Maximum entropy and population heterogeneity in continuous cell cultures

Jorge Fernandez-de-Cossio-Diaz, Anders Wallqvist, Roberto MuletPLOS Computational Biology 2019 · OpenReview

TL;DR

This paper introduces a novel framework using the maximum entropy principle to model cell heterogeneity in continuous cultures. This approach reveals how cell-to-cell variability can significantly alter system dynamics, leading to increased byproduct accumulation and larger cell populations, offering new insights for optimizing bioprocesses.

What Problem Does This Paper Address?

Continuous cell cultures, widely used in biotechnology, are known for complex, non-linear behaviors like multi-stability and hysteresis. While mathematical models aim to control these systems, they often assume a homogeneous cell population. This is a significant simplification, as even clonal populations exhibit cell-to-cell variability. The paper tackles the challenge of understanding and modeling this heterogeneity, which is crucial for accurately predicting and controlling the behavior of these complex biological systems. Ignoring this variability can lead to inaccurate models and missed opportunities for process optimization.

Methods

The core of the paper's methodology is the application of the Maximum Entropy Principle (MaxEnt). This principle is used to derive the phenotypic distribution of cells within a chemostat, considering the interplay between individual cell metabolism and the extracellular environment. The authors couple this with a Flux Balance Analysis (FBA) approach, a standard technique for modeling metabolic networks.

To handle the complexity of genome-scale metabolic networks, they employ an Expectation Propagation (EP) algorithm. This is a message-passing algorithm that provides a computationally tractable way to approximate the distribution of metabolic fluxes, especially for large networks where exact calculations are infeasible. They also introduce a crowding constraint to account for physical limitations within the cell.

For computational efficiency with large networks like the CHO-K1 model, they implement a model reduction strategy: after solving the system at maximum heterogeneity (β = ∞, equivalent to FBA), they remove reactions that are inactive across all conditions, resulting in a smaller, more tractable network.

Datasets

The paper does not explicitly mention using external datasets for training or evaluation in the traditional machine learning sense. Instead, it utilizes:

  • A simplified metabolic model: This serves as a "toy model" for demonstrating core concepts and allowing for exact analytical solutions.
  • A genome-scale metabolic network of the Chinese Hamster Ovary (CHO) cell line: This is a well-established model used in industry, providing a more realistic and complex test case for their framework. The parameters for this model are derived from existing literature and experimental data.

Key Insights and Ideas

  • Heterogeneity is not just noise; it's a driver of system behavior: The paper moves beyond viewing cell-to-cell variability as a nuisance and demonstrates how it fundamentally alters the dynamics of continuous cultures.
  • Toxic byproducts are managed differently in heterogeneous populations: The accumulation of toxic byproducts, like lactate and ammonia, can be a limiting factor in homogeneous cultures. Heterogeneity allows the population to collectively manage these byproducts more effectively, leading to larger cell numbers.
  • MaxEnt provides a principled way to model heterogeneity: Instead of making ad-hoc assumptions about the mechanisms of variability, MaxEnt offers a general framework to derive the distribution of cell phenotypes.
  • The "dynamical landscape" of cell cultures is highly sensitive to heterogeneity: The presence or absence of multi-stable regimes, the transitions between metabolic states, and the overall stability of the system are all profoundly affected by the level of cell-to-cell variation.
  • Homogeneous models can miss crucial phenomena: The paper highlights that byproducts predicted to be negligible or absent in homogeneous models can accumulate significantly in heterogeneous populations, with important implications for understanding and controlling bioprocesses.

Key Results

  • Simplified Model:
    • Heterogeneity can lead to larger cell populations by mitigating the impact of toxic byproducts.
  • The presence of heterogeneity can abolish or shrink multi-stable regimes observed in homogeneous models.
  • Byproducts that are negligible in homogeneous models can accumulate significantly in heterogeneous ones.
  • CHO-K1 Genome-Scale Model:
    • Qualitative findings from the simplified model are recapitulated, but with richer phenomenology.
  • Specific metabolites like formate and ammonia show increased accumulation with heterogeneity, while others like histidine and valine are driven to zero.
  • The study confirms that heterogeneity can enable larger cell densities or faster dilution rates under certain conditions, a surprising result given that higher β (more homogeneous) typically favors maximum growth rate.
  • The paper demonstrates that the system can exhibit multi-stability, with unstable steady states disappearing as heterogeneity increases.

Limitations and Future Work

  • Simplifications: The model assumes that cell mass composition (parameters like maintenance demand and biomass yield) is invariant across cells. While this is a reasonable approximation for a given cell type, it is a simplification. The authors note this is not essential for the formal model but simplifies calculations.
  • "Mean-field" Approximation: The current approach treats cells as interacting with the entire population rather than specific nutrients. Relaxing this "mean-field" approximation would require further work but is expected to yield similar core conclusions.
  • Parameter Uncertainty: The authors acknowledge that some enzymatic flux costs are unknown or poorly annotated, leading them to perform simulations with random perturbations to assess the robustness of their findings.
  • Future Work: The authors suggest that their framework can be extended to other models of continuous cell cultures and metabolic networks. They also hint at a connection between their model and the evolutionary dynamics of cell populations, which could be explored further. The potential for inducing heterogeneity to benefit industrial settings where cell numbers are limited by toxicity is a key implication for future research and application.

Summary updated

Abstract

Continuous cultures of mammalian cells are complex systems displaying hallmark phenomena of nonlinear dynamics, such as multi-stability, hysteresis, as well as sharp transitions between different metabolic states. In this context mathematical models may suggest control strategies to steer the system towards desired states. Although even clonal populations are known to exhibit cell-to-cell variability, most of the currently studied models assume that the population is homogeneous. To overcome this limitation, we use the maximum entropy principle to model the phenotypic distribution of cells in a chemostat as a function of the dilution rate. We consider the coupling between cell metabolism and extracellular variables describing the state of the bioreactor and take into account the impact of toxic byproduct accumulation on cell viability. We present a formal solution for the stationary state of the chemostat and show how to apply it in two examples. First, a simplified model of cell metabolism where the exact solution is tractable, and then a genome-scale metabolic network of the Chinese hamster ovary (CHO) cell line. Along the way we discuss several consequences of heterogeneity, such as: qualitative changes in the dynamical landscape of the system, increasing concentrations of byproducts that vanish in the homogeneous case, and larger population sizes.
BibTeX
@article{fernandezdecossiodiaz2019maximum,
  author = {Jorge Fernandez-de-Cossio-Diaz and Anders Wallqvist and Roberto Mulet},
  title = {{Maximum entropy and population heterogeneity in continuous cell cultures}},
  journal = {PLOS Computational Biology},
  year = {2019},
  url = {https://openreview.net/forum?id=W8oJj7okl7},
}
Figure