#NephJC Chat
Tuesday, December 15, at 9 pm Eastern
Wednesday, December 16, at 9 pm Indian Standard Time
Wednesday, December 16, at 9 pm GMT
Nature 2020 Nov 11. doi: 10.1038/s41586-020-2941-1. Online ahead of print
Decoding myofibroblast origins in human kidney fibrosis
Christoph Kuppe, Mahmoud M Ibrahim, Jennifer Kranz, Xiaoting Zhang, Susanne Ziegler, Javier Perales-Patón, Jitske Jansen, Katharina C Reimer, James R Smith, Ross Dobie, John R Wilson-Kanamari, Maurice Halder, Yaoxian Xu, Nazanin Kabgani, Nadine Kaesler, Martin Klaus, Lukas Gernhold, Victor G Puelles, Tobias B Huber, Peter Boor, Sylvia Menzel, Remco M Hoogenboezem, Eric M J Bindels, Joachim Steffens, Jürgen Floege, Rebekka K Schneider, Julio Saez-Rodriguez, Neil C Henderson, Rafael Kramann
PMID: 33176333
Introduction
Two years ago, the kidney took the stage in the cutting-edge and rapidly evolving arena of single-cell RNA sequencing (scRNA-seq), a technology that captures gene expression profiles of individual cells within a larger population of cells. Adapting a metaphor from a colleague: let’s pretend the kidney is the pictured acai bowl (top image). Older transcriptomic methods capture gene expression data resembling a smoothie blended from that bowl, whereas scRNA-seq gives you information on what all the individual components of the bowl look like (bottom image).
The ability to examine each individual component, or cell, allows us to study the kidney at a previously impossible high resolution.
The kidney is a wonderfully complex organ consisting of many cell types that work in concert to execute life-sustaining tasks;` a podocyte is not a resident macrophage, which is not an intercalated cell or endothelial cell. In addition, within a cell type, there are subpopulations with different phenotypes, as well as cells transitioning from one stage to another during development, injury, and repair. Therefore, to address the mechanistic underpinnings driving kidney disease, we must not only interrogate perturbations in molecular pathways and but also delineate which cell types are undergoing and/or driving those perturbations. This type of cell-specific resolution is critical in both understanding mechanisms of disease and identifying the appropriate cell types for therapeutic targeting.
In the context of chronic kidney disease (CKD), understanding the molecular drivers of fibrosis is the key to developing novel preventative and therapeutic strategies. Kidney fibrosis, the final common pathway of different causes of kidney injury, occurs when myofibroblasts are activated, which is accompanied by an accumulation of extracellular matrix (ECM) and loss of tissue-specific cell lineages. Pathologic deposition of ECM has deleterious functional consequences such as displacing proximal and distal tubules, compromising the filtration barrier by expanding the glomerular mesangium, and altering hemodynamics of the microcirculation. Therefore, targeting ECM production is a logical strategy to prevent scarring of kidney tissue.
To do this, we must understand (1) which cell types produce ECM, (2) whether and how those cell populations are expanded in disease, and (3) the regulatory landscape of ECM production from those causal cell types. The cellular origin of myofibroblasts has long been debated, which has hindered this progression in understanding. Fibroblasts, epithelial cells, myeloid cells, pericytes, and endothelial cells have all been proposed as direct precursors to myofibroblasts.
This week, we will be discussing the recent groundbreaking study from Dr. Rafael Kramann’s team that tackles these challenges and offers new insights into myofibroblast pathobiology in the kidney. In “Decoding myofibroblast origins in human kidney fibrosis,” published in Nature last month, Kuppe et al. reported their high-resolution single cell map of the tubulointerstitium of healthy and fibrotic human kidneys. They generated scRNA-seq data from kidney samples of patients with and without CKD and validated key findings in mouse models of kidney fibrosis. Their work identified cell populations in the tubulointerstitium that produce the vast majority of ECM in fibrosis and regulatory pathways of ECM-producing cells, allowing them to propose a working model. Their efforts also resulted in the identification of a novel potential therapeutic target for reducing kidney fibrosis.
A final note before we dive into Kuppe’s study: we last visited the world of scRNA-seq during our December 2018 NephJC discussion of two seminal kidney science papers that leveraged the technology. Since then, many more studies on single-cell kidney biology have contributed essential knowledge to the field, and other molecular biology tools are being developed and optimized to facilitate multi-level omics on the single-cell level. Single-cell technologies are here to stay and are worth understanding, so we are breaking down the details of the methods used in Kuppe’s study so that hopefully everyone in the NephJC community will be able to follow not only this week’s discussion but also future single-cell studies in kidney science.
For a brief overview of scRNA-seq technology and the type of data typically presented from it, check out this NephJC video and short blog. For a more detailed introduction to scRNA-seq, check out this review and this best practices paper.
For ease of reference, a table of terms and abbreviations used in the study:
Here are several scRNA-seq web-based resources to check out if you would like to browse content or explore a specific gene of interest:
Broad Institute Single Cell Portal
Denby Lab Gene Atlas of UUO Mouse Model
Human Cell Atlas Data Portal
Humphreys Lab KIT Kidney Interactive Transcriptomics
Kidney Cell Atlas Interactive Viewer of Adult and Fetal Kidneys
ReBuilding a Kidney Gene Expression Databases
ReBuilding a Kidney scRNA-Seq Visualizations
Susztak Lab Mouse Kidney Single Cell Atlas
University of Michigan NephroCell
The Study
Methods
Before we dive into the results, let’s review a few key methodologies used in this paper. In addition to performing scRNA-seq, the authors employed well established molecular biology methods such as in situ hybridization, and those will be briefly described in the context of the presented results. Here we will present in broad strokes the approach to the single cell experiments. Outlining all of the steps in scRNA-seq is beyond the scope of this summary.
Single-cell RNA sequence analyses
The kidney comprises numerous cell types– including epithelial, mesenchymal, endothelial and immune cells (see Figure 1 in the Results section). Whereas bulk sequencing can mask heterogeneity between and within cell classes, single-cell sequencing (scRNA-seq) is a powerful methodology that can provide granular information on each individual cell. ScRNA-seq enables us to not only capture different cell states within a tissue at one time point, but we can apply machine learning to the cross-sectional information and actually reconstruct how the cellular landscape has evolved over time and space.
The authors performed scRNA-seq of:
Human kidney samples: In total, there were 11 healthy tissue donors and 10 donors with CKD secondary to hypertension. Because proximal tubule cells represent the vast majority of the tubulointerstitium, this cell type accordingly dominates datasets when scRNA-seq is performed on an unselected sample. To enrich for other cell types of interest, the authors sorted cells by whether they bear a cell surface marker found in the proximal tubule (CD10) using FACS and only sequenced a fraction of CD10+ cells. As a result, 61% of the cells sequenced were CD10-, indicating an enrichment for mesenchymal cells compared to the <20% yield in unbiased studies.
Mouse kidney samples: Later in their analyses, the authors sequenced a subset of mouse kidney cells (Pdgfrb+/Pdgfra+) in order to validate their hypothesis that these cells mediate fibrosis post-injury. The samples had been collected from three mice following unilateral ureteral obstruction (UUO), a type of experimental acute kidney injury (see procedural video here). These mice were genetically engineered to produce a fluorescent red protein in cells upon expression of Pdgfrb (Pdgfrb-CreER-tdTomato).
The authors then conducted various computational analyses to learn about the cell types represented in their scRNA-seq dataset. Here is a quick rundown of how and why each type of analysis was conducted:
Differential gene expression analyses: The authors compared gene expression among single cells of their dataset to learn about gene expression programs that define and distinguish cellular phenotypes. For example, they assigned an ECM score to each cell based on the quantity of intracellular mRNA coding for fibrous matrix proteins present, and from this, they were able to compare the degree of ECM production between different cell types. As is standard for sequencing data, the mRNA expression levels are normalized before comparison.
Data visualization using dimension reduction: Dimension reduction means taking a large dataset with many dimensions (e.g., thousands of cells each defined by thousands of gene expression values) and projecting it into 2-D space based on differences within the dataset. Principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), UMAP, and diffusion mapping are commonly used dimension reduction techniques that differ in how they characterize these “differences.” Of these, the authors use UMAP and diffusion mapping. UMAP is used in every figure of the manuscript to visualize clustering relationships between cell populations based on differences in their mRNA expression. The UMAP dimension reduction technique is relatively new (2018) and is generally considered to more accurately represent data compared to t-SNE.
Pseudotime trajectory reconstruction: The authors use diffusion mapping in lineage tree inference, which arranges cells in 2-D space according to the most likely cellular evolution process over time.
Examples of UMAP (left) and diffusion map (right) embedding from main text Figure 2. As shown here, labelling clusters by cell type and colouring data points by scaled gene expression provides further information regarding the differences between cellular subpopulations.
Results
To gain deep mechanistic insight into kidney fibrogenesis, the authors addressed the following questions:
What does human CKD look like in single-cell resolution?
First, the authors captured a high-resolution landscape of CKD by constructing a single-cell atlas of human kidney fibrosis. Because tubulointerstitial fibrosis is a hallmark of different types of CKD, the authors focused their efforts on the tubulointerstitium. In prior scRNA-seq studies, some cell populations in the kidney cortex were likely masked by the large number of proximal tubule cells in the cortex. To enrich for non-proximal tubule populations, as mentioned in Methods, the authors split the kidney cells into a non proximal tubule group (CD10-) and a proximal tubule group (CD10+) group in order to remove the overwhelming number of proximal tubule cells so other cell types could be studied in greater detail.
This left us with 53,672 non-proximal tubule cells (CD10-) from both patients with and without CKD to be profiled. These cells included immune, some epithelial, endothelial, and mesenchymal cells, as well as even a small number of neurons (Figure 1a-b). In simple terms, each dot on the UMAP in Figure 1b represents a cell, and through dimensionality reduction (i.e., scaling down the large amounts of data to a digestible form), a computational pipeline was able to produce a visualization of how the cells cluster together.
33,690 proximal tubule cell (CD10+) cells were also profiled, with 7 proximal tubule cell clusters identified. Among the proximal tubule cells from patients with CKD, there was gene expression enrichment for pathways related to dysregulated intracellular metabolism.
What is the origin of ECM in human CKD?
From the unbiased approach of scRNA-seq, the authors found that the vast majority of ECM in fibrotic human kidneys originates from mesenchymal cells. They generated a single-cell ECM scoring system based on core genes (for collagens, proteoglycans, and glycoproteins) published in this published ECM gene atlas. Figure 1g (left figure) confirms that when non-proximal tubule cells (CD10-) were clustered by ECM score and stratified by patient eGFR, almost all cells with high ECM scores belonged to patients with eGFR < 60 mL/min.
Extended Data Figure 2r (middle figure) shows a scaled UMAP visualization of ECM scores for each CD10- cell, grouped by cell types. Here, bright red indicates a high ECM score. The mesenchymal portion of the plot (dashed circle), which contains more red than other clusters, is enriched with cells with high ECM scores. Figure 1h (right figure) breaks this down further with violin plots of ECM scores stratified by cell type and patient eGFR. The lighter shades of color on the left half of each “violin” represents eGFR > 60 mL/min, and the darker shades on the right half represents eGFR < 60 mL/min. For endothelial cells (endo), epithelial cells (epi), and immune cells (im), there are more cells belonging to patients with eGFR > 60 mL/min with lower ECM scores (bottom of the y axis). This is why the endothelial, epithelial, and immune violins skew towards the left and down for these cell types. In contrast, the mesenchymal cells (Mes), represented in shades of brown, have more cells distributed towards the middle to upper half of the ECM score scale, with more cells present in the eGFR < 60 mL/min group. Among the mesenchymal cell types, more fibroblasts and myofibroblasts were seen in CKD samples.
Validating the link between mesenchymal cells and ECM, gene expression heatmaps reveal that collagen (Extended Data Figure 2s), proteoglycan (Extended Data Figure 2t), and glycoprotein genes are upregulated and enriched in mesenchymal cells (demarcated by brown boxes).
Based on the scRNA-seq data, mesenchymal cells seem to be responsible for producing most of the ECM in fibrotic kidneys. What are the mesenchymal cells?
Zooming in on just the mesenchymal cell population, trajectory inference analysis was performed. Trajectory inference is a computational method in which the relationships of each cell to other cells along a trajectory is inferred and visually represented in an ordered fashion. Typically, trajectory inference analysis is applied in the context of visualizing cell lineage and progression of a process such as development or cellular injury. Here, trajectory inference analysis suggests that myofibroblasts, rarely found in non-diseased tissue, arise from pericytes and fibroblasts (Extended Data Figure 3a) and reveals the cell type heterogeneity of the mesenchymal cell population. The origin of myofibroblasts has been the subject of much debate, and this study provides evidence that scientists in both the fibroblast and pericyte camps were partially correct but, without unbiased data, missing the whole picture.
But what about those epithelial cells we usually think about?
Prior studies had attributed the epithelial mesenchymal transition (EMT) of tubular epithelial cells to be a source of myofibroblasts and ECM (review of different kidney fibrogenesis hypotheses here and here). However, data from this study suggest that epithelial cells play only a minor role based on ECM scoring. Violin plots in Extended Data Figure 3e reveal that injured proximal tubule cells (iPT in the figures) are enriched in patients with eGFR < 60 mL/min (darker purple, right side of violin), in contrast to healthy proximal tubule (PT in the figures) cells. On gene ontology analysis of cell-specific biological processes (Extended Data Figure 3j), injured proximal tubule cells had different gene expression patterns enriched in pathways related to embryonic development, consistent with a shift from an epithelial state to a less differentiated state. Although these data demonstrate that a small portion of ECM gene expression in fibrotic kidneys comes from injured proximal tubule cells (Extended Data Figure 3e), the majority is still from mesenchymal cells.
If myofibroblasts are pathologic in the kidney, where do they come from?
To narrow the focus on potential sources of myofibroblasts, the investigators isolated 37,380 cells (via flow cytometry) from four kidneys of patients with eGFR > 60 mL/min and four kidneys of patients with eGFR < 60 mL/min. These cells were positive for the marker PDGFR‐β, which is a tyrosine kinase receptor expressed in kidney mesenchyme and implicated in kidney fibrogenesis. scRNA-seq data confirmed that the majority of cells expressing the ECM collagen gene COL1A1 also expressed the PDGFRB gene, consistent with past studies demonstrating the relevance of PDGFR-β in myofibroblast function.
Based on trajectory inference and diffusion map analysis of ECM-producing PDGFRB+ cells (Figure 2c), the authors concluded that the three major sources of human kidney myofibroblasts are pericytes and two different fibroblast lineages.
As expected, the myofibroblasts (labeled MF1 at bottom of lineage tree) have high ECM scores as well as high expression levels of COL1A1 and POSTN, which encodes periostin, a matricellular ligand for integrins involved in wound healing (Figure 2c). They also are enriched with cells from patients with eGFR < 60 mL/min (salmon colored dots upper right, Extended Data Figure 5a). Pericytes (Pe) are marked by NOTCH3, whereas the fibroblast lineages are marked by MEG3 and CXCL12 (Figure 2c, Extended Data Figure 5a). These markers and lineages were validated through in situ hybridization (labeling mRNA of interest and capturing on microscopy) and will be referenced in the authors’ proposed model at the end of this section. In fibrotic kidneys, triple-positive MEG3+/NOTCH3+/POSTN+ cells (open yellow circle, Extended Data Figure 5k) were detected. When a cell is positive for three markers, each marking a different cell type, that cell is likely in a state of transition among the three cell types, and, in this case, possibly in the middle of the lineage tree (solid yellow circle).
Additional analysis of the relationship between ligands from one cell type to the receptors of a different cell type identified TGFb signaling (red box, a known signaling pathway in fibrosis) as a mechanism by which myofibroblasts promote differentiation of pericytes (Extended Data Figure 6a) and fibroblasts (Extended Data Figure 6g).
Using motif enrichment analysis for regulatory transcription factors that promote the expression of their target genes, the authors identified potential key regulators of the fibroblast to myofibroblast transition such as including AP-1.
“The human scRNA-seq data capture CKD vs. non-CKD kidneys at one snapshot in time. Can the findings be recapitulated in a model system, and what can that model teach us in terms of mechanisms?”
Here is where mouse model validation comes in. First, the authors performed UUO surgery, an injury model that induces kidney fibrosis, on Pdgfrb-CreER-tdTomato mice (Figure 3a). As a brief reminder from Methods, this mouse model has an inducible fluorescent reporter (TdTomato, red) that is activated in cells expressing PDGFRβ. This mouse is used to mark cells with a red tag and see what happens to them after some perturbation occurs (UUO in this case). First, they wanted to see where fibrosis was occurring. For this, they looked for Col1a1 mRNA expression in kidney tissue of this mouse line through in situ hybridization (Figure 3b) and found that at day 10 after UUO, almost all Col1a1-mRNA expressing cells, i.e., ECM-producing cells, co-express PDGFRβ as indicated by the tdTomato tag (Figure 3b-c).
In a different mouse model with PDGFRβ expressing cells labeled with eGFP, GFP+ cells were sorted on flow cytometry and profiled with scRNA-seq after UUO (Figure 3d). These cells included pericytes, parietal epithelial cells, smooth muscle cells, mesangium, and fibroblasts (Figure 3e). Over time, pericytes and smooth muscle cells decreased in abundance (red), while mesangial cells and ECM-producing cells increased in abundance (green, Figure 3f). As seen in the human data, myofibroblasts marked by Postn were also positive for PDGFRɑ and PDGFRβ (Figure 3g, circles).
Concordance of double positivity for PDGFRɑ and PDGFRβ in Col1a1 expressing cells was seen in mouse UUO and human CKD in situ hybridization (Extended Data Figure 7h-k), thus validating that PDGFRɑ/PDGFRβ double positive cells are the main source of ECM.
Because ECM in kidney fibrosis mostly originates from PDGFRɑ+/PDGFRβ+ cells, the authors took a closer look at these double-positive cells in mouse kidney fibrosis experiments. They performed scRNA-seq on 7,245 PDGFRɑ+/PDGFRβ+ cells of mice that underwent UUO. There were two primary groups within the mesenchymal cells of these mice: fibroblast 1 (Fib1, marked by Scara5 and Meg3) and myofibroblasts of several subtypes (labeled as MF, Figure 3k).
Trajectory inference analysis (Figure 3l-n) shows that early states of differentiation are fibroblast 1 (Meg3+, Scara5+) and myofibroblast 2 (Col14a1+, Ogn+), whereas myofibroblast 1a (Nkd2+), myofibroblast 1b, and myofibroblast 3b are terminal states on the other end.
In human data, fibroblast 1 (purple box) is distinct from myofibroblasts subtypes on supervised classification (Extended Data Figure 9a). The authors then asked whether these groups were distinct cell types with distinct regulatory transcriptional programs. Using ATAC-Seq, which as a brief reminder profiles open chromatin regions where gene transcription occurs, they found that transcriptional regulation for fibroblasts 1 differed from that of myofibroblasts of PDGFR-ɑ+/PDGFR-β+ kidneys after UUO (Figure 3o).
Taken together, the following schematic in Extended Data Figure 9c sums up the authors’ model of how high-ECM expressing myofibroblasts (myofibroblast 1, NKD2+) arise from pericytes (NOTCH3+), fibroblasts 1 (SCARA5+), and fibroblasts 2 (CXCL12+).
Now that we have all that data and a proposed model for how mesenchymal cells transition into myofibroblasts in kidney fibrosis, how does this help develop novel therapeutics to combat CKD?
From their data, the authors identified NKD2 as a potential therapeutic target for kidney fibrosis. As mentioned above, NKD2 marks terminal myofibroblasts, and it is also involved in the Wnt and TNFa pathways. In humans, it co-localizes with high ECM-producing cells and is found in most PDGFR-ɑ+/PDGFR-β+ cells in kidneys with a higher interstitial fibrosis score (Figure 4d-e).
When NKD2 was overexpressed in PDGFR-β+ kidney cells in culture, ECM regulator and glycoprotein gene expression was upregulated on bulk RNA-seq. When NKD2 was knocked out, ECM-related genes were downregulated (Figure 4h). To establish therapeutic relevance, the authors grew stem cell derived kidney organoids and treated them with IL1b, which induces fibrosis in the model system. When NKD2 expression was knocked down in the IL1b-treated organoids, COL1A1 expression was reduced (Figure 4k), suggesting that inhibiting or knocking down NKD2 may be a potential therapeutic strategy for preventing the progression of fibrosis in the kidney.
Discussion
Kuppe et al.’s study is a tour de force that successfully delineated cell types and cell transitions driving human kidney fibrosis. With the high resolution of single-cell data, the authors were able to resolve the origin of myofibroblasts in the kidney, which cell types and genes are involved in the myofibroblast transition, and identify a potential therapeutic target for kidney fibrosis.
Of note, immune cells, which play important roles in profibrotic signaling, were captured but are missing from the proposed model. Decoding immune cell crosstalk with mesenchymal cells will be important for identifying additional therapeutic strategies. In addition, glomerular cells were not emphasized in this study, although a similar approach may offer insights into mesangial expansion and molecular drivers of glomerulosclerosis outside of the podocyte.
Despite those caveats, this study is a shining example of how unbiased discovery approaches can facilitate meaningful bedside to bench and back to bedside translational research. As scRNA-seq, spatial transcriptomics, single-cell proteomics, and single-cell metabolomics continue to evolve rapidly, we will be gaining even deeper understanding of how kidney development, fibrosis, and regeneration occur.
Summary by Jennie Lin,
Northwestern University, Chicago
and
Queen’s University and NSMC intern, class of 2020