Spatial Transcriptomics and the Kidney

Abstract

Purpose of review

The application of spatial transcriptomics technologies to the interrogation of kidney tissue is a burgeoning effort. These technologies share a common purpose in mapping both the expression of individual molecules and entire transcriptomic signatures of kidney cell types and structures. Such information is often superimposed upon a histologic image. The resulting datasets are readily merged with other imaging and transcriptomic techniques to establish a spatially anchored atlas of the kidney. This review provides an overview of the various spatial transcriptomic technologies and recent studies in kidney disease. Potential applications gleaned from the interrogation of other organ systems, but relative to the kidney, are also discussed.

Recent findings

Spatial transcriptomic technologies have enabled localization of whole transcriptome mRNA expression, correlation of mRNA to histology, measurement of in situ changes in expression across time, and even subcellular localization of transcripts within the kidney. These innovations continue to aid in the development of human cellular atlases of the kidney, the reclassification of disease, and the identification of important therapeutic targets.

Summary

Spatial localization of gene expression will complement our current understanding of disease derived from single cell RNA sequencing, histopathology, protein immunofluorescence, and electron microscopy. While spatial technologies continue to evolve rapidly, their importance in the localization of disease signatures is already apparent. Further efforts are required to integrate whole transcriptome and subcellular expression signatures into the individualized assessment of human kidney disease.

Introduction

Spatial anchoring is essential to define the relationship between cells and structures within a tissue. Modern spatial transcriptomics platforms enable spatial localization of whole transcriptome mRNA expression, often overlaid upon histological information from the same tissue section. Gene expression profiles are then mapped back to their original location, enabling a direct link between gene expression of neighboring cells or histologic structures. By integrating cell type definitions derived from single cell or nuclear RNA sequencing (sc/snRNAseq) to improve cell type specificity, novel data outputs can be acquired which complement traditional histopathologic assessment of the kidney. These outputs include cell neighborhood definitions, receptor ligand interactions of epithelial and immune cells, and localization of relevant downstream pathways activated in disease pathology. Other highly sensitive spatial technologies may enable high resolution subcellular localization of mRNA within the kidney’s cells. Together, these outputs facilitate the ongoing development of transcriptomic atlases of the kidney.

In this discussion, the various spatial transcriptomic technologies are first introduced, followed by an evaluation of recent examples in which spatial interrogation has helped to advance our understanding of kidney disease.

Spatial transcriptomic interrogation technologies

Spatial transcriptomic technologies can be broadly grouped into those that characterize mRNA localization in tissue at the regional, cellular, or subcellular resolution. These technologies seek to link mRNA expression with a histological address for each cell type. Recent advancements have improved strategies to extract information efficiently. Six technology types worthy of highlighting include:

i). Microdissection:

Microdissection of the kidney can be accomplished through manual dissection (1) of the nephron or slide-based laser capture of labelled structures and cells (2). Traditional slide based technologies have isolated glomeruli from the tubulointerstitium, but this technique can be adapted to acquire specific cell types from all nephron subsegments using a rapid antibody-based immunofluorescence stain protocol (3). After tissue isolation, mRNA is extracted, cDNA is synthesized, and sequencing is performed in a manner analogous to bulk tissue processing for whole transcriptome measurement. The acquired tissue can be subjected to multiple interrogation techniques including proteomics (4) or bisulfite sequencing to measure methylation of cytosine residues (5). While microdissection is a form of spatial anchoring, it suffers from a number of drawbacks. First, once the tissue is dissected, the underlying histology or immunofluorescence information is lost. Second, to acquire sufficient tissue to measure mRNA expression, multiple glomeruli or renal tubular cells must be pooled, yielding an average signature with significantly less specificity than sc/snRNAseq technologies.

ii). Fluorescence in situ hybridization (FISH):

Multiplexed FISH provides targeted expression of a set of genes at subcellular resolution with high sensitivity, even for lowly expressed transcripts. In separate rounds, probes bind to mRNA, and subsequent reporter probes bind to the encoding probes. Quantitation of fluorescence in situ hybridization is obtained by counting fluorescent signals. These techniques have been used to define erythropoietin-producing cells in the murine kidney (6) and characterize co-expression of protein and mRNA (mIFISH) in human transplant kidney biopsy specimens (7). Multiplexed FISH technologies such as MERFISH (8, 9) and seqFISH+ (10) can now reach a repertoire of 10,000 co-expressed targets within a tissue. The broad application of such nearly whole transcriptome approaches is presently limited by cost, challenges in probe design, and labor-intensive experimental processes.

iii). in situ sequencing (ISS):

ISS technologies also use fluorescence output, but in contrast to FISH, the mRNA is read and converted to cDNA nucleotide by nucleotide and then sequenced with rolling circle amplification within the tissue. Examples include fluorescent in situ sequencing (FISSEQ) (11, 12) and spatially-resolved transcript amplicon readout mapping (STARmap) (13, 14). These technologies yield subcellular resolution for over 1000 multiplexed targets; however, the sensitivity tends to be lower as compared to FISH. Greater multiplexing further reduces sensitivity and increases sequencing time. Limited data is available in the application to kidney disease at the time of this review.

iv). in situ capturing (ISC):

ISC methods, sometimes referred to as solid phase-based capture technologies, utilize tissue adhered to a slide that contains equidistant capture probes, each barcoded to allow back-mapping of the mRNA signature to its original spatial location. ISC-based technologies like Slide-seq (15) and 10x Visium (16) can reduce experimental complexity, eliminate probe design requirements, and provide whole transcriptome expression. Spatial relationships are rebuilt with barcoded localization and expression quantification. Slide-seq has near single-cell resolution because the slide area is covered by 10-micron beads. The 10x Visium platform has a larger capture area, but expression is mapped over a hematoxylin and eosin stained histological image in the same section for a streamlined experimental design. Each capture zone contains five thousand spots/probes of 55 microns diameter that overlie multiple cells. Thus, slide-seq lends itself to neighborhood analyses of juxtaposed cells, while the Visium platform requires strategies for spot deconvolution for component cells. In our experience, Visium detects approximately 2,500 unique genes per 55 μm spot and 20,000 unique genes per sample. Slide-seq’s 10 μm beads detect fewer genes per spot, but when scaled to area, the 2 techniques detect comparable numbers of genes. The overall cost for one Visium sample, including reagents, array slide, tissue preparation, cDNA synthesis, and sequencing is approximately $2,000. At present. Slide-seq’s commercial availability is limited at the time of writing.

v). Full Transcriptome Spatial RNA Analysis:

The GeoMx Human Whole Transcriptome platform (offered by NanoString) is a slide based technology for FFPE or frozen samples. Tissue is adhered to a slide and visualized on the GeoMx instrument, which contains both an ultraviolet laser and microscope. Regions of interest (ROI) are selected from the slide which then undergoes ultraviolet light dissociation and the resulting RNA from each ROI is captured in a 96-well plate (17). Each well contains barcodes allowing the RNA signature to be remapped to the ROI after cDNA synthesis and sequencing. As compared to each Visium assay, the approximate cost per sample is less and total genes detected per sample is similar (~18,000). However, the GeoMx assay generally yields fewer ROIs than Visium spots and requires supervision to select each region.

vi). in silico reconstruction:

Using the expression signature of genes derived from sc/snRNAseq technologies, spatial localization can be inferred for each cell type. The technique essentially builds a virtual tissue atlas by applying computational methods. This approach, which has been applied to kidney, includes both localization of transcripts and the relative expression of contributing cells (18).

Application of spatial transcriptomics to experimental model systems of the kidney

As described above, multiple technologies are available to localize the comprehensive mRNA expression profile of the kidney. The application of in situ capturing methods to the kidney has received considerable attention in the last year. The exploration of kidney disease model systems with spatial transcriptomic ISC technologies complements recent sc/snRNAseq endeavors, by localizing cell types and injury signatures in situ. Janosevic et al. investigated a murine endotoxemia model at seven time points, defining the timeline of sepsis progression with single-cell RNA sequencing (19). Using pseudotime and velocity field analysis, the authors identify time-dependent phenotypical changes in epithelial and macrophages populations, as well as changes in receptor-ligand interactions, and the activity of genes associated with recovery in later time points. Using spatial transcriptomics, they localize the novel proximal tubule S3-Type 2 (identified in scRNAseq) to the outer medulla of the murine kidney (20).

A septic murine model was also investigated by Melo Ferreira et al. where spatial transcriptomics localized the septic injury in the murine kidney and, using a single cell reference dataset, described differences in the immune cell population when compared to a sham kidney (21). Since each capture spot of the 10X Visium platform covers multiple cells, the signature of immune cell infiltration may be disguised by the dominant transcriptomic signature of the more abundant epithelial cells. In order to uncover the subtle signature of infiltrating immune cells, the authors suppressed epithelial cell signatures from a reference scRNAseq dataset, before mapping it to the spatial transcriptomic sample. The authors applied the same methodology to an Ischemia-Reperfusion injury (IRI) model, identifying a subpopulation of proximal tubules involved in neutrophil chemotaxis localized in the outer medulla of the IRI murine model. The immune cell localization was validated for each model with codetection by indexing (CODEX) immunofluorescence.

Subsequently, the localization of sex specific characteristics of the IRI murine model were identified (22). Specifically, the authors found female mice were relatively protected from IRI with a comparable degree of injury in a 34-minute IRI model as compared to a 22-minute model in their male counterparts. Kidneys were harvested at multiple time points to define the progression of injury. The authors utilized SPOTlight and Giotto to deconvolute the cell distribution in their spatial transcriptomics samples, describing differentially expressed genes and cell population changes along the course of IRI.

Using a combination of scRNAseq and spatial transcriptomics in a murine embryonic model, Sanchez-Ferras et al. described in great detail the development of the Nephric Duct (23). Using scRNAseq, the authors defined four unique cell populations and their contribution to the development of the Nephric Duct. They describe the contribution of Gata3 and Tfap2a/2b genes as regulators of morphogenesis. In this work, spatial transcriptomics was used to validate the spatial segregation of the four Nephric duct progenitor populations.

Building a human kidney atlas

ICS-based spatial transcriptomic technologies may prove an important tool in the construction of a human kidney atlas in health and disease. The molecular interrogation of human kidney biopsy specimens has garnered considerable attention from consortia (24–26). Many spatial transcriptomic technologies are slide based and can be performed in the existing workflow of kidney biopsy processing of paraffin-embedded or Optimal Cutting Temperature (OCT) frozen blocks. Conceivably, a CLIA-validated version of a future spatial transcriptomic assay could one day supplement the existing assays a nephropathologist uses to interpret a biopsy specimen. An early application of 10X Visium spatial transcriptomics, interrogation of a healthy human nephrectomy revealed the ability to map most cell types derived from the Kidney Precision Medicine Project sc/snRNAseq atlas (21, 27). A strong alignment between the mapped spatial transcriptomic defined cell types was observed with the underlying histology, showing better than 90% alignment in all cell types.

Another ICS-based technology, Slide-seqV2, was similarly used to explore spatial relationships in the human kidney (28). The technology was applied to cortical and medullary sections of renal nephrectomies to validate appropriate cell type localization. The authors then apply Slide-seqV2 in a diabetic murine model, detecting the expansion of the juxta-glomerular apparatus and an increase in podocytes injury markers as compared to a sham rodent (28). In a UMOD-C125R knockin kidney disease model, the authors found evidence that thick ascending limb, fibroblast, and macrophage cells contribute to disease-specific neighborhoods in the medulla of the injured mice, identifying injury-related pathways in those regions that could be missed in single-cell technologies where spatial anchoring is missing.

Both of these ICS-based spatial transcriptomic technologies were adapted by Lake et al. to create a comprehensive human atlas of the kidney in health and disease (29). With over 80 kidney samples, the authors apply single-cell and single-nuclei sequencing, combined with single-nucleus chromatin accessibility and mRNA expression sequencing (SNARE-seq) (30), to characterize and uncover over 100 cell populations. Several of these populations were subtypes of common cells which were in putative injury cell states referred to as adaptative and degenerative cell types. Visium and SlideSEQ were combined to localize these novel cell populations and to establish niches of populations associated with altered cell states. 3D tissue cytometry further defined those niches and better characterized the immune cell components of the neighborhoods. With the combination of SNARE-seq and sc/snRNAseq, the authors then define trajectories for the altered cell states, which were associated with disease progression.

The link between immune cell infiltration and altered epithelial expression was also assayed using the GeoMx platform (17). In this study, the authors identified differentially expressed genes in five regions of interest in a kidney allograft with T-cell mediated rejection, and compared expression to two healthy control biopsies, demonstrating feasibility of the technique in kidney tissue.

Of considerable importance is the link between the human kidney and the mouse in order to better understand the translatable elements of common disease models. To this end, Raghubar et al. describe the species differences between mammalian kidney in mice and humans (31). The study converts human genes to murine orthologs and describes the observed differentially expressed genes. This analysis was performed in aggregate pseudo-bulk samples, without a direct comparison of individual renal structures or cell types. In the human sample, glomerular spots were defined by the expression of common marker genes and validated with immunofluorescence. The authors describe receptor-ligand pairs co-expressed within and between glomerular spots, describing over 100 potential receptor-ligand interactions. Finally, spatial transcriptomics was explored as a tool to localize disease-related single nucleotide polymorphisms.

Deconvolution

Due to the larger diameter of barcoded capture beads in certain in situ capture technologies, the cell type classification may not achieve single cell resolution currently available sc/snRNAseq datasets. Unsupervised classification of the beads often results in broad cell types, while underrepresented cell type subpopulations might not be captured with significant frequency to be correctly defined. To overcome this shortcoming, a common strategy applied to spatial transcriptomics is the use of a reference single-cell or single-nuclei dataset to inform the classification of the spatial spots or beads (32). An example is provided in Figure 1. Studies using Slide-seqV2(28, 29) have successfully applied Robust Cell Type Decomposition (RCTD) (33) and Seurat (34, 35) for deconvolution. RCTD deconvolutes by first calculating the mean expression of each cell type in the reference dataset. It then fits a statistical model where the expression of each gene is a Poisson distribution of a linear combination of the mean expression profiles. Seurat, on the other hand, projects both datasets to a common dimensionally reduced space and finds pairs of elements from both datasets sharing nearest neighbors. These pairs are called anchors and are then used to transfer scores associated with each cell type label from the reference sc/snRNAseq dataset to each bead.

Figure 1: Cell type mapping in spatial transcriptomics.

A) H+E image with a glomerulus and a transition from the distal tubule (DCT) into a collecting duct (CD). B) Podocin (NPHS2) is expressed in the glomerulus. The thiazide sensitive sodium chloride cotransporter (SLC12A3) expression decreases from right to left as the DCT transitions to CD. C) Schematic indicating that transfer scores are calculated for >100 snRNAseq clusters based on all genes expressed and then mapped onto the spatial transcriptomics sample in Seurat version 3, deconvoluting the proportion of expression in each spot corresponding to a snRNAseq cluster. D) Glomerular spots contain signature from podocytes (POD) and glomerular endothelial cells (EC-GC). Damaged POD (dPOD = cell state) signature is also seen. From right to left, a gradual transition is seen from DCT and connecting tubule (CNT) signatures to CD intercalated (MIC-A) and principal cell (tPC-IC) signatures.

Both technologies can be readily applied to Visium, but due to its spot size, multiple cell types are expected in each spot. Therefore, multiple deconvolution strategies have been designed. To uncover the transcriptomic signal of immune cells underlying the more abundant epithelial cell types, Seurat V3 can be used to map a subset of the reference dataset with immune cells and fibroblasts (21). Such an approach allowed the authors to localize immune cells, confirmed with immunofluorescence. In a later publication, the methodology is expanded to deconvolute each spot, where the contribution of each cell type in the reference is proportional to the transfer score obtained from Seurat (29). This approach helped to uncover niches of immune cells surrounding injured epithelial tissues. This cell type deconvolution strategy showed results consistent with the associated histological image and was cross-validated with the results obtained in their previous work.

Designed to deconvolute Visium spatial transcriptomics, SPOTlight (36) has been applied to kidney samples (20, 30). This method defines cell type topics, with the distribution of gene expression defining that cell type, in the reference dataset. Weights are then used to build each individual spot signature from the cell type topics. These weights can be interpreted as the proportion of each cell type captured by the spot. Other deconvolution methods have been proposed, such as Giotto (37), deconvoluting spatial transcriptomics (DSTG)(38), Tangram (14), or stereoscope (16), but applications in the kidney have yet to be examined.

Applications from other organ systems relevant to the kidney

To overcome the limitations of current spatial technologies, Chen et al. united in situ sequencing and in situ capture technologies to study brains of Alzheimer’s disease murine models (39). The authors combined immunofluorescence in sequential sections to the in situ capture technology. This strategy allowed identification of groups of genes related to amyloid plaque accumulation. When applied to the kidney, this approach could allow investigation of the relation between the whole transcriptome in the context of protein expression localization. Applications could include a link between extracellular matrix protein expression to neighboring cellular transcriptomic response in order to better define in situ pro-fibrotic processes or better characterize any immune cell infiltration which contributes to this process.

Using human embryonic hearts in four different developmental stages, Asp et al. describe the formation of multiple spatial and genetic patterns, uncovering new localized cell types (40). The Asp approach has particular relevance to the kidney given its very rich spatial structure and the unique dynamics of nephron morphogenesis. The approach could expand upon the Nephric duct developmental progression discussed above (23).

Another application of spatial transcriptomics described the structure of human white adipose tissue (41). The authors measure the propensity of cell types in close approximation to each other. The characterized white adipose tissue was found to be more organized than previously understood. The structure of the nephron is well described, but this method is complementary to the one presented by Lake et al., and could be applied to define the localization of cell states and injured cell types (29).

Lessons learned from the tumor microenvironment

A key advantage of spatial transcriptomics is the ability to define regions based on changes in expression signatures, even when the histology does not coincide with overt signs of disease. The tumor microenvironment is known to be heterogeneous with complex variables defining tumorigenesis, metastasis, and drug resistance (42). To better understand this, spatial transcriptomics has been applied to the tumor microenvironment to extract information about its architecture and tumor margins. Based on early studies, the tumor microenvironment and its near margins are composed of specialized tumor and muscle cells (43) and the margin is a transcriptionally distinct region from the tumor itself or tissue more distant from tumor margins. Analogously, spatial transcriptomics could be applied to seemingly normal histologic regions of human kidney biopsy specimens which border a patchy distribution of acute tubular necrosis or tubulointerstitial fibrosis and atrophy, in order to better understand the adaptive mechanisms of the renal penumbra adjacent to injury.

Conclusion

In the last two years, the application of spatial transcriptomic technologies to the kidney has exploded, with important insights gleaned from disease models and the creation of a spatially anchored human kidney atlas. The outputs of these technologies may one day complement the interpretation of kidney biopsy specimens by nephropathologists. Nonetheless, spatial transcriptomic techniques are constantly evolving and our understanding of cell-cell neighborhoods and localization of cell states is still a nascent endeavor. The sheer quantity of recent high impact studies reviewed here portends a vibrant landscape lies beyond the bleeding edge. Let’s go chart that landscape.

Keypoints.

1. Multiple spatial transcriptomic technologies have been developed to localize mRNA expression signatures. The utility and application of these technologies is constantly evolving.

2. Spatial transcriptomics has been applied to the kidney to better understand the pathogenesis underlying disease models and to create of a spatially anchored human kidney atlas.

3. Insights from other organ systems and the tumor microenvironment provide strategies which can be applied to define the distribution of injury cell states in the kidney.