Inference of the drivers of collective movement in two cell types: Dictyostelium and melanoma

Elaine A. Ferguson, Jason Matthiopoulos, Robert H. Insall, Dirk HusmeierJournal of The Royal Society Interface 2016 · DOI

TL;DR

This paper introduces a novel statistical inference framework to identify the key drivers of collective cell movement by fitting a suite of mathematical models to experimental data. They found that in both Dictyostelium and melanoma cells, movement is primarily driven by cells creating their own chemical gradients through depletion, with melanoma cells also showing evidence of overcrowding effects.

What Problem Does This Paper Address?

Understanding how cells move collectively is crucial for many biological processes, from wound healing to cancer metastasis. While various mechanisms like diffusion and chemotaxis (movement up chemical gradients) are known to influence cell migration, pinpointing which mechanisms are dominant in specific cell types and conditions is challenging. This is particularly true because chemical gradients can arise from external sources or be self-generated by the cells themselves, for instance, by depleting a chemical in their environment. Existing mathematical models for cell movement are complex (often partial differential equations - PDEs), and fitting them to experimental data to infer these underlying drivers has been computationally prohibitive and methodologically underdeveloped. This paper aims to bridge this gap by developing and applying a robust statistical inference methodology to identify the most likely drivers of collective cell movement.

Methods

The core of this paper lies in its novel methodology for inferring cell movement drivers. They defined six candidate models, formulated as advection-diffusion-reaction PDEs, each incorporating different hypotheses for cell movement (e.g., diffusion only, chemotaxis to self-generated gradients, receptor saturation, overcrowding, and cell-cell interactions).

The key innovation is their approach to fitting these complex models to experimental data:

  • Pseudo-posterior inference: Instead of traditional Markov Chain Monte Carlo (MCMC) methods, which can be computationally intensive and prone to getting stuck in local optima for these PDEs, they employed a "poor man's" approximation. This involves generating numerous bootstrap samples of the original data and then optimizing each candidate model on each bootstrap sample. This yields a collection of parameter sets that approximates the posterior distribution of the parameters.
  • Weighted Log-Likelihood: To address potential biases in standard log-likelihood calculations, especially when cell numbers increase over time, they used a weighted log-likelihood function.
  • Model Selection with WAIC: They used the Widely Applicable Information Criterion (WAIC), a more robust model selection statistic than AICc or BIC, to compare the performance of the six candidate models. WAIC is particularly useful as it accounts for poorly determined parameters.

This approach allows for parallelization and automation, overcoming many computational hurdles.

Datasets

The methodology was tested on two distinct datasets:

  1. Dictyostelium discoideum cells: Data from an under-agarose assay where cells were exposed to folate (a chemoattractant) that they could deplete from their environment. This dataset was previously collected and analyzed by Tweedy et al. [7].
  2. Human melanoma cells: Data from a direct visualization chamber assay where cells migrated between wells containing fetal bovine serum (FBS). It was previously established that these cells respond to lysophosphatidic acid (LPA), a component of FBS, which they deplete. This dataset was previously collected and analyzed by Muinonen-Martin et al. [3].

Both datasets were reduced to one spatial dimension for analysis, focusing on movement along the primary direction of spread.

Key Insights and Ideas

  • Self-generated gradients are king: The most significant finding is that in both Dictyostelium and melanoma cell migration assays, the primary driver of collective movement is the cells' ability to create their own chemical gradients by depleting a chemical in their environment. This is a powerful demonstration of the self-generation mechanism.
  • Receptor saturation matters: Incorporating the concept of receptor saturation for chemoattractants significantly improved model fits compared to simpler models. This highlights that how cells sense gradients, not just the gradients themselves, is critical.
  • Overcrowding plays a role in melanoma: For melanoma cells, the model that included overcrowding effects (cells hindering each other's movement at high densities) was a better fit than simpler models. This suggests that the physical constraints of a dense cell population are important for their collective behavior.
  • No strong direct cell-cell interactions detected: For melanoma, there was no evidence for direct attractive or repulsive interactions between cells. For Dictyostelium, while some statistics hinted at it, the evidence was weak, likely due to the short timescales and ranges of such interactions in the experimental setup.
  • Temporal dynamics are crucial: The study revealed that cell behavior changes over time. For instance, the diffusion rate of Dictyostelium cells varied, and their chemotactic responsiveness increased over the assay duration. Melanoma cells showed a decreasing trend in their responsiveness.

Key Results

  • Model Selection:
    • For Dictyostelium, the "receptor saturation" model was selected as the best fit by WAIC.
  • For melanoma, the "overcrowding" model was selected as the best fit by WAIC.
  • Model Performance: The selected models provided visually good agreement with the experimental data, significantly outperforming simpler diffusion-only or basic chemotaxis models.
  • Parameter Dynamics: The study successfully inferred time-varying parameters for key drivers like diffusion and chemotactic responsiveness, revealing dynamic changes in cell behavior during the assays.

Limitations and Future Work

  • Computational Cost: While their pseudo-posterior approach is more efficient than traditional MCMC, it is still computationally intensive, requiring many optimizations.
  • Numerical Instabilities: The authors acknowledge that numerical instabilities in solving the PDEs can limit the parameter space explored, potentially leading to under-represented regions in their pseudo-posteriors. This is a known issue for advection-diffusion models when the Péclet number is high.
  • Simplification of In Vivo Complexity: The study systems are simpler than in vivo environments, where multiple cell types interact. The authors suggest that their methods could be extended to more complex scenarios.
  • Future Directions: The authors emphasize that their statistical inference framework is a powerful tool for guiding targeted experimental work. By identifying the most likely drivers, future experiments can focus on confirming these mechanisms and exploring their nuances in more complex biological contexts. They also suggest that extending the methodology to individual-based models could provide further insights, particularly for phenomena like contact inhibition of locomotion.

Summary updated

Abstract

Collective cell movement is a key component of many important biological processes, including wound healing, the immune response and the spread of cancers. To understand and influence these movements, we need to be able to identify and quantify the contribution of their different underlying mechanisms. Here, we define a set of six candidate models-formulated as advection-diffusion-reaction partial differential equations-that incorporate a range of cell movement drivers. We fitted these models to movement assay data from two different cell types: Dictyostelium discoideum and human melanoma. Model comparison using widely applicable information criterion suggested that movement in both of our study systems was driven primarily by a self-generated gradient in the concentration of a depletable chemical in the cells' environment. For melanoma, there was also evidence that overcrowding influenced movement. These applications of model inference to determine the most likely drivers of cell movement indicate that such statistical techniques have potential to support targeted experimental work in increasing our understanding of collective cell movement in a range of systems.
BibTeX
@article{ferguson2016inference,
  author = {Elaine A. Ferguson and Jason Matthiopoulos and Robert H. Insall and Dirk Husmeier},
  title = {{Inference of the drivers of collective movement in two cell types: <i>Dictyostelium</i> and melanoma}},
  journal = {Journal of The Royal Society Interface},
  year = {2016},
  doi = {10.1098/rsif.2016.0695},
}
Figure