introduction to modelling signalling cascades in yeastemilie/icysb/wp-content/uploads/2011... ·...

38
ICYSB, 2011 Schaber 1/38 Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology Laboratory Institute of Experimental Internal Medicine Otto-von-Guericke-University Magdeburg

Upload: others

Post on 26-Jun-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 1/38

Introduction to Modelling Signalling Cascades in Yeast

Jörg Schaber

Systems Biology LaboratoryInstitute of Experimental Internal Medicine

Otto-von-Guericke-UniversityMagdeburg

Page 2: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 2/38

• The modelling process• A simple step-by-step example

– The Sho1 branch of the HOG pathway• Basic concepts

– Signalling motifs– Model selection

Outline

Page 3: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 3/38

• Modelling requires verbal hypothesis be made specific and conceptually rigorous.

Why Models?

• Modelling highlights gaps in our knowledge.

• Modelling provides quantifiable as well as qualitative predictions.

• Modelling is ideal for analysing complex interactions before experimental tests.

• Modelling is a low-cost, rapid test bed for candidate interventions.

• Well designed models are readily portable and adaptable for many purposes.

Page 4: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 4/38

ObservationPrior knowledge • The observation of a

natural phenomena drives a scientific question, which is represented by a testable hypothesis.• This is the most important step as it sets the stage for the rest.• Requires intuition and talent.

• The word model is the abstract representation of the processes that might explain data based on the hypothesis.• Often represented by a diagram.• Defines the components and their interactions.• Defines systems boundaries.

• The math model is the formalised word model.• The formalism depends on the processes and data, e.g.

• dynamic systems: ODEs• gene networks: boolean models

• The verification is a first qualitative evaluation of the model.• Checks whether the model is in principle able to explain the data, e.g.

• if the data shows oscillations, the model should be able to oscillate as well.

• The validation is a quantitative evaluation of the model.• Checks whether the model is able to explain the data quantitatively.• Fitting model to the data.• Parameter optimization

• A successful validation consolidates our trust that we have captured the most important processes to explain the data.• This justifies an analysis of model properties, e.g.

• sensitivities• robustness

• The analysis may give indications for useful predictions.• Sensitivity analysis may suggest a parameter with high impact on certain behaviour.• We can define the ‘most informative’ experiment.

ObservationPrior knowledge

Math ModelMath Model

Analysis

Validation

Verification

Word Model

Hypothesis

Prediction

Analysis

Validation

Verification

Word Model

Hypothesis

The Modelling Process in 8 Steps

• A successful prediction consolidates our trust in the model and the hypothesis and that we have identified the most important processes to explain the data.

• Usually a prediction is not successful.• Often we have to change our model/hypothesis in the course of the modelling process.• Each modelling round deepens our understanding.

Page 5: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 5/38

A Step-by-Step ExampleThe Sho1-branch of the HOG pathway

Page 6: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 6/38

The Data• It was shown that

a) Sho1 de-oligomerizes upon osmotic shockb) Hog1 phosphorylates de-oligomerized Sho1c) Phosphorylated Sho1 is less able to transmit the

signal

Oligomer-deficient

Oligomer-deficientPhospho-mimic

SEM: standard errorof the mean, i.e.SE/n

Hao et al. (2007) Curr. Biol. 17

Page 7: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 7/38

The Hypothesis

– Phosphorylation of Sho1 by Hog1 constitutes a negative feedback loop.

– This negative feedback leads to the rapid attenuation of Hog1 signalling.

– Might be important to dampen crosstalk to pheromone signalling pathway.

Page 8: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 8/38

The Word Model(s)

• Signal: high osmolarity• Hog1 de-sensitizes Sho1

Hao et al. (2007) Curr. Biol. 17

• No glycerol accumulation

Page 9: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 9/38

Formalising Word Models

Pbs2P

Hog1

‘Biologist’ notation

Not very useful (for modelling), becauseinteractions not clear.

Pbs2P

Hog1P

‘Systems Biologist’ notation

Hog1v1

v2

More useful, because each interaction is made specific.This facilitates mathematical formulation.

Page 10: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 10/38

Formalising Word Models

Pbs2P

Hog1PHog1v1

v2

• arrows between components indicate transformations, i.e. biochemical reactions, mass flows. They determine changes in concentrations, numbers, etc.

21

21

1

1

vvdt

dHog

vvdt

PdHog

Note that Pbs2P does not change.

Page 11: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 11/38

Formalising Word Models

Pbs2P

Hog1PHog1v1

v2

• arrows on arrows indicate modifying interactions (enzymatic reactions), i.e. no (net) mass flows or concentrations changes involved from emanating components.

Biochemical notation

Hog1 + Pbs2P -> Hog1P + Pbs2P

Pbs2P neither consumed nor produced (netto-wise)

Page 12: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 12/38

Formalising Word Models

Pbs2P

Hog1PHog1v1

v2

Biochemical notation

v1 : Hog1 + Pbs2P -> Hog1P + Pbs2Pv2 : Hog1P -> Hog1

Most simple mathematical formulation: mass action kinetics, i.e. linear multiplication of substrates.

v1 : k1 ·Hog1·Pbs2Pv2 : k2 ·Hog1P

21

21

1

1

vvdt

dHog

vvdt

PdHog

Page 13: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 13/38

Formalising Word Models

Fus3

Fus3- Ste5

Ste5

v1

v2

Biochemical notation

v1 : Fus3 + Ste5 -> Fus3-Ste5v2 : Fus3-Ste5 -> Fus3 + Ste5

Most simple mathematical formulation: mass action kinetics, i.e. linear multiplication of substrates.

v1 : k1 ·Fus3·Ste5v2 : k2 ·Fus3-Ste5

21

21

21

53

5

3

vvdt

StedFus

vvdt

dSte

vvdt

dFus

Page 14: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 14/38

Kinetic rate laws and signalling motifs

Pv1

Constant fluxand simple mass action degradation

v2

PkkdtdP

21

Pv1

Modified constant fluxand simple mass action degradation

v2 PkSkdtdP

SkdtdS

2

3

1

S v3

2

2

1

10

kk

P

Pkk

2

1

kk

0)0( tP

Page 15: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 15/38

Pv1

Modified conversion with signal degradationMass action kinetics

v2

1

)1( 2

3

1

PQ

PkPSkdtdP

SkdtdS

S v3

Q

Pv1

Modified conversion with signal degradationMichaelis-Menten kinetics

v2

S v3

Q

PKmPk

PKmPSk

dtdP

SKmSk

dtdS

2

2

1

3

3

1)1(1

k3

k3 /2

Km3

Kinetic rate laws and signalling motifs

Page 16: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 16/38

Pv1

Modified conversion with signal degradationHill kinetics

v2

S v3

Q

22

2

2

11

1

11

33

3

3

2

3

)1()1(

hh

h

hh

h

hh

h

PKmPk

PKmPSk

dtdP

SKmSk

dtdS

k3 /2

Km3

k3

Kinetic rate laws and signalling motifs

Page 17: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 17/38

Pv1

Inhibited conversion with signal degradation

v2

S v3

QPk

KS

PkdtdP

SkdtdS

h

i

2

3

1

)1(1

1/2

Ki

1

Kinetic rate laws and signalling motifs

Page 18: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 18/38

Pv1

Modified conversion with negative feedback

v2

S

Q

Pk

KP

PSkdtdP

h

i

2

1

)1(1

Observation:• System reaches new steady state.• No ‘overshoot’

Possible explanation:• Feedback comes too fast.

Kinetic rate laws and signalling motifs

Page 19: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 19/38

Pv1

Modified conversion with delayed negative feedback

v2

S

Q

Pk

KtP

PSkdtdP

h

i

2)(1

)1(1

Observation:• System ‘overshoots’, but still reaches new steady state.• damped oscillations• The feedback to P depends directly from P itself -> ‘transient feedback’

Kinetic rate laws and signalling motifs

Page 20: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 20/38

Modified conversion with negative feedback

Let’s do the math: Calculate steady-states of P(S) of a simplified system(all constants = 1)

PPPS

dtdP

1

)1(

Ki

261S-1-21

1)1(0

SSP

PPPS

In the steady-state:

Our analysis suggests that in a system with transient negative feedback, the steady- state depends on the input signal.

Page 21: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 21/38

The Mathematical Model

• Signal: high osmolarity• Hog1 de-sensitizes Sho1• No glycerol accumulation• Transient delayed feedback• Michaelis-Menten kinetics (20

parameters)• Model was fitted to Hog1

activation data

Hao et al. (2007) Curr. Biol. 17

Page 22: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 22/38

Validation

Orig. 1 M KClOrig. 0.5 M KClOrig. 0.25 M KCl

Single shock Double shock

- Model fits data well. - Simulations show damped oscillations and increasing steady states- possible spurious effects due to over-parameterization

Page 23: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 23/38

Predictions

Orig. P-Hog1

Orig. Sho1aOrig. Sho1i

0.4 M KCl tripple shock (0, 30, 60 min)

Page 24: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 24/38

Conclusion after first modelling round

• Transient feedback model fits single and double shock data well, but• shows increasing steady states with increasing

external osmolarity (not supported by data),• shows damped oscillations (not supported by data, but

might be due to over-parameterization)• Transient feedback model is not able to predict triple

shock experiment, because of desensitization.

Possible solution: a) the signal has to be removed by adaptationb) no desensitization

Page 25: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 25/38

Sh1Sho1

Pbs1Hog1

P

P

Fps1

Glycerol

Hog1 P

Gpd

Sh1Sho1

Hog1 P

• Increasing the ambientosmotic pressure leads to a rapid passive loss of water and cell skrinkage

• This leads to closure of the glycerol channel and activation of two parallel signaling branches that both activate Hog1.

• Activated Hog1 translocates to the nucleus and triggers production of enzyme that enhance glycerol production.

• Increased glycerol equilibrates water potential differences and forces water back into the cell leading to volume adaptation.

Recalling the biology

• After adaptation water potential differences return to original levels.

Page 26: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 26/38

New Hypothesis

– The signal is the water potential difference (differences in osmolarity) rather than merely external osmolarity.

– The main feedback is via glycerol accumulation.

Page 27: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 27/38

The new word model

Signal

OuterOsmolarity

Hog1 P-Hog1

Glycerol

v3v1

v2

– Signal = OuterOmolarity- Glycerol

– 3 reactions with mass actions, i.e. 3 parameters

Page 28: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 28/38

The mathematical model

Signal

OuterOsmolarity

Hog1 P-Hog1

Glycerol

v3v1

v2

PHogkdtdG

PHogkPHogGOkdt

PdHog

1

1)11)((1

3

21

– For simplicity we assume Hog1+Hog1P=1

– OuterOsmolarity (O) fixed input function.

– The feedback depends on increasing Glycerol, which stays even if Hog1P=0

=> integral feedback

Page 29: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 29/38

Verification of the model

PHogkPHogkPHogGOk

101)11)((0

3

21

Let’s do the math: Calculate steady-states

01

PHogGO

In the steady-state:

1. In the steady state Hog1P=0 independent from the outer osmolarity.

2. Only the internal glycerol concentration is dependent from the outer osmolarity.

Page 30: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 30/38

Validation of the model

Single shock

- Simulations show reasonable fits and perfect adaptation

Double shock1 M KCl0.5 M KCl0.25 M KCl

Page 31: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 31/38

Prediction

- Simple model reacts to triple shock and shows perfect adaptation.

P-Hog1lInnerOsmolarityOuterOsmolarity

0.4 M KCl tripple shock (0, 30, 60 min)

Page 32: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 32/38

RemarksIntegral Feedback (3 parameters) Transient feedback (20 parameters)

• From the quality of the fit, the transient feedback model is much better than the integral feedback model.• The quality of the fit is usually measured by the sum of squared residuals.

1 M KCl0.5 M KCl0.25 M KCl

• If it weren’t for the prediction, which model is better?

n

ii

n

iii rptfyypSSR

1

2

1

2),((),(

r1

r2

r3

Page 33: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 33/38

RemarksIntegral Feedback (3 parameters) Transient feedback (20 parameters)

• Intuitively: both model capture the basic features of the data.• In other words: The main information in the data is reproduced by the models.• But: the information/parameter is much higher in the simpler model.=> The parameters of the simpler model are more informative.

1 M KCl0.5 M KCl0.25 M KCl

Page 34: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 34/38

The principle of parsimony• If we take the number of parameters as a measure for structural properties of the data, we want to have few parameters, i.e. the most important structural features, and a good data representation.• We do not want to have additional parameter to ‘fit the errors’ or spurious effects.• The simpler the model, the easier to analyse.• Therefore, it is advisable to have a model that is as simple as possible and as complex as necessary: this is the principle of parsimony.

“Everything should be madeas simple as possible, but not simpler.”

• Parsimony can be measured by the Akaike Information Criterion (AIC).

nSSRnkAIC log2

• the lower AIC, the better the model approximates the data in terms of parsimony (k number of parameters, n number of data).

Page 35: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 35/38

Model Selection• The AIC can be used as a model selection criterion

# parameters SSR AICTransient FB 20 0.049 164.842Integral FB 3 0.251 -38.045

• Only from the fits, the simper model would have been selected as the best approximating model.

Page 36: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 36/38

Conclusions concerning feedback in Sho1-branch

• According to the AIC, the data does not support a model with transient feedback, but rather a model with integral feedback.• Therefore, for the adaptation and attenuation process, the proposed feedback of Hog1 on Sho1 is not necessary.• The proposed feedback of Hog1 to Sho1 may modulate the signal and serve other purposes (crosstalk, stabilisation), but cannot explain adaptation to single and multiple shock.

Page 37: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 37/38

Final Remarks on Modelling• The integral feedback model has several shortcoming:

• glycerol is only accumulated, never lost. Adaptation to a lowering in external osmolarity cannot be modelled• initial Hog1P is zero

All models are wrong, but some are useful.

• Our model was developed to address the question, whether or not the (multiple) osmotic shock data can be explained by a transient feedback, nothing else.

• All models are tailored to address specific questions. No model can explain everything.• The clearer our hypotheses are formulated, the better models can be developed to address these.

Page 38: Introduction to Modelling Signalling Cascades in Yeastemilie/icysb/wp-content/uploads/2011... · Introduction to Modelling Signalling Cascades in Yeast Jörg Schaber Systems Biology

ICYSB, 2011 Schaber 38/38

Final Remarks on Modelling

• The ‘truth’ (full reality) in biological sciences has infinite complexity and, hence, can never be revealed with only finite samples and a ‘model’ of those data.

• It is a mistake to believe that there is a simple “true model” and that during data analysis this model can be uncovered and its parameters estimated.• We can only hope to identify a model that provides a good approximation to the data available.

• Uncertainty about the biology leads to multiple hypotheses about the underlying processes explaining a set of data.• I recommend the formulation of multiple working hypotheses, and the building of a small set of models to clearly and uniquely represent these hypotheses.• The best approximating model can then be identified by model selection criteria.