Aging Promotes Metabolic Dysfunction-associated Steatotic Liver Disease by Inducing Ferroptotic Stress

Abstract

Aging is an inevitable consequence of living that undermines cellular resiliency to cause tissue degeneration and eventual multi-organ dysfunction. Susceptibility to biological consequences of aging varies among organs and individuals. Stressors that challenge resiliency unmask these differences. Hence, aging increases the risk for failure of many metabolically-stressed organs and exacerbates liver degeneration related to obesity and diabetes. We discovered that aging promotes ferroptosis, a type of metabolism-regulated cell death, in hepatocytes. Hepatocytes that have launched the ferroptotic death program adapt their metabolism to survive but become damaged and dysfunctional. Metabolic stressors amplify this, increasing the burden of ferroptosis-adapted cells and liver damage. Blocking ferroptotic signaling in old livers reduces accumulation of adapted hepatocytes and reverts livers to a more youthful resilient state despite exogenous stressors. Ferroptosis is also induced in other organs that are damaged by chronic metabolic stress, identifying ferroptosis as a tractable conserved mechanism for aging-related multi-organ dysfunction.

INTRODUCTION

Worldwide, about two thirds of the people who die each day die of age-related causes. That proportion approaches 90% in industrialized nations with populations that are enriched with senior and geriatric citizens¹. Aging has a global impact on the burden of disease because it promotes metabolic stress, tissue dysfunction, organ degeneration and death². Susceptibility to the effects of aging is controlled by inherited factors and social determinants of health that determine resilience, i.e., how well the body copes with stressors. Consequently, biological aging may proceed at a faster or slower rate than predicted by chronological age alone³. Factors that accelerate biological aging amplify the incidence and prevalence of metabolic dysfunction and organ degeneration attributed to obesity and diabetes, related conditions that impose major health and economic burdens³. The health and longevity of model organisms can be extended significantly by interventions that retard or reverse cumulative epigenetic effects of aging in cells^4,5. Alleviating forces that accelerate aging may also be an effective strategy to extend the health span in people^6,7, and merits consideration given the ongoing obesity and diabetes pandemics. However, characterization of aging-related transcriptomic and epigenetic changes has mostly been conducted using samples of blood^3,4,6–8, a tissue with high turnover. Hence, it is unclear when and how the accelerated aging process is best manipulated to slow (or reverse) organ degeneration in people. In order to develop diagnostic and therapeutic targets for aging-related multi-organ dysfunction/degeneration, it is necessary to define tractable mechanisms that promote aging and determine whether they are conserved, both across tissues within an individual and among individuals with similar types of degenerative disease.

To address these knowledge gaps, we determined how aging effects hepatocytes. We selected these cells as models for proof-of-principle experiments for several reasons. First, hepatocytes critically regulate systemic metabolism and energy balance^9,10. Second, the healthy liver’s regenerative capacity is greater than that of any other vital organ and thus, liver has inherent mechanisms that protect it from aging¹¹. Third, these mechanisms can be overwhelmed because the incidence and prevalence of the most common type of degenerative liver disease, metabolic dysfunction-associated steatotic liver disease (MASLD), are increasing in parallel with population aging and other diseases that associate with aging-exacerbated metabolic dysfunction (e.g., obesity and type 2 diabetes)^12,13. Finally, changes in the epigenome of DNA in blood suggest the rate of biological aging is increased in MASLD patients and greatest in those with the most severe liver degeneration¹⁴. Our new results demonstrate that aging promotes ferroptosis, a specific type of regulatable, iron-dependent cell death in hepatocytes, and indicate that this death process is also increased in other organs that are experiencing aging-related degeneration. Further, the findings show how hepatocytes that have launched this death signaling drive liver degeneration. Most importantly, the data reveal that preventing ferroptosis reverses both metabolic dysfunction and liver degeneration to revert old livers to a more youthful and resilient state, even when extrinsic metabolic stressors persist.

RESULTS

Hepatocytes in old livers are enriched with pathways related to tissue degeneration

To clarify how aging effects hepatocytes, we performed RNA-seq analysis in primary hepatocytes from three young (3 months old) and three old (2 years old) mice⁵. Principle component analysis demonstrated tight clustering of transcriptomes within each age group and revealed significant differences in the transcriptomes of the young versus old groups of hepatocytes (Figs 1A, B). Gene expression analysis identified over 4200 genes that were significantly up-regulated, and over 4800 genes that were significantly down-regulated, in the old versus young mouse hepatocytes (Fig 1C). Remarkably, six of the top 10 KEGG pathways that were differentially up-regulated in the old mouse hepatocytes relate to degenerative diseases (e.g., Prion disease, Parkinson’s disease, Huntington disease, Alzheimer disease, Amyotropic lateral sclerosis, Pathways of neurodegeneration) and three of the top 10 KEGG pathways that were down-regulated in the old mouse hepatocytes promote longevity and cell proliferation (Wnt signaling, Longevity regulating pathway, FOXO signaling) (Suppl Fig 1).

Figure 1. Liver accumulation of biologically-old hepatocytes parallels MASLD severity in people.

(A) Primary hepatocytes were isolated from young and old mouse livers and sent for global transcriptomic profiling. (B) Principle component analysis (PCA) of transcriptomes of young versus old hepatocytes (n = 3 mice/group). (C) DEGs with an adjusted p value of < 0.05 and log2 fold change >3 in old hepatocytes were used to construct the aging hepatocyte gene signature (AHGS, 496 DEGs). (D) Scatter plots of deconvolution analysis using AHGS in bulk liver RNA-seq data of Duke MASLD human cohort (GSE213623) demonstrated that AHGS correlates with BMI in both control and MASLD cohorts, (E) but correlates with chronological age in only MASLD patients. (F) Scatter plots show AHGS correlates positively with serum AST, Fib4 score but inversely with albumin levels in MASLD patients. (G) Boxplots of AHGS progressively increases with hepatocyte ballooning, portal inflammation and fibrosis severity during MASLD progression, and distinguishes patients with advanced fibrosis (F3F4) from healthy controls. AHGS was also applied to deconvolute liver transcriptomic data set from a Japanese MASLD cohort (GSE33814) (H) and a European MASLD Cohort (GSE135251) (I). In both cohorts, the AHGS is enriched in patients with MASH, and distinguishes them from those with only ‘simple’ steatosis. Boxplots showed the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. p-values were calcualted using Pearson Correlation test in D, E, F, and Wilcoxon Rank Sum test in G, H, I.

Aging Hepatocyte Gene Signature distinguishes hepatocytes from old and young livers

Next, we constructed an aging hepatocyte gene signature (AHGS) using the transcriptomic data from our primary mouse hepatocytes. We reasoned that the optimal gene signature should only include genes that meet stringent criteria for significance and that are validated in independent datasets and shown to maximize prediction power. By applying cutoffs of adjusted p-values (0.05 or 0.01) with log2FoldChange (from 1 to 5) to the upregulated differentially expressed genes (DEGs) in old hepatocytes, we generated multiple gene lists containing various numbers of DEGs ranging from 2,651 (cutoff: p = 0.05 & Log2FC>1) to 276 (cutoff: p = 0.01 & Log2FC>5) (Suppl Fig 2A). We found that the AHGS with an adjusted p value of < 0.01 & log2 fold change >3 has the best balance between size of gene set and high Jaccard similarity index between two independent datasets (Suppl Fig 2B, see Methods and Materials section for detailed description). This cutoff resulted in an AHGS of 496 genes (Fig 1C, Suppl Table 1), which was used for further analysis. Characterization of this AHGS by GSEA demonstrated that it is enriched with genes related to metal metabolism and antigen binding, immune response, cell stress, and cell death (Suppl Figs 3A, B), and correlates with pathways involved in pathologies related to metabolic dysfunction and tissue degeneration in humans, such as hyperinsulinemia, obesity and scarring (Suppl Fig 3C). Thus, the AHGS captures detrimental effects of chronological aging that promote tissue damage and thereby, susceptibility to degeneration.

Liver AHGS enrichment increases with severity of damage/degeneration in human livers

Aging is an acknowledged risk factor for both systemic metabolic dysfunction¹⁵ and maladaptive repair of liver injury in people^12,13. Therefore, we examined the effect of chronological age per se on liver damage in a Duke Health System population of overweight/obese adult patients with biopsy-proven metabolic dysfunction-associated steatotic liver disease (MASLD, n= 299) that we described previously^16,17 (Suppl Table 2). As expected, chronological age positively correlates with histologic indicators of MASLD severity (Suppl. Fig. 4A). Further Gene Set Enrichment Analysis (GSEA) of bulk liver RNA seq data from these MASLD patients and age- and BMI- matched controls without MASLD (n=69) (Suppl Table 2, GSE213623) revealed that genes related to longevity and its associated mechanisms (e.g. sirtuins; nicotinamide; FoxO signaling) are depleted, while genes related to senescence, a hallmark of aging, are highly enriched in the transcriptomics of MASLD patients (Suppl. Figs. 4B, C; Suppl Figs 5A, B).

To examine how the net effects of chronological aging on human livers relate to aging in hepatocytes, we used the mouse AHGS to deconvolute the bulk liver RNA seq data sets from these subjects. We observed a positive correlation between BMI and liver enrichment with the AHGS in both controls and MASLD patients (Fig 1D). This was anticipated because obesity is a major source of hepatic metabolic stress and metabolic stress accelerates biological aging^18,19. Unexpectedly, however, liver enrichment with the AHGS did not increase with chronological age in controls, whereas we found a highly significant correlation between increasing chronological age and liver enrichment with the AHGS in the similarly obese MASLD cohort (Fig 1E, Suppl Fig 4D). To assess the generalizability of the findings in our control cohort, we extended our analysis to include RNA-seq data from 226 normal liver samples from GTEx v8 (https://gtexportal.org/home/), a publicly available data set derived from healthy tissues/cells of over 900 individuals^20,21.GTEx v8 served as an ideal healthy control for our purpose since it covers a broad age spectrum. Application of the AHGS to the 226 normal human liver samples across the six age groups from GTEx v8 demonstrated that the AHGS is not enriched in liver as age increases (Suppl Fig 4E). Hence, these aggregate results support the concept that the AHGS captures detrimental responses to chronological aging that promote tissue damage/degeneration (i.e., biological aging). This suggests that the livers of obese people with MASLD in our cohort are biologically older than the livers of people without MASLD who are similarly obese and the same chronological age. Consistent with this concept, liver enrichment with the AHGS positively correlated with serum AST and the FIB-4 score (a serum biomarker of liver fibrosis severity), but demonstrated a strong negative correlation with serum albumin in subjects with MASLD (Fig 1F). Further analysis demonstrated that liver enrichment with the AHGS also increased in parallel with the severity of hepatocyte ballooning and portal inflammation, histologic markers of MASH, as well as fibrosis stage (Fig 1G). Collectively, these results confirm that the AHGS reflects the consequences of aging-related metabolic stress that promote liver damage and thus, it captures MASLD-related pathogenic responses that are sensitive to the aging process.

To further evaluate the generalizability of the AHGS, we used it to deconvolute independent MASLD transcriptomic datasets from a Japanese cohort (GSE167523), a European cohort (GSE135251), and a German cohort (GSE33814). We derived the AHGS by comparing transcriptomes of hepatocytes isolated from young and old male mice, and MASLD severity in humans is significantly impacted by both age and sex^12,13,22. Nevertheless, despite variations in age and sex across the human cohorts (Suppl Table 3), the AHGS is consistently enriched in patients with MASH (a.k.a. NASH, nonalcoholic steatohepatitis) versus either healthy controls or subjects with ‘simple steatosis’ in each cohort (Figs 1H, I; Suppl Fig 6). Notably, the AHGS differentiates livers with advanced versus mild MASLD or control liver with reasonable accuracy in the Duke cohort, and performs at least this well in the other three cohorts (all AUC > 0.80) (Figs 1G–I; Suppl Fig 6). Together, these results suggest that obese men and women with MASLD accumulate hepatocytes that exhibit hallmarks of aging as they become more elderly, while this does not occur in similarly obese people without MASLD. Further, the severity of liver damage and dysfunction in people with MASLD parallels the accumulation of these biologically-old cells.

Aging hepatocytes in MASLD lose metabolic functions and exhibit ferroptotic stress

The aforementioned conclusions are based on bulk RNA seq analyses and thus, findings may reflect gene expression in various types of liver cells. Therefore, we analyzed a publicly available, single-nucleus RNA-seq dataset generated from two MASLD patients and two healthy controls²³ (GSE174748) to confirm that hepatocytes with the AHGS accumulate in MASLD livers. We focused our analysis on the nine clusters that are most enriched for hepatocyte markers and depleted for non-hepatocyte markers (Suppl Figs 7A, B). Further analysis of this putative hepatocyte population revealed that it is comprised of several smaller communities based on transcriptomic similarities/differences (Fig 2A). Deconvolution analysis with our AHGS indicates that over 17% of hepatocytes in the livers with MASLD exhibit an aging gene signature, compared to only 3% of the hepatocytes in the healthy control livers (Fig 2B). Clinical data are not available for this human data set and thus, it is conceivable that our findings might be biased by differences in sex or chronological age of the control livers and livers with MASLD. To address this concern, we performed a similar analysis on age-matched male mice that we fed with either CDA-HFD or control diet for 22 weeks. By this time point, all mice in the CDA-HFD group have developed significant MASH and liver fibrosis, whereas livers are uniformly normal in mice fed control diets²⁴. Hepatocyte enrichment with the AHGS is even more striking in the mouse model of advanced MASH than in humans with MASLD (44% versus 17%), while the proportion of hepatocytes with the AHGS is similar in healthy murine and human livers (≤5% in both species) (Figs 2A–D, Suppl Figs 7B, D). In both murine control and MASLD datasets, cluster 4 is the largest AHGS-enriched hepatocyte community (Figs 2A–D), and GSEA reveals that this community is relatively depleted of genes and pathways associated with hepatocyte-specific metabolic functions (Figs 2E, F). In contrast, the aging hepatocyte cluster displays enrichment with genes and pathways related to insulin secretion, extracellular matrix formation and cancer (Figs 2E, F). These results affirm the link between the AHGS and liver degenerative processes in human MASLD and complement the analysis of bulk liver RNA seq data from the Duke MASLD cohort which demonstrated enrichment with pathways related to oxidative stress, DNA damage, telomere stress, cellular senescence and the senescence associated secretory phenotype (Suppl Figs 4, 5). Together, the findings suggest that biologically-old hepatocytes contribute to both metabolic dysfunction, injury, and maladaptive repair in livers with MASLD.

Fig. 2. Aging hepatocytes in MASLD lose metabolic functions and exhibit ferroptotic stress.

(A, C) Uniform manifold approximation and projection (UMAP) visualization of hepatocyte clusters in single-nucleus RNA-seq dataset generated from 2 MASLD patients or 2 healthy controls (GSE174748), or mice fed with CDA-HFD diet or chow diet for 22 weeks (n = 4 mice/group). (B, D) Deconvolution analysis on above human or mouse snRNA-seq data indicates that MASLD livers accumuate hepatoctyes with AHGS. AUCell algorism was used to identify clusters that is enriched with AHGS, and “aging cells” were defined by Area Under Curve (AUC) scores across expression ranking of all genes per cell with adjusted cutoff. (E, F) GSEA reveals that “old hepatocytes” in cluster 4 are relatively depleted of genes and pathways associated with hepatocyte-specific metabolic functions, but enriched with genes and pathways related to programmed cell death, insulin secretion and extracellular matrix formation.

Liver of MASLD patients exhibits ferroptotic stress

Our GSEA identified ‘diseases of programmed cell death’ as one of the most upregulated pathways in liver biopsies from overweight/obese MASLD patients (Figs 3A, B; Suppl Fig 4C). The AHGS itself is enriched for pathways related to insulin resistance (e.g., ‘hyperinsulinemia’, ‘abnormal circulating insulin concentration’, ‘type 2 diabetes’, ‘obesity’) and tissue degeneration (e.g., ‘scarring’), as well as biological processes that regulate ‘metal ion transport’, ‘response to cell stress’ and ‘programmed cell death’ (Suppl Figs 3A, B). These findings intrigued us because emerging evidence links type 2 diabetes and obesity, two major metabolism-related degenerative diseases, with ferroptosis, a metabolism-regulated cell death process that eventuates in lethal iron-catalyzed peroxidation of membrane lipids^25–27. Further, a recent report demonstrated ferroptotic stress in young mice with MASLD²⁸. While genes associated with other cell death pathways such as apoptosis, necroptosis, and pyroptosis are also enriched in MALD liver (Suppl Figs 7E–G), ferroptosis has been suggested as the root death mechanism in metabolic diseases^29,30.

Figure 3. Livers of MASLD patients exhibit ferroptotic stress.

(A, B) GSEA of Duke MASLD cohort (GEO213623) reveals that human MASLD liver displays enrichment with genes and pathways related to programmed cell death. Hepatic mRNA expression of ferroptosis-related genes is altered with MASLD (C) and fibrosis stage (D). (E) Ferroptosis marker score was construbted using the 9 verified ferroptosis marker genes (FTH1, GPX4, CHAC1, HSPB1, NFE2L2, PTGS2, SLC40A1, TF, and TFRC) from the FerrDb V2 database. (F) Representative staining for ACSL4 and GPX4 and the quantification of the positively stained area in MASLD patients versus healthy controls. Data are graphed as mean ± SEM. (G) ‘Iron accumulation signature’ (IAS) score was constructed using the128 genes that are strongly induced across most cell types in response to iron in vivo. For both panel E and G, their average expression in each single cell was calculated by the AddModuleScore function of Seurat. Feature plot showed single cells on dimensional reduction plot, with cell color indicating their expression score. Violin plot showed the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution in the healthy and MASLD group. p-values were calcualted using: permutation testing in A, B; Wilcoxon Rank Sum test in C, D, E, G; two-tailed Student’s t-test in F.

To determine whether ferroptotic stress also associates with MASLD in humans, we compared expression of key ferroptosis-related genes in livers of Duke MASLD cohort¹⁶. mRNA expression of ACSL4, which encodes a critical enzyme that generates lipid substrates for peroxidation²⁵ progressively increased, while transcripts that encode GPX-4 and GCH-1, enzymes that detoxify lipid peroxides²⁵, progressively decreased during fibrosis progression in the livers of MASLD patients (Figs 3C, D). Reciprocal changes in the gene expression of ferroptosis inducers (ACSL4) and suppressors (GPX-4 and GCH-1) suggest that ferroptotic stress increases with MASLD severity. Expression of SLC7A11, PTGS2, FSP1/AIFM2 also tended to increase with fibrosis severity in MASLD livers (Figs 3C, D). While these mRNAs encode cytoprotective genes that oppose ferroptosis, they are also used as markers of ferroptotic stress²⁵ and thus, their upregulation is thought to reflect adaptive responses aimed at preventing ferroptosis. To examine whether age affected the expression of ferroptosis-related genes in patients with the same fibrosis stage, we stratified the Duke MASLD cohort patients with the same fibrosis stage into young (age ≤ 35), middle age (35 <age< 55) and old groups (age ≥ 55) and then compared the expression of these ferroptotic stress-associated genes. Expression of ACSL4 transcripts is significantly higher in livers of older MASLD patients compared to their younger counterparts with the same fibrosis stage, but we were unable to demonstrate evident age-related differences of the other markers of ferroptotic stress in whole liver RNA (Suppl Fig 8A).

However, gene expression in bulk liver RNA reflects the sum of transcripts in multiple cell types and thus, significant cell-specific differences in gene expression may be obscured. Therefore, to better characterize ferroptotic stress in hepatocytes per se, we employed a comprehensive approach that utilized all nine verified ferroptosis marker genes from the FerrDb V2 database to construct a robust ferroptosis gene signature and applied this signature to the hepatocyte subpopulations of the single-nuclei RNA-seq dataset from human healthy and MASLD livers (GSE174748)²³. Hepatocytes in the MASLD patients were significantly enriched with this ferroptosis marker gene signature compared to hepatocytes of healthy controls (Fig 3E). To assure that differences in hepatocyte protein expression matched differences in hepatocyte mRNA expression, we compared protein expression of the key ferroptosis inducer (ACSL4) and ferroptosis suppressor (GPX-4) in liver sections from Duke controls and MASLD subjects. Accumulation of ASCL4 was significantly increased in hepatocytes of MASLD subjects, while hepatocyte accumulation of GPX4 was significantly decreased in MASLD subjects (Fig 3F), providing further evidence that hepatocyte ferroptotic stress is increased in MASLD liver.

Ferroptosis is an iron-dependent cell death process²⁵ and many studies associate whole liver iron accumulation with increased MASLD severity^31–35. However, existing literature also demonstrates diverse patterns of hepatic iron deposition in MASLD, including exclusive deposition of iron in hepatocytes (HC), predominant reticuloendothelial system (RES) iron accumulation, and a mixed pattern of iron overload involving both HC and RES^31–35. The publicly available MASLD and control liver single cell RNA seq data sets provide an opportunity to evaluate cell type-specific expression of genes that regulate iron homeostasis. Therefore, we examined the expression of genes related to iron homeostasis in hepatocytes from this dataset as our other analyses suggested that aging promotes ferroptosis in these cells (Figs 3E, F). The results show that relative to hepatocytes from healthy livers, hepatocytes from MASLD livers have significantly up-regulated genes that promote cellular uptake/accumulation of iron (TF, TFRC, TFR2, HFE, FTL, FTH1 and HAMP) (Suppl Fig 8B). Expression of FPN mRNAs (which encode the only known iron exporter) was also increased but the biological relevance of this is uncertain as increased HAMP is predicted to reduce FPN activity via a post-translational mechanism³⁶. A recent publication reported an ‘iron accumulation signature’ (IAS) that includes 128 genes that are strongly induced across most cell types in response to iron in vivo³⁷. To consolidate the results from our analysis of individual iron-related genes, we performed deconvolution analysis by calculating the IAS enrichment score in this snRNA-seq dataset. The results reveal that the IAS is also significantly enriched in hepatocytes of MASLD liver (Fig 3G), along with the decreased expression of anti-ferroptotic protein GCH1 (Suppl Fig 8C). Thus, the aggregate findings suggest that hepatocytes in MASLD livers accumulate iron and support the other evidence demonstrating these cells are experiencing significant ferroptotic stress.

Ferroptotic stress and lipotoxicity are increased in livers of old mice

The findings in human livers with MASLD link the AHGS with ferroptotic stress, suggesting that hepatocyte aging and MASLD might be causes and/or consequences of ferroptotic stress. GSEA revealed that old mouse hepatocyte transcriptomes are enriched with pathways associated with MASLD (a.k.a., NAFLD) (Fig 4A). Therefore, we compared parameters of MASL (‘simple’ steatosis), MASH (hepatic lipotoxicity and inflammation), and liver degeneration (fibrosis) in old and young mice to determine how chronological aging per se impacts MASLD pathogenesis/progression. Oil red O staining (Figs 4B, C) and triglyceride quantitation (Fig 4D) demonstrated increased steatosis in livers of old mice and GSEA confirmed aging-related changes in signaling that are known to promote lipid accumulation in hepatocytes, including down-regulation of insulin signaling and up-regulation of fatty acid transport (Fig 4A). F4/80 staining and PCR analysis of various proinflammatory cytokines/chemokines revealed increased inflammation in old livers (Figs 4B, C, E). These aging-related increases in liver inflammation are accompanied by increased lipotoxicity as we found that livers and hepatocytes of old mice accumulate more products of lipid peroxidation, including 4-HNE (Figs 4B, C) and malondialdehyde (MDA) (Suppl Fig 9A). Together, these findings indicate that chronological aging per se evokes sufficient metabolic stress to promote MASH in mice. MASH stimulates repair responses and thus, increases the risk for liver fibrosis. We found that compared to young mouse livers, livers of old mice have more fibrosis basally as exemplified by greater Sirius red staining (Fig 4E) and increased expression of myofibroblast markers (Fig 4F). Aging and the associated tissue alterations are known to develop with large interindividual heterogeneity^3,38. Therefore, we also analyzed a published dataset (GSE132042) that compared the transcriptomes of very old mouse liver (≥ 24 months old, n=7) to young mouse liver (≤ 3 months old, n=12) to assess the generalizability of our findings. AHGS deconvolution of this dataset revealed an enrichment of the AHGS in the livers of aged mice (Suppl Fig 9B). Subsequent GSEA analysis confirmed that the transcriptomes of aged mouse livers are enriched with genes associated with hepatocyte aging (Suppl Fig 9C), lipid metabolism (Suppl Fig 9D), inflammation (Suppl Fig 9E), and fibrotic activity (Suppl Fig 9F).

Figure 4. Ferroptotic stress and lipotoxicity are increased in livers of old mice.

Livers of chow-fed young and old mice were analyzed to determine if old livers exhibited features of MASLD. (A) GSEA demonstrated that the transcriptome of old hepatocytes is enriched with genes invloved in MASLD (a.k.a NAFLD) and fatty acid transprot, but depleted with genes involved in insulin signaling. (B) Representative staining and (C) quantification of the positively stained area. (D) Liver triclyceride levels. (E) RT-PCR for proinflammatory markers. (F) Western blotting for fibrosis markers, αSMA and Vim. (G) GSEA demonstrated that transcriptome of old hepatocytes is enriched with genes invloved in iron uptake and transport and ferroptosis. (H) Western blotting for proteins involved in iron metabolism and ferroptosis. Data are graphed as mean ± SEM. n = 8 mice/group for Oil Red O staining (B), liver triclyceride (D) and RT-PCR analyes (E) due to a larger number of archived frozen liver tissues; n = 4 mice/group for IHC analysis due to the limited availability of the FFPE liver blocks from aged mice.*p < 0.05 versus young liver. p-values were calcualted using: permutation testing in A, G; two-tailed Student’s t-test in C, D, E, F, H.

Lipid peroxidation can be catalyzed by iron and lead to cell death via ferroptosis. Cellular accumulation of iron is known to increase with chronological age in many tissues, including liver³⁹ and brain⁴⁰ and ferroptosis has been implicated in the pathogenesis of various neurodegenerative diseases that worsen with age⁴¹. We found that signaling pathways associated with those diseases are significantly enriched in old mouse hepatocytes (Suppl Fig 1). GSEA revealed that old mouse hepatocytes are also significantly enriched with pathways related to iron uptake and transport, as well as ferroptosis (Fig 4G). Immunostaining and western blot analyses confirmed that compared to young hepatocytes, old hepatocytes express higher levels of proteins that regulate iron uptake and accumulation (e.g., Tfrc, and Fth1), as well key lipid peroxidization enzymes (e.g., ACSL4) (Figs 4B, C, H). Together, these data indicate that aging itself increases ferroptotic stress in hepatocytes and is sufficient to initiate the entire spectrum of MASLD in mice.

Aging increases ferroptotic stress and promotes hepatocyte susceptibility to lipotoxicity

Early hypotheses about MASLD pathogenesis postulated that hepatocytes in steatotic livers are viable but vulnerable to damage from super-imposed stressors (i.e., ‘second hits’)⁴². To determine if this is also true of old hepatocytes that are under aging-related ferroptotic stress we subjected primary cultures of hepatocytes from old and young mouse livers to superimposed lipotoxic stress by treating them with oleic and palmitic acid (OPA) for 7 days. Oil red O staining confirmed that old hepatocytes are more steatotic than young hepatocytes, both at baseline and after fatty acid challenge (Figs 5A, D). In addition, old hepatocytes exhibit increased expression of Acls4 and greater lipid peroxidation both basally and in response to superimposed lipotoxic stress (Figs 5B, D). Aging is known to impair the liver’s regenerative capacity⁵ and accumulation of senescent cells is a characteristic of aging tissues⁴³. β-galactosidase reactivity is a hallmark of senescent cells⁴³. We observed that hepatocyte isolates from old livers are enriched with β-galactosidase-positive hepatocytes basally and found that this is exacerbated by superimposed lipotoxic stress. (Figs 5C, D). Together, these results indicate that hepatocytes in old livers are viable, but more vulnerable than hepatocytes in young livers, and demonstrate that hepatocyte senescence and susceptibility to lipotoxicity increase with aging-related ferroptotic stress.

Figure 5. Aging increases ferroptotic stress and promotes hepatocyte susceptibility to lipotoxicity.

Primary hepatocytes from old and young mice (n = 3 mice/group) were exposed to lipotoxic stress by treating them with oleic and palmitic acid (OPA) for 7 days Compared to young hepatocytes, old hepatocytes are: more steatotic, as shown by Oil red O staining (A); afflicted with more ferroptotic stress, as measured by Acsl4 and 4-HNE staining (B); and are more senescent, as evidenced by β-gal staining (C), at either baseline or in response to lipotoxic stress. (D) Quantification of the positively stained areas. Data are graphed as mean ± SEM. *p < 0.05 versus young hepatocytes without OPA treatment. ^#p < 0.05 versus young hepatocytes with OPA treatment. p-values were calculated using two-tailed Student’s t-test.

Ferrostatin-1 protects hepatocytes from aging-related ferroptotic stress and senescence during diet-induced MASH

Hepatocyte lipotoxcity is thought to drive progression from “simple” hepatic steatosis (the earliest stage of MASLD) to steatohepatitis and thus, increase the risk for progressive liver damage that can culminate in cirrhosis and increase the risk for liver cancer⁴⁴. Our studies in cultured primary hepatocytes suggest that aging-related vulnerability to progressive MASLD may be due to aging-related increases in ferroptotic stress. To evaluate this hypothesis, we fed young and old mice a choline-deficient amino acid restricted high fat diet (CDA-HFD) for 5 weeks to induce MASH and then randomized the old mice to receive either vehicle or ferrostatin-1 (Fer1), a ferroptosis inhibitor²⁵, via i.p. injection every other day during the last 8 days of CDA-HFD feeding. We elected to use this MASH-inducing diet for this experiment because this diet induces robust histologic and biochemical indicators of fibrosing MASH in a relatively short time and mimics the effects of dietary choline deficiency that are known to promote MASLD in people^45,46. In addition, our initial studies had demonstrated this protocol recapitulates aging-related changes in hepatocyte populations that occur in human MASLD livers (Fig 2). Results in old Fer-1 treated mice were compared to vehicle treated old mice and young mice that were continued on the CDA-HFD and treated with vehicle (Fig 6A; Suppl Fig 10A). CDA-HFD increased ferroptotic stress more in old mice than young mice, as evidenced by increased liver and serum levels of lipid peroxidation products (e.g., MDA, 4HNE) in old vehicle-treated mice versus young vehicle-treated mice at the end of the CDA-HFD challenge (Figs 6B–D). Hepatic expression of Tfrc and Acsl4, is also higher in aged mice (Figs 6C, D; Suppl Figs 10B, C). Remarkably, treating old mice with Fer1 during the last week of CDA-HFD feeding eliminated the age-related effects on all of these parameters such that old Fer1-treated mice resembled young vehicle-treated mice with regards to hepatic expression of Tfcr and Acsl4, as well as liver and serum accumulation of lipid peroxidation products (Figs 6B–D; Suppl Figs 11B, C).

Figure 6. Ferrostatin-1 protects hepatocytes from aging-associated ferroptotic stress and senescence during diet-induced MASH.

(A) Young and old mice were fed with a CDA-HFD diet for 6 weeks. Old mice were intraperitoneally injected with ferrostatin1 (Fer1) or its vehicle every other day during the last 8 days of feeding (young CDAHFD treated with veh, n=9; old CDAHFD treated with veh, n=6; old CDAHFD treated with Fer-1, n=6). (B) MDA levels in liver tissue or serum of MASH mice. (C) Representative IHC stainings for Acsl4, 4-HNE and (D) corresponding densitometric analysis of positively stained areas. (E) Representative IF stainings for DNA damage markers, 8-OHdG and γH2AX, and (F) corresponding quantification of the positively stained areas. (G) Representative stainings for senescent markers, SA-β-gal, and (H) corresponding quantification of the positively stained areas. Ven diagram identified DEGs (old vs young) that reversed by Fer1 treatment. (I) Expression of senescent marker p16 detected by RT-PCR. (J) PCA reveals that Fer1 shifts the liver transcriptomes of old mice toward those of young vehicle-treated mice. (K, L) Deconvolusiton analysis reveals that Fer1 reverses old liver enrichment with both the AHGS and SenMayo (n = 4 mice/group). Boxplot shows the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. (M) GSEA demonstrated that transcriptome of old mice treated with Fer1 is depleted with genes involved in senescence and SASP, but enriched with genes involved in DNA repair. Data are graphed as mean ± SEM for B, D, F, H, I. ^#p < 0.05 versus young mice + veh; ^$p < 0.05 versus old mice + veh. p-values were calcualted using: one-way ANOVA in B, D, F, H; Wilcoxon Rank Sum test in K, L; permutation testing in M.

Ferroptotic cell death results when oxidative damage to vital cellar components outstrips processes to repair that damage²⁵. Cells with oxidative DNA damage that survive during ferroptotic stress evoke programs the mobilize cyclin dependent kinase inhibitors, such as p21 and p16, to arrest progression through the cell cycle and thereby, prevent their proliferation²⁵. Although these ferrosenescent cells themselves are not regenerative, they remain viable and are metabolically-active, exhibiting increased activity of certain lysosomal enzymes (e.g., beta galactosidase(β-gal)) and acquiring a dynamic senescence-associated secretory phenotype (SASP) that helps to shape the microenvironment of injured tissues⁴⁷. Ferrosenescence has been implicated in the pathogenesis of degenerative diseases that are exacerbated by aging⁴⁸. To determine whether ferrosenescence might be involved in aging-related susceptibility to progressive MASLD, we compared markers of DNA damage and senescence in young and old CDA-HFD-fed mice and assessed how Fer1 impacted those outcomes. Young mice fed CDA-HFD for 6 weeks exhibit rare immunostaining for markers of oxidative DNA damage (8-OHdG) and double strand DNA breaks (γH2AX) (Figs 6E, F) and faint staining for β-gal activity (Figs 6G, H). These mice also have significantly higher expression of p21 protein than young mice on chow diets (Suppl Fig 10D), supporting previous publications showing that MASH-inducing diets induce liver cell senescence in young mice^17,49. Interestingly, these markers of DNA damage are five- to eight- fold higher in old CDA-HFD-fed mice than in young CDA-HFD-fed mice, and this aging-exacerbated DNA damage is paralleled by increased expression of p21, p16 and β-gal activity (Figs 6E–I; Suppl Fig 10D). Remarkably, Fer1 treatment of old CDA-HFD-fed old mice reverted all of these senescence markers to levels observed in young CDA-HFD-fed mice (Figs 6E–I; Suppl Fig 10D).

To further confirm these remarkable phenotypic changes in old liver (vs young-vehicle) and the beneficial effects of Fer1 treatment (vs old-vehicle), we performed RNA-seq analysis of whole liver RNA isolated from 4 mice/each treatment group at the end of CDA-HFD feeding. PCA plot revealed that Fer1 treatment shifted the liver transcriptomes of old mice toward those of young vehicle-treated mice (Fig 6J). Overlapping the DEGs by Venn Diagram identified 520 DEGs that are upregulated in old liver (compared to young vehicle) but suppressed by Fer1 treatment (compared to old vehicle) (Suppl Fig 11), and 658 DEGs that are suppressed in old liver but increased by Fer1 treatment (Suppl Fig 12). Characterization of the Fer1-sensitive DEGs that are up-regulated in old vs. young livers demonstrated enrichment with genes involved in epithelial-mesenchymal transition (EMT), inflammation, and TGF-beta signaling (Suppl Fig 11), while the Fer1-sensitive DEGs that are down-regulated in old vs. young livers are enriched with genes involved in liver metabolic functions (e.g. oxidative phosphorylation; xenobiotic metabolism; bile acid metabolism) (Suppl Fig 12). GSEA also revealed that Fer1 down-regulated pathways for oxidant stress and oxidant stress-related lethality (Suppl Fig 10E), complementing the other evidence that Fer1 suppresses MASH-related ferroptotic stress in the old livers. We also conducted senescence deconvolution based on SenMayo gene set⁵⁰, which includes 125 genes related to cellular senescence. Aged liver enrichment with both the AHGS (Fig 6K) and the SenMayo signature (Fig 6L) is also reversed by Fer1, suggesting that alleviating ferroptotic stress shifts old livers to a younger and healthier phenotype. GSEA further demonstrated that relative to livers from CDA-HFD old mice that were treated with vehicle, livers from old CDA-HFD-fed mice that received Fer1 down-regulate pathways involved in DNA damage, telomere stress, senescence, and the SASP phenotype and become significantly enriched with pathways involved in DNA repair (Fig 6M). Together, these results implicate ferroptotic stress as a critical driver of liver cell senescence in MASLD.

Ferrostatin-1 prevents aging-related exacerbation of MASH

Senescent hepatocytes accumulate in humans with MASLD, and reducing the liver burden of senescent cells improves MASLD in preclinical models by abrogating the pro-inflammatory and pro-fibrogenic actions of the senescence associated secretory phenotype (SASP)^17,49. Findings of the present study identify a key role for ferroptotic stress in promoting MASLD-related accumulation of senescent hepatocytes, particularly in older individuals, and our RNA seq analysis demonstrates that ferrostatin reduces SASP-related signaling in old mice fed CDA-HFD. To determine how this impacts MASLD severity we compared the degree of steatosis, inflammation and fibrosis in young and old CDA-HFD-fed mice and assessed how Fer1 impacted these parameters in old CDA-HFD-fed mice. Liver histology and quantitative morphometric analyses confirmed that old livers developed more fibrosis than young livers, although aging had only minor impact on the severity of hepatic steatosis or macrophage accumulation induced by feeding CDA-HFD for 6 weeks (Figs 7A, B). Compared to livers of young CDA-HFD-fed mice, livers of CDA-HFD-fed old mice also had higher protein expressions of αSMA and Vim, and mRNAs of Col1a1 and PDGFRβ, consistent with histologic evidence that aging exacerbates fibrogenic responses to metabolic stress (Figs 7C, D). In addition, CDA-HFDs induced greater expression of TNFα and IL6 mRNAs in old livers (Fig 7D), suggesting that aging also exacerbates pro-inflammatory immune responses. This is consistent with the above evidence that aging promotes hepatocyte senescence (Fig 6) and suggests a pro-inflammatory and pro-fibrogenic role for the SASP in this context. Since Fer1 treatment greatly alleviates aging-associated ferroptotic stress and senescence (Fig 6), we asked whether it improves aging-related exacerbation of MASH. Indeed, we observed that treating old mice with Fer1 abolished aging-related increases in liver steatosis, fibrosis, and inflammation during CDA-HFD exposure (Figs 7A–D), reverting the liver phenotype of old mice to that of young mice despite ongoing exposure to exogenous metabolic stress. Consistent with this remarkable finding, RNA seq analysis revealed that treating old CDA-HFD-fed mice with Fer1 also reverted the activities of pathways involved in the inflammatory response and fibrosis to levels of young mice that were fed CDA-HFD (Fig 7E). Together, these results demonstrate that reducing the liver burden of ferroscenescent cells improves the health-span of old livers.

Figure 7. Ferrostatin-1 prevents aging-related exacerbation of MASH.

Young and old mice were fed with a chow diet or CDA-HFD diet for 6 weeks. CDA-HFD fed mice were intraperitoneally injected with Fer1 or its vehicle every other day during the last 8 days of feeding (n = 3 mice/young chow; n = 9 mice/ young CDAHFD veh; n = 6 mice/ old CDAHFD veh; n = 6 mice/ old CDAHFD Fer1). (A) Representative stainings for H&E, Oil Red O, Sirius Red and F4/80. (B) Quantification of the positively stained areas for Oil Red O, Sirius Red and F4/80. (C) Expression of fibrotic markers, aSMA and Vim, detected by western blotting. (D) Expression of fibrotic and inflammatory markers detected by RT-PCR. (E) GSEA demonstrated that transcriptome of old mice is enriched with genes involved in fibrosis and inflammation, and these gene signatures are reversed by Fer1 treatment in the old mcie. Data are graphed as mean ± SEM. *p < 0.05 chow fed mice; ^#p < 0.05 versus young mice + veh; ^$p < 0.05 versus old mice + veh. p-values were calcualted using: one-way ANOVA in B, C, D; permutation testing in E.

The collective data from mice and humans demonstrate consistent robust correlations between iron-dependent lipotoxicity (i.e., ferroptotic stress), hepatocyte senescence, biological aging, and MASLD pathogenesis/progression. However, although chronological aging is known to increase both hepatic accumulation of iron and risk for MASLD^31–35, our findings clearly demonstrate that humans exhibit significant inter-individual differences in susceptibility to aging-related iron accumulation/toxicity. This suggests that ferroptotic stress must also be mitigated by other variables in these high-risk populations. Recent publications have reported that iron suppresses hepatocyte expression and activity of FXR (the nuclear hormone receptor for endogenous bile acids) and showed that FXR deficiency exacerbates hepatic iron toxicity and dysregulates homeostasis of both iron and bile acids⁵¹. These observations intrigued us because defects in FXR signaling and bile acid homeostasis are common in individuals with MASLD and a recent publication demonstrated that certain bile acids and other FXR agonists can activate FXR to inhibit ferroptosis in HepG2 cells and primary mouse hepatocytes⁵². These findings in liver cells align with studies demonstrating that FXR protects against nephrotoxicity by up-regulating the transcription of genes that inhibit ferroptosis^53,54. To investigate whether differences in FXR signaling might underlie differences in ferroptosis susceptibility in MASLD, we analyzed our bulk liver RNA-seq datasets derived from the three groups of mice that were subjected to the 6-week CDA-HFD regimen. Our GSEA revealed a substantial reduction in bile acid metabolism in the livers of old mice compared to young mice, and this decline was effectively reversed by Fer1 treatment (Suppl Fig 13A). Importantly, the expression levels of FXR/Nr1h4 and the Biocarta_FXR_Pathway (the only available FXR-associated pathway in the GSEA database) were down-regulated in the livers of old mice, and both of them were restored to youthful levels with Fer1 treatment (Suppl Figs 13B, C). Thus, our mouse data support the earlier reports that iron-related stress suppresses FXR expression and activity in hepatocytes⁵¹, and reveal a plausible mechanism that may help to explain how aging promotes MASLD. Namely, aging increases iron accumulation; this suppresses FXR signaling; the loss of FXR activity re-enforces iron accumulation and FXR suppression, further reducing expression of FXR target genes that inhibit ferroptosis; this increases ferroptotic stress and consequent hepatocyte lipotoxicity and senescence; ferrosenescent hepatocytes cannot regenerate but have a SASP that promotes inflammation and fibrosis; the progressive accumulation of ferrosenescent hepatocytes drives liver degeneration and MASLD progression. Our other data indicate that livers of people with MASLD are particularly sensitive to the detrimental effects of aging (Fig 1). Hence, extension of this reasoning predicts that FXR expression will be lower in people with MASLD than people without MASLD, and decline as fibrosis severity increases within the MASLD cohort. Analysis of the Duke cohort dataset supports both of these concepts (Suppl Fig 13D). Additionally, comparison of the hepatocyte clusters in the single-nucleus RNA-seq data sets confirms that FXR expression is reduced in hepatocyte populations from MASLD patients compared to healthy controls (Suppl Fig 13E).

Liver aging correlates with dysfunction of multiple organ systems

The liver is a master regulator of systemic bile acid and iron homeostasis and its ability to modulate both processes helps control metabolism, energy balance, oxidant stress, inflammation and fibrosis in other tissues^55,56. People with MASLD and advanced liver fibrosis are at increased risk both for degeneration of other organs and all-cause mortality⁵⁷. The AHGS captures changes in signaling that disrupt iron and bile acid homeostasis and promote biological aging in hepatocytes by initiating the ferroptosis death program. We found that liver enrichment with the AHGS increases with BMI, regardless of chronological age (Fig 1D), suggesting that obesity-related metabolic stress accelerates biological aging in liver. This is intriguing given growing evidence linking ferroptosis to various obesity-related degenerative diseases^58–60 including obesity itself²⁷. Obesity increases the risk factor for hyperinsulinemia and type 2 diabetes (T2D), and the odds for MASLD-related liver degeneration are increased in people withT2D⁶¹. Further analysis of our Duke Health System cohort demonstrates that liver enrichment with the AHGS is greater in people with T2D than in non-diabetics (Fig 8A). Interestingly, Fer1 reversed activation of signaling pathways in hepatocytes that promote insulin secretion in old mice (Fig 8B) and dysregulated ferroptosis has been implicated in pancreatic beta cell dysfunction in T2D⁶². As mentioned earlier, our GSEA demonstrates that transcriptomes of old mouse hepatocytes are highly enriched with signaling pathways that associate with aging-exacerbated neurodegenerative diseases that can be improved by targeting ferroptosis⁶³ (Suppl Fig 1). To screen for a potential interaction between liver aging and dysfunction in other organs, we used the AHGS to deconvolute RNA data sets from three additional organs that often become dysfunctional in MASLD patients (pancreas, kidney, and heart). Because accumulation of senescent cells is an acknowledged hallmark of aging⁴³, the AHGS results were validated using the SenMayo gene signature that other researchers derived by comparing bone and bone marrow cells from old and young individuals⁵⁰.

Figure 8. Liver aging correlates with dysfunction of multiple organ systems.

(A) AHGS was applied to deconvolute bulk liver RNA-seq data that includes 368 liver biopsies from healthy controls or patients with MASLD (GSE213623). AHGS is enriched in both controls and MASLD patients with diabetes and the enrichment is statistically significant in MASLD patients with diabetes. (B) Young and old mice were fed with a CDA-HFD diet for 6 weeks. Old mice were intraperitoneally injected with Fer1 or its vehicle every other day in the last 8 days. GSEA of the transcriptomics of livers from these mice revealed that Fer1 reversed activation of signaling pathways in hepatocytes that promote insulin secretion in old mice. (C) AHGS is significantly enriched in both failing hearts with preserved ejection fraction (HFpEF; n=41) and failing hearts with reduced ejection fraction (HFrEF, n=59), comparing to Normal (n=45). The SenMayo senescence score is also higher in HFrEF and HFpEF. (D) AHGS is significantly enriched in patients with diabetes-chronic kidney disease (D-CKD; n=19), Diabetes (n=21), and hypertension (n=19) (left panel). By merging Normal, Diabetes and hypertension into Control, significant AHGS enrichment in D-CKD group persisted relative to this combined control group. The SenMayo senescence score is also higher D-CKD than combined Control. (E) Bulk RNA-seq was conducted in human pancreatic islets from 89 deceased donors. In contrast to Normal (n=51) AHGS is enriched in donors with IGT (n=15) and T2D (n=11), and AHGS enrichment strongly correlates with HbA1c levels. The SenMayo senescence score is also higher in IGT and T2D. Boxplot shows the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. p-values were calcualted using: Wilcoxon Rank Sum test in A, C, D, E; permutation testing in B; Pearson Correlation test in E middle panel.

For the heart analysis, we examined publicly available RNAseq data derived from 45 healthy and 100 failing hearts⁶⁴. We found that failing hearts are significantly enriched with both the AHGS and SenMayo (Fig 8C). Interestingly, we observed enrichment of AHGS in heart failure with both preserved ejection fraction (HFpEF) and reduced ejection fraction (HFrEF) (Fig 8C), conditions that strongly associates with obesity-related systemic metabolic dysfunction and MASLD⁶⁵.

For the kidney analysis, we downloaded microarray-based transcriptome expression data from 79 normal and fibrotic human kidneys that Kang and colleagues had segregated into four groups⁶⁶. Compared to the healthy control group, the 3 disease groups (Diabetic chronic kidney disease, D-CKD, Diabetes, and hypertension) each showed significant enrichment with the AHGS signature (Fig 8D). After we merged data from the diabetes and hypertension groups with the healthy kidney control group, significant AHGS enrichment in D-CKD group persisted relative to this combined control group (Fig 8D). Moreover, the SenMayo senescence score was also higher in D-CKD than in the combined controls (Fig 8D).

For the pancreas analysis, we reanalyzed bulk RNA-seq data derived from human pancreatic islets of 89 donors with different levels of insulin sensitivity and hyperglycemia⁶⁷. The AHGS score strongly correlated with HgbA1c, and was significantly enriched in donors with T2D and impaired glucose tolerance (IGT) (Fig 8E). SenMayo analysis also showed significantly higher enrichment of senescence genes in islets from T2D donors than normal donors (Fig 8E).

These studies do not indicate whether aging-related liver degeneration induces metabolic stress that triggers other organ dysfunction (or vice versa). Nevertheless, the collective analyses of all four organs demonstrate that the liver-derived AHGS generally differentiates diseased from healthy organs. This finding is important because it indicates that conserved mechanisms control organ degeneration and identifies iron-sensitive targets as critical mediators of the process.

DISCUSSION

We discovered a key mechanism that controls hepatic susceptibility to aging and suggest that this mechanism may also regulate aging, and thus degeneration, in other tissues. Briefly, our findings show that hepatocytes in old livers have reduced metabolic resiliency that increases their exposure to factors that cause ferroptosis, a type of lipotoxicity. Ferroptosis results from iron catalyzed peroxidation of membrane lipids and thus, requires co-localization of certain types of lipids, ROS, and iron. Hence, ferroptotic stress is regulatable by all of the metabolic pathways that control the bioavailability of these three factors. Our data indicate that hepatocytes can survive ferroptotic stress by up-regulating FXR-sensitive mechanisms that enhance anti-oxidant defenses, restrict accumulation of lipid species that are peroxidation targets, and/or limit iron bioavailability. However, when these adaptations fail to occur, or are not sufficient to abrogate oxidative damage to membranes, proteins and DNA, aging livers accumulate dysfunctional and senescent hepatocytes.

These ‘old’ hepatocytes are resistant to FXR and other critical pro-regenerative/hepato-trophic factors, including insulin⁶⁸. Our study design does not permit us to determine if hepatocyte insulin resistance is a cause or consequence of aging. Nor can we determine the origin(s) of the insulin resistance in ‘old’ hepatocytes since aging associates with impaired insulin signaling in many tissues⁶⁹. Nevertheless, evidence that ‘old’ hepatocytes are FXR-deficient and insulin resistant during ferroptotic stress helps to explain why the burden of aged and senescent hepatocytes is greatest in mice and humans with the most severe liver fibrosis. More specifically, hepatocyte FXR deficiency and insulin resistance inhibit hepatocyte replication and thus, impair liver regeneration; liver fibrosis is a biomarker of impaired liver regeneration (i.e., increased liver degeneration/aging).

Fer-1 specifically inhibits ferroptosis by decreasing both labile iron and lipid radicals that drive peroxidation of membrane phospholipids⁷⁰. Importantly, our experiments with Fer-1 demonstrate that alleviating ferroptotic stress significantly reduces accumulation of senescent hepatocytes and improves liver metabolism, inflammation and fibrosis in old mice during exogenous metabolic stress imposed by MASH-inducing diets. Pathway analyses of RNA seq data from old mouse hepatocytes and hepatocytes from humans with MASLD/NAFLD indicate that senescing hepatocytes drive liver inflammation and fibrogenesis in MASLD by generating secretomes that promote inflammation and fibrosis. Together, these results support the concept that aging exacerbates MASLD-related liver degeneration by promoting accumulation of senescent hepatocytes that have adapted to survive ferroptotic stress. Remarkably, the findings in mice show that by reducing ferroptotic stress, it is possible to reverse adverse hepatic effects of biological aging accelerators even when chronological age is very advanced. Our analyses of several large human data sets suggest that regardless of chronological age, people with MASLD are more vulnerable to the hepatic effects of aging than controls. Importantly, the results also indicate that the severity of liver degeneration in MASLD patients parallels that of biological aging. Therefore, efforts to test the efficacy and safety of ferroptosis inhibitors as potential treatments for MASLD are justified. Interestingly, many treatments that have shown some benefit in MASLD have also been reported to inhibit ferroptosis, including caloric restriction^71,72, exercise^71,73, vitamin E^25,74, pioglitazone^74,75, obeticholic acid and other FXR agonists⁵¹, SGLT2 inhibitors^76,77, and GLP1 agonists^78,79.

Many of these interventions have also been reported to have some benefit in other types of degenerative disease that associate with aging-related metabolic dysfunction including type 2 diabetes (T2D), heart failure with preserved ejection fraction (HFpEF), and neurodegenerative disorders^30,63,80. Our data suggest that the mutual therapeutic benefits may reflect shared aging-sensitive pathogenic mechanisms across tissues in individuals with these disorders. For example, we discovered that the aging hepatocyte gene signature is enriched in the livers of people with T2D. Remarkably, we also found that the gene signature of old hepatocytes is significantly enriched in RNAseq data sets derived from diseased pancreatic islets from insulin-resistant subjects, kidneys with chronic damage related to diabetes, and failing versus healthy human hearts. In addition, KEGG pathway analysis of old mouse hepatocytes showed that six of the top ten up-regulated pathways define signaling in neurodegenerative disorders. Importantly, our new findings complement and extend previous publications about ferroptosis and tissue degeneration related to metabolic stress. Oxidative stress and dysregulated iron and lipid homeostasis have been noted in pancreatic beta cells during T2D⁶², chronic kidney disease⁸¹, cardiomyocytes in HFpEF⁸⁰, and both neurons and neural stromal cells in neurodegenerative diseases⁶³, suggesting that ferroptotic stress is increased in all of these conditions. This concept is further supported by emerging evidence that ferrostatin is beneficial in preclinical models of these diseases^62,63,80.

In summary, aging is known to facilitate cellular metabolic stress and tissue degeneration that leads to multiple organ failure and limits longevity. In several tissues, degeneration associates with ferroptosis, a regulated cell death process that is both a consequence and a cause of metabolic dysfunction. The effectors of ferroptotic stress are certain types of membrane lipids, ROS and iron and thus, susceptibility to ferroptosis is controlled by mechanisms that regulate the bioavailability of these factors. The liver has a key role in regulating systemic lipid metabolism, energy balance, oxidant stress and iron homeostasis. We discovered that aging increases ferroptotic stress in hepatocytes themselves and demonstrated that damaged hepatocytes that have adapted to survive chronic ferroptotic stress generate a secretome that drives liver degeneration (Summary Cartoon). The results raise the possibility that livers enriched with these ‘undead’ hepatocytes may also amplify ferroptotic stress in other tissues by disrupting inter-organ signaling that normally constrains ferroptotic stress. If true, this would help to explain why liver fibrosis severity is a biomarker for both other types of organ failure and increased all-cause mortality in people with systemic metabolic dysfunction⁵⁷. More research is needed to evaluate this hypothesis and to define the master regulators of metabolic resiliency as these factors ultimately control exposure to ferroptosis effectors and thus, susceptibility to aging.

Summary Cartoon: Aging Promotes MASLD and Multi-Organ Dysfunction by Inducing Ferroptotic Stress.

Hepatocytes in old livers exhibit ferroptotic stress. The accumulation of hepatocytes with ferroptotic stress induces features of MASLD in old mice, and strongly correlates with MASLD severity in people. In old livers, chronic ferroptotic stress amplifies the susceptibility of hepatocytes to lipotoxicity when a secondary metabolic stress is imposed, excerbating DNA damage and enhancing accumulation of senescent hepatocytes with a Senescence-Associated Secretory Phenotype (SASP). Release of SASP factors from the ferrosenescent hepatocytes promotes fibrosis and inflammation, driving progression of MASL to MASH. The ferroptosis inhibitor Fer1 suppresses ferroptotic stress, reduces the burden of senescent cells and improves MASH and fibrosis, suggesting that targeting the ferroptosis pathway and its associated pathobiology may help to extend the health span of old livers and thus, prevent aging-related exacerbation of MASLD. Moreover, liver aging induces dysfunction of multiple organ systems, including heart, kidney and pancreas.

METHODS and MATERIALS

Animal studies

Mice.

A total of 57 C57BL/6J male mice were obtained from Jackson Laboratories (Bar Harbor, ME) or NIA Aged Rodent Colonies, and housed in a barrier facility on 12 h:12 h light cycle with free access to water. Unless otherwise noted, all young mice were studied at 3 months of age and all old mice were studied at 2 years of age. All studies described below were approved by the Duke University Institutional Animal Care and fulfilled National Institutes for Health and Duke University IACUC requirements for humane animal care.

Liver perfusion and primary hepatocyte isolation.

A total of 12 mice were used for these studies. In experiment 1, primary hepatocytes were harvested from 3 young mice and 3 old mice. Freshly isolated hepatocytes were processed to isolate RNA for bulk RNA seq analysis as we described⁵. In experiment 2, primary hepatocytes were isolated from an additional 3 young mice and 3 old mice and cultured for up to four days to assess the effects of age on susceptibility to lipotoxicity. Viability of primary hepatocyte preparations was determined by trypan blue exclusion and generally >95%. Cell purity was evaluated with an assessment of the relative percentage of hepatocytic-appearing cells using standard microscopy, and it was generally > 96% hepatocytes. Hepatocytes were plated in DMEM F12 medium (Gibco) containing 5% FBS and 1% streptomycin and penicillin. After 4 hours, cells were washed and treated with 200μM of oleate (Sigma-Aldrich, St. Louis, Missouri) and 100μM palmitic acid (Sigma-Aldrich, St. Louis, Missouri) in DMEM F12 (Gibco, Billings, Montana) containing 2% FBS (HyClone, Logan, Utah), 1% Penicillin/streptomycin (Gibco), 1% L-glutamine (Gibco), 0.1% Insulin-transferrin-selenium (Gibco), and 40ng/ml Dexamethasone (Sigma-Aldrich). The cells were maintained for 4 days and fixed with 4% paraformaldehyde (Santa Cruz, Dallas, Texas) for following immunofluorescent analysis.

Phenotypic comparison of livers from young and old mice.

Liver tissues were harvested from a distinct cohort comprised of 8 young mice and 8 old mice. Parameters of MASL (‘simple’ steatosis), MASH (hepatic lipotoxicity and inflammation), and liver degeneration (fibrosis) were compared to determine how chronological aging per se impacts MASLD pathogenesis/progression.

CDA-HFD studies.

To examine the impact of age on MASLD susceptibility, additional young (3 months old, n = 9) and old (2 years old, n = 12) mice were fed with a CDA-HFD diet (A06071302, choline-deficient, L-amino acid defined diet with 60 kcal% fat; Research Diets, Inc, New Brunswick, NJ) for 6 weeks. CDA-HFD induces robust histologic and biochemical indicators of fibrosing MASH in a relatively short time^82–85 and mimics the effects of dietary choline deficiency that are known to promote MASLD in people^45,46. For the exploration of ferroptosis’s role in age-dependent MASLD progression, old mice received intraperitoneal injections of 10 mg/kg Ferrostatin-1 (Fer-1, Cayman, Ann Arbor, Michigan) or its vehicle every other day during the last 8 days of the study, with injections consistently administered at ~ 2 pm (n = 6 per group). Existing literature has reported the anti-ferroptotic efficacy of this Fer-1 regimen in various in vivo murine models, including those studying age-related degenerative diseases^86–88, and various liver diseases such as hemochromatosis and iron overload‒induced liver fibrosis^89,90, hepatic ischemia-reperfusion injury⁹¹, thioacetamide-induced acute liver injury⁹², alcoholic liver diseases⁹³, and NAFLD^94–96. In a separate experiment, 12 weeks old C57BL/6J male mice were fed with either chow or CDA-HFD for 22 weeks, and nuclei were isolated from liver tissues for snRNA-seq analysis (n = 4 mice/group). Mice were not subjected to fasting before sacrifice, and sacrifices were consistently conducted between 2 – 4 pm. At the end of diet administration in both CDA-HFD studies, mice were sacrificed, blood was obtained, and liver tissues were fixed in phosphate-buffered formalin for histological analysis, or flash-frozen in liquid nitrogen and stored at −80 °C.

RNA-seq library construction and analyses

Total mRNA was extracted using Trizol reagent following the manufacturer’s instructions, converted to cDNA using the High-Capacity cDNA Reverse Transcription Kit (ThermoFisher Scientific, Waltham, MA), and sequenced on Illumina NovaSeq sequencer with 150 bp paired-end reads by Novogene company. Raw reads was processed by following steps: trimmed off adapter and low-quality reads, cleaned reads aligned to the reference genome using STAR v2.7.5⁹⁷, and quantified gene counts using featureCounts⁹⁸. Differential expression genes (DEG) were identified using DESeq2⁹⁹, and pathway gene enrichment was analyzed by gene set enrichment analysis (GSEA) software (version 4.2.2) using the whole transcriptome dataset. The RNA-seq raw files and gene counts will be uploaded to the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus database upon the acceptance of the manuscript.

Development and Application of Aging Hepatocyte Gene Signature (AHGS)

AHGS development.

To develop a gene signature of hepatocyte aging, we first reanalyzed our recently published RNA-seq dataset (GSE181761) generated from freshly isolated hepatocytes from young (3months) and old (2 years) C57BL/6J mice⁵. Differentially expressed genes (DEGs) were quantified using DESeq2⁹⁹ and those that were up-regulated in old hepatocytes with different cutoffs of adjusted p-values (0.05 or 0.01) and log2FoldChange (from 1 to 5) were identified (Suppl Fig 2A). We reasoned that the optimal gene signature should contain a limited number of genes and could be validated across independent datasets. Therefore, we next compared these DEG gene sets with DEGs in bulk liver RNA-seq data sets from a different cohort of mice that were fed a CDA-HFD diet for 6 weeks (vehicle treated young mice; vehicle treated old mice; old mice treated with Fer1, n = 4 per group) using the same approach. Finally, we compared DEGs from those two independent sets (i.e., DEGs from primary hepatocytes of old vs. young mice fed with chow-fed diet, and DEGs from liver tissues of old vs. young mice fed with CDA-HFD for 6 weeks) using Jaccard index.

The Jaccard index was calculated using following equation:

$$J=\frac{n_o}{n_1+n_2-n_o}$$

Where $n_o$ represents number of overlapping DEGs, $n_1$ represents number of DEGs in dataset 1, and $n_2$ represents number of DEGs in dataset 2. The $J$ of 1 suggests the two datasets completely overlap, and $J$ of 0 suggests no overlapping. As shown in Suppl Fig 2B, for adjusted p-value < 0.05, the Jaccard index decreased continuously as the log2FC cutoff increased; but for adjusted p-value < 0.01, the Jaccard index demonstrated an upward trend with an increment of the Log2FC cutoff. Notably, the Jaccard index reached its peak between a Log2FC of 2 and 3 (Suppl Fig 2B). Consequently, we opted to utilize the DEG sets defined by an adjusted p-value of 0.01 and a Log2FC of 3 as the AHGS.

Deconvolution of independent RNA-seq data sets using AHGS.

Enrichment scores of AHGS were calculated using the Gene Set Variation Analysis (GSVA)¹⁰⁰ based on normalized gene expressions per cohort.

The predictive power of this gene signature was firstly evaluated in 5 independent MASLD cohorts:

1. Duke MASLD cohort¹⁶ (GSE213623; n=368). This cohort was generated from healthy obese controls (no histologic features of chronic liver disease, n = 69) or biopsy-proven MASLD (n = 299) who had consented to have liver, blood and clinical data archived in the Duke University Health System (DUHS) MASLD Clinical Database and Biorepository. A summary of patient characteristics is detailed in Supplemental Table 2.

2. German MASLD cohort¹⁰¹ (GSE33814; n=25). Transcriptome analysis of microarray data was performed in liver tissue surgical samples (normal n=13; steatohepatitis n=12) using Affymetrix HG-U133 Plus 2.0 array. Raw “.Cel” data were downloaded using R GEOquery package, then normalized using R gcrma method.

3. Japanese MASLD cohort¹⁰² (GSE167523; n=98). This is a retrospective cohort study conducted on liver tissue samples from a total of 98 patients. Among these patients, 51 had simple steatosis, while the remaining 47 had non-alcoholic steatohepatitis (NASH). Gene count matrix of bulk RNA-seq was downloaded from NCBI GEO database and normalized by DESeq2.

4. European MASLD cohort¹⁰³ (GSE135251; n=216). Bulk RNA-seq were conducted on liver biopsy from 206 MASLD and 10 healthy controls. The fibrosis degrees of MASLD vary from F0 to F4 (cirrhosis). The MASLD of this cohort comes from France, Germany, Italy and UK. DESeq2 was used to normalize the raw counts for further AHGS deconvolution analysis.

5. Single nuclei RNA-seq on liver of controls and MASLD patients²³ (GSE174748; n=4). Single nuclei RNA-seq data were generated from healthy livers (n=2) and MASLD explants (n=2); information about sex and age for these subjects was not disclosed in this study. Seurat software was used to perform quality control, normalization, cell clustering on raw gene counts. AUCell¹⁰⁴ was used to identify clusters that is enriched with AHGS, and “aging cells” were defined by Area Under Curve (AUC) scores across expression ranking of all genes per cell with adjusted cutoff.

Enrichment score of AHGS was also assessed in publicly available data sets from healthy livers of:

1. Young and old mice: RNA seq data was generated from very old mouse livers (≥ 24 months old, n=7) and young mouse livers (≤ 3 months old, n=12) (GSE132042)¹⁰⁵. Transcriptomes of these livers were also compared to examine whether aged livers exhibit MASLD-like gene profiles.

2. Humans across the age spectrum: RNA seq data generated from 226 healthy human livers from GTEx v8 (https://gtexportal.org/home/).

We also tested the AHGS in 3 other dysfunctional human organ systems:

1. Heart failure⁶⁴ (gene count was obtained from https://zenodo.org/record/4287217). Bulk RNA-seq was performed on left or right ventricular septal endomyocardial biopsies from heart failure with preserved ejection fraction (HFpEF), heart failure with reduced ejection fraction (HFrEF), and control. A total of 145 samples were included in this study, including HFpEF (n=41), HFrEF (n=59) and Normal (n=45). DESeq2 were used to normalize the raw counts and GSVA was applied for AHGS deconvolution analysis.

2. Kidney fibrosis⁶⁶ (gene count was obtained from E-MTAB-2502 from EMBL-EBI website https://www.ebi.ac.uk/biostudies/arrayexpress/studies/E-MTAB-2502). Microarray data were downloaded and comprised gene expression from 79 normal and fibrotic human kidney tubule samples. Four groups were defined by the authors, including Control (n=20), Diabetes-chronic kidney disease (D-CKD), Diabetes (n=21), and hypertension (n=19). Gene probe annotation is based on Affymetrix HG-U133A_2 Array. For genes having multiple probes, we took the mean of multiple probes as their gene expression in our analysis.

3. Pancreatic Islets⁶⁷ (gene count was obtained from GSE50244). Bulk RNA-seq was conducted in human pancreatic islets from 89 donors. Following Fadista and colleagues, three groups were defined based on level of HbA1c: Normal with HbA1c < 6% (n=51), impaired glucose tolerance (IGT) with 6% <= HbA1c < 6.5% (n=15), and type 2 diabetes (T2d) with HbA1c >= 6.5% (n=11). DESeq2 was used to normalize the raw counts and GSVA was applied for AHGS deconvolution analysis.

Available information on sex and age for the above human cohorts was collected from the original manuscripts and summarized in Supplemental Table 3.

Quantitative real time PCR (qRT-PCR)

Total RNA was isolated from whole liver tissue or cultured cells using Trizol (ThermoFisher Scientific) reagent. Complementary DNAs (cDNA) were produced using Superscript II Reverse Transcriptase (Life Technologies, Carlsbad, CA) in accordance with the manufacturer’s instructions after measuring the RNA’s concentration and purity. cDNA was quantified by qRT-PCR using SYBR Green Super-mix (Life Technologies) using primers listed in Supplementary Table 4. Results were normalized to the housekeeping gene S9 based on the threshold cycle (Ct) and relative fold change was determined using the 2-ΔΔCt method.

Immunoblot

Protein was extracted from isolated cell pellets or whole liver tissue using RIPA buffer with protease inhibitors (Sigma-Aldrich). Equal quantities of protein were loaded and subjected to SDS-PAGE gel electrophoresis using 4%−20% Criterion gels (BioRad, Hercules, CA). Subsequently, the proteins were transferred onto PVDF membranes and probed with the following primary antibodies: αSMA (Abcam, ab32575), Vimentin (Abcam, ab92547), Acsl4 (Santa Cruz, sc-271800), MDA (Abcam, ab243066), Tfrc (ThermoFisher Scientific; 13–6800), P21 (Abcam, ab188224), or β-tubulin (Abcam, ab6046). Blots were then incubated with HRP-conjugated secondary antibodies and visualized with Image Studio^™ Lite Ver 5.2 (LI-COR Biosciences).

MDA assay

The concentrations of MDA in total liver and serum of mice were analyzed using mouse MDA ELISA kits (MyBioSource, San Diego, CA; MBS741034), in strict accordance with the manufacturer’s instructions and detected using a microplate reader (Tecan, Männedorf, Switzerland).

Triglyceride quantification.

Triglyceride levels in total liver tissues were measured using Triglyceride Assay Kit (Abcam, ab65336), in strict accordance with the manufacturer’s instructions and detected using a microplate reader (Tecan, Männedorf, Switzerland).

Histopathological analysis, Immunohistochemistry (IHC) and Immunocytochemistry (ICC)

Liver tissue samples were fixed in formalin, embedded in paraffin, and sectioned. The sections were subjected to various staining techniques for histopathologic evaluation as detailed in our earlier publications (REFS). Hematoxylin and eosin (H&E) staining was performed to assess overall liver histopathology. Picrosirius red staining (Sigma-Aldrich, 365548) was used to assess fibrosis intensity across whole liver sections.

For immunohistochemistry (IHC), the slides were dewaxed, hydrated, and treated with 3% hydrogen peroxide for 10 minutes to block endogenous peroxidase activity. Antigen retrieval was performed by heating the sections in 10 mmol/L sodium citrate buffer (pH 6.0) for 10 minutes. Sections were then blocked with Dako protein block solution (Agilent Technologies, Santa Clara, CA) for 1 hour and incubated overnight at 4°C with specific primary antibodies: Acsl4 (Santa Cruz, sc-271800), 4-HNE (Abcam, ab46545), F4/80 (Cell Signaling, Danvers, MA; 70076), Tfrc (Abcam, ab214039). Polymer-horseradish peroxidase secondary antibodies were applied for 1 hour at room temperature, followed by detection using the Dako 3,3’-Diaminobenzidine Substrate Chromogen System. To detect liver DNA damage, sections were incubated with anti-gamma H2A.X (Abcam, ab11174) and anti-8-Hydroxy-2’-deoxyguanosine (Abcam, ab48508) primary antibodies, and then with Alexa Fluor^™ 488 and Alexa Fluor^™ 594 secondary antibodies (ThermoFisher Scientific) for fluorescence detection. Images were acquired and processed using Leica Microsystems.

Frozen liver tissue samples were also used. Sections were cut at a thickness of 20 μm, fixed with 10% formalin, and stained with Oil Red O (Sigma-Aldrich, O0625) for 10 minutes to visualize lipid accumulation. Cellular senescence in the liver was evaluated by SA-β-gal staining using a commercially available kit (Cell Signaling, 9860), following the manufacturer’s instructions. Results were examined using light microscopy.

For immunocytochemistry (ICC), cells were washed, fixed in 4% paraformaldehyde, permeabilized, blocked with normal goat serum, and incubated overnight with anti-gamma H2A.X (Abcam, ab11174) and anti-8-Hydroxy-2’-deoxyguanosine (Abcam, ab48508) primary antibodies. After washing with PBS, cells were incubated with Alexa Fluor^™ 488 and Alexa Fluor^™ 594 secondary antibodies (ThermoFisher Scientific) for 1 hour at room temperature. Nuclei were visualized using 4’,6-diamidino-2-phenylindole (DAPI). Images were acquired and processed using Leica Microsystems.

Statistics

Data were expressed as mean ± SEM. Statistical significance between two groups was evaluated using the student’s t test, while comparisons of multiple groups were assessed by one-way analysis of variance (ANOVA), followed by Student–Newman–Keul’s test. p ≤ 0.05 was considered to be statistically significant. Plots from RNA-seq datasets are generated using R ggplot2. Wilcox non-parameters test or t-test was used to compare enrichment score among groups, as implemented in R ggpubr software.

Supplementary Material

1

Suppl. Fig. 1 Top 10 upregulated/downregulated KEGG pathways in old versus young mouse hepatocytes. Red arrows point to the upregulated pathways of interest, and blue arrows point to the downregulated pathways of interest.

Suppl. Fig. 2 Identification of the optimal aging hepatocyte gene signature (AHGS). (A) Number of DEGs in each gene set after applying different adjusted p-value cutoffs (p = 0.05 or 0.01) and log2 fold changes (1 – 5) for DEGs. (B) Jaccard index, examining similarity of overlapping DEGs of primary hepatocytes from old vs young mice fed with chow diet, and liver tissues from old vs young mice fed with CDA-HFD diet, revealed that AHGS with p-value<0.01 and Log2 FC>3 resulted in the best gene set by showing a reduced gene number and higher Jaccard similarity.

Suppl. Fig. 3 Characterization of the aging hepatocyte gene signature (AHGS). (A) GO:MF (Molecular Function) clustering of the AHGS. Gene Set Enrichment Analysis (GSEA) of AHGS further identified (B) upregulated GO:BP (GO Biological Process) pathways and (C) upregulated HPO (Human Phenotype Ontology) pathways.

Suppl. Fig. 4 Age is a major risk factor of MASLD development. (A) Re-analysis of the Duke MASLD cohort (GSE213623) revealed that the chronological age positively correlates with MASLD histological markers including hepatocyte ballooning, portal inflammation and fibrosis in MASLD patients. (B, C) GSEA using KEGG revealed that genes related to longevity and its associated mechanisms (e.g. nicotinamide; FoxO signaling) are depleted, while genes related to programmed cell death are highly enriched in the transcriptomics of MASLD patients. Red arrows point to the pathways of interest. (D) AHGS enrichment increases with chronological age in MASLD patients but not control subjects. (E) AHGS enrichment score with chronological age in RNA-seq data of 226 normal liver samples from GTEx v8 database (https://gtexportal.org/home/). p-values were calculated using Wilcoxon Rank Sum test. Boxplot shows the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. **p < 0.01; ***p < 0.001; ****p < 0.0001.

Suppl. Fig. 5 Aging-associated mechanisms are altered during MASLD development. GSEA of bulk liver RNA seq data from Duke MASLD patients (GSE213623) revealed that (A) genes related to longevity and its associated mechanisms (e.g. NAD metabolism, sirtuins) are depleted, (B) while genes related to senescence are highly enriched in the transcriptomics of MASLD patients. p-values were calcualted using permutation testing.

Suppl. Fig. 6 AHGS distinguishes MASH patients from healthy controls. (A) AHGS was applied to deconvolute bulk liver RNA seq data (GSE135251, heathy control n = 13; MASH n = 12). AHGS is enriched in transcriptomes of MASH patients and distinguishes MASH patients from healthy controls. Boxplot shows the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. p-values were calculated using Wilcoxon Rank Sum test.

Suppl. Fig. 7 snRNA-seq analysis and bulk RNA-seq GSEA of MASLD liver. (A, C) Uniform manifold approximation and projection (UMAP) visualization of liver cells in single-nucleus RNA-seq dataset generated from two MASLD patients or two healthy controls (GSE174748), or mice fed with CDA-HFD diet or chow diet for 22 weeks. (B, D) We focused our analysis on the clusters that are most enriched for hepatocyte markers and depleted for non-hepatocyte markers. GSEA of bulk liver RNA-seq data from Duke MASLD patients (GSE213623) revealed that genes related to (E) apoptosis, (F) pyroptosis and (G) necroptosis are all enriched in the liver transcriptomics of MASLD patients. p-values were calcualted using permutation testing.

Suppl. Fig. 8 Livers of MASLD patients exhibit ferroptotic stress. (A) Duke MASLD cohort patients (GSE213623) with the same fibrosis stage were stratified into young (age ≤ 35), middle age (35 <age< 55) and old groups (age ≥ 55). Expression of ferroptotic-associated genes was compared among the different age groups. (B) Expression of genes related to iron homeostasis and (C) ferroptotic stress was compared between MASLD patients versus controls, and in patients with different fibrosis stages. The average expression of in each single cell was calculated by the AddModuleScore function of Seurat. Feature plot showed single cells on dimensional reduction plot, with cell color indicating their relative expression levels. Violin plot showed the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. p-values were calculated using Wilcoxon Rank Sum test.

Suppl. Fig. 9 Ferroptotic stress, lipotoxicity, inflammation and collagen metabolism are increased in livers of old mice. (A) MDA levels were measured by western blots in livers and primary hepatocytes from chow-fed young (3 months old, n = 4) and old (2 years old, n = 4 – 5). Data are graphed as mean ± SEM.*p < 0.05 versus young liver or hepatocytes. p-values were calcualted using one-tailed Student’s t-test. (B) A published dataset (GSE132042) was re-analyzed to compare the transcriptomes of very old mouse liver (≥ 24 months old, n=7) to young mouse liver (≤ 3 months old, n=12). AHGS was enriched in the transcriptome of liver of aged mice. Boxplot shows the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. p-values were calculated using Wilcoxon Rank Sum test. GSEA analysis further revealed that the transcriptomes of aged mouse liver are enriched with genes associated with (C) hepatocyte aging, (D) lipid metabolism, (E) inflammation and (F) collagen activity. p-values were calcualted using permutation testing.

Suppl. Fig. 10 Ferrostatin-1 protects hepatocytes from aging-related ferroptotic stress and senescence during diet-induced MASH. Young and old mice were fed with a chow diet or CDA-HFD diet for 6 weeks. Old mice were intraperitoneally injected with Fer1 or its vehicle every other day during the last 8 days of feeding (n = 3 mice/young chow; n = 9 mice/ young CDAHFD veh; n = 6 mice/ old CDAHFD veh; n = 6 mice/ old CDAHFD Fer1). (A) Body weights and liver weights were measured on the day of sacrifice. (B) Expression of ferroptosis-related proteins detected by western blotting. (C) Representative Tfrc staining and quantification of the positively-stained areas. (D) Expression of senescent marker p21 detected by western blotting. (E) GSEA demonstrated that transcriptome of old mice treated with Fer1 is depleted with genes involved in response to oxidative stress and related mortality. Protein expression was quantified by densitomeric analyses of western blots. Data are graphed as mean ± SEM. ^#p < 0.05 versus young mice + veh; ^$p < 0.05 versus old mice + veh. p-values were calcualted using: one-way ANOVA for A, B, C, D; permutation testing in E.

Suppl. Fig. 11 GSEA of the DEGs that upregulated in old mice but reversed by ferrostatin1. Venn diagram identified 520 DEGs that were upregulated in old mice (vs young mice + veh) but reversed by Fer1 (vs old mice + veh). GSEA further identified 14 significantly enriched hallmark pathways in these DEGs.

Suppl. Fig. 12 GSEA of the DEGs that downregulated in old mice but reversed by ferrostatin1. Venn diagram identified 658 DEGs that were downregulated in old mice (vs young mice + veh) but reversed by Fer1 (vs old mice + veh). GSEA further identified 7 significantly enriched hallmark pathways in these DEGs.

Suppl. Fig. 13 FXR and ferroptotic stress interact to modulate age-dependent susceptibility to MASLD. Young and old mice were fed with CDA-HFD diet for 6 weeks. These mice were intraperitoneally injected with Fer1 or its vehicle every other day during the last 8 days of feeding. Transcriptomes of liver tissues were analysed by RNA-seq (n = 4 mice/group). (A) GSEA demonstrated that liver transcriptome of old mice is depleted of genes involved in bile acid metabolism and this is reversed by Fer1 treatment in the old mcie. (B) Expression of FXR/Nr1h4. (C) GSEA demonstrated that liver transcriptome of old mice is depleted of genes involved in FXR pathway activity, and this is reversed by Fer1 treatment in the old mcie. (D) Expression of FXR/Nr1h4 in liver of Duke NAFLD cohorts (GSE213623). (E) Expression of FXR/Nr1h4 in hepatocytes subpopulations from single-nucleus RNA-seq dataset of two MASLD patients or two healthy controls (GSE174748). The average expression of in each single cell was calculated by the AddModuleScore function of Seurat. Feature plot showed single cells on dimensional reduction plot, with cell color indicating their relative expression levels. Violin plot showed the upper quantile (75%), median (50%) and lower quantile (25%) of overall data distribution. p-values were calcualted using: permutation testing in A, C; Wilcoxon Rank Sum test in B, D, E.

Abbreviations used in this paper:

αSMA

alpha-smooth muscle actin

AHGS

aging hepatocyte gene signature

ALT

alanine aminotransferase

AST

aspartate aminotransferase

BSA

bovine serum albumin

BMI

body mass index

CDA-HFD

choline-deficient L-amino acid-defined high-fat diet

DEG

differentially expressed genes

DMSO

dimethyl sulfoxide

ECM

extracellular matrix

FBS

fetal bovine serum

Fer1

ferrostatin 1

GPX4

glutathione peroxidase 4

GSEA

Gene Set Enrichment Analysis

GEO

Gene Expression Omnibus

GSH

glutathione

HCC

hepatocellular carcinoma

H&E

Haemotoxylin and Eosin

Hep

hepatocyte

HSC

hepatic stellate cell

ICC

immunocytochemistry

IF

immunofluorescence

IHC

immunohistochemistry

i.p.

intraperitoneal

KEGG

Kyoto Encyclopedia of Genes and Genomes

MASLD

metabolic dysfunction-associated steatotic liver disease

OA

oleic acid

PA

palmitic acid

PBS

phosphate buffered saline

qRT-PCR

quantitative reverse-transcription polymerase chain reaction

ROS

reactive oxidative stress

SASP

senescence-associated secretory phenotype

TUNEL

terminal deoxynucleotidyl transferase (TdT) dUTP Nick-End Labeling

VIM

vimentin