It is difficult to extend computational methods based on isotopomer or cumomer balances to the IMFA problem, since they result in a large system of ordinary differential equations (ODE's) which must be solved to simulate the transient response of the network. An alternative method based on a recently developed elementary metabolite unit (EMU) decomposition of the network allows for a reduction in system size by 90% in comparison to the cumomer formulation. This decomposition traces the atom transitions through the network to identify the smallest collection of mass isotopomers that need to be simulated in order to describe the available measurements. An ODE solver has been customized to efficiently handle the cascaded linear systems generated by the EMU treatment. This solver makes use of a partial analytical solution, reducing the integration of ODE's to numerical quadrature. As a result of these developments, computational time can be dramatically reduced such that IMFA is feasible for biochemical reaction networks of realistic size and complexity.
As an illustration of our approach, we report results from applying IMFA to yeast cultures growing on 13C labeled glucose. High density fed-batch and chemostat cultures are used to maintain the system at metabolic steady state throughout the course of the experiments. Labeled substrate is introduced by switching the feed glucose from an unlabeled to a labeled supply. Culture samples are withdrawn from the fermentor and quenched immediately using a cold methanol/water solution. The biomass fraction is extracted with a methanol/chloroform/water mixture, and the resulting samples are derivatized to enable analysis by GC/MS. The mass isotopomer distributions of fragments obtained in this manner are used to estimate reaction fluxes and metabolite levels by fitting the IMFA network model to the measurements. Nonlinear statistical methods are applied to characterize the goodness-of-fit and to compute accurate confidence intervals for all estimated parameters.