Single-Cell Sequencing the Glomerulus, Unraveling the Molecular Programs of Glomerular Filtration, One Cell at a Time

Because of the complex heterogeneous nature of kidney structure and function in health and disease states, the mechanisms responsible for renal disease development and progression remain incompletely understood. Studies of genome-wide mRNA profiles of renal tissues have provided a starting point to map the molecular underpinning of the disease pathogenesis and progression. Gene expression profiles provide a snapshot of the transcriptional programs active in a given tissue or tissue compartment and allow key biologic mechanisms to be uncovered. Successes of this approach include the identification of noninvasive biomarkers from tissue gene expression studies of AKI¹ and CKD.² These results have provided the starting points for identification of novel therapeutic approaches.^3,4 The majority of these studies have been done on bulk mRNA samples, where the transcriptomic expression is averaged over thousands of cells.

At least 18 different kidney cell types have been recognized in the kidney. Aberration in any of those cell types could lead to renal disease manifestations, prompting studies to resolve gene expression levels down to a specific kidney cell type. Initial studies related cell type–specific mRNA levels (i.e., Nephrin or Podocin) in microdissected glomeruli to disease states and outcomes⁵ followed by the development of genome-wide deconvolution strategies with a specific focus on renal cell lineages.⁶ In parallel, approaches have been pursued to scale down expression profiling from bulk tissue to microdissected renal tissue compartments and single-cell quantitative RT-PCR analyses of individual isolated cells.⁷

The development of genome-scale profiling via RNA-Sequencing analysis at the single-cell transcriptome level (scRNA-Seq) has gained wider acceptance in the renal field over the last 2 years (R. Menon, et al., unpublished data).^8–10 Many collaborative efforts, including Gudmap (https://www.gudmap.org/tutorials/kidney-dev/), Rebuilding the Kidney (https://www.rebuildingakidney.org/), AMP (https://amp-ralupus.stanford.edu/about/ra-lupus-amp-project), and the Kidney Precision Medicine Project (https://kpmp.org/), are using scRNA-Seq as one of the strategies to generate a cellular atlas of normal and disease kidneys. The technology of obtaining expression profiles from single cells is rapidly evolving. Studies reported in this issue of the Journal of the American Society of Nephrology and elsewhere use mainly two single-cell RNA-Sequencing approaches are on the basis of the amount of kidney tissue sample available: shallow sequencing of large numbers of cells using a method like Dropseq or deep sequencing of a smaller number of cells using 10XGenomics technology, a similar droplet-based method.

A variety of computationally intensive bioinformatic tools have been used to identify the different cell types from single-cell transcriptomic data. Because most of the published transcriptomic disease biopsy studies have been performed on bulk mRNA, increased attention to the development of computational methods that can correctly model and analyze mixture samples is critical. Integrating the cell type–specific genes identified from the single-cell transcriptomic analyses along with such tools will predict the cellular composition in the mixture samples more accurately.

One of the unique features of scRNA-Seq analysis is the de novo identification of cell populations using an unsupervised clustering strategy. This approach can identify by their gene expression profile distinct renal cell populations including so far unknown cell (sub-)types and rare infiltrating cells. After the different cell clusters are identified, they can be further characterized and/or matched to known cell types by finding the cluster-specific differentially expressed genes.¹⁰ For instance, the unsupervised method can identify podocytes as a distinct cell cluster, although they make up <1% of the total kidney cell population. A cluster of interest can be further subclustered to reveal the different cellular states (i.e., immature, mature, or proliferating). This feature is valuable when studying a disease biopsy sample, because it may have different states of identical cell types or infiltrating ectopic cells. As an example, Figure 1A shows a distinct cell cluster identified by its gene expression in a renal biopsy of a patient with focal proliferative lupus nephritis. Close examination of the mRNAs defining this cluster identified among others the expression of LILVRA4, a marker for the infiltrating plasmacytoid dendritic cells. It is imperative to choose the right clustering algorithm and the associated parameters to identify the correct number of biologically meaningful clusters. Careful biologic characterization of the clusters using additional externally validated information is also essential.

Figure 1.

Single-cell RNA sequencing driven identification of unique cell types and transcripts in adult human kidney. (A) Feature tSNE plot showing the expression of LILRA4 (red). The expression is observed in a specific single-cell cluster among all clusters identified from an unsupervised clustering analysis of RNA-Sequencing analysis at the single-cell transcriptome level data from a focal proliferative lupus nephritis kidney biopsy sample. No expression of the gene in a cell is represented as green dots. LILRA4 is a marker for plasmacytoid dendritic cells, suggesting the infiltration of these cells in disease kidney. (B) Violin plot showing the expression of LINC00839 in the clusters identified from single-cell transcriptomic analysis of human adult kidney. Relative average mRNA expression is given along the y axis. Unique high expression of this long intergenic noncoding RNA is observed in podocytes compared with all other clusters. tSNE, T-distributed stochastic neighbor embedding.

Promising machine learning approaches, such as use of minimal spanning trees or nonlinear data embedding, can be applied to single-cell data to infer the order of cells along a biologic process. The data-driven arrangement of cell states into “pseudoprogression” trajectories enables intermediate cellular transitions in disease progression or differentiation to be inferred. When studying kidney diseases using gene expression data from clinical samples, this approach will allow the identification and mapping of the gene expression profile associated with the disease stage.

Another major concern in adult kidney single-cell transcriptome analysis is the method for dissociation of the tissue to a single-cell suspension for sequencing purposes. When the cells are being dissociated, cellular membranes can be damaged, resulting in mRNA leakage and degradation. The percentage of mitochondrial gene read content has emerged as a quality measure for the cell viability in single-cell studies. High percentage of mitochondrial reads indicates that the cells are less viable, because they have lost cytoplasmic mRNAs. Studies are ongoing for optimizing the dissociation process by testing the effect of different concentrations of dissociation enzymes.⁸

The results from the published single-cell transcriptomic studies on adult and fetal human kidney confirm the power of these methods to contribute to understanding a complex organ like the kidney. As an informative example, C.E. Gillies et al.¹¹ discovered 894 glomerular and 1767 tubulointerstitial expression quantitative trait loci associated with nephrotic syndrome; by integrating scRNA-Seq from kidney tissue, they could assign the quantitative trait loci to their putative cellular source. Single-cell transcriptomic analysis can identify cell type–specific expression of key regulatory molecules, including transcription factors and long non-coding RNA. Figure 1B shows the unique expression of LINC00839 in podocytes; the functional role of this molecule and its role in podocytes are not known.

It is critical that the single-cell datasets be made publicly available (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE94333, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE109205, and https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE107585).

Shared datasets will serve the research needs for understanding of renal regulatory programs and ultimately serve the broader goal of defining renal disease in molecular terms.

Disclosures

None.