Profiling Cell Heterogeneity and Fructose Transporter Expression in the Rat Nephron by Integrating Single-Cell and Microdissected Tubule Segment Transcriptomes

Abstract

Single-cell RNA sequencing (scRNAseq) is a crucial tool in kidney research. These technologies cluster cells based on transcriptome similarity, irrespective of the anatomical location and order within the nephron. Thus, a transcriptome cluster may obscure the heterogeneity of the cell population within a nephron segment. Elevated dietary fructose leads to salt-sensitive hypertension, in part, through fructose reabsorption in the proximal tubule (PT). However, the organization of the four known fructose transporters in apical PTs (SGLT4, SGLT5, GLUT5, and NaGLT1) remains poorly understood. We hypothesized that cells within each subsegment of the proximal tubule exhibit complex, heterogeneous fructose transporter expression patterns. To test this hypothesis, we analyzed rat kidney transcriptomes and proteomes from publicly available scRNAseq and tubule microdissection databases. We found that microdissected PT-S1 segments consist of 81% ± 12% cells with scRNAseq-derived transcriptional characteristics of S1, whereas PT-S2 express a mixture of 18% ± 9% S1, 58% ± 8% S2, and 19% ± 5% S3 transcripts, and PT-S3 consists of 75% ± 9% S3 transcripts. The expression of all four fructose transporters was detectable in all three PT segments, but key fructose transporters SGLT5 and GLUT5 progressively increased from S1 to S3, and both were significantly upregulated in S3 vs. S1/S2 (Slc5a10: 1.9 log2FC, p < 1 × 10^-299; Scl2a5: 1.4 log2FC, p < 4 × 10^-105). A similar distribution was found in human kidneys. These data suggest that S3 is the primary site of fructose reabsorption in both humans and rats. Finally, because of the multiple scRNAseq transcriptional phenotypes found in each segment, our findings also imply that anatomical labels applied to scRNAseq clusters may be misleading.

Keywords: SGLT2 inhibitors; hexoses; salt-sensitive hypertension; sugar transport.

Figure A1

All rat scRNAseq clusters identified in the rat kidney were assigned to different cell classes by conducting a genomewide Pearson correlation with pseudobulk transcriptomes from human snRNAseq cell classes. (A) Normalized Pearson correlation coefficient between rat scRNAseq clusters (horizontal axis) and human snRNAseq cell classes (vertical axis) yielded: (1) Epithelial, 24 clusters; (2) Stromal, 1 cluster; (3) Immune, 8 clusters; and (4) Endothelial, 3 clusters, while no cluster in the rat dataset correlated with neural cells. (B) Uniform manifold approximation and projection (UMAP) of individual cell clusters. (C) Principal component analysis (PCA) projection. (D) t-distributed stochastic neighbor embedding (tSNE) projection of individual clusters.

Figure A2

The tubular epithelial cell cluster identities were annotated using a curated list of cell markers from HuBMAP Kidney v1.2. See, the main manuscript for details.

Figure A3

Cluster expression of gene markers for glomerular visceral epithelial cells. (A) Podocytes and (B) glomerular parietal epithelial cells (PEC). (C) Normalized Pearson correlation between rat scRNAseq cell clusters and human regional transcriptomes from the glomerular and tubulointerstitial regions (GLO.region and TI.region, respectively). Publicly available human kidney regional transcriptomics data were obtained from the Kidney Tissue Atlas (atlas.kpmp.org (accessed on 13 April 2023)).

Figure A4

Correlation between rat scRNAseq clusters transcriptomes and bulk transcriptomes from rat microdissected nephron segments. A genomewide Pearson correlation was calculated and normalized. For each microdissected segment, the values on the table represent the number of standard deviations that the correlation with each cluster deviates from the average of all clusters. We used the “Median” gene expression of each segment as originally published by Knepper and collaborators. The use of the median gene expression value is robust against contamination with other segments as far as such contamination is present in less than 50% of the samples. Bulk transcriptomics of microdissected S1 segments primarily correlate with the PT.S1 scRNAseq cluster but also present a strong association with the PT.S2 scRNAseq cluster. Microdissected S2 segments present similar correlations with all three PT subsegment clusters. Microdissected S3 segments present the strongest correlation overall with the PT.S3 cluster. Microdissected segments of the loop of Henle present a strong correlation with the corresponding cell types characterized in the scRNAseq. Notably, microdissected DCT correlated with the DCT cluster expressing the sodium-chloride symporter (NCC, Slc12a3) but also with the MTAL cluster. The PC cluster dominates the correlations in CNT and collecting duct segments. Finally, the IMCD cell type was most highly correlated with the inner medullary collecting duct (IMCD) segment transcriptome.

Figure A5

Correlation between rat scRNAseq cell cluster regions and transcriptomes from human kidneys. Normalized Pearson correlation between rat cell clusters and human single-nucleus (sn)RNAseq sub-regional transcriptomes. Of note, the rat scRNAseq dataset contains nearly 10% of the cell count of the KPMP snRNAseq and with less sequencing depth. As such the tubular epithelial clusters identified in the rat (horizontal axys), provided enough resolution for comparison with the subRegion (subclass.l1, vertical axys) but not the cell type (subclass.l2) identified in humans.

Figure A6

Normalized expression of sugar transporters in rat scRNAseq proximal tubule clusters. (A) Glucose transporters and (B) fructose transporters.

Figure A7

Normalized RNA transcripts of glucose transporters in rat scRNAseq proximal tubule clusters (A) compared to their corresponding gene products expressed in “million copies per cell” (MCPC) in microdissected rat proximal tubule segments (B). The order and colors of GeneIDs on the left panels match that of the corresponding gene products (Protein) on the right panels. Glucose and fructose transporter NAGLT1 are not shown in this figure, as it has already been shown in Figure 5 of the main manuscript.

Figure A8

Normalized RNA transcript expression in snRNAseq proximal tubule clusters from human kidneys. (A) Fructose transporters, (B) glucose transporters, and (C) fructokinase (KHK) and triokinase (TKFC).

Figure 1

Cell type assignments and principal component analysis (PCA) projections of rat kidney cell clusters. (A) Cell type annotation using HuBMAP Kidney v1.2 cell markers. (B) PCA projections show that Epithelial cells (green) separate from other cell classes (Immune: cyan, Endothelial: pink and Stromal (purple) along PC1, and (C) PCA projections show tubular epithelial cells separate along PC2 resembling the anatomy of the nephron (Proximal Tubules: pink, Thin Limbs: army green, Thick Ascending Limbs: green, Distal Convoluted Tubule: cyan and Collecting ducts: purple).

Figure 2

Expression of enzymes from the sorbitol pathway. (A) A dot plot of rat scRNAseq data show that sorbitol dehydrogenase (Sord), which converts sorbitol to fructose, is predominantly expressed in proximal tubules, while aldolase reductase (Akr1b1), which converts glucose to sorbitol, is predominantly expressed in medullary segments. (B) Analysis of human snRNAseq transcriptomes shows that the expression of aldolase reductase (AKR1B1) and sorbitol dehydrogenase (SORD) genes in the human kidney resembles that of the rat.

Figure 3

Expression of enzymes necessary to metabolize fructose into synthetic pathways. (A,B) Enzymes specific to fructose metabolism fructokinase (Khk) and triokinase (Tkfc) are mostly restricted to proximal tubules. (C,D) The enzymes shared with glycolysis and gluconeogenesis, aldolase B (Aldob), and triosephosphate isomerase (Tpi1) are widely expressed in other segments in addition to proximal tubules.

Figure 4

Sugar transporters are differentially expressed across proximal tubule cell types. (Upper table) log2 fold change (Log2FC) and p values on each segment, as compared to the other two segments together. (Lower panels) t-distributed stochastic neighbor embedding (tSNE) projections of proximal tubule clusters showing a transcript density map of the differentially expressed sugar transporter in rat proximal tubule scRNAseq clusters. Only differentially expressed sugar transporters are shown in the figure. The tSNE analysis was run in all cells (Appendix A in Figure A1D) and then the clusters corresponding to proximal tubules were extracted for visualization. Proximal tubule cells were not reclustered to create this figure.

Figure 5

Normalized RNA transcripts of fructose transporters in rat scRNAseq proximal tubule clusters (A) compared to their corresponding proteins expressed in “million copies per cell” (MCPC) in microdissected rat proximal tubule segments (B). Normalized RNA transcripts of fructokinase (Khk) and Triokinase (Tkfc) in rat scRNAseq proximal tubule clusters (C) compared to their corresponding proteins expression in microdissected rat proximal tubule segments (D). The order and colors of GeneIDs on the left panels (A,C) match that of the corresponding protein on the right panels (B,D).