Ethics

This research complied with all relevant ethical regulations and was approved and overseen by the following ethics review boards. Fully anonymized liver biopsies from patients with non-alcoholic fatty liver disease were obtained from the Imperial Hepatology and Gastroenterology Biobank which is fully REC-approved by the Oxford C Research Ethics Committee under REC reference 16/SC/0021. Informed written consent was provided by the donors. Mouse liver fibrosis experiments were performed according to German law and with the approval of the Regierungspräsidium Karlsruhe (G139/19). All other mouse procedures were performed under license, according to the UK Home Office Animals (Scientific Procedures) Act 1986, ARRIVE 2.0, and approved by the Imperial College’s animal welfare and ethical review body (aging experiments, PPL 70/8700; liver cancer initiation and senolysis experiments, PPL 70/09080).

Cell lines

Both female (IMR90, MCF7, HEK-293T) and male (A549, SK-HEP-1, HCT116, MRC-5) cell lines were used in this study. A549 (CCL-185), HCT116 (CCL-247), HEK-293T (CRL-11268), SK-HEP-1 (HTB-52), SK-MEL-103 (HTB-70), MRC-5 (CCL-171) and MCF7 (HTB-22) cells were obtained from the American Type Culture Collection (ATCC). Early passage IMR90 cells (ATCC CCL-186) were obtained from Coriell Institute. IMR90 ER:RAS cells were generated by retroviral infection of IMR90 cells and have been described elsewhere⁴⁵. A549, HCT116, HEK-293T, IMR90, SK-HEP-1, SK-MEL-103, and MCF7 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco), supplemented with 10% fetal bovine serum (FBS, Sigma F7524) and 1% antibiotic-antimycotic solution (Gibco). MRC-5 cells were cultured in Eagle’s Minimum Essential Medium (EMEM), supplemented with 10% FBS and 1% antibiotic-antimycotic solution (Gibco). For inducing quiescence, the media was replaced with 0.5% FBS in DMEM.

Senescence induction

The following drugs were used to induce senescence in different cell lines after culturing cells in 96-well (Nunc, Thermo Fisher) or 100 mm dishes (Corning, 430167): A549, 2 μM etoposide (Eto, Sigma–Aldrich, E1383), 0.2 μM doxorubicin (Doxo, Selleck chemicals, E2516), 2 μM alisertib (Ali, Selleck chemicals, S1133), 1 μM barasertib (Bara, Selleck chemicals, S1147); SK-MEL-103, 0.25 μM Eto, 0.1 μM Doxo, 0.25 μM Ali, 0.5 μM Bara; SK-HEP-1, 0.25 μM Eto, 0.1 μM Doxo, 0.1 μM Ali, 0.5 μM Bara; MCF7, 1.5 μM Eto, 0.1 μM Doxo. HCT116, 2 μM Eto, 0.5 μM Doxo. IMR90, 50 μM Eto, 2 μM Doxo (washed and media replaced 24 h post-treatment). Cell culture media with or without drugs was changed every 72 h. Senescence was assessed after 7 days of drug treatment unless otherwise stated.

Screen for drugs inducing senescence

676 drugs from the Target selective library (Selleck Chemicals) and Protein Kinase Inhibitor Library II (EMD Calbiochem®, Cat. No. 539745) were used to screen for drugs inducing senescence selectively in cancer cells. Drugs (at 10 μM) were added 24 h after seeding cells. Cell media was changed 72 h after and cells were fixed 5 days after cells were seeded. Cells were then fixed in 4% PFA, DAPI-stained, and images acquired (see High Throughput Microscopy). We screened the drugs in A549 lung adenocarcinoma and IMR90 fibroblasts in parallel, with biological triplicates per cell line. The percentage of senescent cells was calculated using the GM algorithm. A toxicity threshold was established against the viability of the positive controls. Samples with a lower cell count than 40% of that of the positive control were considered toxic and were excluded from analysis. This filtering excluded 69 drugs and data from cells treated with the remaining 607 drugs was taken forward for normalization. B-score normalization of the predictions of senescence induction was carried out (see B-score normalization analysis). For a drug to be considered a hit, at least 2 of the 3 replicates would need to have a B-score >15. Drug candidates were then classified based on the predicted capacity to induce senescence in the cell lines tested based on the established threshold. Senescent B scores for all screened drugs have been included in the source data.

For follow-up studies (described in Fig. 6 and Supplementary Fig. 13), we used the following drug concentrations: doxorubicin, 0.5 μM; MLN8054 (Selleck Chemicals, S1100), 3 μM; ZM447439 (Selleck chemicals, S1103), 2 μM; SC144 (Selleck chemicals, S7124), 1 μM; ARQ 621 (Selleck chemicals, S7355); 0.1 μM; Niraparib tosylate (MK-4827, Selleck chemicals, S7625), 10 μM; AG-14361 (Selleck chemicals, S2178), 10 μM. Drug stocks (10 mM) were prepared in dimethylsulfoxide (DMSO) and stored at −20 °C.

B-score normalization analysis

To analyze the drug screen, senescence prediction was normalized by B-score using the R package CellHTS2 (https://doi.org/10.18129/B9.bioc.cellHTS2)⁴⁶. Value normalization was performed using the plate-averaging method and on separate batches for A549 and IMR90 cells.

Antibodies

The following antibodies were used for the immunofluorescent and immunohistochemistry experiments: mouse monoclonal anti-bromodeoxyuridine (BrdU) (3D4; BD Biosciences, 555627) 1:2000; rabbit polyclonal anti-53BP1 antibody (Novus Biologicals, NB100-304) 1:1000; goat polyclonal anti-uPAR (Novus Biologicals, AF534), 1:200; mouse monoclonal anti-Nras (Santa Cruz, sc-31), 1:500; rabbit monoclonal anti-LINE-1 ORF1p (Abcam, ab216324) 1:500; mouse monoclonal anti-phospho-Histone H2A.X (Ser139) (Sigma–Aldrich, 05-636) 1:250; rabbit polyclonal anti-p21 (2947 S; Cell Signaling) 1:2000; mouse monoclonal anti-p53 (DO-1, Santa Cruz, sc-126) 1:100; rabbit monoclonal anti-p21 (EPR18021, Abcam) 1:700; rabbit recombinant monoclonal anti-GFP antibody [EPR14104] (ab183734) 1:500; Alexa Fluor 488/594 conjugated, (Thermo Fisher Scientific, A11029/A11032) goat anti-mouse IgG (H + L), 1:2000; Alexa Fluor 488/594 conjugated, (Thermo Fisher Scientific, A11034A11037) goat anti-rabbit IgG (H + L), 1:2000; Alexa Fluor 488/594 conjugated, (Thermo Fisher Scientific, A11055/A11058) donkey anti-goat IgG (H + L), 1:2000.

Vectors

LentiGuide-Puro (Addgene, #52963) was used to express GFP, and pBabe puro IRES-mCherry (Addgene, #128038) for mCherry expression. Cells were FACS sorted and expanded. Cells expressing the construct were selected in 2 μg/mL of puromycin and kept under selection in 0.5 μg/mL of puromycin.

Immunofluorescent staining of cells

Cells were grown in 96-well plates (Nunc^TM MicroWell^TM, 167008, Thermo Fisher Scientific), fixed with 4% PFA (w/v) for 20 min, and then permeabilized in Triton X-100 0.2% diluted in PBS for 10 min. Cells were then blocked with 1% bovine serum albumin (BSA) (w/v) for 25 min. Cells were incubated with the primary antibody diluted in a blocking solution for 1 h and washed thrice in PBS. Cells were incubated with the secondary antibody (Invitrogen, Alexa Fluor^TM) diluted in blocking solution for 30 min and after washing thrice on PBS, 1 μg ml⁻¹ of DAPI was added for 12 min and washed with PBS thrice.

Cytochemical SA-β-galactosidase assay

Cells were grown on 6-well (Nunc^TM, 140675) or 96-well plates and fixed with 0.5% glutaraldehyde (w/v) (Sigma–Aldrich) for 15 min, then washed with 1 mM MgCl₂/PBS at pH 6.0 and incubated in X-Gal solution (5 mM K₃(Fe(CN)₆), 5 mM K₄(Fe(CN)₆) and 1 mg ml⁻¹ of X-Gal by Thermo Scientific) for 6 h and 8 h in cancer cells and fibroblasts respectively, at 37 °C. Brightfield images were acquired using DP20 digital camera attached to an Olympus CKX41 inverted light microscope, at 4x and 10x magnification.

Senolytic assay

Drugs were diluted to the required concentration in DMSO and stored at −20 °C. Cells were induced to senesce for 7 days and then cultured with 1 μM ABT-263 (Selleck Chemicals; S1001) or 5 μM ABT-737 (Selleck Chemicals; S1002) for 72 h. Cells were then fixed in 4%PFA (w/v) for 20 min and stained with DAPI for 12 min. Cells were then washed thrice in PBS.

Irradiation-induced DNA damage

To induce DNA damage cells cultured in 96-well or 6-well plates were exposed to ionizing radiation (15 Gy). Media was changed 24 h later, and DNA damage was assessed by immunofluorescence at the required time point.

Fluorescent SA-β-galactosidase assay

Cells grown in 96-well plates were washed in phosphate-buffered saline (PBS) and incubated with 33 μM C₁₂FDG (Abcam, ab273642) diluted in DMSO for 1 h. Cells were then fixed with 4% paraformaldehyde (PFA) for 15 min, washed thrice in PBS, and stained with 1 μg ml⁻¹ of DAPI. Images were taken using a high-throughput fluorescent microscope IN Cell Analyzer 2500HS (Cytiva) with a 10× objective for quantification. The percentage of SA-β-galactosidase positive cells was calculated using IN Carta^TM Image Analysis Software (version 1.14) based on cellular fluorescence intensity using an arbitrary threshold to define positive cells.

Gene expression analysis

Total RNA was extracted using RNeasy® Minikit (Qiagen). cDNA was produced using Superscript II reverse transcriptase (Invitrogen) and Random Hexamers (Invitrogen). Quantitative real-time PCR (RT-qPCR) was performed using SYBR^TM Green PCR master mix (Applied Biosystems) in a CFX96 RT-PCR system C1000 Touch (Bio-Rad). For data normalization, GAPDH expression was used. The primer pairs used are:

GAPDH: GAAGGTGAAGGTCGGAGTC; TTGAGGTCAATGAAGGGG

CDKN1A: CGTGTCACTGTCTTGTACCCT; GCGTTTGGAGTGGTAGAAATCT

CDKN2A: CGGTCGGAGGCCGATCCAG; GCGCCGTGGAGCAGCAGCAGCT

IL1A: AGTGCTGCTGAAGGAGATGCCTGA; CCCCTGCCAAGCACACCCAGTA

IL1B: TGCACGCTCCGGGACTCACA; CATGGAGAACACCACTTGTTGCTCC

High-throughput microscopy

Cells were cultured in 96-Well Flat-Bottom plates (Thermo Fisher) and CellCarrier-96 Ultra Microplates (Perkin Elmer) were used. Plates were analyzed using IN Cell Analyzer 2500HS high content analysis (HCA) imaging at a magnification of 20× or 40×, with a binning of 1 × 1. TIF files obtained in HCA were analyzed using IN Carta^TM Image Analysis Software (Cytiva, version 1.14). The acquired images had the following characteristics: width and length of 663.005 μm (2040 pixels), at a 3.0769 pixels per μm resolution (20×); width and length of 331.5 μm (1020 pixels), at a 0.3250024 pixels per μm resolution (40x). The following nuclear features were extracted using the In Carta software (See “Software” section): Area (in μm²), form factor (object roundness), elongation (object short axis/object long axis), compactness (average radius of the object), chord ratio (object min. chord ratio/object max. chord length), gyration radius (average radius of the shape), displacement (distance between the nucleus center of gravity and the cell center of gravity, normalized by the gyration radius of the nucleus). Where protein expression signal was analyzed intensity measures were also acquired.

To capture nuclear morphology parameter measurements, images were thresholded based on DAPI, primary objects identified, and measurements performed (see “Software” section). For quality control and to exclude artifacts, the cell segmentation pipeline performed a noise removal step, excluding shapes when image contrast was low. A sensitivity threshold was also established to accurately detect true nuclei events and a typical diameter of a nucleus was also established, to further exclude non-conforming structures. Overexposed targets were eliminated establishing a minimum DAPI intensity threshold and objects touching edges were excluded from the analysis to avoid partially acquired nuclei.

Libraries of nuclear parameters of senescent and normal cells in culture

To develop the algorithm training sets (summarized in Supplementary Table 1), the indicated cells were seeded in 96-well plates and cultured and treated with the indicated drugs or DMSO as a control. Seven days after treatment, cells were fixed, and stained with DAPI for imaging. Each plate contained 30 wells with cells treated with the senescence inducer and 30 wells treated with DMSO (as a control). Datasets for each cell line and senescence induction contained data derived from at least three plates, resulting in a total of between 0.1 × 10⁶ and 0.9 × 10⁶ cells per condition (see Supplementary Table 1 for details). Training datasets (as indicated in Supplementary Table 1) were then generated by randomly selecting 10,000 normal cells and 10,000 treated cells for each training set. For the General Model, randomized samples from all training datasets (3 cell lines, A549, SK-MEL-103, SK-HEP-1; 4 conditions: Etoposide, Doxorubicin, Alisertib, Barasertib) were taken to develop the classifier, as noted in Supplementary Table 1. Independent training sets (with different randomizations of the same libraries) were constructed for the classification tree and random forest algorithms.

Software

The following packages were utilized for classification trees, random forest building, and related analysis. For Classification tree (CT) and random forest (RF) algorithms python version 3.7.7 was used. The following packages were also utilized: scikit-learn and derived packages;⁴⁷, pandas, numpy, matplotlib. pyplot, seaborn, and csv. Area, Form Factor, Elongation, Compactness, Chord Ratio, Gyration, and Displacement were used as nuclear features. Analysis of public software CellProfiler (version 4.2.4)²¹ was performed using a nucleus detection workflow. For High Content Analysis (HCA), InCarta software (Molecular Devices, version 1.14) was used. For B-score analysis R (version 4.3.1) and packages BiocManager (version 1.30.22) and cellHTS2 (version 2.64) were used.

Generation of classifiers to identify senescence in cell culture

For classification tree (CT)-based classifiers, preliminary classification trees were built using sklearn, providing 30% of the training set as test size. After assessing initial accuracy, AUC, and ROC curves, cost complexity pruning was performed to avoid overfitting. The optimal value of alpha was calculated and applied to develop the classification trees. Obtained classification tree branches were eliminated where redundancy occurred. For random forest (RF)–based classifiers, the test size was set at 0.5. For CT classifiers the classes were established in a binary manner, where 0 equalled growing, normal cells and 1 represented senescent (treated) cells. For RF classifiers senescence probability was estimated and values > 0.5 were considered as senescent. Classifiers were ultimately tested on new experiments (test data) and the accuracy of prediction was assessed. For the Voting-Based Clustering Algorithm (VCA), the input from all algorithms described in Supplementary Table 2 (except the CellProfiler-based classifiers) was considered. Relabeling of partitions was avoided and opted for a democratic vote system with equal weight per classifier algorithm.

Algorithm performance metrics

Algorithm accuracy on the test data was measured by area under the curve (AUC) and receiver operating characteristic (ROC) curve (True Positive Rate vs False Positive Rate) assessment and posterior testing on new data not belonging to the training datasets. Algorithm accuracy was also measured using the following metrics in a dataset of co-cultures of senescence and normal cells:

TP true positive; TN true negative; FP false positive; FN false negative.

Co-cultures of senescent and non-senescent cells

The setup of these experiments is described in detail in Supplementary Fig. 5. Briefly, 10⁵ cells (for DMSO) and 10⁶ cells (for treatment) were seeded in 100 mm dishes. 24 h after plate seeding media was washed once with PBS (Gibco^TM, Thermo Fisher, 10010023), and DMSO or senescence-inducing drug was added to the media (DMEM, 10% FBS, 1% antibiotic-antimycotic). Media (with drug or DMSO) was replaced every 72 h. 6 days post-treatment, senescent and control cells were trypsinized and counted using a Guava Muse Cell Analyzer. DMSO-treated and drug-treated cells were seeded in separate master plates (96-well Round (U) Bottom Plate, Thermo Fisher, 163320) as indicated in Supplementary Fig. 5c. Those master plates were used to generate the co-culture plate. After 24 more hours, cells were cultured in C₁₂FDG (33 μM, diluted in DMSO) for 30 min, fixed in 4% PFA for 15 min, and stained with DAPI.

Senescence classifiers based on CellProfiler data

To develop algorithms utilizing CellProfiler (version 4.2.4)²¹, we produced a new training set consisting of 4 plates of DMSO-treated and etoposide-treated A549 cells. 7 days after treatment cells were fixed, stained with DAPI, and imaged using an INCell Analyzer 2500HS. The acquired TIF files were then analyzed utilizing a bespoke nuclear workflow CellProfiler protocol (Dapi_CellProfiler.cpproj, see “Code Availability” section). The following features were considered for CT and RF algorithm development: Area, Bounding Box Area, Compactness, Convex Area, Eccentricity, Equivalent Diameter, Extent, Form Factor, Major Axis Length, Maximum Feret Diameter, Maximum Radius, Mean Radius, Median Radius, Minimum Feret Diameter, Minor Axis Length, Perimeter, and Solidity. To capture nuclear morphology parameter measurements, images were thresholded based on DAPI, primary objects identified, and measurements performed. From the resulting parameter files, cells were grouped by treatment, and randomized, and 10,000 cells were extracted per condition, following the same procedure to develop CT- (AECP) and RF-based (AERFCP) classifiers.

Mouse experiments

Mice were kept on a 12-h light/dark cycle and between 21–23 °C temperature and 45–65% humidity levels under specific pathogen-free barrier conditions within individually ventilated cages with ad libitum access to standard chow food (SDS RM1/3 [E] LBS Serving Biotechnology) and water. C57BL/6 J littermate mice were used unless otherwise specified. As sex is not a factor in the scope of the study design, liver sections of both male and female mice were used (as detailed below) for different experiments. Animal welfare was monitored and euthanasia practices were performed according to the requirements of the aforementioned practice licenses and regulatory frameworks.

Liver fibrosis

Eight-week-old male C57BL6/J mice were treated twice a week with either corn oil (n = 6) or carbon tetrachloride CCl₄ (0.5 mL/kg) (n = 7) by intraperitoneal injection for 6 weeks to induce liver fibrosis as described before⁴⁸. Mice were sacrificed at the indicated time points and analyzed for senescence.

Aging experiment

Male C57BL/6 J littermates were used. Mice that were 90 days old (n = 13) were used for the young cohort and 600 days old mice (n = 7) for the old cohort. Mice were sacrificed at the indicated time points and analyzed for senescence.

Liver cancer initiation and senolysis experiments

Hydrodynamic tail vein injection (HDTVI) was carried out in female C57BL/6 J (Charles River UK) mice aged 5–6 weeks using 25 μg of a transposon expressing Nras^G12V or Nras^{G12V, D38A} along with 5 μg of SB13 transposase-expressing plasmid. All plasmids were prepared with GenElute HP Endotoxin-Free Maxiprep kit (Sigma). For HDTVI, vectors were diluted in normal saline to a final volume of 10% body weight. HDTVI was performed within 7–8 s.

For liver cancer initiation experiments, mice transduced with transposon vectors co-expressing GFP and either Nras^G12V (n = 6) or Nras^{G12V, D38A} (n = 9) were used.

For the experiment described in Supplementary Fig 15c, d an additional cohort of mice transduced with transposon vectors co-expressing GFP and Nras^G12V (n = 12) was used, Mice were culled 9 days (after HDTVI) and livers were collected for paraffin embedding.

For senolysis experiments, a transposon vector co-expressing Nras^G12V and Gaussia luciferase (Gluc) was used. On day 5 after HDTVI mice were given either a senolytic drug (n = 9) or vehicle (n = 8) intraperitoneally (i.p.) daily for 4 days. 24 h after the last drug injection mice were culled and livers collected for paraffin embedding.

Immunohistochemical staining of tissue sections

Mouse liver tissue sections were deparaffinized in Histoclear^TM (Scientific laboratory supplies) for 5 min, and washed in decreasing concentrations of ethanol, until a final 5 min wash in dH₂O Heat-induced epitope retrieval (HIER) was then performed in a pressure cooker for 20 min in citrate-based at pH 6.0 (VectorLab, H-3300-250) or tris-based at pH 9.0 (VectorLab, H-3301-250), following the antibody manufacturer’s instructions. For intracellular expression stains and sections were washed in Triton X-100 0.2% in PBS for 10 min and washed in PBS for 5 min. Slides were then incubated in BLOXALL blocking solution (VectorLab, SP-6000), washed in PBS, and exposed to Animal-Serum Free serum (Cell Signaling, 15019 L) diluted in dH₂O for 30 min. Slides were then incubated with primary antibody overnight in a humidified chamber at 4 °C. Slides were washed twice in PBS for 5 min and incubated with secondary antibody SignalStain® Boost IHC detection reagent Mouse/Rabbit, HRP (Cell Signalling Technology, 8125) for 30 min. After, slides were washed in PBS and incubated in SignalStain DAB (CST, 8059) for 5 min or until the HRP signal was visible and the reaction stopped in dH₂O. Cells were then stained for Hematoxylin (DAKO, Mayer’s Hematoxylin, S3309) for 30 s and washed in dH₂O. When necessary, slides were further stained in Eosin Y (Sigma–Aldrich, HT110132-1L). Slides were, dehydrated in 75% ethanol for 1 min and 100% ethanol for 5 min, washed in Histoclear for 5 min, and mounted in DPX (Sigma–Aldrich).

Immunofluorescence staining of tissue sections

For Immunofluorescence staining, deparaffinization, and antigen retrieval were performed as described previously (see “Immunohistochemical staining of tissue” sections). Mouse liver samples were incubated overnight in the primary antibody previously diluted in antibody diluent (Dako). Samples were washed in PBS three times for 5 min. Samples were then incubated in secondary antibody SignalStain® Boost detection reagent (Cell Signalling Technology, 8125) for 45 min. The signal was then amplified using Thermo Fisher AlexaFluor^TM 488/647 Tyramide SuperBoost^TM Kit (B40958) following the manufacturer’s instructions. To perform double staining, samples were incubated in HCl 0.02 N for 20 min after the first antibody signal amplification step. Samples were then washed in PBS for 5 min and peroxidase blocking was reapplied for 20 min. Samples were then incubated in Animal-Serum Free blocking solution (Cell Signaling, 15019 L) diluted in H₂0 for 1 h and the second primary antibody incubation was performed overnight. The signal was then amplified using a different wavelength-reactive SuperBoost^TM Kit, using antibodies raised in different hosts to avoid cross-reactivity. Samples were then washed three times in PBS for 5 min and incubated with DAPI (1 μg ml⁻¹ in PBS) for 5 min. Samples were washed thrice in PBS for 5 min and mounted in 50% glycerol in PBS.

Slide image acquisition and analysis

Slides containing preclinical and clinical liver samples were acquired using 40x brightfield objective on a Zeiss AxioScan Z.1 or Leica Aperio AT2 slide scanner and analysis was performed using QuPath version 0.3.0, adjusting the built-in cell acquisition parameters to immunofluorescent and immunohistochemical samples to maximize the accuracy of cell and nuclear detection for feature extraction and signal quantification. A pixel size of 0.5 μm was established, and to accurately detect nuclei and avoid artifacts a background and median filter radius was established, together with a Gaussian filter to reduce noise and a minimum nuclear area. To further ensure accurate detection, an intensity threshold and a maximum background intensity were set. For immunofluorescence nuclear detection, DAPI was used as a detection channel and the same artifact filters were incorporated into the pipeline. The following features were extracted: centroids X and Y (coordinates in μm, for single cell positioning in the slide), nucleus area, nucleus perimeter, nucleus circularity, nucleus max caliper, nucleus min caliper, and nucleus eccentricity. Where immunohistochemical staining was performed, DAB optical density (OD) mean and total DAB were also acquired (nuclear of cellular, corresponding to protein expression localization). The indicated data were then extracted and analyzed. For human patient samples, a circularity threshold (nuclei superior to 0.7) was established to ensure that captured nuclei belonged predominantly to hepatocytes and not to fibroblasts or immune cells. The number of cells per sample (as evaluated by acquired nuclei) varied between 7 × 10⁴ and 1 × 10⁵. Samples with less than 10,000 cells were excluded from the analysis.

Senescence scoring system in tissue sections

To assess senescence in liver tissue sections, we took advantage of liver sections of mice transduced with Nras^G12V using hydrodynamic tail vein injection (HDTVI), stained with anti-p21^Cip1 antibodies, and counterstained with Hematoxylin. 4 slides were scanned and analyzed using QuPath. Cells were classified as p21^Cip1-positive or -negative based on DAB nuclear mean intensity (>0.2 for p21^Cip1-positive cells and <0.2 for p21^Cip1-negative cells). Nuclear features (area, perimeter, maximum caliper, minimum caliper, eccentricity, and circularity) were extracted for p21^Cip1-positive and p21^Cip1-negative cells (see “Slide image acquisition and analysis” section). Data from 4.32 × 10⁵ p21^Cip1-negative cells was used to obtain average measurements, which defined the ideal normal (non-senescent) cell. p21^Cip1-positive cells were ranked based on the p21^CIP1 staining intensity. The average parameters of the top one hundred p21^Cip1-positive cells were selected to define the ideal senescent cell. Therefore, for each nuclear feature, an ideal parameter value for normal cells (P^N) and an ideal value for senescent cells (P^S) were defined (shown in Supplementary Table 5).

For each cell, we performed the following operation:

Where:

n = number or features

p = acquired parameter value

P^N = ideal parameter value for normal cells

P^S = ideal parameter value for senescent cells

T = summation of absolute values of differences from acquired parameters

Consequently, individual features from single nuclei obtained a feature score, which was then corrected by the weighted effect of each feature in the summation of the absolute values of the differences (ΣT), which allows to proportionalise the effect of the features. Individual feature scores were then aggregated to provide a single value (that we termed the cell senescence score, CSS) for each cell. To calculate the tissue senescence score (TSS), for each tissue section, we plotted curves showing the distribution of individual cell senescence scores for all the cells present in that section. Thus, the TSS from a given tissue sample relies upon the distribution of its individual CSS values. To obtain the TSS value, that describes senescence presence in the sample, we scored the sections based on the percentage of CSS values between 1 and 5 (CSS values associated with nuclear senescent features), thus obtaining a unique metric for the tissue section. We evaluated other ranges of CSS values (as shown in Supplementary Table 4) and chose 1–5 as the one better correlating with the percentage of senescent cells present in the tissues. Importantly, the same operation and metrics to calculate CSS/TSS values (without adjusting or changing any of the parameters for subsequent experiments) were applied to all preclinical and clinical liver models used in this study.

Senescence assessment in samples from patients with mild fatty liver disease

Human liver biopsies were fully anonymized and acquired from the Imperial Hepatology and Gastroenterology Biobank, therefore no regard for sex and gender was considered. Sections were deparaffinized, and hydrated, and then heat-mediated antigen retrieval was performed in citrate-based pH 6.0 solution. The endogenous peroxidase was quenched with 3% hydrogen peroxide. The sections were incubated with mouse monoclonal to p16^INK4a (CINtec, 9511, clone E6H4), followed by rabbit anti-mouse IgG. The sections were subsequently incubated with anti-rabbit IgG conjugated with polymeric horseradish peroxidase linker (Leica Bond Polymer Refine Detection, DS9800). DAB was used as the chromogen and the sections were then counterstained with hematoxylin and mounted with DPX. IHC was performed on Leica BOND III. Serial sections were stained with H&E and used to calculate tissue senescence scores. Slides were scanned with NanoZoomer 2.0HT (Hamamatsu, Japan). NDP.scan 3.2.12 software was used for digital image acquisition and NDP.view2 software was used for image viewing. 36 samples were processed and imaged, but 2 samples with less than 10,000 cells were excluded from the analysis.

Statistical analysis

We used GraphPad Prism (Version 9.4.0) for statistical analysis. Two-tailed, unpaired Student’s t-tests were used to estimate statistically significant differences between groups, as well as one-way ANOVA when required. Pearson correlation analysis was performed utilizing a two-tailed option, with a 95% confidence interval. Simple linear regression was also performed to display the corresponding fit line. To study the cumulative distributions between treated and control nuclear features we performed the Kolmogorov-Smirnov (K-S) test and detailed the maximum absolute difference (D) and the associated P value.

Reporting summary

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