Biology Reference
In-Depth Information
allows for both prediction of mutants with desirable properties and identifi cation of
conditions that support the expression of these properties.
Notwithstanding the generally good agreement between experimental results and
simulations of our model, several of the discrepancies encountered refl ect pitfalls in-
herent to CB modeling that go beyond the scope of our study:
First, the high number of blocked reactions and the mismatches with the BIOLOG
data show that there are still many areas of the metabolism that require thorough ex-
ploration. The genes encoding transport-related are particularly relevant, as for most
of them, neither the translocated compound nor the mechanism of translocation is
known. Furthermore, it should be highlighted that the genome still has 1,635 genes an-
notated as “hypothetical” or “conserved hypothetical”, more than 800 genes annotated
as putative, and over 800 for which the functional annotation gives no information
beyond the protein family name. It is thus likely that a fraction of the hypothetical and
non-specifi cally annotated genes in the current
P. putida
annotation are responsible
for unknown metabolic or transport processes, or that some might code for proteins
that add redundancy to known pathways. This observation is common to all genomes
sequenced so far and illustrates a major hurdle in the model building process (and
hence, its usefulness) that can be overcome only through extensive studies in func-
tional genomics.
Second, although we carefully constrained the
in silico
fl ux space through FBA
and FVA and obtained distribution spaces roughly consistent with those experimen-
tally determined via
13
C- fl ux analysis, these approaches are inherently limited as they
assume growth as a sole metabolic objective and ignore any effects not explicitly
represented in a CB metabolic model. It has been shown that FBA using objective
functions other than growth can improve predictive accuracy under certain conditions
[53]. Kinetic limitations also may play a very important role in determining the extent
to which a particular reaction or pathway is used. Teusink et al. [52] showed that in the
case of
L. plantarum
these factors may lead to false predictions.
Third, the reconstruction includes causal relationships between genes and reac-
tions via GPRs but it lacks explicit information regarding gene regulation. The regula-
tion of gene expression causes that there are many genes in the cell that are expressed
only under certain growth conditions. Therefore, the
in silico
fl ux space is generally
larger than the true
in vivo
fl ux space of the metabolic network. This, in turn, may in-
fl uence the robustness of the metabolic network and the essentiality of some reactions
and genes. The lack of regulatory information and of the genetic interactions involved
is likely to be one of the causes for faulty predictions of the viability of mutant strains.
Adding this information will be an important step in the further development and im-
provement of the accuracy of the reconstruction.
Fourth, although our analyses indicated that growth yield is relatively insensitive
to changes in biomass composition, these analyses also suggest that factors other than
the structure of the metabolic network play an important role in defi ning the relation-
ship between the growth yield and environmental conditions. The prediction of the
exact growth yield requires the precise measurement of maintenance values, which
may vary substantially from one condition to the other [44-46]. As the maintenance
