Cellular senescence is a heterogeneous and complex process caused by altered morphology, the synthesis of inflammatory factors, and sustained proliferative arrest. Typical senescence research focuses on biomarkers, which limits our comprehensive understanding of biology. Advances in single-cell technology have revealed a wide range of senescent subtypes. This study aimed to provide a comprehensive explanation of the functional and heterogeneous diversity of senescent cells by using a senescence subtype classifier based on observable unique phenotypes (SenSCOUT). An integrated imaging and machine learning framework is used to classify and profile the senescence subtypes at the single-cell level.
This study used primary dermal fibroblasts from young and aged donors and exposed them to various senescence-inducing conditions, including atazanavir, bleomycin (BLEO), hydrogen peroxide (H2O2), doxorubicin (DOX), and replicative exhaustion. The researchers used high-content imaging to visualize nuclear and cellular morphology, and then performed staining for typical senescence biomarkers, including β-galactosidase, p16, p21, Lamin B1 (LMNB1), and High Mobility Group Box 1 (HMGB1), after 8 days of development. About 50,000 cells were analyzed to confirm the induction of senescence caused by increased expression of senescence markers, such as β-galactosidase, p16, and p21, and reduced levels of proliferative markers, including EdU and Ki-67. The senescence-associated secretory phenotype (SASP) was apparent, and increased levels of SASP components were found in the senescent cells.
By using methods to decrease dimensionality, such as Uniform Manifold Approximation and Projection (UMAP), and clustering algorithms like k-means. It discovered unique senescence-associated clusters differentiated by changes in cellular morphology. These clusters contain completely non-senescent cells, senescent cells, and intermediate phases. Nuclear morphology was a useful and distinctive imputation method based on single-cell morphology, which yields 85% accuracy in predicting senescence biomarker expression. The researchers also trained a machine learning model using an Xception convolutional neural network, which achieved 89% accuracy in classifying non-senescent and senescent cells. The model used nuclear and cellular morphology to predict senescence. The nuclear and cell size showed a strong correlation with senescence. Senescence scores were obtained using a threshold, and various subtypes of senescent cells were found, like clusters C10, C7, and C11. These clusters exhibited different responses based on senescence-inducing conditions and donor age.
The study showed a significant correlation between chronological age and higher senescence scores, indicating that older cells tend to accumulate a greater senescence burden. Cells with a higher baseline senescence score were found to be more prone to further senescence induction, specifically when treated with doxorubicin (DOX). The researchers delved into the potential of senotherapies, specifically examining the impact of a senolytic drug combination, such as dasatinib (D) and quercetin (Q), on different types of senescence. The results showed that these subtypes exhibited varying sensitivities to the D+Q treatment, with cluster C7 showing a substantial reduction in cell viability.
The study used one-way ANOVA and post hoc Tukey tests for statistical comparisons, and linear regression correlations were analyzed using R. Data normalization and log transformation were applied. Experiments were performed with at least three biological and two technical replicates. Software used included CellProfiler for single-cell segmentation, TrackPy for dynamic tracking, and Python for data analysis.
In conclusion, this research introduces a comprehensive framework for profiling cellular senescence based on single-cell morphological features and machine learning. It gives a more detailed understanding of senescence as a process that differs by subtype and is influenced by biological factors and therapeutic interventions. This approach holds great promise for advancing aging research and developing targeted senotherapies.
Reference: Kamat P, Macaluso N, Li Y, et al. Single-cell morphology encodes functional subtypes of senescence in aging human dermal fibroblasts. Sci Adv. 2025;11(17):eads1875. doi:10.1126/sciadv.ads1875
