EVERY OBJECT THAT BIOLOGY STUDIES IS A SYSTEM OF SYSTEMS' Francois Jacob 1974'

1
EVERY OBJECT THAT BIOLOGY STUDIES IS A SYSTEM OF
SYSTEMS. Francois Jacob 1974. Systems Biology.
The study of biological systems requires an
understanding of control and design structures,
elements of structural stability, resilience and
robustness which are not easily constructed from
mechanistic or reductionist information. All
these design elements represent emergent
properties that result from interactions between
mutually dependent variables that compose the
system.
2
SIGNAL PROCESSING AND INFORMATION TRANSFER. In
unicellular organisms, protein-based circuits act
in place of a nervous system to control
behaviour in the larger and more complicated
cells of plants and animals many thousands of
proteins, functionally connected to each other,
carry information from the plasma membrane to the
genome. The imprint of the environment on the
concentration and activity of many thousands of
proteins in a cell is, in effect, a memory trace
like a random access memory containing
ever-changing information about the cells
surroundings. Because of their high degree of
interconnection, systems of interacting proteins
act as neural networks trained by evolution to
respond appropriately to patterns of extra
cellular stimuli. The wiring of these networks
depends on diffusion-limited encounters between
molecules and for this and for other reasons they
have unique features not found in conventional
computer-based neural networks. (20-30 of
cytoplasmic volume is protein and oddly enough
protein interaction behaviour is nearer modelled
neural networks, than real neural
networks). Dennis Bray. Nature 1995.
3

• BRIEF OUTLINE OF LECTURE.
• Discuss several of these issues using data from
  Rainey lab on Pseudomonas fluorescens.
• Raineys interest is in pleiotropy how one gene
  affects many traits. Describe biology first.
• Describe system and examine briefly robustness,
  metabolic flux change and protein co-expression
  network. Broaden to include briefly ve and ve
  feedback and computation.
• Conformational Spread as another design feature.

4

• A SIMPLE MODEL
• Bacterial populations offer opportunities for
  simplification of investigation of evolution.
• Adaptive evolution of Pseudomonas fluorescens in
  experimental microcosms.
• Still (i.e. unshaken) media provide a number of
  new ecological niches one of these is the medium
  interface with air.
• Standard shape in shake culture is smooth (SM).
  At surface layer wrinkly morphs appear (wrinkly
  spreader, WS fuzzy spreader FS)

Rainey Travisano Nature (1998) Buckling et al
Nature (2000) Kassen et al Nature (2000) Buckling
Rainey Nature (2002) Buckling et al Science
(2003) Rainey Rainey Nature (2003) MacLean et
al PNAS (2004) Kassen et al Nature, (2004)
5
THE WS AND FS MORPHS APPEAR QUICKLY IN STILL
MEDIA AND START TO DISAPPEAR QUICKLY WHEN THE
MEDIUM IS SHAKEN.
Still
Shaken after being still for 7 days
WS
SM
Spatially heterogeneous
Spatially homogeneous
FS
days
6
COMPETITION FOR OXYGEN (and lack of shaking ) MAY
DRIVE DIVERSIFICATION.
AGITATED STILL
0
36hr
24hr
0hr
6
Oxygen saturation
The generation of the WS phenotype is the result
of information processing and signal
transduction. One of the signals may be a
depletion of oxygen but the morphs only appear
after mutation.
The WS and FS morphs are the result of simple
mutations.
7

• THE WS PHENOTYPE IS THE RESULT OF SIMPLE
  MUTATIONS.
• Cooperation is the key to WS success.
• Cooperation caused by simple mutations in
  negative regulators of cyclic dimeric GMP
  (c-diGMP) synthesis. Over-production of
  c-diGMP-dependent adhesive polymers, mainly
  acetylated cellulose, is induced, forming a mat
  or biofilm.
• Polymers are costly to the individual.
• But costs to individuals are traded against
  enhanced fitness of the group (the polymer aligns
  the interests of individuals with those of the
  group). Group selection.
• The group-level trait is an emergent property.
  (Gould 2002)

Rainey Rainey Nature (2003)
8
THE WS SIGNAL TRANSDUCTION SYSTEM

• The WS operon consists of 7 genes in tandem
  wspA-F, R. The mutations that occur naturally are
  usually in Wsp-F. Different mutations in Wsp-F
  can cause different phenotypes.
• WSP is a typical chemosensory system like
  chemotaxis. A number of these Wsp gene products
  share sequence similarity with chemotactic
  proteins. Chemotaxis is the most well understood
  signal transduction system in bacteria.

CANONICAL SIGNAL TRANSDUCTION IN BACTERIA.
These are known as two component systems. At
least 50 different inputs.
H and D phosphates unstable
PROCESSING MECHANISMS IN BACTERIA OFTEN DESCRIBED
AS A NEURAL NETWORK
9
SIGNAL TRANSDUCTION OF SM
Periplasmic space
Periplasmic space
Ligand receptor
Ligand receptor
A
A
b
a
activates
D
D
B
B
P
F
E
E
-ve feedback
P
ATP
R
Feedforward
R
10
SIGNAL TRANSDUCTION OF WS MUTATION
Periplasmic space
Periplasmic space
Ligand receptor
Ligand receptor
A
A
b
a
D
D
B
B
F
E
E
ATP
Very large gain of function
R
c-diGMP
cellulose
GTP
11
DIFFERENT MUTATIONS IN WspR CHANGE PHENOTYPE. A
SM B over-expressed R Elimination of
phosphorylation site in R (D67N) and
over-expression of R caused WS phenotype.
MIMICS CONFORMATIONAL CHANGE INDUCED BY
PHOSPHORYLATION!! C WspR R129C (receiver
domain) D WspR D159G (receiver domain) E
WspR F252S (c-diGMP domain) G-J repeats in WS
background.
12
SIMILARITY OF DESIGN OF WSP TO CHEMOTAXIS.
Periplasmic space
Periplasmic space
Ligands receptors
Ligand receptor
A
A
CHEMOTAXIS
WSP
D
B
W
F
B
E
P
ATP
ATP
P
Methylation, demethylation and information flow
determine level of CheYP.
R
Y
tumbling
cellulose
13
CHEMOTAXIS SIMPLIFIED
Input ligand
Chemotactic movement consists of smooth swimming
and tumbling. When attractant binds to MCP, Che
A activity (CheAP) is reduced decreasing CheYP
and decreasing tumbling. But this is modified by
methylation and demethylation of E by Che R and
CheBP. Attractants diminish information flow to
tumbling methylation increases it.
MCP Che W Che A
Che R
Em
E
CheBP
MCP Specialised Receptors that overlap in
function.
OUTPUT Tumbling (CheYP)
14
SPECIFIC ASPECTS OF THE CHEMOTACTIC SYSTEM ENSURE
IT IS ROBUST AGAINST CHANGES IN
CONSTITUENTS.(Expect WSP to be the same)
Barkall and Leibler (Nature 1997) modelled the
chemotactic system using a set of differential
equations.They predicted that adaptation
precision to concentrations of different ligands
would be robust to cell constituent changes.
Their model concerns changes in attractant levels
and the necessary adaptation as gradient of
attractant increases
Adaptation precision is basically chemotactic
velocity before and after stimulation with
increased levels of attractants.
15
EXPERIMENTAL SUPPORT FOR MODEL
At the heart of the two state model is a
functional receptor Che A. The active receptor
phosphorylates output protein (CheY) with
probabilities dependent on methylation and ligand
occupancy.
(M)
Output
CheA
The model deals with changes in signalling in an
ongoing system.
CheB
MeOH
Barkall/Leibler model confirmed by Alon et al
(Nature 1999) who genetically varied protein
constituents up to 12 fold. Measured adaptation
precision which varied only from 0.9 to 1.1. In a
sense this is a form of homeostasis or
homeokinesis. But system is ROBUST.
16
FURTHER EVIDENCE FOR CHEMOTACTIC ROBUSTNESS
(KOLLMANN ET AL. NATURE 2005)
Individuality of expression CheY fused to YFP
Che Z fused to CFP.
YFP
Adaptation precision
However bacteria seem to lack ve feedback
17
VARIATION IN NORMAL HEALTHY REPRODUCING HUMAN
BEINGS.
The classic text is ROGER WILLIAMS, Biochemical
Individuality, (1956). Variation/kilo body
weight. Stomach 6 fold. Thymus 13 fold. Testicles
10 fold. Thyroid 25. Cells in bone marrow.
Proerythroblasts 20 fold,. myelocytes 23 fold,
lymphocytes 9 fold. Inorganic composition 4 fold.
Endocrines. Thyroid 10 fold, oestrogen 12
fold, androgens 4-5 fold. Feedback accounts and
overlapping controls accounts for the apparent
stability of overall function. and overlapping
controls.
18
A COMMON TOPOLOGY FOR CONTROL Use repellent as
model for chemotaxis
(M)
Output
CheA
MeOH
19
CHEMICAL NEURONS FROM CHE-LIKE MODEL.
Simple design elements as in Che are common in
transduction. Okamoto, Arkin, Ross and others
have modelled their theoretical behaviour. They
show that behaviour is diode-like, similar to
McCulloch-Pitts neuron. They have been called
chemical neurons.
Okamoto Ross cyclic system
Cellular Ca2 oscillations
Ca2i
Diode-like output of protein-P in face of
variable input
20
CHEMICAL NEURON BEHAVIOUR DEVELOPED.
Ross has taken this much further using coupled
chemical neurons. He has indicated how logic
gates AND, NOR, OR etc.) can be constructed. In
addition he has constructed a universal Turing
machine, a binary decoder/adder and stack memory.
Relevance to reality? The interactions require
ve and ve feedback loops (inhibitory neurons)
Build up a cascade
Memory stack
The reality MAPK-P in vivo
Behaviour of neurons in a decoder/adder
Behaviour of single neuron
21
POTENTIAL IN ROSS ANALYSIS FOR PATTERN
RECOGNITION.
Using a number of coupled bistable systems
(Hjelmfelt et al., Science 1993) demonstrated the
potential for programming patterns (memory).
Provided the presented pattern differed by no
more than 10 errors, complete replication of
programmed pattern was observed after a number of
trials (Fig 1). The further away presented
pattern was from programmed pattern average
number of final errors increased.
Time series obtained from a chemical network.
22
LONG TERM MEMORY CREATED BY BISTABLE SYSTEM
CONSTRUCTED FROM VE AND VE FEEDBACK IN YEAST
GALACTOSE SYSTEM. Acar et al. Nature 2005.
Yeast galactose metabolism contains multiple
nested feedback loops. One of these ve feedback
loops (Gal3p) generates two stable expression
states with a persistent memory of previous
galactose consumption. Memory persists for
several hundred generations. Blunt and pointed
arrows indicate inhbition and activation
respectively
Many papers and reviews now indicate the
importance of nested ve and ve feedback in
conferring robustness, bistability, memory and
precision in development.
23
ANOTHER ELDERLY EXAMPLE OF POSSIBLE CELL
COMPUTATION.
Chemotaxis circuitry
Parallel distributed processing
Parallel distributed networks learn by adjusting
the strengths of connections through trial and
error. Could cells learn the same way?
Experimentql deletion of bits leads only to
overall degradation. Hologram? Barabasi-Error
tolerance of complex networks?
24

• BUT BACK TO WSP.
• Seems to be generally agreed that stable
  eucaryote development requires nested and ve
  feedback to operate. -ve feedback limits the
  length and strength of the signal ve feedback
  enables long term memory systems to develop.
  Many examples exist. Do bactera have ve
  feedback?Freeman. Nature 2000.
• In Wsp however neither ve or ve feedback
  apparently. The signal transduction system works
  apparently without regulation. But the mutation
  generates a stable phenotype thus must be
  robust. Stability can be expected to result from
  other constraints within the system e.g.
  hydrolysis of c-diGMP, feedback from cellulose
  wall etc.
• Systems have a kind of informational entropy
  that carries them to a more probable state of
  optimal stability. Stafford Beer. Nature 1965.

25
BACK TO WS AGAIN THE CHANGE IN PHENOTYPE IS LOSS
OF A NEGATIVE FEEDBACK. What is interesting is
that reliance of switching is placed on a
mutation, normally in WspF, to generate the new
phenotype. Yet this niche must be common for
Pseudomonas. But the penalty for specialisation
is a reduction in metabolic flexibility a
fitness trade off.
Different WspF mutants
Ability to metabolise different substrates
compared to wild type. Black indicates failure
grey weak growth.
Ability to metabolise amino acids reduced.
26
EFFECTS OF WSPF-R301S REVEALED AT THE PROTEOME
LEVEL.

• Comparison of protein expression
• Ancestral (SM) vs. LSWS (wspF R301S)
• 1985 protein spots identified (from five 2D gels)
• 81 differences (Plt0.005, paired t-test)
• 10 false positives expected (0.005 x 2000)
• 30 further 2D gels
• Heteroscedasticity accounts for 30 spots
• 52 spots verified as real changes.
• Note that many modifications are usually
• required for orderly change in any system.

27
CHANGES IN PROTEIN SPOTS AS A RESULT OF A SINGLE
BASE CHANGE IN SM (WSP-F) TO PRODUCE WS

• No spot corresponds to a known WS gene
• 49 (of 52) spots identified by mass spectrometry
• 42 spots show greater expression in LSWS
• Most associated with transport and degradation of
  amino acids

28
CORRELATION ANALYSIS OF CHANGES IN 52 KNOWN
SPOTS. One way of expressing the data.
A number of replicate experiments (15) are
performed. Because conditions are always slightly
different, the spot size of the 52 identified,
changes. Correlations between the 52 are
calculated and correlations above threshold
indicated as links.
29
CORRELATION TREE BETWEEN 52 PROTEINS THAT CHANGE
Instead of linking all spots together, only the
highest correlation is indicated to construct
this correlation tree. Trees of spot
correlations among replicates. (a) SM ancestral
tree (b) WS strain tree. Symbols refer to
proteins that catabolize amino acids. They
correlate more closely in Wsp.
30

• IS CONFORMATION SPREADING RESPONSIBLE FOR THE
  CORRELATION TREE?
• There is a problem. The Wsp-F mutation causes a
  reduction in the ability to utilise (catabolise)
  amino acids but it is these proteins that are
  increased in amount in Wsp.
• An additional mechanism, conformational
  spreading, may be responsible. Conformational
  spreading results from a ligand-bound
  conformational change in a nucleating allosteric
  protein that spreads this new conformation in
  adjacent unbound proteins. Well known examples
  are prions, amyloid plaques?This is a mechanism
  that is neither ve or ve feedback but can
  amplify very small signals.
• In E.coli it was observed that 2nM aspartate was
  sufficient to reduce tumbling bias. That is about
  1-2 molecules bound/cell out of 1000 receptors.
  The gain (calculated as the change in tumbling
  bias divided by the change in occupancy) was
  about 60.

31
EXTREME AMPLIFICATION OF ATTRACTANT SIGNAL
E.coli was variably transformed with CheYp
(fluorescently labelled) and measurements made of
tumbling bias. Because of kinetics of Che
system, CheYp equates directly with receptor
occupation. Hill coefficient is at least 10.
Cluzel et al., Science 2000.
CheY-P
Sourjik and Berg PNAS 2001 used fluorescent
versions of CheY CheZ for FRET. Stage between
aspartate binding and CheYP has an amplification
30-40 times greater than simple receptor binding
models suggest. Total amplification at least 60
fold. 2 molecules bound is 120 receptors
activated.
32
CONFORMATIONAL SPREAD
X marks ligand bound white receptors activated
(or inactivated). Many examples of heterologous
spreading, e.g. transient transduction complexes,
integrins. PH, SH domains
Bacterial receptor cluster
SOLID STATE CIRCUITRY. Allosteric proteins are
found throughout cells of all kinds. Living
cells are to a first approximation made of a
gel-like matrix of protein molecules. Aggregates
of myriad catalytic and structural roles link
membranes to the cytoskeleton in a complete web
of protein-protein interactions permeated by an
aqueous solution of ions and small molecules. The
topology of this this matrix is not known in
detail and probably changes dynamically with the
changing physiology of the cell. But there is
good reason to think it is physically continuous
over long distances. It seems likely that any
two locations of the cytoplasm are connected by a
chain of protein molecules. Bray and Duke, 2004.
Ann Rev Biophys.
33
BACK TO WSP AGAIN SUMMARY
receptors
What I suggest is going on here is that
conformational spread from the activation and
high activity of the Wsp transduction pathway
affects the stability (against protease) of local
nearby proteins.
Che-YP
Whether the receptors are localised in a patch
(as in E.coli here or distributed throughout the
plasma membrane is not known. But variable
degrees of activation according to circumstance
could account for variations in apparent
correlative change in protein co-expression
network. Over-expression of Wsp-R is already
known to activate in the absence of
phosphorylation thus directly indicating
conformational change and spread.
SUMMARY. -ve and ve feedback are now joined by
conformational spread as design modules in
biological systems. Solid state circuitry joins
memory as important elements enabling computation
to occur.
34
SIGNIFICANT CHANGES IN CONNECTIVITY.

• The WspF R301S (LSWS) mutation results in
  significant changes in the correlation skew for
  four proteins

ABC transporter
PBP
Ancestral SM
Derived LSWS
Frequency distributions of correlations to all
other proteins
35
TRADITIONAL STRUCTURE FOR NEGATIVE FEEDBACK THAT
PROVIDES FOR CERTAIN ELEMENTS OF ROBUSTNESS.
Reference signal
36
(No Transcript)
37
The sine qua non of behavioural intelligence
systems is the capacity to predict the future-to
model likely behavioural outcomes in the service
of inclusive fitness. This logic is already
evident in a primitive sense in chemotaxis in
Escherichia coli information transduced by
environmental sensors directs behavioural
responses in a manner that increases the
probability of the attainment of bioenergetic
resources in the next moment. La Cerra and
Bingham, PNAS 1998. Some of the most fundamental
features of brains such as sensory integration,
memory, decision-making and the control of
behaviour can all be found in these simple
organisms (Allmann, 1999).
38
INACTIVE STATE
ACTIVE STATE
CHEMOTAXIS
Receptors
Ligands receptors
A
A
activates
inactivates
W
W
B
B
ATP
P
P
Che A phosphorylates output protein (CheY) with
probabilities dependent on methylation ()and
ligand occupancy (-).
Y
Y
Tumbling increased
39
BACTERIAL CHEMOTAXIS.

• Chemotaxis in Escherichia coli is a combination
  of smooth swimming and of random tumbling caused
  by counter rotation of the flagellar motor. When
  an attractant (amino acids, sugars) is sensed,
  active CheA is diminished, information flow to
  CheY to form CheYp reduced. Tumbling rates are
  reduced and enhanced smooth swimming up the
  gradient occurs. To enable adaptation to
  increasing concentrations of attractant, Che A
  methylation induces active state and CheBp
  demethylation activity is decreased. Consequently
  the rate of information flow is kept constant
  through the system enabling adaptation to large
  changes in attractant levels. The active receptor
  phosphorylates output protein (CheY) with
  probabilities dependent on methylation and ligand
  occupancy.
• Some of the most fundamental features of brains
  such as sensory integration, memory,
  decision-making and the control of behaviour can
  all be found in these simple organisms (Allmann,
  1999).