An unbiased ranking of murine dietary models based on their proximity to human metabolic dysfunction-associated steatotic liver disease (MASLD)

Abstract

Metabolic dysfunction-associated steatotic liver disease (MASLD), previously known as non-alcoholic fatty liver disease, encompasses steatosis and metabolic dysfunction-associated steatohepatitis (MASH), leading to cirrhosis and hepatocellular carcinoma. Preclinical MASLD research is mainly performed in rodents; however, the model that best recapitulates human disease is yet to be defined. We conducted a wide-ranging retrospective review (metabolic phenotype, liver histopathology, transcriptome benchmarked against humans) of murine models (mostly male) and ranked them using an unbiased MASLD 'human proximity score' to define their metabolic relevance and ability to induce MASH-fibrosis. Here, we show that Western diets align closely with human MASH; high cholesterol content, extended study duration and/or genetic manipulation of disease-promoting pathways are required to intensify liver damage and accelerate significant (F2+) fibrosis development. Choline-deficient models rapidly induce MASH-fibrosis while showing relatively poor translatability. Our ranking of commonly used MASLD models, based on their proximity to human MASLD, helps with the selection of appropriate in vivo models to accelerate preclinical research.

Fig. 1. Study design.

In this study, we collected retrospective information from 598 animals (509 WT/GA mice, 89 WT/GA rats): 336 animals subjected to treatment (MASLD-inducing conditions: 315 animals; or CCl₄: 21 animals) and 262 animals as controls for MASLD-inducing conditions (247 animals) or CCl₄ (15 animals), returning 39 models (that is, study designs) aimed at modeling MASLD, and two time points for CCl₄ (positive controls of MASLD-independent fibrosis). Details of the study designs (numerosity, species, background, genetic manipulation, diet, time point and room temperature) are provided in Supplementary Table 1. For all the studies, phenotypic information (Supplementary Table 5), centralized histopathology assessment (Supplementary Table 6) and liver transcriptomics (Supplementary Table 4) were available. These data were integrated into an unbiased binary score (MHPS) ranking the models in terms of their metabolic relevance and ability to induce MASH-fibrosis. Created with BioRender (agreement number GG26BHMS6Y).

Fig. 2. Phenotypic and histologic characterization of the models.

Phenotypic changes observed in the MASLD models compared to their matched controls were profiled as the log₂ fold change (log₂FC) across measures of BW, blood triglycerides (TGs) and cholesterol, LW:BW% ratio, and ALT and AST. The red–blue color gradient indicates the level of increase–decrease of the measure in the MASLD models compared to their controls, while an asterisk indicates a significant change at P < 0.05 (two-sided Mann–Whitney U-test). The two panels of horizontal bars give an overview of the complete histological profiles, in which the total length indicates the activity score (CRN NAS) and fibrosis. In addition, NAS components (steatosis, ballooning and inflammation) are represented by the stacked bar (yellow, green and blue, respectively) lengths. All models are grouped according to their macro-categories (detailed by the leftmost annotations). Source data

Fig. 3. DRPs in human and murine MASLD.

Selection of KEGG-affected pathways characterizing the ‘early disease development’, ‘all disease stages’ and ‘disease progression’ groups. The ‘early disease development’ group includes statistically significant (P < 0.05) modulated pathways in ‘mild vs controls’ but not in ‘moderate–severe vs mild’ human disease stages comparisons. The ‘all disease stages’ group includes statistically significant and homogeneously modulated pathways at all human disease stages (‘mild vs controls’ and ‘moderate–severe vs mild’ in both UCAM–VCU and EPoS); the ‘disease progression’ group includes homogeneously modulated and statistically significant pathways in ‘moderate–severe vs mild’ comparisons in both UCAM–VCU and EPoS but not in ‘mild vs controls’ comparisons. FGSEA calculated a normalized enrichment score (NES; the enrichment score normalized to mean enrichment of random samples of the same size) for each pathway and determined statistical significance using permutation testing (two-sided), adjusting for multiple comparisons to control the false discovery rate (FDR). The human and murine datasets are represented in a color-scale matrix showing the NES. * and $ symbols denote statistical significance ($, P < 0.05; *, FDR < 0.05). All models are grouped according to their macro-categories, as indicated by the panel on the top of the heatmap. Source data

Fig. 4. Agreement of murine DEGs and DRPs with human MASLD.

Heatmap showing the agreement between murine MASLD models and human data based on the list of significant DEGs, DRPs, or the whole transcriptome. The percentage (%) of agreement between murine and human datasets defines the proportion of DEGs and DRPs statistically modulated and in the same direction compared to the human reference datasets (defined for ‘early disease development’, ‘all disease stages’ and ‘disease progression’ comparisons) or the proportion of DEGs statistically modulated and in the same direction compared to the human disease stage comparisons (defined for the whole transcriptome). All models are grouped according to their macro-categories as indicated in the graphic legend of the figure. Data are represented in a color-scale matrix showing the percentage of agreement and refer to DEGs (Supplementary Table 4; whole dataset or genes defined for ‘early disease development’, ‘all disease stages’ and ‘disease progression’, respectively) and DRPs (Fig. 3). In parenthesis, we show the results of the hypergeometric test (one-sided) performed on the same comparison groups, indicating the statistical significance (NS, non-significant (P > 0.05)). Source data

Fig. 5. Highly performing genes in human and murine MASLD.

Selection of statistically significant DEGs (complete list in Supplementary Table 4) enriching statistically significant modulated pathways (complete list in Fig. 3), thus being highly biologically relevant genes associated with the different stages of MASLD development or progression. To assess the statistical significance between the compared groups, a Wald test statistic (two-sided hypothesis testing) was deployed to compare the coefficients of explanatory variables in a regression model, representing the gene expression differences among the compared groups. The human and murine datasets are represented in a color-scale matrix showing the log2FC. * and $ symbols denote statistical significance ($, P < 0.05; *, adjusted P < 0.05 (Benjamini–Hochberg correction)). As the graphic legend indicates, all models were grouped according to their macro-categories. Source data

Fig. 6. The MHPS—metabolic relevance and progressive MASLD.

a,b, Comparison of the MASLD models, performed based on the MHPS that incorporates the PHPS (details in Supplementary Table 5), the HHPS (details in Supplementary Table 6) and the DHPS (see Supplementary Fig. 4). The average of these normalized scores (MHPS) ranks the murine models (from high to low) based on their metabolic relevance (a) or their ability to induce MASH-fibrosis (b). A detailed description of the different components is provided in Extended Data Figs. 3 and 4, respectively. For both a and b, the total length of the horizontal bars indicates the MHPS, while the length of the stacks within each bar indicates the relative contribution from the three evidence layers: PHPS, HHPS and DHPS. A reference panel to the right indicates BW (significant increases in red; decreases in blue) and fibrosis score (*). Macro-categories are indicated by the color panel to the left of the plots. c, Correlation among the two MHPS outputs (‘metabolic relevance’ vs ‘ability to induce MASH-fibrosis). Specific models are highlighted based on their performance within the two rankings. Yellow dots represent models that score high with both rankings and represent the best approximation to human MASH. Red dots are models that score highly for metabolic relevance but are less relevant for MASH-fibrosis. Grey dots are models that score highly for MASH-fibrosis but have less metabolic relevance. Panels a and b provide a specific reference to the position in the scatter. Source data

Fig. 7. Key experimental components contributing to MHPS ‘metabolic relevance’ and ‘ability to induce MASH-fibrosis’ outputs.

Relationship between study design parameters and the MHPS of metabolic relevance and ability to induce MASH-fibrosis evaluated by PLSR. a, VIP for each study parameter (among those detailed in Supplementary Table 1) for the two components of the model. Parameters contributing the most to the PLSR model are characterized by having VIPs > 1 (black solid line) in any of the two components of the model. b, Clustered heatmap of the most influential study parameters (VIP > 1), indicating their correlation with the MHPS metabolic relevance and ability to induce MASH-fibrosis (color-scale matrix showing a positive correlation represented in red; negative correlation represented in blue). Source data

Fig. 8. Effects of treatments in a selection of WD and choline-deficient models.

Response of the best-ranked diet (GAN-C2), another WD (AMLN-C2) and a CDHFD (CDAHFD-F45) to treatments mimicking lifestyle intervention (CR, caloric restriction; REV, chow reversal) and semaglutide pharmacological treatment (SEMA, 30 nmol per kg per day). a, Simplified study designs (details in Supplementary Table 7). b, Effect on the phenotype and histology. Phenotypic changes observed in the treatment or dietary models compared to their matched dietary models or controls, respectively, were profiled as log₂FC across measures of BW, blood TGs and cholesterol, LW:BW%, and ALT and AST. The red–blue color gradient indicates the level of increase–decrease of the measure in the models compared to their respective controls; * indicates statistical significance (P < 0.05; two-sided Mann–Whitney U-test). For the histological changes, the Mann–Whitney U-test was used to calculate P values for the differences in the ordinal scores (P < 0.05 are shown with *). The color scale indicates the signed P value: −log₁₀(P value) for up-regulation and +log₁₀(P value) for down-regulation. c, Effect of treatments on a selection of biologically relevant DEGs as described in Fig. 5. The human and murine datasets are represented in a color-scale matrix showing the log₂FC. * and $ symbols denote statistical significance (two-sided Wald test statistic and adjustment for multiple testing using the Benjamini–Hochberg correction; $: P < 0.05; *: adjusted P < 0.05). d, Effect of treatments on pathways as described in Fig. 3. FGSEA calculated a NES for each pathway and determined statistical significance using permutation testing (two-sided), adjusting for multiple comparisons to control the FDR. The human and murine datasets are represented in a color-scale matrix showing the NES. $, P < 0.05; *, FDR < 0.05). Source data

Extended Data Fig. 1. PCA plot of RNASeq data comparing the different mouse MASLD models.

The plot uses the normalised batch-corrected gene expression data. Each point represents a mouse model. The Controls (triangles) are clearly separated from the MASLD models (circles) according to the first principal component (PC1). Source data

Extended Data Fig. 2. PCA plot of RNASeq data comparing the different rat MASLD models.

The plot uses the normalised batch-corrected gene expression data. Each point represents a rat model. The Controls (triangles) are separated from the MASLD models (circles) according to the first principal component (PC1). Source data

Extended Data Fig. 3. Details of the MHPS components (PHPS, HHPS, DHPS) – Metabolic Relevance.

Bar plots show the detailed composition of the MHPS components (PHPS, HHPS, and DHPS) utilised for the output focusing on metabolic relevance. Models are ordered globally according to the MHPS, where each barplot gives an overview of the total PHPS, HHPS or DHPS, the associated features, and their corresponding scores (see Supplementary Tables 5 and 6 and Supplementary Fig. 4 for an overview of the scoring strategies). For each model, the total scores of PHPS and HHPS are represented by red vertical lines to overcome the fact that some features are scored negatively. For PHPS and HHPS, the length of the stacked bars represents the scores’ absolute values (before normalising to the interval 0-1). DHPS directly shows the normalised score. Feature scores included are abbreviated as follows: PHPS: BW = Body weight, TG = Triglycerides, CL = Cholesterol, L/B = Liver Weight/Body weight (%), LE = Liver enzyme levels of ALT & AST; HHPS: C1 = Topography of histological lesions, C2 = Type of steatosis, C3 = Hepatocyte ballooning, C4 = Lobular inflammation, C5 = Mallory-Denk bodies; DHPS: DEG = Differentially Expressed Genes, DRP = Differentially Regulated Pathways. Source data

Extended Data Fig. 4. Details of the MHPS components (PHPS, HHPS, DHPS) – Ability to induce MASH-Fibrosis.

Bar plots show the detailed composition of the MHPS components (PHPS, HHPS, and DHPS) for the output focusing on the ability to induce MASH-Fibrosis. Models are ordered globally according to the MHPS, where each barplot gives an overview of the total PHPS, HHPS or DHPS, associated features, and corresponding scores (see Supplementary Tables 5 and 6 and Supplementary Fig. 4 for an overview of the scoring strategies). For each model, the total scores of PHPS and HHPS are represented by red vertical lines to overcome the fact that some features are scored negatively. For PHPS and HHPS, the length of the stacked bars represents absolute scores' absolute values (before normalising to the interval 0-1). DHPS shows the normalised score directly. Feature scores included are abbreviated as follows: PHPS: BW = Body weight, TG = Triglycerides, CL = Cholesterol, L/B = Liver Weight/Body weight (%), LE = Liver enzyme levels of ALT & AST; HHPS: C1 = Topography of histological lesions, C2 = Type of steatosis, C3 = Hepatocyte ballooning, C4 = Lobular inflammation, C5 = Mallory-Denk bodies, F(C6) = Fibrosis; and for DHPS: DEG = Differentially Expressed Genes, DRP = Differentially Regulated Pathways. Source data

Extended Data Fig. 5. Representative images of MASLD histopathology features in the 6NTAC-GAN-C2-TN42W model.

a. Panlobular distribution of steatosis, grade 3, H&E stain, x50; b. Ballooned hepatocyte (arrow), H&E stain, x200; c. Necroinflammatory focus (arrowhead), H&E stain, x200; d. Periportal and extensive sinusoidal fibrosis, stage 2, Sirius red stain, x50 (THV terminal hepatic venule, PT portal tract).

Extended Data Fig. 6. Representative images of MASLD histopathology features in the 6NTAC-GAN-C2-RT42W model.

a. Panlobular distribution of steatosis, grade 3, H&E stain, x50; b. Necroinflammatory foci (arrowheads), H&E stain, x200; c. Ballooned Hepatocyte (arrow), H&E stain, x200; d. Periportal and perisinusoidal fibrosis, stage 2, Sirius red stain, x50 (THV terminal hepatic venule, PT portal tract).

Extended Data Fig. 7. Representative images of MASLD histopathology features in the AZ-OB-GAN-23W model.

a. Panlobular distribution of steatosis, grade 3, H&E stain, x50; b. Non-classical ballooned hepatocyte (arrow) and necroinflammatory foci (arrowheads), H&E stain, x200; c. Periportal and perisinusoidal fibrosis, stage 2, Sirius red stain, x50 (THV terminal hepatic venule, PT portal tract).

Extended Data Fig. 8. Representative images of MASLD histopathology features in the OB-GAN-C2-12W model.

a. Panlobular distribution of steatosis, grade 3, H&E stain, x50; b. Necroinflammatory focus (arrowhead), H&E stain, x200; c. Ballooned hepatocyte (arrow), H&E stain, x200; d. Periportal and perisinusoidal fibrosis, stage 2, Sirius red stain, x50 (THV terminal hepatic venule, PT portal tract).

Extended Data Fig. 9. Representative images of MASLD histopathology features in the MC4R-WD-C0.2-21W model.

a. Panlobular distribution of steatosis, grade 3, H&E stain, x50; b. Necroinflammatory foci (arrowheads), H&E stain, x200; c. Ballooned hepatocytes (arrows), H&E stain, x200; d. Periportal and perisinusoidal fibrosis, stage 2, Sirius red fast green stain, x50 (THV terminal hepatic venule, PT portal tract).

Extended Data Fig. 10. Representative images of MASLD histopathology features in the LDLR-CDHFHCD-F42-C1-8W model.

a. Panlobular distribution of steatosis, grade 3, H&E stain, x50; b. Periportal and perisinusoidal fibrosis, stage 2, Sirius red stain, x50; c) Necroinflammatory foci (arrowheads), H&E stain, x200; d. Ballooned hepatocyte (arrow), H&E stain, x200 THV terminal hepatic venule, PT portal tract).