A framework for discrete stochastic simulation on 3D moving boundary domains

Brian Drawert, Stefan Hellander, Michael Trogdon, Tau‐Mu Yi, Linda PetzoldThe Journal of Chemical Physics 2016 · DOI

TL;DR

This paper introduces a novel computational framework for simulating biochemical reactions within complex, three-dimensional biological systems whose boundaries are constantly changing. By coupling a discrete stochastic simulation method (RDME) with a moving mesh algorithm, the authors enable the study of systems where cell shape and internal chemistry are interdependent, a crucial aspect of many biological processes.

What Problem Does This Paper Address?

Many biological systems, particularly at the cellular level, involve complex spatial arrangements of molecules and dynamic changes in cell shape. Traditional simulation methods often assume static environments or spatially homogeneous conditions, which fail to capture the intricate interplay between biochemical reactions and cell mechanics. For instance, processes like cell polarization, tip growth, and cell division involve cell boundaries that deform based on internal biochemical states, and vice versa. Accurately modeling these phenomena requires methods that can handle both spatial stochasticity and time-dependent geometries. Existing methods for spatial stochastic simulations (like particle-tracking) can become computationally prohibitive for large systems, while mesoscopic methods (like RDME) typically assume static domains. This paper bridges this gap by developing a method to simulate spatial stochastic biochemical reactions within complex, 3D, time-dependent domains.

Methods

The core of the proposed method is an operator splitting approach that decouples the reaction-diffusion dynamics from the domain movement. The simulation proceeds in discrete time steps ($\tau$). In each step:

  1. Solve Biochemistry: The reaction-diffusion master equation (RDME) is solved for the current biochemical state within a static mesh representing the domain at the beginning of the time step. This is efficiently handled using the Next Subvolume Method (NSM) on unstructured meshes, which allows for complex geometries.
  2. Calculate Velocity Field: A user-defined function, based on the current biochemical state, determines the velocity of the domain's boundary. This is the crucial step that couples the biochemistry to the mechanics.
  3. Evolve Geometry: The mesh is updated by moving its points according to the calculated velocity field over the time step $\tau$. This is implemented using the FEniCS/Dolfin finite element package.
  4. Transfer State: The biochemical state (molecule positions) is transferred from the old mesh to the newly evolved mesh. This involves sampling particle positions within their voxels on the old mesh and assigning them to the closest corresponding voxels on the new mesh, with special handling for species restricted to subdomains (e.g., membrane-bound proteins).

An adaptive step size selection mechanism is also introduced to control the error by adjusting $\tau$ based on the maximum allowed boundary displacement ($d_{max}$). This ensures that the diffusion of molecules is sufficiently faster than the boundary movement to maintain accuracy.

Datasets

Not discussed in the paper. The paper focuses on developing and validating a computational framework using simulated data and comparing it to existing theoretical models and microscopic simulations.

Key Insights and Ideas

  • Bridging the Gap: The primary insight is the successful integration of spatial stochastic simulations (RDME) with time-dependent, complex geometries, enabling the study of coupled biochemical and mechanical processes in cells.
  • Operator Splitting for Dynamics: The operator splitting approach effectively decouples the computationally intensive RDME simulations from the geometry evolution, making the overall simulation tractable.
  • Biochemistry Dictates Mechanics: The framework explicitly models how the internal biochemical state can drive changes in cell shape, a critical aspect of many biological phenomena like cell polarization and growth.
  • Time Scale Separation is Key: The method relies on the assumption that diffusion is faster than boundary movement. The authors characterize the error introduced by this separation and provide an adaptive time-stepping scheme to manage it.
  • Unstructured Meshes for Complexity: The use of unstructured meshes (via PyURDME and FEniCS/Dolfin) is crucial for handling the complex and evolving 3D geometries encountered in biological systems.

Key Results

  • Convergence Demonstrated: The method was validated by comparing its results to microscopic simulations and analytical solutions. Example 1 showed that the error decreases as the time step decreases, confirming convergence.
  • Accuracy in Dynamic Systems: Example 2 demonstrated the method's ability to accurately simulate a density-dependent switch for polarization in an expanding and contracting sphere, showing good agreement with theoretical steady-state values. The error was shown to increase with higher boundary velocities and larger time steps, as expected.
  • Biologically Relevant Simulation: Example 3 successfully simulated the formation of the yeast mating projection, a process where biochemistry and mechanics are tightly coupled. The simulation results showed a qualitative match with experimental microscopy images, highlighting the method's utility for studying real biological problems.

Limitations and Future Work

  • Time Scale Separation Assumption: The method's accuracy is dependent on the assumption that diffusion is faster than boundary movement. While an adaptive time-stepping scheme helps manage this, significant discrepancies in time scales can lead to larger errors.
  • Error Characterization: While the paper characterizes error related to time step and boundary velocity, further work could explore the impact of mesh quality and resolution more deeply.
  • Future Directions:
    • Developing methods that more closely couple the dynamics of the moving boundary and the biochemistry, potentially avoiding the error introduced by operator splitting.
  • Applying the framework to other biologically relevant problems involving coupled biochemistry and domain movement, such as tip growth in fungal hyphae, chemotaxis, cell migration, and cell division.
  • Extending the analysis to systems where the time scale separation is less pronounced.

Summary updated

Abstract

We have developed a method for modeling spatial stochastic biochemical reactions in complex, three-dimensional, and time-dependent domains using the reaction-diffusion master equation formalism. In particular, we look to address the fully coupled problems that arise in systems biology where the shape and mechanical properties of a cell are determined by the state of the biochemistry and vice versa. To validate our method and characterize the error involved, we compare our results for a carefully constructed test problem to those of a microscale implementation. We demonstrate the effectiveness of our method by simulating a model of polarization and shmoo formation during the mating of yeast. The method is generally applicable to problems in systems biology where biochemistry and mechanics are coupled, and spatial stochastic effects are critical.
BibTeX
@article{drawert2016framework,
  author = {Brian Drawert and Stefan Hellander and Michael Trogdon and Tau‐Mu Yi and Linda Petzold},
  title = {{A framework for discrete stochastic simulation on 3D moving boundary domains}},
  journal = {The Journal of Chemical Physics},
  year = {2016},
  doi = {10.1063/1.4967338},
}
Figure