Metabolically re-modeling the drug pipelineruppin/drug_remodelling_review.pdf · Metabolically 1...
Transcript of Metabolically re-modeling the drug pipelineruppin/drug_remodelling_review.pdf · Metabolically 1...
Metabolically re-modeling the drug pipelineMatthew A Oberhardt1,2,3, Keren Yizhak1,2,3 and Eytan Ruppin1,2
Available online at www.sciencedirect.com
Costs for drug development have soared, exposing a clear
need for new R&D strategies. Systems biology has meanwhile
emerged as an attractive vehicle for integrating omics data and
other post-genomic technologies into the drug pipeline. One of
the emerging areas of computational systems biology is
constraint-based modeling (CBM), which uses genome-scale
metabolic models (GSMMs) as platforms for integrating and
interpreting diverse omics datasets. Here we review current
uses of GSMMs in drug discovery, focusing on prediction of
novel drug targets and promising lead compounds. We then
expand our discussion to prediction of toxicity and selectivity of
potential drug targets. We discuss successes as well as
limitations of GSMMs in these areas. Finally, we suggest new
ways in which GSMMs may contribute to drug discovery,
offering our vision of how GSMMs may re-model the drug
pipeline in years to come.
Addresses1 School of Computer Sciences, Tel Aviv University, Tel Aviv 69978,
Israel2 The Sackler School of Medicine, Tel Aviv University, Tel Aviv 69978,
Israel
Corresponding authors: Oberhardt, Matthew A ([email protected]),
Yizhak, Keren ([email protected]) and Ruppin, Eytan
([email protected])3 These authors contributed equally to this study.
Current Opinion in Pharmacology 2013, 13:778–785
This review comes from a themed issue on New technologies
Edited by Andreas Gescher and Marco Gobbi
For a complete overview see the Issue and the Editorial
Available online 31st May 2013
1471-4892/$ – see front matter, # 2013 Elsevier Ltd. All rights
reserved.
http://dx.doi.org/10.1016/j.coph.2013.05.006
IntroductionDespite huge advances in biology in the last few decades,
the resources pumped into drug development become
ever larger and yet pipelines are not producing more
drugs [1–3]. Industry analysts have blamed this in part
on stagnation in R&D due to risk aversion in Big Pharma
[4], and yet with strains of bacteria emerging that are
resistant to nearly every existing antibiotic [5] and many
of the most devastating diseases in the developed and
developing world still lacking cures, new drugs are des-
perately needed. A major pitfall of traditional approaches
(such as the ‘one drug, one target’ paradigm) is their
inability to account for the full complexity of biological
systems [6]. With many systems biology techniques now
being developed specifically to navigate this complexity,
Current Opinion in Pharmacology 2013, 13:778–785
a synthesis between the traditional drug pipeline and
systems methods may soon become not an R&D luxury,
but an industry necessity [4].
Systems biology addresses biological questions not from
the perspective of individual protein or gene actors, but
rather taking into account full interacting cellular net-
works. Modern omics technologies, as well as huge
advances in computational capabilities, have made sys-
tems biology both appropriate and crucial for fully utiliz-
ing the resources for biological discovery that are available
today. One of the most influential paradigms within
computational systems biology is constraint-based mod-
eling (CBM), which focuses on building and analyzing
genome-scale metabolic models (GSMMs). CBM
imposes context-specific constraints on a space of possible
metabolic behaviors and allows prediction of numerous
metabolic phenotypes, including growth rates, nutrient
uptake rates, gene essentiality and more (Figure 1) [7].
The CBM field has been rapidly expanding, and today
dozens of manually curated GSMMs are available in
addition to over a hundred bacterial GSMMs built
through an automated system called SEED [8]. In
2007, two genome-scale models of human metabolism
were published [9,10], followed by a new version that has
been published very recently [Recon2 [11�]]. While these
models are generic and thus not specific to any mature
cell-type or tissue-type, they have successfully served as a
basis for context-specific models of human tissues [12–22]. GSMMs have been used in areas including metabolic
flux analysis [23,24], studies of network evolution [25],
and metabolic engineering [26,27]. These uses and many
others have been thoroughly reviewed elsewhere [28,29].
The increasing availability of high-quality GSMMs pre-
sents new opportunities in drug development and design
both for combating pathogens and for treating human
disorders, which together form the focus of this review.
Determining the best drug targetsTraditional drug development left the determination of
targets for a second step following the discovery of lead
compounds with potential therapeutic efficacy (i.e. phe-
notypic screens), whereas the genomics era brought a
surge of interest in the reverse ‘target-based’ approach,
that is, determining good potential gene or protein targets
first, and then screening for suitable small molecules to hit
them [30]. Despite great interest in what seemed a more
rational approach to drug development, target based
screens have under-performed phenotypic screening in
drug development [30,31]. One of the reasons cited for
this failure is not target based screening itself, but the
www.sciencedirect.com
Metabolically re-modeling the drug pipeline Oberhardt, Yizhak and Ruppin 779
Figure 1
Lipids
AminoAcids
Nucleotide s
Cellula rBiomas s
‘Om ics ’ da ta
INPUT 3
Growth mediu m
INPUT 1
High
Low
Moderate
Glucos e Sulfat e
Genome-Scal e Metabol ic Mod el
Byprodu ctsecretion rates
OUTPUT 1
Growth rate andgene esse ntia lit y
X
OUTPUT 2
A
B
C
D
E
F
A
B
C
D
E
F
Prediction of flu xrates at wil d-typeand pertu rbat ion
states
X
OUTPUT 3Substrate upta kerat e
10gGlu cos e
1gSulfate
..
..
INPUT 2
Current Opinion in Pharmacology
Inputs and outputs to GSMMs. A genome-scale metabolic model (GSMM) integrates information on compounds present in the growth medium,
substrate uptake rates (if known), and information on the internal state of a cell (such as gene knockouts or large-scale omics datasets) to predict the
steady state fluxes and activities of all reconstructed pathways in metabolism. Outputs of a GSMM analysis typically include growth yields (or growth
rates if substrate uptake data is available), gene essentiality information, and prediction of wild-type or perturbed flux distributions across the whole
metabolic network.
need to integrate new technologies in order to predict
better targets [32]. There has been a major focus in the
GSMM community on using models to predict drug
targets, as GSMMs can directly predict the effects of
perturbations such as gene knockouts on growth and thus
lend themselves naturally to this task. Therefore, we
focus here on the ways GSMMs have been used to predict
novel drug targets. Drugs serve two major purposes:
either to kill or to functionally alter a target tissue or
cell. Here, we review GSMM methods for both of these
aims.
Drugs to kill
Selective killing of certain target cells while sparing
healthy human cells is the fundamental goal of many
drugs. This goal is common, for example, in targeting
both bacterial pathogens as well as cancer. About a third
of current antibiotics target metabolic genes (derived
from DrugBank: [33]), and metabolism comprises a large
potential pool for more targets. Because of their basic
structure and capabilities, GSMMs are a natural choice for
rational drug target development.
www.sciencedirect.com
GSMMs have been recognized for quite a while as
vehicles to predict lethal metabolic drug targets in patho-
gens [34]. However, despite many proofs-of-principle
such as accurate prediction of essential genes using flux
balance analysis (FBA) (e.g. [35]) and ample predictions
of possible targets in over a decade of use, these efforts
have led to few promising new drug leads, and to our
knowledge, no new drugs. Part of the reason for this is that
many key factors essential in a good drug target have been
out of the scope of GSMMs — for example, the ‘drugg-
ability’ of a target, which can involve its cellular localiz-
ation, its three-dimensional structure, and its potential
binding with known classes of inhibitors, among other
factors [32,36].
More recently, merging druggability information with
GSMM-based essentiality predictions has shown tremen-
dous potential. A recent study aimed to repurpose already
FDA-approved drugs to kill the deadly parasite, Leishma-nia major, combining GSMM predictions of gene essenti-
ality and synthetic lethality (i.e. lethality of two genes in
combination) along with ‘druggability’ scores and known
Current Opinion in Pharmacology 2013, 13:778–785
780 New technologies
Figure 2
Eicosanoid pathwaymetabolites identified
as cancer specificmetabolic features;
Agren et al. [16]
Novel antibiotic leadcompounds against E.
coli and S. aureusdeveloped to target FAbiosynthesis pathway;
Shen et al. [38]
Suggested role of ACL inproducing acetylcholinein Alzheimer disease;
Lewis et al. [21]
hypertensive side effectcaused by the drugtorcetrapib linked toreduced secretion of
PGI2 in kidney; Chang etal. [20]
Glucose
G6P
F6P
FBP
G3P
DHAP
3PG
2PG
PEP
Lactate Ac-CoA
To PentosePhosphate Pathway
R5PNucleotideSynthesis
TCACycle
Glycolysis
Serine
Glycine
DHF
mTHF
THF
dUMP
dTMP
Amino acidSynthesis
3PHP
NAD NADH
SerineP-Serine
HemeHeme
Bilirubin
Heme Biosynthesis
Succinyl-CoA
Citrate
Citrate Ac-CoA
Fatty Acid Oxidation
Folate Metabolism
Serine Biosynthesis
NADH
OxidativePhosphorylation
ATP
Acyl-CoAAcyl-CoA
Fatty Acid Biosynthesis
ArachidonicAcid
HPETE PGG2
Leukotrines PGI2
EicosanoidMetabolism
HMOX
HMOX identified andvalidated as synthetic lethalwith FH in renal cell cancer;
Frezza et al. [46]
PDH
ACLLipid
Synthesis
Halofantrine, a malarial drug associatedwith the vacuolar ATP synthase catalyticsubunit A, found to be active against L.
major; Chavali et al. [37]
DHFR
DHFR, a knownchemotherapeutic,identified as cancer
target; Folger et al. [15]
Chemical analogs toFolate metabolism
intermediates found aspromising antibiotic
leads in V. vulnificus;Kim et al. [45]
Pyruvate
1,3BPG
Brain GSMM identifiedknown altered activityof PDH in Alzheimerdisease; Lewis et al.
[21]
Human targets Bacterial targets
Key:
Current Opinion in Pharmacology
Drug target related predictions using GSMMs. A map of metabolism is shown, with descriptions of some of the highest confidence drug targets (or
lead molecules) that have been identified in GSMM studies. Human (blue, solid border) and bacterial (orange with dashed border) targets are both
shown.
chemical-protein interactions [37�]. This led to successful
in vitro killing of L. major by Halofantrine, an antimalarial
agent, as well as synergistic killing by four repurposed
drug pairings (Figure 2). Such repurposing efforts, especi-
ally when aimed at neglected diseases (of which L. majorinfection is one), may have enormous impact.
Recent literature includes several other mergers of
GSMM predictions with structural or druggability data
in order to discover new drugs. In a study of Escherichia coliand Staphylococcus aureus, predicted enzyme targets in the
FASII pathway were virtually screened for binding
against a library of small molecules. Out of 41 predicted
inhibitors, eight proved strongly and nine weakly active in
extracted enzyme assays [38]. Another study of the same
two organisms also used GSMMs alongside protein-mol-
ecule binding predictions to identify drugs leads that
target the histidine biosynthesis pathway, which showed
marked in vitro efficacy [39]. Some recent efforts have
Current Opinion in Pharmacology 2013, 13:778–785
also brought the focus from genes to metabolites, analyz-
ing ‘choke points’ (i.e. reactions uniquely consuming or
producing a metabolite) [40,41] or metabolite essentiality
[42,43]. Metabolite-centric methods have the advantage
that small molecules can be sought based directly on their
similarity to the essential metabolite, and will often have
the potential to hit multiple enzymatic targets, a desirable
drug attribute [6]. This line of reasoning led to identifi-
cation of chorismate as a prime drug target for four
prominent pathogens [44]. Chorismate is an important
biochemical intermediate in plants and microorganisms
and a precursor of metabolites such as aromatic amino
acids, salicylic acid, and more. Importantly, there is no
significant homology between genes that encode for
reactions that utilize chorismate and the human genome,
suggesting that blocking this compound might affect the
pathogen and not the human host. A later study from
the same group screened live cells of Vibrio vulnificuswith 352 analogs of five promising target metabolites,
www.sciencedirect.com
Metabolically re-modeling the drug pipeline Oberhardt, Yizhak and Ruppin 781
as identified through metabolite-centric GSMM
methods. This screen yielded a small molecule inhibitor
with bacteriocidal properties in the range of currently
used antibiotics [45��].
These studies demonstrate how combining GSMM-
based target predictions with information on druggability
and enzyme/metabolite structural analysis can lead to
novel drug candidates. These pipelines are new, but as
they mature we may begin to see their effects in the drug
industry. So far, attempts to use GSMMs to develop lethal
drugs have focused mostly on microbial pathogens, but
there is increased focus and interest in turning GSMM
methods toward cancer. The power of GSMM in this
arena was recently shown in a study of the role of
fumarase, a TCA-cycle enzyme whose loss in the germ-
line can cause leiomyomatosis and renal carcinoma [46].
In the study, a kidney cancer GSMM was used to predict
what metabolic mechanisms these cells might use to
survive without fumarase. A novel pathway was predicted
and then confirmed in vitro in immortalized kidney cancer
cells. This in turn led to promising new drug targets in the
haem oxygenase pathway that selectively killed cancer
cells that were fumarase negative while sparing fumarase
positive cells (Figure 2), a major step forward in selective
targeting of this type of cancer.
Drugs to alter
While most drugs targeting microbes aim to kill the target
cells, the majority of human-cell targeted drugs obviously
aim not to kill the cells but rather to alter their function in
a therapeutic manner. For example, 16 of the 20 most
prescribed therapeutic drug classes in the US in 2011 aim
to alter the functions of human cells (derived from: [47]).
GSMMs are attractive tools for developing new drugs in
this area, as they provide a large-scale, high confidence
vehicle for predicting genome-wide effects of a pertur-
bation. Hundreds of past studies have explored the
effects of drugs or metabolic alterations (such as gene
knockouts or changes in environmental conditions) on
microbes, providing proofs of principle for many types of
uses of GSMMs in this arena (see, e.g. [48]). In an out-
standing example, a GSMM of E. coli was recently used to
build a range of ensemble models that account for pro-
duction of reactive oxygen species (ROS) [49]. Gene
deletions were then predicted that increase endogenous
production of ROS, and knockout of these genes in vitrowas shown to increase susceptibility of E. coli to oxidants
and to antibiotics. This study shows the power of GSMMs
in developing adjuvant therapies to microbial pathogens,
and more broadly, shows how prediction of metabolic
alterations can suggest new targets for drugs.
Until very recently, usage of GSMMs to develop drugs
that alter cells’ state has been limited because of the lack
of a GSMM for human cells. However, with the recent
introduction of human metabolic reconstructions ([9] and
www.sciencedirect.com
a recent update [11�]), methods to explore drug-related
alterations in human metabolism are emerging. Generally
speaking, many of the techniques mentioned in the ToKill section can also be applied to altering metabolism,
but rather than inhibiting production of cellular biomass
(and hence killing a cell), the goal is to inhibit the
production of a metabolite known to have systemic
impact on disease (e.g. cholesterol). Conversely, our lab
recently developed a new GSMM method to address the
problem of restoring diminished functions to cells. The
method, which identifies gene targets most likely to
globally transform a diseased metabolic state toward a
healthy one, identified novel lifespan extending genes,
which were then successfully experimentally validated in
yeast. In a promising follow-up, the method identified the
eicosanoids pathway, a known age related pathway, as a
high-priority target for age-reversal in mammals (Yizhak
et al., personal communication).
Despite these few examples, this area of drug develop-
ment using GSMMs is still relatively underdeveloped.
Potential future applications include developing adjuvant
therapies (as mentioned above), reducing the metastatic
potential of cancer, and even predicting dietary interven-
tions that may augment or replace a certain drug [50].
Using GSMMs to predict drug toxicity andselectivityDespite extensive advancements in drug discovery, drug
toxicity and selectivity remain major drivers of failures
and cost during the drug development process. These
failures result from unexpected negative systems-level
effects of hitting the appropriate target, as well as to off-
targeting in both primary and secondary tissues. Early
pre-clinical elimination or modification of troublesome
lead compounds is hence an important task that may
significantly bolster drug industry R&D efficiency.
GSMMs fit naturally into this scheme, as they are well
suited to predicting system-wide effects of metabolic
perturbations. Importantly, these effects can be evaluated
at multiple levels that include both diseased and healthy
tissues, and thus can give valuable insight into selective
targeting.
The prediction of ligand binding sites using protein
structures is a crucial step toward identifying drug side-
effects. This type of structural analysis can identify as-yet
uncharacterized drug targets that may participate directly
in drug response. While ligand binding predictions go
beyond GSMM-based methods, the integration of both
approaches is complementary and hence of much value.
Such a combined approach was recently applied [20] to
investigate the hypertensive side effect of the cholesteryl
ester transfer protein inhibitor torcetrapib in the context
of human renal function. Focusing on 41 predicted off-
targets of torcetrapib, the authors used a renal metabolic
model to show that the inhibition of PTGIS precludes the
Current Opinion in Pharmacology 2013, 13:778–785
782 New technologies
secretion of PGI2. This metabolite has been linked to
blood pressure and is thus expected to have a hyperten-
sion effect. As mentioned earlier, GSMMs have also been
used alongside ligand binding models in the prediction of
novel antimicrobial compounds [38,39], an approach that
shows marked promise.
Many drugs work to inhibit a certain enzyme or function of
the cell, but some drugs are rather designed to stimulate or
trigger a response. Clearly, inhibitory and stimulatory
responses should be modeled differently. Our lab has
recently developed CBM-based method termed EDGE
that is designed to address this challenge (Wagner et al.,personal communication). Given a species metabolic
model, EDGE works to systematically predict whether
the over-activation of a metabolic enzyme is expected to be
toxic to the organism or not. As an initial proof of concept,
EDGE was shown to successfully detect genes whose
expression is toxic in microorganisms, both in rich and
in poor media. EDGE therefore may help predict unin-
tended drug side effects in future studies in humans.
The capacity of a drug to produce the desired effect only
at a selected target cell or tissue is another major chal-
lenge in drug design. This problem is encountered at
various levels: first, antibiotics that are designed to kill
only a certain spectrum of bacteria (bacteria versus bac-
teria); second, antibiotics that should spare the host
human cells (bacteria versus human); and third, drugs
that must target diseased cells while keeping healthy
tissues in our body intact (human versus human). The
recent development of GSMMs, both for hundreds of
bacterial species and for numerous human tissues, pro-
vides for the first time an opportunity to address these
issues. GSMMs have been used to target specific bacterial
spectra [44], and reducing likely toxicity by filtering out
gene or metabolite targets present in humans is standard
practice in predicting novel antibacterial drug targets
[45��]. Furthermore, GSMMs have been used to study
host–pathogen interactions, resulting in a joint Mycobac-terium tuberculosis — alveolar macrophage model [51] that
displayed higher accuracy than the non-joined model in
predicting gene essentiality. Although the tuberculosis
study did not focus on drug development, the growth of
M. tuberculosis is both so sluggish and so interconnected
with its macrophage host that such models may be
required in order to boost both effectiveness and selec-
tivity of future drugs.
The issue of using GSMMs to develop selective drug
targets has already been initially addressed in the context
of cancer. An attempt to accomplish this goal was taken by
Folger et al. [15] who built a metabolic model of cancer
metabolism and identified drug targets that can poten-
tially affect cell proliferation. In parallel, the effect of
these targets on the production of ATP in a generic non-
proliferating human cell was examined, resulting in a set
Current Opinion in Pharmacology 2013, 13:778–785
of predictions that is highly enriched with FDA-approved
metabolic anticancer drugs (Figure 2). As mentioned
previously, a GSMM of the kidney cancer was used to
determine new drug targets in the haem pathway that are
selective for certain cancer cells, relying on utilizing
synthetic lethality [46]. Finally, Facchetti et al. [52]
developed a novel method for identifying drug targets
by inhibiting a predefined objective function such as cell
growth in one cell type while preserving the metabolism
of another. This analysis correctly identified known che-
motherapeutic agents, lending a strong proof-of-principle
to its use in novel drug discovery. Since these studies
have been published, multiple GSMMs of human tissues
have been reconstructed. Utilizing them, we can there-
fore expect a more thorough analysis of drug selectivity in
a broad span of human metabolic disorders in the near
future.
Future directionsGSMMs lend themselves naturally to the early stages
of drug development, most notably the determination of
new targets for target-based screens. However, as direct
drivers of innovation and as scaffolds for interpretation of
complex large-scale datasets, the potential of GSMMs in
drug development is yet largely untapped. We put some
current and projected future uses in Figure 3.
One of the great strengths of GSMMs is their ability to
integrate high throughput data to, for example, model
complex lifestyles of pathogens in vivo or in situ. As such,
GSMMs have been used to explore the progression of a
chronic cystic fibrosis lung infection [53], lifecycle stages
of the parasites causing Chagas and Malaria [54,55], and
features of other complex diseases [51,56]. Large efforts
such as the ENCODE project [57] have recently revealed
many new areas of biology, such as genome-wide human
DNA methylations, copy number variations, and SNPs
[58], which may also soon be integrated with GSMMs
following along the lines of recent multi-system modeling
efforts [59,60�]. GSMMs provide a solid framework for
contextualizing such data, and may be integrated into
future drug development pipelines in ways analogous to
how physiologically based pharmacokinetic (PBPK)
models already have been [61].
Additionally, GSMMs may be important for developing
entirely new classes of drugs, such as probiotic treatments
in which engineered bacteria outcompete pathogens in
the human gut. Such ideas are not so outlandish con-
sidering, for example, the significantly higher cure rates of
Clostridium difficile infections via fecal transplant (a
probiotic treatment) than via antibiotics [62]. GSMMs
are natural platforms for use in developing such probio-
tics, as they can integrate with 16s rRNA or metagenomic
data to produce predictive community models [63,64],
and can also aid in engineering of specific probiotic strains
[65].
www.sciencedirect.com
Metabolically re-modeling the drug pipeline Oberhardt, Yizhak and Ruppin 783
Figure 3
Develop ment& clini cal trial sNovel screens? Drug
UsageResista nce
Future GSMM possibili ties
Designingprobiotics orother novel
lead pipelines
Repurposingdrugs
Reducingtoxicity-baseddrug failures
Finding new/ better drug
targets
Targetingcertain lifestages ofparasites
Reducingbacterial
resistance atpopulation level
Reducing cancer/bacterial resistance
in patientsDeveloping
personalizedmedicines
Determiningmechanisms/targets of lead
compounds
Developingphenotypic
assays
Curre nt GSMM application sor proofs of prin cipl e
DrugPip elin e
Development& clinical trialsNovel screens? Drug
UsageResistance
Repurposingdrugs
Reducingtoxicity-baseddrug failures
/ better drugtargets
Rb
res
Reducing cancer/bacterial resistance
in patientsDeveloping
personalizedmedicines
Determiningmechanisms/targets of lead
Developing
Target based screens
Phenotypic screens
Current Opinion in Pharmacology
Current and future uses of GSMMs for drug discovery. GSMMs have potential to aid in many areas of the drug discovery process, although most
studies to date focus on determining better drug targets for target based screens, and a few other select applications (top, green). The drug pipeline is
portrayed along with current (green) and future potential (orange) applications of GSMMs toward drug discovery.
As mentioned before, GSMMs can also be used to suggest
new ways to repurpose existing drugs [37,66]. Drug
repurposing sits far later in the drug development pipe-
line than target discovery, the prediction of toxicity, and
most of the other applications previously discussed (see
Figure 3), and thus represents a break from the emphasis
on the very early stages of drug development.
Many other portions of drug development have not even
seen a preliminary GSMM study. This is especially
striking in phenotypic screening, as despite clear poten-
tial, GSMMs have never (to our knowledge) been used in
either developing phenotypic screens or in interpreting
the mechanisms of lead compounds from them, even in
proof of principle. Concerns that arise later in the life-
times of drugs, such as the common occurrence of resist-
ance in bacteria and cancer, also have not yet been
addressed using GSMMs, despite the demonstrated use-
fulness of GSMMs in examining important resistance
factors such as gene promiscuity [67] and cellular evol-
ution [68,69]. Finally, GSMM methods subtle enough to
predict differential phenotypes of a tissue across many
humans, a requirement for personalized medicine, are
only just now being developed (Yizhak et al., personal
communication). Such methods will also help in
www.sciencedirect.com
developing drugs that target certain tissues over others.
GSMMs are obviously not magic bullets, but once they
are noticed and adopted by the pharmaceutical industry,
they will undoubtedly add value and insight that would
be difficult to gain otherwise.
AcknowledgementsKY is partially supported by a fellowship from the Edmond J. SafraBioinformatics Center at Tel-Aviv University and is grateful to the AzrieliFoundation for the award of an Azrieli Fellowship. MO gratefullyacknowledges support from the Whitaker Foundation and from the DanDavid Foundation. ER’s research is supported by grants from the IsraeliScience Foundation (ISF), the Microme project (EU FP7), and by theIsraeli Centers of Research Excellence, Gene Regulation in ComplexHuman Disease Center (41/11).
References and recommended readingPapers of particular interest, published within the period of review,have been highlighted as:
� of special interest�� of outstanding interest
1. Kessel M: The problems with today’s pharmaceuticalbusiness — an outsider’s view. Nat Biotechnol 2011,29:27-33.
2. Paul SM, Mytelka DS, Dunwiddie CT, Persinger CC, Munos BH,Lindborg SR, Schacht AL: How to improve R&D productivity: thepharmaceutical industry’s grand challenge. Nat Rev DrugDiscov 2010, 9:203-214.
Current Opinion in Pharmacology 2013, 13:778–785
784 New technologies
3. Munos B: Lessons from 60 years of pharmaceutical innovation.Nat Rev Drug Discov 2009, 8:959-968.
4. Cuatrecasas P: Drug discovery in jeopardy. J Clin Invest 2006,116:2837-2842.
5. Boucher HW, Talbot GH, Bradley JS, Edwards JE, Gilbert D,Rice LB, Scheld M, Spellberg B, Bartlett J: Bad bugs, no drugs:no ESKAPE! An update from the Infectious Diseases Society ofAmerica. Clin Infect Dis 2009, 48:1-12.
6. Hopkins AL: Network pharmacology: the next paradigm in drugdiscovery. Nat Chem Biol 2008, 4:682-690.
7. Reed JL: Shrinking the metabolic solution space usingexperimental datasets. PLoS Comput Biol 2012, 8:e1002662.
8. Henry CS, DeJongh M, Best AA, Frybarger PM, Linsay B,Stevens RL: High-throughput generation, optimization andanalysis of genome-scale metabolic models. Nat Biotechnol2010, 28:977-982.
9. Duarte NC, Becker SA, Jamshidi N, Thiele I, Mo ML, Vo TD,Srivas R, Palsson BO: Global reconstruction of the humanmetabolic network based on genomic and bibliomic data. ProcNatl Acad Sci U S A 2007, 104:1777-1782.
10. Ma H, Sorokin A, Mazein A, Selkov A, Selkov E, Demin O,Goryanin I: The Edinburgh human metabolic networkreconstruction and its functional analysis. Mol Syst Biol 2007, 3.
11.�
Thiele I, Swainston N, Fleming RMT, Hoppe A, Sahoo S,Aurich MK, Haraldsdottir H, Mo ML, Rolfsson O, Stobbe MD et al.:A community-driven global reconstruction of humanmetabolism. Nat Biotechnol 2013, 31:419-425 (advance onlinepublication).
This article presents Recon 2, a community-driven consensus GSMM,which is the most comprehensive, model-ready representation of humanmetabolism to date. Recon 2 is shown to outperform its predecessorRecon 1 in predicting metabolite biomarkers, and may serve as a scaffoldfor integrating diverse data types in various human diseases.
12. Sahoo S, Thiele I: Predicting the impact of diet andenzymopathies on human small intestinal epithelial cells. HumMol Genet 2013.
13. Bordbar A, Mo ML, Nakayasu ES, Schrimpe-Rutledge AC, Kim Y-M, Metz TO, Jones MB, Frank BC, Smith RD, Peterson SN et al.:Model-driven multi-omic data analysis elucidates metabolicimmunomodulators of macrophage activation. Mol Syst Biol2012, 8.
14. Jerby L, Shlomi T, Ruppin E: Computational reconstruction oftissue-specific metabolic models: application to human livermetabolism. Mol Syst Biol 2010, 6.
15. Folger O, Jerby L, Frezza C, Gottlieb E, Ruppin E, Shlomi T:Predicting selective drug targets in cancer through metabolicnetworks. Mol Syst Biol 2011, 7:501.
16. Agren R, Bordel S, Mardinoglu A, Pornputtapong N, Nookaew I,Nielsen J: Reconstruction of genome-scale active metabolicnetworks for 69 human cell types and 16 cancer types usingINIT. PLoS Comput Biol 2012, 8:e1002518.
17. Wang Y, Eddy J, Price N: Reconstruction of genome-scalemetabolic models for 126 human tissues using mCADRE. BMCSyst Biol 2012, 6:153.
18. Shlomi T, Cabili MN, Herrgard MJ, Palsson BØ, Ruppin E:Network-based prediction of human tissue-specificmetabolism. Nat Biotechnol 2008, 26:1003-1010.
19. Gille C, Bolling C, Hoppe A, Bulik S, Hoffmann S, Hubner K,Karlstadt A, Ganeshan R, Konig M, Rother K et al.: HepatoNet1: acomprehensive metabolic reconstruction of the humanhepatocyte for the analysis of liver physiology. Mol Syst Biol2010, 6.
20. Chang RL, Xie L, Xie L, Bourne PE, Palsson B: Drug off-targeteffects predicted using structural analysis in the context of ametabolic network model. PLoS Comput Biol 2010, 6:e1000938.
21. Lewis NE, Schramm G, Bordbar A, Schellenberger J, Andersen MP,Cheng JK, Patel N, Yee A, Lewis RA, Eils R et al.: Large-scale insilico modeling of metabolic interactions between cell types inthe human brain. Nat Biotechnol 2010, 28:1279-1285.
Current Opinion in Pharmacology 2013, 13:778–785
22. Mardinoglu A, Agren R, Kampf C, Asplund A, Nookaew I,Jacobson P, Walley AJ, Froguel P, Carlsson LM, Uhlen M et al.:Integration of clinical data with a genome-scale metabolicmodel of the human adipocyte. Mol Syst Biol 2013, 9.
23. Rantanen A, Rousu J, Jouhten P, Zamboni N, Maaheimo H,Ukkonen E: An analytic and systematic framework forestimating metabolic flux ratios from 13C tracer experiments.BMC Bioinformatics 2008, 9:266.
24. Schuetz R, Zamboni N, Zampieri M, Heinemann M, Sauer U:Multidimensional optimality of microbial metabolism. Science2012, 336:601-604.
25. Fong SS, Joyce AR, Palsson BO: Parallel adaptive evolutioncultures of Escherichia coli lead to convergent growthphenotypes with different gene expression states. GenomeRes 2005, 15:1365-1372.
26. Burgard A, Maranas C: Optimization-based framework forinferring and testing hypothesized metabolic objectivefunctions. Biotechnol Bioeng 2003, 82:670-677.
27. Byrne D, Dumitriu A, Segre D: Comparative multi-goal tradeoffsin systems engineering of microbial metabolism. BMC Syst Biol2012, 6:127.
28. Bordbar A, Palsson BO: Using the reconstructed genome-scalehuman metabolic network to study physiology and pathology.J Intern Med 2012, 271:131-141.
29. Chavali AK, D’Auria KM, Hewlett EL, Pearson RD, Papin JA: Ametabolic network approach for the identification andprioritization of antimicrobial drug targets. Trends Microbiol2012, 20:113-123.
30. Swinney DC, Anthony J: How were new medicines discovered?Nat Rev Drug Discov 2011, 10:507-519.
31. Rask-Andersen M, Almen MS, Schioth HB: Trends in theexploitation of novel drug targets. Nat Rev Drug Discov 2011,10:579-590.
32. Bunnage ME: Getting pharmaceutical R&D back on target. NatChem Biol 2011, 7:335-339.
33. Knox C, Law V, Jewison T, Liu P, Ly S, Frolkis A, Pon A, Banco K,Mak C, Neveu V et al.: DrugBank 3.0: a comprehensive resourcefor ‘omics’ research on drugs. Nucleic Acids Res 2011,39:D1035-D1041.
34. Schuster S, Dandekar T, Fell DA: Detection of elementary fluxmodes in biochemical networks: a promising tool for pathwayanalysis and metabolic engineering. Trends Biotechnol 1999,17:53-60.
35. Oh YK, Palsson BO, Park SM, Schilling CH, Mahadevan R:Genome-scale reconstruction of metabolic network inbacillus subtilis based on high-throughput phenotyping andgene essentiality data. J Biol Chem 2007, 282:28791-28797.
36. Hopkins AL, Groom CR: The druggable genome. Nat Rev DrugDiscov 2002, 1:727-730.
37.�
Chavali AK, Blazier AS, Tlaxca JL, Jensen PA, Pearson RD,Papin JA: Metabolic network analysis predicts efficacy of FDA-approved drugs targeting the causative agent of a neglectedtropical disease. BMC Syst Biol 2012, 6:27.
Chavali et al. present a computational pipeline for repurposing alreadyFDA-approved drugs toward killing of the neglected tropical diseaseparasite, L. major. Genes predicted to be essential to L. major eitheralone or in combination are compared to target genes of existing drugs,and likely targets are then filtered based on several criteria such asdruggability and toxicity. This led to successful in vitro killing of L. majorby Halofantrine, an antimalarial agent, as well as synergistic killing by fourrepurposed drug pairings.
38. Shen Y, Liu J, Estiu G, Isin B, Ahn YY, Lee DS, Barabasi AL,Kapatral V, Wiest O, Oltvai ZN: Blueprint for antimicrobial hitdiscovery targeting metabolic networks. Proc Natl Acad Sci U SA 2010, 107:1082-1087.
39. Henriksen ST, Liu J, Estiu G, Oltvai ZN, Wiest O: Identification ofnovel bacterial histidine biosynthesis inhibitors using docking,ensemble rescoring, and whole-cell assays. Bioorg Med Chem2010, 18:5148-5156.
www.sciencedirect.com
Metabolically re-modeling the drug pipeline Oberhardt, Yizhak and Ruppin 785
40. Yeh I, Hanekamp T, Tsoka S, Karp PD, Altman RB: Computationalanalysis of Plasmodium falciparum metabolism: organizinggenomic information to facilitate drug discovery. Genome Res2004, 14:917-924.
41. Rahman SA, Schomburg D: Observing local and globalproperties of metabolic pathways: ‘load points’ and ‘chokepoints’ in the metabolic networks. Bioinformatics 2006,22:1767-1774.
42. Kim HU, Kim TY, Lee SY: Genome-scale metabolic networkanalysis and drug targeting of multi-drug resistant pathogenAcinetobacter baumannii AYE. Mol Biosyst 2010,6:339-348.
43. Kim PJ, Lee DY, Kim TY, Lee KH, Jeong H, Lee SY, Park S:Metabolite essentiality elucidates robustness of Escherichiacoli metabolism. Proc Natl Acad Sci U S A 2007,104:13638-13642.
44. Kim TY, Kim HU, Lee SY: Metabolite-centric approaches for thediscovery of antibacterials using genome-scale metabolicnetworks. Metab Eng 2010, 12:105-111.
45.��
Kim HU, Kim SY, Jeong H, Kim TY, Kim JJ, Choy HE, Yi KY,Rhee JH, Lee SY: Integrative genome-scale metabolic analysisof Vibrio vulnificus for drug targeting and discovery. Mol SystBiol 2011, 7:460.
This study presents a full pipeline for novel drug target development usingGSMMs along with structural data. Briefly, a GSMM for V. vulnificus isdeveloped and used to predict essential metabolites, that is, those whoseremoval fully obstruct biomass production. Next, the essential metabo-lites (and related enzymes) are filtered to remove those present in humanmetabolism. Finally, analogues of these metabolites were screenedagainst live cells, yielding a high-priority lead compound with promisinganti-bacterial properties.
46. Frezza C, Zheng L, Folger O, Rajagopalan KN, MacKenzie ED,Jerby L, Micaroni M, Chaneton B, Adam J, Hedley A et al.: Haemoxygenase is synthetically lethal with the tumour suppressorfumarate hydratase. Nature 2011, 477:225-228.
This article presents a metabolic model of the kidney that emulateshereditary leiomyomatosis and renal-cell cancer (HLRCC), both of whichare conditions caused by a germline mutation in the enzyme fumaratehydratase (FH). The authors use the GSMM to identify an enzyme in thehaem oxygenase pathway that kills FH-deficient cells while sparinghealthy cells. This target was validated experimentally in vitro in immor-talized kidney cancer cells.
47. Parsippany (Ed): The Use of Medicines in the United States:Review of 2011. NJ: IMS Institute for Healthcare Informatics; 2012.
48. Oberhardt MA, Palsson BO, Papin JA: Applications of genome-scale metabolic reconstructions. Mol Syst Biol 2009, 5:320.
49. Brynildsen MP, Winkler JA, Spina CS, MacDonald IC, Collins JJ:Potentiating antibacterial activity by predictably enhancingendogenous microbial ROS production. Nat Biotechnol 2013,31:160-165.
This study presents an ensemble modeling approach that integrates ROSproduction into a GSMM of E. coli, following the observation that ROSproduction enhances killing by antibiotics. The model is then used topredict gene deletions that increase ROS production within cells. Thedeletions are then shown in vitro to increase susceptibility of E. coli tooxidants and antibiotics. This study represents an exemplary use ofGSMMs to develop new adjuvant therapies to assist current antibiotics.
50. Neal EG, Chaffe H, Schwartz RH, Lawson MS, Edwards N,Fitzsimmons G, Whitney A, Cross JH: The ketogenic diet for thetreatment of childhood epilepsy: a randomised controlled trial.Lancet Neurol 2008, 7:500-506.
51. Bordbar A, Lewis NE, Schellenberger J, Palsson BO, Jamshidi N:Insight into human alveolar macrophage and M. tuberculosisinteractions via metabolic reconstructions. Mol Syst Biol 2010, 6.
52. Facchetti G, Zampieri M, Altafini C: Predicting andcharacterizing selective multiple drug treatments formetabolic diseases and cancer. BMC Syst Biol 2012, 6:115.
53. Oberhardt MA, Goldberg JB, Hogardt M, Papin JA: Metabolicnetwork analysis of Pseudomonas aeruginosa during chronic
www.sciencedirect.com
cystic fibrosis lung infection. J Bacteriol 2010,192:5534-5548.
54. Roberts SB, Robichaux JL, Chavali AK, Manque PA, Lee V,Lara AM, Papin JA, Buck GA: Proteomic and network analysischaracterize stage-specific metabolism in Trypanosomacruzi. BMC Syst Biol 2009, 3:52.
55. Huthmacher C, Hoppe A, Bulik S, Holzhutter HG: Antimalarialdrug targets in Plasmodium falciparum predicted by stage-specific metabolic network analysis. BMC Syst Biol 2010,4:120.
56. Lobel L, Sigal N, Borovok I, Ruppin E, Herskovits AA: Integrativegenomic analysis identifies isoleucine and CodY as regulatorsof Listeria monocytogenes virulence. PLoS Genet 2012,8:e1002887.
57. Rosenbloom KR, Dreszer TR, Long JC, Malladi VS, Sloan CA,Raney BJ, Cline MS, Karolchik D, Barber GP, Clawson H et al.:ENCODE whole-genome data in the UCSC Genome Browser:update 2012. Nucleic Acids Res 2012, 40:D912-D917.
58. Jamshidi N, Palsson BO: Systems biology of SNPs. Mol Syst Biol2006, 2:38.
59. Chandrasekaran S, Price ND: Probabilistic integrative modelingof genome-scale metabolic and regulatory networks inEscherichia coli and Mycobacterium tuberculosis. Proc NatlAcad Sci U S A 2010, 107:17845-17850.
60.�
Karr JR, Sanghvi JC, Macklin DN, Gutschow MV, Jacobs JM,Bolival B Jr, Assad-Garcia N, Glass JI, Covert MW: A whole-cellcomputational model predicts phenotype from genotype. Cell2012, 150:389-401.
In this article, the authors present the first ever whole-cell computationalmodel of a cell. The model captures the life cycle of the human pathogenMycoplasma genitalium, including every known gene function as well asall known molecular components and interactions. The model was vali-dated against a broad range of data, and provides insights into manycellular behaviors, including in vivo rates of protein-DNA association andan inverse relationship between the durations of DNA replication initiationand replication.
61. De Buck SS, Sinha VK, Fenu LA, Nijsen MJ, Mackie CE,Gilissen RA: Prediction of human pharmacokinetics usingphysiologically based modeling: a retrospective analysis of 26clinically tested drugs. Drug Metab Dispos 2007,35:1766-1780.
62. van Nood E, Vrieze A, Nieuwdorp M, Fuentes S, Zoetendal EG, deVos WM, Visser CE, Kuijper EJ, Bartelsman JF, Tijssen JG et al.:Duodenal infusion of donor feces for recurrent Clostridiumdifficile. N Engl J Med 2013, 368:407-415.
63. Freilich S, Zarecki R, Eilam O, Segal ES, Henry CS, Kupiec M,Gophna U, Sharan R, Ruppin E: Competitive and cooperativemetabolic interactions in bacterial communities. Nat Commun2011, 2:589.
64. Zhuang K, Ma E, Lovley DR, Mahadevan R: The design of long-term effective uranium bioremediation strategy using acommunity metabolic model. Biotechnol Bioeng 2012,109:2475-2483.
65. Kim TY, Sohn SB, Kim HU, Lee SY: Strategies for systems-levelmetabolic engineering. Biotechnol J 2008, 3:612-623.
66. Li L, Zhou X, Ching WK, Wang P: Predicting enzyme targets forcancer drugs by profiling human metabolic reactions in NCI-60 cell lines. BMC Bioinformatics 2010, 11:501.
67. Nam H, Lewis NE, Lerman JA, Lee DH, Chang RL, Kim D,Palsson BO: Network context and selection in the evolution toenzyme specificity. Science 2012, 337:1101-1104.
68. Pal C, Papp B, Lercher MJ, Csermely P, Oliver SG, Hurst LD:Chance and necessity in the evolution of minimal metabolicnetworks. Nature 2006, 440:667-670.
69. Yizhak K, Tuller T, Papp B, Ruppin E: Metabolic modeling ofendosymbiont genome reduction on a temporal scale. MolSyst Biol 2011, 7.
Current Opinion in Pharmacology 2013, 13:778–785