Single-cell transcriptomics: a novel precision medicine technique in nephrology

Sample preparation for single-cell analysis

There are several experimental considerations for scRNAseq such as tissue preservation, tissue dissociation, and cell count.

Obtaining high-quality single-cell suspensions is a key determinant of the success of single-cell studies. The single-cell preparation process is the most frequent cause of unwanted technical variation and batch effects in single-cell studies. To obtain a single-cell suspension from kidney samples, fresh tissue is dissociated by mechanical disruption and enzymatic digestion. An automatic tissue dissociator is typically used to minimize batch effects. Because different tissues have different characteristics, the protocol must be optimized for each purpose. Excessive tissue dissociation causes cell damage and reduces viability, which can result in unwanted transcriptional changes, ambient mRNAs, and greater amounts of mitochondrial mRNAs [2,3]. In contrast, insufficient tissue dissociation leads to multiplets in the data. In addition, a single-cell suspension obtained by tissue dissociation should be filtered through a cell strainer of appropriate size or immersed in cell debris removal solution. Cell counting is also critical because use of an excessive cell density can result in capture of multiple cells simultaneously. In contrast, underloading can cause loss of information due to empty droplets. Multiplets can be removed using computational tools such as Scrublet and DoubletFinder [4,5]. It is also important to determine the appropriate number of cells to be analyzed, which requires consideration of sample heterogeneity and abundance of the target cell type. A large number of cells must be sequenced to analyze small populations of renal cells (20,000 cells are required to obtain 100 target cells if the target cell type frequency is 0.5%), as approximately 60% of the kidney is composed of proximal tubule cells [6].

scRNA-seq analysis is hindered by difficulties in isolating cells or if cells are damaged. Single nucleus RNA sequencing (snRNA-seq) overcomes these limitations by isolating a single nucleus rather than a single-cell [7,8]. Although scRNA-seq requires fresh tissue, snRNA-seq can be performed on frozen tissue, and can be applied to tissues that are difficult to separate because of intertwined cells [8]. Furthermore, snRNA-seq reduces cell stress and the composition bias generated during the separation step. Nevertheless, information on cytoplasmic RNA cannot be obtained because only nuclear RNAs are analyzed, and various intron sequences are present. Therefore, the method used should be selected based on the aim of the experiment.

Single RNA sequencing technologies

scRNA-seq methods typically involve RNA capture, reverse transcription, complementary DNA (cDNA) amplification, sequencing library construction, and high-throughput sequencing.

scRNA-seq is mainly divided into full-length scRNA-seq and tag-based scRNA-seq according to library construction method. Full-length scRNA-seq can be utilized not only to measure gene expression levels but also to identify transcript isoforms, alternative splicing, and single-nucleotide polymorphisms within the transcripts [9–12]. It has high sequencing coverage and mapping efficiency; however, it has limited cell throughput (hundreds of cells), a relatively large batch effect, considerable sample preparation time, and high cost per cell because the samples must be prepared independently [9,13–15]. Smart-seq2 and Quartz-seq are representative full-length scRNA-seq methods [16,17]. The tag-based scRNA-seq technique estimates transcript abundance by sequencing the 3′-end of transcripts in a large number of cells (tens of thousands to millions of cells) [9,13,18,19]. To distinguish cell types and to determine the transcript copy number accurately, cDNA molecules are labeled with barcode sequences such as cell barcodes and unique molecular identifiers (UMIs). Tag-based scRNA-seq is subdivided into droplet-based [19–21], microwell-based [22,23], and split-pool barcoding-based technologies [24,25], according to the labeling method. Droplet-based technology uses a microfluidic device to generate aqueous droplets formed by combining water and oil streams. In each droplet, one cell and one barcoded bead pair are encapsulated, and each single-cell mRNA is captured by oligo-dT of the barcoded bead. During cDNA synthesis, cell barcodes and UMIs are added to the cDNAs [19–21]. Although droplet-based barcoding generates limited information on the 3′-end of the mRNA, it improves the throughput of single-cell analysis and reduces the time, labor, cost per cell, and batch effect by simplifying the experimental procedure. Microwell-based technology captures mRNA by loading cells on a microwell plate, washing doublets with capillaries, and adding barcoded beads [22,23]; it is simple and economical. Split-pool barcode methods identify a single-cell by combinatorial indexing without the need for separation to obtain a single-cell. Single-cell combinatorial indexing RNA sequencing (sci-RNA-seq) and split-pool ligation-based transcriptome sequencing (SPLiT-seq) are typical split-pool barcoding methods [24,25]. These techniques distribute permeabilized cells across 96- or 384-well plates; the first molecular index is introduced to the mRNA of cells in each well, with in situ reverse transcription. After the first barcoding, the cells are pooled, and the cells are distributed across another plate. Thereafter, a second barcode is added, resulting in a unique combination of barcodes for each cell [24,25].

Another challenge of scRNA-seq is the loss of cell location and orientation information during tissue dissociation. Much effort has focused on recreating a near-realistic cell environment. Because cells exist in a three-dimensional (3D) space and interact within that space, spatial transcriptome analysis was developed to evaluate cell types and their locations. Single-molecule fluorescence in situ hybridization (smFISH) was used to identify RNA location and copy number; however, it cannot analyze multiple cells simultaneously [26,27]. To overcome this limitation, multiplexed error robust FISH (MERFISH), which involves labeling of multiple RNAs in a single-cell, and sequential-FISH (seq-FISH), which involves multiple rounds of FISH, have been developed [28,29]. In addition, spatially resolved transcript amplicon readout mapping, which labels RNA with a DNA probe, and fluorescent in situ sequencing, which overcomes the shortcomings of existing padlock probes, are available [30,31]. Visium, a spatial transcriptomics technology from 10× Genomics, provides a multidimensional view of tissue biology by means of high-throughput mRNA analysis of intact tissue sections [32]. Computing technologies such as Seurat, DistMap, and novoSpaRc are also evolving [33–35], and efforts are focused on understanding the functions of cells in 3D space.

scRNA-seq has several limitations that need to be overcome. For example, dissociation may cause stress or alter the cell proportion [2,36]. Also, the batch effect may result from use of different protocols, sample handling, and platforms [37,38]. In addition, the transcript efficiency per cell is lower compared to bulk RNA-seq [39,40]. Finally, dropout occurs with droplet-based scRNA-seq [20,21,41].

Computational analysis of scRNA-seq data

The raw data generated by sequencing are processed using analysis pipelines such as Cell Ranger [21], SEQC [42], and zUMIs [43] to a gene-by-cell data matrix.

Quality control filters select only high-quality cells, facilitating identification of cell type, and screening of failed samples so that their data can be recovered or removed from analysis. This requires generation of quality metrics such as the number of UMIs per cell, number of genes detected per cell, and proportion of mitochondrial genes [44–46].

Normalization is essential for accurate comparison of gene expression between samples. Because the gene expression count depth of the same cells can vary due to the diversity inherent in each step—such as single-cell capture, reverse transcription, and sequencing—the gene expression counts are scaled by the total number of sequencing reads or counts per cell [45,47,48].

After normalization, clustering analysis is performed to separate the cells based on gene expression patterns and identify the cell types. The Seurat package allocates cells to clusters based on the principal component scores obtained from the expression of the most variable genes [49,50]. For visualization of the cell clusters, t-distributed stochastic neighbor embedding and uniform manifold approximation and projection are typically performed [51,52]. The identified clusters are then assigned into known cell types based on cell type-specific markers or by automatic cell assignment software. There are several cell-marker databases for cell type classification, such as Cell Finder [53], CellMarker [54], and PanglaoDB [55]. Despite marker-based cell type classification being commonly used, these methods cannot distinguish other cell types that express the same marker from the target cell type, and can complicate cell type classificaition by the heterogeneity of cell states. The various automated cluster annotation methods combine annotation and clustering, and the automatic cell identification methods are compared elsewhere [56]. Downstream analyses—such as trajectory analysis, differential expression analysis, gene set analysis, and gene regulatory networks—have been reviewed by others [45,47,57].

Integration of single-cell transcriptomics with other -omics data

Single-cell sequencing enables simultaneous analysis of the genome, epigenome, transcriptome, and proteome of one cell. By integrating genomic and transcriptomic information, it is possible to confirm the effect of DNA copy number variation on gene expression, transcript changes according to genomic changes, and the effect of mutations in the coding or non-coding region on transcript expression [58]. Genome and transcriptome sequencing (G&T-seq) combines whole-genome amplification and Smart-seq2 [59]. Single-cell methylome and transcriptome (scM and T-seq) was developed to analyze the relationship between DNA methylation and the transcriptome in a single-cell [60,61].

Moreover, technologies that integrate chromatin information and the transcriptome are now available. Single-cell combinatorial indexing-chromatin accessibility and mRNA (sci-CAR) integrates single-cell transcriptome analysis (sci-RNA-seq) and epigenetic analysis (sci-ATAC-seq) [62]. This allows assessment of the relationships between differentially expressed genes and their regulatory chromatin regions.

RNA and protein determine the properties of a biological system; however, because of the distinct half-lives of mRNA and protein, and the effects of post-transcriptional modification, it is difficult to evaluate the correlation between mRNA and protein levels [58]. Simultaneous identification of the transcriptome and proteome at the single-cell level has enabled exploration of RNA and protein abundance. Cellular indexing of transcriptomes and epitope sequencing (CITE-seq) and RNA expression and protein sequencing (REAP-seq) enables simultaneous analysis of proteins and transcripts at the single-cell level using oligonucleotide-labeled antibodies, and can detect proteins barcoded into multiple antibodies and more than 20,000 genes [63,64].

Several techniques unaffected by cellular heterogeneity have been developed to overcome the limitations of single transcriptome studies. Single-cell multiomics techniques can distinguish subtypes of cells from heterogeneous cell populations [65–67]. In addition, the accuracy of lineage trajectory analysis can be enhanced by integrating gene expression and the epigenetic changes that occur during cell division and delivery to daughter cells [68]. Single-cell multiomics data can reveal correlations between different -omics data, and presumably, may effectively reveal the causal relationship between -omics and technology development.

Kidney single-cell atlas

Several scRNA-seq studies of human and mouse diseased and normal kidneys have been conducted; the results were applied to produce a variety of databases (Table 1).

Table 1

Single-cell analysis databases

The creation of a single-cell atlas offers several advantages. First, by typing differentiated cells, their stage of development and differentiation can be determined. Second, it provides insight into the mechanisms of disease development and progression. By analyzing diseased tissue at the single-cell level and comparing it with the cell atlas of healthy tissue, it is possible to evaluate the heterogeneity of cells in the diseased tissue and so determine the cause of the disease. Analyzing single cells in diseased tissues can identify the factors that contribute to differences between individuals, allowing personalized treatment and facilitating the development of new drugs and biomarkers.

Park et al. [6] constructed a cell atlas using scRNA-seq data of 57,979 cells from healthy mouse kidneys. They identified 18 kidney epithelial and immune cell types as well as novel transitional cell types located between intercalated cells (ICs) and principal cells (PCs) in the collecting ducts. Moreover, they revealed that Notch signaling is responsible for the transition of ICs to PCs. Karaiskos et al. [69] performed single-cell profiling of mouse glomerulus cells. They identified glomerular cell types such as podocytes, mesangial cells, and endothelial cells. They also revealed novel marker genes for all glomerular cell types and assessed the transcriptional heterogeneity of each cell type via subclustering the endothelial cells and podocytes. Chen et al. [70] performed scRNA-seq on mouse collecting duct cells. Ransick et al. [71] anatomically dissected male and female kidneys for anatomy-guided scRNA-seq. They confirmed sexual, spatial, and temporal diversity in nephrons and the collecting system.

Liao et al. [72] identified 10 normal human cell clusters via scRNA-seq of 23,366 kidney cells from three human donors. Proximal tubule cells and collecting duct cells were classified into three and two subtypes, respectively. For gene expression profiling of human fetal kidney development, Wang et al. [73] applied scRNA-seq to 3,543 renal cells spanning several embryonic stages and classified the major cell types. Moreover, they identified two subpopulations in the cap mesenchyme and assessed the molecular heterogeneity of the cap mesenchyme. Furthermore, they identified the transcription factors and signaling pathways involved in nephron tubule segmentation during fetal kidney development. Wu et al. [74] performed scRNA-seq of human kidney allograft biopsy samples and confirmed the proinflammatory response of allograft rejection by comparison with healthy kidney epithelial transcriptomes. Understanding the characteristics and origins of organ-specific tumors is crucial for making treatment decisions. Young et al. [75] performed single-cell profiling of human renal tumors and normal tissue from pediatric and adult kidneys. They confirmed that Wilms tumor, a pediatric kidney cancer, was derived from abnormal fetal cells. The origin of the tumor was predicted by matching the transcriptome of adult kidney cancer to a specific subtype of proximal convoluted tubular cells. Lake et al. [7] optimized the snRNA-seq pipeline for clinical specimens to define the molecular transition states of more than 10 nephron segments in the proximal tubules and collecting ducts. Using the results they described the anatomical nephron organization, providing a starting point for building a molecular and physiological atlas that can be used to identify variations in kidney diseases. Stewart et al. [76] evaluated the spatiotemporal immune topology in human kidneys by scRNA-seq. scRNA-seq of mature and fetal kidneys revealed the asymmetric distribution of immune cells, and the contribution of each cell type based on the nephrogenesis stage was analyzed. This kidney immune cell atlas provides insight into the pathogenetic mechanisms and therapeutic targets in immune and infectious kidney diseases.

Single-cell studies of human and mouse kidneys have shown the complexity of kidney tissues; however, some scRNA-seq studies missed known cell subtypes. This is because sample preparation, dissociation, and batch effects influence the detection of rare or sensitive cells. A multiplex approach to droplet snRNA-seq (Mux-seq) was applied to minimize batch variation [77]. This allows scRNA-seq of several human biopsies simultaneously, enabling identification of kidney cell populations in health and disease.

Single-cell analysis of kidney diseases

scRNA-seq of the kidneys of patients with lupus nephritis (LN) has been reported. Der et al. [78] reported that the type 1 interferon response in tubular epithelial cells was greater than that in healthy controls, and confirmed clinically relevant disease signatures in kidney and skin biopsies. Arazi et al. [79] performed scRNA-seq of the kidney tissues of patients with LN and healthy controls, and found that leukocytes in the kidney were active and that their activation status differed before and after the inflammatory reaction. Subgroups of innate and adaptive immune cells expressing various transcription factors related to systemic lupus erythematous were identified by analyzing cluster-specific expression of risk-related genes in genome-wide association studies. scRNA-seq of LN tissues confirmed the molecular signature associated with prognosis, enabling improvement of patient care and stratification.

Transcriptome profiling of kidney tissue or isolated glomeruli provides insight into the pathogenesis of kidney fibrosis. Wilson et al. [80] performed snRNA-seq using 23,980 nuclei from control and diabetic kidney cortex samples, and found gene expression changes in diabetic glomeruli, mesangial cells, endothelial cells, and diabetic proximal convoluted tubules and ascending limbs. In addition, differential expression of ligand-receptors was observed in the various glomerular cell types. Differential expression analysis of leukocytes revealed that infiltrating immune cells contribute to the synthesis of kidney risk inflammatory signature markers. Such gene expression changes may enable identification of biomarkers and signaling pathways early in diabetic nephropathy. To evaluate the mechanism of early diabetic kidney disease, Fu et al. [81] performed scRNA-seq of kidney glomerular cells in a diabetic mouse model. They identified five distinct cell clusters and novel glomerular cell-specific markers. Comparison of the scRNA-seq data between diabetic and normal mouse kidneys revealed identical cell clusters, but the immune cell population was increased in diabetic mice. Moreover, there were remarkable gene expression changes in endothelial and mesangial cells in the diabetic mouse kidney. These analyses identified key factors involved in diabetic kidney disease progression and thereby assisted development of new treatment approaches [81]. Using the reversible unilateral ureteric obstruction model (R-UUO), changes in renal injury and repair at the single-cell level were noted [82]. scRNA-seq revealed myeloid cell heterogeneity in damaged and recovering kidneys by identifying new monocyte and macrophage subsets not observed in the kidney [82]. The data will enable identification of potential therapeutic targets and the development of therapeutics for kidney diseases.

Intratumoral heterogeneity interferes with marker-based cancer treatment because targeted therapy removes only a specific population of tumor cells. To evaluate the molecular and cellular heterogeneity in renal cell carcinoma, Kim et al. [83] performed transcriptome profiling of primary and metastatic renal cell carcinoma with single-cell resolution. Metastatic cancer cells exhibited distinct gene expression patterns with increased metastatic and aggressive signatures compared to primary cancer cells. Based on transcriptome profiling and drug screening, the drug sensitivity and activation status of signaling pathways were predicted, and the correlation between the predicted signature and measured data was verified. sci-RNA-seq allows identification of cell subsets with active signaling pathways and selection of the most effective drug combination. Such approaches will overcome the intratumoral heterogeneity that hampers precision medicine. The immune cells of the tumor microenvironment (TME) are crucial in determining the response to cancer immunotherapy. Nevertheless, the role of immune cells in clear cell renal cell carcinoma (ccRCC) is unclear, and most patients do not respond to these treatments. Vishwakarma et al. [84] applied scRNA-seq to the tumors and peripheral blood immune cells from patients with ccRCC to characterize the TME. They identified several intratumoral CD8 T cell states that characterize the effector, memory, and exhausted subpopulations together with multiple states of tumor-associated macrophages and dendritic cells. Moreover, they found intratumoral cytotoxic and regulatory CD4 T cell clusters, which led to the establishment of tumor-infiltrating effectors and memory cells. These results will facilitate research on TME to identify new therapeutic targets and biomarkers.