Design of Pathway-Level Bioprocess Monitoring and Control Strategies Supported by Metabolic Networks - Measurement, Monitoring, Modelling, and Control of Bioprocesses - page 195

engineer the environment of cells more efficiently, namely for optimization of
culture medium composition and design of process monitoring and control strat-
egies that target intracellular control variables.
Metabolic networks can be used to interpret metabolic footprinting data and to
study how the extracellular environment can be manipulated to control intracel-
lular processes [ 7 ]. A few studies have addressed the development of dynamic
macroscopic models for process control derived from metabolic networks. Haag
et al. [ 8 ] showed that, for a class of macroscopic dynamic models, systems with
complex intracellular reaction networks can be represented by macroscopic
reactions relating extracellular components only with equivalent ''input-output''
behavior. Following a similar approach, Provost and Bastin [ 9 ] have developed
macroscopic dynamic models for Chinese hamster ovary (CHO) cultures wherein
the reaction mechanism is defined by the elementary modes (EMs) of the meta-
bolic network. An elementary mode can be defined as a minimal set of metabolic
reactions able to operate at steady-state, with the enzymes weighted by the relative
flux they need to carry for the mode to function [ 10 ]. As a result, each elementary
mode can be viewed as a metabolic subnetwork, which, under the steady-state
assumption, can be equivalently represented by a macroscopic reaction involving
only extracellular substrates and end-products.
The main difficulty in macroscopic dynamic modeling based on elementary
modes lies in the definition of the elementary mode weighting factors. As dis-
cussed later, any particular set of metabolic fluxes, or fluxome (i.e., phenotypic
state), can be represented as a weighted sum of elementary modes. The magnitude
of a weighting factor thus quantifies the contribution of the particular elementary
mode to the overall phenotypic state. In Provost and Bastin [ 9 ], the elementary
mode weighting factors were modeled by Michaelis-Menten kinetic laws as
functions of extracellular concentrations. The analogy between Michaelis-Menten
kinetics and elementary mode weighting factors is, however, not founded on
mechanistic principles. Moreover, this approach gives rise to very complex non-
linear systems, which are difficult to identify. In Teixeira et al. [ 11 ] we developed
hybrid macroscopic models structured by elementary flux modes for baby hamster
kidney (BHK) cells. Instead of Michaelis-Menten kinetic laws, empirical mod-
eling, namely artificial neural networks, was employed to model the elementary
mode weighting factors as functions of extracellular physicochemical variables.
Another difficulty in macroscopic dynamic modeling based on elementary
modes lies in the typically very high number of elementary modes. Indeed, the
number of elementary modes increases exponentially with the size and complexity
of the network [ 12 ]. However, most of these elementary modes are not active at
preset environmental conditions [ 13 ]. It is thus not necessary to use the full set of
elementary modes for a specific application. Of particular interest is the subset of
elementary modes describing a collection of measured phenotypic data. The
importance of this lies in the fact that the internal fluxes are not independently
distributed but strictly constrained by external fluxes through the pathways at
steady-state [ 14 ]. Therefore, the challenge is how to select the subset of elemen-
tary modes that describe a physiological state of interest. Effective reduction of