCH, bone morphogenetic protein (BMP), and transforming growth factor beta (TGF-beta) pathway activity, as well as low levels of LEF1, which has been shown to indicate induction towards differentiation [10,19]. The classification of NPCb as “primed NPC” by comparison to the Lindström data set supported the interpretation of NPCb as a state distinct from NPCa that is primed to differentiate. The fact that we detected NPCb only at w16 and w18 leads us to speculate that NPCb could be the result of continuous changes in the nephrogenic niche over the course of development. The third NPC cluster, NPCc, appeared together with NPCb in pseudotime and was distinguished from the other NPCs by higher expression of genes involved in or regulated by retinoic acid signaling. The retinoic-acid binding protein CRABP2 has been identified as an NPC marker in other reports [10,57]. We speculate that NPCc are the result of spatially varying concentrations of retinoic acid, which is produced in the cortical interstitium [58]. Finally, NPCd appeared between NPCb–c and PTA in pseudotime and were clearly distinguished from the other NPC clusters by increased proliferation, at least as far as that can be inferred from gene expression data. In agreement with our analysis, NPCd were classified as “proliferating cells” by comparison with the Lindström data set [19]. NPCd cells also lowly expressed markers of induction towards differentiation (such as LEF1, LHX1, WNT4), which indicates a transitory state between induced and/or primed NPCs and PTA. The suggested developmental flow from NPCa via NPCb–c to NPCd was supported by a gradual decrease of OSR1, which is a well-known marker of the early CM.

By in situ detection of CITED1, SIX2, and other genes, we also explored the spatial localization of the different NPC clusters. NPCa seemed to reside closest to the tip of the UB, the induced and/or primed NPC b and c were situated closer to the stalk, and NPCd were closest to the PTA. This finding is consistent with the recent report of NPCs streaming from their niche at the UB tip towards the UB branch point to form the PTA and RV [19]. On this path, the cells gradually lose the NPC transcriptional program, and differentiation is induced. In the mouse, trajectories of NPCs also seem to have a large stochastic component: NPCs repeatedly detach from the UB and attach again and also shuttle back and forth between the “uninduced region” at the UB tip and the “committed region” around the stalk of the UB [59]. This observation could indicate that varying expression levels of genes such as CITED1 occur as a consequence of cell migration and are not necessarily functionally relevant. Indeed, Cited1 knockout has no adverse effects on kidney development in the mouse [60]. Taken together, our results support a model in which (self-renewing) NPCa reside at the tip of the UB, probably in close proximity or even in contact with the UB. Movement away from the UB tip, toward the stalk, is accompanied by decreased CITED1 expression and transformation to the (induced and/or primed) NPCb–c states. Arrival at the stalk of the UB is characterized by the NPCd expression state, increased proliferation, and eventually transformation to the PTA. It is conceivable that cells sometimes visit the different NPC states in reverse order, which would reconcile this model with the observed high, multidirectional motility of NPCs [59].

Proximal–distal patterning

In the prevailing model of mammalian kidney development, self-renewing NPCs are not prepatterned to develop into a certain lineage. When the developing nephron first displays signs of proximal–distal patterning is an important, outstanding question. Mugford and colleagues have found evidence that the PTA, which succeeds the NPCs, is already polarized [26]. A recent study by Lindström and colleagues [19] proposed an intriguing mechanism that couples temporal and spatial cues: whereas NPCs that are recruited to the PTA early develop into distal cell types, NPCs that are integrated later contribute to the proximal compartment. In our study, we identified the PTA by known marker genes (CCND1, LHX1, and WNT4) and high proliferation. We were unable to detect any substructure within the PTA, which might be due to the limited resolution of our scRNA-seq method. RV/CSB, the next developmental stage, however, was split in two clusters (a and b). Pseudotime analysis suggested that RVCSBa was a heterogeneous cluster comprising early RV cells (which appeared before RVCSBb) and the distal segment of the RVCSB. This observation is consistent with the time-dependent recruitment model by Lindström and colleagues [19] in the sense that in that model, distal specification precedes proximal patterning.

Podocyte development

Single-cell transcriptomics studies of various organs have brought to light many new, intermediate cell states. This has provoked the question of whether we should consider gene expression in complex tissues as a continuum rather than a collection of distinct expression profiles. In developmental systems, it is certainly useful to think about gene expression change as a continuous process. Nevertheless, there are clearly distinct intermediate cell states even within linear developmental paths. In our study, we observed that RVCSBb gave rise to Pods via an intermediate state, the SSBpod, which directly preceded the Pods in pseudotime. Specific expression of OLFM3 made it likely that this cluster is identical to previously identified podocyte precursors, which were marked by this gene [19,46]. Menon and colleagues defined a cluster of “immature” or “early” podocytes, characterized by high OLFM3 and low MAFB expression. In that study, podocytes showed increased MAFB expression but loss of OLFM3 [46]. Lindström and colleagues located OLFM3 positive cells to the proximal part of the SSB [19]. In our study, we confirmed all of these observations: OLFM3 was localized to the visceral part of the proximal segment of the SSB, and OLFM3 negative Pods showed higher expression of mature podocyte markers compared to the OLFM3 positive SSBpod. Functional annotation analysis of genes that were differentially expressed between SSBpod and Pods revealed enrichment of a binding site for AP-1. This transcription factor has been found to be important for the development of the skin [61], neural precursor cells [62], and the heart valve [63] in mice. A role of AP-1 in kidney development has not been described yet, and further research is needed to elucidate its potential function. The analysis of the kidneys from different gestational ages showed high similarity of cell types across different ages with the exception of Pods. These displayed a systematic change in expression pattern, which might indicate the continued maturation of Pods over time. This observation is in agreement with a study by Brunskill and colleagues [64] in the mouse, which compared embryonic (E13.5 and E15.5) with adult podocytes (defined as MafB positive cells). That study found hundreds of genes that were differentially expressed between embryonic and adult podocytes. Furthermore, targeted experiments are needed to demonstrate the possible maturation of podocytes in human kidney development.

In summary, we have leveraged a combination of single-cell transcriptomics and in situ imaging to study the intricate structure of the developing human kidney. The transcriptomics data, accessible via a web application (http://www.semraulab.com/kidney), will be a valuable starting point for discovering gene regulatory mechanisms or finding new disease mechanisms.

Ethics statement

The collection and use of human material in this study was approved by the Medical Ethics Committee from the Leiden University Medical Center (P08.087). The gestational age was determined by ultrasonography, and the tissue was obtained by vacuum aspiration from women undergoing elective abortion. The material from six embryos (w9, male; w11, male; w13, female; w15, female; w16, male; and w18, female) was donated with written informed consent. Questions about the human material should be directed to S. M. Chuva de Sousa Lopes (Lopes@lumc.nl).

Experimental methods

Quantification and statistical analysis