Hepatocellular senescence induces multi-organ senescence and dysfunction via TGFβ

Abstract

Cellular senescence is not only associated with ageing but also impacts physiological and pathological processes, such as embryonic development and wound healing. Factors secreted by senescent cells affect their microenvironment and can induce spreading of senescence locally. Acute severe liver disease is associated with hepatocyte senescence and frequently progresses to multi-organ failure. Why the latter occurs is poorly understood. Here we demonstrate senescence development in extrahepatic organs and associated organ dysfunction in response to liver senescence using liver injury models and genetic models of hepatocyte-specific senescence. In patients with severe acute liver failure, we show that the extent of hepatocellular senescence predicts disease outcome, the need for liver transplantation and the occurrence of extrahepatic organ failure. We identify the TGFβ pathway as a critical mediator of systemic spread of senescence and demonstrate that TGFβ inhibition in vivo blocks senescence transmission to other organs, preventing liver senescence induced renal dysfunction. Our results highlight the systemic consequences of organ-specific senescence, which, independent of ageing, contributes to multi-organ dysfunction.

Main

Cellular senescence is a state of permanent cell cycle arrest accompanied by a hyper-secretory phenotype (senescence-associated secretory phenotype, SASP) and is associated with both injury and ageing-related pathologies within affected organs^1,2. Removal of senescent cells is beneficial to both organ function and organism survival^3,4,5,6. Severe acute injury of any large organ is associated with systemic effects including multi-organ failure, of which acute liver failure (ALF) is a paradigm. ALF is itself associated with senescence induction and subsequent regenerative failure⁷. Studies, both in vitro and in vivo, have shown that senescence can be transmitted in a paracrine manner within affected organs^8,9,10,11; however, whether senescence can spread systemically to distant organs remains unknown. Here, we use acute liver senescence as an exemplar model, independent of systemic ageing, to test whether senescence can be transmitted between organs in an endocrine manner. We demonstrate that systemic transmission of senescence affects multiple organs associated with target organ dysfunction and identify the transforming growth factor β (TGFβ) signalling pathway as a critical mediator of this process.

Liver senescence spreads to other organs

To model tissue-restricted senescence we used a genetic mouse model of conditional, hepatocyte-specific excision of the p53-binding domain of MDM2 (Mdm2^E5/E6fl; R26^{LSL-tdTomato/LSL-tdTomato} mice) (Fig. 1a). Intravenous administration of the hepatocyte-specific AAV8-TBG-Cre vector resulted in widespread genetic recombination in the hepatocytes (ΔMdm2^Hep), unlike the empty vector AAV8-TBG-Null (control) (Extended Data Fig. 1a,b). The genetic recombination was hepatocellular specific, with no detectable Cre-mediated recombination occurring in extrahepatic tissues, as previously reported¹² (Extended Data Fig. 1b–f). MDM2 inactivation resulted in p53 accumulation in the hepatocytes and subsequent upregulation of its target gene, the cell cycle inhibitor p21, which is itself a key marker of senescence (Fig. 1b,c and Extended Data Fig. 1a,b). This established a senescent phenotype in the liver manifested by increased senescence-associated β-galactosidase (SA β-Gal) activity and a senescence-associated transcriptome (Extended Data Fig. 2a–c).

Fig. 1: Hepatocellular senescence results in senescence and dysfunction in other organs.

a, A schematic of the experimental approach; 8–12-week-old Mdm2^E5/E6fl; R26^{LSL-tdTomato/LSL-tdTomato} mice were intravenously injected with 2 × 10¹¹ GC of AAV-Cre or AAV-Null and culled 4 days later. The downstream targets of MDM2 are highlighted. b, Representative images of p21 IHC in liver cells of ΔMdm2^Hep and control mice; n = 16/19 control/ΔMdm2^Hep mice, respectively. c, Automated quantification of p21⁺ liver cells; n = 4/5 control/ΔMdm2^Hep mice, respectively; unpaired two-tailed t-test. d, Representative images of p21 IHC on mouse kidney sections. e, Manual quantification of p21⁺ renal tubular cells; the data are presented per field of view (FOV), n = 16/19 control/ΔMdm2^Hep mice (additional controls shown in Extended Data Fig. 1g); Welch’s two-tailed t-test. f, Representative images of p21 IHC on brain sections. g, Manual quantification of p21⁺ brain cells; n = 4/6 control/ΔMdm2^Hep mice, respectively; two-tailed Welch’s t-test. h, Representative images of p21 IHC in lung sections. i, Automated quantification of p21⁺ lung cells; n = 6/9 control/ΔMdm2^Hep mice, respectively; unpaired two-tailed t-test. j, Representative images of p21 IHC in liver sections of Kras^G12D/Kras^WT mice. k, Automated quantification of p21⁺ liver cells; n = 8/11 Kras^WT/Kras^G12D mice, respectively; two-tailed Mann–Whitney test. l, Representative images of p21 IHC on kidney sections of Kras^G12D/Kras^WT mice. m, Manual quantification of p21⁺ renal tubular cells; n = 14/16 Kras^WT and Kras^G12D mice, respectively; two-tailed Mann–Whitney test. n, The plasma levels of cystatin C in ΔMdm2^Hep or control mice 4 days post AAV-Cre or AAV-Null injection; n = 9/10 control/ΔMdm2^Hep mice, respectively; two-tailed Welch’s t-test. o, The urine levels of alanine and threonine in ΔMdm2^Hep mice pre and post AAV-Cre injection; the dots represent the average peak area of n = 3/4/5 mice at days −2 and 0, 4 and −1 and 3, respectively; two-way ANOVA comparing each timepoint to induction day (day 0). p, The proportion of time spent by the ΔMdm2^Hep/control mice in the new arm of the Y maze at day 4; n = 7/9 control/ΔMdm2^Hep mice, respectively; unpaired two-tailed t-test. q, The share of stable versus unstable oscillations of hippocampal brain slices; n = 13/14 brain slices from four control/ΔMdm2^Hep mice each, respectively. All the bars are mean ± s.e.m., each dot represents one sample per mouse and the numbers are P values. The scale bars are 50 μm and 5 μm in the inset magnifications. The source numerical data are available in Source data.

Source data

We next examined other organs for senescence induction in this model. In the kidney, we observed an increase in p21 expression and in SA β-Gal activity, along with a senescent transcriptional signature (Fig. 1d,e and Extended Data Figs. 1g and 2d,e). Additionally, kidneys of ΔMdm2^Hep mice contained increased concentration of several SASP markers (Extended Data Fig. 2f). In addition to the kidney, we observed increased expression of p21 in the brain and the lungs of ΔMdm2^Hep mice (Fig. 1f–i). Therefore, in this model, we observe renal, brain and lung senescence in response to acute genetically induced and hepatocellular restricted senescence.

In ΔMdm2^Hep mice, p53 accumulation in hepatocytes also induces moderate liver injury and dysfunction as evidenced by an increase in plasma levels of alanine aminotransferase (ALT), alkaline phosphatase (ALP) and bilirubin, as well as an increase in hepatic cleaved caspase 3 (CC3), a marker of apoptosis (Extended Data Fig. 2g–j). We also examined a further liver injury and senescence model induced by carbon tetrachloride (CCl₄)⁷, observing here that renal p21 expression and dysfunction also occurred (Extended Data Fig. 2k–n). To address whether systemic transmission of senescence is driven by liver senescence or by liver injury, we induced liver senescence in the absence of injury through oncogene activation. Using hepatocyte-specific expression of oncogenic Kras^G12D, we observed liver senescence without histological or biochemical evidence of liver injury or dysfunction (Fig. 1j,k and Extended Data Fig. 2o–r). Importantly, similar to the ΔMdm2^Hep model, renal p21 expression was increased in mice that expressed KRAS^G12D in their hepatocytes (Kras^G12D mice) compared with the control (Kras^WT) mice (Fig. 1l,m and Extended Data Fig. 2s,t).

When near-global hepatocellular recombination (~80%) of hepatic Mdm2 occurs¹³, the animals become systemically unwell and require to be killed humanely 4 days following induction. To observe the effects of hepatic senescence in a longer-term model, we titrated the hepatic induction through reduction of the AAV8-TBG-Cre vector. When approximately 67% of hepatocytes are induced (so-called recovery-Mdm2 model), a reduction in p21⁺ hepatocytes was observed at day 4 compared with the ΔMdm2^Hep model (Extended Data Fig. 3a–c). In the recovery-Mdm2 model, we tracked renal p21 expression over time, observing a delayed p21 induction, again associated with hepatic injury and senescence. This resolved as the hepatic senescence signature was lost over time (Extended Data Fig. 3b–e). When p53-driven hepatic senescence was further titrated down, the renal signature was lost (Extended Data Fig. 3d), as it was in the Kras^G12D model also. Therefore, we conclude that a critical mass of liver senescence is able to drive the systemic spread of senescence to multiple distant organs.

Transmitted senescence is associated with organ dysfunction

We then proceeded to explore whether extrahepatic senescence was associated with organ dysfunction. To do this, we first measured plasma cystatin C, a marker of renal filtration efficiency, and the levels of urine amino acids as a readout for renal tubular function. Plasma cystatin C and several urinary amino acids were perturbed in the ΔMdm2^Hep mice, consistent with renal dysfunction (Fig. 1n,o and Extended Data Fig. 4a). Urinary amino acid loss was also observed in the recovery-Mdm2 model, temporally consistent with the resolution of the renal senescent phenotype (Extended Data Fig. 3f). To assess brain functionality, we performed cognitive function studies. First, we utilized the Y-maze test, which assesses the natural exploratory instinct of mice. Normal murine behaviour consists of preferential exploration of the novel arm of a Y maze after an acclimatization period before opening the novel arm. ΔMdm2^Hep mice spent less time in the novel arm compared with control mice, consistent with impaired cognitive function (Fig. 1p and Extended Data Fig. 4b). To further corroborate this finding, we performed electrophysiology on hippocampal slices from these mice and measured the gamma (30–80 Hz) oscillation area power and frequency in response to a cholinergic agonist, carbachol. Following carbachol administration, brain sections from healthy mice (control) produced network gamma oscillations of increasing power and stable frequency before stabilizing after 120–150 min, as expected (Extended Data Fig. 4c,d). In comparison, brain sections from ΔMdm2^Hep mice exhibited weaker and less stable oscillations consistent with abnormal hippocampal function (Fig. 1q and Extended Data Fig. 4c,d). Hence, the senescence observed in multiple organs in response to liver senescence is associated with organ-specific dysfunction.

To explore whether these findings are of functional relevance in human disease, we interrogated a cohort of patients with acute liver-specific disease (acute indeterminant hepatitis), each undergoing diagnostic liver biopsy in the early stages of this severe disease (Fig. 2a). Consistent with previous reports, routine biochemistry and clinical parameters at clinical presentation did not define either outcome or subsequent multi-organ dysfunction (including both renal and cerebral failure—hepatic encephalopathy) in this cohort (Supplementary Table 1). We find, however, that elevated levels of hepatocellular senescence markers p21 and γH2A_X on the diagnostic biopsy predict subsequent survival (Fig. 2b,c). Serum creatinine levels, indicative of renal dysfunction, increased over time in patients who did not survive (Fig. 2d), and liver biopsy p21 levels in hepatocytes correlate with both subsequent renal (Fig. 2e) and cerebral dysfunction (Fig. 2f). In contrast to prior clinical scoring systems, which use markers during progressive disease in its later stage to define patient outcome and the need for transplantation, hepatic p21 levels in this cohort predict this on admission with features of severe acute hepatitis.

Fig. 2: Hepatocellular senescence in human severe acute indeterminate hepatitis correlates with development of subsequent renal dysfunction.

a, A schematic of patient stratification and outcomes. b, Multiplex immunofluorescence staining for hepatocytes (HNF4α, red), senescent cells (p21, green), DNA damage (γH2AX, yellow) and nuclei (DAPI, grey) in human patient liver tissue. c, Quantification of p21⁺ and γH2AX⁺ hepatocytes in survivors versus non-survivors. n = 17 patients in both groups; two-tailed Welch’s t-test. d, The serum creatinine levels in survivors versus non-survivors over the course of 28 days post hospital admission. n = 16/14, 13/13, 13/14, 12/12, 10/6 and 9/5 patients on days 0, 2, 3, 7, 14 and 28 for the survivors/non-survivors groups, respectively; two-tailed paired t-test. The bars are the mean ± s.e.m. e, Change in serum creatinine from first (D0) to last recorded (DLast) day of admission; two-tailed paired t-test. n = 16 in both groups. f, Quantification of p21⁺ hepatocytes in patients who developed hepatic encephalopathy (HE) versus those ones who did not. N = 10 and n = 24 in the ‘HE’ and ‘no HE’ groups, respectively; two-tailed Mann–Whitney test. On all violin graphs, each dot represents one biological sample from individual patients, and the numbers are P values. Scale bars, 50 μm. The source numerical data are available in Source data.

SASP factors induce the systemic transmission of senescence

To uncover the changes occurring to the kidney in response to liver senescence, we performed single-cell RNA sequencing (scRNA-seq) on kidneys of ΔMdm2^Hep and control mice (Fig. 3a). From a total of 30,236 single-cell transcriptomes and after stringent quality control (Extended Data Fig. 5a), 24,215 cells were used for downstream analysis. Initial clustering of the control and ΔMdm2^Hep cells together resulted in eight clusters (Extended Data Fig. 5b–e), with each cluster containing cells both from ΔMdm2^Hep and from control mice. Again, there was no evidence of activity of the AAV8-TBG-Cre vector on the genetic reporter in the kidney (Extended Data Fig. 5f).

Fig. 3: scRNA-seq reveals transcriptional changes, including amino acid transporter expression within the proximal tubular compartment in response to liver senescence.

a, A schematic of the scRNA-seq experiment. Three kidneys from control mice and three kidneys from ΔMdm2^Hep mice were dissociated, and droplet-based scRNA-seq was performed on them on a 10x Chromium chip. b, UMAP plots of all 24,215 cells (control and ΔMdm2^Hep). The cells are coloured on the basis of the broad clusters (left) or experimental group (right) (green, control; blue, ΔMdm2^Hep). c, UMAP plots showing the distribution of the cells that have a positive score for the p21 gene signature in the control (top) and the ΔMdm2^Hep (bottom) cells. The inset image is a representative kidney section from a ΔMdm2^Hep mouse stained for p21 by IHC with highlighted tubules (dashed lines). d, Pie charts showing the share of p21^high, TGFβ^high PTCs and JAK–STAT^high PTCs and mesenchymal cells in the control and ΔMdm2^Hep samples. Contingency was tested with the chi-square test (one-tailed). e, A bar chart showing the total number of cells with a positive score for the senescence transcriptional signature in the control and ΔMdm2^Hep samples (chi-square test (one-tailed)). f, A heat map showing the significantly differentially expressed Slc transporter genes in the PTC, DTC and LOH compartments. The gene IDs in the red frames are genes encoding transporters associated with amino acid transport. g, A schematic of the p21^KO experiment. Six Mdm2^E5/E6fl; p21^WT and six Mdm2^E5/E6fl; p21^KO mice were injected with AAV-Cre and culled 4 days later. h, The plasma levels of cystatin C in Mdm2^E5/E6fl; p21^WT and Mdm2^E5/E6fl; p21^KO mice 4 days post AAV injection. n = 6 for both groups; unpaired two-tailed t-test. The bars are the mean ± s.e.m., and the numbers on the graphs are P values. The source numerical data are available in Source data.

By colouring the cells on the uniform manifold approximation and projection (UMAP) plot depending on experimental cohort (control or ΔMdm2^Hep), we observed that, particularly in the broad cluster of proximal tubular cells (PTC), the ΔMdm2^Hep cells deviated from the control cells indicating a shift in their transcriptome (Fig. 3b). To further investigate this, we performed unsupervised pathway analysis in the PTC cluster, comparing ΔMdm2^Hep and control cells. This revealed several pathways associated with senescence (wound healing response and down-regulation of apoptosis) were enriched in the ΔMdm2^Hep PTCs (Extended Data Fig. 6a). Next, to identify the cell identity of the p21-expressing cells we observed in kidney tissue by immunohistochemistry (IHC), we created a p21 gene signature comprised of genes associated with p21 expression. Most cells with a high score for this signature (p21^high cells) belong to the PTC compartment, in agreement with the p21 IHC in kidney sections that showed preferential localization of the p21⁺ cells in the renal tubules (Fig. 3c and Extended Data Fig. 6b). In addition, there was a significant increase of p21^high cells in the ΔMdm2^Hep PTCs identified in the scRNA-seq data (Fig. 3d), consistent with the increase in p21⁺ kidney cells observed in tissue. Additionally, using a second transcriptional senescence signature specifically of renal tubular senescence, developed from a murine renal injury induced senescence and validated in both murine renal injury and ageing and senescent human renal epithelium¹⁴, we also observed an increase in senescent epithelial cells in the kidney following hepatic senescence (Fig. 3e and Extended Data Fig. 6c) but identify these as a generally separate population to those PTC with the p21 gene signature (Extended Data Fig. 6d).

An important function of PTCs is the uptake of filtered substances, including amino acids. As we observed suppression of pathways related to polarity and amino acid transport in the ΔMdm2^Hep PTCs (Extended Data Fig. 6a), we examined the expression of several transporters of the solute carrier (Slc) family across the different parts of the renal tubule. We observed marked differential expression of several genes encoding SLC transporters specifically in the PTCs but not in the distal tubular cells (DTCs) or in the loop of Henle (LOH) (Fig. 3f). Some of these transporters (namely, SLC6A6, SLC7A7, SLC25A39, SLC7A8 and SLC7A13) are involved in amino acid transport in the PTCs^15,16,17, consistent with the increase in amino acids observed in the urine of ΔMdm2^Hep mice. We also observed that in the p21^high PTCs, the expression of several Slc transporter genes was altered compared with PTCs without the p21 signature and specifically in the ΔMdm2^Hep mice (Extended Data Fig. 6e). Therefore, the effects of liver senescence upon the kidney are particularly focused upon the PTCs, resulting in impaired PTC polarity and function.

Next, to explore the p21 dependence of renal senescence and dysfunction, we performed the ΔMdm2^Hep model on a p21-knockout (p21^KO) background (Mdm2^E5/E6fl; Cdkn1a^KO). Here, we observed a lack of reversal in the renal dysfunction observed in ΔMdm2^Hep animals (Fig. 3g,h) implying that secondary organ dysfunction is not dependent upon p21 in either the primary senescent (hepatocyte) or the secondary organs.

To investigate the mechanism of senescence transmission from the liver to the kidney, we hypothesized that SASP factors secreted by the liver reach the kidney via the circulation. This hypothesis was supported by an enrichment of the ΔMdm2^Hep liver transcriptome for a SASP gene signature (Extended Data Fig. 7a). To test this hypothesis, we treated cells in vitro with plasma from ΔMdm2^Hep or control mice (Fig. 4a). We performed this using both wild-type (WT) murine embryonic fibroblasts (MEFs) and neuronal cells differentiated from human neural stem cells (NS cells). In both cases, we observed a notable increase in SA β-Gal activity after treatment with ΔMdm2^Hep plasma compared with plasma from healthy controls (Fig. 4b,c). Next, we performed cytokine arrays on plasma from ΔMdm2^Hep and control mice. While plasma concentration of many of the screened factors remained unchanged, we observed a significant increase in the levels of a number of classical SASP markers: TGFβ1, TGFβ2, TGFβ3, C–C motif chemokine ligand 2 (CCL2), hepatocyte growth factor, angiopoietin and leukaemia inhibitory factor (LIF) in the plasma of ΔMdm2^Hep mice (Supplementary Table 2). We referred to our renal transcriptomics analysis to assess cross-correlation to the downstream pathways (that is, the TGFβ and the JAK–STAT (Janus kinase–signal transducer and activator of transcription) signalling pathways) stimulated by these factors in the kidneys of ΔMdm2^Hep mice, observing elevations in both but, particularly, the TGFβ pathway in the relevant senescent PTC population (Extended Data Fig. 7b–e). The TGFβ pathway was also elevated at the whole kidney level in the transcriptomes of the ΔMdm2^Hep (Extended Data Fig. 7f) and Kras^G12D (Extended Data Fig. 2u) models and has previously been associated both with intra-organ paracrine senescence⁷ and renal dysfunction¹⁸.

Fig. 4: Hepatic SASP in plasma induces senescence.

a, A schematic of the in vitro plasma treatment. WT MEFs and neuronal cells were treated with plasma from control/ΔMdm2^Hep mice and stained for SA β-Gal. b, Manual quantification of SA β-Gal⁺ WT MEFs. Each dot represents one technical replicate. Each of the two biological replicates (one mouse’s plasma sample, with 1/3 technical replicates shown as red/black points, respectively) from both control/ΔMdm2^Hep; n = 3 technical replicates for no plasma control. One-way ANOVA compares all the technical replicates. c, Manual quantification of SA β-Gal⁺ NS cell-derived neuronal cells. Each dot represents the average of the technical triplicates of one biological replicate for each group (n = 6 biological replicates). For the ‘no plasma’ groups, each bar is the mean of four technical replicates (n = 1 biological replicate), one-way ANOVA. d, The plasma levels of TGFβ1, TGFβ2, TGFβ3, CCL2 and LIF. For the TGFβ ligands: n = 9/7, CCL2: n = 9/10, LIF: n = 7/10 control/ΔMdm2^Hep plasma samples, respectively; unpaired two-tailed t-test, Welch’s t-test and Mann–Whitney U test for TGFβ1, TGFβc3 and TGFβ2 and CCL2 and LIF, respectively. Each dot represents one biological replicate (one mouse’s plasma sample). e, Quantification of western blot band signal intensity (Extended Data Fig. 8c) normalized to β-actin; n = 5 each, two-tailed Welch’s t-test/Mann–Whitney for pSMAD2/SMAD2/pSMAD3/SMAD3. f, A schematic of TGFβ inhibitor treatment. g, Plasma-treated neuronal cells were treated with either TGFβR1i (AZ12601011, AstraZeneca) or vehicle (DMSO), stained for SA β-Gal and manually quantified. Each dot represents the average of a technical triplicate from one biological replicate (one mouse’s plasma sample); n = 5 each, one-way ANOVA. h, A schematic of the transfer of conditioned media (CM) in murine slice culture experiments. The livers of Mdm2^E5/E6fl/R26^{LSL-tdTomato/LSL-tdTomato} mice were collected 2 days after AAV-Null/AAV-Cre (n = 6 each). CM was collected 48 h later and added, with TGFβR1i or vehicle, to ex vivo-cultured WT murine kidney slices. i, Manual quantification of kidney slice p21⁺ cells, n = 6 each; unpaired two-tailed t-tests. j, A schematic of a human kidney slice experiment. k, Human kidney slices were treated with TGFβ1 with or without TGFβR1i ex vivo, and p21⁺ renal cells were manually quantified; n = 6 each group, one-way ANOVA. All the bars are the mean ± s.e.m., and the numbers on the graphs are the P values. The source numerical data are available in Source data.

We, therefore, further studied the mechanistic role of the TGFβ pathway in the induction of PTC senescence and dysfunction. We confirmed, by western blotting, that phosphorylated (p)Smad2 and pSmad3 levels (messengers of the canonical TGFβ pathway) were elevated on whole kidney lysates in the ΔMdm2^Hep model (Fig. 4e and Extended Data Fig. 7g). At the single-cell level, we observed that the ΔMdm2^Hep PTCs had a higher score for a transcriptional TGFβ gene signature compared with control PTCs (Fig. 3d) and this was validated in tissue with in situ hybridization for Smad7, a transcriptional target of TGFβ signalling being overexpressed in senescent PTCs (Extended Data Fig. 7b,c). The increase in plasma TGFβ ligands is associated with their increased hepatic production, as evidenced by their transcriptional upregulation in this organ (Extended Data Figs. 2c and 7h), including by a subset of senescent hepatocytes (Extended Data Fig. 7i) and, particularly, the non-parenchymal cells (Extended Data Fig. 7h). This also corresponded with increased expression of the TGFβ receptor (Tgfbr1-3), both in the liver and in the whole kidney, and by renal PTCs, specifically (Extended Data Fig. 7j–l).

We next performed functional studies using plasma and media transfer experiments to test the role of TGFβ for transmission of senescence from plasma to neurons and the kidney. Initially, we tested this in vitro in plasma-treated neuronal cells and observed a significant reduction in the number of SA β-Gal⁺ cells after TGFβR1 inhibition (Fig. 4f,g). Next, using an ex vivo precision-cut liver and kidney slice (PCLS and PCKS) system, we tested if the circulating plasma changes came directly from the senescent liver (Fig. 4h). In this setting, transferring media from senescent PCLS to the PCKS induced renal senescence (Fig. 4i). We further tested TGFβ receptor inhibition, showing p21 induction was inhibited using the TGFβR1 inhibitor AZ12601011 (ref. ¹⁹). Next, for human relevance, we treated healthy human PCKS with the TGFβ ligand, inducing p21 expression, an effect also suppressed by TGFβR inhibition (Fig. 4j,k).

TGFβ inhibition prevents systemic transmission of senescence

Having associated TGFβ pathway activation with senescence transmission in vitro, we went on to functionally test this pathway’s role in senescence and dysfunction induction/transmission in vivo. Using the TGFβR1 inhibitor in the ΔMdm2^Hep model (Fig. 5a) resulted in inhibition of the TGFβ pathway in the kidney, as expected (Fig. 5b and Extended Data Fig. 8a) and a reduction p21⁺ cells in the kidney, brain and lung of these mice (Fig. 5c–f) without affecting the genetically induced liver senescence (Extended Data Fig. 8b,c). A similar reduction in renal p21⁺ cells was also observed in the Kras^G12D model following TGFβR1 inhibitor (TGFβR1i) treatment (Extended Data Fig. 8d,e). The improvement of the senescence phenotype was accompanied by an improvement in renal function, as indicated by a significant reduction of plasma cystatin C and urine amino acids (Fig. 5g–i and Extended Data Fig. 8f). We tested the effects of senomorphics and senolytics in the ΔMdm2^Hep model. Using rapamycin as a senomorphic did not affect genetically driven p21 overexpression in the liver but ameliorated p21 expression in both the kidney and lung (Extended Data Fig. 8g–i). Using senolytics (either oubaine or navitoclax) resulted routinely in progressive liver damage and adverse mouse outcome. We tested the requirement for renal Tgfbr1 expression for the renal PTC p21 phenotype using the CCl₄ model. Genetic Tgfbr1 deletion in the renal²⁰ and/or hepatic epithelia showed that renal TGFβR1 epithelial expression is required for the renal PTC p21 phenotype resulting from liver disease (Fig. 5j–l). We, therefore, conclude that inhibition of the TGFβ signalling pathway prevents the systemic transmission of senescence to and dysfunction in the kidney.

Fig. 5: Inhibition of TGFβ signalling prevents induction of senescence in the extrahepatic organs.

a, A schematic of the in vivo treatment with TGFβ receptor inhibitor (TGFβR1i). ΔMdm2^Hep mice were treated with TGFβR1i or vehicle by oral gavage twice daily, starting 24 h post AAV-Cre. b, A quantification of western blot band signal intensity (Extended Data Fig. 8a); n = 5 each group, unpaired two-tailed t-tests. c, Representative images of p21 IHC in kidney, brain and lung sections of ΔMdm2^Hep mice treated with vehicle or TGFβR1i (n = 3–13 mice for each group as stated below). d, A manual quantification of p21⁺ renal cortical cells; the data are presented as mean p21⁺ cells per field of view (FOV) for each mouse; n = 13 (biological replicates) mice for each group, two-tailed Mann–Whitney test. e, a Manual quantification of p21⁺ brain cells; the data are presented as p21⁺ cells per square millimetre, n = 3/5 vehicle/TGFβR1i treated, respectively; two-tailed Welch’s t-test. f, Automated quantification of p21⁺ lung cells; n = 7/8 vehicle/TGFβR1i treated, respectively; two-tailed Mann–Whitney test. g, Plasma cystatin C of vehicle or TGFβR1i treated at cull. Each dot represents the data from one mouse; n = 7 mice for each group, unpaired two-tailed t-test. h,i, The levels (peak area ratio to creatinine) of alanine (h) and threonine (i) in the urine of ΔMdm2^Hep mice before and after AAV-Cre injection; n = 4/5 vehicle/TGFβR1i treated, respectively, per timepoint, two-tailed Welch’s t-test comparing vehicle- or TGFβR1i-treated mice at day 4. j, A schematic of the genetic ablation of TGFβ signalling in the kidney. TGFBR1(Alk5)^fl/fl ± AhCre^+/− mice were administered either 2 × 10¹¹ GC ml⁻¹AAV-Cre (ΔAlk5^Liver) or 3 x 80 mg kg⁻¹β-NF (ΔAlk5^Kidney), respectively, or control induction (2 × 10¹¹ GC ml⁻¹AAV-TBG-Null/vehicle, respectively) and were collected 4 days later. k,l, Automated quantification of p21⁺ liver cells (k) and manual quantification of p21⁺ renal cortical cells (l) in ΔAlk5^Liver and ΔAlk5^Kidney mice; the data are presented as percentage of total liver cells or as p21⁺ cells per FOV. n = 6/8 for ΔAlk5^Liver/ΔAlk5^Kidney mice, respectively; Brown–Forsythe and Welch ANOVA test (liver) or one-way ANOVA (kidney). The bars are the mean ± s.e.m., and the numbers on the graphs are P values. The scale bars are 50 μm and 5 μm in the inset magnifications. The source numerical data are available in Source data.

Discussion

The SASP is a central mediator of the non-autonomous effects of senescent cells. Here, we demonstrate that senescence can be transmitted to and affect the function of distant organs in a systemic manner. In the context of acute injury, senescence has often been described as part of a finely tuned mechanism with overall beneficial effects for wound healing^21,22. SASP factors have been shown to induce reprogramming in neighbouring cells, facilitating tissue regeneration^23,24,25. However, following severe injury, this mechanism may have the opposite effect, systemically, through excessive SASP production, including senescence itself. In turn, this excessive stimulus for senescence can be associated with compromised organ function.

Systemic transmission of senescence may be relevant to several diseases. Here, we use a series of models of hepatocyte-specific senescence to model an acute senescence phenotype, such as the one observed during ALF. ALF is, itself, characterized by sequential multi-organ failure, typically beginning with the kidney progressing to also involve the brain, lungs and other organs. This clinical progression may, at least in part, be underpinned by the systemic transmission of senescence. The data in patients with acute indeterminate hepatitis, showing that increased hepatocellular expression of p21 at initial presentation before multi-organ failure can predict ensuing multi-organ failure, requirement for liver transplant and/or death, provide evidence for a biomarker that both allows early risk stratification and selection of patients for specific therapies. Similarly, the observation that TGFβ signalling is a central driver of systemic transmission of senescence may pave the way for therapeutic approaches in diseases where this phenomenon occurs. Whilst this effect may either be independent of solely p21-dependent senescence or a phenomenon related to TGFβ activity outside of senescence, it is in line with the beneficial effects of senolytics and senomorphics that have been elegantly demonstrated on numerous pathologies^5,26,27,28. Further research is required to dissect out the direct causal link between senescence in the primary tissue and the systemic effects and how they are affected by factors such as disease site or specific senescence phenotype, chronicity of senescence and the interaction with concurrent or pre-existing senescence in other organs. Our results demonstrate that systemic transmission of senescence can induce systemic organ dysfunction, which may be central to multi-systemic sequelae in many diseases.

Methods

Ethics

This study was conducted in accordance with all relevant ethical regulations. The clinical study was approved by the London-Hampstead Research Ethics Committee (07/Q0501/50) and was in accordance with the declaration of Helsinki as reported before²⁹. This study reported multi-lobular necrosis as a predominant histopathological feature being significantly more frequent in non-survivors²⁹. All mouse experiments were carried out in accordance with the UK Home Office regulations (licence 70/8891; protocol 2 or licence PP0604995; protocols 3 and 4 or licence P3F79C606, protocol 2) following ethical approval by the local Animal Welfare and Ethical Review Body at the University of Glasgow.

Patient selection and data collection

The study included 34 consecutive patients with severe acute indeterminate hepatitis who were admitted to a single hospital and underwent transjugular liver biopsy or liver transplantation. The participants were not compensated for participation. Retrospective anonymized data and samples were obtained, following ethical approval and consistent with the UK Human Tissue Act, without informed consent. Of the 34 patients, 14 underwent liver transplantation, and 3 died within 3 months. They were defined as non-survivors. Seventeen patients who recovered spontaneously were defined as survivors. All 34 liver tissue biopsies analysed in this study were obtained at the time of enrolment (baseline biopsies). The following clinical data were collected from the patients: sex, age, date of histopathological examination, history of chronic disease and results of biochemical tests at baseline and at follow-up up to 28 days or the last value before death. These include serum levels of ALT, aspartate transaminase, bilirubin, ALP, international normalized ratio, creatinine, prothrombin time, albumin and hepatic encephalopathy.

Animal studies

Mice were bred and housed in a licensed, pathogen-free facility, under a 12 h light–dark cycle, at stable temperature (19–23 ^oC) and humidity (55 ± 10%). The mice were bred on a mixed C57Bl6 background, were housed in conventional cages and had ad libitum access to food and water (standard CRM(E) chow; Special Diets Services no. 801730). All experiments were performed on 8–12-week-old male and female mice, according to the guidelines of the Animal Welfare and Ethical Review Body, and are reported in agreement with the ARRIVE guidelines³⁰.

Treatment with AAV was performed as described previously¹². Briefly, mice were injected with either AAV8.TBG.PI.Cre.rBG (AAV-Cre) (Addgene, 107787-AAV8) or AAV8.TBG.PI.Null.bGH (AAV-Null) (Addgene, 105536-AAV8) at a dose of 2 × 10¹¹ genomic copies (GC) ml⁻¹unless otherwise stated, in a final volume of 100 μl sterile PBS via a single tail vein injection. Mice in this study weighed 21.9–31.9 g at the time of induction.

For the mouse model of hepatocyte-specific inactivation of Mdm2, male mice homozygous for the Mdm2tm2.1Glo allele (ID: MGI2385439176 (ref. ³¹)) and the Gt(ROSA)26Sortm14(CAG-tdTomato)Hze allele (ID: MGI3809524177 (ref. ³²)) were used (Mdm2^E5/E6fl; R26^{LSL-tdTomato/LSL-tdTomato} mice). This model was also crossed to create a p21^KO model (allele ID: MGI1888950 (ref. ³³)). For the induction of a RAS-induced liver senescence, 8–12-week-old male and female mice heterozygous for the Kras^tm4Ty allele (ID: MGI:2429948178 (ref. ³⁴)) and the Gt(ROSA)26Sortm14(CAG-tdTomato)Hze allele (ID: MGI3809524177 (ref. ³²)) were used (Kras^LSL-D12D/WT; R26^{LSL-tdTomato/LSL-tdTomato} mice). Epithelial TGFBR1 deletion experiments used 1 μl g⁻¹of CCl₄ diluted 1:4 CCl₄ in corn oil administered day 3 in mice with alleles AhCre and TGFBR1^fl (IDs: MGI3052655 (ref. ³⁵) and MGI2388050 (ref. ³⁶), respectively). Mice with AhCre^+/− TGFBR1^fl/fl were used for epithelial induction utilizing β-naphthoflavone (β-NF; Sigma) versus β-NF-treated AhCre^−/− Tgfbr1^fl/fl controls, while AhCre^−/− TGFBR1^fl/fl mice were treated with either AAV-TBG-Cre or AAV-Cre-Null as described above at 2 × 10¹¹ GC per mouse. Genetic controls used both AhCre^−/− Tgbfr1^fl/fl controls given respective controls (β-NF vehicle or AAV-Cre-Null) and AhCre^−/− TGFBR1^fl/fl mice given no induction agent but time matched CCl₄ at day 3. The acute CCl₄ model used 1 μl g⁻¹of 1:4 CCl₄ in corn oil administration as previously described⁷. For treatment with TGFβR1i, the mice received either TGFβR1i or vehicle (0.5% hydroxypropyl methylcellulose (HPMC) and 0.1% Tween 80) by oral gavage twice daily¹⁹. The dose of TGFβR1i was 50 mg kg⁻¹in 100 ml PBS⁷. Senolytic and senomorphic agents were administered separately; navitoclax (ABT263), oubaine and rapamycin (or vehicle controls) were administered at 100 mg kg⁻¹(gavage twice weekly), 1 mg kg⁻¹(intraperitoneal injection (i.p.) on days 1 and 3 or twice weekly) and 250 μg (i.p. daily), respectively, were given from days 1 to 7 in the ΔMdm2^Hep model and day 7 in the Recovery-Mdm2 model. Rapamycin mice in the ΔMdm2^Hep model were collected at day 4. All mice receiving either ABT263 or oubaine either deteriorated clinically and were humanely killed or died shortly (within 6 h) after receiving the senolytics. β-NF was given by 3 × 80 mg i.p. injections for epithelial induction in mice with the AhCre allele as previously described³⁷.

The mice were killed by CO₂ inhalation in a CO₂ chamber, cervically dislocated and then weighed. Blood was collected immediately via cardiac puncture for whole blood analysis (EDTA buffer-coated tubes; Sarstedt) and plasma biochemistry (lithium–heparin coated tubes; Sarstedt). The plasma was separated by centrifugation (2,350gfor 10 min at room temperature) within 1–3 h after culling and stored immediately at −80 ^oC. After weighing the liver, the caudate and left median lobes lobe were snap frozen on dry ice for protein and RNA extraction and for histology studies, respectively. The remaining liver was fixed in 10% neutral buffered formalin (NBF) for 22–24 h before transfer to 70% ethanol for further processing. The left kidney was immediately cut in half, and both halves were snap frozen on dry ice for protein and RNA extraction and for histology. The right kidney, heart, brain and lungs were fixed in 10% NBF for 22–24 h and then changed to 70% ethanol.

Assessment of cognitive function

Y-maze test

The mice were individually placed into a testing area measuring 25.66 cm × 17.53 cm onto a Samsung Galaxy Tab 2 10.1 for 5 min. Steps and distance travelled were recorded using MouseTrapp software. The Y-maze arena had three arms of 40 cm identified as A1 A2 A3, each with a differentiating marker at the end of the arm. The mice were assigned different start arms in a rotating allocation and were tested before the start of the experiment and on day 4, with differing arm allocations each time. In the T1 phase, the mice were placed into the maze with only two arms open for 5 min to explore the arena. The mice were then removed to a clean cage for 1 min (fresh cage used per cage of mice). Then, in the T2 phase, the mice were returned to the maze in the starting position with the novel arm opened for 2 min and allowed to explore. A camera was set up above the maze to record movements and the video files analysed via Ethovision XT13 software.

Brain slice electrophysiology

The animals were humanely killed by anaesthetic overdose with inhaled isoflurane and intramuscular injection of ketamine (≥100 mg kg⁻¹) and xylazine (≥10 mg kg⁻¹) as previously described³⁸. The mice were then transcardially perfused with at least 25 ml of sucrose-rich artificial cerebrospinal fluid—composed of 252 mM sucrose, 3.0 mM KCl, 1.25 mM NaH₂PO₄, 24 mM NaHCO₃, 2.0 mM MgSO₄, 2.0 mM CaCl₂ and 10 mM glucose. The brain was removed and sliced at 450 μm horizontal slices with a Leica VT1000S vibratome in ice-cold sucrose-rich artificial cerebrospinal fluid. The slices were trimmed to the hippocampal region and maintained at 32–34 ^oC at an air–liquid interface between normal artificial cerebrospinal fluid (sucrose replaced with 126 mM NaCl) and humidified 95% O₂/5% CO₂. Oscillations were evoked with 10 μM cholinergic agonist carbachol, to activate transmission through acetylcholine receptors. Extracellular recording electrodes were filled with normal artificial cerebrospinal fluid (resistance 2–5 MΩ), and field recordings taken from the border between stratum radiatum and stratum lacunosum moleculare in CA3. The recordings were taken with an Axoclamp-2B amplifier (Axon Instruments) and extracellular data filtered at 0.001–0.4 kHz using Bessel filters. Mains noise was deducted with a Humbug (Digitimer) and data redigitized at 10 kHz using an ITC-16 interface (Digitimer). Axograph 4.6 software (Axon Instruments) was used for data acquisition and analysis.

To generate power spectra Axographs we used Fourier analysis using 60 s per 10 min recording. This was used to calculate peak frequency and area power (area under the peak). The mouse gamma frequency oscillation was measured at frequencies between 15 and 49 Hz. The oscillations were categorized as stable when area power measured within 10% for three consecutive 10 min recording intervals.

Histology

Murine samples

Formalin-fixed, paraffin-embedded sections 4 μm thick were used for simple IHC and for multiplex immunofluorescence. The sections were subjected to heat-induced antigen retrieval, followed by protein blocking to reduce non-specific staining. Incubation with primary antibody overnight at 4 ^oC or for 1 h at room temperature was followed by secondary antibody incubation and signal detection. The details on the antibodies can be found in Supplementary Table 3.

Photos were taken with a Zeiss Axiovert 200 microscope using a Zeiss Axiocam MRc camera. The stained slides were scanned using a Leica Aperio AT2 slide scanner (Leica Microsystems) at 20× magnification. Automated quantification of positively stained cells or area was performed using the HALO image analysis software (V3.1.1076.363, Indica Labs). Manual quantification of p21⁺ and BrdU⁺ kidney cells was performed on 20 random fields at 20× magnification. For p21 IHC quantification on brain sections, total brain area was calculated using the HALO software, and the p21⁺ cells were manually counted in the whole brain tissue area.

For multiplex immunofluorescence, 4 μm tissue sections underwent heat-induced antigen retrieval by boiling (in a waterbath) in citrate buffer (10 mM Na Citrate (Sigma, W302600), 0.05% Tween 20 (Sigma, P1379), pH 6) for 25 min and were subsequently cooled down for 20 min in the retrieval solution. Peroxidase quenching with 3% H₂O₂ (Sigma, 95321) was followed by biotin (Invitrogen, 4303) and protein blocking (Abcam, ab64226). The sections were incubated with the primary antibodies either for 1 h at room temperature or overnight at 4 ^oC, followed by 45 min with the secondary antibodies (conjugated to fluorophor) together with 4,6-diamidino-2-phenylindole (DAPI, 1 mg ml⁻¹). Sudan black B was then used to quench autofluorescence. An aqueous mounting solution (DAKO, S1964) was used for mounting.

The anti-p21 antibody required additional signal amplification, which was achieved by using the tyramide signal amplification system. Briefly, after incubation with the anti-p21 primary antibody, the sections were incubated with a secondary anti-rat biotinylated antibody for 30 min, followed by a 30 min incubation with an avidin–HRP (horse radish peroxidase) complex (Vectastain ABC, Vector, PK-7100). Then, the sections were incubated with tyramide signal amplification fluorescein isothiocyanate (FITC; PerkinElmer, NEL741B001KT) for 6 min (in the dark). After that, the sections that were subjected to a 5 min heat-induced antigen retrieval to remove the anti-p21 antibody complex underwent protein blocking and then were incubated with the other primary antibodies, as described above. The images were taken using a Zeiss 710 upright confocal Z6008 microscope. The Opera Phenix scanner (PerkinElmer) was used to scan the stained sections at 20× magnification. For the analysis of scanned sections, the Columbus software (PerkinElmer) was used to identify hepatocytes and to quantify immunofluorescence staining intensity by hepatocyte in 20 random fields of view.

In situ hybridization was performed in an autostainer (Leica Bond R_x) using the 2.5 LSx RNAScope kit (Bio-Techne, 322700) according to the manufacturer’s instructions. The probes against Smad7 messenger RNA (Bio-Techne, 429418), TGFβ1 (Bio-Techne, 407758), TGFβ2 (Bio-Techne, 406188) and TGFβR1 (Bio-Techne, 431048) were used for the detection of the respective mRNA, and PPIB (Bio-Techne, 313918) was used as a positive control of gene expression.

Patient samples

Multiplex immunofluorescence staining was performed as previously described^29,39,40. Formalin-fixed, paraffin-embedded liver samples were deparaffinized and rehydrated in xylene (Roth) and ethanol (Roth). Antigen retrieval was performed with Tris–EDTA buffer (pH 9) or universal antigen retrieval (Abcam) in a water bath at 98 °C for 30 min, followed by a cooling period of 20–30 minutes. The tissues were blocked with 2% normal goat serum (Thermo Fisher Scientific) to prevent non-specific antibody binding. The slides were incubated overnight at 4 °C with primary antibodies diluted in antibody dilution solution (Life Technology) and stained for 30 min with fluorescently labelled secondary antibodies (Supplementary Table 3) together with DAPI nuclear counterstain (Sigma Aldrich). After scanning the entire slide with a Zeiss Axio Observer7, the images were merged, and the background was subtracted. After each run, the antibodies were stripped by using the 2-mercaptoethanol/SDS method³⁹, and the staining was repeated in multiple cycles over an 3 day period. Subsequently, all scans were aligned, hyperstacked and concatenated using the plugin FIJI HyperStackReg V5.6. For binary images, cell segmentation was performed using Ilastik software (v 1.3.3). Cell identification and counting, as well as fluorescence intensity measurement, were performed using CellProfiler v3.1.9 and plugin FIJI.

SA β-Gal staining

Staining for SA β-Gal was performed as described previously⁴¹. Briefly, 10-mm-thick cryosections or cultured cells were fixed in 2% paraformaldehyde or 0.25% glutaraldehyde in PBS for 15 or 5 min, respectively. This was followed by three washes with PBS 1 mM MgCl₂ (pH 5.5 or 6 for mouse or human cells and tissues, respectively) and incubation with staining solution (1 mM MgCl₂, 0.5 mM K₃Fe(CN)₆ 0.5 mM C₆FeK₄N₆·3H₂O and 1 mg ml⁻¹X-Gal in PBS, pH 5.5 or 6 for mouse or human cells and tissues, respectively) overnight (liver sections and cells) or for 2.5 h (kidney sections). After three washes with PBS, the cryosections were counterstained with eosin and mounted, while cells (on coverslips) were mounted immediately. For quantification, SA β-Gal⁺ and SA β-Gal⁻ cultured cells were counted in 20 random fields of view.

RNA extraction

Whole-tissue RNA was extracted using the Qiagen RNeasy kit (Qiagen, 74104), including the optional DNase step, as described previously¹². Briefly, 20–30 mg of snap-frozen tissue were homogenized in 600 ml buffer RLT and 1% β-mercaptoethanol using the Precellys Evolution homogenizer (Bertin Technologies), and the RNA was eluted in 30 μl RNase-free water. The RNA concentration was estimated with the Nanodrop 2000m and only samples with a 260/280 ratio ≥2 were used for further analysis.

Complementary DNA generation and qPCR

cDNA was generated using the QuantiTect Reverse Transcription Kit (Qiagen, 205311) according to the manufacturer’s instructions from 1 μg RNA. A PTC-200 Gradient cycler (MJ Research) was used to perform the genomic DNA wipeout and reverse transcription steps. A sample-free reaction and a reaction without the reverse transcriptase served as the negative controls. A real-time quantitative polymerase chain reaction (qPCR) was performed with the SYBR Green system (Qiagen, 204145) using a QuantStudio 5 real-time polymerase chain reaction system in a 384-well-plate setting (final reaction volume 10 μl per well). All primers used were purchased from Qiagen, as shown in Supplementary Table 4. Each biological replicate (mouse) was run in triplicate, and the 18S ribosomal RNA (Rn18S) was used as a house keeping gene for normalization. Relative expression was calculated using the ΔΔCt method.

Whole-tissue (bulk) RNA-seq

For bulk RNA sequencing (RNA-seq), the RNA was extracted as described above. Briefly, the RNA was tested on an Agilent 2200 TapeStation (D1000 ScreenTape) using RNA ScreenTape and only samples with a RIN >7 were further processed for library preparation. A total of 20 ng ml⁻¹of RNA were used to prepare libraries using the TruSeq Stranded mRNA Kit. The Agilent 2200 Tapestation was used to assess library quality and Qubit (Thermo Fisher Scientific) was used to check concentration. The libraries were then run on the Illumina NextSeq 500 using the High Output 75 cycles kit (paired end, 2 × 36 bp cycle, dual index (I5 and I7 Illumina)).

For the bioinformatics analysis, raw data quality checks and trimming were performed using FastQC (versions 0.11.7 and 0.11.9 for Mdm2 and Kras, respectively), FastP (v0.19.3/0.20.1) and FastQ Screen (v0.12.0/0.14.0). The reads were aligned to the mouse genome and annotation (GRCm38.92 version) using HiSat2 version 2.1.0183. Determination and statistical analysis of expression levels was done by a combination of HTSeq version 0.9.1184, the R environment (v3.4.4/4.1.2), utilizing packages from Bioconductor (v3.6/3.15) and differential gene expression analysis based on the negative binomial distribution using the DESeq2 package (v1.18.1186). Pathway analysis was performed using MetaCore (v2018/2022) from Clarivate Analytics (https://portal.genego.com/).

scRNA-seq on kidneys

Three ΔMdm2^Hep and three control mice were culled by CO₂ inhalation; the blood samples were collected by cardiac puncture, and 30–40 ml cold PBS were used to perfuse the circulatory system via the left heart. The six mice were culled and their kidneys processed on two different days. Two mice (one ΔMdm2^Hep and one control) were culled together on one day, and the others (two ΔMdm2^Hep and two control) were culled 2 months later. On each occasion, the left kidney was used for dissociation and generation of single-cell suspension, while the right kidney was fixed in 10% NBF. The renal capsule of the left kidney was removed, and the kidney was cut into equally sized pieces and was dissociated using a multi-tissue dissociation kit 1 (Miltenyi, 130-110-201) as per the manufacturer’s instructions. A total of 0.25 g of kidney tissue were placed in a GentleMACS C tube with dissociation buffer (2.35 ml serum-free RPMI (Roswell Park Memorial Institute) culture medium (Gibco), 100 μl enzyme D, 50 μl enzyme R and 12.5 μl enzyme A). Kidney dissociation was performed in a GentleMACS dissociator using the ‘heart_01_01’ programme (15 s). The samples were placed in a waterbath (37 ^oC) for 30 minutes and then placed back in the dissociator (‘heart_01_02’ programme, 30 s).

After the second round of dissociation, 8 ml of sterile PBS and 10% FBS were added in the C tubes to stop the reaction. The samples were passed through a 40 μm cell strainer into a 50 ml falcon tube. All subsequent steps were performed on wet ice or at 4 ^oC. The samples were spun at 300g for 5 min and resuspended in 5 ml of cold PBS. They were spun again at 300g for 5 min, resuspended in 1 ml red blood cell lysis buffer (8.29 g NH₄Cl, 1 g KHCO₃ and 37.2 g Na₂EDTA in PBS) and incubated for 30 s on wet ice. The samples were topped with 9 ml of cold PBS and washed twice, as described previously (spin at 300g for 5 min and resuspended in PBS). The samples were resuspended in cold PBS and were subjected to debris removal using the debris removal solution (Miltenyi) according to the manufacturer’s instructions. Finally, the samples were resuspended in 10 ml cold PBS, 10% FBS and 2.5 mM EDTA.

Cell viability and concentration were determined using the Trypan Blue assay and flow cytometry (viability ≥90%). A total of 20,000–40,000 cells were loaded on a 10x Chromium chip (one sample per lane). Cleanup, reverse transcription, cDNA amplification and library preparation were performed using the Chromium Single Cell 3′ Reagent Kits (v3), as per the manufacturer’s instructions.

The samples were sequenced in the Illumina NextSeq 500 using the 2 × 150 bp kit with the following read length parameters: 26 bp Read1, cell barcode and Unique Molecular Identifier (UMI); 8 bp I7 index, sample index; and 98 bp Read2, transcript read. CellRanger v.4.0 (with default parameters) was used to demultiplex Illumina BCL output files and align reads to the ensemble GRCm38.99 reference genome with the addition of the AAV8-TBG-Cre and AAV8-TBG-Null sequences.

All other steps were performed using R v.4.0 and packages from Bioconductor v.3.12. The raw matrices generated by CellRanger (v.4.0) were transformed to SingleCellExperiment objects and they were filtered to exclude empty droplets using the DropletUtils v.1.10 package. SingleCellExperiment objects were merged together to perform further downstream analysis. Following similar parameters used in a previous scRNA-seq analysis murine kidney⁴², cells with <75 or >3,000 expressed genes or >50% mitochondrial gene expression were filtered out, and only genes expressed in more than ten cells with at least one UMI were kept for further analysis.

The normalization by deconvolution method designed by Lun et al.⁴³ was used to normalize and log transform the counts with functions from Scater package v.1.18. Highly variable genes were computed with default parameters, and the top 10% were used to perform a principal component analysis and UMAP using the top 20 PCs. After examining UMAP plots coloured by batch run, it was determined that batch correction was not required.

Initial clustering was performed with functions from the Seurat package v.4.0 with default parameters and with eps value of 0.5 and resolution sequence of 0.1 to 1 by 0.1. Markers of each cluster were identified by performing a pairwise differential analysis between each pair of clusters with a minimum difference of 20% of cells expressing the gene and log₂ fold change threshold of >0.25 and only keeping the differentially expressed markers in all the comparisons. Reclustering of control cells with a new principal component analysis and UMAP was performed, and the main markers were used to manually classify clusters into different cell types. ΔMdm2^Hep cells were projected onto the reference UMAP and assigned to the identified clusters. The same methodology was followed for the identified control PTCs, DTCs and the initial cluster ‘mesenchymal cells’ for the projection and assignment of the ΔMdm2^Hep cells. Scores for senescence, p21, proliferation, TGFβ, JAK–STAT and regeneration signatures were computed using the AddModuleScore Seurat function. The genes included in each list used to create the signatures can be found in Supplementary Table 5. All transcriptomic data will be made publicly available on the Gene Expression Omnibus (GSE189726) repository at the time of publication. Alternatively, cells expressing senescent features were identified by using an unbiased cell type classifier ‘singleR’, trained on a reference transcriptomic dataset of previously validated murine renal scRNA-seq data¹⁴. The hyperparameters used were threshold of 0.4, 150 differentially expressed (DE) genes, quantile of 0.6 and fine-tune parameter set to true.

Differential gene expression (DGE) analysis was performed as described by Giustacchini et al.⁴⁴. Briefly, the log₂ fold change was calculated between groups, and a non-parametric Wilcoxon test was used to compare the expression values. Fisher’s exact test was used to compare expressing cell frequency (percentage of cells per group with at least one UMI count). The Pvalues from both tests were combined using Fisher’s method and adjusted using Benjamini–Hochberg. The genes were considered to be differentially expressed if the P adjusted values were <0.05. The heat maps comparing relative gene expression by cell groups were computed using the ‘DoHeatmap’ function from Seurat R package or ‘plotGroupedHeatmap’ function from the scater R package. In both cases, the scaled value from each group was calculated by substracting the average logcount gene expression from the total mean gene expression and divided by the standard deviation. A two-colour range scale was used to convert scale values into colour intensity. Gene Set Enrichment Analysis was performed in an unsupervised manner using the ‘gseGO’ function from the ClusterProfiler R package; enrichment scores and P values were calculated by the function using an empirical phenotype-based permutation test.

Protein extraction and western blotting

Snap-frozen tissue was homogenized in 300 μl of protein extraction buffer (50 mM HEPES, 100 mM NaF, 150 mM NaCl, 10 mM Na₄P₂0₇, 10 mM EDTA, 1% Triton X100, 0.1% SDS and 0.5% Na deoxycholate in ddH2O), 1:100 protease inhibitor (Thermo Fisher, 1862209) and 1:100 phosphatase inhibitor (Thermo Fisher, 1862495) in ddH₂O using the Precellys Evolution homogenizer (Bertin technologies). The lysate was passed through a 25G needle five to eight times and was then spun twice at 20,800gfor 10 min at 4 ^oC. The Bradford assay was used to measure protein concentration using the Coomasie Plus reagent (Thermo Fisher, 23236) on a 96-well-plate setting. The absorbance was measured immediately at 596 nm and a standard curve was automatically created using a Spectramax reader.

Western blotting was performed on precast gels using the XCell SureLoc Mini-Cell and XCell I Blot Module (Invitrogen, 10572913), and the protein samples were mixed with loading buffer (4× NuPAGE loading buffer (Invitrogen, 11549166) and 5% β-mercaptoethanol) at a concentration of 20 μg μl⁻¹. The samples were heated for 5 min at 95 ^oC and then were spun for 2 min at 20,800g. A 20 μl sample and 5 μl of protein ladder (Biorad, 1610373) were loaded on a 4–12%, Bis–Tris, 1.0 mm, ten-well precast gel (Invitrogen, NP0321PK2), and the gel was run at 120 V for 1 h and 50 min in a MOPS running buffer (Invitrogen, NP-0001). Transfer onto polyvinylidene difluoride membranes was performed by wet transfer, using the NuPAGE transfer buffer (Invitrogen, NP-0006) for 1 h at 30 V. Transfer efficiency was assessed by Ponceau S staining. The membranes were blocked for 1 h with 5% BSA buffer and then incubated with the primary antibody (diluted in 5% BSA) overnight at 4 °C. This was followed by 45 min incubation with the HRP-conjugated secondary antibody and enhanced chemiluminescence incubation for the appropriate amount of time (2 min for pSmad2, pSmad3, Smad2 and Smad3 and 5 s for the β-actin). Visualization of the bands was performed using the Chemidoc imaging system (Biorad) and quantification and densitometry was done with ImageJ.

ELISA and cytokine arrays

For the detection of cystatin C in murine plasma, the ‘mouse/Rat cystatin C’ immunoassay kit (R&D Technologies, MSCTC0) was used, according to the manufacturer’s instructions. The plate was scanned in a plate reader at 450 nm (wavelength correction was set to 570 nm) within 10 min after assay completion. The standard curve concentration calculations were performed on the ‘Myassays’ website (https://www.myassays.com/).

Cytokine arrays on murine plasma and tissue samples were performed by Eve Technologies, using the TGFβ1, 2, 3 magnetic bead kit for the measurement of the TGFβ ligands and the Milliplex MAP mouse cytokine magnetic bead panel (discovery assay ‘Mouse Cytokine Array/Chemokine Array 31-Plex (MD31)’).

Metabolomics on mouse urine

For the detection of urine amino acids, urine was collected from the mice by free urination 2 and 3 days before AAV-Cre injection on injection day, as well as 3 and 4 days post AAV-Cre injection. The urine collected by free urination after scruffing the mice was diluted 1:50 in cold metabolite extraction buffer (50% methanol, 30% acetonitrile and 20% water) and were vortexed for 30 s. The samples were then centrifuged at 20,800gfor 10 min at 4 °C. Liquid chromatography–mass spectrometry was performed as described previously⁴⁵. The source data are presented in Supplementary Table 6.

Cell culture

MEFs

WT MEFs were isolated from one E13-E14 WT C57Bl/6 embryo. The MEFs were cultured on 10 cm Petri dishes (Corning) in Dulbecco’s modified Eagle medium supplemented with 10% FBS, 1% penicillin–streptomycin, 1% ʟ-glutamine (cDMEM) at low O₂ (3%). The MEFs cultures were confirmed to be free of mycoplasma contamination. For the plasma treatments, passage (P)3–P5 MEFs were trypsinized, and the cell density was determined by using the CASY cell counter (Cambridge Bioscience, 5651808). A total of 30,000 cells (in 1 ml medium) were plated on 24-well-plate wells (with a round coverslip in each well) and were incubated with cDMEM with 1% plasma from either ΔMdm2^Hep or control mice for 24 h. The cDMEM with plasma was changed with fresh cDMEM every 2 days, and 6 days after the first medium change, the cells were stained for SA β-Gal as described above. The coverslips were mounted on the slides, and the SA β-Gal⁺ cells were quantified by manual counting of 20 random fields of view of using a Zeiss Axiovert 200 microscope at 20× magnification.

NS cell-derived neuronal cells

Neuronal cells were derived from human foetal neural crest progenitors as previously described⁴⁶. The cells were passaged using trypsin-EDTA solution and were seeded into a geltrex coated 48-well cell culture plate at a density of 5,000 cells per well with 250 μl of proliferation medium. They were allowed to proliferate for 2 days before the proliferation medium was withdrawn and replaced with differentiation medium. The growth medium was replaced with differentiation medium for 16 days before beginning the experimental procedure. The composition of the proliferation and differentiation media has been described previously⁴⁷.

For the plasma treatment experiments, plasma samples were thawed on ice, vortexed and heat inactivated for 30 min at 60 ^oC. They were diluted in media to the desired concentration (1:100) and added to the plate for 24 h. For the additional treatment with TGFβR1i, either TGFβR1i (0.2 mM) or vehicle (dimethylsulfoxide, DMSO) were mixed with the same plasma-containing medium and stayed on the cells for 24 h. Two days later, the cells were stained for SA β-Gal, and the positive cells were quantified as described in ‘SA β-Gal staining’ section.

PCLS and PCKS

Ethics

Human kidney tissue will be collected from donor kidneys declined for transplantation and accepted for research, through the Newcastle Transplant Tissue Biobank under project IOT054.

Kidney slice culture

Eight millimetre human kidney and murine kidney (WT) or liver (Mdm2flfl or ΔMdm2^Hep) tissue cores were generated using an 8 mm Stiefel biopsy punch, placed in a metal mould and 3% low gelling temperature agarose and allowed to set. The slices were then generated and cultured according to methods previously described^48,49. Human kidney slices were cultured with 10 ng TGFβ1 to induce fibrosis and treated with ±10 μM TGFβR1i SB-525334 (Sigma Aldrich, S8822, batch number: 000014491).

Media transfer experiment

Murine PCLS were generated from liver (Mdm2^flfl or ΔMdm2^Hep), and the medium was collected and pooled after 24 and 48 h and concentrated using Amicon Ultra-15 Centrifugal Filter Units (Merck, UFC9003). WT murine kidney slices were cultured in PCKS medium containing 0.5% v/v of concentrated cultured PCLS-conditioned medium (equivalent to 25% PCLS condition medium at final concentration) from the livers of control Mdm2^flfl or ΔMdm2^Hep mice, with or without 200 nM AZ12601011.

Statistical analyses and graphs

The Prism 9 Software (GraphPad Software) was used for statistical analyses. The Shapiro–Wilk test was used to assess data normality. For normally distributed data, the one-way analysis of variance (ANOVA), two-way ANOVA, the Brown–Forsythe and Welch ANOVA test, paired and unpaired t-tests or the Welch’s t-test, were used to test for statistical significance between data groups. The Kruskal–Wallis test or the Mann–Whitney test were performed for non-parametric data. All statistical tests comparing two groups were two-tailed. All figures were created using the Scribus Software (v1.4.7, G.N.U. general public licence). Unless otherwise stated, all data points on the line or bar graphs represent the mean ± standard error of the mean (s.e.m.), and each dot represents a single mouse.

Statistics and reproducibility

No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications^3,7,11,12. For animal experiments, the biological replicate sizes were chosen taking into account the variability observed in pilot and prior studies using AAV-TBG-Cre in the Mdm2 model. For all experiments, the animal/sample assignment was matched for sex and age-matched controls and were assigned based upon randomly assigned mouse identification markings. Batched staining and analyses alongside controls were used throughout. The investigators were not blinded for the in vivo experiments. No animals or data points were excluded from analyses. The technical staff administering the therapy were blinded to the mouse genotypes. All subsequent tissue handling and analysis were blinded and/or performed using standardized automated analyses where possible. Western blot studies were performed without blinding, given that samples were run by condition for visualization. In the figure legends, n represents the number of mice, unless explicitly stated. The data distribution for normality and testing of equal variances were assessed using Prism 9 Software. No animals or data points were excluded from analyses.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.