CoChat

The DOI matches its Crossref record; 58 of the 58 references it cites were found in a registry or an index.

Nuclear morphology is a deep learning biomarker of cellular senescence

Open access

Published version of a preprint

Abstract

Cellular senescence is an important factor in aging and many age-related diseases, but understanding its role in health is challenging due to the lack of exclusive or universal markers. Using neural networks, we predict senescence from the nuclear morphology of human fibroblasts with up to 95% accuracy, and investigate murine astrocytes, murine neurons, and fibroblasts with premature aging in culture. After generalizing our approach, the predictor recognizes higher rates of senescence in p21-positive and ethynyl-2’-deoxyuridine (EdU)-negative nuclei in tissues and shows an increasing rate of senescent cells with age in H&E-stained murine liver tissue and human dermal biopsies. Evaluating medical records reveals that higher rates of senescent cells correspond to decreased rates of malignant neoplasms and increased rates of osteoporosis, osteoarthritis, hypertension and cerebral infarction. In sum, we show that morphological alterations of the nucleus can serve as a deep learning predictor of senescence that is applicable across tissues and species and is associated with health outcomes in humans.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s43587-022-00263-3

Cites

58 of 58 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Nuclear morphology is an accurate senescence predictor in cultured cells. a, Analysis workflow. DNN, deep neural network. b, Sample nuclei for control, RS- and IR-induced senescent cells. c, Area of identified nuclei (RS n = 6,976, IR n = 19,193, control n = 68,971; mean ± 95% confidence interval (CI), Tukey multiple comparison). d, Convexity of identified nuclei (same as panel c). e, Aspect ratio of identified nuclei (same as panel c). f, Scatter plot of individual nuclei, with overall distributions for each at the top and right margins. g, Cell cycle analysis after exposure to several doses of IR; mn, multinucleated cells (n = 4, mean ± 95% CI). h, Accuracy of a deep neural network predictor on test data. i, Receiver–operating characteristic (ROC) curve of the deep neural network. j, Percentage of nuclei in each state classified as senescent for independent cell lines. k, Distribution of prediction probabilities for several doses of IR for three fibroblast cell lines. p(sen), predicted senescence score. l, Distribution of p21 intensities for several doses of IR for three fibroblast cell lines. m, Distribution of PCNA intensities for several doses of IR for three fibroblast cell lines.
Fig. 1 Nuclear morphology is an accurate senescence predictor in cultured cells. a, Analysis workflow. DNN, deep neural network. b, Sample nuclei for control, RS- and IR-induced senescent cells. c, Area of identified nuclei (RS n = 6,976, IR n = 19,193, control n = 68,971; mean ± 95% confidence interval (CI), Tukey multiple comparison). d, Convexity of identified nuclei (same as panel c). e, Aspect ratio of identified nuclei (same as panel c). f, Scatter plot of individual nuclei, with overall distributions for each at the top and right margins. g, Cell cycle analysis after exposure to several doses of IR; mn, multinucleated cells (n = 4, mean ± 95% CI). h, Accuracy of a deep neural network predictor on test data. i, Receiver–operating characteristic (ROC) curve of the deep neural network. j, Percentage of nuclei in each state classified as senescent for independent cell lines. k, Distribution of prediction probabilities for several doses of IR for three fibroblast cell lines. p(sen), predicted senescence score. l, Distribution of p21 intensities for several doses of IR for three fibroblast cell lines. m, Distribution of PCNA intensities for several doses of IR for three fibroblast cell lines.CC BY · PubMed Central
Fig. 2 Analyzing quiescence, density, mixtures and morphology. a, Correlation between predicted senescence and nearby SA-β-gal regions, showing all and 90% confidence predictions only for RS and IR groups. b, Correlation between predicted senescence and multiple markers, showing all, filtered for markers with strong signals and filtered with 90% confidence predictions only. c, Percentage of EdU-positive cells for serum-starved quiescent, serum-starved and restored, IR-treated and control cells. d, Predicted senescence for quiescence, serum-starved, serum-starved and restored, IR-treated and control cells. e, Pearson correlation between density and predicted senescence (RS n = 7, IR n = 4, control n = 4; mean ± 95% CI). f, Prediction accuracy for several mixtures with different ratios of senescent and nonsenescent cells; each point represents mean accuracy for each of three cell lines. g, Samples of mixtures with different ratios of cells, showing predicted senescence per nuclei. h, Accuracy of deep neural networks trained and predicting after different normalization methods. acc, accuracy. i, Correlation between morphological metrics and predicted senescence by class. BG, background.
Fig. 2 Analyzing quiescence, density, mixtures and morphology. a, Correlation between predicted senescence and nearby SA-β-gal regions, showing all and 90% confidence predictions only for RS and IR groups. b, Correlation between predicted senescence and multiple markers, showing all, filtered for markers with strong signals and filtered with 90% confidence predictions only. c, Percentage of EdU-positive cells for serum-starved quiescent, serum-starved and restored, IR-treated and control cells. d, Predicted senescence for quiescence, serum-starved, serum-starved and restored, IR-treated and control cells. e, Pearson correlation between density and predicted senescence (RS n = 7, IR n = 4, control n = 4; mean ± 95% CI). f, Prediction accuracy for several mixtures with different ratios of senescent and nonsenescent cells; each point represents mean accuracy for each of three cell lines. g, Samples of mixtures with different ratios of cells, showing predicted senescence per nuclei. h, Accuracy of deep neural networks trained and predicting after different normalization methods. acc, accuracy. i, Correlation between morphological metrics and predicted senescence by class. BG, background.CC BY · PubMed Central
Fig. 3 Predictions from deep ensembles increase accuracy. a, Heatmap of variation in predictions by members of ensemble; 500 sample nuclei as rows, ensemble members as columns; blue is control, and white is senescent. b, Heatmap of per-class accuracy for control (Ctrl) and senescent (Sen) by ensemble model. c, Accuracy of deep ensemble. d, ROC curve for the deep ensemble. e, Accuracy of single model, BNNs, deep ensemble and bagging (Bag). f, Accuracy of deep ensemble with normalized samples. g, ROC curve for the deep ensemble with normalized samples. h, Accuracy of RS-only model. i, Accuracy of IR-only model.
Fig. 3 Predictions from deep ensembles increase accuracy. a, Heatmap of variation in predictions by members of ensemble; 500 sample nuclei as rows, ensemble members as columns; blue is control, and white is senescent. b, Heatmap of per-class accuracy for control (Ctrl) and senescent (Sen) by ensemble model. c, Accuracy of deep ensemble. d, ROC curve for the deep ensemble. e, Accuracy of single model, BNNs, deep ensemble and bagging (Bag). f, Accuracy of deep ensemble with normalized samples. g, ROC curve for the deep ensemble with normalized samples. h, Accuracy of RS-only model. i, Accuracy of IR-only model.CC BY · PubMed Central
Fig. 4 Senescence can be predicted across cell types and species. a, ɣH2AX foci per nuclei (RS n = 1,537, IR n = 5,365, control n = 9,971; mean ± 95% CI, Tukey multiple comparison). b, 53BP1 foci per nuclei (same as panel a). c, Correlation between foci count and predicted senescence. d, Representative micrographs; scale bar, 10 μm; HGPS, Hutchinson–Gilford progeria syndrome; AT, ataxia telangiectasia; CS, Cockayne syndrome. e, Nuclear area (AT n = 4,340, CS n = 4,524, P1 n = 9,948, P2 n = 4,924, P3 n = 4543, control 1 n = 14,480, control 2 n = 9,875, control 3 n = 15,074, control n = 9,371, control 5 n = 7,542, mean ± 95% CI). f, ɣH2AX foci per nuclei (same as panel e). g, 53BP1 foci per nuclei (same as panel e). h, Predicted probability of senescence per nuclei (same as panel e). i, Nuclei with nearby SA-β-gal regions. j, Correlation between predicted senescence and nearby SA-β-gal (left axis) and number of nuclei (right axis) with increasing thresholds. k, DAPI intensities (same as panel e). l, Representative micrographs of senescent murine astrocytes and neurons; scale bar, 10 μm. m, Nuclear area of murine astrocytes (IR n = 4,888, control n = 13,549; mean ± 95% CI, Tukey multiple comparison). n, Nuclear area of murine neurons (IR n = 62,847, control n = 33,303; mean ± 95% CI, Tukey multiple comparison). o, ɣH2AX foci per nuclei (IR n = 2,016, control n = 5,202; mean ± 95% CI, Tukey multiple comparison). p, 53BP1 foci per nuclei (same as panel o). q, ɣH2AX foci per nuclei (IR n = 63,241, control n = 33,533; mean ± 95% CI, Tukey multiple comparison). r, 53BP1 foci per nuclei (same as panel q). s, Predicted senescence (same as m). t, Predicted senescence (same as panel q). PCC, Pearson correlation coefficient.
Fig. 4 Senescence can be predicted across cell types and species. a, ɣH2AX foci per nuclei (RS n = 1,537, IR n = 5,365, control n = 9,971; mean ± 95% CI, Tukey multiple comparison). b, 53BP1 foci per nuclei (same as panel a). c, Correlation between foci count and predicted senescence. d, Representative micrographs; scale bar, 10 μm; HGPS, Hutchinson–Gilford progeria syndrome; AT, ataxia telangiectasia; CS, Cockayne syndrome. e, Nuclear area (AT n = 4,340, CS n = 4,524, P1 n = 9,948, P2 n = 4,924, P3 n = 4543, control 1 n = 14,480, control 2 n = 9,875, control 3 n = 15,074, control n = 9,371, control 5 n = 7,542, mean ± 95% CI). f, ɣH2AX foci per nuclei (same as panel e). g, 53BP1 foci per nuclei (same as panel e). h, Predicted probability of senescence per nuclei (same as panel e). i, Nuclei with nearby SA-β-gal regions. j, Correlation between predicted senescence and nearby SA-β-gal (left axis) and number of nuclei (right axis) with increasing thresholds. k, DAPI intensities (same as panel e). l, Representative micrographs of senescent murine astrocytes and neurons; scale bar, 10 μm. m, Nuclear area of murine astrocytes (IR n = 4,888, control n = 13,549; mean ± 95% CI, Tukey multiple comparison). n, Nuclear area of murine neurons (IR n = 62,847, control n = 33,303; mean ± 95% CI, Tukey multiple comparison). o, ɣH2AX foci per nuclei (IR n = 2,016, control n = 5,202; mean ± 95% CI, Tukey multiple comparison). p, 53BP1 foci per nuclei (same as panel o). q, ɣH2AX foci per nuclei (IR n = 63,241, control n = 33,533; mean ± 95% CI, Tukey multiple comparison). r, 53BP1 foci per nuclei (same as panel q). s, Predicted senescence (same as m). t, Predicted senescence (same as panel q). PCC, Pearson correlation coefficient.CC BY · PubMed Central
Fig. 5 Senescence can be predicted across tissues and species. a, Predicted senescence for murine skin nuclei by EdU state (n = 7, mean ± 95% CI). p(sen), predicted senescence score. b, Predicted senescence for murine testis nuclei by EdU state (n = 4, mean ± 95% CI). p(sen), predicted senescence score. c, Mean probability of predicted RS senescence by p21 state across thresholds (mean ± 95% CI). p(RS), predicted RS senescence score. d, Mean probability of predicted IR senescence by p21 state across thresholds (mean ± 95% CI). p(IR), predicted IR senescence score. e, Analysis workflow. f, Nuclear area (n = 5 mice per group, mean ± 95% CI, Wald test with t-distribution). g, Nuclear convexity (same as panel f). h, Nuclear aspect ratio (same as panel f). i, Prediction percent for RS senescent (same as panel f). j, Prediction percent for IR senescence (same as panel f). k, PCNA intensity after senescence induction in three fibroblast cell lines (Doxo n = 30,957, ATV/r n = 119,669, Anti n = 106,920, Ctr n = 84,449; mean ± 95% CI). Doxo, doxocyline; ATV/r, atazanavir/ritonavir; Anti, antimycin A; Ctr, control. l, Nuclei area (as in panel k). m, Nuclei convexity (same as panel k). n, Nuclei aspect ratio (same as panel k). o, DAPI intensity (same as panel k). p, Predicted probability of RS senescence (same as panel k). q, Accuracy of doxorubicin-only model. r, Accuracy of ATV/r-only model. s, Accuracy of antimycin A-only model. t, Accuracy of unified model.
Fig. 5 Senescence can be predicted across tissues and species. a, Predicted senescence for murine skin nuclei by EdU state (n = 7, mean ± 95% CI). p(sen), predicted senescence score. b, Predicted senescence for murine testis nuclei by EdU state (n = 4, mean ± 95% CI). p(sen), predicted senescence score. c, Mean probability of predicted RS senescence by p21 state across thresholds (mean ± 95% CI). p(RS), predicted RS senescence score. d, Mean probability of predicted IR senescence by p21 state across thresholds (mean ± 95% CI). p(IR), predicted IR senescence score. e, Analysis workflow. f, Nuclear area (n = 5 mice per group, mean ± 95% CI, Wald test with t-distribution). g, Nuclear convexity (same as panel f). h, Nuclear aspect ratio (same as panel f). i, Prediction percent for RS senescent (same as panel f). j, Prediction percent for IR senescence (same as panel f). k, PCNA intensity after senescence induction in three fibroblast cell lines (Doxo n = 30,957, ATV/r n = 119,669, Anti n = 106,920, Ctr n = 84,449; mean ± 95% CI). Doxo, doxocyline; ATV/r, atazanavir/ritonavir; Anti, antimycin A; Ctr, control. l, Nuclei area (as in panel k). m, Nuclei convexity (same as panel k). n, Nuclei aspect ratio (same as panel k). o, DAPI intensity (same as panel k). p, Predicted probability of RS senescence (same as panel k). q, Accuracy of doxorubicin-only model. r, Accuracy of ATV/r-only model. s, Accuracy of antimycin A-only model. t, Accuracy of unified model.CC BY · PubMed Central
Fig. 6 Nuclear morphology predicts senescence and multiple diseases in humans. a, Predicted probability of senescence by p21 state in human dermis, across thresholds (mean ± 95% CI). b, Analysis workflow. Sen, senescent. c, Nuclear area per patient (n = 148, mean ± 95% CI, Wald test with t-distribution). d, Nuclear convexity per patient (same as panel c). e, Nuclear aspect ratio per patient (same as panel c). f, Predicted percentage of RS senescence (n = 169, mean ± 95% CI, Wald test with t-distribution). g, Predicted percentage of IR senescence (same as panel f). h, Volcano plot of ICD-10 chapters based on IR senescence residuals and P values (two-sided chi-squared test). i, Volcano plot of ICD-10 chapters based on RS senescence residuals and P values (same as panel i). j, Volcano plot of ICD-10 chapters based on doxorubicin senescence residuals and P values (same as panel i). k, Volcano plot of ICD-10 chapters based on ATV/r senescence residuals and P values (same as panel i). l, Volcano plot of ICD-10 chapters based on antimycin A senescence residuals and P values (same as panel i). m, Volcano plot of ICD-10 chapters based on unified senescence residuals and P values (same as panel i). n, Disease conditions, percentage of individuals with negative or positive residuals, P value (two-sided chi-squared and Fisher’s exact tests) and risk ratio ± 95% CI. NS, not significant.
Fig. 6 Nuclear morphology predicts senescence and multiple diseases in humans. a, Predicted probability of senescence by p21 state in human dermis, across thresholds (mean ± 95% CI). b, Analysis workflow. Sen, senescent. c, Nuclear area per patient (n = 148, mean ± 95% CI, Wald test with t-distribution). d, Nuclear convexity per patient (same as panel c). e, Nuclear aspect ratio per patient (same as panel c). f, Predicted percentage of RS senescence (n = 169, mean ± 95% CI, Wald test with t-distribution). g, Predicted percentage of IR senescence (same as panel f). h, Volcano plot of ICD-10 chapters based on IR senescence residuals and P values (two-sided chi-squared test). i, Volcano plot of ICD-10 chapters based on RS senescence residuals and P values (same as panel i). j, Volcano plot of ICD-10 chapters based on doxorubicin senescence residuals and P values (same as panel i). k, Volcano plot of ICD-10 chapters based on ATV/r senescence residuals and P values (same as panel i). l, Volcano plot of ICD-10 chapters based on antimycin A senescence residuals and P values (same as panel i). m, Volcano plot of ICD-10 chapters based on unified senescence residuals and P values (same as panel i). n, Disease conditions, percentage of individuals with negative or positive residuals, P value (two-sided chi-squared and Fisher’s exact tests) and risk ratio ± 95% CI. NS, not significant.CC BY · PubMed Central
Extended Data Fig. 1 Cell culture models of senescence. a Schematic of Ionizing radiation (IR)-induced senescence. b Representative growth curve of cells undergoing replicative senescence (RS). c Representative micrographs of SA-β-gal activity and DAPI in control, IR and RS cells; scale bars, 10 μm. d Relative p21 mRNA expression levels by qPCR (n = 3, ± SEM). e Relative p16 mRNA expression by qPCR (n = 3, ± SEM). f Relative IL-6 mRNA expression by qPCR (n = 3, ± SEM). g Representative immunohistochemistry micrographs of nuclei with p21, p16, and p53 staining in control and IR cells; scale bars, 10 μm. h Distribution of p16 intensities for IR and control for three fibroblast cell lines (n = 9,196-27,716). i Distribution of p21 intensities for IR and control for three fibroblast cell lines (n = 13,678-32,730). j Distribution of p53 intensities for IR and control for three fibroblast cell lines (n = 12,844-31,100). k Cell count following irradiation (n = 3 cell lines, mean ± 95% CI). l DAPI intensities for IR, RS, and control for three fibroblast cell lines (RS n = 6,976, IR n = 19,193, control n = 68,971; mean ± 95% CI, two-sided Student’s t-test). m Gating strategy for cell cycle analysis (data shown is for 10 Gy treated fibroblasts), mn: multinucleated cells
Extended Data Fig. 1 Cell culture models of senescence. a Schematic of Ionizing radiation (IR)-induced senescence. b Representative growth curve of cells undergoing replicative senescence (RS). c Representative micrographs of SA-β-gal activity and DAPI in control, IR and RS cells; scale bars, 10 μm. d Relative p21 mRNA expression levels by qPCR (n = 3, ± SEM). e Relative p16 mRNA expression by qPCR (n = 3, ± SEM). f Relative IL-6 mRNA expression by qPCR (n = 3, ± SEM). g Representative immunohistochemistry micrographs of nuclei with p21, p16, and p53 staining in control and IR cells; scale bars, 10 μm. h Distribution of p16 intensities for IR and control for three fibroblast cell lines (n = 9,196-27,716). i Distribution of p21 intensities for IR and control for three fibroblast cell lines (n = 13,678-32,730). j Distribution of p53 intensities for IR and control for three fibroblast cell lines (n = 12,844-31,100). k Cell count following irradiation (n = 3 cell lines, mean ± 95% CI). l DAPI intensities for IR, RS, and control for three fibroblast cell lines (RS n = 6,976, IR n = 19,193, control n = 68,971; mean ± 95% CI, two-sided Student’s t-test). m Gating strategy for cell cycle analysis (data shown is for 10 Gy treated fibroblasts), mn: multinucleated cellsCC BY · PubMed Central
Extended Data Fig. 2 Performance characteristics. a Accuracy of raw model, tested on independent cell lines. b ROC curve of the raw model for independent cell lines. c Accuracy of normalized model. d ROC curve for the normalized model. e Accuracy of Xception-based Bayesian neural network. f ROC curve for the Xception BNN. g Accuracy of InceptionV3-based Bayesian neural network. h ROC curve for the InceptionV3 BNN.
Extended Data Fig. 2 Performance characteristics. a Accuracy of raw model, tested on independent cell lines. b ROC curve of the raw model for independent cell lines. c Accuracy of normalized model. d ROC curve for the normalized model. e Accuracy of Xception-based Bayesian neural network. f ROC curve for the Xception BNN. g Accuracy of InceptionV3-based Bayesian neural network. h ROC curve for the InceptionV3 BNN.CC BY · PubMed Central
Extended Data Fig. 3 Improving performance by adjusting thresholds. a Histogram of predicted probabilities for Bayesian neural network. b Histogram of predicted probabilities for deep ensemble. c Accuracy and percent of samples evaluated with different classifier thresholds for single model. d Accuracy and percent of samples evaluated with different classifier thresholds for Xception-based Bayesian neural network. e Accuracy and percent of samples evaluated with different classifier thresholds for deep ensemble. f Accuracy and percent of samples evaluated with different classifier thresholds for deep ensemble with normalized samples. g Accuracy and percent of samples evaluated with different classifier thresholds for ensemble of RS-only for normalized samples. h Accuracy and percent of samples evaluated with different classifier thresholds for ensemble of IR-only for normalized samples.
Extended Data Fig. 3 Improving performance by adjusting thresholds. a Histogram of predicted probabilities for Bayesian neural network. b Histogram of predicted probabilities for deep ensemble. c Accuracy and percent of samples evaluated with different classifier thresholds for single model. d Accuracy and percent of samples evaluated with different classifier thresholds for Xception-based Bayesian neural network. e Accuracy and percent of samples evaluated with different classifier thresholds for deep ensemble. f Accuracy and percent of samples evaluated with different classifier thresholds for deep ensemble with normalized samples. g Accuracy and percent of samples evaluated with different classifier thresholds for ensemble of RS-only for normalized samples. h Accuracy and percent of samples evaluated with different classifier thresholds for ensemble of IR-only for normalized samples.CC BY · PubMed Central
Extended Data Fig. 4 Development of Senescent Phenotype. a Predicted probability of senescence for several time points of IR-induced senescent cells for three fibroblast cell lines (n = 35,191-106,549). b Distribution of predicted probability split by time points. c Mean PCNA intensity per nucleus for several time points for three fibroblast cell lines (n = 35,191-106,549). d Mean p21 intensity per nucleus for several time points for three fibroblast cell lines (n = 35,191-106,549). e Mean DAPI intensity per nucleus for several time points for three fibroblast cell lines (n = 35,191-106,549). f p21 for several time points, split by predicted senescence state (p(sen)<0.1 n = 15,011, p(sen)>0.9 n = 6626; mean ± 95% CI). g PCNA for several time points, split by predicted senescence state (same as f). h DAPI for several time points, split by predicted senescence state (same as f). i Area for several time points, split by predicted senescence state (same as f). j Convexity for several time points, split by predicted senescence state (same as f). k Aspect for several time points, split by predicted senescence state (same as f). l Cell counts for several time points.
Extended Data Fig. 4 Development of Senescent Phenotype. a Predicted probability of senescence for several time points of IR-induced senescent cells for three fibroblast cell lines (n = 35,191-106,549). b Distribution of predicted probability split by time points. c Mean PCNA intensity per nucleus for several time points for three fibroblast cell lines (n = 35,191-106,549). d Mean p21 intensity per nucleus for several time points for three fibroblast cell lines (n = 35,191-106,549). e Mean DAPI intensity per nucleus for several time points for three fibroblast cell lines (n = 35,191-106,549). f p21 for several time points, split by predicted senescence state (p(sen)<0.1 n = 15,011, p(sen)>0.9 n = 6626; mean ± 95% CI). g PCNA for several time points, split by predicted senescence state (same as f). h DAPI for several time points, split by predicted senescence state (same as f). i Area for several time points, split by predicted senescence state (same as f). j Convexity for several time points, split by predicted senescence state (same as f). k Aspect for several time points, split by predicted senescence state (same as f). l Cell counts for several time points.CC BY · PubMed Central
Extended Data Fig. 5 DNA Damage Foci and Other Senescence Indicators. a Representative immunohistochemistry micrographs of nuclei with DNA damage foci staining of gH2AX and 53BP1; scale bars, 10 μm. b 2D histogram (density heatmap) of predicted senescence and foci count per senescence type. c 2D histogram (density heatmap) of predicted senescence and foci count for premature aging diseases. d 2D histogram (density heatmap) of predicted senescence and foci count for murine astrocytes. e Predicted senescence of skin nuclei for high and low DAPI (DAPI low n = 7, DAPI high n = 7; mean ± 95% CI, two-sided Student’s t-test). f Predicted senescence of testis nuclei for high and low DAPI (DAPI low n = 4, DAPI high n = 4; mean ± 95% CI, two-sided Student’s t-test). g Predicted probability for RS senescence (n = 5, mean ± 95% CI, Wald Test with t-distribution). h Predicted probability for IR senescence (same as g).
Extended Data Fig. 5 DNA Damage Foci and Other Senescence Indicators. a Representative immunohistochemistry micrographs of nuclei with DNA damage foci staining of gH2AX and 53BP1; scale bars, 10 μm. b 2D histogram (density heatmap) of predicted senescence and foci count per senescence type. c 2D histogram (density heatmap) of predicted senescence and foci count for premature aging diseases. d 2D histogram (density heatmap) of predicted senescence and foci count for murine astrocytes. e Predicted senescence of skin nuclei for high and low DAPI (DAPI low n = 7, DAPI high n = 7; mean ± 95% CI, two-sided Student’s t-test). f Predicted senescence of testis nuclei for high and low DAPI (DAPI low n = 4, DAPI high n = 4; mean ± 95% CI, two-sided Student’s t-test). g Predicted probability for RS senescence (n = 5, mean ± 95% CI, Wald Test with t-distribution). h Predicted probability for IR senescence (same as g).CC BY · PubMed Central
Extended Data Fig. 6 Senescence metrics in human dermal fibroblasts. a Sample of human dermis, stained with H&E and DAB for p21; scale bars, 10 μm. b Predicted probability of senescence from IR model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). c Predicted probability of senescence from Doxorubicin model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). d Predicted probability of senescence from ATV/r model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). e Predicted probability of senescence from antimycin-A model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). f Predicted probability of RS senescence (n = 169, mean ± 95% CI, Wald Test with t-distribution). g Predicted probability of IR senescence (same as f). h Predicted probability of doxorubicin senescence (same as f). i Predicted probability of ATV/r senescence (same as f). j Predicted probability of antimycin-A senescence (same as f). k Percent of doxorubicin senescence (same as f). l Percent of ATV/r senescence (same as f). m Percent of antimycin-A senescence (same as f). n 2D histograms (density heatmaps) of dermal nuclei.
Extended Data Fig. 6 Senescence metrics in human dermal fibroblasts. a Sample of human dermis, stained with H&E and DAB for p21; scale bars, 10 μm. b Predicted probability of senescence from IR model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). c Predicted probability of senescence from Doxorubicin model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). d Predicted probability of senescence from ATV/r model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). e Predicted probability of senescence from antimycin-A model and number of nuclei for p21- and p21+ nuclei in human dermis (mean ± 95% CI). f Predicted probability of RS senescence (n = 169, mean ± 95% CI, Wald Test with t-distribution). g Predicted probability of IR senescence (same as f). h Predicted probability of doxorubicin senescence (same as f). i Predicted probability of ATV/r senescence (same as f). j Predicted probability of antimycin-A senescence (same as f). k Percent of doxorubicin senescence (same as f). l Percent of ATV/r senescence (same as f). m Percent of antimycin-A senescence (same as f). n 2D histograms (density heatmaps) of dermal nuclei.CC BY · PubMed Central
Extended Data Fig. 7 Additional Metrics and Clinical Codes. a Correlation coefficient matrix of metrics from nuclei. b Contour distribution of aspect versus area of nuclei for individuals under 40. c Contour distribution of aspect versus area of nuclei for individuals over 60. d Volcano plot of conditions based on IR senescence residuals and p values (two-sided Fisher’s exact test). e Volcano plot of conditions based on RS senescence probability residuals and p values (two-sided Fisher’s exact test). f Volcano plot of conditions based on doxorubicin senescence probability residuals and p values (two-sided Fisher’s exact test). g Volcano plot of conditions based on ATV/r senescence probability residuals and p values (two-sided Fisher’s exact test). h Volcano plot of conditions based on antimycin-A senescence probability residuals and p values (two-sided Fisher’s exact test). i Volcano plot of conditions based on probability of unified senescence probability residuals and p values (two-sided Fisher’s exact test). j Volcano plot of conditions based on percent IR senescence residuals and p values (two-sided Fisher’s exact test). k Volcano plot of conditions based on percent RS senescence residuals and p values (two-sided Fisher’s exact test). l Volcano plot of conditions based on percent doxorubicin senescence residuals and p values (two-sided Fisher’s exact test). m Volcano plot of conditions based on percent ATV/r senescence residuals and p values (two-sided Fisher’s exact test). n Volcano plot of conditions based on percent antimycin-A senescence residuals and p values (two-sided Fisher’s exact test). o Volcano plot of conditions based on percent unified senescence residuals and p values (two-sided Fisher’s exact test).
Extended Data Fig. 7 Additional Metrics and Clinical Codes. a Correlation coefficient matrix of metrics from nuclei. b Contour distribution of aspect versus area of nuclei for individuals under 40. c Contour distribution of aspect versus area of nuclei for individuals over 60. d Volcano plot of conditions based on IR senescence residuals and p values (two-sided Fisher’s exact test). e Volcano plot of conditions based on RS senescence probability residuals and p values (two-sided Fisher’s exact test). f Volcano plot of conditions based on doxorubicin senescence probability residuals and p values (two-sided Fisher’s exact test). g Volcano plot of conditions based on ATV/r senescence probability residuals and p values (two-sided Fisher’s exact test). h Volcano plot of conditions based on antimycin-A senescence probability residuals and p values (two-sided Fisher’s exact test). i Volcano plot of conditions based on probability of unified senescence probability residuals and p values (two-sided Fisher’s exact test). j Volcano plot of conditions based on percent IR senescence residuals and p values (two-sided Fisher’s exact test). k Volcano plot of conditions based on percent RS senescence residuals and p values (two-sided Fisher’s exact test). l Volcano plot of conditions based on percent doxorubicin senescence residuals and p values (two-sided Fisher’s exact test). m Volcano plot of conditions based on percent ATV/r senescence residuals and p values (two-sided Fisher’s exact test). n Volcano plot of conditions based on percent antimycin-A senescence residuals and p values (two-sided Fisher’s exact test). o Volcano plot of conditions based on percent unified senescence residuals and p values (two-sided Fisher’s exact test).CC BY · PubMed Central
Extended Data Fig. 8 Table of Significant Conditions and Additional Conditions. Table of conditions with significant p values from two-sided chi-squared test (neoplasms and malignancy) or two-sided Fisher’s test (all others). Not corrected for multiple comparisons. * indicates negative association, others are positive.
Extended Data Fig. 8 Table of Significant Conditions and Additional Conditions. Table of conditions with significant p values from two-sided chi-squared test (neoplasms and malignancy) or two-sided Fisher’s test (all others). Not corrected for multiple comparisons. * indicates negative association, others are positive.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

2 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 55 references

Cited by

See all 246

Metadata

Authors
Heckenbach, I., Mkrtchyan, G. V., Ezra, M. B., Bakula, D., Madsen, J. S., Nielsen, M. H., Oró, D., Osborne, B., Covarrubias, A. J., Idda, M. L., Gorospe, M., Mortensen, L., Verdin, E., Westendorp, R., & Scheibye-Knudsen, M.
Affiliations
  • University of Copenhagen (DK)
  • Gubra (Denmark) (DK)
  • University of California, Los Angeles (US)
  • Institute of Genetic and Biomedical Research (IT)
  • National Institute on Aging (US)
  • Buck Institute for Research on Aging (US)
Venue
Nature Aging
Published
2022-08-15
Volume
2
Issue
8
Pages
742-755
ISSN
2662-8465 (ISSN-L)
OpenAlex
W4291793058
Cited by
232
References
55
Citations by year
2026: 59 · 2025: 63 · 2024: 67 · 2023: 36 · 2022: 6
FWCI
26.63
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Aging (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion)
  • Research Square (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Research at the University of Copenhagen (University of Copenhagen) (publishedVersion, cc-by)
  • UnissResearch (Università degli Studi di Sassari) (submittedVersion, cc-by)
Funders
  • U.S. Department of Health & Human Services | NIH | Office of Strategic Coordination (U54 AG075932)
  • Novo Nordisk Fonden (NNF17OC0027812)
  • Lundbeckfonden (R324-2019-1492)
  • Ministry of Higher Education and Science | Forskerakademiet (0238-00003B)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Cell Biology › Telomeres, Telomerase, and Senescence
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Aging › Genetics, Aging, and Longevity in Model Organisms
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Biophysics › Cell Image Analysis Techniques
Keywords
cellular senescence, nuclear morphology, human fibroblasts, astrocytes, neurons, premature aging, p21, senescent cells, aging, age-related diseases, osteoporosis, osteoarthritis, hypertension, cerebral infarction, cancer risk, liver tissue
MeSH terms
Deep Learning, Aging, Animals, Humans, Biomarkers, Cellular Senescence, Aging, Premature, Mice

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Nuclear morphology is a deep learning biomarker of cellular senescence. https://app.cochat.ai/papers/10.1038/s43587-022-00263-3
The paper
Heckenbach, I., Mkrtchyan, G. V., Ezra, M. B., Bakula, D., Madsen, J. S., Nielsen, M. H., Oró, D., Osborne, B., Covarrubias, A. J., Idda, M. L., Gorospe, M., Mortensen, L., Verdin, E., Westendorp, R., & Scheibye-Knudsen, M. (2022). Nuclear morphology is a deep learning biomarker of cellular senescence. Nature Aging. https://doi.org/10.1038/s43587-022-00263-3

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.