Patient-Specific Predictions of Thrombosis

Platelet aggregation is an important regulator of hemostatic function in response to a blood vessel injury.  Overactivity of platelet aggregation in situations such as atherosclerotic plaque rupture can cause occlusion of the vessel, while underactivity results in excessive bleeding.  Several complex biological and physical mechanisms contribute to the growth of a stable platelet mass at the site of injury while limiting the final extent of growth to prevent occlusion.  A detailed model of this system would provide insight on the coupling between biological signaling and fluid flow as well as predict strategies for anti- or pro-thrombotic therapy.  Patient-specific models would enable predictions of resistances to common anti-platelet therapies such as asipirin and clopidogrel.  A multiscale model is built upon patient-specific platelet phenotyping and is compared to patient-specific platelet deposition to collagen in a microfluidic chamber.

The multiscale model has four major modules: lattice Boltzmann (LB), finite element (FEM), lattice kinetic Monte Carlo (LKMC), and neural network (NN).  The NN is trained on a single donor's pairwise agonist scanning (PAS) experiment that measures the rise in intracellular calcium to three main platelet agonists: adenosine diphosphate (ADP), thromboxane A2 (TXA2), and convulxin (CVX).  The LKMC method follows the motion of platelets within the fluid due to diffusion and convection.  Platelet bonding is captured in LKMC through a bonding model that links prediction of intracellular calcium from the NN to bonding kinetics.  The concentrations of soluble platelet agonists, which are released from activated platelets, are tracked using FEM.  Finally, LB solves for the velocity of the fluid around the growing platelet aggregate.  Due to the efficiency of the multiscale framework, the entire ‘active zone' of the experiment can be simulated in 2 dimensions.

The multiscale model is directly compared to experimental results of platelet aggregation for 3 donors with 3 antiplatelet therapies: COX inhibition, P₂Y₁ inhibition, and prostacyclin receptor activation.  The model predicts that donor 1 has consistently larger platelet aggregates than donors 2 or 3, which is confirmed by experiment.  The model also correctly predicts the donor-specific potency of antiplatelet therapies.  In general, prostacyclin receptor activation had the largest effect on platelet aggregation due to a decrease in collagen signaling.  P₂Y₁ inhibition and COX inhibition both reduce propagation of the aggregate once a monolayer is formed, however P₂Y₁ inhibition has a greater effect.  These trends are also seen in the experimental results.  The model also tracks the morphology of the platelet aggregate and shear rate (force) distribution along the solid-fluid boundary.

.

Figure 1. Simulation of platelet aggregation on collagen surface.  Circles – platelets (Black – fully unactivated, White – fully activated).  Lines – streamlines of the blood flow.  Background color – ADP concentration (TxA₂ concentration not shown).  Red bar – collagen patch.