Bioinformatics Euroschool CMBI 2005

Systems Biology

Bas Teusink
Something old? Something new?
 Lecture
What is Systems Biology?
Dynamical systems theory
Metabolic control analysis

Methods and models in Systems Biology

Data-driven approaches: example from mice
Knowledge-driven approaches: substrate-accelerated death in
microorganisms, , and humans
Hybrid approaches: the future?
 Systems concept
A system represents a set of components together with
the relations connecting them to form a whole unity
Defining a system divides reality into the system itself
and its environment
The number of interconnections within a system is
greater than the number of connections with the
environment
Systems can include other systems as part of their
construction
allows complex systems to be put together from known simple
ones (system of systems)
concept of modularity!
 Systems Biology

New? NO and YES

Systems theory and theoretical biology are old

Experimental and computational possibilities are

new
 Omics-revolution shifts paradigm to
complete systems

-integrative bioinformatics
-(network) modelling
 Systems Theory
The behaviour of a system depends on:

Properties of the components of the system

(can be subsystems)
The interactions between the components

THUS:
You cannot understand a system via pure
reductionism (studying the components in
isolation)
 Systems behaviour: oscillations

700 8.5

650 7.5
NAD(P)H fluorescenc

concentration (mM)
600 6.5

heat flux (W)

550 Glc-6P 5.5

500 4.5
4 mM KCN
450 3.5
20 mM glucose
Fru-
400 1,6bP 2.5
8 8.5 9 9.5
0 10 20
time (min)
Mixing systems:
Dynamical systems theory, bifurcation theory, chaos theory

2D 3D

stable oscillations chaos:

attractor: limit cycle attractor: .
 Simple example of systems behaviour: steady state

1.8

1.6
Isolated enzyme:
1.4

S 1 P
enzyme activity
1.2

v_isolated
1
v_system
0.8

0.6
Enzyme in system:
0.4

S 1 P 2 X 0.2

0
0 2 4 6 8 10 12
[E1]
Metabolic control analysis: a useful theory for formalizing
control and regulation

Enzyme in system:

S 1 P 2 X 2

1.8

1.6

1.4

enzyme activity
Control: 1.2

v_isolated
1
v_system
0.8

J %change in flux J 0.6

Ce1 = %change in e1
0.4

0.2

0
0 2 4 6 8 10 12
[E1]

ei Cei =
J
1
Metabolic control analysis: a useful theory for formalizing
control and regulation

S 1 P 2 X

How important is feedback via X for regulating the flux?

J %change in flux J %change in e1 J

RX = = Ce1
%change in X %change in X
 Holism vs. reductionism
Systems biology seems the opposite
approach of reductionism ...
But systems biology is not holistic!
Systems biology requires detailed
information of:
the components of the system
the interactions
This allows building models of the system
 models: different questions,
different models

-summary of existing knowledge

-tool for interpretation of data

-tool for prediction of behaviour/ rational design

-hidden-Markov models, black-white-grey models, core

models, kinetic models, genome-scale models, Boolean
models
 What is Systems Biology?

Systems Biology is seen as the next step after

Genomics, as the discipline that collects and
manages all the generated data and subsequently
integrates them and interprets them in terms of
biological models.

-integrative bioinformatics
-engineering
-statistics
-modeling/systems theory
-biology
 Lecture
What is Systems Biology?
Dynamical systems theory
Metabolic control analysis

Methods and models in Systems Biology

Data-driven approaches: example from mice
Knowledge-driven approaches: substrate-accelerated death in
microorganisms, , and humans
Hybrid approaches: the future?
 Data driven: How to analyze the data?
Different ways to Rome

-Clustering

-Multivariate analysis (pattern recognition techniques)

-Correlation analysis

-Network reconstruction

-visualisation
Clustering: need a measure of
similarity/distance

Data from Garber et al.

PNAS (98), 2001.
just visualisation of data is also useful...
glucose

ribose
 Mice on a high fat diet for 20 weeks

WT VLDLR-/-

3 dietary fatty acids (fish oil, trans FA, CLA)

1 drug (fenofibrate)

Proteomics and metabolomics

Goudriaan et al. ATVB (2001)
Figure 2

e
rat

rat
fib

fib
il

il
ho

ho
LIVER CYTOPLASM

no

no
A

A
A

A
Fis

Fis
CL

Fe

CL

Fe
TV

TV
glucose Galactokinase Senescence marker protein 30 oxidation
metabolism Ketohexokinase Selenium binding protein 56kDa and aging
Fructose biphosphatase 1 Selenium binding protein 56kDa
Malate dehydrogenase T-complex protein 1, beta subunitt
Alpha enolase Peroxyredoxin 6
Alpha enolase Glutathione S-transferase Mu 2
Thriosephophate isomerase Gluthatione S-transferase Mu 1
Phosphoglycerate mutase Heat shock protein 74 kDa
Isocitrate dehydrogenase Epoxide hydrolase
Adenosine kinase Catalase
Adenosine kinase Aldehyde dehydrogenase
lipid Long chain acetyl-CoA dehydrogenase Aldehyde dehydrogenase
metabolism Long chain acyl-CoA thioester hydrolase Aldehyde dehydrogenase
Long chain acyl-CoA thioester hydrolase L-gulonolactone oxidase
Mitochondrial long-chain acyl-CoA thioester Thioether S-methyltransferase homocysteine
Acyl-CoA thioester hydrolase Adenosylhomocysteinase metabolism
Acyl-CoA thioesterase Glycine-N-methyltransferase
Apolipoprotein E Sepiapterin reductase BH4
Apolipoprotein A1 precursor Phenylanaline-4-hydroxylase metabolism
Adipophilin Delta-aminoevulinate dehydratase haem
2-hydroxyphytanoyl-Coa lyase Hydroxymethylbiliane biosynthesis
CTP:phosphocholine cytidyltransferase b2 Pyrophosphatase energy
Isovaleryl CoA dehydrogenase Nicotinate-nucleotide pyrophosphorylase metabolism
Phosphatidylethanolamine-binding protein N-sulfotransferase enzymes
Hypothetical GDSL-like Lipase/ acylhydrolase Proteasome beta subunit
Annexin A5 Nucleoside diphosphate kinase B
protein Formimimo cyclodeaminase Guanidinoacetate N-methyltransferase
metabolism Ornithine aminotransferase Purine nucleoside phosphorylase
Glutamine synthetase Ubiquitin / ribosomal protein CEP52
Glutamine synthetase Tetranectin carcinogenesis marker
Histidine ammonia-lyase Tubulin alpha-2 chain structural proteins
Arginase-1liver
3-hydroxyanthranilate 3,4-dioxygenase
Cysteine sulfinic acid decarboxylase >200% >150% >125% >100% <100% <75% <50% <10%
 Multivariate analysis: Principle
component analysis

LIVER CYTOPLASM LIVER MEMBRANE

1.0 0.6

0.8 0.4 TVA

Fenofibrate
0.6 0.2 Fish oil
PC 2 (20.94%)

CLA
0.4 0.0

PC 2 (30.72%)
Fenofibrate
0.2 -0.2

CLA
0.0 -0.4 Control

-0.2 Fish oil -0.6

TVA
Control
-0.4 -0.8
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 -0.6 -0.4 -0.2 0.0 0.2 0.4
PC 1 (47.47%) PC 1 (40.68%)
Correlation analysis: example from mice 1
2
3
Glutathione S-transferase Mu 2
Long chain acyl-CoA thioester hydrolase B
Cysteine sulfinic acid decarboxylase
41 4 Long chain acyl-CoA thioester hydrolase A
37 5 Catalase
6 liverFC
38 7 3-hydroxyanthranilate 3,4-dioxygenase
39 8 Adenosine kinase
40
36
9 Peroxyredoxin 6
10 plasma cholesterol
29 28 35
34
11 Senescence marker protein 30
13 liverCE
33
30 14 Phenylanaline-4-hydroxylase
27
15 Glycine-N-methyltransferase
25 31 16 Ornithine aminotransferase
26
24
17 liver weigth
23 18 Tetranectin
32
19 Adenosylhomocysteinase
22
20 Aldehyde dehydrogenase
21 Pyrophosphatase
22 Tubulin alpha-2 chain
20 9
Metabolic Syndrome X ! 23 Delta-aminoevulinate dehydratase
21
24 Selenium binding protein 56kDa
19 7
8
25 Malate dehydrogenase
6
26 Thioether S-methyltransferase
27 Formimimo cyclodeaminase
28 Nicotinate-nucleotide pyrophosphorylase
10 5 29 T-complex protein 1, beta subunit
4
30 Fructose biphosphatase 1
13 11 31 plasma FFA
32 Glutamine synthase
2 33 plasma glucose
14
3 34 Ketohexokinase
35 Sepiapterin reductase
36 Alpha enolase
16 17 18 1
37 Phosphatidylethanolamine-binding protein
15
38 plasma TG
39 liver TG
40 Nucleoside diphosphate kinase B
Good for fishing, but tricky 41 Proteasome beta subunit
 How to reconstruct? That is the question
m samples l signals

m samples

n components
n components

l signals
= .

Data = interaction signal

-1 +2

Key problem: no unique solution

 Lecture
What is Systems Biology?
Dynamical systems theory
Metabolic control analysis

Methods and models in Systems Biology

Data-driven approaches: example from mice
Knowledge-driven approaches: substrate-accelerated death in
microorganisms, , and humans
Hybrid approaches: the future?
Biological knowledge
most knowledge-based models are
concerned with metabolic systems
(and signalling cascades)

Feedback loop

genes
enzyme
transcription translation
mRNA

Feedback
Degradation loop

Metabolic Reactions
Biological knowledge-based modeling

precursor 1 M1 2 M2 3 Amino acid

Ordinary differential equations:

v1 = 1(Mi, p)
dM1 / dt = v1 v2
dM2 / dt = v2 v3 v2 = 2(Mi, p)
v3 = 3(Mi, p)

Small systems!
Biological knowledge-based modeling:

Yeast glycolysis
Substrate-accelerated death
)
60

50

40

tps1 -
30

20

10 tps1
F ructose 1,6-bisphosphate (mM)

0
0 1 2 3 4 5 10 20 30 40 50

time (mi n)

3
The biological story:
ATP (mM)

2
No brake: all ATP consumed at
1 hexokinase
tps1
0
0 1 2 3 4 5 10 20 30 40 50
time (min)
Turbo design: optimal strategy for ATP production
In silico tps1 phenotype
danger of turbo design:
If
-the activity of enzyme 1 exceeds the capacity of enzyme 2
-enzyme 1 is insensitive to intermediate I

Then
intermediate I will accumulate, despite a link via adenine
nucleotides

ATP 2 ATP

1 I I II I 2

extreme case: v1 = 2 v2
Turbo explanation of the tps1 phenotype

-Tps1 serves to control the turbo

engine of glycolysis
-Well-designed control structure: both
turbo and checked
liver: portal blood: high
postprandial peaks in glucose
HK inhibited by regulatory
protein -

muscle: arterial
blood: glucose
homeostasis
HK inhibited by -
product glucose 6P
Different niches, different purposes, different regulations

yeast: extremely high variation in glucose:

HK inhibited by regulatory protein

liver: high variation in glucose:

HK inhibited by regulatory protein

muscle: slight variation in glucose:

HK inhibited by product glucose 6P

beta cells: sensing of glucose:

HK not inhibited at all, but low activity

trypanosoma brucei: no turbo design:

HK not inhibited at all, normal activity
 conclusion:

knowledge-based modeling gives fundamental insight, but

of relatively small systems

how to proceed to genomes and functional genomics data?

hybrid approaches:
networks and states on a genome scale
Genome-scale metabolic models: constraint-
based modeling

precursor 1 M1 2 M2 3 Amino acid

Ordinary differential equations:

v1 = 1(Mi, p)
dM1 / dt = v1 v2 = 0
dM2 / dt = v2 v3 = 0 v2 = 2(Mi, p)
v3 = 3(Mi, p)

Large systems!
Constraint-based modeling
v1 v2
S

Kinetic model: With 1 constraint: With 2 constraints:

[S], vmax and Km Steady state: v1= v2 Steady state: v1= v2
known Capacity (v1max)

v2 v2 v2

v1 v1 vmax v1
Unique solution Solution space Limited solution space
The Model Development Process

Physiology

Genomics
biochemistry

gene

protein

reaction
Stoichiometric models: constraint-
based analysis
An Optimal
Solution EP or EM

Network
Elementary modes:
Allowable
Reconstruction
Genomics, Physiology Solution Space
1
what are all the
possibilities?
and Biochemistry 2

Application of v2 3
Constraints c
b

4
v1
a
v3
Energy and
5
Biomass Constituents

6
1. Linear Optimization (MILP)
2. Elementary Modes/Extreme Pathways
3. Phenotypic Phase Plane Analysis
4. Gene Deletions
5. Finding Objective Functions
6. Full Phase Space Sampling
elementary modes

Recipe for linking

genome data to
transcriptome data
Stoichiometric models: constraint-
based analysis
An Optimal
Solution EP or EM

Network
Biology vs
Allowable
Reconstruction
Genomics, Physiology Solution Space
1
engineering: what is
the objective function?
and Biochemistry 2

Application of v2 3
Constraints c
b

v1
4
a
Evolution &
Energy and
Biomass Constituents
v3
5
microbes: optimal
biomass formation
6
1. Linear Optimization (MILP)
2. Elementary Modes/Extreme Pathways
3. Phenotypic Phase Plane Analysis
4. Gene Deletions
5. Finding Objective Functions
6. Full Phase Space Sampling
Constraint-based modeling:
predictions of microbes in evolution

T = 37 C

T = 30 C
Flux balance analysis:
optimal flux distribution

200
Corr.Coeff.=0.97
WT (FBA)
FBA v ( (Theor.) 150
P-value=10
C 0.4
-11
8
7

predictions, 9
)

following 100

3
10

Edwards and
Fluxes

1413 11
50
2 12 1

Palsson 456
16

1517
PNAS, 2000
0

0 50 100 150 200

vi (exper)
Fluxes (Experim.)
(percent, relative to glucose uptake)

Data from Uwe Sauers lab

(Emmerling et al., J.Bacteriol.2002)
Alternative for short term adaptation
MOMA: Minimalization Of Metabolic Adjustments

Use the strategy of

homeostasis to solve
underdetermined
system
Concept:

perturbation adaptation

response: regulation

State I State II

If there is a purpose for the regulatory design, we can predict state II without
knowing the details of the underlying regulation
How to move forward? Integrate data-driven
and knowledge-based modelling
Systems Biology: an iterative process

Experimental design and

method development

Hypothesis and prediction Experimental protocols

Data management/
Bioinformatics

Experimental data
Model building acquisition

Experimental data
 The Genomics Demand for
Modeling & Simulation

High Throughput Data Cellular Complexity

Data volume & cellular

complexity demand
formulation of in silico models

Model-Driven design, prototyping, and manufacturing

?
Acknowledgements and collaborations

Willem de Vos
Astrid Mars
Douwe Molenaar Magda Faijes
Arno Wegkamp Christof Francke Steven Rothuizen
Marc Stevens Richard Notebaart
Mariela Serrano Frank van Enckevort
Filipe Santos Jos Boekhorst
Jeroen Hugenholtz Robert Kerkhoven
Michiel Kleerebezem Michiel Wels Systems Biology of LAB
Eddy Smid Roland Siezen
Oscar Kuipers
Jan Kok
Hans Westerhoff
Stanley Brul
Timo Breit
Bert Poolman
Mariet vd Werf
Sef Heijnen Age Smilde
Walter van Gulik
Wouter van Winden