Elsevier

Metabolic Engineering

Volume 12, Issue 2, March 2010, Pages 150-160
Metabolic Engineering

Application of dynamic flux balance analysis to an industrial Escherichia coli fermentation

https://doi.org/10.1016/j.ymben.2009.07.006Get rights and content

Abstract

We have developed a reactor-scale model of Escherichia coli metabolism and growth in a 1000L process for the production of a recombinant therapeutic protein. The model consists of two distinct parts: (1) a dynamic, process specific portion that describes the time evolution of 37 process variables of relevance and (2) a flux balance based, 123-reaction metabolic model of E. coli metabolism. This model combines several previously reported modeling approaches including a growth rate-dependent biomass composition, maximum growth rate objective function, and dynamic flux balancing. In addition, we introduce concentration-dependent boundary conditions of transport fluxes, dynamic maintenance demands, and a state-dependent cellular objective. This formulation was able to describe specific runs with high-fidelity over process conditions including rich media, simultaneous acetate and glucose consumption, glucose minimal media, and phosphate depleted media. Furthermore, the model accurately describes the effect of process perturbations—such as glucose overbatching and insufficient aeration—on growth, metabolism, and titer.

Introduction

Escherichia coli is an important host organism for the production of many biopharmaceutical proteins. Although production of properly folded full-length antibodies has been achieved in E. coli (Simmons et al., 2002), humanized glycosylation remains the province of eukaryotic hosts. Even with this limitation, many therapeutic proteins can be produced in E. coli, with a good chance for higher volumetric productivity rates than mammalian hosts (mass protein/time/volume). It is therefore important to continue to develop process-scale computational models of E. coli to aid in the understanding, design, and troubleshooting of processes at conditions common in the industrial manufacture of recombinant proteins. This effort is further motivated by the FDA's process analytical technology (PAT) initiative, which foresees computer modeling as an integral part of demonstrating process understanding. In addition, a reasonable model that captures the majority of a process's salient features may be used as a technician training tool or for testing the effect of minor process deviations in silico.

A very robust and generalizable method of modeling cellular metabolism is one that relies solely on fundamental mass and energy balances. Following this approach, flux balance analysis (FBA) is especially well-developed and suitable for describing cellular metabolism (Feist and Palsson, 2008). FBA provides a general framework on which the modeler can hang a wide variety of cellular information and heuristic hypotheses over the system of cellular mass and energy balances. Examples of additional constraints include, but are not limited to, transcript levels (Covert and Palsson, 2002), thermodynamic information (Henry et al., 2007), and intracellular crowding (Beg et al., 2007). Despite the promise of this approach in aiding the solution to a variety of metabolic engineering problems (Bro et al., 2006; Pharkya and Maranas, 2006; Lee et al., 2007), few descriptions of applying FBA to industrial fermentations are available. Part of this dearth may be due to the perceived steady-state nature of FBA models, despite papers showing how they may be extended to describe dynamic systems (Mahadevan et al., 2002; Beg et al., 2007). Other challenges include the accurate description of simultaneous, multiple substrate consumption and the general complexity of industrial fermentation processes, which often span a wide range of metabolic states and nutrient concentrations. Ramakrishna and co-workers have long recognized that a maximal growth rate objective may be a useful way to understand multi-substrate conditions (e.g. Dhurjati et al., 1985; Ramakrishna et al., 1996; Young et al., 2008), although other possible cellular objectives have been discussed (Schuetz et al., 2007; Ow et al., 2009).

In this paper, we describe the utilization of the FBA modeling approach in validating and improving our understanding of industrial E. coli fermentations. We show that a maximum growth rate objective function is a useful metabolic objective and that FBA is a powerful way to interpret complex fermentation data sets, reveal unexpected behavior, and demonstrate process understanding.

Section snippets

Process description

The host and 10L analog of the process modeled here have previously been described by Andersen et al. (2001). The 1000L version of the process used component concentrations and conditions similar to the 10L fermentation. The base medium contained sugar as the primary carbon source, but also contained significant quantities of yeast extract and a protein hydrolysate (Andersen et al., 2001). The process was operated as a fed-batch 32 h in length, and glucose-limited to control the dissolved oxygen

General phases of metabolism

The process consisted of at least five distinct metabolic phases. Fig. 1 presents an overview of process variables, simulation results, and predicted metabolic fluxes:

  • 1.

    Rich media growth phase

  • 2.

    Transitional growth phase

  • 3.

    Acetate reconsumption growth phase

  • 4.

    Growth on glucose and methionine phase

  • 5.

    Phosphate depletion growth phase

Although the division into these phases is somewhat arbitrary, we feel it is a useful way to discretize the changes occurring in the process. A description of the phases can be

Acknowledgements

The authors would like to thank Stephen Meadows and Dr. Christina J. Lee for help with editing, Ramin Afrasiabi and Jane Gunson for assistance in obtaining samples, and Craig Azzolino, Dr. Michael Laird, and Dr. Robert Kiss for their helpful advice and insight.

References (30)

  • H. Akashi et al.

    Metabolic efficiency and amino acid composition in the proteomes of Escherichia coli and Bacillus subtilis

    Proc. Natl. Acad. Sci. USA

    (2002)
  • D.C. Andersen et al.

    Metabolic oscillations in an E. coli fermentation

    Biotechnol. Bioeng.

    (2001)
  • Q.K. Beg et al.

    Intracellular crowding defines the mode and sequence of substrate uptake by Escherichia coli and constrains its metabolic activity

    Proc. Natl. Acad. Sci. USA

    (2007)
  • P. Dhurjati et al.

    A cybernetic view of microbial growth: modeling of cells as optimal strategists

    Biotechnol. Bioeng.

    (1985)
  • A.M. Feist et al.

    A genome-scale metabolic reconstruction for Escherichia coli K-12 MG1655 that accounts for 1260 ORFs and thermodynamic information

    Mol. Syst. Biol.

    (2007)
  • Cited by (97)

    • A Constraint-based modeling approach to reach an improved chemically defined minimal medium for recombinant antiEpEX-scFv production by Escherichia coli

      2022, Biochemical Engineering Journal
      Citation Excerpt :

      No need of significant amount of experimentations has made this method cost-effective and convenient for medium formulation. Constraint-based modeling has been used in different studies for overproduction of a wide range of products such as recombinant proteins [30–33]. For example, Swayambhu et al. used this method to overproduce siderophore compounds in E. coli.

    View all citing articles on Scopus
    1

    Present address: Amyris Biotechnologies, Inc., 5885 Hollis St, Suite 100, Emeryville, CA 94608, United States.

    View full text