Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Apr 28.
Published in final edited form as: Liver Int Commun. 2025 Apr 28;6(2):10.1002/lci2.70012. doi: 10.1002/lci2.70012

Cell-Type Deconvolution Reveals Dynamic Changes in MASLD

Jeff JH Kim 1, Yang Dai 1,*
PMCID: PMC12490751  NIHMSID: NIHMS2058017  PMID: 41048656

Abstract

Background & Aims:

Metabolic-associated steatotic liver disease (MASLD) is among the most prevalent liver disorders worldwide, with many patients progressing to metabolic-associated steatohepatitis (MASH) characterized by fibrosis and inflammation. The current lack of effective treatments for MASH highlights the urgent need to deepen our understanding of its underlying mechanisms. Examining cellular dynamics — specifically, changes in cell type proportions across disease stages — offers a promising avenue for gaining such insights. However, previous deconvolution analyses have been limited to a few cell types, and a comprehensive analysis encompassing diverse cell populations and their unique subtypes has yet to be conducted.

Methods:

In this study, we employed MuSiC deconvolution to analyze two bulk RNA sequencing datasets spanning the MASL to MASH spectrum across both fibrosis staging and Non-Alcoholic Fatty Liver Disease Activity Score (NAS) staging.

Results:

Our analysis reveals distinct proportional trends in 10 different cell types, including hepatocytes, cholangiocytes, two subpopulations of hepatic stellate cells, endothelial cells, and immune cells such as kupffer cells, TREM2+ macrophages, and plasma B cells. In addition to deconvolution analysis, we integrated cell type proportion data with transcriptomic profiles, significantly enhancing the performance of random forest models in classifying fibrosis stages compared to using transcriptomic data alone.

Conclusions:

The study’s findings highlight critical cellular dynamic changes across MASLD progression, advancing our understanding of the disease mechanisms and potentially informing the development of more effective therapeutic strategies.

Keywords: MASLD, MASH, Cell-Type Deconvolution, Fibrosis Staging, Disease Progression

Lay Summary:

Metabolic-associated steatotic liver disease (MASLD) is a common liver condition that can progress to a more severe stage called metabolic-associated steatohepatitis (MASH), which involves liver scarring and inflammation. Currently, there are no effective treatments for MASH, so understanding how different liver cell types change during disease progression is critical. In this study, we analyzed liver tissue samples and identified key changes in the proportions of various cell types across disease stages and we also found that combining this cell type information with gene activity data improved our ability to predict disease severity.

Introduction

Metabolic-associated steatotic liver disease (MASLD) represents a growing global health concern, leading to severe liver conditions and significant clinical implications.1,2 MASLD, previously known as non-alcoholic fatty liver disease (NAFLD), encompasses a spectrum of liver abnormalities ranging from steatosis to the more severe Metabolic Associated Steatohepatitis (MASH), characterized by hepatocyte ballooning, inflammation, and damage.3,4 Understanding the progression from MASLD to MASH, as determined by either the fibrosis stage or the Non-Alcoholic Steatohepatitis (NAS) score, is important for early diagnosis and therapeutic intervention.5,6

Accurate staging of fibrosis and NAS is essential for assessing disease progression and guiding clinical management.7 Fibrosis is graded from F0 to F4, where F0 indicates the absence of fibrosis, and F4 signifies cirrhosis with extensive scarring.8 On the other hand, the NAS score evaluates three key histological features: steatosis, lobular inflammation, and hepatocyte ballooning. Each feature is scored individually—steatosis and inflammation on a scale of 0 to 3 and ballooning on a scale of 0 to 2—resulting in a total NAS score ranging from 0 to 8.9 However, it is important to recognize that while the NAS score reflects disease activity, it does not necessarily correlate with the fibrosis score, as these two assessments arise from distinct biological processes.10

In recent years, cell-type deconvolution techniques have emerged as powerful tools to dissect the cellular composition of tissues.11 These computational methods leverage transcriptomic data to estimate the proportions of different cell types within a bulk tissue sample, providing insights into the cellular dynamics underlying disease progression. One such technique, MuSiC, utilizes cell-type specific gene expression from single-cell RNA-seq (scRNA-seq) data to characterize cell type compositions from bulk RNA-seq data in complex tissues.12 By appropriately weighting genes that show cross-subject and cross-cell consistency, MuSiC enables the transfer of cell type-specific gene expression information from one dataset to another without the need for a predetermined signature matrix, unlike counterparts such as CIBERSORTx.13

While scRNA-seq offers detailed insights into cellular heterogeneity, generating large datasets with sufficient coverage for comprehensive analysis is often impractical due to high costs and technical limitations.14 Therefore, relying solely on scRNA-seq for determining cell type composition can be challenging. To address this issue, deconvolution techniques leverage bulk RNA-seq to provide broad coverage of patient samples, while incorporating scRNA-seq to ensure single-cell genomic precision.15 By distinguishing the contributions of different cell populations, the mechanisms of fibrosis and inflammation in MASH can be better highlighted, ultimately leading to a more comprehensive understanding of the disease and improved patient outcomes.

A previous study conducted by Pantano et al. applied MuSiC deconvolution to the stages of MASLD and MASH, revealing critical insights into the roles of different cell types in the disease pathogenesis.16 Through cell type deconvolution, the study demonstrated the loss of hepatocytes and the gain of hepatic stellate cells (HSCs), macrophages, and cholangiocytes proportions with advancing fibrosis. However, the study did not capture the comprehensive heterogeneity of liver cell types beyond the four cell types, nor did it provide information on their specific subpopulations. Additionally, the study did not examine cell type composition across NAS score staging.

This study aims to leverage MuSiC deconvolution to dissect the cellular composition of liver tissues in the progression of MASLD to MASH by observing two progression lines: fibrosis and NAS score staging. By integrating bulk RNA-seq with scRNA-seq data, a comprehensive analysis will be provided to highlight the significance of specific cell types in disease progression, offering new perspectives for future research.

Materials and Methods

Datasets Used

Raw scRNA-seq data from the Payen et al. dataset (GSE158723) was obtained from the Gene Expression Omnibus (GEO). This dataset profiles the transcriptomes of over 25,000 individual liver cells from two healthy human donors using droplet-based RNA sequencing. It was chosen for its comprehensive coverage of unique liver cell types. The original study identified 22 distinct cell subpopulations, covering the heterogeneity of endothelial cells, macrophages, HSCs, and lymphoid cells.17

Raw bulk RNA-seq data from the Govaere et al. dataset (GSE135251) and Pantano et al. dataset (GSE162694) were imported with 19,861 and 31,572 genes, respectively. The Govaere et al. dataset is a comprehensive multicenter study investigating transcriptional alterations during liver disease progression. It includes 216 snap-frozen liver biopsies, with 206 representing various fibrosis stages of MASLD and 10 serving as controls.18 The Pantano et al. dataset consists of subjects from the Massachusetts General Hospital NAFLD Cohort, comprising 112 adults diagnosed with NAFLD based on imaging or liver histology, along with 31 controls.19 Most participants underwent bariatric surgery with liver biopsies, while others had liver transplants or clinically indicated biopsies. Tissue samples were either flash-frozen or preserved in RNAlater before RNA sequencing. All biopsies were pathologically assessed and assigned fibrosis and NAS scores. Table 1 details the sample distribution across disease stages in both datasets. These bulk RNA-seq datasets were selected for their high sequencing quality and large sample sizes, providing robust coverage across the MASLD spectrum.

Table 1.

Number of samples in each fibrosis stage (Control, MASL, MASH F0–F4) and NAS stage (Stage 0–8) across Govaere et al. (GSE135251) and Pantano et al. (GSE162694) datasets

GSE135251 GSE162694 GSE135251 GSE162694
Control 10 31 Stage 0 10 33
MASL 51 NA Stage 1 11 12
MASH F0 5 35 Stage 2 21 9
MASH F1 29 30 Stage 3 26 10
MASH F2 53 27 Stage 4 38 13
MASH F3 54 8 Stage 5 47 19
MASH F4 14 12 Stage 6 37 12
Stage 7 18 9
Stage 8 8 NA
Not Graded 0 26
Open in a new tab

RNA-seq Data Analysis

The Seurat suite (v5.0.1) was used to analyze scRNA-seq dataset Quality control measures were implemented to filter out low-quality cells, excluding those with fewer than 100 detected genes, total read counts below 100, or mitochondrial gene expression exceeding 20%. To account for variations in sequencing depth, log transformation normalization and scaling were applied using Seurat. Dimensionality reduction was performed using Principal Component Analysis (PCA), retaining the top 50 principal components for downstream analysis. The top 200 most variable genes were selected to enhance the resolution of clustering. Uniform Manifold Approximation and Projection (UMAP) was used to visualize the data within a two-dimensional space. Differential expression analysis was conducted using the Wilcoxon rank-sum test to identify differentially expressed genes within each cell cluster. Cell type annotations were assigned based on established marker gene expression profiles and further validated using SingleR, with reference to the Human Primary Cell Atlas to ensure accuracy.

For the bulk RNA-seq data, genes with more than 10 counts across at least three samples were retained. Normalization was performed using the DESeq2 package to account for differences in sequencing depth and library size. Gene Ontology (GO) enrichment analysis was performed using the clusterProfiler package to identify biological processes associated with differentially expressed genes throughout MASLD progression. For GSE135251, the groups included Control, MASL/MASH F0, MASH F1/F2, and MASH F3/F4, while for GSE162694, the groups were Control, MASH F0, MASH F1/F2, and MASH F3/F4. Differential expression analysis was performed by comparing each fibrosis stage to its preceding stage to capture significant expression changes along the continuum of liver fibrosis. Genes were classified as differentially expressed based on adjusted p-values < 0.05 and log2 fold-change thresholds > 0.58. Enriched GO terms were visualized using dot plots to highlight key biological processes.

Cell Type Deconvolution

Cell type deconvolution of RNA-seq data was performed using MuSiC, chosen for its ability to account for interindividual variability in bulk RNA-seq data, its robustness in estimating cell type proportions from heterogeneous tissue samples, and its independence from a predefined signature matrix. Annotated single-cell RNA-seq data were processed using SingleCellExperiment to generate a reference dataset. Normalized bulk RNA-seq data were then deconvoluted using this single-cell reference to estimate cell type proportions. The analysis was conducted separately for each stage of liver fibrosis and NAS score. Average cell type proportions for each stage were calculated and visualized to highlight compositional differences. To compare cell type proportions across stages, a non-parametric ANOVA (Kruskal-Wallis test) and linear regression for trend analysis were performed to assess the change across disease progression.

Pseudobulk and Downsampling Analysis

To validate the accuracy of the MuSiC algorithm for cell type deconvolution, we generated 50 pseudobulk samples from the Payen et al. dataset and deconvoluted them using its annotated single-cell reference data (Figure S1). Cell types with deconvolution results yielding a Pearson correlation coefficient (R) below 0.7 were excluded from further analysis. Additionally, bulk RNA-seq datasets from Govaere et al. and Pantano et al. were used in a downsampling analysis to assess the robustness of MuSiC in detecting rare cell types (Figure S2). Gene expression counts were systematically downsampled from 90% to 20% to evaluate the impact of reduced sequencing depth on deconvolution accuracy. MASLD stage was selected for the Govaere et al. dataset, while MASH F0 was chosen for the Pantano et al. dataset, as these stages had larger sample sizes and comparable staging characteristics. Statistical differences and trends of the cell type proportions predicted from MuSiC across the disease stages were analyzed using nonparametric ANOVA and linear regression analysis.

Random Forest for MASLD Fibrosis Stage Classification

A Random Forest machine learning model was implemented to classify liver fibrosis stages based on gene expression profiles using the scikit-learn library in Python. The Govaere et al. dataset, containing gene expression data and corresponding fibrosis stage labels, was imported into a pandas DataFrame. To address class imbalance and improve model performance, fibrosis stage labels were adjusted by merging stages with similar pathological characteristics: MASL was combined with MASH F0, MASH F1 with F2, and MASH F3 with F4. Feature selection was performed using the ANOVA F-test, selecting the top N genes based on their F-scores, where N was optimized for model performance. This step reduced dimensionality while retaining the most informative genes for classification. A 10-fold cross-validation was employed to evaluate the model’s generalizability, with the average accuracy across all test folds reported as an estimate of performance. Random Forest models were trained using 500 decision trees on the selected gene features. Additionally, cell type proportion data was incorporated alongside gene expression values to assess its impact on fibrosis stage classification. Stratified k-fold cross-validation was repeated, and the accuracy of the combined model was compared to the transcriptomics-only approach to determine whether cell type proportions improved classification performance.

Results

Bulk RNA-seq and Single Cell Datasets Show Distinctive Clustering Patterns

The PCA plots of the bulk RNA-seq datasets reveal distinct clustering patterns when organized by fibrosis stage and NAS scores (Figure 1). In the Govaere et al. dataset, the first and second principal components (PC1 and PC2) account for 9.55% and 7.99% of the variance, respectively (Figure 1A,B). Samples from different stages form a moderately dispersed cluster with some overlap, suggesting a continuum of variation rather than distinct, discrete clusters. In contrast, the Pantano et al. dataset exhibits clearer stage separation when organized by fibrosis stage, with PC1 and PC2 explaining 14.23% and 12.89% of the variance, respectively (Figure 1C,D). Notably, MASH F4 samples cluster more distinctly from the other stages, indicating a stronger separation in advanced disease. However, when organized by NAS scores, the clustering is less pronounced, with the exception of NA values, which form a more distinct cluster from the other stages.

Figure 1. Principal Component Analysis of the two bulk RNA-seq datasets.

Figure 1.

Open in a new tab

PCA plot of bulk RNA-Seq dataset filtered based on fibrosis (A) and NAS score (B) of the Govaere et al. dataset. PCA plot of bulk RNA-Seq dataset filtered based on fibrosis (C) and NAS score (D) of the Pantano et al. dataset (D).

The UMAP plot of the single-cell dataset (Payen et al.) reveals 17 distinct cell populations, comprising 8 non-immune and 9 immune cell types (Figure 2A). Closely related subtypes are observed to cluster together, with immune cells forming a tight grouping similar to that seen among cholangiocytes and HSCs. Endothelial subtypes cluster in proximity to immune cells, while hepatocytes remain distinct and separate from other cell types. To verify cell type annotations, a dot plot illustrating the expression profiles of key marker genes across different cell populations is presented in Figure 2B. Table 2 provides a detailed list of these marker genes, their supporting literature, and the biological functions of the corresponding cell types.

Figure 2. Single cell RNA-seq data analysis.

Figure 2.

Open in a new tab

UMAP of single cell RNA-seq dataset of Payen et al. single-cell RNA-Seq dataset (n=1) annotated with the corresponding identity (A) and the cluster identity verification with gene markers (B).

Table 2.

Identified cell types from Payen et al. (GSE158723) single-cell RNA-seq, their marker genes, and primary biological functions

Cell Type Genes Function
B Cell BANK144, MS4A145 Part of the adaptive immune system; produces antibodies.
Cell Cycle PBK46, TOP2A47 Represents proliferating cells across different cell types.
Cholangiocyte EPCAM48, KRT1949 Epithelial cells lining bile ducts; involved in bile production and secretion.
Dendritic Cell A CD1C50, CLEC10A51 Antigen-presenting cells
Dendritic Cell B XCR152, IDO153 Antigen-presenting cells
Dendritic Cell Plasmacytoid LILRA454, GZMB55 Important for antiviral immunity; produce large amounts of interferon-alpha.
Hepatic Stellate Cell A RGS556, COLEC1057 Produce collagen-rich extracellular matrix in liver fibrosis
Hepatic Stellate Cell B DPT58, TPM259 Produce collagen-rich extracellular matrix in liver fibrosis
Hepatocyte CYP3A460, ALB61 Main functional liver cells; involved in metabolism, detoxification, and protein synthesis.
Lymphatic Endothelium PDPN62,63, CCL2164,65 Lines lymphatic vessels; facilitates fluid homeostasis and immune cell trafficking.
Macrophage A TIMD466, MARCO67,68 Innate immune cells involved in tissue homeostasis, immune defense, and repair
Macrophage B TREM269, GPNMB70 Innate immune cells involved in tissue homeostasis, immune defense, and repair
Neutrophil S100A8, S100A971 Innate immune cells that defend against bacterial and fungal infections.
Plasma B Cell POU2AF172, IGHG273 Terminally differentiated B cells that secrete large amounts of antibodies.
Sinusoidal Endothelium STAB174, OIT375 Specialized liver endothelium; plays a role in blood filtration and immune tolerance.
T Cell CD3D, CD3E76 Adaptive immune cells involved in immune responses, including helper and cytotoxic functions.
Vascular Endothelium VWF77, ACKR178,79 Lines blood vessels; regulates blood flow, coagulation, and immune cell trafficking.
Open in a new tab

Unique Cell Type Composition Changes Are Seen Across Fibrosis and NAS Score Staging

Cell type deconvolution using MuSiC identified 10 distinct cell type proportions, comprising 7 non-immune and 3 immune cell types, across the two bulk RNA-seq datasets. Distinct changes in cell type proportions were observed across fibrosis stages (Figures 3A and 3B). Hepatocyte proportions significantly declined across MASLD stages in both the Govaere (Slope = −0.0178, p = 7.81e-06) and Pantano (Slope = −0.0451, p = 1.37e-15) datasets, while Cholangiocyte proportions significantly increased in the Pantano dataset (Slope = 0.0085, p = 3.81e-05). Cholangiocyte also showed significant one-way ANOVA results in both Govaere (Kruskal-Wallis, p = 3.6e-05) and Pantano (Kruskal-Wallis, p = 2.4e-06) datasets. The two HSC sub-populations exhibited notable increases. RGS5+ HSC showed a continuous rise in both Govaere (Slope = 0.0122, p = 5.23e-07) and Pantano (Slope = 0.0197, p = 1.69e-11). Similarly, DPT+ HSCs showed marked increases in Govaere (slope = 0.0018, p = 5.4e-06) and Pantano (slope = 0.0105, p = 2.57e-10) datasets. For sinusoidal endothelium, a significant increase was observed in the Pantano dataset (slope = 0.004, p = 0.000116), while no significant change was detected in the Govaere dataset. Lymphatic Endothelium exhibited a modest increasing trend in both Govaere (Slope = 5e-04, p = 1.18e-05) and Pantano (Slope = 5e-04, p = 0.00592) datasets, though only Govaere exhibited a significant difference among stages (Kruskal-Wallis, p = 4.1e-05).

Figure 3. Proportion of liver cell types across fibrosis score staging.

Figure 3.

Open in a new tab

Cell type deconvolution results of the liver bulk RNA-seq data for 10 liver cell types across fibrosis score staging for the Govaere et al. dataset (A) and the Pantano et al. dataset (B).

Kupffer cell proportions did not show statistically significant changes in either dataset. However, TREM2+ macrophage proportions significantly increased in both Govaere (Slope = 0.0015, p = 5.23e-05) and Pantano (Slope = 4e-04, p = 0.00216) datasets. Plasma B Cells exhibited significant differences among stages in both Govaere (Kruskal-Wallis, p = 8.8e-05) and Pantano (Kruskal-Wallis, p = 0.00024) datasets, with a significant increasing trend in Pantano (Slope = 5e-04, p = 6.59e-05). Lastly, cell cycle analysis showed statistically significant changes in the Pantano dataset (Kruskal-Wallis, p = 0.021; Slope = 4e-04, p = 0.0145).

Cell type deconvolution based on NAS score revealed unique biological insights across MASH progression (Figures 4A and 4B). Hepatocyte proportions significantly declined with increasing NAS scores in both Govaere (Slope = −0.0124, p = 1.74e-05) and Pantano (Slope = −0.016, p = 1.49e-05) datasets. Cholangiocyte proportions showed significant differences in the Pantano dataset (Kruskal-Wallis, p = 0.05). RGS5+ HSC proportions demonstrated consistent increases across both Govaere (Slope = 0.0078, p = 1.98e-05) and Pantano (Slope = 0.0114, p = 3.74e-08) datasets. DPT+ HSC proportions showed a modest increase in Govaere (slope = 6e-04, p = 0.0468), with significant stage-wise differences observed in both Govaere (Kruskal-Wallis, p = 0.0052) and Pantano (Kruskal-Wallis, p = 0.002). Sinusoidal endothelium proportions increased across NAS stages in both Govaere (slope = 0.0026, p = 0.018) and Pantano (slope = 0.0025, p = 0.000523). Lymphatic endothelium proportions exhibited significant differences among stages in both Govaere (Kruskal-Wallis, p = 0.027) and Pantano (Kruskal-Wallis, p = 0.026) datasets.

Figure 4. Proportion of liver cell types across NAS score staging.

Figure 4.

Open in a new tab

Cell composition deconvolution of the liver bulk RNA-Seq data for 10 liver cell types across NAS score staging for the Govaere et al. dataset (A) and the Pantano et al. dataset (B).

Kupffer cell proportions showed significant differences in the Pantano dataset (Kruskal-Wallis, p = 0.026). TREM2+ macrophage proportions showed increasing trend in the Govaere dataset (Slope = 2e-4, p = 0.013). Plasma B Cells exhibited an increasing trend in the Pantano dataset (Slope = 2e-04, p = 0.0389). Additionally, Govaere showed an increasing trend for cell cycle-related proportions (Slope = 3e-04, p = 0.0396).

Addition of Cell Type Proportions as Features Improves Accuracy of MASLD Fibrosis Staging Classification in Random Forest Models

To assess whether predicted cell proportions enhance the classification of MASLD fibrosis stages, Random Forest models were trained using the Govaere dataset with two feature sets: 1) transcriptomic-only features (200 most variate genes) and 2) transcriptomic features and the cell type proportions. The Area Under the Curve (AUC) using the 10-fold cross-validation had a median value of approximately 0.76 with an interquartile range (IQR) from about 0.74 to 0.80 when using the transcriptomic-only features (Figure 5A). With the incorporation of cell type proportion data, the median AUC increased to approximately 0.80, with a tighter IQR between about 0.78 and 0.83, indicating a significant improvement (p = 0.0322).

Figure 5. Random Forest model performance based on the transcriptomic data and cell type proportions to predict liver fibrosis staging using the Govaere dataset.

Figure 5.

Open in a new tab

Comparison of the Area Under the Curve (AUC) between Transcriptomic Data and enhanced model including cell type proportions (A). The boxplots show AUC values for models based solely on transcriptomic data versus those augmented with cell type proportion data, with a p-value indicating statistical significance in performance differences. Precision (B), recall (C), and F1-score (D) of the Random Forest classifier in distinguishing between different stages of liver fibrosis (MASL/MASH F0, MASH F1-F2, MASH F3-F4) using a variable number of top-ranked gene features filtered based on lowest ANOVA p-values.

Precision

Model precision remained relatively stable across different feature set sizes for all fibrosis stages (Figure 5B). For MASL/MASH F0, precision fluctuated with increasing feature set size before stabilizing around 0.75. The MASH F1-F2 group showed less variation but exhibited lower precision than MASL/MASH F0, reaching stability at approximately 0.65. The MASH F3-F4 group had the highest stabilized recall, around 0.90. Both macro and weighted average precision values improved with increasing feature set sizes, plateauing at approximately 0.80.

Recall

The recall metric demonstrated distinct trends across fibrosis stages (Figure 5C). For MASL/MASH F0, recall remained relatively unchanged with increasing feature set size, stabilizing at approximately 0.65. The MASH F1-F2 and MASH F3-F4 groups exhibited considerable improvements, both reaching stability around 0.80. Macro and weighted average recall values also improved with increasing feature set sizes, leveling off at approximately 0.80.

F1 Score

The F1 score followed a similar pattern to recall across all groups (Figure 5D). The MASL/MASH F0 group stabilized around 0.65, while both MASH F1-F2 and MASH F3-F4 groups improved with increasing feature set sizes, stabilizing at approximately 0.75 and 0.85, respectively. The macro and weighted average F1 scores increased with larger feature sets, reaching approximately 0.80 for feature sets larger than 125.

The top 200 genes used in the Random Forest models were significantly enriched for various biological processes (Figure S3). Among these, ’fat cell differentiation’ was the most significantly represented process, exhibiting the highest gene ratio and an adjusted p-value, suggesting its potential involvement in the pathogenesis or progression of fibrotic changes in the liver. Other enriched processes included ’response to peptide hormone’ and ’regulation of glial cell differentiation.’

Gene Ontology (GO) Enrichment Analysis

Gene Ontology (GO) enrichment analysis identified distinct biological processes associated with fibrosis progression across the Govaere and Pantano datasets, revealing both similarities and differences at each stage of comparison. Only fibrosis stage comparisons were available due to the low number of differentially expressed genes (DEGenes) in the NAS stage DESeq2 analysis. In MASL/MASH F0 vs. Control, upregulated processes in the Govaere dataset included cellular component disassembly, alcohol metabolic process, and various lipid metabolic processes, suggesting early cellular and metabolic reprogramming before fibrosis onset (Figure S4A, column 1). In contrast, upregulated genes in the Pantano dataset were primarily enriched in processes related to cellular and immune activation and structural organization, indicating a more pronounced immune and fibrotic response at early disease stages (Figure S5A, column 1). Downregulated genes in the Govaere dataset for MASL/MASH F0 vs. Control were enriched in epithelial cell differentiation, muscle cell differentiation, fat cell differentiation, and cell chemotaxis, while in the Pantano dataset, downregulated genes were associated with steroid metabolism, terpenoid metabolism, and other metabolic processes (Figure S4A, column 2; Figure S5A, column 2).

In the MASH F1/F2 vs. MASL/MASH F0 comparison, upregulated genes in the Govaere dataset were enriched in response to xenobiotic stimulus, defense against bacteria, and cell recognition, whereas the Pantano dataset showed enrichment in leukocyte chemotaxis, myeloid leukocyte migration, and chemokine responses (Figure S4B, column 1; Figure S5B, column 1). Downregulated genes in the Govaere dataset were associated with response to peptide hormone and response to copper-like ions, while the Pantano dataset also showed downregulation of response to copper-like ions (Figure S4B, column 2; Figure S5B, column 2). In the final comparison, MASH F3/F4 vs. MASH F1/F2, the Govaere dataset exhibited upregulated genes involved in extracellular matrix organization, muscle processes, and kidney development, while the Pantano dataset showed enrichment in positive regulation of cell activation, leukocyte activation, and cell-cell adhesion (Figure S4C, column 1; Figure S5C, column 1). Interestingly, downregulated genes in the Govaere dataset were linked to cell-cell adhesion at the plasma membrane and neutrophil movement, whereas downregulated genes in the Pantano dataset were associated with modulation of chemical synaptic transmission and regulation of membrane potential (Figure S4C, column 2; Figure S5C, column 2).

Discussion

In this study, we utilized the MuSiC deconvolution algorithm to predict the cell type proportions of parenchymal and non-parenchymal cells and their trends across the MASLD spectrum. In addition, we demonstrated that cell type proportions can be used as important features to improve the accuracy of machine learning models in classifying fibrosis stages.

Our study successfully deconvoluted three immune and seven non-immune cell types involved in MASLD progression across both fibrosis and NAS staging. In comparison, a prior deconvolution study by Pantano et al. identified the proportions of four major cell types—hepatocytes, macrophages cholangiocytes, and HSC.20 Our findings not only expanded the range of identified cell types but also revealed trends in two distinct progression trajectories: fibrosis and NAS staging. Our analysis demonstrated highly similar proportional trends with Pantano et al.’s analysis, with a nearly identical downward trajectory for hepatocytes and upward trajectories for macrophages, cholangiocytes, and HSC (Figure 3B). This strong concordance reinforces the reliability of our deconvolution approach.

A key strength of our study is its ability to deconvolute not only major liver cell types but also rarer, functionally distinct subpopulations. Notably, we identified macrophage subtypes, including kupffer cells and TREM2+ macrophages, as well as endothelial subtypes such as sinusoidal and lymphatic endothelium. Additionally, we resolved the heterogeneous subtypes of HSCs, which are implicated in fibrosis and carcinogenesis, potentially through epithelial-mesenchymal transition.21,22 This level of resolution enhances our understanding of cellular dynamics in MASLD progression.

In addition, our findings provide a comprehensive overview of cellular dynamics across MASLD progression. The significant decline in hepatocyte proportions across the MASLD stages likely reflects progressive hepatocyte loss and diminished functionality, as supported by existing literature.23,24 The observed increase in cholangiocytes suggests adaptive responses to bile duct damage, leading to cholangiocyte proliferation and exacerbating liver fibrosis through epithelial-mesenchymal transition.25,26 The pronounced increase in the two subtypes of HSCs across MASLD stages highlights their differential role in the MASLD pathogenesis. Notably, DPT+ HSCs exhibit a more significant elevation in advanced stages of MASH compared to RGS5+ HSCs, which show a steadier rise throughout the spectrum. This suggests that DPT+ HSCs may have a more critical role in the advanced stages of MASH. Future research should further explore the temporal dynamics of these HSC subtypes and their precise contributions to the fibrotic pathway. The increases observed in both endothelial subpopulations point to endothelial remodeling within the liver. Literature supports the growing recognition of endothelial cells in MASH pathogenesis, aligning with our findings. Lymphangiogenesis, known to occur during chronic liver diseases including MASH, may explain the increase in lymphatic endothelial cells.27,28,29 Similarly, sinusoidal endothelial cells are known to undergo dysfunctional changes, with increased involvement during advanced stages of MASH.30,31

The dynamic changes in immune cell populations reflect the immune system’s complex involvement during the progression of MASH. The elevation of TREM2+ macrophages, indicates their increased participation in inflammatory processes throughout MASH. Notably, TREM2+ macrophages have been linked to immunosuppression in cancer, including HCC, making their rise particularly significant.32,33 Additionally, recent studies indicate that hepatic danger signaling triggers the induction of TREM2+ macrophages, driving steatohepatitis through MS4A7-dependent inflammasome activation.34 However, TREM2+ macrophages have also been shown to exert protective effects against fibrosis and inflammation, underscoring their functional complexity in MASH.35,36,37 Furthermore, the activation of plasma B cells in MASH has been previously investigated, with reports confirming their involvement in the disease progression.38,39 This aligns with the increasing trend observed in our deconvolution results, further supporting their role in MASH pathogenesis.

The significant decrease in hepatocyte proportions across NAS stages mirrors the trend observed in fibrosis staging, reinforcing the central role of hepatocyte loss in MASH progression. This consistent decline suggests that hepatocyte proportion may serve as a key indicator of disease severity across NAS staging criteria. Interestingly, while both HSC subtypes are elevated in fibrosis staging, only the RGS5+ HSC subtype shows a significant increase in NAS staging. This selective elevation suggests a more nuanced role for RGS5+ HSCs in driving or exacerbating inflammation in MASH, highlighting distinct functional specializations within HSC subpopulations. A deeper understanding of these specialized roles and their interactions could provide valuable insights into MASLD pathogenesis and uncover novel therapeutic targets for inflammation management. Endothelial cell changes, particularly the significant alterations in lymphatic and sinusoidal endothelium, underscore the importance of vascular remodeling in NAS progression. Notably, sinusoidal endothelial cells exhibit an increasing trend across NAS staging in both datasets, supporting existing literature that highlights their role in maintaining immune homeostasis through the recruitment of immunosuppressive leukocytes.40,41 The significant variation in plasma B cells and TREM2+ macrophages across both fibrosis and NAS staging indicates that these immune cells play a critical role in both fibrotic and inflammatory processes in MASH.

The fewer significantly elevated cell type proportions observed in NAS staging compared to fibrosis staging may be influenced by the distribution of samples. NAS covers eight stages, compared to the five to six stages in fibrosis staging, resulting in fewer samples per stage for NAS. This difference in sample distribution could contribute to the reduced statistical power and fewer significant findings in cell type proportions for NAS staging. However, it is important to consider that biological factors may also contribute to this finding. Fibrosis involves extensive ECM remodeling and changes in various cell types, whereas steatosis, inflammation, and ballooning primarily affect hepatocyte numbers.42,43 Therefore, while sample distribution is a factor, alternative biological explanations must be considered. A larger dataset with more samples per NAS stage could help clarify whether the observed differences are due to statistical power or biological variation across the stages.

Furthermore, not all cell types identified in the single-cell dataset were successfully deconvoluted in the two bulk RNA-seq datasets. Many immune cell populations like the T Cell, B Cell, and Dendritic Cells could not be deconvoluted in either dataset. The inability to deconvolute many immune cells can be attributed to the overlaps of gene expression with other immune cell populations, which can obscure the distinct transcriptional signatures required for accurate deconvolution. The variability in deconvolution success across datasets suggests that some cell types may be inherently more challenging to deconvolute from bulk RNA-seq data.

Our success in enhancing the Random Forest model highlights the value of cell type proportion data in improving the accuracy of predicting fibrosis stages in MASLD. Integrating these proportions with transcriptomic data not only enhances model performance but also offers a more detailed understanding of cell type contributions across different fibrosis stages. Future research could build on this by incorporating additional features, such as other omics data, to further enhance model performance and deepen insights into disease mechanisms. Additionally, investigating the biological processes associated with the most predictive genes could provide important insights into the underlying mechanisms of fibrosis progression, paving the way for more targeted therapeutic interventions and the identification of biomarkers.

Overall, pseudobulk deconvolution results in relatively accurate estimations of cell type proportions (Figure S1). Both major and rare cell types alike showed adequete deconvolution results. Lymphatic endothelium, TREM2+ macrophage, and cell cycle showed relatively greater degree of underestimation than other major cell types. This could be attributed to the overlap of signatures with other similar cell types or due to weaker single cell transcriptional signature. The downsampling analysis shows solid detection of cell types for both major and rare cell types even in heavily downsampled samples (Figure S2). This was evident in both datasets where they differed in sequencing depth and RNA protocol.

Future research should focus on improving deconvolution methods to better distinguish between closely related cell types, potentially through the integration of additional data types like ATAC-seq and spatial transcriptomics. Additionally, selecting datasets with clear stage-specific expression profiles will be crucial for accurately capturing the complexity of cellular changes in disease progression. With improved single-cell RNA-seq data that captures a greater diversity of cell types, we anticipate even better results in future studies. One limitation of our study is the lack of independent validation of our results. Therefore, future research should also focus on further validating these findings in larger, more diverse cohorts to ensure their generalizability. Advanced computational techniques and improved single-cell RNA-seq data could further enhance the deconvolution of cell types and the accuracy of predictive models. Furthermore, implementing advanced algorithms, such as deep learning or neural networks to predict MASLD staging, could potentially reveal even better classification results. Finally, exploring integrating other data types, such as proteomics and metabolomics, could provide a more comprehensive understanding of MASLD.

Conclusion

This study leveraged MuSiC deconvolution to dissect the cellular composition of liver tissues in the progression of MASLD to MASH by observing two progression lines: fibrosis score staging and NAS score staging. By integrating bulk RNA-seq with scRNA-seq data, a comprehensive analysis highlights the significance of specific cell types in disease progression, offering new perspectives for future research and clinical practice.

Supplementary Material

Supinfo

Supplementary Figure 1. Pseudobulk cell type deconvolution of single cell dataset using MuSiC. Fifty pseudobulk samples were generated and deconvoluted using matched single-cell data to assess the performance of the MuSiC algorithm, as described by Payen et al.

Supplementary Figure 2. Downsampling analysis of Govaere et al. (MASL) and Pantano et al (MASH F0) bulk RNA-seq dataset deconvoluted using Payen et al. single-cell dataset. The plots show the proportion of various cell types across downsampled datasets. Nonparametric ANOVA and linear regression analyses were performed to assess statistical differences and trends, with p-values reported for each cell type. This analysis evaluates the consistency of cell type proportions and the robustness of the deconvolution approach under downsampling conditions.

Supplementary Figure 3. Gene Set Enrichment Analysis of the 200 most varied genes. The genes were selected via ANOVA for the lowest p-values, which were used in random forest model to classify the stages of fibrosis.

Supplementary Figure 4. Functional enrichment result of Govaere et al. dataset of upregulated and downregulated DE genes comparing MASL/MASH F0 and Control (A), MASH F1/F2 and MASL/MASH (B), MASH F3/F4 and MASH F1/F2 (C).

Supplementary Figure 5. Functional enrichment result of Pantano et al. dataset of upregulated and downregulated DE genes comparing MASH F0 and Control (A), MASH F1/F2 and MASL (B), MASH F3/F4 and MASH F1/F2 (C).

Acknowledgements

The authors would like to express their sincere gratitude to Dr. Salman Khetani for his guidance and critical review of this manuscript. We also thank Dr. Gregory H. Underhill for his ongoing collaborative efforts that have significantly contributed to this work.

Funding Statement

This research was supported by the National Institute of Diabetes and Digestive and Kidney Diseases of the National Institutes of Health under award number R01 DK115747.

Footnotes

Conflict of Interest Statement

The authors declare that there are no conflicts of interest related to this study.

Data Availability Statement:

The single-cell RNA-seq data used in this study can be accessed from the Gene Expression Omnibus (GEO) under accession number GSE158723. The bulk RNA-seq datasets are available under GSE135251 and GSE162694. The analysis code is available upon reasonable request.

References

  • 1.Miao Lei, et al. “Current status and future trends of the global burden of MASLD.” Trends in Endocrinology & Metabolism (2024). [Google Scholar]
  • 2.Anstee Quentin M., et al. “From NASH to HCC: current concepts and future challenges.” Nature reviews Gastroenterology & hepatology 16.7 (2019): 411–428. [DOI] [PubMed] [Google Scholar]
  • 3.Michelotti Gregory A., Machado Mariana V., and Diehl Anna Mae. “NAFLD, NASH and liver cancer.” Nature reviews Gastroenterology & hepatology 10.11 (2013): 656–665. [DOI] [PubMed] [Google Scholar]
  • 4.McPherson Stuart, et al. “Evidence of NAFLD progression from steatosis to fibrosing-steatohepatitis using paired biopsies: implications for prognosis and clinical management.” Journal of hepatology 62.5 (2015): 1148–1155. [DOI] [PubMed] [Google Scholar]
  • 5.Brunt Elizabeth M., et al. “Nonalcoholic fatty liver disease (NAFLD) activity score and the histopathologic diagnosis in NAFLD: distinct clinicopathologic meanings.” Hepatology 53.3 (2011): 810–820. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Goodman Zachary D. “Grading and staging systems for inflammation and fibrosis in chronic liver diseases.” Journal of hepatology 47.4 (2007): 598–607. [DOI] [PubMed] [Google Scholar]
  • 7.Koch Lisa K., and Yeh Matthew M.. “Nonalcoholic fatty liver disease (NAFLD): diagnosis, pitfalls, and staging.” Annals of diagnostic pathology 37 (2018): 83–90. [DOI] [PubMed] [Google Scholar]
  • 8.Brunt Elizabeth M., et al. “Nonalcoholic steatohepatitis: a proposal for grading and staging the histological lesions.” Official journal of the American College of Gastroenterology| ACG 94.9 (1999): 2467–2474. [Google Scholar]
  • 9.Kleiner David E., et al. “Design and validation of a histological scoring system for nonalcoholic fatty liver disease.” Hepatology 41.6 (2005): 1313–1321. [DOI] [PubMed] [Google Scholar]
  • 10.Ekstedt Mattias, et al. “Low clinical relevance of the nonalcoholic fatty liver disease activity score (NAS) in predicting fibrosis progression.” Scandinavian journal of gastroenterology 47.1 (2012): 108–115. [DOI] [PubMed] [Google Scholar]
  • 11.Titus Alexander J., et al. “Cell-type deconvolution from DNA methylation: a review of recent applications.” Human molecular genetics 26.R2 (2017): R216–R224. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Wang Xuran, et al. “Bulk tissue cell type deconvolution with multi-subject single-cell expression reference.” Nature communications 10.1 (2019): 380. [Google Scholar]
  • 13.Steen Chloé B., et al. “Profiling cell type abundance and expression in bulk tissues with CIBERSORTx.” Stem Cell Transcriptional Networks: Methods and Protocols (2020): 135–157. [Google Scholar]
  • 14.Kulkarni Ashwinikumar, et al. “Beyond bulk: a review of single cell transcriptomics methodologies and applications.” Current opinion in biotechnology 58 (2019): 129–136. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Newman Aaron M., et al. “Determining cell type abundance and expression from bulk tissues with digital cytometry.” Nature biotechnology 37.7 (2019): 773–782. [Google Scholar]
  • 16.Pantano Lorena, et al. “Molecular characterization and cell type composition deconvolution of fibrosis in NAFLD.” Scientific reports 11.1 (2021): 18045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Payen Valéry L., et al. “Single-cell RNA sequencing of human liver reveals hepatic stellate cell heterogeneity.” Jhep Reports 3.3 (2021): 100278. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Govaere Olivier, et al. “Transcriptomic profiling across the nonalcoholic fatty liver disease spectrum reveals gene signatures for steatohepatitis and fibrosis.” Science translational medicine 12.572 (2020): eaba4448. [Google Scholar]
  • 19.Pantano Lorena, et al. “Molecular characterization and cell type composition deconvolution of fibrosis in NAFLD.” Scientific reports 11.1 (2021): 18045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Pantano Lorena, et al. “Molecular characterization and cell type composition deconvolution of fibrosis in NAFLD.” Scientific reports 11.1 (2021): 18045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Barry Anna E., et al. “Hepatic stellate cells and hepatocarcinogenesis.” Frontiers in cell and developmental biology 8 (2020): 709. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Yang Ming-Chen, et al. “Hepatic stellate cells secretes type I collagen to trigger epithelial mesenchymal transition of hepatoma cells.” American journal of cancer research 4.6 (2014): 751. [PMC free article] [PubMed] [Google Scholar]
  • 23.Ribeiro Paulo S., et al. “Hepatocyte apoptosis, expression of death receptors, and activation of NF-κ B in the liver of nonalcoholic and alcoholic steatohepatitis patients.” Official journal of the American College of Gastroenterology| ACG 99.9 (2004): 1708–1717. [Google Scholar]
  • 24.Kanda Tatsuo, et al. “Apoptosis and non-alcoholic fatty liver diseases.” World journal of gastroenterology 24.25 (2018): 2661. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Sato Keisaku, et al. “Mechanisms of cholangiocyte responses to injury.” Biochimica et Biophysica Acta (BBA)-Molecular Basis of Disease 1864.4 (2018): 1262–1269. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Glaser Shannon S., et al. “Cholangiocyte proliferation and liver fibrosis.” Expert reviews in molecular medicine 11 (2009): e7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Tamburini Beth A. Jiron, et al. “Chronic liver disease in humans causes expansion and differentiation of liver lymphatic endothelial cells.” Frontiers in immunology 10 (2019): 1036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Vollmar Brigitte, et al. “Lymph vessel expansion and function in the development of hepatic fibrosis and cirrhosis.” The American journal of pathology 151.1 (1997): 169. [PMC free article] [PubMed] [Google Scholar]
  • 29.Jeong Jain, Tanaka Masatake, and Iwakiri Yasuko. “Hepatic lymphatic vascular system in health and disease.” Journal of hepatology 77.1 (2022): 206–218. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Miyao Masashi, et al. “Pivotal role of liver sinusoidal endothelial cells in NAFLD/NASH progression.” Laboratory investigation 95.10 (2015): 1130–1144. [DOI] [PubMed] [Google Scholar]
  • 31.Hammoutene Adel, and Rautou Pierre-Emmanuel. “Role of liver sinusoidal endothelial cells in non-alcoholic fatty liver disease.” Journal of hepatology 70.6 (2019): 1278–1291. [DOI] [PubMed] [Google Scholar]
  • 32.Tan Jizhou, et al. “TREM2+ macrophages suppress CD8+ T-cell infiltration after transarterial chemoembolisation in hepatocellular carcinoma.” Journal of hepatology 79.1 (2023): 126–140. [DOI] [PubMed] [Google Scholar]
  • 33.Park Matthew D., et al. “TREM2 macrophages drive NK cell paucity and dysfunction in lung cancer.” Nature immunology 24.5 (2023): 792–801. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Zhou Linkang, et al. “Hepatic danger signaling triggers TREM2+ macrophage induction and drives steatohepatitis via MS4A7-dependent inflammasome activation.” Science Translational Medicine 16.738 (2024): eadk1866. [Google Scholar]
  • 35.Hendrikx Tim, et al. “Soluble TREM2 levels reflect the recruitment and expansion of TREM2+ macrophages that localize to fibrotic areas and limit NASH.” Journal of hepatology 77.5 (2022): 1373–1385. [DOI] [PubMed] [Google Scholar]
  • 36.Luo Qiujie, et al. “TREM2 insufficiency protects against pulmonary fibrosis by inhibiting M2 macrophage polarization.” International immunopharmacology 118 (2023): 110070. [DOI] [PubMed] [Google Scholar]
  • 37.Sun Huifang, Feng Jianguo, and Tang Liling. “Function of TREM1 and TREM2 in liver-related diseases.” Cells 9.12 (2020): 2626. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Barrow Fanta, et al. “Microbiota-driven activation of intrahepatic B cells aggravates NASH through innate and adaptive signaling.” Hepatology 74.2 (2021): 704–722. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Barrow Fanta, et al. “The emerging role of B cells in the pathogenesis of NAFLD.” Hepatology 74.4 (2021): 2277–2286. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Hammoutene Adel, and Rautou Pierre-Emmanuel. “Role of liver sinusoidal endothelial cells in non-alcoholic fatty liver disease.” Journal of hepatology 70.6 (2019): 1278–1291. [DOI] [PubMed] [Google Scholar]
  • 41.Wilkinson Alex L., Qurashi Maria, and Shetty Shishir. “The role of sinusoidal endothelial cells in the axis of inflammation and cancer within the liver.” Frontiers in physiology 11 (2020): 990. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Friedman Scott L. “Mechanisms of hepatic fibrogenesis.” Gastroenterology 134.6 (2008): 1655–1669. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Sanyal Arun J. “Past, present and future perspectives in nonalcoholic fatty liver disease.” Nature reviews Gastroenterology & hepatology 16.6 (2019): 377–386. [DOI] [PubMed] [Google Scholar]
  • 44.Gómez Hernández Gonzalo, Morell María, and Alarcón-Riquelm Marta E.. “The role of BANK1 in B cell signaling and disease.” Cells 10.5 (2021): 1184. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Ernst James A., et al. “Isolation and characterization of the B-cell marker CD20.” Biochemistry 44.46 (2005): 15150–15158. [DOI] [PubMed] [Google Scholar]
  • 46.Dougherty JD, et al. “PBK/TOPK, a proliferating neural progenitor-specific mitogen-activated protein kinase kinase.” Journal of Neuroscience 25.46 (2005): 10773–10785. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Uusküla-Reimand Liis, and Wilson Michael D.. “Untangling the roles of TOP2A and TOP2B in transcription and cancer.” Science advances 8.44 (2022): eadd4920. [Google Scholar]
  • 48.Sun Tung-Tien, et al. “Keratin classes: molecular markers for different types of epithelial differentiation.” Journal of Investigative Dermatology 81.1 (1983): S109–S115. [Google Scholar]
  • 49.Schnell Ulrike, Cirulli Vincenzo, and Giepmans Ben NG. “EpCAM: structure and function in health and disease.” Biochimica et Biophysica Acta (BBA)-Biomembranes 1828.8 (2013): 1989–2001. [DOI] [PubMed] [Google Scholar]
  • 50.Piccioli Diego, et al. “Functional specialization of human circulating CD16 and CD1c myeloid dendritic-cell subsets.” Blood, The Journal of the American Society of Hematology 109.12 (2007): 5371–5379. [Google Scholar]
  • 51.Heger Lukas, et al. “CLEC10A is a specific marker for human CD1c+ dendritic cells and enhances their toll-like receptor 7/8-induced cytokine secretion.” Frontiers in immunology 9 (2018): 744. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Bachem Annabell, et al. “Expression of XCR1 characterizes the Batf3-dependent lineage of dendritic cells capable of antigen cross-presentation.” Frontiers in immunology 3 (2012): 214. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Mellor Andrew L., and Munn David H.. “IDO expression by dendritic cells: tolerance and tryptophan catabolism.” Nature Reviews Immunology 4.10 (2004): 762–774. [Google Scholar]
  • 54.Cao Wei, et al. “Regulation of TLR7/9 responses in plasmacytoid dendritic cells by BST2 and ILT7 receptor interaction.” Journal of Experimental Medicine 206.7 (2009): 1603–1614. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Cheng Menglan, et al. “Characterization of species-specific genes regulated by E2–2 in human plasmacytoid dendritic cells.” Scientific Reports 5.1 (2015): 10752. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Bahrami Arya J., et al. “Regulator of G-protein signaling-5 is a marker of hepatic stellate cells and expression mediates response to liver injury.” PLoS One 9.10 (2014): e108505. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Zhang Mengfan, et al. “The C-type lectin COLEC10 is predominantly produced by hepatic stellate cells and involved in the pathogenesis of liver fibrosis.” Cell Death & Disease 14.11 (2023): 785. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58.Buechler Matthew B., et al. “Cross-tissue organization of the fibroblast lineage.” Nature 593.7860 (2021): 575–579. [DOI] [PubMed] [Google Scholar]
  • 59.MacParland Sonya A., et al. “Single cell RNA sequencing of human liver reveals distinct intrahepatic macrophage populations.” Nature communications 9.1 (2018): 4383. [Google Scholar]
  • 60.Martínez-Jiménez Celia P., et al. “Transcriptional regulation and expression of CYP3A4 in hepatocytes.” Current drug metabolism 8.2 (2007): 185–194. [DOI] [PubMed] [Google Scholar]
  • 61.Rothschild Marcus A., Oratz Murray, and Schreiber Sidney S.. “Serum albumin.” Hepatology 8.2 (1988): 385–401. [DOI] [PubMed] [Google Scholar]
  • 62.Pan Yanfang, and Yago Tadayuki. “Transcriptional regulation of podoplanin expression by Prox1 in lymphatic endothelial cells.” Microvascular research 94 (2014): 96–102 [DOI] [PubMed] [Google Scholar]
  • 63.Bianchi Roberta, et al. “Postnatal deletion of podoplanin in lymphatic endothelium results in blood filling of the lymphatic system and impairs dendritic cell migration to lymph nodes.” Arteriosclerosis, thrombosis, and vascular biology 37.1 (2017): 108–117. [DOI] [PubMed] [Google Scholar]
  • 64.Ulvmar Maria H., et al. “The atypical chemokine receptor CCRL1 shapes functional CCL21 gradients in lymph nodes.” Nature immunology 15.7 (2014): 623–630. [DOI] [PubMed] [Google Scholar]
  • 65.Russo Erica, et al. “Intralymphatic CCL21 promotes tissue egress of dendritic cells through afferent lymphatic vessels.” Cell reports 14.7 (2016): 1723–1734. [DOI] [PubMed] [Google Scholar]
  • 66.MacParland Sonya A., et al. “Single cell RNA sequencing of human liver reveals distinct intrahepatic macrophage populations.” Nature communications 9.1 (2018): 4383. [Google Scholar]
  • 67.Ramachandran P, et al. “Resolving the fibrotic niche of human liver cirrhosis at single-cell level.” Nature 575.7783 (2019): 512–518. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Zigmond Ehud, et al. “Infiltrating monocyte-derived macrophages and resident kupffer cells display different ontogeny and functions in acute liver injury.” The Journal of Immunology 193.1 (2014): 344–353. [DOI] [PubMed] [Google Scholar]
  • 69.Nakamura Kyohei, and Smyth Mark J.. “TREM2 marks tumor-associated macrophages.” Signal Transduction and Targeted Therapy 5.1 (2020): 233. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 70.Ripoll Vera M., et al. “Gpnmb is induced in macrophages by IFN-γ and lipopolysaccharide and acts as a feedback regulator of proinflammatory responses.” The Journal of Immunology 178.10 (2007): 6557–6566. [DOI] [PubMed] [Google Scholar]
  • 71.Ryckman Carle, et al. “Proinflammatory activities of S100: proteins S100A8, S100A9, and S100A8/A9 induce neutrophil chemotaxis and adhesion.” The Journal of Immunology 170.6 (2003): 3233–3242. [DOI] [PubMed] [Google Scholar]
  • 72.Zhao C, et al. “POU2AF1, an amplification target at 11q23, promotes growth of multiple myeloma cells by directly regulating expression of a B-cell maturation factor, TNFRSF17.” Oncogene 27.1 (2008): 63–75. [DOI] [PubMed] [Google Scholar]
  • 73.Nutt Stephen L., et al. “The generation of antibody-secreting plasma cells.” Nature Reviews Immunology 15.3 (2015): 160–171. [Google Scholar]
  • 74.Olsavszky Victor, et al. “Exploring the transcriptomic network of multi-ligand scavenger receptor Stabilin-1-and Stabilin-2-deficient liver sinusoidal endothelial cells.” Gene 768 (2021): 145284. [DOI] [PubMed] [Google Scholar]
  • 75.Li Zhi-Wen, et al. “Oit3, a promising hallmark gene for targeting liver sinusoidal endothelial cells.” Signal Transduction and Targeted Therapy 8.1 (2023): 344. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 76.Clevers JC, et al. “The T cell receptor/CD3 complex: a dynamic protein ensemble.” Annual review of immunology 6 (1988): 629–662. [Google Scholar]
  • 77.Ruggeri ZM “Von Willebrand factor, platelets and endothelial cell interactions.” Journal of thrombosis and haemostasis 1.7 (2003): 1335–1342. [DOI] [PubMed] [Google Scholar]
  • 78.Thiriot Aude, et al. “Differential DARC/ACKR1 expression distinguishes venular from non-venular endothelial cells in murine tissues.” BMC biology 15 (2017): 1–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 79.Liao Zheting, et al. “Single-cell transcriptome analysis reveals aberrant stromal cells and heterogeneous endothelial cells in alcohol-induced osteonecrosis of the femoral head.” Communications Biology 5.1 (2022): 324. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supinfo

Supplementary Figure 1. Pseudobulk cell type deconvolution of single cell dataset using MuSiC. Fifty pseudobulk samples were generated and deconvoluted using matched single-cell data to assess the performance of the MuSiC algorithm, as described by Payen et al.

Supplementary Figure 2. Downsampling analysis of Govaere et al. (MASL) and Pantano et al (MASH F0) bulk RNA-seq dataset deconvoluted using Payen et al. single-cell dataset. The plots show the proportion of various cell types across downsampled datasets. Nonparametric ANOVA and linear regression analyses were performed to assess statistical differences and trends, with p-values reported for each cell type. This analysis evaluates the consistency of cell type proportions and the robustness of the deconvolution approach under downsampling conditions.

Supplementary Figure 3. Gene Set Enrichment Analysis of the 200 most varied genes. The genes were selected via ANOVA for the lowest p-values, which were used in random forest model to classify the stages of fibrosis.

Supplementary Figure 4. Functional enrichment result of Govaere et al. dataset of upregulated and downregulated DE genes comparing MASL/MASH F0 and Control (A), MASH F1/F2 and MASL/MASH (B), MASH F3/F4 and MASH F1/F2 (C).

Supplementary Figure 5. Functional enrichment result of Pantano et al. dataset of upregulated and downregulated DE genes comparing MASH F0 and Control (A), MASH F1/F2 and MASL (B), MASH F3/F4 and MASH F1/F2 (C).

NIHMS2058017-supplement-Supinfo.pdf (1.5MB, pdf)

Data Availability Statement

The single-cell RNA-seq data used in this study can be accessed from the Gene Expression Omnibus (GEO) under accession number GSE158723. The bulk RNA-seq datasets are available under GSE135251 and GSE162694. The analysis code is available upon reasonable request.

RESOURCES