predicting erbb network signaling dynamics marc r birtwistle 16 sept 2011 georgia health sciences...

63
Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Upload: alaina-parrish

Post on 02-Jan-2016

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Predicting ErbB Network Signaling Dynamics

Marc R Birtwistle16 Sept 2011

Georgia Health Sciences University

Dept. of BiostatisticsResearch Forum

Page 2: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

• Background– ErbB signaling network– Why we build mathematical models to help us

understand signal transduction

• How Cells Generate and Interpret Ligand-Specific Spatiotemporal Signaling in the ErbB Network

• Cell-to-Cell Variability in Protein Expression and ErbB Signaling

Outline

Page 3: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

ErbB Signaling Network

Yarden and Sliwkowski, Nat. Rev. Mol. Cell Biol., 2; 127-137 (2001)

Page 4: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

ErbB Receptors and Targeted Pharmaceuticals for Cancer Treatment

4

Citri and Yarden, Nat. Rev. Mol. Cell Biol., 7; 505-516 (2006)

• Sometimes they’re successful, but many times they’re not.

• Why? How can we do better?

Page 5: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

5

Our Hypothesis: The Collective Network Behavior is Most Important

Oda et al., Mol. Sys. Biol., 1; (2005)

Page 6: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Example: A Negative Feedback Amplifier Inherently Resists Perturbations Within the Feedback Loop

MEK

ERK

Proliferation, Migration, etc

BXB-ER

4-OHT

Without Feedback

U0126

With Feedback

Sturm, O, Orton, R, Grindlay, J, Birtwistle, MR, Vyshemirsky, V, Gilbert, D, Calder, M, Pitt, A, Kholodenko, B and Kolch, W. The mammalian MAPK/ERK pathway exhibits properties of a negative feedback amplifier. Sci Signal 3(153), (2010)

U0126

EGF

Ras

Raf

MEK

ERK

Proliferation, Migration, etc

ErbB1

Page 7: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

• Need to know:– Magnitude of Effects on D

A strong; C weak D upA weak; C strong D down

– Dynamics of Interactions with DA slow; C fast D down then upA fast; C slow D up then down

– Localization with DA local; C distant D upA distant; C local D down

Qualitative Knowledge is Not Enough to Predict Outcomes

7

A Mathematical Model Helps Us Keep Track of Quantitative, Spatiotemporal Aspects of ErbB Signaling

Page 8: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Modeling the ErbB Signaling Network in the MCF-7 Breast Cancer Cell Line

Where does the specificity come from?

MCF-7 breast adenocarcinoma cells; HRG induces differentiation (lipid droplet accumulation); EGF induces only proliferation

Adapted from Yarden and Sliwkowski, Nat Rev Mol Cell Biol 2001

Page 9: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Specificity Can Emerge from the Spatiotemporal Dynamics of Signaling

Adapted from Marshall, Cell, 1995

PC-12 Cells

Transient

Sustained

Page 10: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

How does the network generate distinct signaling dynamics?

How does the network interpret different dynamics?

Page 11: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

MCF-7 cells; HRG induces differentiation (lipid droplet accumulation); EGF induces only proliferation

Birtwistle et al., Ligand-depednent responses of the ErbB signaling network: experimental and modeling analyses. Mol Syst Biol, 2007

What is Controlling ErbB Signaling Dynamics?

Transient

Sustained

Page 12: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Constructing a Model of Short-Term ErbB Signaling

Birtwistle et al., Mol Syst Biol, 2007

Biological Knowledge

Wiring Diagram

Kinetic Scheme

•An ordinary differential equation (ODE) model describing signaling from EGF and HRG to ERK and Akt over a 30 minute time course•117 species (ODEs)•96 net reactions (combined forward and reverse)

Data

Page 13: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Generating Hypotheses—Sensitivity Analysis

Birtwistle et al., Mol Syst Biol, 2007

Perturb Parameters for Negative Regulatory Processes—Simulate Effects on Signaling Dynamics

PTP-1BERK to Receptors

ERK to SOS

ERK to Gab

Receptor to RasGAP

Receptor Trafficking

Simulated ppERK dynamics-10-fold ErbB2 overexpression

Quantitatively corroborated by proteomic data of Wolf-Yadlin et al. (2006) and Kumar et al. (2007); 10 and 30 min after EGF stimulation, ERK activation is 1.15 to 2-fold higher in ErbB2-overexpressing human mammary epithelial cells

Page 14: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

How Does This Receptor and

Negative Feedback Control Hypothesis

Apply to the Classical PC-12 Cell

System?

von Kriegsheim, Baiocchi, Birtwistle et al., Cell fate decisions are specified by the ERK interactome Nat Cell Biol, 2009

Page 15: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Ras vs. Rap1 Activation?

15

York et al. (1998) observed that both NGF and EGF induce transient Ras activation but only NGF induced sustained Rap1 activation. The hypothesis therefore was that this

sustained Rap1 activation leads to sustained NGF-induced ERK activation.

Using FRET-based reporters for Ras and Rap1 activity, however, Mochizuki et al. (2001) showed that both Ras and Rap1 activation are sustained in response to NGF

Dominant negative Ras blocks EGF and NGF-induced ERK activation…

…without affecting Rap1 activation

von Kriegsheim, Baiocchi, Birtwistle et al., Nat Cell Biol, 2009

Page 16: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Ligand-Dependent Positive Feedback?

16

Santos et al. (2007) found that NGF induces positive feedback from ERK to Ras, whereas EGF induces negative feedback from ERK to Ras. This led to the hypothesis

that the NGF-induced positive feedback is what sustains ERK activity

Test: Stimulate PC-12 cells with NGF in the presence of U0126 (inhibitor of ERK activation) to break the positive feedback, and then measure upstream Ras and MEK activation.

If positive feedback is responsible for sustained ERK activity, then Ras and MEK activation should be transient with U0126.

Mechanisms for generating transient vs. sustained ERK signaling in PC-12 cells remain unclear…

von Kriegsheim, Baiocchi, Birtwistle et al., Nat Cell Biol, 2009

Page 17: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

A Quantitative, Interaction Proteomics Approach Identifies New Players

17

Neurofibromin-1 (NF1) Phosphoprotein enriched in astrocytes 15 (PEA-15)

•The predominant RasGAP in PC12 cells

•Displays transient vs. sustained dissociation from Ras, which is may be due to ERK phosphorylation of NF1

•A cytoplasmic anchor for ERK when unphosphorylated

•Phosphorylated by either ERK dependent mechanisms (S104) or by Akt directly (S116), and then dissociates from ERK, allowing ERK to enter the nucleus

•Potential positive feedback from ERK to Ras •Positive crosstalk from the Akt pathway

von Kriegsheim, Baiocchi, Birtwistle et al., Nat Cell Biol, 2009

Unfortunately these players don’t reveal a mechanisms for generating transient vs. sustained ERK signaling

Page 18: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

A Systems-Level View of EGF and NGF Signaling in PC-12 Cells

18

•PKC and RKIP control ERK dynamics (Santos et al., 2007).

•PKC phosphorylates and inactivates RKIP (Corbit et al., 2003).

•PKC is activated via the PLC pathway, which also displays transient vs. sustained behavior:

•PLC signaling is predominantly effective at the plasma membrane where its substrate PIP(4,5)2 is present (Haugh et al., 2002; Di Paolo et al., 2006).

•NGF signaling is shifted towards the plasma membrane by the p75 neurotrophin (NTR) receptor (Saxena et al., 2005), whereas EGF signaling is shifted towards the cell interior due to rapid ligand induced-internalization.

•Active PLC induces Ras activation through the GRP family of guanine exchange factors in PC-12 cells (as phorbol esters) (Buday et al., 2008; Brose et al., 2002)

•Distinct from the canonical SOS pathway which is subject to negative feedback.

von Kriegsheim, Baiocchi, Birtwistle et al., Nat Cell Biol, 2009

Page 19: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

A Systems-Level View of EGF and NGF Signaling in PC-12 Cells

19

•Describes signaling from EGF and NGF to ERK over a 30 minute time course in PC-12 cells

•System of 35 ordinary differential equations based on standard chemical reaction kinetics rates

•73 kinetic parameters whose values are constrained by our own data and those of Sasagawa et al. (2005) and Santos et al. (2007).

Based on this model, the main difference between NGF and EGF signaling (sustained vs transient) is receptor localization

Page 20: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Experimentally Testing the Receptor Localization Hypothesis

20

•TGF activates EGFR, but dissociates from EGFR in the endosomes causing receptor recycling. TGF should therefore sustain ERK signaling relative to EGF

•Cbl activity is needed for EGFR internalization, and the Cbl-70Z mutant is an inactive dominant negative. Therefore 70Z should sustain EGF-induced ppERK.

•Knock-down of PLC should make NGF-induced signaling more transient.

von Kriegsheim, Baiocchi, Birtwistle et al., Nat Cell Biol, 2009

Page 21: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

MCF-7 cells; HRG induces differentiation (lipid droplet accumulation); EGF induces only proliferation

Nakakuki*, Birtwistle* et al., Ligand-specific c-Fos responses emerge from the spatiotemporal control of ErbB network dynamics. Cell, 2010

How Are Different ERK Dynamics Interpreted?

Transient

Sustained

Page 22: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

c-Fos Acts as a Sensor for the ERK Activation Dynamics

Nakakuki*, Birtwistle*, et al., Cell, 2010

Adapted from Murphy et al., Nat Cell Biol 2002

Why does sustained ERK signaling cause transient c-

fos mRNA expression?

How robust are the all-or-nothing pc-Fos responses?

Page 23: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Building a Mechanistic Model

ODE model based on mass-action and Michaelis-Menten kinetics

Parameter estimation with a genetic algorithm

Nakakuki*, Birtwistle*, et al., Cell, 2010

Page 24: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Model Predicts that DUSPs Control the c-fos mRNA Kinetics

EGF HRG

Nakakuki*, Birtwistle*, et al., Cell, 2010

Page 25: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

DUSP Knockdown Doesn’t Significantly Affect HRG-induced c-fos mRNA Kinetics

Nakakuki*, Birtwistle*, et al., Cell, 2010

EGFHRG

Page 26: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

What’s wrong with the current model?

Hypothesis: c-Fos induces or perhaps is its own transcriptional repressor

Nakakuki*, Birtwistle*, et al., Cell, 2010

Page 27: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Evidence for the Refined Model

EGF HRG

Nakakuki*, Birtwistle*, et al., Cell, 2010

Page 28: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

How Robust is This System?The “Core” Model

Training: 10 nM

Validation: 1, 0.1 nM

Nakakuki*, Birtwistle*, et al., Cell, 2010

Page 29: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

How Robust is This System?

Cascade of coherent feedforward loops

Robustness to Input Disturbances Robustness to Internal Perturbations

Robustness: Sum over all inverse, absolute parameter sensitivity coefficients for the time-integrated pc-Fos response.

Inner CFL Intact

Inner CFL Broken

Nakakuki*, Birtwistle*, et al., Cell, 2010

Integral Negative Feedback

Page 30: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

How General is the Core Model?Predicting the responses in PC-12 cells

Predicting the responses to EGF+PMA in MCF-7 cells

Nakakuki*, Birtwistle*, et al., Cell, 2010

Page 31: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Quantifying the Transient vs. Sustained Paradigm

Approximating the ppERK dynamics

A master relationship between the transience of ppERK signaling and the integrated pc-Fos

response

Nakakuki*, Birtwistle*, et al., Cell, 2010

Page 32: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

• Generation:– Ligand-specific, spatiotemporal signaling patterns are

generated by negative feedback coupled with receptor trafficking.

• Interpretation:– Transcriptional circuits consisting of integral negative

feedback and cascades of coherent feedforward loops provide robust interpretation of transient vs. sustained signaling dynamics.

Conclusions-I

Page 33: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Cell-to-Cell Variability in Protein Expression and ErbB Signaling

Page 34: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Cell-to-Cell Heterogeneity in ErbB Network Signaling and Phenotypes

It is widely thought that cell-to-cell variability in protein expression levels plays a major role

Sigal et al., 2006Spudich et al., 1976McAdams and Arkin, 1997to name a few…

Page 35: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Burst-like gene expression in mammalian cells

35

Raj et al., Stochastic mRNA Synthesis in Mammalian Cells, PLoS Biology, 2006

Large cell-to-cell variability in mRNA levels of the YFP reporter gene

Variability seems to arise from changes in local chromatin structure

Page 36: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Cell-to-cell variability in protein expression in response to a topoisomerase-1 inhibitor

(camptothecin)

Cohen et al., Dynamic Proteomics of Individual Cancer Cells in Response to a Drug, Science, 2008

Library of cell lines expressing YFP-tagged proteins under control of the endogenous

promoter and automated microscopy methods

Large cell-to-cell variability Divergent

responses

Cell-to-cell variability in DDX5 can partially determine whether a cell lives or dies in response to the topo-1 inhibitor

Page 37: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

• What kind of probability distribution characterizes this cell-to-cell variability in protein expression?

• What are the implications for ErbB signaling dynamics?

Questions

37

Page 38: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

The “Standard” Gene Expression Model

38

•Stochastic simulations using the Gibson and Bruck method in Copasi

•Six parameters characterizing the model were combinatorially varied to make 6400 total combinations

•Each parameter combination was simulated 707 times to get an approximate steady-state protein abundance distribution

Birtwistle, MR, Kholodenko, BN, Kolch, W. Mammalian Protein Expression Noise: The Gamma Distribution and Implications for Overexpression and Knockdown Experiments. To be submitted.

Page 39: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Results are well-modeled by a gamma distribution

39

SSE=0.24 SSE=0.66

SSE=0.12 SSE=0.21

Fre

qu

en

cy

Fre

qu

en

cy

SSE

Protein Abundance [#]

gamma lognormal Weibull

Fre

qu

en

cy

SS

E

)()(

/)1(

obsk

obs

Nk

k

eNNf

obs

obsobs

Gamma pdf:

Gamma distribution describes data better than lognormal or Weibull distributions

Page 40: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Some Flow Cytometry Measurements Compared to the Gamma Distribution

Model

40

LS174 cells

A2780 cells

MEFs with inducible K-Ras downregulation

Total K-Ras [AU]

Birtwistle, MR, Kholodenko, BN, Kolch, W. Mammalian Protein Expression Noise: The Gamma Distribution and Implications for Overexpression and Knockdown Experiments. To be submitted.

Page 41: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Predictions of the Gamma Distribution Model: Noise Scaling with the Mean

41

kobsobs;2 kobs obs 2;

1

kobs

CV 0.5

1

kobs

Tet-regulated systems and RNAi should affect obs but leave kobs constant (Raj et al., 2006)

Birtwistle, MR, Kholodenko, BN, Kolch, W. Mammalian Protein Expression Noise: The Gamma Distribution and Implications for Overexpression and Knockdown Experiments. To be submitted.

Page 42: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Comparison to Flow Cytometry Experiments

42

Inducible RKIP Downregulation by shRNA in LS174 cells

Inducible WT K-Ras downregulation in MEFs without endogenous Ras

Inducible K-Ras V12 downregulation in in MEFs without endogenous Ras

Birtwistle, MR, Kholodenko, BN, Kolch, W. Mammalian Protein Expression Noise: The Gamma Distribution and Implications for Overexpression and Knockdown Experiments. To be submitted.

Page 43: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Cell-to-Cell Heterogeneity in EGF-induced ERK Signaling

Birtwistle et al., Mixed Analog-Digital Responses Determine Single Cell and Population Switches in MAPK Signaling. Under Revision

EGF-induced ERK activation in HEK293 cell populations measured by flow cytometry

Cell-to-cell variability in total ERK levels doesn’t affect bimodality

Page 44: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

What Kind of ERK Cascade Might Account for the Observed Behavior?

Raf pRaf

RasGTP

v1

v2

MEK pMEK

v3

v6

ppMEK

v4

v5

ERK pERK

v7

v10

ppERK

v8

v9

v11

Feedback• positive—PF (Fa=5)• negative—NF (Fa=0.5)• none—US (Fa=1)

• Model adapted from Markevich et al., Mol Syst Biol, 2006

• Topology changed by varying the feedback strength (Fa)

• Total protein levels (RasGTP, Raf, MEK, ERK) sampled from a gamma distribution

• RasGTP dynamics estimated from experimental data

1. Distributions of active ERK display bimodal/shouldering behavior with increasing EGF dosea. Characteristic of bistable or

ultrasensitive dose-responsepositive or no feedback

2. The ERK-on population mean exhibits analog behavior at shorter times, but effective digital behavior at longer times and becomes smaller as time progressesa. Characteristic of negative feedback

Birtwistle et al., Mixed Analog-Digital Responses Determine Single Cell and Population Switches in MAPK Signaling. Under Revision

Page 45: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Comparing Model Simulations with the Observed Population ResponsesData Model

Birtwistle et al., Mixed Analog-Digital Responses Determine Single Cell and Population Switches in MAPK Signaling. Under Revision

All three models give rise to bimodal/shouldering behavior

Only negative feedback model shows proper dose-response

DataModel

Page 46: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Ongoing Testing of the Modeling Predictions

Negative Feedback should be dominant over the investigated time scales

Desensitization of RasGTP levels should underly the mixed analog-digital behavior of the ERK-on population mean

Birtwistle et al., Mixed Analog-Digital Responses Determine Single Cell and Population Switches in MAPK Signaling. Under Revision

DataModel

Page 47: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

• Cell-to-cell variability in protein expression seems to be well-described by a gamma distribution

• Bimodal population responses of ERK activation can emerge from a negative feedback system combined with cell-to-cell variability in protein expression

Conclusions-II

Page 48: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

• Systems Biology Ireland– Kolch Group– Kholodenko Group (at

Thomas Jefferson University as well)

– von Kriegsheim Group

Acknowledgements

48

• Ogunnaike Group at the University of Delaware, USA

• Hatekeyama Group at RIKEN, Yokohama, Japan

• Funding and Partners– Marie Curie

International Incoming Fellowship

– EMBO Long-term fellowship

Page 49: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

EXTRA

Page 50: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Protein Expression: A simple bursting model

50

11 bN

2121deg bebN tk

32)(

132deg21deg bebebN tkttk

1t 2t

3t

1b

2b

3b

1

1deg

1

1deg

1

1

k

jj

k

jj

tk

i

k

iii

k

i

tk

ik

e

b

ebN

•Limit as k∞ will give us the desired distribution, which we denote f(Nss)•Nss is a sum of identically distributed, yet non-identically weighted exponential random variablesNot identically gamma; closed-form solution unknown

/1

)(;1

~ BeBfExpB

Page 51: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Run Some Simulations…stochastic sampling of burst sizes and between-burst intervals

Calculate Nss in 10,000 “cells” for different values of , kdeg, and

51

Nss: Random variable; protein abundance directly after an expression burst: Mean number of proteins produced per expression burstkdeg: First-order protein degradation constant: Mean waiting time between expression bursts

Page 52: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Simulation Results are Well-Modeled by the Gamma Distribution

N ss ~Gam(keff , eff ); f (N ss )N ss

(keff 1)e Nss / eff

eff keff (keff )

Page 53: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Observed Protein Abundance Distribution

53

Nobs

t

Nss

t

tobs

obstkssobs eNN deg

Observed protein abundance, Nobs, is an explicit function of two random variables:

Nss: Number of proteins after an expression burst; gamma

tobs: Time of observation after an expression burst; uniform between 0 and t

Page 54: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Run Some Simulations…stochastic sampling of burst sizes, between-burst intervals, and

observation time

Calculate Nobs in 10,000 “cells” for different values of and kdeg

54

Nobs: Random variable; observed protein abundance: Mean number of proteins produced per expression burstkdeg: First-order protein degradation constant: Mean waiting time between expression bursts

Page 55: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

The Gamma Distribution Again

55

Page 56: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

56

What about a more realistic gene expression model?

Page 57: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

More Knockdown Experiments

57

A2780 cells

MDA-MB-237 and SW480 cells; Lapan et al., 2008

Fold Mean Expression Change = 0.57

Fold Mean Expression Change = 0.61

Fold Mean Expression Change = 0.71

Page 58: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

What Does This Mean for Knockdown Experiments?

58

Page 59: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

What Does This Mean for Knockdown Experiments?

59

Dox [ng/mL]R

ela

tiv

e R

KIP

A

bu

nd

an

ce

LS174 colon carcinoma cells with tet-inducible RKIP downregulation by shRNA expression

Assay RKIP levels by flow cytomtery

Page 60: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Can We Get a “Good” Knockdown?

60

Endogenous Ras-less MEFs with engineered tet-off WT K-Ras expressionK-Ras levels measured by flow cytometry

Page 61: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Some Fluorescence Microscopy Measurements Compared to the

Gamma Distribution Model

61

Raj et al., 2006, CHOs stably expressing CFP or YFP

Lapan et al., 2008, PTEN data from MDA-MB-231 cells; STAT3 data from SW480

Page 62: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Predictions of the Gamma Distribution Model: Noise Scaling with the Mean

62

2

2kobs obs

2

kobs obs

obs

Based on modeling, protein half-life and burst frequency affect kobs. So far we’ve have had trouble experimentally manipulating these

cleanly. Birtwistle, MR, Kholodenko, BN, Kolch, W. Mammalian Protein Expression Noise: The Gamma Distribution and Implications for Overexpression and Knockdown Experiments. To be submitted.

Page 63: Predicting ErbB Network Signaling Dynamics Marc R Birtwistle 16 Sept 2011 Georgia Health Sciences University Dept. of Biostatistics Research Forum

Data from yeast support this scaling behavior

63

Log scale makes inverse proportionality trend appear linear

43 GFP tagged protein proteins under 11 different environmental conditions in S. cerevisiae (Bar-Even et al, 2006)

Birtwistle, MR, Kholodenko, BN, Kolch, W. Mammalian Protein Expression Noise: The Gamma Distribution and Implications for Overexpression and Knockdown Experiments. To be submitted.