Genome-Scale Models of Metabolic and Regulatory Networks - Systems Biology: Networks, Models, and Applications - page 242

three times higher growth rates than were experimentally observed.
After experimental evolution, which resulted in growth rate increase
for all strains, 39 of the 50 cases tested (78%) reached the in silico pre-
dicted growth rates (figure 8.4). For the majority of the remaining cases
the experimental evolution reduced the discrepancy between experimen-
tal data and in silico predictions, but the in silico predictions were still
in most cases higher than the experimentally measured growth rates.
Based on the results discussed above, it appears that the FBA
predictions of at least exchange fluxes match those obtained for exper-
imentally evolved strains, and in some cases even for nonevolved
strains. This indicates that, in general, experimental evolution under
constant evolutionary pressure results in removal of any regulatory
and other constraints so that only the basic stoichiometric and maxi-
mum nutrient uptake constraints remain. However, even after
experimental evolution there are cases where the model either over- or
underpredicts compared to the experimental data. Such discrepancies
between model predictions and in vivo behavior may indicate either
missing reactions or additional constraints that are still active in
evolved strains. These constraints may, for example, be due to kinetic or
transcriptional regulation and identifying them will allow developing
improved constraint-based modeling strategies.
GENOME - SCALE ANALYSIS OF DELETION PHENOTYPES
In addition to analyzing in detail the physiology of evolved and
nonevolved strains, as described above, genome-scale metabolic models
have been widely used for qualitative prediction of growth phenotypes
for large nonevolved knockout strain collections. Although, based
on the data for evolved E. coli knockout strains, it is known that FBA
generally overpredicts growth rates for nonevolved knockout strains,
it should still be able to predict qualitative growth phenotypes (e.g.,
slow/normal growth) correctly. Large-scale knockout studies have
been performed in E. coli [56,57], S. cerevisiae [19,58], H. pylori [21],
and H. influenzae [59]. Overall the FBA predictions of gene deletion
phenotypes have been accurate in 60-90% of the cases studied,
depending on the completeness of the metabolic network for each
organism. While the correct prediction rates have been quite high for
all the organisms studied so far, the interesting part of the deletion
Figure 8.4 Adaptive evolution of E. coli metabolic gene knockout strains on
specific carbon sources. The x -axis corresponds to days of experimental
evolution (up to 50 days). The y -axis shows the strain growth rate measured as
a function of days of evolution together with the in silico predicted growth rate
(horizontal line). Data for at least two independently evolved strains is shown
in each panel. For more details see ref. [55].