Single-cell and bulk transcriptomics of the liver reveals potential targets of NASH with fibrosis

Abstract

Fibrosis is characterized by the excessive production of collagen and other extracellular matrix (ECM) components and represents a leading cause of morbidity and mortality worldwide. Previous studies of nonalcoholic steatohepatitis (NASH) with fibrosis were largely restricted to bulk transcriptome profiles. Thus, our understanding of this disease is limited by an incomplete characterization of liver cell types in general and hepatic stellate cells (HSCs) in particular, given that activated HSCs are the major hepatic fibrogenic cell population. To help fill this gap, we profiled 17,810 non-parenchymal cells derived from six healthy human livers. In conjunction with public single-cell data of fibrotic/cirrhotic human livers, these profiles enable the identification of potential intercellular signaling axes (e.g., ITGAV-LAMC1, TNFRSF11B-VWF and NOTCH2-DLL4) and master regulators (e.g., RUNX1 and CREB3L1) responsible for the activation of HSCs during fibrogenesis. Bulk RNA-seq data of NASH patient livers and rodent models for liver fibrosis of diverse etiologies allowed us to evaluate the translatability of candidate therapeutic targets for NASH-related fibrosis. We identified 61 liver fibrosis-associated genes (e.g., AEBP1, PRRX1 and LARP6) that may serve as a repertoire of translatable drug target candidates. Consistent with the above regulon results, gene regulatory network analysis allowed the identification of CREB3L1 as a master regulator of many of the 61 genes. Together, this study highlights potential cell-cell interactions and master regulators that underlie HSC activation and reveals genes that may represent prospective hallmark signatures for liver fibrosis.

Figure 1

Generation of scRNA-seq data for human liver non-parenchymal cells. (a) Workflow of sample preparation, sequencing and bioinformatic analysis. (b) UMAP visualization of single cells profiled in this study. Each dot represents a cell that is color-coded by cell type. HSCs, hepatic stellate cells; LSECs, liver sinusoidal endothelial cells; Vascular ECs, vascular endothelial cells; MDMs, monocyte-derived macrophages; NK cells, natural killer cells. (c) Representative gene expression and distribution of known marker genes for each population in UMAP plots. Normalized expression values are shown. (d) Dotplot displaying the top two marker genes for each cell type identified. Size of the dot represents proportion of the cell population that expresses each gene. Color indicates level of expression. (e) Barplot representing the relative contribution of cells from each donor for each cell type. NPC, non-parenchymal cells derived from healthy liver. (f) UMAP visualization of single cells profiled in this study together with human liver cells integrated from Ramachandran et al. (2019). (g) UMAP visualization of mesenchymal cells. The direction of cell differentiation inferred from estimated RNA velocities are plotted as streamlines on the UMAP. (h) UMAP visualization of mesenchymal cells. Cells colored by ECM score. (i) Heatmap showing the top 10 differentially expressed genes for each mesenchymal cell type. VSMCs, vascular smooth muscle cells; Meso, mesothelial cells.

Figure 2

Intercellular communications in the sinusoidal signaling niche. (a) Heatmap showing the number of potential ligand-receptor pairs between any two liver cell types predicted by CellphoneDB. (b) Dotplot displaying putative ligand-receptor interactions between aHSCs and other cell types. Dot size represents statistical significance of the indicated interactions. Dot color indicates the average expression level (log₂-transformed) of the receptor from aHSCs and the ligand from another cell type. (c–f) Gene expression of ITGAV, LAMC1, TNFRSF11B, and VWF at progressive disease stages. The RNA-seq data were downloaded and reanalyzed from Gerhard et al. (2018). Mann–Whitney U tests (two-sided) were performed for statistical comparisons. logCPM, log-transformed counts per million.

Figure 3

Cell type-specific gene regulatory networks (regulons) of liver mesenchymal cells. (a) Heatmap of the inferred regulons for human liver mesenchymal cells. (b) Regulon activity-based UMAP colored according to the regulon activity of CREB3L1, RUNX1, TWIST1 and TWIST2 showing the cell type specificity of regulons. (c) UMAP embedding of mouse liver mesenchymal cells from uninjured and fibrotic (up to 6 weeks of CCl₄ treatment) mouse livers retrieved from Dobie et al. (2019). Left, cells colored by cell type/state. Right, cells colored by condition/cell origin. (d) Heatmap of the inferred regulons for mouse liver mesenchymal cells. (e) Regulon activity-based UMAP colored by the regulon activity of Creb3l1 and Twist1 showing the cell type specificity of regulons.

Figure 4

Comparison of expression patterns between hepatic and pancreatic stellate cells. (a) UMAP embedding of human pancreatic cells retrieved from Baron et al. (2016). Left, cells colored by cell type. Right, cells colored by donor. qPSCs, quiescent PSCs; aPSCs, activated PSCs. (b) Volcano plot of differential gene expression analysis between activated and quiescent stellate cells for liver (left) and pancreas (right). Significantly upregulated (Sig.Up) and downregulated (Sig.Down) genes in activated stellate cells are shown in green and pink, respectively. 16 and 6 genes shared among the top 50 up- or down-regulated genes between liver and pancreas are labelled, respectively. (c) Heatmap of the inferred regulons for human pancreatic cells. (d) Regulon activity based UMAP colored according to the regulon activity of CREB3L1 and RUNX1 showing the cell type specificity of regulons.

Figure 5

Meta-analysis of existing RNA-sequencing data sets for NASH and liver fibrosis. (a) Differential gene expression analysis between fibrotic and healthy (as reference) livers. Only the top 10 up- or down-regulated genes are labelled with gene symbols. Significantly upregulated (Sig.Up) and downregulated (Sig.Down) genes are shown in green and pink, respectively. (b) Computational deconvolution of bulk RNA-seq data of 191 human liver samples (normal, 36; steatosis, 50; inflammation, 52; fibrosis, 53). The RNA-seq data were retrieved from Gerhard et al. (2018). Mann–Whitney U tests (two-sided) were performed for statistical pairwise comparisons. (c) Differential gene expression analysis between aHSCs and qHSCs (as reference) isolated from healthy and fibrotic (CCl₄) mouse livers. The RNA-seq data were retrieved from Marcher et al. (2019). Significantly upregulated (Sig.Up) and downregulated (Sig.Down) genes are shown in green and pink, respectively. Only the top 10 up- or down-regulated genes are labelled with gene symbols. (d) Differential gene expression analysis between aHSCs and qHSCs (as reference). Healthy liver-derived qHSCs were activated on plastic in vitro. The RNA-seq data were retrieved from Marcher et al. (2019). Significantly upregulated (Sig.Up) and downregulated (Sig.Down) genes are shown in green and pink, respectively. Only the top 10 up- or down-regulated genes are labelled with gene symbols. (e) Venn diagram showing all possible logical relations between three sets of up-regulated genes in aHSCs compared with qHSCs, and a set of up-regulated genes in bulk human fibrotic livers in comparison to healthy controls. Only the 15,252 protein-coding one-to-one orthologs for human, mouse, and rat were considered in this analysis (Supplementary Table 6). (f) Functional enrichment analysis for the 61 shared upregulated genes. (g) Regulatory gene network of the top three regulators (motifs) and their corresponding target genes among the 61 genes. (h–m) Contextualization of candidate therapeutic targets for liver fibrosis with time-course RNA-seq data. (h,i) High-fat (HF) diet-induced NASH mouse model; (j,k) Thioacetamide (TAA)-induced liver fibrosis rat model; (l,m) Bile duct ligation (BDL)-induced liver fibrosis rat model. Left, PCA of the samples studied. Right, expression levels of the genes of interest. TPM, transcripts per million. Adjusted p-values < 0.05 are colored in green.