From imaging to computational domains for physics-driven molecular biology simulations: Hindered diffusion in platelet masses

When formed in vivo, murine hemostatic thrombi exhibit a heterogeneous architecture comprised of distinct regions of densely and sparsely packed platelets. In this study, we utilize high-resolution electron microscopy alongside machine learning and physics-based simulations to investigate how such clot microstructure impacts molecular diffusivity. We used Serial Block Face – Scanning Electron Microscopy (SBF-SEM) to image select volumes of hemostatic masses formed in a mouse jugular vein, producing high-resolution 2D images. Images were segmented using machine learning software (Cellpose), whose training was augmented by manually segmented images. The segmented images were then utilized as 2D computational domains for Lattice Kinetic Monte-Carlo (LKMC) simulations. This process constitutes a computational pipeline that combines purely data-derived biological domains with physics-driven simulations to estimate how molecular movement is hindered in a hemostatic platelet mass. Using our pipeline, we estimated that the 2D hindered diffusion rates of a globular protein range from 2% to 40% of the unhindered rate, with denser packing regions lending to lower molecular diffusivity. These data suggest that coagulation reactions rates, thrombin generation and activity, as well as platelet releasate activity may be drastically impacted by the internal geometry of a hemostatic thrombus.Author summary: Hemostasis and coagulation are two exquisitely complex, intertwined, and tightly regulated biological processes. Dysregulation of either process may lead to severe health consequences or death. Coagulation reactions have been extensively studied under static laboratory conditions, which are different from in vivo conditions. It is therefore imperative to understand if and how the chemical reactions underlying coagulation are regulated by the environment where they occur. In vivo experimentation enables us to image hemostasis, but not chemical reactions. Physics-driven molecular simulations of chemical reactions can bridge the gap, provided the physical environment is correctly represented computationally. The present work serves as a much-needed foundation for image-to-computation for physics based molecular simulations in biological domains.