The Interplay Between Gene
Regulatory and Metabolic Reaction
Networks
Dr Andrzej M. Kierzek
Lecturer in Bioinformatics
Faculty of Health and Medical Sciences
University of Surrey
Guildford, GU2 7XH, UK
[email protected]
Phenotypic switching
• Diauxic shift. Escherichia coli cells chose
glucose from a mixture of carbon sources
available in the medium.
• Antibiotic production. Under stress, (e.g.
starvation) Streptomyces coelicolor produces
numerous complex organic substances that
kill other bacterial species in the
environment.
• Persistence. Mycobacterium tuberculosis
switches to very slow growth and can persist
inside macrophage cells for many years. It
can later switch to faster growth and cause
disease.
• Cancer. The cells of healthy tissue change
their behaviour, escape the control of the
organism and develop a tumor.
Single cell of
E. coli
S. coelicolor
producing
actinorhodin.
Tuberculosis
infection.
Dividing
leukemia
cells.
Phenotypic switching involves the
interplay between gene regulatory and
metabolic reaction networks.
Metabolic reaction network of E. coli
(EcoCyc)
Genes directly and indirectly
(network distance of 2 edges)
affected by Crp protein in E.
coli (RegulonDB).
Timescale separation between molecular
events occurring in gene regulation and
metabolism is used to justify consideration of
gene regulatory and metabolic reaction
networks as different levels of system
organisation.
Outline
I.
Modelling phenotypic switching without
separation of the system into different
organisation levels.
II. Modelling steady state of metabolic reaction
network using Flux Balance Analysis.
III. The interplay between gene regulatory and
metabolic reaction networks.
I. Modelling phenotypic switching
without separation of the system into
different organisation levels.
Enzyme activity, gene regulation, signal transduction, transport are all examples of
molecular interaction events and should not be arbitrarily divided into different
categories.
The model of glucose, lactose and
glycerol metabolism in E.coli
All network interactions are modeled as
chemical reaction events without splitting
them into arbitrary types (gene regulatory,
metabolic, signaling)
The model includes 94 substances and 120
reactions. For this well studied system a lot of
quantitative data was available in the
literature.
Substances represent promoters, transcripts,
proteins and small molecules.
Gene expression, enzymatic reactions,
transport and signal transduction cascade
(PTS system) are modeled.
Numbers of molecules and reaction rates
span 7 orders of magnitude. New algorithm
was formulated to allow simulation of such
systems.
Puchalka & Kierzek, BIOPHYSICAL JOURNAL, 86:1357-1372 2004
Stochastic phenotypic
switching
Balaban NQ, Merrin J,
Chait R, Kowalik L,
Leibler S. Bacterial
persistence as a
phenotypic switch.
Science. 2004
305(5690):1622-5.
“A fraction of a genetically homogeneous microbial
population may survive exposure to stress such as
antibiotic treatment. Unlike resistant mutants, cells
regrown from such persistent bacteria remain
sensitive to the antibiotic.”
Stochastic chemical kinetics
The system (reaction network)
s = (S1,...,SN)
molecular species
x = (X(t)1,...,X(t)N)
Xi(t) the number of molecules of Si present in
the reaction environment at time t.
r = (R1,...,RM)
chemical reactions
νμ = (ν1,...,νN)
state change vector of the reaction Rμ. Eg:
Rμ:S1+S2ÆS3, νμ=(-1,-1,1,0,...,0)
V
volume of the reaction environment
Stochastic chemical kinetics
Propensity function
•
Chemical reactions occur as a result of
random, reactive collisions between
molecules.
•
For every reaction Rμ, there exists a
propensity function aμ(x) such that:
aμ(x)dt = probability, given the state of
the system x, that one individual reaction
Rμ will occure somewhere inside the
reaction volume V in the next
infinitisemal time interval (t,t+dt)
(fundamental hypothesis).
•
Propensity function can be defined if the
system is well stirred or if the number of
non-reactive molecular collisions is much
larger than the number of reactive
molecular collisions.
Rm: A + B Æ AB
aμ(x) = cμ #A #B
cμ stochastic rate constant of Rμ [1/s]
Stochastic chemical kinetics
Exact stochastic simulation with Gillespie algorithm.
Set initial state of the system
x0=(X1(t0),...,XN(t0))
2.
For every reaction Rμ compute aμ(x)
3.
Compute the sum of propensity func.
a0 = ∑j=1,M aj(x)
4.
Randomly select reaction Rμ and waiting
time τ such that:
P(τ,μ)dτ = aμ exp(-a0τ) dτ
P(μ) = (aμ / a0)
P(τ) = a0 exp(-a0τ) dτ
5.
Update the system: x(t+τ) = x(t) + νμ
6.
Set simulation time to t + τ
7.
Go to 2.
100
'S'
'P'
'E'
'ES'
90
Number of molecules
1.
80
70
60
50
40
30
20
10
0
0
2
4
6
8
Time [s]
S+E->ES c = 1 1/s
ES->E+S c = 10 1/s
ES->P+E c = 1 1/s
Initial conditions: #S = 100; #E = 20
10
Stochastic chemical kinetics
Exact stochastic simulation with Gillespie algorithm.
Set initial state of the system
x0=(X1(t0),...,XN(t0))
2.
For every reaction Rμ compute aμ(x)
3.
Compute the sum of propensity func.
a0 = ∑j=1,M aj(x)
4.
Randomly select reaction Rμ and waiting
time τ such that:
P(τ,μ)dτ = aμ exp(-a0τ) dτ
P(μ) = (aμ / a0)
P(τ) = a0 exp(-a0τ) dτ
5.
Update the system: x(t+τ) = x(t) + νμ
6.
Set simulation time to t + τ
7.
Go to 2.
100
Number of molecules
1.
'S'
'P'
'E'
'ES'
80
60
40
20
0
0
2
4
6
8
Time [s]
S+E->ES c = 1 1/s
ES->E+S c = 10 1/s
ES->P+E c = 1 1/s
Initial conditions: #S = 100; #E = 20
10
Detailed kinetic model of prokaryotic gene expression.
Kinetic model of LacZ gene expression
Reaction
Stochastic
rate constant
[1/s]
Meaning
PLac+RNAP->PLacRNAP
0.17
RNA polymerase binding. RNAP – RNA polymerase. PLac –
promoter, PLacRNAP closed RNAP/promoter complex.
PLacRNAP->PLac+RNAP
10
RNA polymerase dissociation.
PLacRNAP->TrLacZ1
1
Closed complex isomerisation. TrLacZ1 – open RNAP/promoter
complex.
TrLacZ1->RbsLacZ+Plac+TrLacZ2
1
Promoter clearance. RBSLacZ – ribosome binding site,
TrLacZ2 – RNA polymerase elongating LacZ mRNA.
TrLacZ2->RNAP
0.015
mRNA chain elongation and RNAP release.
Ribosome+RbsLacZ ->RbsRibosome
0.17
Ribosome binding. Ribosome – ribosome molecule,
RbsRibosome – ribosome/RBS complex.
RbsRibosome ->Ribosome+RbsLacZ
0.45
Ribosome dissociation
RbsRibosome->TrRbsLacZ+RbsLacZ
0.4
Ribosome binding site clearance. TrRbsLacZ – ribosome
elongating LacZ protein chain.
TrRbsLacZ->LacZ
0.015
LacZ protein synthesis.
LacZ->dgrLacZ
6.42e-5
Protein degradation. dgrLacZ – inactive LacZ protein.
RbsLacZ->dgrRbsLacZ
0.3
Functional mRNA degradation. dgrRbsLacZ – inactive mRNA.
Kierzek A.M, et al. J. Biol. Chem. 276, 8165-8172 (2001)
Bacterial genes expressed under the control of weak promoters
show significant random fluctuations (bursts) in the number of
protein molecules.
Gene expression level is decreased by decreasing TRANSCRIPTION intiation frequency:
Kierzek A.M, et al. J. Biol. Chem. 276, 8165-8172 (2001).
Bacterial gene can be expressed at very low level, without
introducing large stochastic fluctuations if it is expressed under
the control of weak Ribosome Binding Site.
Gene expression level is decreased by decreasing TRANSLATION intiation frequency:
Kierzek A.M, et al. J. Biol. Chem. 276, 8165-8172 (2001).
Two major technical problems
1. Exact stochastic simulation is too time
consuming for systems including
intensive metabolic reactions
2. It is difficult to formulate and
parameterize a model of complex system
in terms of elementary reaction
mechanisms.
Hybrid simulations algorithms are constructed by combining exact
stochastic simulation, executed for the subset of “slow” reactions, with
one of the methods below, used to simulate fast reactions in the system.
τ-leap method:
Assumption: The propensity functions of ALL reactions in the system are so large that they will
not considerably change during the time τ .
X(t+τ) = X(t) + ∑j=1,...,M kj vj
kj = P (aj(x)τ)
Integration of the chemical Langevin equation:
Assumption: The propensity functions of ALL the reactions in the system are so large
that kj >> 1
X(t+τ) = X(t) + ∑j=1,...,M kj vj
kj = N (aj(x)τ, aj(x)τ)
Integration of the deterministic chemical kinetics by Euler method:
Assumption: Propensity functions of ALL the reactions in the system Æ ∞
X(t+τ) = X(t) + ∑j=1,...,M aj(x)τ vj
Maximal timestep method: Example of hybrid simulation algorithm.
J. Puchałka & A.M. Kierzek BIOPHYS. J., 86:1357-1372 2004
Numerical test: CPU
Xenon 2Ghz. Exact
stochastic simulation 124
min/trajectory. Maximal
timestep method 1
min/trajectory.
Benchmark model from
Kierzek, A.M. (2002)
Bioinformatics 18 470481
Maximal timestep method integrates Gibson & Bruck algorithm and Gillespie’s tau-leap method.
1.
Set maximal timestep κ = 10-3 s.
2.
Divide reactions into “slow” and “fast” reaction subset Sslow, Sfast according to maximal possible
number of reaction occurrences (<100) and probability of reaction to occur within κ (<10-4).
3.
For each reaction Rμ compute propensity function a μ(x) and the time τ μ at which this reaction is going
to occur.
4.
Find Rmin from Sslow such that: τmin = min(τ1,.., τO).
5.
If τmin–t < κ: Execute elementary reaction Rmin; For each Rj from Sfast compute kj=P(aj(x)(τmin-t)) and
update the state of the system: X(τmin) = X(t) + ∑j=O+1,...,M kj vj
6.
Else: Do not execute any reactions from Sslow; For each Rj from Sfast compute kj=P(aj(x) κ) and update
the state of the system: X(t+κ) = X(t) + ∑j=O+1,...,M kj vj
Quasi steady state approximation
Example (Rao CV, Arkin AP, J CHEM PHYS 118 (11): 4999-5010 2003):
R1:S1 + S2 Æ S3
R2:S3 Æ S2 + S1
R3:S3 Æ S2 + S4
Assumption: dP(z|y;t)/dt = 0 z = (X2(t),X3(t)) y =(X1(t),X4(t))
Then the proces can be modeled by the following single reaction:
R1:S1 Æ S4
a1(X1(t)) = Vmax X1(t) / ( Km + X1(t) )
Quasi steady state approximation
Example of complex reaction mechanism used in E. coli diauxic shift model.
Puchalka & Kierzek, BIOPHYSICAL JOURNAL, 86:1357-1372 2004
The model of glucose, lactose and
glycerol metabolism in E.coli
All network interactions are modeled as
chemical reaction events without splitting
them into arbitrary types (gene regulatory,
metabolic, signaling)
The model includes 94 substances and 120
reactions. For this well studied system a lot of
quantitative data was available in the
literature.
Substances represent promoters, transcripts,
proteins and small molecules.
Gene expression, enzymatic reactions,
transport and signal transduction cascade
(PTS system) are modeled.
Numbers of molecules and reaction rates
span 7 orders of magnitude. New algorithm
was formulated to allow simulation of such
systems.
Puchalka & Kierzek, BIOPHYSICAL JOURNAL, 86:1357-1372 2004
Population heterogeneity during growth on the
mixture of carbon sources
Switch from glucose
to the mixture of
lactose and glycerol
A) LacZ protein
B) GlpF protein
C) cAMP
D) external glycerol
Puchalka & Kierzek, BIOPHYSICAL JOURNAL, 86:1357-1372 2004
Conclusions
Stochastic fluctuations in gene regulatory
network may propagate to the level of
metabolic network and result in
epigenetically inheritable changes in cell
physiology.
Detailed kinetic models of selected, well
known model systems uncover fundamental
dynamic properties of molecular interaction
networks in the living cell.
Detailed kinetic models involve too many
unknown parameters to address the
challenge of using high throughput
experimental data of molecular biology to
predict behaviour of the particular cell/tissue
under experimental conditions of interest.
Shall we wait next 100 years until we have enough of quantitative
enough data to build detailed dynamic models addressing this:
... and besides, can we ever have enough detailed information? The model n+1
can always be more reductionistic than model n. Some people spend months of
CPU time to compute from the first principles that atomic mass of proton is 1.
II. Modelling steady state of metabolic
reaction network using Flux Balance
Analysis.
It is possible to analyse steady state flux distributions in genome scale metabolic reaction
networks. Moreover, detailed dynamic simulations show that metabolic reactions quickly
reach steady state and that the gene regulatory and signal transduction processes switch
between different steady states of metabolic reaction network.
Flux Balance Analysis – a constraint based
approach
Bxt
Find maximal dX/dt if the following
constraints are satisfied:
Cxt
dX
= F5 + F8
dt
0 < F1 ≤ 100
Value to be maximised
(objective function)
Axt
growth
Dxt
Adapted from FluxAnalyzer software (Steffen
Klamt,MPI Magdeburg)
The linear programming algorithm finds
the largest possible value of vx.
However, there are many possible
values of fluxes (F1,..,F8) that result in
the same maximal value of vx.
dAxt
dt
dDxt
dt
dBxt
dt
dCxt
dt
0 < F3 ≤ 100
= − F1
= − F6
= F3
Transport of
extracellular
(external,
unbalanced)
metabolites.
= − F4
0 = F2 − F3
0 = F2 + F4 − F5
0 = F7 − F8
− 100 < F4 ≤ 100
0 < F5 ≤ 100
0 < F6 ≤ 100
0 < F7 ≤ 100
0 = F1 − F2 − F5 − F7
0 = F6 − F7
0 < F2 ≤ 100
0 < F8 ≤ 100
Minimal and maximal
reaction capacities
(bounds). R4 is the only
reversible reaction in the
system.
Steady state (flux balance)
assumption for intracellular
(internal) metabolites.
Chemostat can be used to force cellular
metabolism to operate under steady state.
RP
M
O2
0
C
pH
Analysis of microarray based gene
essentiality screen in the context of the
constraint-based model of TB bacillus
metabolism.
GSMN-TB: a web-based genome-scale network model of
Mycobacterium tuberculosis metabolism.
Dany JV Beste*, Tracy Hooper*, Graham Stewart, Bhushan Bonde,
Claudio Avignone-Rossa, Michael E Bushell, Paul Wheeler, Steffen
Klamt, Andrzej M Kierzek#, Johnjoe McFadden#
Genome Biology 2007, 8(5):R89
* Joint first authors
# Joint senior authors
Web software available at: http://sysbio.sbs.surrey.ac.uk/
GSMN-TB: Genome Scale Metabolic Reaction
Network of Mycobacterium Tuberculosis
Statistics of the GSMN-TB model
Reaction Class
Number
Enzymatic conversions
723
Transport reactions
126
Total number of reactions
849
Orphan reactions
210
Genes
726
Internal metabolites
638
External metabolites
101
Total number of metabolites
739
Biomass formula of GSMN-TB model
0.214 PROTEIN + 0.036 RNA + 0.022 DNA + 0.050
SMALLMOLECULES + 0.006 PE + 0.016 TAGbio + 0.040 PIMS +
0.186 LAM + 0.208 MAPC + 0.035 P-L-GLX + 0.007 CL + 0.054
LM + 0.001 TREHALOSEDIMYCOLATE + 0.001
TREHALOSEMONOMYCOLATE + 0.001
POLYACYLTREHALOSE + 0.001 DIACYLTREHALOSE + 0.0001
MPD + 0.002 DIM + 0.029 PGL-TB + 0.005 SL-1 + 0.1 GLUCAN +
47 ATP = 1 BIOMASS + 47 ADP + 47 PI
Modelling of M. tuberculosis metabolism by
maximisation of growth rate seems to be a grotesque
idea because M. tuberculosis is a dangerous
pathogen due to its particularly slow growth.
However,
If the maximal theoretical flux towards biomass equals 0
than the M. tuberculosis culture cannot grow.
If the maximal theoretical flux towards any metabolite in the
system changes as the result of gene activation or other
perturbation, the perturbation is likely to affect synthesis of
the metabolite.
Alternative, feasible flux distributions represent metabolic
states accessible to stochastic phenotypic switching.
..... moreover, when TB grows in chemostat ....
minimal glycerol uptake flux
(mmol/g DW/h )
1.2
predicted
glycerol
consumption
rate
1
0.8
0.6
0.4
measured
glycerol
consumption
rate
0.2
0
0.005
0.015
0.025
growth rate (1/h )
0.035
..... minimisation of carbon source consumption
leads to interesting research hypothesis.
minimal glycerol uptake flux
(mmol/g DW/h )
1.2
………
Predicted glycerol
consumption when
TWEEN hydrolisis to
oleic acid and oleic
acid consumption
are taken into
account.
1
0.8
0.6
0.4
0.2
0
0.005
0.015
0.025
growth rate (1/h )
0.035
Screening for essential genes by
Transposon Site Hybridisation (TraSH)
EZ:TN BCG mutant
input pool
(2500 mutants)
Mutant
output pool
Label transposon
flanking regions by
PCR incorporation of
Cy3-dCTP
Label transposon
flanking regions by
PCR incorporation of
Cy5-dCTP
Abundance of mutants in output pool
is quantified relative to abundance in
the input pool by co-hybridisation of
labelled transposon flanking regions
Comparison of gene essentiality
prediction with TraSH data.
For each gene in the model {
Constrain all fluxes that require this gene to 0.
Run FBA.
If the flux towards BIOMASS is less than cut-off {
the gene is essential.
} else {
the gene is non-essential
}
}
Compare list of predicted essential genes with the list of genes with TraSH
microarray signal ratio (insertion probe/genomic probe) lower than cut-off.
Classify genes to the following categories.
TP
FP
TN
FN
essential in the
essential in the
non-essential in
non-essential in
model and essential in experiment.
model, non-essential in experiment.
the model, non essential in experiment
the model, essential in experiment
To study the interplay between single gene activity and global metabolism of
TB bacillus in vitro we used the protocol shown above to compare predictions
of GSMN-TB model with TrASH data of Sassetti et al. Mol Microbiol 2003,
48:77-84.
Receiver Operating Characteristics
(ROC) of gene essentiality prediction.
Each ROC curve shows 100 points corresponding to
sensitivity and specificity of the model predictions
obtained for growth rate thresholds varying in the
range from 0.0 to 0.1 (increment 0.001). The growth
rate threshold has no effect on prediction accuracy.
The LP optimisation is effectively used as a
qualitative test of BIOMASS producibility and it is
irrelevant whether TB bacillus grows with maximal
rate or not.
Different curves correspond to TraSH ratio
thresholds of 0.05, 0.1, 0.2, 0.6, 1. The TraSH ratio
cutoff has considerable influence on prediction
accurracy.
The best ROC curve corresponds to the following
prediction scores: Sensitivity 71%, Specificity
80%, Correct predictions 78%.
Sensitivity = TP/(TP + FN)
Specificity = TN/(TN+FP)
A test of prediction significance
independent on microarray signal
threshold.
Distributions of the raw TraSH signal ratios
for genes present in the model. Blue line
shows distribution for genes that were
predicted by the model to be essential for
growth. Red line shows distribution of TraSH
ratio among genes predicted to be
nonessential for BIOMASS production.
Medians of two distributions
significantly different by means of nonparametric U-test (p-value < 2e-16)
Extension to expression data: Perform
this analysis for every metabolite in the
network, not only for BIOMASS. For each
metabolite in the network identify genes that
influence and do not influence its
producibility. Compare microarray signal for
both groups of genes and decide whether
metabolite is differentially affected by
gene expression program.
GSMN-TB: WWW server for constrained based modelling of
TB-bacillus metabolism (http://sysbio.sbs.surrey.ac.uk)
The list of four simulation
protocols.
USER
The choice of one of four simulations: 1)
Computation of maximal growth rate. 2) Flux
Variability Analysis.
3) Reaction essentiality
scan. 4) Gene essentiality prediction.
M. Tuberculosis genome
scale metabolic reaction
model with default set of
constraints.
RESULTS
The input form for media conditions,
objective function and the gene to be
inactivated.
Constraints, objective
function and genotype
defined by the user.
Linear programming
computations
HTML formatting of
results file.
SERVER
Results file
III. The interplay between gene
regulatory and metabolic reaction
networks.
Metabolism can be modelled on genome scale but the global metabolic flux distribution
cannot be easily measured. Global state of gene regulatory network can be measured
with cDNA microarrays but the genome scale gene regulatory network cannot be
modelled. Therefore, it is useful to analyse experimental gene expression profiles in the
context of genome scale metabolic reaction network model.
Analysis of Differentially Affected Metabolites
(ADAM) in S. coelicolor.
This research is a part of BBSRC funded project:
Title: The interplay between two-component signal transduction systems and the genome
scale metabolic network of Streptomyces coelicolor
Principal Investigator: Andrzej M. Kierzek
Co-applicants: Colin Smith, Claudio Avignone-Rossa, Michael Bushell.
Postdoctoral researchers: Nick Allenby, John Jim
Funding: 714,000 GBP, 48 months
Aim: To investigate induction of secondary metabolism by PhoPR and AbsA1A2 two
component systems in S.coelicolor.
Experimental methods: Chemostat and fermentor cultures, microarray expression
profiling, ChIP on chip, global mRNA decay, enzyme assays, metabolite profiling.
Modelling: Experimental data will be integrated by constraint-based modelling of global
metabolism in S. coelicolor, including dynamic, quasi steady-state simulations.
Monitoring dynamics of gene expression and
antibiotic production in fermentor cultures of S.
coelicolor
6
Head of experimental group:
Colin Smith
5
WT_A
Dry Weight (g/L)
4
WT_B
3
PhoP_A
2
PhoP_B
Experimental work:
Nicola Cattini
Nick Allenby
Giselda Bucca
Microarray data analysis:
Emma Laing
1
0
0
50
100
150
Time (h)
Phosphate limited fermentor cultures of S. coelicolor wild
type and PhoP mutant have been set. Samples have
been taken at 31h, 38h, 42h, 46h, 60h, 81h. The gene
expression has been assayed by cDNA microarrays and
metabolite concentration in the media has been
measured.
Analysis of data in context of
GSMN model:
Andrzej Kierzek
Genome Scale Metabolic Reaction network of
Streptomycete (Borodina I, Krabben P, Nielsen J.
Genome Res. 2005 Jun;15(6):820-9.)
971 reactions, 500 metabolites, 711 genes
Simulation of metabolite producibility
Unlimited glucose
uptake
Unlimited CO2 and
organic acid secretion
Unlimited nitrate
uptake
Growth rate
Unlimited sulphate
uptake
GSMN Model
Limiting phosphate
uptake, 0.13 mmol/
gDW/ h
Unlimited oxygen
uptake
Inactivate selected gene
Compute maximal
theoretical
synthesis rate of
selected
metabolite.
In silico experiment.
For each gene g and metabolite m:
Glucose, phosphate (limiting), sulphate,
NH3, O2
Glucose, phosphate (limiting),
sulphate, NH3, O2
GSMN, gene g
knock-out
GSMN, wild type
Metabolite m
Metabolite m
Does inactivation of g change maximal
possible flux towards m?
Producibility plot.
600
500
metabolite ID
400
Metabolite
affected by
many genes
(GLYCOLATE)
300
200
100
0
0
1000
2000
3000
4000
gene ID
5000
6000
7000
8000
Essential gene that affects most of
metabolites (SCO4738).
Producibility plot for S. coelicolor growing on mininimal medium. Each dot represent a genemetabolite pair such that inactivation of the gene affects production of the metabolite. Genes
are ordered according to their position on the chromosome.
Analysis of Differentially Affected Metabolites
(ADAM) in S. coelicolor.
For each metabolite m create two groups of
numbers:
Group 1
Group 2
Expression fold
changes for genes
that influence
production of m
Expression fold
changes for genes
that do not influence
production of m
Compute p-value for the null hypothesis that
medians of these two groups are equal, using
Mann-Whitney test.
Analysis of Differentially Affected Metabolites
(ADAM) in S. coelicolor.
Example result:
Changes in metabolite producibility are observed
only between t=38h and t=46h
All three antibiotics
included into the GSMN
model are on the list of
differentially affected
metabolites !!!!!
Genes which expression most affects metabolites
which producibility changes between t= 38h and
t=46h
Differential expression of genes
marked in yellow has been
validated by QRT-PCR.
Both metabolite and gene lists
suggest interesting scenario of
events involving the interaction
between energy metabolism and
antibiotic production.
Analysis of Differentially Affected
Metabolites is able to identify
interesting genes which microarray
signal ratio is too low to be
considered significant by standard
approaches.
gene
number of phoP/WT signal
affected
ratio
metabolites
Analysis of Differentially Affected
Metabolites in cancer cells ?????
Cancer cells undergo major changes in metabolism. The classical
chemotherapy drugs such as Methotrexate act by inhibiting metabolic enzymes
which are required for metabolism of cancer cells, but are less active in healthy
tissues (eg: DHFR).
Analysis of changes in gene expression program of cancer cells with respect to
healthy tissue is one of the major applications of cDNA microarrays.
The first reconstruction of the human genome scale metabolic reaction network
has been published in January 2007 by Palsson’s group and made fully
available in BiGG database (Proc Natl Acad. Sci U S A 104(6):1777-82
(2007)).
Analysis of Differentially Affected Metabolites in cancer cells could exploit
microarray datasets and provide insight into the global metabolic changes in
cancers cells. Previously unnoticed genes which expression changes in cancer
cells could also be identified.
Analysis of Differentially Affected Metabolites in
cancer cells ?????
The human GSMN model
is available in SBML
format. SBML file has
been easily converted into
the file format of our
constraint-based modelling
software.
SBML has become a true standard for exchange of complex biochemical reaction
network models.
Analysis of Differentially Affected
Metabolites in cancer cells ?????
The human GSMN model contains 3311 reactions, 2766 metabolites and 1496
genes. This makes computation of producibility plot time consuming.
Elimination of metabolites which are not producible and reactions which are not
active in “wild type” allows reduction of the model to 2496 reactions, 1957
metabolites and 684 genes.
The human GSMN contains 404 external metabolites. After model reduction
there are still 241 of them. Inactivation of genes in general does not influence
metabolism because most of pathway intermediates can be transported.
To perform ADAM analysis on human GSMN and expression profiles of tumor
tissues we have to decide which of 404 external metabolites are not actually
available in excess in the cancer’s cell environment.
Conclusions
1. Detailed kinetic models of selected, well known model systems uncover fundamental
dynamic properties of molecular interaction networks in the living cell. For example
dynamic simulations show that, stochastic fluctuations in gene expression may
propagate to the level of metabolic processes and result in epigenetically inheritable
changes in cell physiology.
2. Detailed kinetic models involve too many unknown parameters to address the
challenge of using high throughput experimental data of molecular biology to predict
behaviour of the particular cell/tissue under experimental conditions of interest.
3. The GSMN-TB model (http://sysbio.sbs.surrey.ac.uk/) achieves good results in
reproducing results of Transposon Site Hybridisation in TB bacillus. Predictive power
of the model can be demonstrated without using arbitrary threshold of microarray
signal.
4. Analysis of Differentially Affected Metabolites (ADAM) explores microarray data in the
context of Genome Scale Metabolic Network model and provides insight into
metabolic state of the cell determined by gene expression program under given
experimental conditions.
5. Definition of boundary conditions is the major problem in using human GSMN to
analysis of cancer metabolism.