Research ArticleLiver disease

TGFβ inhibition restores a regenerative response in acute liver injury by suppressing paracrine senescence

See allHide authors and affiliations

Science Translational Medicine  15 Aug 2018:
Vol. 10, Issue 454, eaan1230
DOI: 10.1126/scitranslmed.aan1230

Setting liver regeneration free

The liver is an excellent model of organ regeneration; however, regeneration may fail in a normal liver after acute severe injury such as acetaminophen poisoning. Bird and colleagues now show that a process that prevents proliferation termed senescence, which is classically associated with aging and carcinogenesis, inhibits the liver’s regenerative cells after acute injury. This senescence can be spread from cell to cell by the signaling molecule transforming growth factor–β (TGFβ). When TGFβ signaling was blocked during acetaminophen poisoning in mice, senescence was impeded, regeneration accelerated, and mouse survival increased. Therefore, targeting senescence induced by acute tissue injury is an attractive therapeutic approach to improve regeneration.


Liver injury results in rapid regeneration through hepatocyte proliferation and hypertrophy. However, after acute severe injury, such as acetaminophen poisoning, effective regeneration may fail. We investigated how senescence may underlie this regenerative failure. In human acute liver disease, and murine models, p21-dependent hepatocellular senescence was proportionate to disease severity and was associated with impaired regeneration. In an acetaminophen injury mouse model, a transcriptional signature associated with the induction of paracrine senescence was observed within 24 hours and was followed by one of impaired proliferation. In mouse genetic models of hepatocyte injury and senescence, we observed transmission of senescence to local uninjured hepatocytes. Spread of senescence depended on macrophage-derived transforming growth factor–β1 (TGFβ1) ligand. In acetaminophen poisoning, inhibition of TGFβ receptor 1 (TGFβR1) improved mouse survival. TGFβR1 inhibition reduced senescence and enhanced liver regeneration even when delivered beyond the therapeutic window for treating acetaminophen poisoning. This mechanism, in which injury-induced senescence impairs liver regeneration, is an attractive therapeutic target for developing treatments for acute liver failure.


After moderate liver injury or resection, the liver regenerates efficiently through hepatocyte proliferation (1, 2). However, after severe acute liver injury, there is a failure of regeneration, and acute liver failure may follow. Acute liver failure can be caused by a variety of insults including viruses, toxins, and medical therapy, with the most common single agent in the Western world being acetaminophen (paracetamol) (3). Annually, there are about 2000 patients affected by acute liver failure in the United States. Despite its relative rarity, acute liver failure is clinically important, because of its high morbidity and mortality in previously healthy individuals. Outcomes in acute liver failure have improved modestly with advances in supportive care (4). However, once acute liver failure of a defined clinical severity is established, no specific medical therapies exist, recovery is unlikely, and, unless liver transplantation occurs, death usually ensues (5). New therapies are needed for the potential treatment window during this progression from acute liver injury to the most severe forms of acute liver failure.

Cells may enter growth arrest in response to stress, which is termed senescence when permanent. Senescence is associated with changes in morphology and lysosomal activity including senescence-associated β-galactosidase (SA-βGal) expression. Senescence is also marked by both a DNA damage response, which includes alterations in chromatin structure (for example, γH2Ax expression), and the activation of a dynamic pro-inflammatory senescence-associated secretory phenotype (SASP) including expression of interleukin-1α (IL-1α) and transforming growth factor–β (TGFβ) (6, 7). When senescence results from oncogenic stress, it reinforces cell cycle arrest in an autocrine manner (8, 9), activates immune surveillance (1013), and induces paracrine senescence via SASP (14, 15). SASP may also modulate fibrosis and regeneration in response to acute tissue injury (11, 16, 17). Hepatocyte senescence involves the induction of genes encoding p53 (TRP53), p21 (WAF1), and p16 (INK4A) (18). It is described in both chronic diseases (19) and steatosis (20), but not in acute liver disease. Fibroblast senescence occurs in the dermis after acute wounding (21) and acute myocardial infarction (22). However, it is not clear whether there is acute epithelial senescence in response to liver injury.

Here, we show that acute liver injury is associated with a suite of senescence markers in previously uninjured hepatocytes. We show that senescence is transmitted between hepatocytes in a feedback loop that is dependent on TGFβ derived from macrophages. Targeting TGFβ signaling after acetaminophen-induced injury reduced senescence development and improved both regeneration and survival in a mouse model of acute liver injury and failure.


Acute liver injury results in proportionate acute hepatocellular senescence

Human liver specimens resected at the time of liver transplantation from patients with hyper-acute fulminant hepatic failure (less than 1 week from jaundice to encephalopathy, with no prior liver disease) showed expression of various senescence-related markers including p21 (Fig. 1A), DcR2, γH2Ax, and SA-βGal (fig. S1). Thus, in the most severe form of liver disease, a previously healthy human liver developed widespread markers of hepatocellular senescence within days of acute insult. To examine a potential relationship between disease severity and senescence induction, we then analyzed a case series of human diagnostic liver biopsy samples from patients with submassive hepatic necrosis. Here, we observed a direct association between hepatic necrosis and hepatocyte senescence, as well as an indirect association between necrosis and hepatocellular proliferation (Fig. 1B and fig. S1). Therefore, worsening acute liver injury in humans results in a proportional expression of senescent markers by hepatocytes, associated with a reduced capacity for liver regeneration.

Fig. 1 Human liver necrosis causes acute hepatocellular senescence.

(A) Representative images of sections of explanted human liver after liver transplantation for severe acetaminophen overdose (n = 8) compared to control healthy human liver. Explanted livers injured by acetaminophen overdose show expression of the senescence marker p21 detected by immunohistochemistry in residual hepatocytes surrounding areas of necrosis. Necrosis interface, dashed white line; CV, central vein; black asterisk indicates area of necrosis. As a control, human liver with normal histology was used (n = 50). Scale bars, 50 μm. (B) A case series (n = 74) of patients with submassive liver necrosis divided into subgroups according to the extent of hepatocellular necrosis is presented. <25%, n = 8; 25 to 50%, n = 16; 50 to 75%, n = 22, >75%, n = 28. The extent of hepatocellular submassive necrosis (defined histologically by globalized confluent necrosis) was quantified by immunohistochemistry for the hepatocellular senescence marker p16 and the proliferation marker Ki67. *P < 0.05, one-way analysis of variance (ANOVA). Mean ± SEM.

To investigate the functional impact of senescence in acute liver injury, we examined the established murine models of acute liver injury induced by carbon tetrachloride (CCl4) (23) or acetaminophen (24). Both injury models resulted in expression of senescence markers (p21, SA-βGal, and p16) by hepatocytes, demonstrating features of growth arrest [absence of 5-bromo-2′-deoxyuridine (BrdU) or Ki67], the DNA damage response (γH2Ax), senescence-associated heterochromatic foci (HMGA2), and SASP (IL-1α) (Fig. 2, A to C, and figs. S2 and S3). We also observed senescence marker expression by hepatocytes in dietary mouse models inducing either hepatocellular steatosis plus injury [choline-deficient ethionine (CDE) diet] or biliary-related liver injury [3,5-diethoxycarbonyl-1,4-dihydrocollidine (DDC) diet] (fig. S4). In both CCl4- and acetaminophen-induced acute liver injury, expression of senescence markers was maximal 2 days after initiation and was lost after hepatocellular recovery (Fig. 2D, figs. S2 and S3, and table S1). In acetaminophen injury, using unbiased transcriptomics, we confirmed a senescence-associated gene expression signature 24 hours after injury (Fig. 2E and tables S2 and S3). At 48 hours after acetaminophen injury, we observed p21 expression particularly focused to hepatocytes surrounding the area of receding necrosis at this time of injury resolution (Fig. 2F and fig. S3C). In acute liver injury–induced senescence, the principal target population was the hepatocyte; however, p21 expression by nonparenchymal cells also occurred (fig. S2), consistent with the previous report by Krizhanovsky et al. (11).

Fig. 2 Toxin-mediated liver injury causes p21-dependent hepatocellular senescence in mice.

(A and B) In murine toxin–induced acute liver injury models, mice were treated with either CCl4 (A) or acetaminophen (B). Treatment with these toxins resulted in pericentral necrosis 2 days after administration as shown by immunohistochemistry for expression of the senescence marker p21 (green); expression of the proliferation marker BrdU (magenta) and the hepatocyte marker hepatocyte nuclear factor 4 α (magenta) is also shown. (C) Immunohistochemistry for expression of the proliferation marker Ki67 is shown 2 days after acetaminophen treatment. Staining indicates hepatocyte proliferation away from but not next to the area of necrosis; red arrows indicate proliferating hepatocytes. (D) Quantification of p21+ hepatocytes after injury; n ≥ 3 for each time point, P < 0.0001 versus time 0, two-way ANOVA. (E) Gene set enrichment analysis (GSEA) plot showing enrichment of the early (24 hours) acetaminophen injury gene expression signature in liver compared to an oncogene-induced senescence (OIS) signature. Gene set: IMR90 ER:RAS OIS cell model (15). Enrichment score is 0.2564; normalized enrichment score is 2.466; nominal P < 0.001. (F) Perinecrotic hepatocytes (brown nuclei) were quantified for p21 expression 2 days after acetaminophen treatment; 74.9% of total perinecrotic hepatocytes expressed p21 (n = 8 mice). (G) Immunohistochemistry for expression of the proliferation marker Ki67 in p21-deficient (p21KO) mice 2 days after acetaminophen-induced liver injury. Ki67 expression indicates proliferating hepatocytes in the perinecrotic area of the injured mouse liver. (H) Quantification of perinecrotic hepatocytes shown in (G). (I) The number of Ki67+ hepatocytes in relation to serum alanine transaminase (ALT; units per liter), a marker of liver injury (n = 5 versus 8 mice; 20 high-power fields were quantified per liver). P = 0.0074, two-tailed t test. Linear regression for wild-type (WT) and p21KO mice, R2 = 0.54 and 0.92, with slope 95% confidence intervals of −0.10 to −0.0045 and 0.082 to 0.28 and probability slope ≠ 0, P = 0.037 and 0.010, respectively. Scale bars, 50 μm. CV, central vein. Dashed white lines, necrosis boundary; asterisk, area of necrosis.

To assess the necessity for p21 in the formation of injury-induced senescence, we performed acetaminophen-induced injury in wild-type and p21-deficient (p21KO) mice (Fig. 2G). We measured the proliferative response within the perinecrotic area where p21 was expressed by wild-type hepatocytes. Injury was equivalent between wild-type and p21KO mice (fig. S3). However, perinecrotic hepatocellular regeneration was increased in p21KO mice compared to wild-type mice (Fig. 2H and table S1). Furthermore, whereas a negative correlation between injury and regeneration existed in wild-type mice, a positive correlation was observed in p21KO animals (Fig. 2I and table S1), indicating that in the absence of p21, injury no longer impeded the regenerative response. Together, these data show that hepatocytes can enter a p21-dependent senescent state after acute injury and that this is associated with impaired local regeneration.

TGFβ-dependent senescence transmission between hepatocytes in vivo

Rapid entry of the liver parenchyma into senescence after injury may represent a precursor to cell death. However, because we observed increasing expression of senescence markers by hepatocytes during the time of necrosis recession, we explored the hypothesis that ongoing senescence may be a cellular response to adjacent injury and existing senescence. We, and others, have shown that OIS not only is cell-autonomous but also spreads via paracrine factors (7, 15). Furthermore, transcriptomic analysis revealed a transition between a gene expression signature associated with SASP-induced senescence and one of cell cycle regulation during the induction of senescence after acetaminophen toxicity (Fig. 3A). Thus, we studied the spread of senescence in two independent genetic mouse models of senescence.

Fig. 3 Non–cell-autonomous senescence in hepatocyte-specific mouse senescence models.

(A) Plots of GSEA normalized enrichment scores comparing gene sets over time observed in the acetaminophen-treated mouse model to the unbiased top 15 ranked hallmark gene sets and the OIS signature from the IMR90 ER:RAS cell model (15). Black borders of data points highlight P < 0.05; raw data are shown in tables S2 and S3. Top and bottom panels show inflammatory and cell cycle arrest gene expression signatures. (B) Diagram showing the use of genetic induction of transgenes in hepatocytes to induce cell-autonomous senescence and assessment of senescence using a combination of markers—p53, p21, and p16. Presence of senescence markers, p21 or p16, in the absence of markers of genetic recombination, p53 or Tomato reporter (Tom), identifies non–cell-autonomous senescence. (C) p53 accumulates in a subpopulation of hepatocytes in the partial ΔMdm2 Hep mouse model where βNF (20 mg/kg) is given to AhCre+ Mdm2f l/f l mice. Immunohistochemical staining for p21/p53 and for p53/p16INK4As was assessed by confocal microscopy. (D) Immunohistochemical staining and confocal microscopic analysis of mouse liver sections for p53 and p21 after deletion of Mdm2 using AAV8-TBG-Cre [2.5 × 1011 genetic copies (GC) per mouse]. (E) Immunohistochemical staining and confocal microscopic analysis of mouse liver sections for p21 expression and GFP staining in a hepatocyte transplant mouse model 94 days after transplantation of GFP-tagged hepatocyte progenitor cells. AhCre+ Mdm2fl/fl mouse recipients were given wild-type (WT) donor cells tagged with GFP and iterative doses of βNF to induce hepatocyte recombination of Mdm2. Dashed white line, border of the engrafted cells. The magnified area is shown in individual color images on the right. (F) Immunohistochemical staining and confocal microscopic analysis of mouse liver sections for p21 expression after hepatocellular TGFβR1 activation by AAV8-TBG-Cre in LSL-TGFβR1-CA mice. (G) Immunohistochemical staining and confocal microscopic analysis of mouse liver sections for p21 expression and red fluorescent protein (RFP) staining to detect tdTomato reporter after reduced dosing of the AAV8-TBG-Cre vector (6.4 × 108 GC per animal) in LSL-TGFβR1-CA R26-LSL-tdTomato mice. (H) After partial ΔMdm2Hep, mice were given the TGFβR1 inhibitor SB525334 or vehicle control. Immunohistochemical staining and confocal microscopic analysis of mouse liver sections for p53 and p21 with quantification of non–cell-autonomous p21 expression; P = 0.0023, two-tailed Mann-Whitney test; n = 6 vehicle control versus n = 7 for SB525334-treated mice. Mean ± SEM. Scale bars, 50 μm. Open arrow, cell-autonomous senescence; closed arrow, non–cell-autonomous senescence; arrowhead, unaffected. TNFα, tumor necrosis factor–α; NFκB, nuclear factor κB; JAK, Janus kinase; STAT3, signal transducer and activator of transcription 3.

Murine double minute 2 (Mdm2) is a key negative regulator of p53, and the p53/p21 pathway is central to senescence induction. We induced up-regulation of p53 in hepatocytes by inducing hepatocyte-specific deletion of Mdm2Mdm2Hep). We achieved this using the AhCre system (25), which expresses Cre recombinase in hepatocytes in response to administration of a xenobiotic chemical [β-naphthoflavone (βNF)] rendering Mdm2 inactive. This resulted in hepatocellular injury as we have previously reported (26). In the ΔMdm2Hep mouse model, we observed a rapid expression of a suite of senescence markers (fig. S5). Using a mitogen cocktail consisting of hepatocyte growth factor (HGF) and triodothyronine (T3) (27), we attempted to promote proliferation of senescent hepatocytes in the ΔMdm2Hep model but were unable to do so, unlike in wild-type hepatocytes (fig. S5). Therefore, these cells appeared to be in a state of functional senescence. Next, we tested p21 dependence of growth arrest in this model. When Mdm2 was deleted in hepatocytes of p21KO mice, we observed rescue of the growth arrest (fig. S5). Thus, the ΔMdm2Hep model induces acute p21-dependent hepatocellular senescence.

To examine spread of senescence, we used reduced titration of βNF to delete Mdm2 in a subpopulation of hepatocytes (partial ΔMdm2Hep). By doing so, we aimed to distinguish cell-autonomous and non–cell-autonomous senescence induction in hepatocytes (Fig. 3B). We define cell-autonomous senescence as being caused by genetic manipulation of that cell (for example, through Mdm2 deletion), whereas we consider non–cell-autonomous senescence as an indirect response to environmental senescence and injury in a genetically unmanipulated cell. In the ΔMdm2Hep mouse model, a cell-autonomous senescence was observed with activation of p21/p16 in association with p53 overexpression within a subgroup of the total hepatocyte population (Fig. 3C and fig. S6). Non–cell-autonomous expression of p21/p16 occurred in a distinct hepatocyte subpopulation in the absence of p53 overexpression. These cells had atypical morphology and more pronounced p21 expression than did their neighboring p21+/p53+ Mdm2-deleted hepatocytes.

Next, we examined a potential geographical relationship between non–cell-autonomous p21 expression and regional p53+ hepatocytes using a further titrated genetic induction of hepatocyte Mdm2 deletion. We observed a lower frequency of non–cell-autonomous p21 expression and geographical clustering of non–cell-autonomous p21 expressing hepatocytes in areas dense with hepatocytes overexpressing p53 (fig. S6).

We then aimed to study senescence transmission from Mdm2-deleted hepatocytes to local hepatocytes in different mouse models. To exclude any extrahepatic effects of AhCre-mediated recombination (including the intestinal epithelium) (28), we used a hepatocyte-specific induction regime using an adeno-associated viral vector, AAV8–TBG (thyroxine binding globulin)–Cre, to induce hepatocyte-specific deletion of Mdm2 (29). As predicted, AAV8-TBG-Cre induced deletion of Mdm2 in a subpopulation of hepatocytes (fig. S6). In this model, non–cell-autonomous p21 expression was also observed (Fig. 3D). We did observe a very low (0.3% hepatocytes) Cre-independent expression of p21 as a result of transfection with the control AAV8 vector (fig. S6). To test transmission of senescence to wild-type hepatocytes, we used a large-scale hepatocyte repopulation model. Using iterative βNF dosing, we found that the livers of AhCre+ Mdm2 fl/fl mice were repopulated by hepatocytes derived from transplanted wild-type green fluorescent protein (GFP)–tagged cells (26). After final βNF dosing and p53 expression by native hepatocytes, transplanted wild-type hepatocytes expressed p21, particularly at the margins of the engrafted nodules (Fig. 3E).

In the ΔMdm2Hep model, we observed activation of the TGFβ pathway (fig. S5). TGFβ receptor 1 (TGFβR1) was expressed by hepatocytes in addition to nonepithelial cells. The TGFβR1 ligand, TGFβ1, was expressed both by nonparenchymal cells and, to a lesser degree, by hepatocytes. Because p21 is a canonical TGFβ signaling target gene and the TGFβ signaling pathway has a role in oncogene-induced paracrine senescence (14, 15), we hypothesized that the TGFβ1 ligand plays a mechanistic role in non–cell-autonomous p21 expression by hepatocytes. To test the functional role of TGFβR1 and TGFβ1 ligand in transmitted senescence, we used a model of hepatocyte-specific TGFβ signal pathway activation. Here, we used LSL-TGFβR1-CA mice, which have a genetically inducible constitutively active (CA) TGFβR1 that is expressed upon removal of a stop codon (LSL) by Cre recombinase (30). We activated this model using hepatocyte-targeted recombination (AAV8-TBG-Cre) and observed TGFβ pathway activation and acute senescence marker induction in hepatocytes. This was accompanied by liver injury and increased paracrine TGFβ1 production (Fig. 3F and figs. S7 and S8). Using this as a further model of senescence induction in vivo, we investigated non–cell-autonomous senescence driven specifically by the TGFβ pathway as a model distinct from Mdm2 deletion. Using lower titrations of AAV8-TBG-Cre, we induced TGFβ pathway activation and the R26-LSL-tdTomato reporter in about 5% of mouse hepatocytes. We observed evidence of non–cell-autonomous spread of senescence to adjacent hepatocytes in response to cell-autonomous TGFβ pathway activation (Fig. 3G). Thus, we observed non–cell-autonomous senescence in mouse models of acute hepatocellular senescence in vivo, suggesting that senescence can spread within the liver epithelium.

To test the necessity of TGFβ signaling for the transmission of senescence, we returned to the ΔMdm2Hep model. Using SB525334, a small-molecule inhibitor of TGFβR1, in the partial ΔMdm2Hep mouse model, we observed reduced hepatocellular pSMAD3 without an effect on hepatocyte ΔMdm2Hep recombination efficiency (fig. S6). SB525334 treatment of partial ΔMdm2Hep resulted in reduced non–cell-autonomous expression of p21 (Fig. 3H and table S4), demonstrating that TGFβ signaling was required for the paracrine induction of non–cell-autonomous p21 in this mouse model.

Hepatocyte senescence induced by acute liver injury is dependent on macrophage-derived TGFβ

Clinically relevant TGFβ inhibitors are currently available (31). Given our findings of TGFβ-dependent transmitted senescence in the genetic mouse models, we examined the functional role of TGFβR1 signaling in senescence formation in acute liver injury. In human fulminant liver failure, senescent hepatocytes showed TGFβ pathway activity (Fig. 4A). Acetaminophen-induced liver injury in mice was accompanied by elevated TGFβ1 (Fig. 4, B and C, and table S5). SMAD7 (a TGFβ pathway target gene) was expressed upon acute liver injury and up-regulated by perinecrotic hepatocytes (Fig. 4D and table S5). These perinecrotic hepatocytes expressed both TGFβR1 and senescence markers adjacent to local TGFβ expression (Fig. 4E and fig. S10). Therefore, we observed evidence of active TGFβ signaling in senescent hepatocytes adjacent to necrosis after acute liver injury.

Fig. 4 TGFβ signaling is activated in acetaminophen-induced hepatocellular senescence.

(A) Representative images showing immunohistochemistry for expression of p21 and pSMAD2/3 in healthy human liver and in liver from patients with fulminant hepatic failure secondary to acetaminophen overdose. White arrows indicate senescent hepatocytes. (B) Representative images showing in situ hybridization for TGFβ1 in the livers of acetaminophen-treated (350 mg/kg) and untreated C57BL/6J mice. TGFβ1 ligand is expressed by nonparenchymal cells with a monocyte-like appearance. CV, central vein. Black asterisk indicates area of necrosis. (C) Enzyme-linked immunosorbent assay (ELISA) of mouse liver TGFβ1 for untreated mice and acetaminophen-treated mice 12 hours after exposure. (n = 6 versus 7, respectively). Mean ± SEM. P = 0.0047, two-tailed Mann-Whitney test. (D) Quantification by in situ hybridization of SMAD7 expression in the perinecrotic region of mouse liver 2 days after acetaminophen treatment. P = 0.0286, compared to equivalent area in uninjured mouse liver, one-tailed Mann-Whitney test. (E) Mouse liver serial sections assessed for expression of SMAD7, TGFβR1, and TGFβ1 ligand by in situ hybridization and for p21 expression by immunohistochemistry 12 hours after acetaminophen treatment. Scale bars, 50 μm.

Macrophages are a known source of TGFβ ligands, particularly in the context of tissue injury (32). Perinecrotic macrophages in murine acetaminophen-induced liver injury expressed TGFβ1 (Fig. 5A). Because both TGFβ and CCL2 (a macrophage chemokine and known SASP component) are associated with severe human acute liver disease (33), we proceeded to examine the functional role of macrophage recruitment and TGFβ expression in senescence induction in our liver injury models. A SASP-related pro-migratory chemokine axis developed in partial ΔMdm2Hep mice with expression of both chemokine ligands and receptors (fig. S11). CCL2 was expressed by nonparenchymal cells in and around the areas of hepatocellular necrosis in acetaminophen injury (Fig. 5B) before an increase in circulating monocytes (Fig. 5C) and then local macrophage accumulation (Fig. 5D and table S6).

Fig. 5 Macrophage recruitment and TGFβ1 production drive hepatocellular senescence and impair hepatocellular regeneration in mice.

(A) A representative mouse liver section assessed for hepatic TGFβ1 ligand production (red) and F4/80+ macrophages (pale blue) by in situ hybridization and F4/80 immunohistochemistry, respectively, 2 days after acetaminophen (350 mg/kg) treatment. CV, central vein. (B) In situ hybridization staining for expression of the CCL2 chemokine. Dashed white line, necrotic interface; black asterisk, area of necrosis. (C) Immunohistochemical staining for F4/80+ macrophages (green) and p21+ hepatocytes (magenta). Scale bars, 50 μm. (D) Quantification of peripheral monocytes in mice after acetaminophen treatment versus fasted untreated mice as baseline (dashed black line). n = 5 mice for each time point. P = 0.0001, one-way ANOVA with Dunnett’s multiple comparison baseline versus day 1. (E) Quantification of immunohistochemical staining for p53 and p21 expression or for BrdU in mouse livers 4 days after partial deletion of Mdm2 (partial ΔMdm2Hep), where βNF (20 mg/kg) is given to AhCre+ Mdm2fl/fl mice, followed by twice daily antibody-mediated CCL2 inhibition (with isotype antibody as the control). Non–cell-autonomous hepatocyte p21 expression (without p53 expression) and proliferation (BrdU) were quantified. P = 0.05, Mann-Whitney (n = 3 mice per group). (F) Liposomal clodronate depletion of macrophages 3 days after partial ΔMdm2Hep compared to phosphate-buffered saline (PBS) control. TGFβ and p21 expression in whole mouse liver were quantified by quantitative reverse transcription polymerase chain reaction (PCR). P = 0.000063, 0.237, and 0.126 for TGFβ1, TGFβ2, and TGFβ3, respectively, and P = 0.025 for p21, t test (n = 4 mice per group). Non–cell-autonomous p21+ hepatocytes were quantified after immunohistochemical staining for p53 and p21. P = 0.035, t test (n = 4 mice per group). (G) Acetaminophen (350 mg/kg) was administered to LysMCre+ TGFβfl/fl or LysMCreWT TGFβfl/fl mouse littermates. Hepatocyte proliferation was assessed by BrdU immunohistochemistry. P = 0.0006, two-tailed t test (n = 10 versus 8 mice). Mean ± SEM.

To examine the role of macrophages in non–cell-autonomous senescence, we returned to the ΔMdm2Hep mouse model. Inhibition of leukocyte recruitment via CCL2 blockade in the ΔMdm2Hep model reduced non–cell-autonomous p21 expression and improved hepatocellular regeneration (Fig. 5E, fig. S11, and table S6). Next, we performed macrophage ablation using liposomal clodronate in the partial ΔMdm2Hep model. This reduced hepatic TGFβ1 expression by 87% (Fig. 5F, fig. S11, and table S6), implying that macrophages are the principal source of the TGFβ1 ligand. Consistent with this hypothesis, both p21 gene expression and non–cell-autonomous p21 expression were reduced when macrophages were depleted in the partial ΔMdm2Hep mouse model (Fig. 5F and table S6). To functionally test the role of macrophage-derived TGFβ1 ligand in liver injury, we used myeloid specific TGFβ1 deletion in the acetaminophen-induced liver injury mouse model (Fig. 5G, fig. S11, and table S6). This resulted in equivalent injury but improved liver regeneration. Therefore, macrophage-derived TGFβ1 is required for optimal induction of paracrine senescence after acute liver injury in mice.

Inhibition of TGFβR1 signaling impairs senescence induction and improves liver regeneration, function, and outcome in acute liver injury

Given the finding of TGFβ-dependent paracrine senescence in the genetic models, we tested whether this effect was also observed in two clinically relevant models of liver injury, the CCl4 and acetaminophen models. We examined the effect of TGFβ signaling disruption in both acute and chronic CCl4 liver injury models. In the acute model, we used the TGFβR1 inhibitor AZ12601011 administered 12 hours after administration of CCl4 (1 μl/g; fig. S12). This resulted in reduced senescence induction, improved liver regeneration, and reduced jaundice. In the chronic liver injury model, CCl4 was given repeatedly over 8 weeks in combination with a genetic depletion approach targeting hepatocellular TGFβR1 (ΔTGFβR1Hep; fig. S12). Again, we observed reduced hepatocellular p21 expression and increased hepatocellular proliferation. Next, we examined ΔTGFβR1Hep in acute acetaminophen-induced injury (fig. S13). We observed early necrosis equivalent to controls, but reduced hepatocellular p21 expression by perinecrotic hepatocytes. There was also an altered distribution of hepatocellular regeneration, with marked proliferation by perinecrotic hepatocytes. Accelerated resolution of necrosis was observed in mice lacking hepatocellular TGFβR1.

To test the clinical utility of TGFβR1 inhibition, we administered AZ12601011 at the time of a lethal acetaminophen dose (Fig. 6A). TGFβR1 inhibition resulted in marked clinical improvement from 6 to 16 hours and permitted survival after acetaminophen dosing (525 mg/kg; Fig. 6B and table S7). At the end point, vehicle-treated mice showed worsened jaundice compared to their AZ12601011-treated counterparts (Fig. 6C and table S7).

Fig. 6 Inhibition of TGFβR1 signaling reduces hepatocellular senescence and restores a proportional regenerative response after acetaminophen treatment in mice.

(A) Cohorts of male C57BL/6J mice were given vehicle control or were treated with the TGFβR1 inhibitor AZ12601011, starting when acetaminophen (525 mg/kg) was administered. Mice were closely monitored throughout the experiment until death or the humane end point was reached, typically between 16 and 18 hours. Initially, the mice treated with the TGFβR1 inhibitor (n = 14) were sacrificed when the control animals reached the end point irrespective of clinical condition (total biological replicates, n = 14 with AZ12601011 and n = 16 with vehicle control; performed over three separate experiments). (B) Separate survival cohorts (n = 5 in each of two experiments) treated with the TGFβR1 inhibitor were compared to simultaneous vehicle controls to examine longer-term survival; P < 0.0001, Gehan-Breslow-Wilcoxon test. (C) At matched end point, the TGFβR1 inhibitor and vehicle control groups were compared for serum bilirubin. P = 0.0162, two-tailed Mann-Whitney test. (D) In an experiment examining delayed TGFβR1 inhibition commencing 12 hours after acetaminophen treatment in male C57BL/6J mice, the TGFβR1 inhibitor SB525334, or vehicle, was given twice daily. (E) Serum bilirubin over time from (D); P > 0.05 and P < 0.01 at days 2 and 4 for SB525334 treatment compared to vehicle control, respectively; two-way ANOVA with Bonferroni correction (n = 8 mice each group). (F) Immunohistochemical staining for hepatocellular p21 expression was quantified; P = 0.049, t test, 30 high-power fields in mouse liver sections were analyzed (n = 8 mice per group). (G) Immunohistochemical staining for BrdU (representative images for mouse liver sections 2 days after acetaminophen treatment and administration of either SB525334 or vehicle control). Effect of treatment upon BrdU+ hepatocytes was quantified in both whole liver (days 2 and 4) and perinecrotic hepatocytes (day 2 only). P = 0.0075 and 0.30 for total BrdU+ hepatocytes at days 2 and 4, respectively, and P < 0.0001 for BrdU+ perinecrotic hepatocytes comparing SB525334 treatment to vehicle control, t test (n = 8 per group, except day 2 vehicle control where n = 6 per group). Scale bars, 50 μm. (H) In individual mice, 2 days after acetaminophen treatment, hepatocytes were analyzed for serum ALT and BrdU staining, and linear regression was performed. R2 = 0.15 and 0.71, with slope 95% confidence intervals of −0.0094 to 0.0038 and 0.0049 to 0.085 and probability slope ≠ 0, P = 0.34 and 0.036, respectively. (I) A nonfatal dose of acetaminophen (450 mg/kg) was administered to male C57BL/6J mice, followed by treatment with AZ12601011 or vehicle control 12 hours later. Serum bilirubin was measured, and p21 expression in hepatocytes was quantified by immunohistochemistry. P = 0.0029 and 0.0017, respectively, comparing AZ12601011 treatment to vehicle control, two-tailed t test, n = 9 per group. Data presented as mean ± SEM.

Conventional treatment of acetaminophen toxicity in humans involves N-acetylcysteine therapy that, to be effective, must be given within 8 hours after exposure for humans or 4 hours for mice (34). Many patients present to medical services too late for this to be effective (35). To model delayed treatment, we used small-molecule TGFβR1 inhibitors commencing either SB525334 or AZ12601011 treatment of mice 12 hours after acetaminophen administration (Fig. 6D). In addition, as liver injury peaks before treatment administration, this strategy was designed to test whether the improvements in clinical outcome were distinct from the reduced hepatocellular injury we observed with synchronous acetaminophen and therapy administration. With SB525334 treatment, downstream signaling through TGFβR1 was inhibited, and necrosis was unchanged (fig. S14). Liver injury was reduced upon TGFβR1 inhibition (fig. S14), along with a resolution of jaundice (Fig. 6E and table S7). Hepatocellular senescence was reduced by TGFβR1 inhibition (Fig. 6F, fig. S14, and table S7) and hepatocellular proliferation increased, both overall and specifically within the perinecrotic area (Fig. 6G and table S7). During untreated acetaminophen-induced injury, an apparent inverse relationship between severity of hepatocellular injury and hepatocellular regeneration was once again observed (Figs. 2I and 6H and table S7). Therefore, severe liver injury in the mouse recapitulates the negative correlation between injury and regeneration observed in severe human disease (Fig. 1). In the mouse, this relationship was reversed upon inhibition of TGFβR1, restoring a proportional regenerative response to liver injury, mimicking genetic deletion of p21. Using a higher dose of acetaminophen to induce nonfatal liver injury, we tested the second clinical compound AZ12601011 in the delayed treatment mouse model. Here, jaundice was once again improved and was associated with an inhibition of hepatocellular senescence (Fig. 6I and table S7). These effects were accompanied by reduced local TGFβ pathway activation in perinecrotic hepatocytes (fig. S14). Therefore, inhibition of TGFβ signaling after acute liver injury reduced hepatocellular senescence and improved liver regeneration and recovery from injury.


In contrast to minor forms of acute liver injury where regeneration occurs efficiently, increasingly severe liver injury exhibits regenerative failure and poorer prognosis (4). Validated clinical scoring systems predict the outcomes of patients who will survive versus those in whom liver regeneration will ultimately fail (36), suggesting a tipping point beyond which recovery is unlikely. The pathophysiological mechanism underlying this remains poorly understood and is a barrier to therapeutic development. Our finding that hepatocyte senescence inhibits liver regeneration may underpin this tipping point. These findings contrast with previous reports in which senescence after injury in other organs may facilitate regeneration (37) and limit fibrosis after liver injury (11).

With our data, we provide a mechanistic model whereby injury-induced senescence is amplified by macrophage-dependent paracrine TGFβ signaling (fig. S15). In our in vivo models, we observed the expression of local TGFβ ligand in response to cell-intrinsic TGFβR pathway activation. This may represent a paracrine positive feedback loop reinforcing and amplifying local TGFβ signaling and downstream senescence.

Senescence is challenging to define and study in vivo. Hepatocytes in the ΔMdm2Hep mouse model showed functional senescence in vivo. They also express a suite of senescence markers (38), including markers of the DNA damage response, growth arrest and the SASP. Likewise, the acetaminophen-induced liver injury model displayed senescence marker expression by hepatocytes. It also had a tissue transcriptomic signature matching that of an in vitro cellular oncogene-induced senescence model (the IMR90 ER:RAS model) (15). Therefore, we conclude that our two mouse models demonstrate senescence in vivo. A similarly rapid senescence program (including TGFβ production, direct Notch target activation, and p16 expression) has been observed in OIS models within 48 hours in vitro (7, 39). This is similar to our in vivo observations, which were also associated with an early inflammatory gene expression signature (for example, TGFβ, IL-6, and nuclear factor κB) within 24 hours after acetaminophen exposure. In our model, the SASP components TGFβ and IL-6 were expressed concurrently within 12 hours of acetaminophen exposure; however, Hoare et al. have shown that they appear sequentially in senescence in vitro (7). The acute injury–induced senescence described here may represent a generic response to severe tissue injury whereby regional regeneration is inhibited. Our study does not delineate the mechanisms that link hepatocellular injury to senescence initiation. Whether the recently described cGAS-STING–driven senescence pathway detecting cytoplasmic chromatin (40) plays a central role in liver injury–induced senescence remains to be studied. Furthermore, the rapid clearance of senescent hepatocytes in our liver senescence models justifies future investigations.

We have recently reported in an in vivo model of hepatocyte growth arrest that cholangiocytes, which expand as a ductular reaction, may act as facultative stem cells for generating hepatocytes with replacement of hepatocytes occurring over weeks to months (29). In comparison, resolution of liver injury and architecture in our acetaminophen-induced and CCl4-induced mouse liver injury models was complete within 1 week and was not accompanied by a ductular reaction. Future studies are required to test whether inhibition of senescence in chronic liver injury models (for example, TGFβ inhibition) may affect regeneration from both the hepatocyte and facultative stem cell pools. However, our data in mouse genetic models suggest that when senescence formation is impaired, the ductular expansion including facultative stem cells is also impaired (fig. S16).

In the chronic liver injury setting, iterative CCl4-induced fibrosis has been associated with nonparenchymal (myofibroblast) senescence and an impaired fibrotic response to injury (11). Consistent with this previous report, we observed senescence marker expression by nonparenchymal cells in both acute and chronic mouse liver injury models. However, after acute injury, the predominant cells expressing senescence markers were hepatocytes. Our observations are consistent with a requirement for chronic or iterative injury to observe persistent populations of nonparenchymal cells with senescence marker expression.

Here, we have investigated TGFβ as a tractable target to interrupt paracrine-induced non–cell-autonomous p21 expression by hepatocytes in the mouse. Our study does not address whether TGFβ inhibition is effective for acute liver injury and failure in man, and further human safety and efficacy studies are required. In addition, study of other SASP components, which may also promote paracrine senescence after liver injury, may be worthwhile (15, 41, 42). In acute liver failure, TGFβ is produced in the injured liver (43). TGFβ tonically inhibits hepatocellular regeneration during health; however, changes in TGFβ ligand and receptor sensitivity facilitate regeneration after partial hepatectomy (44, 45). TGFβ is believed to restrict hepatocellular regeneration rather than just acting as a brake during the termination phase of liver regeneration (1, 46). Clinical oncology trials of TGFβR1 or TGFβ inhibitors in humans are currently underway [for example, NCT02452008 and NCT02581787, respectively; see also (31)]. TGFβR2, by acting as a coreceptor of TGFβR1, may serve as a further potential target for drug development. Long-term therapy using TGFβ inhibition raises potential concerns such as carcinogenesis, autoimmunity, or cardiac valvulopathies (47). However, these concerns may prove to be less relevant for the short periods of therapy (<1 week) required for acute liver failure, a condition in which the prognosis is otherwise grave. A further relevant concern relates to the potential physiological role for TGFβ in mechanically stabilizing the local environment through a fibrotic response during acute liver injury. We did not observe hemorrhagic transformation (microscopic bleeding into the tissue parenchyma) after TGFβR1 inhibition in our treated mice, but further studies that ensure efficacy and safety of TGFβR1 inhibition are required.

SASP components including chemokines (for example, CCL2/CCR2 and CX3CL1/CX3CR1) that promote macrophage recruitment and local TGFβ expression within areas of necrosis are well described in human fulminant hepatic failure (33, 48). Macrophage recruitment in liver injury shares similarities to the p21-dependent recruitment observed during hepatic oncogene-induced senescence (40) and the clearance of early hepatocellular carcinoma (13).

We have shown that severe acute hepatic necrosis induces the spread of senescence to remaining viable hepatocytes, which impairs hepatocyte-mediated regeneration. This process is therapeutically modifiable, thus providing the potential for developing future therapies to treat this devastating condition.


Study design

This study was designed to examine the role of injury-induced senescence in the mammalian liver. With patient consent and ethical approval, we used archival human tissue retrieved as part of routine clinical care. The use of human tissues for this study was approved by the Local Commission for Medical Ethics of the University of Leuven and the University of Edinburgh. Murine in vivo models were used for mechanistic dissection and preclinical compound testing. The n for murine models was based on the predicted variance in the model and was powered to detect 0.05 significance of 30% magnitude; in the event that no predicted variance was inferable from previous work, preliminary experiments were performed using n = 3 mice. Animals were randomly assigned to experimental groups before experimental readings; no animals were excluded from analysis (two mice in Fig. 6G did not receive BrdU). No blinding was performed during experimental administration of treatments to mice; vehicle controls were used, and no bias was applied during husbandry or during tissue harvesting. Histological sections were assigned a randomized blinded code before quantification by a separate researcher, and the randomization was decoded at the time of final data analysis.

Human tissue

Human liver biopsies from a clinical series of cases of submassive hepatic necrosis (but not necessarily progressing to acute liver failure, n = 74; viral hepatitis, n = 13; drug-induced hepatitis, n = 21; and cryptogenic hepatitis, n = 40) were assessed histologically using hematoxylin and eosin, CK19, p16, and Ki67 staining and evaluated by an expert pathologist (T.A.R.), who also performed cellular quantification using ×400 magnification fields. Diagnoses were based on clinical and radiological data and confirmed by histology. Control human tissue was obtained from the Brain Bank, University of Edinburgh, comprising cases of sudden unexpected death. These cases were reviewed by a pathologist before their inclusion as normal control tissue.

Animal models

Animal welfare conditions have been previously described (26). Briefly, male and female animals were housed in a specific pathogen-free environment and kept under standard conditions with a 12-hour day/night cycle and access to food and water ad libitum. Eight-week-old male C57BL/6J mice were purchased from Charles River UK. All animal experiments were carried out under procedural guidelines and severity protocols and within the UK with ethical permission from the Animal Welfare and Ethical Review Body and the Home Office (UK) or in CNIO (Spanish National Cancer Research Centre), Spain, performed according to protocol 193 approved by the Institute of Health “Carlos III” Ethics Committee for Research and Animal Welfare and protocol 194 approved by the Autonomous Community of Madrid. As described previously (26), AhCre+/WT mice were crossed with both Mdm2 f l/f l and Mdm2 f l/+ mice to generate AhCre+ Mdm2 f l/f l and AhCreWT Mdm2 f l/f l and AhCre+ Mdm2 f l/+ controls and then subsequently crossed with p21KO (49) and TGFβR1f l/f l (50) animals. LysMCre mice were crossed with TGFβ1f l/f l animals. Litters from LysMCrehet TGFβ1f l/f l × LysMCreWT TGFβ1f l/f l crosses were used for experimental and control animals. Power calculations were not routinely performed; however, animal numbers were chosen to reflect the expected magnitude of response, taking into account the variability observed in previous experiments. Genotyping, BrdU administration, and intraperitoneal injection of βNF (10 to 80 mg/kg; Sigma-Aldrich) were performed as previously described (26), with BrdU given 2 hours before tissue harvest. AAV8 recombination was performed as previously described (51). Briefly, viral particles [6.4 × 108, 2 × 1011, or 2.5 × 1011 GC per mouse as specified] of AAV8.TBG.PI.Cre.rBG (UPenn Vector Core, catalog number AV-8-PV1091) were injected via tail vein in 100 μl of PBS into male AhCreWT Mdm2f l/f l, LSL-TGFβR1-CAHom (30, 52), or wild-type mice. Control male AhCreWT Mdm2f l/f l or LSL-TGFβR1-CAHom mice received equal AAV8.TBG.PI.Null.bGH (UPenn Vector Core, catalog number AV-8-PV0148) injection. Cell transplantation was performed as previously described (26); AhCre Mdm2f l/f l recipient mice received βNF (10 mg/kg, intraperitoneally) 4 days before cell transplant of 5 × 106 GFP-expressing cells suspended in 200 μl of PBS and injected intrasplenically after laparotomy. Transplanted 7-AADCD31CD45Ter119EpCAM+CD24+CD133+ hepatic progenitor cells from wild-type mice fed the CDE diet were transfected using 1 μg of vector with a puromycin-resistant CAG (cytomegalovirus β-actin β-globulin)–GFP before transplantation. The transplantation control group received 200 μl of PBS only. Recipient mice received intraperitoneal injections of βNF (20 mg/kg) every 10 days after transplantation to induce persistent liver injury. Mice were sacrificed, and the livers were harvested 12 weeks after cell transplantation. HGF (250 μg/kg; R&D Technologies) was administered via tail vein injection. T3 (Sigma-Aldrich) was dissolved in solution (0.01 M NaOH and 0.9 M NaCl) at 0.4 g/liter. This solution was then neutralized with 2 M HCl up to before T3 precipitation and stored at −20°C. T3 (4 mg/kg) was administered to mice via subcutaneous injection. CDE- and DDC-supplemented dietary protocols were as previously described (53). Clodronate liposomes (200 μl) or control PBS (200 μl) was injected intravenously as previously described (53). TGFβR1 antagonists: SB525334 (10 mg/kg; Tocris Bioscience) was given twice daily in 10% polyethylene glycol, 5% dimethyl sulfoxide, and 85% saline vehicle by gavage; AZ12601011 (50 mg/kg; AstraZeneca) (47) was given twice daily in 0.5% (hydroxypropyl)methyl cellulose/0.1% Tween 20 vehicle by gavage. Acetaminophen was prepared as previously described (8) and delivered at 350 or 450 mg/kg by single intraperitoneal injection of 20 μl/g after a 10-hour fast. Acetaminophen (525 mg/kg) was administered by injection of 30 μl/g. CCl4 was delivered by weekly intraperitoneal injection for 8 weeks at 0.75 ml/kg or by single dose at 1 ml/kg 1:3 in corn oil. CCL2 inhibitory antibody (#AF-4679-NA, R&D Systems) was administered (10 μg per injection) daily for 4 days by tail vein injection of a stock (100 μg/ml) diluted in PBS.

Animal tissue harvesting and serum analysis

Mice were sacrificed by CO2 inhalation or cervical dislocation and blood was harvested by cardiac puncture. Organs were harvested and stored in paraffin blocks after fixation in 10% formalin (in PBS) for 18 hours before embedding. Blood hematology was performed using an IDEXX ProCyte Dx analyzer on blood collected in EDTA. Serum analysis used commercial kits according to the manufacturer’s instructions for ALT (Alpha Laboratories Ltd.), microalbumin (Olympus Diagnostics Ltd.), and aspartate aminotransferase and alkaline phosphatase (both from Randox Laboratories).

Immunohistochemistry and in situ hybridization

Three-micrometer-thick paraffin sections were stained for BrdU, p16, HMGA2, γH2Ax, DcR2, and pSMAD3 [AB6326, clone BU1/75; AB54210, clone 2D9A12; AB52039 and AB81299, clone EP854(2)Y; AB108421, clone EPR3588(2); and AB52903, clone EP823Y, respectively; Abcam]; p53 (VP-P956, clone CM5; Vectorn); p21 [clones BMK-2202 (Santa Cruz Biotechnology) and HUGO 291H]; Ki67 (M7249, clone TEC-3; Dako); pSMAD2/3 and pSMAD2 (Cell Signalling #8828, clone D26F4, #3101); CYP2D6 (gift from R. Wolfe, University of Dundee); and the ductular cell marker panCK (Z0622, Dako). Species isotype (Santa Cruz Biotechnology) staining controls were routinely performed. Detection was performed with 3,3′-diaminobenzidine (DAB) (Dako) followed by counterstaining with hematoxylin or, alternatively, with Alexa Fluor 488, Alexa Fluor 555, or Alexa Fluor 650 (A21206, A21434A21436/S32355, and A21448, respectively; Invitrogen) with 4′,6-diamidino-2-phenylindole (DAPI)–containing Vectashield mounting media (Vector Laboratories). Histochemical detection of SA-βGal was performed as previously described (54). In situ mRNA hybridization was performed using RNAscope LS probes for TGFβ1, TGFβR1, CCL2, SMAD7, and PPIB control (407758, 406208, 469608, 429418, and 313918; Advanced Cell Diagnostics) as per the manufacturer’s instructions.

ELISA for murine TGFβ1 ligand was performed using the Mouse/Rat/Porcine/Canine TGFβ1 Quantikine ELISA Kit (R&D Systems) according to the manufacturer’s protocol. Whole liver tissue samples were homogenized in radioimmunoprecipitation buffer (50 mM tris, 150 mM NaCl, 1% Triton X-100, 0.5% deoxycholate, and 0.1% SDS) supplemented with NaF and protease and phosphatase inhibitors and cleared by centrifugation. Protein concentration was determined by BCA assay (Thermo Fisher Scientific #23225). The samples were diluted 1:4. To allow TGFβ activation, 20 μl of 1 M HCl and 20 μl of 1.2 M NaOH/0.5 M Hepes were added to each 100-μl sample. Optical density was measured using a Safire II microplate reader (Tecan) at 450 nm (reference wavelength, 540 nm).

Microscopy and cell counting

Images were obtained on a Zeiss Axiovert 200 microscope using a Zeiss Axiocam MRc camera. Cell counts were performed manually on blinded slides and consecutive nonoverlapping fields at ×200 magnification. Perinecrotic hepatocytes were defined as those contacting the area of necrosis. Confocal image analysis was performed using a Leica SP5 system with the pinhole set to 1 airy unit. DAPI and Alexa Fluor 488 and Alexa Fluor 555 were detected using band paths of 415 to 480 nm, 495 to 540 nm, and 561 to 682 nm for 405-, 488-, 543-nm lasers, respectively. Serial sections were aligned manually in Adobe Photoshop CS5; images were color-deconvoluted using ImageJ using hematoxylin/DAB settings (version 1.5). For RNAscope and quantification of necrosis, slides were scanned on an SCN400F slide scanner (Leica) and the files were analyzed using Halo v2.0 Image Analysis Software (Indica Labs) as previously described. For perinecrotic SMAD7 quantification, perinecrotic and uninjured pericentral areas were manually defined by drawing a ring (200 μm radius from vein or necrosis) around 10 centrilobular structures per sample. Results are expressed as probe copies per area for RNAscope. Necrosis was defined after validation of classifier definition of healthy liver, hemorrhagic necrosis, and nonhemorrhagic necrosis with results expressed as percentage area of necrosis. All scale bars are 50 μm.

Real-time PCR and gene expression analysis

Total RNA was extracted from 30- to 50-mg tissue samples previously stored in RNAlater at −80°C, using a combination of TRIzol reagent (Invitrogen) and Qiagen RNeasy Mini system (Qiagen) according to both manufacturers’ instructions. Genomic DNA decontamination, reverse transcription, and real-time PCR were performed using reagents and primers (QuantiFast and QuantiTect, respectively; Qiagen) on an ABI Prism 7500 cycler, except for chemokine/chemokine receptor analysis, which was performed as previously described (55). Data were collected using the LightCycler system after normalization to the housekeeping gene peptidylprolyl isomerase A (Ppia) or Gapdh for chemokine/chemokine receptor data. All samples were run in triplicate.

RNA-seq analysis

Total RNA was extracted from 30- to 50-mg tissue samples as described above. Purified RNA was tested on an Agilent 2200 TapeStation using RNA screentape. Libraries for cluster generation and DNA sequencing were prepared following an adapted method from Fisher et al. (56) using the Illumina TruSeq Stranded mRNA LT Kit. Quality and quantity of the DNA libraries were assessed on Agilent 2200 Tapestation (D1000 screentape) and Qubit (Thermo Fisher Scientific), respectively. The libraries were run on the Illumina NextSeq 500 using the High Output 75 cycles kit (2 × 36 cycles, paired-end reads, single index). Quality checks on the raw RNA sequencing (RNA-seq) data files were done using fastqc version 0.10.1 and fastq_screen version 0.4.2. RNA-seq reads were aligned to the GRCm38 (57) version of the mouse genome using tophat2 version 2.1.0 (58) with Bowtie version (59). Expression values were determined and statistically analyzed by a combination of HTSeq version 0.5.4p3, the R 3.4.2 environment, using packages from the Bioconductor data analysis suite, and differential gene expression analysis based on the negative binomial distribution using DESeq2 (60). GSEA was performed using the Broad Institute Online Platform. An OIS signature was defined by the top 100 up-regulated genes in the IMR90 ER:RAS model (15).

Statistical analysis

Prism software (GraphPad Software Inc.) was used for all statistical analyses; t tests were used for normally distributed samples (D’Agostino-Pearson omnibus test was used to assess Gaussian distribution) with Welch’s correction if variances differed (F test). One- or two-way ANOVA was used to compare multiple (>2) samples or groups, respectively. Mean number of hepatic progenitor cells per ×200 magnification field from 30 fields for each mouse was compared. Data are presented as mean ± SEM throughout; n refers to biological replicates in all instances unless otherwise stated.


Fig. S1. Senescence markers in human acute liver disease.

Fig. S2. Senescence in the acute CCl4 model.

Fig. S3. Senescence in the acute acetaminophen model.

Fig. S4. Senescence in acute dietary models of liver injury.

Fig. S5. Hepatocyte Mdm2 deletion model.

Fig. S6. Non–cell-autonomous senescence marker induction.

Fig. S7. Hepatocyte TGFβ pathway activation model.

Fig. S8. Hepatocyte TGFβ pathway promotes hepatic TGFβ ligand production.

Fig. S9. TGFβ pathway activity in the acetaminophen model.

Fig. S10. Serial sections of the TGFβ pathway and senescent hepatocytes.

Fig. S11. Macrophage recruitment, TGFβ secretion, and induced senescence.

Fig. S12. TGFβR1 inhibition in acute and chronic CCl4 models.

Fig. S13. Genetic deletion of hepatocyte TGFβR1 in the acetaminophen model.

Fig. S14. Therapeutic TGFβR1 inhibition in the acetaminophen model.

Fig. S15. Schematic representation of paracrine-induced senescence in acute liver injury.

Fig. S16. Ductular reaction responses in murine models and human disease.

Table S1. Source data for Fig. 2.

Table S2. RNA-seq gene: Hallmarks.

Table S3. RNA-seq GSEA: Selected ranked hallmarks and OIS signature.

Table S4. Source data for Fig. 3.

Table S5. Source data for Fig. 4.

Table S6. Source data for Fig. 5.

Table S7. Source data for Fig. 6.


Acknowledgments: We thank G. Lozano (Department of Cancer Genetics, University of Texas) for Mdm2fl mice and A. S. Rocha for constructive input into experimental design. Cl2MDP (or clodronate) was a gift from Roche Diagnostics GmbH. We thank Cancer Research UK (CRUK) Beatson Histological services for their assistance. We acknowledge the use of the GSEA software and Molecular Signature Database (61). Funding: T.G.B. was funded by the Wellcome Trust (WT081604AIA and WT107492Z). S.J.F. was funded by a Medical Research Council (MRC) project grant (G1000868), the Wellcome Trust, the Sir Jules Thorn Charitable Trust, and Scottish Enterprise. J.P.I. was supported by an MRC program grant. O.J.S. was supported by CRUK grant # A12481. U.A. was supported by NIH grant # R01 DK98414. Author contributions: T.G.B., U.A., M.S., T.A.R., J.P.I., O.J.S., and S.J.F. designed the study. O.G., K.J.S., and T.A.R. collected and analyzed human tissue. T.G.B., M.M., R.A.R., T.J., C.K., and S.R.M. performed in vivo experiments. E.L.-G. conducted experiments in p21KO mice. L. Boulter and W.-Y.L. conducted experiments with liposomal clodronate and performed transplantation, respectively. T.G.B., M.M., D.F.V., S.F.-G., A.M.C., T.H., M.C., P.G., C.N., S.B., J.S., and R.J.B.N. performed additional experiments and analysis. A.D.C., A.H., and W.C. performed bioinformatics and statistics. N.V.R., L. Bartholin, and S.T.B. provided resources. T.G.B. wrote the manuscript. T.G.B., M.M., M.S., T.A.R., O.J.S., and S.J.F. reviewed and edited the manuscript. T.G.B., O.J.S., and S.J.F. provided project administration, and T.G.B., A.R.C., U.A., M.S., O.J.S., and S.J.F. raised project funding. Competing interests J.P.I. has consulted for Novartis. M.S. is cofounder and advisor of Senolytic Therapeutics S.L. (Spain) and Senolytic Therapeutics Inc. (USA). All other authors declare that they have no competing interests. Data and materials availability: All of the data associated with this study can be found in the paper or the Supplementary Materials. The data for this study have been deposited in Gene Expression Omnibus at National Center for Biotechnology Information accession number GSE111828. The following materials were provided under materials transfer agreements: AAV8 vector (Penn Vector Core), LSL-TGFβR1-CA mice (INSERM), and AZD12601011 (AstraZeneca).

Stay Connected to Science Translational Medicine

Navigate This Article