Nuclear morphometrics coupled with machine learning identifies dynamic states of senescence across age

Abstract

Cellular senescence is an irreversible state of cell cycle arrest with a complex role in tissue repair, aging, and disease. However, inconsistencies in identifying cellular senescence have led to varying conclusions about their functional significance. We developed a machine learning-based approach that uses nuclear morphometrics to identify senescent cells at single-cell resolution. By applying unsupervised clustering and dimensional reduction techniques, we built a robust pipeline that distinguishes senescent cells in cultured systems, freshly isolated cell populations, and tissue sections. Here we show that this method reveals dynamic, age-associated patterns of senescence in regenerating skeletal muscle and osteoarthritic articular cartilage. Our approach offers a broadly applicable strategy to map and quantify senescent cell states in diverse biological contexts, providing a means to readily assess how this cell fate contributes to tissue remodeling and degeneration across lifespan.

Fig. 1. Senescent conversion alters nuclear morphology.

A Schematic of the nuclear morphometric pipeline (NMP) developed to identify senescence (Created in BioRender. Mapkar, S. (2025) https://BioRender.com/w2xaf4w). B H₂O₂ treatment of C2C12 cells induces a senescent phenotype in a dose-dependent manner. MFI – mean fluorescence intensity. n = 3 independent experimental replicates. C Representative FACS histogram for Spider SA-β-Gal staining. The graph quantifies replicates relative to untreated and unstained. Blue – untreated, Red – 300 µM, Orange – 500 µM. n = 3 independent experimental replicates. D SASP assessment by qRT-PCR. n = 4 independent experimental replicates, where each column represents one replicate. Green – increased expression, Orange – decreased expression, White – equivalency to untreated. E Representative images and quantification of the nuclear morphology between treatment groups. Note, image Look-Up Tables (LUTs) and exposure times are equivalent. Scale bar – 50 μm. n = 3 independent experimental replicates. Error bars represent the mean ± standard error of the mean (SEM), and statistical tests were conducted using unpaired two-sided t tests with no adjustments. *p ≤ 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Source data are provided as a Source Data file. Precise p-values are listed in Supplementary Table 2.

Fig. 2. Nuclear morphometrics define senescence and can be reduced to a unifying score.

A A UMAP of reduced single-cell nuclear morphology with 5 groups resolved by the k-means clustering algorithm. Inserts are representative nuclei from each cluster of H₂O₂ treated C2C12 cells. Image Look-Up Tables (LUTs) and exposure times are equivalent. Scale bars – 5 μm. B Further reduction creates a one-dimensional senescent “score” correlating with nuclear phenotype and treatment condition. C Orange (senescent) and red (senescent like) clusters exhibit a strong senescence phenotype and are quantified in the bar graph. n = 3 independent experimental replicates. D Quantification of senescent-associated markers in clusters identified as senescent by the NMP. MFI – mean fluorescence intensity; NS – non-senescent; SEN – senescent. n = 3 independent experimental replicates. E Representative UMAPs and a line graph quantifying the decrease in SnCs with increasing senolytic concentration, as detected by the NMP. Gray – untreated, Blue – 300 µM, Green – 500 µM. Violin plots visualize senescent score (y-axis) to demonstrate decrease in the density of low-scoring, morphologically senescent nuclei, as shown inside the red box. Comparisons are made within treatment groups. n = 3 independent experimental replicates. Error bars represent the mean ± standard error of the mean (SEM), and statistical tests were conducted using unpaired two-sided t tests with no adjustments except for (E), which used a two-way ANOVA.*p ≤ 0.05, **p < 0.01, ***p < 0.001. Source data are provided as a Source Data file. Precise p-values are listed in Supplementary Table 2.

Fig. 3. The NMP is applicable across varying mechanisms of senescence induction.

A,B Quantification of senescence markers following (A) etoposide and (B) doxorubicin treatment of C2C12s. n = 3 independent experimental replicates. C, D Representative images and quantification of nuclear morphometrics of C2C12s treated with (C) etoposide - Etop and (D) doxorubicin - Dox. Note the dosage response to each small molecule inducer. Exposure times and Look-Up Tables (LUTs) are held constant. Scale bar – 200 μm. n = 3 independent experimental replicates. E, F Large UMAPs consisting of reduced single-cell nuclear morphology with 5 groups resolved by the k-means clustering algorithm for (E) etoposide and (F) doxorubicin. Inserts are representative nuclei from each cluster. Scale bars – 20 μm. One-dimensional senescent “score,” in small UMAPs, correlating with nuclear phenotype and treatment condition, shows dose-dependent quantification in violin plots. Orange (senescent) and red (senescent like) clusters exhibit a strong senescence phenotype and are quantified in the horizontal color bar graph. n = 3 independent experimental replicates. Gray bar graphs are the quantification of senescent-associated markers in clusters identified as senescent by the NMP. n = 3 independent experimental replicates. MFI – mean fluorescence intensity. NS – non-senescent; SEN – senescent. Error bars represent the mean ± standard error of the mean (SEM), and statistical tests were conducted using unpaired two-sided t-tests with no adjustments. *p ≤ 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Source data are provided as a Source Data file. Precise p-values are listed in Supplementary Table 2.

Fig. 4. The NMP identifies a dynamic state of senescence in regenerating skeletal muscle.

A Schematic demonstrating the experimental approach (Created in BioRender. Mapkar, S. (2025) https://BioRender.com/mxe0n7o). B–D Fibroadipogenic progenitors - FAPs, E–G: Satellite Cells - SCs. B, E Representative UMAPs reducing nuclear phenotypes for the noted experimental conditions and the NMP-derived senescent score. Representative associated heat map displays phenotypic differences at noted days post injury (DPI), Red – increased from uninjured, Blue – decreased from uninjured. The representative associated violin plot visualizes differences in score. C, F Representative UMAPs defining the age-dependent cell representation in the senescent population. Circle plots quantify the relative representation of cells per age in the SnC cluster. Blue – young, Red – aged, Green – geriatric. D, G Line graphs show the age-dependent changes for the noted DPI, and the bar graphs denote those of 3 DPI. Significance is relative to the untreated. n = 3 independent biological replicates per age. Teal – Uninjured, Purple – 3 DPI, Orange – 21 DPI. H Visualization of which cell populations contribute to total senescence during SkM injury at different age groups. Enriched Endothelial Cells – Enr. ECs. Immune Cells – ICs. All UMAP cell numbers from each condition are equivalent. Error bars represent the mean ± standard error of the mean (SEM), and statistical tests were conducted using unpaired two-sided t-tests with no adjustments. **p < 0.01, ***p < 0.001, ****p < 0.0001. Source data are provided as a Source Data file. Precise p-values are listed in Supplementary Table 2.

Fig. 5. The NMP can effectively identify FAPs with senescent characteristics in vivo in the regenerating muscles of young and geriatric mice.

A, D Representative images of geriatric skeletal muscle (SkM) cross-sections identifying PDGFRα + FAPs, Laminin+ encircled muscle fibers, (A) Ki67 + proliferating cells or (D) γH2A + damage, and DAPI to identify and resolve nuclear morphology. B, E Quantification of (B) proliferating or (E) damaged FAPs relative to young. Violin plots quantify the noted marker per single cell (each dot). MFI – mean fluorescence intensity. n = 3 independent biological replicates per age. C, F UMAPs showing the clustering of single FAPs by the NMP with senescent scores indicated by color (lower scores – purple – are highly senescent cells), and expression level of the noted marker by the size of the circle. Bar graphs quantify the amount of the noted marker in the non-SnC (NS) or SnCs (SEN). n = 3 independent biological replicates per age. Scale bars – 50 μm, inset scale bars – 10 μm. Error bars represent the mean ± standard error of the mean (SEM) and statistical tests were conducted using unpaired two-sided t-tests with no adjustments. *p ≤ 0.05, **p < 0.01, ***p < 0.001. Source data are provided as a Source Data file. Precise p-values are listed in Supplementary Table 2.