Biology Reference
In-Depth Information
Regulated reactions in E. coli were identified by using
the Gibbs energies of formation for metabolites in each
reaction
[96]
. Reactions that were far from equilibrium
(having a large absolute value of the Gibbs energy for the
reaction) were identified as likely being under some form
of active regulation (due to high potential for wasted
energy if left unregulated). This hypothesis was verified
through quantitative metabolomic data, which in turn was
used to further constrain the boundaries of reactions in the
model.
Essential genes in E. coli have been predicted by the
model by optimizing biomass formation for single-gene
KOs at ~91% accuracy compared to experimental data
[104]
across the mutant strains in the Keio gene deletion collection
[113]
. In a similar approach, essential metabolites for E. coli
have been predicted computationally
[107]
by blocking flux
through all reactions producing a compound. In vivo
experiments were used to test the prediction by knocking out
multiple non-lethal reactions producing a compound and
evaluating the growth rate of the organism. Compounds
found by the model to be essential (e.g., tetrahydrofolate)
exhibited at least a 50% decrease in growth rate following
these knockouts, whereas non-essential compounds (e.g., 1-
deoxy-
D
-xylulose 5-phosphate) showed little to no change
following knockouts. This approach has subsequently
proved useful for identifying new antibiotics that are analogs
for essential metabolites
[114]
.
It has been reported that some FBA-predicted in silico
growth rates are higher than experimental values for wild-
type
[63]
, mutant
[62]
, and engineered
[62]
E. coli.In
these cases, it was theorized that the discrepancies were
the result of suboptimal utilization of the metabolic
network. To test this, adaptive laboratory evolution
experiments (ALE) were conducted in which the cell
cultures were maintained at exponential growth for
hundreds of generations
[61,115]
, allowing for selection of
the fastest-growing strains. In these experiments, growth
improved during ALE, often leading the phenotypes that
were consistent with the model-predicted optimal pheno-
types. By analyzing transcriptomic and proteomic data
from evolved E. coli
[111]
, it was found that the gene and
protein expression of these evolved strains changed
consistent with model-predicted usage. Essential and
optimal genes for the model (coding for reactions carrying
flux in the optimal solutions) showed significant increases
in expression in the evolved strains, whereas genes
encoding enzymes that were not used by the model showed
significant decreases in expression.
Phenotypic evaluation of metabolic models thus repre-
sents a very broad area of study, with potential uses in
increasing our understanding of cellular phenomena. Even
though it has been one of the most popular uses for
genome-scale metabolic models, the area of phenotype
prediction and evaluation has many unexplored avenues of
focus that will yield insights into how the gene content of
a cell correlates with cell phenotypes.
Fundamental Properties of Biological
Networks
The extent and careful curation of genome-scale metabolic
networks (e.g., 92% of 1260 genes in the iAF1260 E. coli
reconstruction
[36]
have experimentally validated func-
tions) makes them ideal targets for holistic analysis of
overarching network properties, including the identification
of co-regulated genes using flux coupling analysis
[116
e
118]
, network organization
[119,120]
, and flux
distribution logic
[121]
.
Analysis of flux distribution across thousands of
optimal growth simulation conditions and 50 000 non-
optimal solutions showed that distribution of fluxes across
E. coli followed a power-law distribution, with only
a fraction of the available reactions in the network carrying
high fluxes
[121]
. Albeit not noteworthy if the distribution
was only observable under limited conditions, the fact that
the phenomenon was observed across such a diverse range
of conditions indicates that the distribution is likely phys-
iologically advantageous in some way. This discovery
holds importance in the area of synthetic biology (e.g., in
building synthetic organisms
[122]
and efforts to create
a minimal organism
[123]
) as well as for metabolic engi-
neering, since perturbations in growth conditions showed
redistribution of fluxes around the high-flux 'backbone'
with little change in the reactions carrying low flux
[75]
,
indicating a degree of plasticity in the network.
Network analysis using in silico tools to identify co-
regulated genes was carried out as early as 2002
[124]
on
a subsystem basis, and on a genome-scale in 2004
[116]
by
looking at flux ratios between reactions to categorize the
coupling type. These results have been experimentally
verified via metabolomic data of single-gene knockouts in
yeast
[117]
. A similar approach looking at metabolite
connectivity was used to calculate hard coupled reaction
(HCR) sets (groups of reactions that are forced to operate in
unison due to mass conservation and connectivity
constraints
[118]
)inMycobacterium tuberculosis. This
analysis suggested several novel drug targets against the
organism that could be used to bypass resistance or
unwanted side-effect concerns of current therapeutics.
Thus, it has been demonstrated how the connectivity of
a network and other properties contribute to the function of
a biological system. Genome-scale metabolic reconstruc-
tions have aided in many such studies. Therefore, as these
properties continue to be identified and characterized, they
can potentially be used for many applications, such as
improving our understanding of disease
[125]
and the
development of treatments
[126]
.
