Title: Evolution of Complex Dynamics in Mathematical and Electronic Models of Genetic Networks
1Evolution of Complex Dynamics in Mathematical and
Electronic Models of Genetic Networks
- Leon Glass
- Isadore Rosenfeld Chair in Cardiology,
- McGill University
2A synthetic oscillatory network of
transcriptional regulators
TetR
LacI
l cI
gene B
gene C
gene A
PC
PA
PB
mRNA A
mRNA B
mRNA C
protein A
protein C
protein B
Elowitz and Leibler, 2000
3Plasmids
Repressilator
Reporter
PLlacO1
ampR
tetR-lite
PLtetO1
kanR
TetR
TetR
SC101 origin
gfp-aav
l PR
l cI
LacI
GFP
lacI-lite
ColE1
l cI-lite
Elowitz and Leibler, 2000
PLtetO1
4Observation in Individual Cells
60
140
250
300
390
450
550
600
GFP Fluorescence
Bright-Field
Fluorescence (a.u.)
time (min)
Elowitz and Leibler, 2000
5Problem How can we develop mathematical models
that represent the dynamics in real networks?
6A differential equation
Glass, Kauffman, Pasternack, 1970s
7Rationale for the equation
- A method was needed to relate the qualitative
properties of networks (connectivity,
interactions) to the qualitative properties of
the dynamics - The equations allow detailed mathematical
analysis. Discrete math problems
(classification), nonlinear dynamics (proof of
limit cycles and chaos in high dimensions)
8The Repressilator
9The Hypercube Representation
10The Hypercube Representation for Dynamics (N
genes)
- 2N vertices each vertex represents an orthant
of phase space - N x 2N-1 edges each edge represents a
transition between neighboring orthants - For networks with no self-input, there is a
corresponding directed N-cube in which each edge
is oriented in a unique orientation
11Cyclic Attractors
- Any attracting cycle on the hypercube corresponds
to either a stable limit cycle or a stable
focus in the differential equation (Glass and
Pasternack, 1978)
12Strategy of Proof of Limit Cycles in N dimensions
The Poincare map is a linear fractional map. a is
a (N-1)x(N-1) matrix, b and x(t) are (N-1)
vectors. The composition of 2 linear fractional
maps is a linear fractional map. From the
Frobenius-Perron theorem, it follows that if the
leading eigenvalue of a is gt1, then there is a
stable,unique limit cycle and if the leading
eigenvalue of a is less than 1, there is a
focus. Conjecture Cycles are robust when step
functions are changed to steep enough sigmoids.
13Chaotic dynamics in the model equations
A network that generates chaos
Mestl, Lemay, Glass, 1996
14A Chaotic System in 4D
15Projection of flows
16Evolution of dynamics
17The number of different networks in N dimensions
Glass, 1975 Edwards and Glass, 2000
18Evolving Rare Dynamics
- Long cycle
- Chaotic dynamics - increased complexity using
topological entropy as a measure of complexity
19An Evolvable Circuit
(J. Mason, J. Collins, P. Linsay, LG, Chaos, 2004)
20Why study electronic circuits?
- It is real
- It leads us to think about issues in real
circuits i.e. not all decay rates will be equal - Circuits could be useful
21The Hybrid Analog-Digital Circuit
22Circuit Elements
23Distribution of Cycle Lengths in Electronic
Circuit (300 random circuits with stable
oscillations)
Choose a target period of 80 ms
24Sample Evolutionary Run
25Optimal Mutation Rate - Data
- Each trial starts with oscillating network
- 25 Trials at each mutation rate for 250
generations - Mutation rates of 2.5, 5, 10, 20, 100
26The Hypercube Representation for Fitness
Landscapes (N genes)
- 2N x 2N-1 vertices
- Each vertex represents a different genome
- Two vertices can have the same fitness, i.e.
there can be neutral mutations
27Fitness Landscape
k is the Hamming distance of the truth table of a
network from a network with stable cycle
The mean improvement as a function of mutation
rate
where,
28Prediction of Optimal Mutation Rate
- Compares favorably with experimentally determined
value of 5-10
29Chaotic Electronic Network
Glass, Perkins, Mason, Siegelmann, Edwards, J.
Stat. Phys. 2005
30Symbolic sequences of cycles that start and end
on vertex 00110
- 6 141343 4 1414 6 143143 6
143413 14 14341523545245 14
14345123545245 18 143453431352353254 8
15421245 8 15421254 12 154351535343
14 15435153535354 16 1543515423424252
62 154512131512121212121212121214241241243242141
24214124214124252 60 154512131512121212121212
121214241241243242424214124254524252 56
15451213151212121212121212121424124124324242424242
545245 62 15451213151212121212121214241241243
242141242141242141242545245 60
15451213151212121212121214241241243242141242141242
1545124252 58 1545121315121212121212121424124
124324214124214124254524252 56
15451213151212121212121214241241243242141242141242
545245 52 15451213151212121212121214241241243
24242424242545245 54 154512131512121212121212
142432421412421412421545124252 46
1545121315121212121212121424324214124214124252
50 154512131512121212121212142432421412421412425
45245 48 154512131512121212121212412412412432
424242424252
Etc.- Can we use these to classify bifurcations
and to compare different types of chaos in
chaotic systems?
31Statistical Properties of Words in Return Map
Zipfs Law!
32Evolving Rare Dynamics
- Long cycle
- Chaotic dynamics - increased complexity using
topological entropy as a measure of complexity
(An upper limit on this can be determined from
the underlying truth table of a network)
33Evolving Complex Dynamics
Wilds, Kauffman, Glass (2007)
34No neutral mutations
35With neutral mutations
36The theoretical limit on entropy increases during
evolution
37Conclusions
- These networks can model some real biological
systems - Electronic networks displaying rare dynamics can
be evolved - Systems with neutral mutations evolve faster than
those that just do hill climbing - During evolution, the neutral subspace decreases
as evolution proceeds
38Properties of Networks Based on Logical Structure
- Extremal stable fixed points
- Limit cycles associated with cyclic attractors
(stability and uniqueness) - Necessary conditions for limit cycles and chaos
- Analysis of chaos in some networks
- Upper limit on topological entropy
39The inverse problem given the dynamics what is
the network?(Research with Ted Perkins, McGill)
40Fibrosis Pathway by STKE
41A simple feedback network
X1 X2(t) X1 X2(t1)
42Flows in phase space and the hypercube
representation for a simple feedback network
11
01
10
00
43Poincare map for the 2D cycle
, where a, b gt0
alt1
x(t1)
x(t)
x(t)
44(No Transcript)
45Main Problems - To relate the interactions in
genetic networks to the dynamics.
- To predict the dynamics given the genetic network
- To determine the genetic network given the
dynamics - To understand how genetic networks evolved
46A differential equation
Glass, Kauffman, Pasternack, 1970s
47(No Transcript)
48The Inverse Problem. Compute the number of
logical states needed to determine connectivity
diagram
Perkins, Hallett, Glass (2004)
49Compute the number of switches needed to
determine the entire network
50Chaotic Electronic Network
Glass, Perkins, Mason, Siegelmann, Edwards, J.
Stat. Phys. 2005
51(No Transcript)
52Inverse problem for the electronic circuit
(parameter values)
53Simulation of differential equation with
parameters just determined
54Sequential recruitment and combinatorial
assembling of multiprotein complexes in
transcriptional activation.
Lemaire, Lee, Lei, Metivier, Glass (PRL,2006)
55Modeling the kinetics
A long sequence of biochemical transformations of
any kind taking place at the promoter
- complex formation
- enzymatic modification
- conformational modification
- ATP-dependent reaction
- etc.
- m unidentified complexes xi
- xi are composed of an arbitrary number of
proteins
One constraint Period ? 40 min, the
experimental period
56Deterministic approach
In the limit of a large number of cells,
x1(t) full solution
x1(t) leading term
x1(t) sum of the 2 first terms
x1(t) sum of the 7 first terms
x1(t) sum of the 4 first terms
57Fitting results
Values of the ci,j (colored regions is 1)
Métivier et al 2003
58Study of gene expression in dogs who have rapid
pacing of the atria or ventricles.
- Experiment designed and carried out in the
laboratory of Dr. S. Nattel, Sophie Cardin and
Patricia Pelletier. Data analysis by Eric Libby - (Circulation Research 2007)
59Affymetrix Gene Chips
- Surveys thousands of genes simultaneously
- Has 11-20 probes for each gene
- Probes are 25 base pairs long either PM or MM
- Species specific
- Millions of sequences in each 500,000 feature
spots
60Chip Hybridization
61The Data
- CEL files
- Raw PM and MM intensities
- Few software packages can open files
62Ventricular Tachypacing
- Fibrotic remodeling
- Immune system response and apoptosis
- Pace right ventricle at 220-240 bpm
63Microarray Setup
- 5 dogs in each group
- 8 groups total
- Compare ATP and VTP in LA
- Compare 24 hr to 1-2 wk treatments
- Compare VTP LA to VTP LV
64B
A
ATP Expression Level (ln L)
ATP Expression Level (ln L)
ATP 24-h
ATP 1-w
xxxx
xxxx
xxxx
Sham Expression Level
Sham Expression Level
C
D
VTP Expression Level (ln L)
VTP Expression Level (ln L)
VTP 24-h
VTP 2-w
xxxx
Sham Expression Level (ln L)
Sham Expression Level (ln L)
Figure 1
65Upregulated
Downregulated
DNA/RNA syn/deg. Cell cycle Apoptosis Immunity ECM
Signal trans. Protein syn/deg. Ribosomes Hormones
Metabolism Cell structure Transport Mitochondria
14 12 10 8 6 4 2 0 2 4
Number of genes
Figure 2
66Hypothetical Network
67Conclusions
- For the model differential equations, and the
electronic models of these equations - The networks can evolve to find new very rare
dynamics - It is possible to determine the networks based on
the observed dynamics in model equations, in
electronic circuits, in drosophila - Problems for complex physiological data
- Noise, biological variability, few time points
(expense!), time delays
68The Inverse Problem of Genetic Networks
- Leon Glass
- Isadore Rosenfeld Chair in Cardiology, McGill
University, Montreal, Quebec
69Plane of the Poincare map
70First analysis
- All reaction rates are assumed irreversible.
- All reaction rates are assumed identical ai a.
Choosing m 200, then a m/T0 5 min-1
Due to the stochastic nature of the system, the
chemical scheme can be interpreted as a
sequential Poisson process in which the duration
t before the next reaction takes place follows
the distribution
p(t) a e-at
After synchronization of the cells,
- the k-th cycle has mean starting time of km/a
- and variance kT02/m
71Fitting the model
Each protein is a component of different
complexes xi(t), but we do not know a priori
which ones.
xi(t)
xi1 (t)
xi2(t)
xm(t)
x1(t)
xi-2 (t)
xi-1 (t)
Po(t)
For the orange protein, co 0, ,0, 1, 1, 1, 0,
, 0
t
Pb(t)
For the blue protein, cb 0, ,0, 0, 0, 1, 0,
, 1
t
Pp(t)
For the purple protein, cp 0, ,0, 0, 0, 0, 1,
, 0
t
72Robustness of the results
We performed different numerical studies to test
the robustness of the fitted sequence
- Different values of m 100, 200, 300 and 400.
- Addition of 2 of noise to the data points.
- Changes in the vertical scaling of the data (up
to 1.4). - Selections of random reaction rates ai
(provided that the period, for the selection of
ai, is close to T0, and the solutions are not too
damped).
Results Although there can be slight changes in
the sequence ci , the main pattern of the ci,j
remain unchanged.
73Biological meaning of the results
- Correlation between the sequences.
- Reconciling ChIP assays and Fluorescence
microscopy-based methods. - It is believed that 5 or 6 functional complexes
are successively formed on the promoter of pS2.
74(No Transcript)
75(No Transcript)
76A differential equation
i(t)
i
t
Xi1
Xi0
Xi1
77(Side issue classification of complex dynamics)
78Mechanism Discovery
- Look up pathways of hormones and most
significantly expressed genes
Fibrosis Pathway by STKE
79(No Transcript)
80Global Results
81Decipher Mechanisms
9
36
533
503
24
4
4
357
14
12
172
1
5
47
29
0
0
0
18 genes
1641 genes
91 genes
82Significant Genes
83(No Transcript)
84(No Transcript)
85(No Transcript)
86(No Transcript)
87Cumulative fraction of ranks
24-h VTP
2-w VTP
B
A
Fraction of ranks
Extracellular matrix
Cell structure/mobility
Immunity
Extracellular matrix
Cell structure/mobility
Metabolism
Ribosome
Metabolism
Others
Mitochondria
DNA/RNA syn/deg
Ribosome
Others
Gene groups deviating from overall VTP expression
pattern
Figure 4
88Gene expression in Drosophila Perkins, Jaeger,
Reinitz, GlassPLOS Computational Biology 2006
89(No Transcript)
90(No Transcript)
91(Perkins, Jaeger, Reinitz, Glass, PLOS
Computational Biology 2006)
92(Perkins, Jaeger, Reinitz, Glass, PLOS
Computational Biology 2006)
93Proposed network for gene control
94Comparison with different models