Poster U13 - Single-cell morphology encodes functional subtypes of senescence in aging human dermal fibroblasts

Introduction: Biological aging is described as the continuous degradation of physiological function that eventually drives age-associated disease and frailty. Cellular senescence is identified as a key driver of adverse aging patterns, linking defective cell-based changes to physiological dysfunctions. Cellular senescence is a state of stress-induced cell cycle arrest typically accompanied by increased expression of CDK inhibitors—e.g., p16 and p21, increased secretion of proinflammatory factors, and a larger cell and nuclei size when cultured in vitro. However, current studies are limited, as they treat senescence as binary (i.e., senescent or not), use few senescence-associated biomarkers to classify senescence, and are unable to discriminate between the heterogeneous and potentially functional subtypes of senescence. To address this, we hypothesized that subtypes of senescent cells with distinct functional behaviors exist and can be identified and classified based on cellular morphologies. These functional subtypes provide valuable insights to: (1) elucidate the temporal evolution of the senescent phenotype, (2) understand the association between abundance of senescent subtypes and age-associated pathologies, and (3) identify novel therapies that appreciate senescence heterogeneity to combat disease.

Materials and

Methods: We induced senescence in primary human dermal fibroblasts (aged 23 and 85) by exposing to induction agents that include doxorubicin, bleomycin, atazanavir, hydrogen peroxide, and replication using serial passaging. Following induction, cells were stained for combinations of two senescence-associated biomarkers (p21, p16, LMNB1, HMGB1, Beta Galactosidase) for time points of 2-, 5-, 8-,12-, and 15-days post induction. Cells/Nuclei were segmented using a combination of computational and deep learning approaches, and the dataset was subsequently clustered based on morphological features computed using custom software to identify subtypes of senescence. Morphological subtypes were further validated with an in-house Xception convolutional neural network (CNN). These subtypes were screened through a panel of commercially available senotherapies to probe differential response modalities.

Results, Conclusions, and Discussions: Senescent cells exist in a heterogeneous collection of morphological clusters that correlate with their senescence-associated protein biomarker expression (Figure 1a-c). We found that the abundance of cells within each of the morphological clusters depended on the mode of induction, age of the sample, and time after induction, highlighting senescence as a dynamic ensemble of phenotypes (Figure 1d). Similarly, senescence associated protein biomarkers expressed unique temporal activity- (turning “on” at different times post induction) dependent on induction agent. Morphological clusters were supplemented with the CNN scoring platform which provided quantitative assessment of senescent burden at true single-cell resolution and identified three senescence morphological subtypes. Importantly, senescence subtypes expressed differential speeds and extents of clearance to senotherapeutic agents (Figure 1e-f).
In this work, we establish a novel biophysical framework to elicit functional behavior of senescent cells at single cell-resolution. We can comprehensively evaluate the senescence burden of cells, classify their morphological subtype, and interface cellular morphology with senescence-associated protein biomarker behavior and therapeutic response. We are currently in the process of ascribing further functional readouts to morphological subtypes based on spatial molecular technologies.

Acknowledgements (Optional):