Cross-platform single cell analysis of kidney development shows stromal cells express Gdnf

Abstract

The developing kidney provides a useful model for study of the principles of organogenesis. In this report we use three independent platforms, Drop-Seq, Chromium 10x Genomics and Fluidigm C1, to carry out single cell RNA-Seq (scRNA-Seq) analysis of the E14.5 mouse kidney. Using the software AltAnalyze, in conjunction with the unsupervised approach ICGS, we were unable to identify and confirm the presence of 16 distinct cell populations during this stage of active nephrogenesis. Using a novel integrative supervised computational strategy, we were able to successfully harmonize and compare the cell profiles across all three technological platforms. Analysis of possible cross compartment receptor/ligand interactions identified the nephrogenic zone stroma as a source of GDNF. This was unexpected because the cap mesenchyme nephron progenitors had been thought to be the sole source of GDNF, which is a key driver of branching morphogenesis of the collecting duct system. The expression of Gdnf by stromal cells was validated in several ways, including Gdnf in situ hybridization combined with immunohistochemistry for SIX2, and marker of nephron progenitors, and MEIS1, a marker of stromal cells. Finally, the single cell gene expression profiles generated in this study confirmed and extended previous work showing the presence of multilineage priming during kidney development. Nephron progenitors showed stochastic expression of genes associated with multiple potential differentiation lineages.

Keywords: Gdnf; ICGS; Kidney development; Multilineage priming; ScRNA-Seq.

Figure 1. Integrative workflow for multi-platform single-cell analysis

Experimental design included dissociation of E14.5 mouse kidney cells, scRNA-Seq data generation with Chromium 10X Genomics, Drop-Seq and Fluidigm HT 800 cell IFC platforms, unsupervised bioinformatics classification of the resulting scRNA-Seq profiles, supervised harmonization of the three datasets and downstream cell-population level analyses.

Figure 2. Unsupervised cell-population analysis for Drop-Seq, Fluidigm and Chromium E14.5 kidney scRNA-Seq data

A) De novo identified cell populations from the software ICGS are shown for each scRNA-Seq platform. The displayed heatmaps were produced by the MarkerFinder algorithm, downstream of the ICGS population predictions, with yellow indicating high relative gene expression and blue or black, low or no gene expression in the associated genes (rows). Prior established embryonic kidney marker genes corresponding to compartments are shown in panel C. Text to the left of each heatmap indicates the statistical enrichment of genes from the Drop-Seq ICGS analysis for the 16 identified populations (MarkerFinder) using the embedded gene-set enrichment analysis tool GO-Elite in AltAnalyze. B–C) t-SNE plot derived from the ICGS heatmaps in panel A, where each dot represents individual cells colored according to its B) ICGS cluster annotation or C) prior established population specific genes. CD: Collecting duct, UT: Ureteric Tip, LOH: Loop of Henle, RV: Renal vesicle, DCSB: Distal comma shaped body, Pod: podocyte, PT: Proximal Tubule, PA: Pre-tubular aggregate, CM: Cap mesenchyme, Endo: Endothelium, NZS: Nephrogenic Stroma, CS: Cortical Stroma.

Figure 3. Classification and comparison of kidney cellular heterogeneity between scRNA-Seq platforms

A) ICGS delineated cellular heterogeneity in the top ~2,000 DropSeq captured barcodes, segregated into 16 populations. B–C) Supervised classification of Fluidigm-800 chip captured libraries and 10X Genomics Chromium In-Drop barcodes using B) K-nearest neighbor (knn) classification against the Drop-Seq population centroids or C) MarkerFinder classified cells (rather than genes) for the top 862 population-specific genes from the Drop-Seq analysis. The upper panel displays the 862 Drop-Seq population specific genes and lower panel the top de novo MarkerFinder-gene results obtained following cell-population assignment. D) Comparison of population-specific genes jointly identified by all three scRNA-Seq technological platforms from the knn or MarkerFinder analysis. E) Number of genes jointly identified population-specific genes for each individual population for the two classification approaches. DCSB: Distal comma shaped body.

Figure 4. Orthogonal scRNA-Seq platforms identify equivalent kidney cell-populations and population-specific genes

A) Cell-population annotation predictions assigned from literature and GUDMAP gene-set annotations. Corresponding kidney compartments are colored according to the ICGS cell cluster colors. B) t-SNE analysis of the harmonized knn classified cell states using the de novo identified MarkerFinder genes for each scRNA-Seq platform. The number of cells present in the plot are indicated. C) Gene expression bar chart (log2) for prior annotated kidney developmental marker genes from the knn classified datasets. D) Comparison of population-specific genes consistently identified by two or more scRNA-Seq platforms or that are specific to a single platform by MarkerFinder-gene analysis.

Figure 5. Cell type specific gene expression validations

A: Rprm, UT: ureteric tip B: Pcp4, Pod: Podocytes, C: Spock2, CM: cap mesenchyme, D: Gpc3, NZS: Nephrogenic zone stroma, E: CRABP1, NZS, F: Penk, CS: Cortical Stroma, G: Alx1, CS, H: Col6a1, MS: Medullary stroma. A, B, D, F, G, and H are from the Allen Brain Atlas, and C from the GUDMAP database. E was validated using immunofluorescence.

Figure 6. Possible cell-cell interactions identified through receptor-ligand interaction analysis

A circos plot displaying prior annotated receptor ligand interactions shared between all possible cell-type combinations. Cell clusters are labeled on the outer most edge of the circle. This analysis was restricted to MarkerFinder cell-state specific genes (Pearson correlation >0.3). Receptors are labeled in red and ligands are labeled in blue. Interconnecting lines are color coded to match the heat map cell clusters. MCD: medullary collecting duct, CD: collecting duct, UT: ureteric tip, LOH: loop of Henle, DCSB, distal comma shaped body, Pod: podocyte, MSSB: mid S-shaped body, PT: proximal tubule, CM: cap mesenchyme, Endo: Endothelium, NZS: nephrogenic zone stroma, CS: cortical stroma, MS: medullary stroma.

Figure 7. Gdnf expression in the nephrogenic zone stroma

Heatmaps using data from all three scRNA-seq platforms show Gdnf expression by stromal cells. Six2, Cited1 and Crym are markers of cap mesenchyme (CM), while Meis1, Foxd1, Crabp1 and Aldh1a2 are expressed in stroma. As expected, cells that clustered with cap mesenchyme as determined by these markers as well as complete gene expression signatures often showed expression of Gdnf. Surprisingly, many cells that strongly clustered with the stromal cell compartment also showed robust Gdnf expression.

Figure 8. Gdnf and Six2 in situ hybridization

A,B: Gdnf and C,D: Six2 in situ hybridization images of E13.5 embryonic kidneys from the Allen Brain atlas. Arrows point to regions of Gdnf expression that are deeper in the kidney than normally seen for Six2, suggesting that they are not cap mesenchyme cells. Arrowheads point to regions overlying ureteric tips, showing low levels of Gdnf expression.

Figure 9. Gdnf expression in stromal cells

A: Gdnf in situ hybridization (ISH) showing Gdnf expression in the nephrogenic zone. B: SIX2 immunohistochemistry (IHC) showing expression in the cap mesenchyme and its early derivatives. C: MEIS1 IHC showing expression in the stroma. D: Gdnf ISH alone, higher magnification (20X). E: Same Gdnf ISH region as in panel D, but with added SIX2 IHC, highlighting the presence of Gdnf positive, SIX2 negative cells, indicating expression of Gdnf by cells that are not cap mesenchyme. E′: 40X view of Gdnf positive, SIX2 negative cells. F: Gdnf ISH alone (20X). G: Gdnf ISH and MEIS1 (IHC) double staining, showing the presence of Gdnf, MEIS1 double positive stroma cells. G′: 40x view of Gdnf positive, MEIS1 positive cells. Arrowheads point to cells positive for Gdnf and negative for SIX2. Dashed lines outline the SIX2 positive CM. Black double arrowheads point to Gdnf negative, MEIS1 positive cells below the nephrogenic zone. Arrows point to Gdnf positive, MEIS1 double positive cells within the nephrogenic zone.

Figure 10. Multi-lineage priming

Heatmap with representative cells from the CM: cap mesenchyme, PA: pretubular aggregate, DCSB: distal comma shaped body, MSSB: mid S-shaped body, PT: proximal tubule, LOH: loop of Henle, and Pod: Podocyte clusters from Drop-seq, Chromium 10X Genomics, and Fluidigm 800-cell. The early progenitor CM cells show stochastic expression of markers of multiple lineages. The MSSB cells are more committed and show elevated expression of proximal tubule associated genes and reduced expression of podocyte marker genes.