Control of disease related signal transduction networks

Transcription

1 Control of disease related signal transduction networks Réka Albert Departments of Physics and Biology Huck Institutes for the Life Sciences Pennsylvania State University

2 Connect within-cell networks to cell behavior through dynamic modeling Cell motion, proliferation, programmed cell death are regulated through signal transduction, gene regulatory and/or metabolic networks. Mutations and deregulations of network nodes may lead to wrong cellular behavior, which often leads to disease. Each component is characterized with a state and with a regulatory function that connects the state of its regulators to its own future state. Mutations and deregulations are incorporated by altered node states or altered regulatory functions. The long-term states of a subset of the nodes can be related to cell behavior.

3 A parsimonious dynamic modeling approach: Boolean modeling with stochastic update Main assumption: components have two main states: ON (1) or OFF (0) The future state of a regulated node depends on the current state of its regulators in the network. This dependence is described by Boolean logic (using NOT, AND, OR) Node states are updated probabilistically (up to two classes). Thank you Eduardo! In1 In2 Out Starting from an initial condition, the system s state vector changes in time and eventually settles down into an attractor (a steady state or a complex attractor). Out= In1 OR In2 Each attractor, or group of attractors that are identical in terms of the relevant node subset, can be associated to a cell behavior or phenotype. All attractors can be identified.

4 Boolean models are a limit of nonlinear dynamical systems If ν is large, the synthesis function approaches T max ρ X B(Y) B(Y)=1 if Y K Y and 0 otherwise Piece-wise linear models (L. Glass) Boolean simplification: Both Y and X are binary variables; X follows Y after an unspecified time delay X* = Y Pioneers: Stuart Kauffmann, Leon Glass, Rene Thomas Logical models (R. Thomas, D. Thieffry) Polynomial dynamical systems (Reinhard Laubenbacher) Conversion into continuous models (SQUAD, ODEfy)

5 Examples of Boolean models by our group Drosophila embryonic development: Albert & Othmer, J Theor Biol 2003 Chaves, Albert, Sontag, J Theor Biol 2005, IEE Proc Sys Biol 2006, Chaves & Albert J R S Interface 2008 Signaling in plants (with Sally Assmann): Li, Assmann, Albert, PLoS Biology 2006 Saadatpour et al., J Theor Biol 2010 Sun, Jin, Albert, Assmann, PLOS Comp Biol 2014 Host-pathogen interactions (with Eric Harvill, Isabella Cattadori): Thakar et al. PLoS Comput Biol 2007, J R S Interface 2009, PLoS Comput. Biol 2012 T cell survival signaling in T-LGL leukemia (with Thomas Loughran): Zhang et al PNAS 2008, Saadatpour et al PLoS Comp Biol 2011 Epithelial to mesenchymal transition in liver cancer (with Thomas Loughran): Steinway et al. Cancer Research 2014, Syst. Biol. and Applications 2015 Plant-pollinator communities (with Katriona Shea): Campbell et al PNAS 2011, PRE 2012, Theor Ecol 2013, Theor Ecol 2014

6 Modeling and experiments on T cell survival in T-LGL leukemia Ranran Zhang Modeling and experiments on epithelial to mesenchymal transition in liver cancer Steven Steinway Control of intracellular network dynamics See Jorge GT Zañudo s talk on Friday Jorge G T Zañudo

7 Steps in constructing a Boolean model of a within-cell network 1. Needed: components; interactions; binary states of components in certain known conditions 2. Construct the interaction network 3. Determine the Boolean regulatory functions for each node. When there are multiple regulators, select the function that best represents the existing knowledge about their combinatorial action. 4. Determine the relevant initial condition(s) 5. Choose a time implementation (update schedule) 6. Analyze the model: determine the attractors and their basins, determine the effects of node perturbations on the attractors/basins 7. Compare with known behavior. If there are discrepancies, revise the network and/or the regulatory functions 8. Use the model to make novel predictions, e.g. of therapeutic strategies.

8 Example 1: Modeling T cell survival Phenomenon: survival of cytotoxic T cells in T-LGL leukemia Constructed: survival signaling network inside T- LGL cells Hypotheses: two protein/mrna states R Zhang, MV Shah, J Yang, SB Nyland, X Liu, JK Yun, R Albert, TP Loughran, PNAS (2008). A Saadatpour, RS Wang, A. Liao, X Liu, TP Loughran, I. Albert, R. Albert, PLoS Comp Biol (2011)

9 T cell survival signaling network Rectangle: intracellular; ellipse: extracellular; diamond: receptor. Upregulated, downregulated, deregulated node; Activation, inhibition edge

10 Stochastic Boolean dynamic model Two states: ON, OFF Inhibitors NOT Conditional activation AND Independent activation OR PI3K*= PDGFR OR RAS TCR *= Stimuli AND NOT CTLA4 Initial state characteristic to resting T cells + Stimuli= ON +? (unknown number of driver deregulations) Stochastic asynchronous update Multiple (>300) replicate simulations for each initial condition Simulation ends when Apoptosis (cell death) is activated. Output of the model: the frequency of node activation among simulations that are still alive

11 Minimum condition to reproduce a T-LGLlike state Two attractors: Normal: Apoptosis=ON (cell death) T-LGL: Apoptosis=OFF, known deregulated states for other nodes C Minimum condition: IL-15 constantly ON, PDGF intermittently ON, Stimuli initially ON. No mutations necessary, it may all be in the environment. Frequency of node activation JAK IL-2 Proliferation Apoptosis Rounds of updating

12 Manipulations that reverse T-LGL survival Assume IL-15, PDGF present Permanently reverse the T-LGL state of a single node Candidate therapeutic target: altering its state causes apoptosis in all simulations Newly identified: SPHK1, NFκB, S1P, SOCS, GAP, BID and IL2RB Validation: NFκB constitutively active in T-LGL Model: NFκB stabilizes at ON, setting it OFF causes total apoptosis Validation: NFκB inhibition induces apoptosis in T- LGL

13 Example 1: Modeling T cell survival Phenomenon: survival of cytotoxic T cells in T-LGL leukemia Constructed: survival signaling network inside T- LGL cells Hypotheses: two protein/mrna states Validation: reproduces the two known outcomes (apoptosis and abnormal survival state); reproduces known key mediators Predicts: minimal initial condition necessary for survival state 10 new manipulations that ensure apoptosis of T-LGL cells 12 additional deregulations in survival state Several predictions were validated experimentally Implications: identifying therapeutic targets for T-LGL leukemia R Zhang, MV Shah, J Yang, SB Nyland, X Liu, JK Yun, R Albert, TP Loughran, PNAS (2008). A Saadatpour, RS Wang, A. Liao, X Liu, TP Loughran, I. Albert, R. Albert, PLoS Comp Biol (2011)

14 Example 2: Modeling epithelial to mesenchymal transition Phenomenon: cell fate change that starts cancer metastasis Constructed: signaling network leading to loss of E-cadherin Hypotheses: two protein/mrna states, two timescales SN Steinway, JGT Zanudo, W Ding, CB Rountree, D Feith, TP Lougran, R. Albert, Cancer Reseach (2014). SN Steinway, JGT Zanudo, PJ Michel, D Feith, TP Loughran, R. Albert, npj Systems Biology and Applications, accepted

15 Epithelial to mesenchymal transition: many signals, one outcome Blue: external signals Green: transcription factors Red: EMT driven by loss of E-cadherin Steinway et al, Cancer Research 2014

16 Stochastic Boolean dynamic model Inhibitors Conditional activation Independent activation NOT AND OR E-cadherin* = β-catenin_memb and (not SNAI1 or not HEY1 or not ZEB1 or not ZEB2 or not FOXC2 or not TWIST1 or not SNAI2) EMT = not β-catenin_memb or (SNAI1 and HEY1 and ZEB1 and ZEB2 and FOXC2 andtwist1 and SNAI2) Initial state characteristic to epithelial cells, then TGFβ initiated Stochastic update with fast (post-translationally regulated) and slow (transcriptionally regulated) nodes. A timestep defined as the average number of updates needed to update a slow node. Multiple (>10000) replicate simulations Output: the frequency of activation of each node versus timesteps

17 Dynamics of the EMT network Two steady states without external signals: Color coded node activity l Epithelial state (E) l Mesenchymal state (M) l The model captures known features of these states Epithelial Mesenchymal SN Steinway, JGT Zañudo, W Ding, CB Rountree, D Feith, TP Loughran, R Albert. Cancer Res 74 (21) (2014).

18 Cancer metastasis Dynamics (EMT) of network the EMT network Two steady states without external signals: Color coded node activity l Epithelial state (E) l Mesenchymal state (M) l Epithelial + TGFβ Mesenchymal Epithelial +TGFβ Mesenchymal SN Steinway, JGT Zañudo, W Ding, CB Rountree, D Feith, TP Loughran, R Albert. Cancer Res 74 (21) (2014).

19 Cancer metastasis Dynamics (EMT) of network the EMT network l Two steady states without external signals: Color coded node activity l Epithelial state (E) l Mesenchymal state (M) l Epithelial + TGFβ Mesenchymal l The model reproduces the known aspects of EMT l The model predicts acbvabon of mulbple signaling pathways during TGFβ induced EMT. l PredicBon validated experimentally Epithelial +TGFβ Time steps Mesenchymal SN Steinway, JGT Zañudo, W Ding, CB Rountree, D Feith, TP Loughran, R Albert. Cancer Res 74 (21) (2014).

20 How do we intervene to suppress EMT? Try all combinabons of knockouts/ overexpressions in the model (up to 4 nodes). Single KOs Double KOs Triple KOs TFs validated experimentally

21 Example 2: Modeling epithelial to mesenchymal transition Phenomenon: cell fate change that starts cancer metastasis Constructed: signaling network leading to loss of E-cadherin Hypotheses: two protein/mrna states, two timescales Validation: reproduces known molecular markers and known key mediators of EMT Predicts: activation of several unexpected pathways successful combinatorial interventions that block TGFβ driven EMT Several predictions were validated experimentally Implications: identifying therapeutic targets to prevent cancer metastasis SN Steinway, JGT Zanudo, W Ding, CB Rountree, D Feith, TP Lougran, R. Albert, Cancer Reseach (2014). SN Steinway, JGT Zanudo, PJ Michel, D Feith, TP Loughran, R. Albert, npj Systems Biology and Applications, accepted

22 A better way to search for candidate interventions Which nodes/groups of nodes determine cell behavior (attractors)? Can we answer this question without dynamic simulations? Yes, through an integration of the interaction network and of the Boolean regulatory rules Can be used to Determine elementary signaling modes (analog of S-systems) Determine each node s contribution to outcome Find the centers of stability in the network Simplify the original network Drive the system to or away from attractors R.S. Wang and R. Albert BMC Systems Biology (2011) JGT Zanudo, R. Albert, Chaos (2013) JGT Zanudo, R. Albert, PLOS Comp. Biol. (2015)

23 Network expansion allows the identification of centers of stability Eliminate negative edges, eliminate AND/OR ambiguity A complementary (negated) node is added for each node; a composite node is added for each AND rule C* = A and (~B) ~C* = ~A or B

24 Network expansion allows the identification of centers of stability A complementary (negated) node is added for each node; a composite node is added for each AND rule Stable motif: the smallest strongly connected component that - Does not contain both a node and its negation. - If it contains composite nodes, it also needs to contain these nodes inputs. The nodes of a stable motif will have a steady state in any attractor of the network.

25 Network expansion allows the identification of centers of stability Stable motif: the smallest strongly connected component that - Does not contain both a node and its negation. - If it contains composite nodes, it also needs to contain these nodes inputs. The nodes of a stable motif will have a steady state in any attractor of the network. A = 0 B = 0 C = 0

26 Network simplification by reduction of stable motifs 1. Create expanded network (complementary, composite nodes). 2. Identify stable motifs. 3. Reduce network using the state of one of these stable motifs. 4. Repeat as necessary The end result is either a steady state of the system, or a partial steady state, in which some nodes have fixed states and others are expected to oscillate. A = Oscillates? B = Oscillates? C = 1 D = Oscillates? E = 0, F = 0 G = 0, H = 1 I = 1 J = Oscillates? K = Oscillates? J. G. T. Zañudo, R. Albert, Chaos 2013

27 Stable motifs in the T-LGL network Ceramide = OFF S1P = ON PDGFR = ON SPHK1 = ON S1P = OFF PDGFR = OFF SPHK1 = OFF J. G. T. Zañudo, R. Albert, Chaos 2013

28 Stable motif succession diagram summarizes all trajectories to the two outcomes grey: OFF black: ON J. G. T. Zañudo, R. Albert, PLOS Comp Biol 2015

29 Control based on stable motifs u Control: influence the system s natural dynamics so that it converges into a target attractor u u A sequence of stable motifs determines an attractor Fix the state of all stable motifs in a sequence all initial conditions go to target attractor. u Reduce the number of nodes whose state needs to be fixed. control of a subset of the nodes may stabilize the motif some motifs in the sequence are not branching points in the dynamics so do not need control J. G. T. Zañudo, R. Albert, PLOS Comp Biol (2015)

30 Control of T cell outcome Leukemia network Ceramide = OFF S1P = ON PDGFR = ON SPHK1 = ON Leukemia (diseased state) S1P = OFF PDGFR = OFF SPHK1 = OFF Apoptosis (normal state) grey: OFF black: ON 9 interventions that lead to apoptosis, each effective for all initial conditions, even when non-permanent. Interventions effective for a continuous version of the model as well.

31 Opposing the state of a motif may block the system from reaching an undesired attractor Ceramide = ON 7 interventions that block the T-LGL attractor from >90% of initial conditions, one of them effective when non-permanent. Interventions effective for a continuous version of the model as well.

32 Example 1: Modeling T cell survival Phenomenon: survival of cytotoxic T cells in T-LGL leukemia Constructed: survival signaling network inside T- LGL cells Hypotheses: two protein/mrna states Validation: reproduces the two known outcomes (apoptosis and abnormal survival state); reproduces known key mediators Predicts: minimal initial condition necessary for survival state 26 new potential therapeutic strategies Several predictions were validated experimentally Implications: identifying therapeutic targets for T-LGL leukemia R Zhang, MV Shah, J Yang, SB Nyland, X Liu, JK Yun, R Albert, TP Loughran, PNAS (2008). A Saadatpour, RS Wang, A. Liao, X Liu, TP Loughran, I. Albert, R. Albert, PLoS Comp Biol (2011) J. G. T. Zañudo, R. Albert, PLOS Comp Biol (2015)

33 In the EMT network any one of eight stable motifs in the leads to the mesenchymal state Plus five smaller ones Blue: node OFF Yellow: node ON SN Steinway, JGT Zanudo, W Ding, CB Rountree, D Feith, TP Lougran, R. Albert, Cancer Reseach (2014).

34 Control sets to reach the epithelial state Stable motif associated with the E state Control of one node in each yellow rectangle is sufficient for convergence to the epithelial state. e.g. SMAD, SNAI1, RAS, SHH knockout, β-catenin_memb const. expr. Black: OFF in the E state White: ON in the E state Blue: part of a SMAD+ X intervention identified by systematic analysis Steinway et al, npj Systems Biology and Applications, accepted

35 Preventative versus corrective interventions Control of SMAD, SNAIL, RAS, SHH, β-catenin_memb sufficient to convergence to epithelial state from any initial state. SNAIL knockout is sufficient to prevent TGFβ-induced EMT. The resulting state has high E-cadherin. In case of deregulations (e.g. gain of function mutation), the second intervention may be more relevant.

36 Example 2: Modeling epithelial to mesenchymal transition Phenomenon: cell fate change that starts cancer metastasis Constructed: signaling network leading to loss of E-cadherin Hypotheses: two protein/mrna states, two timescales Validation: reproduces known molecular markers and known key mediators of EMT Predicts: activation of several unexpected pathways centers of stability points of no return combinatorial interventions to - prevent EMT Several predictions were validated experimentally - recover the epithelial state Implications: identifying therapeutic targets to prevent cancer metastasis SN Steinway, JGT Zanudo et al., Cancer Reseach (2014). JGT Zanudo, R. Albert, PLOS Comp. Biol. (2015) SN Steinway, JGT Zanudo et al. npj Systems Biology and Applications, accepted

37 Acknowledgements Ranran Zhang T-LGL model and experiments Assieh Saadatpour Network reduction, attractor analysis Ruisheng Wang Steven Steinway Jorge G T Zañudo Key mediator analysis EMT model and experiments Stable motif analysis, control Colin Campbell Attractor stabilization with added edges National Science Foundation