2.1 Patient Selection

Patients whose liver biopsies were obtained between January 2019 and June 2022 were screened by two researchers specialising in liver disease. The inclusion criterion was liver biopsy histology indicating fatty liver. The exclusion criteria were as follows: (1) the presence of other viral liver disease (hepatitis A, hepatitis C and hepatitis E) or HIV; (2) MASLD with increased alcohol intake (alcohol intake exceeding 30–60 g/day in males and 20–50 g/day in females every week); (3) autoimmune hepatitis, drug-induced liver injury, Wilson's disease, total parenteral nutrition or toxic liver disease; (4) malignant tumour and (5) other autoimmune diseases. Healthy control liver tissue was obtained from patients with unconjugated hyperbilirubinemia whose liver biopsy histology indicated Gilbert syndrome with normal liver tissue. The biopsy samples were fixed in formalin, embedded in paraffin and stained with haematoxylin–eosin, Oil Red O, Mesh and Masson. All liver tissue sections were assigned the steatosis, activity and fibrosis (SAF) scoring system by two liver pathologists [13]. Disagreements on the pathology scores were discussed or resolved with the assistance of a third pathologist in the group. Samples from healthy controls, patients with simple steatosis and MASH were used for space transcriptome sequencing.

2.2 Spatial Transcriptomic Assay

Each formalin-fixed paraffin-embedded (FFPE) tissue sample was embedded, sectioned and placed on a capture area of a gene expression slide. Standard fixation and staining techniques were used to visualise tissue sections on slides under brightfield microscopy according to the recommended protocol CG000409. FFPE tissues were permeabilised to release ligated probe pairs that bind to the spatially barcoded oligonucleotides present on the spots. Spatial barcodes were added via an extension reaction. The barcoded molecules were then pooled and subjected to downstream processing to generate a sequencing-ready library according to the recommended protocol CG000407. The resulting 10× barcoded library was compatible with standard NGS short-read sequencing on Illumina sequencers for massive transcriptional profiling of entire tissue sections.

2.3 Visium Data Processing

In this study, Space Ranger (version 1.3.1) was used for data preprocessing, gene expression quantification and point identification, after which the corresponding gene expression of the effective cells was revealed. The Seurat package (version 4.1.1) was used to analyse and cluster the spots. The gene expression matrix was pre-processed using the Seurat SCTransform function [14]. The top 2000 highly variable genes (top 2000) were extracted for principal component analysis (PCA) to reduce the dimensionality of the data, and the top 15 significant principal components were subjected to cluster analysis at 0.5 resolution. Uniform manifold approximation and projection (UMAP) and t-distributed stochastic neighbour embedding (t-SNE) were then used to visualise the clusters. Marker genes for each cluster and subgroup were identified by comparing the expression of genes in spots from specific clusters or subgroups to that of others using the Seurat FindMarkers function and filtered using p < 0.05 and log-scale fold change ≥ 0.25 as the threshold.

2.4 Enrichment Analysis

ClusterProfiler software was used to enrich cluster markers by gene ontology (GO) [15] and Kyoto Encyclopedia of Genes and Genomes (KEGG) [16]. The marker genes identified using the Seurat software were adjusted using the Benjamini–Hochberg multiple test according to the Wilcoxon test and log-scale folding change (logFC. threshold ≥ 0.25). The results were visualised using the R package.

2.5 Cell Clustering and Deconvolution Analysis

The R package Seurat (v4.03) was used to cluster the cells in the merged matrix. Cells with < 300 genes, or more than 25% of mitochondrial expression were first filtered out as low-quality cells (Figure S1). Normalised data with default parameters were used to normalise the expression level for each cell [17]. The FindIntegrationAnchors and IntegrateData functions were used to integrate the samples prepared using different 10× Chromium chemistries. The ScaleData function was used to scale and centre the counts in the data set. PCA was performed on the genes that showed variable expression, and the first 30 PCs were used for cell clustering and UMAP dimensional reduction. The cluster marker genes were identified using the FindAllMarkers function. The cell types were annotated by overlapping the cluster markers with the canonical cell type signature genes. An anchor-based integration method was implemented for the integration of spatial RNA-seq data with matched scRNA-seq data using the FindIntegrationAnchors command, and the cell type labels were then transferred to the spatial data using the TransferData command [18]. Cell-type prediction values for spatial RNA-seq spots were saved as an assay and used in further analysis.

2.6 Cell–Cell Interactions

CellPhoneDB, a curated database of ligands, receptors and their subunit interactions, was used to identify ligand–receptor interactions that occur among the gene products expressed in the samples [19]. Following the identification of different cell types in the spatial transcriptomic data, we used the ‘statistical_analysis’ function in CellPhoneDB v2.0 to calculate the likelihood of interaction between each pair of signalling cells. Intercellular communication was modelled using the law of mass action to bind gene expression to receptor ligands and their cofactors of known signals. The number of ligand–receptor pairs was deduced from the average gene expression between cell groups.

2.7 Pseudotime Analysis

We performed pseudotime analysis using Monocle v3 software with the default parameters. First, the UMAP algorithm was used to reduce the dimensionality of the data. We then inferred developmental trajectories and subset cells based on their branches in the trajectory.

2.8 Transcription Factor Regulon Analysis

Analysis of the regulatory network and regulon activity was performed by pySCENIC [20]. The regulon activity (measured in AUC) was analysed using the AUCell module of pySCENIC, and the active regulons were determined based on the AUCell default threshold. The differentially expressed regulon was identified by the Wilcoxon rank-sum test in the ‘FindAllMarkers’ function in the R package Seurat with the following parameters: min.pct = 0.1, logfc.threshold = 0.25, pseudocount.use = F and only.pos = T. A heatmap showing the scaled expression of regulon activity was generated.

3.1 The Spatial Transcriptome Reveals the Spatial Distribution Characteristics of MASH Fibrosis

Twelve samples—four from healthy controls, four from patients with simple steatosis and four from patients with MASH—were subjected to spatial transcriptome sequencing. Following quality inspection, five of the samples were ultimately selected for use in the analysis: one from a healthy control (pathology number: 191518), one from a simple steatosis patient (pathology number: 20201869, SAF score: S2A0F1) and three from MASH patients (pathology numbers: 20201896, SAF score: S2A4F2; 20 232 357, SAF score: S2A4F3; 181 394, SAF score: S1A4F4). The unique molecular identifiers (UMIs) corresponding to each spot of every chip were detected as depicted in Figure S2. PCA linear dimension reduction was conducted on the spatial transcriptomic data using Seurat_4.1.1, and the elbowplot function was used to select the best resolution parameters for cluster analysis (Figure S3). The clustering results were visualised using t-SNE and UMAP, as demonstrated in Figure 1. After the first 30 principal components, which showed stable standard deviations, were selected for downstream data analysis, the samples were divided into 12 clusters via dimensionality reduction and clustering, as illustrated in Figure 2A. The regions of Clusters 3 and 9 were found to highly overlap with the inflammatory and fibrotic regions, respectively, identified by hepatology (Figure 2B–D). The classification of identical RNA capture spots as fibrotic by conventional histology and spatial transcriptomics exhibited remarkable consistency (fibrosis: 80% mean congruency (Figure S4)), indicating the potential of spatial transcriptomics for delineating fibrotic tissue regions in patients with MASLD and aiding in the characterisation of tissue types when combined with conventional histological assessment. Fibrosis in MASH was primarily distributed in the lobules, and a small amount was also observed in the portal area.

Subsequently, a functional analysis of Cluster 9 was conducted. GO analysis revealed that biological processes (BPs), which included ECM receptor interaction, were mainly concentrated in the ECM and in structural tissues. Molecular functions (MFs) included functions related to glycosaminoglycans and integrins and were mainly enriched in ECM structural components and signal molecule transmission. The cell composition (CC) terms were mainly related to the positive regulation of cell adhesion, as shown in Figure 3A. According to the KEGG analysis, the differentially expressed genes were predominantly enriched in the NF-κ B signalling pathway and ECM-receptor interaction, as illustrated in Figure 3A. GSEA enrichment analysis focused on ECM-related pathways (Figure 3B) such as NABA Core Matrisome, Extracellular Matrix Organisation, lipid metabolism and biological oxidation. NABA Core Matrisome and Extracellular Matrix Organisation represent collections of genes that encode proteins associated with the core ECM, ECM glycoproteins and related structural elements. Through differential analysis of subgroups, 303 differentially expressed genes were obtained from Cluster 9. In summary, the functional clustering results preliminarily confirmed Cluster 9 as the fibrosis region.

3.2 Single-Cell and Spatial Transcriptomics Reveal the Key Role of HSCs in MASH Fibrosis

Through integration of spatial transcriptome data with the single cell-sequencing data set GSE189175, a total of seven cell types were identified from GSE189175 (Figure 4A). Compared to those in the healthy control and simple steatosis groups, the proportions of inflammatory cells, such as Kupffer cells and T cells, increased in the MASH group, whereas the proportion of HSCs decreased, and the proportion of myofibroblasts significantly increased (Figure 4B–E). Cell communication analysis revealed that the main type of cell-to-cell communication, including endocrine, paracrine and autocrine communication, was secretion, followed by ECM-receptor cell communication (Figure 4F). In MASH fibrosis, hepatocytes, HSCs, myofibroblasts and bile duct cells exhibited strong communication with each other (Figure 4G), and hepatocytes, HSCs and myofibroblasts were the primary cell types communicating via collagen signalling pathways (Figure 4H). Additional communication pathways included the FN1, LAMININ, PTPRM, THBS and VISFATIN signalling pathway networks (Figure S5).

3.3 Key Genes Specifically Overexpressed in the MASH Fibrosis Region

Six genes with high- and low-specific expression in Cluster 9, namely, ADAMTSL2, PTGDS, S100A6, PPP1R1A, ASS1 and G6PC, were screened in combination with pathology (Figure 5). Notably, in addition to being highly expressed in Cluster 9, ADAMTSL2, PTGDS and S100A6 also exhibited relatively high expression in Cluster 3, while PPP1R1A, ASS1 and G6PC showed the opposite pattern. The average expression levels of ADAMTSL2, PTGDS and S100A6 indicated by the pathological HE staining map correlated positively with the increase in the degree of fibrosis and aligned strongly with the distribution of fibrosis. Conversely, the expression levels of PPP1R1A, ASS1 or G6PC within the fibrotic region decreased as the degree of fibrosis increased (Figure 6). Overall, these findings indicate that these six genes serve as specific markers of MASLD fibrosis.

Through the analysis, it was found that the expression of six key genes varied in different cell types; ADAMTSL2 showed high specificity in myofibroblasts (Figure 7A). To investigate the potential involvement of ADAMTSL2+ myofibroblasts in liver fibrosis, SCENIC analysis was used to categorise myofibroblasts into two groups, ADAMTSL2+ and ADAMTSL2− based on the expression of ADAMTSL2. The specific transcription factors (TFs) of ADAMTSL2+/ADAMTSL2− myofibroblasts were predicted using SCENIC, and TFs with high relative activity scores were identified (Figure 7B). The numbers of ADAMTSL2+ and ADAMTSL2− myofibroblasts were positively regulated by five TFs: IRF8, FLI1, SOX4, GATA6 and TBX2. These five TFs were found to be highly expressed in Cluster 9, suggesting their potential significance as TFs for ADAMTSL2+ myofibroblasts. Subsequent analysis of the disparities in cellular signalling pathways between the two groups (Figure 7C) revealed the involvement of ADAMTSL2+ myofibroblasts in fibrosis development. Notably, ADAMTSL2+ myofibroblasts were implicated in the TNF signalling pathway and in the regulation of the production of structural constituents of the ECM.

To investigate the evolutionary relationships among MASLD cells, we used Monocle v3 to analyse the developmental trajectory of these cells. Trajectory analysis was performed to allow us to infer the pathways involved in cell-state transitions within the MASLD clusters. Pseudotemporal ordering was performed (Figure 8A) with the cell state at time 0 set as ‘hepatocyte’ on the basis of the results of the RNA velocity analysis. At the outset of the pseudotime, the majority of the cells were typical hepatocytes. As the pseudotime progressed, an increase in the number of T cells and Kupffer cells was observed. Subsequently, under the influence of T cells and Kupffer cells, the proportion of HSCs and myofibroblasts gradually increased, and the cells were clustered densely at the end of the pseudotime (Figure 8B). These findings indicate that hepatic cells experience diverse evolutionary trajectories during the progression of MASLD fibrosis. Specifically, the main contributing factor to the events that occur in the early stage of MASH is the inflammation caused by infiltration by T cells and Kupffer cells. This inflammation subsequently triggers the conversion of HSCs into myofibroblasts, leading to the initiation of liver fibrosis.

In a further search for genes that are differentially expressed across the pseudotime trajectory, we used plot cells to identify the first six genes whose expression is clearly branch-dependent. These are ACAP3, ATAD3B, MIB2, MTND2P28, CCNL2 and RP5-857K21.4 (Figure 8C). ACAP3 and RP5-857K21.4 showed obvious branch-dependence in the hepatocellular-T lymphocyte-HSCs branch, while ATAD3B, MIB2, MTND2P28 and CCNL2 showed obvious branch-dependence in the T lymphocyte/Kuffer cell-HSCs branch. By observing the regulatory roles of key genes in the trajectory of cellular development, we found that G6PC and ASS1 were highly expressed in the early stages of development of MASLD disease and expressed at low levels in the middle and late stages (Figure 8D). This finding suggests that these two genes may have antifibrotic functions, as their reduced expression may promote the progression of fibrosis.