Biology Reference
In-Depth Information
that conferred the neuroprotective response. Microarray
data from Alzheimer's disease patients provided additional
support for the prediction of the role of GAD. It was found
that GAD expression was highest in brain regions that were
recalcitrant to neuron loss in AD, whereas the most
severely affected regions showed the lowest expression
levels in non-AD control patients. These models can thus
be used for computational exploration of diseased states in
human tissue in order to better understand the metabolic
basis for a disease, as well as to potentially identify novel
therapeutic targets.
Recently, another study used HT data to create a model
for non-small cell lung cancer
[136]
that was used to
identify novel drug targets. Using FBA, single-gene dele-
tions and synthetic lethals (i.e., double deletions resulting
in a lethal phenotype only when both genes are deleted) that
reduced growth rate in the model were identified as
potential drug targets. The set of 52 individual gene dele-
tions included all but one of the metabolic anticancer drug
targets approved by the FDA, as well as 13 genes targeted
by non-cancer drugs currently being tested for cancer
therapy applications. This left 31 potentially new targets for
anticancer drugs. Additionally, novel drug targets were
identified by using the synthetically lethal gene pairs and
HT data on cancer and healthy cell gene expression data.
By finding pairs where one gene was deleted in cancer cells
but not in healthy cells, drugs targeting the second gene of
a synthetically lethal pair would be lethal only to cancer
cells, as healthy cells would have the first gene of the pair
still active. For example, in healthy kidney cells the model
predicted that haem oxygenase (Hmox1) and fumarate
hydratase (FH) would function as a synthetic lethal gene
pair. In hereditary leiomyomatosis and renal-cell cancer
(HLRCC) there is often a germline mutation in the FH gene
that prevents its expression, making Hmox1 a potential
target for targeted treatment. An experimental validation
[144]
showed that shRNA silencing Hmox1 was lethal only
to cells missing FH, thereby demonstrating that synthetic
lethal gene pairings are a viable approach to developing
targeted therapeutic approaches for cancer treatment.
Interpreting the vast quantities of data generated from
high-throughput experiments has long been a challenge in
the life sciences, with the ability to create data far out-
pacing the ability to analyze it. Metabolic models thus
represent a potential context for data analysis, thus helping
in the interpretation of this data in new biologically
meaningful ways.
A Context for the Analysis of Large Data Sets
The interpretation of the vast amounts of data from various
high-throughput (HT) techniques presents an ever-growing
challenge for biological science
[127]
. Although statistical
approaches in data analysis constantly reveal novel insights
into cell properties
[128]
, metabolic reconstructions
provide a context-based approach toward HT data analysis
[129]
. Applications range from utilizing HT data to
constrain the model
[130,131]
, to the use of the model for
HT data visualization
[132]
, and even guiding tissue-
specific reconstruction abstractions from genome-scale
models
[133
e
140]
.
Although the correlation between gene expression and
protein expression is not exact
[141,142]
, using microarray
data to set constraints on in silico models has been shown to
be effective in predicting metabolic behavior in yeast
[133]
.
Alternative data types, e.g., fluxomic data, have also been
used to predict internal fluxes in E. coli
[130]
by con-
straining extreme pathways of the model. Altogether, these
applications function to provide an experimental basis for
simulation constraints on models that further reduce the
range of feasible phenotypes in the solution space for
simulations, resulting in models that have greater predictive
power.
Other studies utilize networks by overlaying HT data on
them to visualize the effects of perturbations
[132]
. Albeit
seemingly trivial, applications of this sort can facilitate the
identification of metabolic hotspots or pathways that are
significantly altered by providing a context for the data
gathered. For example, one study
[143]
overlaid gene
expression data after gastric bypass surgery on the human
metabolic model and found that the expression levels of
many reactions clustered together in metabolic pathways in
a similar way as would be seen in the skeletal muscle of
rhesus monkey when subjected to caloric restriction. Since
the data compared gene expression data from patients
before and a year after the surgery, this suggested that even
after weight stabilization has occurred in these patients,
their skeletal muscle metabolism is still showing the effects
of caloric restriction.
A more recent application of gene expression data has
been used in conjunction with a genome-scale model of
human metabolism
[143]
. Construction of tissue-specific
models of metabolism is facilitated by reducing the parent
model based on the presence or absence of transcripts or
proteins in HT data
[133]
. This approach was used to
generate models for 10 different tissue types.
A later study
[134]
generated tissue-specific models for
three neuron types and glial cells in the brain using similar
techniques. The models were able to predict the resistance
of GABAergic neurons to Alzheimer's damage and identify
a mechanism underlying this response. This mechanism
pointed to an enzyme, glutamate decarboxylase (GAD),
Multi-Cell and Microbial Community
Metabolism
The analysis of metabolic models can provide insight into
physiological phenomena for organisms in isolation.
However, outside the laboratory cells often grow in diverse
