(G-247) Dynamic Feature-Based Approach for Categorizing AVM Endothelial Cells

Introduction:: Cellular heterogeneity is a ubiquitous feature of biological systems that results from the existence of diverse cell phenotypes at multiple levels of organization, ranging from single cells to tissues and organs These cell phenotypes are characterized by a range of biophysical features, encompassing genetic makeup, morphology, behavior, and functionality. Heterogeneity in cellular phenotypes is prevalent in various diseases, including Arteriovenous malformations (AVMs), a type of vascular anomaly that can lead to severe cardiovascular complications. Recent research has emphasized the significance of angiogenic potential in the development and treatment of AVMs, as AVM endothelial cells exhibit considerable genetic mutations in angiogenesis signaling pathways. Since angiogenesis is closely associated with dynamic behaviors such as cellular migration and proliferation, a quantitative approach is necessary for classifying heterogeneous cell phenotypes based on their dynamic features. In this study, we propose a dynamic feature-based approach to classify heterogeneous cell phenotypes, which accounts for time-variability and the number of division cycles. We apply this approach to classify AVM endothelial cell phenotypes and group them into subcategories based on their angiogenic potential.


Materials and Methods:: AVM endothelial cells were obtained from the School of Medicine at Kyungpook National University. We captured time-lapse images of AVM endothelial cells every 30 minutes for 48 hours and utilized a deep neural network model called Cellpose (Stringer C et al., 2021) with fine-tuning to predict cell masks. We then tracked and hierarchically labeled the doubling lineage, a linear sequence from a parent cell to a daughter cell (Figure 1.(b)) using customized codes referred to as Jaqaman et al.(2008). We extracted dynamic features comprising 123 morphology variation features and 50 motility features from a linear mask sequence along the doubling lineage. We assessed the angiogenic potential by measuring the number of divisions and average migration speed along the doubling lineages. Based on the number of divisions, we identified the state for the divisions as Doubling_1, Doubling_2, Doubling_3, Doubling_4, and Doubling_no. We divided the average migration speed into high, middle, and low-migration speed states and identified 15 properties of angiogenic potential by combining each subdivided state. We utilized PCA of dynamic features and K-means clustering algorithm to classify AVM endothelial cells into 6 phenotypes based on dynamic features. These 6 phenotypes were then categorized into 3 subgroups, high-angiogenic condition, intermediate-angiogenic condition, and low-angiogenic condition, based on the distribution of angiogenic potential for doubling lineages in each phenotype by the hierarchical clustering algorithm.


Results, Conclusions, and Discussions:: PCA found that the variance of solidity and the positive tendency of aspect ratio are the two dynamic features that best discriminate the different cell phenotypes in 2-dimensional space. The positive tendency is defined as being proportional to the frequency and speed of morphological changes, and a higher solidity value indicates a star-like shape. We applied K-means clustering to the principal components and identified 6. Phenotypes 3&4 exhibited a high positive tendency of aspect ratio, while phenotypes 2&5 and phenotypes 1&6 exhibited intermediate and low positive tendencies, respectively. We also observed that phenotypes 1&4&5 exhibited high variance of solidity, while phenotypes 2&3&6 exhibited low variance of solidity. Using hierarchical clustering, we categorized the 6 phenotypes into 3 subgroups: high-angiogenic, intermediate-angiogenic, and low-angiogenic conditions, based on their distribution of angiogenic potential. The phenotypes under the high-angiogenic condition exhibit a higher number of divisions and migration speeds than those under the intermediate-angiogenic and low-angiogenic conditions. For those under intermediate-angiogenic condition, the distribution of angiogenic potential is relatively even across intermediate levels of both the number of divisions and migration speeds. However, under the low-angiogenic condition, phenotype 1 tends to have non-mitotic and non-actively migratory behavior.


This study presents a quantitative approach for categorizing heterogeneous cell phenotypes based on their dynamic features, which has been applied to AVM endothelial cells. This approach provides a more comprehensive understanding of the dynamic behaviors of AVM endothelial cells compared to traditional approaches that rely solely on morphology or gene profiles, by accounting dynamic features including cell division and migration. Our categorization result implies that there may be a mechanism regulating angiogenesis in AVM development that is specific to different phenotypes with their angiogenic potential. These findings have implications for the understanding and treatment of AVMs, and may provide new insights into the underlying mechanisms of AVM development. Furthermore, this approach may have potential applications for the study of other diseases.




Acknowledgements (Optional): : This research was supported by National Research Funding granted by the Korean Government (NRF-2021R1A2C3008408).

References (Optional): : [1] Wu, Zhenqin et al. “DynaMorph: self-supervised learning of morphodynamic states of live cells.” Molecular biology of the cell vol. 33,6, 2022

[2] Wu, Pei-Hsun et al. “Single-cell morphology encodes metastatic potential.” Science advances vol. 6,4 eaaw6938. 22 Jan. 2020