AI-assisted imaging screening reveals mechano-molecular tissue organizers and network of signaling hubs.
Cristina Bertocchi,Juan Jose Alegria,Sebastian Vasquez-Sepulveda,Rosario Ibanez-Prat,Aishwarya Srinivasan,Ignacio Arrano-Valenzuela,Barbara Castro-Pereira,Catalina Soto-Montandon,Alejandra Trujillo-Espergel,Gareth I. Owen,Pakorn Kanchanawong,Mauricio Cerda,Giovanni Motta,Ronen Zaidel-Bar,Andrea Ravasio
DOI: https://doi.org/10.1101/2024.11.14.623670
2024-11-15
Abstract:Cadherin-mediated adhesions are crucial mechanical and signaling hubs that connect cells within a tissue and probe the mechanics of the surrounding environment. They constitute a physical link between the actin cytoskeleton of neighboring cells, providing the mechanical coordination needed for morphogenetic processes, tissue homeostasis, collective migration, and regeneration. Disruptions in adhesion mechanisms are closely linked to the breakdown of epithelial structure and the emergence of disease-related traits characteristic of cancer progression. The cadhesome network comprises over 170 structural and regulatory proteins involved in cadherin-mediated adhesion. While this network is essential for coordinating tissue responses to mechanical stress, its complexity has historically limited our understanding of how individual components contribute to force transmission and tissue homeostasis. Recent technological advances offer tools to investigate large molecular networks in cellular function and pathology (functional omics). Leveraging these advances, we developed an experimental and analytical platform combining high-throughput gene silencing, imaging, and artificial intelligence (AI) to systematically profile each role of each protein in tissue formation, mechanical stability, and response to induced tension. Using EpH4 cells as an epithelial tissue model, we performed systematic silencing in triplicate, capturing a range of tissue phenotypes under baseline and tension-inducing conditions. Machine learning methods were used to analyze complex imaging data, quantify tissue ruptures, characterize junctional organization, and measure tension states of the tissue. By incorporating machine learning algorithms, we automated image feature extraction, clustering, and classification, enabling an unprecedented quantitative evaluation of tissue mechanics at scale. Our machine learning models allowed us to identify significant patterns, including protein-specific responses to tension and their roles in tissue-level mechanical integrity. Finally, we constructed a protein interaction network detailing the roles of each protein, their physical interactions, and known links to cancer. The network analysis revealed three prominent mechanotransductive and signaling subnetworks centered around E-cadherin, EGFR, and RAC1. Our study provides a foundational framework for investigating mechanosensing proteins and it offers a scalable blueprint for discovering potential therapeutic targets in diseases like cancer, where tissue mechanics play a crucial role.
Cell Biology