Logical modelling of cellular decision processes with GINsim C. - - PowerPoint PPT Presentation

Logical modelling of cellular decision processes with GINsim

JOBIM, Rennes, July 5h, 2012

• C. Chaouiya, A. Naldi, L. Spinelli, P. Monteiro, D. Berenguier,
• L. Grieco, A. Mbodj,S. Collombet, A. Niarakis, L. Tichit,
• E. Remy & D. Thieffry

Cell proliferation, differentiation

• r death...

How are decisions taken?

Dynamical modelling

Why ?

• To gain rigorous, global, functional understanding of the

(complex) underlying networks

• To predict the behaviour of the system in novel situations
• To design novel experiments

How ?

• Regulatory charts/maps/graphs (CellDesigner, Cytoscape)
• Qualitative modelling: Boolean / multilevel discrete

networks

• Quantitative modelling: ODE, PDE, Stochastic equations

Boolean networks - Stuart Kauffman (1969)

The Boolean vector x represents the state of the system Random connections, nodes with predefined degree Canalizing Boolean functions Focus on asymptotic behaviour Two types of attractors: stable states and (simple) cycles Deterministic behaviour (only one possible following state)

xt +1 = B(xt )

Kinetic logic - René Thomas (1973)

Xi (image or logical function) specifies whether gene i is currently transcribed xi (logical variable) denotes the presence (above a threshold of the functional product of gene i

X = B(x)

t Xi

Gene i switched ON Gene i switched OFF 1

xi

1 Delay dOFF Delay dON

Asynchronous, potentially non deterministic behaviour!

Logical modelling of regulatory networks

 A graph describes the interactions between genes or regulatory products Logical rules/parameters

KA = 2 IFF (C=0) KA = 0 otherwise

 Discrete levels of expression associated to each regulatory component and interaction

KC = 1 IFF (B=1) AND (C=0) KC = 0 otherwise B C C

1

C

Decision trees

2

C C=0 1 KB = 1 IFF (A=1) KB = 0 otherwise

1

A 2 1

[1] [2]

B C A

Logical modelling of regulatory networks

 A graph describes the interactions between genes or regulatory products Logical rules/parameters

2

C

 Discrete levels of expression associated to each regulatory component and interaction

B C

1 1

A

Decision diagrams

KA = 2 IFF (C=0) KA = 0 otherwise KB = 1 IFF (A=1) KB = 0 otherwise KC = 1 IFF (B=1) AND (C=0) KC = 0 otherwise C=0 1 2 1

[1] [2]

B C A

Logical state transition graphs

Asynchronous updating (R Thomas, L Glass)

ABC C↑ C↓ B↓ B↓ A↑

Regulatory graph + Logical rules => simulations / dynamical analysis State transition graph

Stable state

[1] [2]

B C A

Logical state transition graphs

ABC

Stable state

+ Logical rules

Cycle Cycle

B↓ C↓ C↓

State transition graph Synchronous updating (S Kauffman)

A↑ C↑ A↑

[1] [2]

B C A

GINsim (Gene Interaction Networks simulation)

analysis toolbox core simulator GINML parser user interface

graph analysis graph editor simulation

State transition graph

Regulatory graph Available at http://gin.univ-mrs.fr/GINsim

Aurélien NALDI Fabrice LOPEZ Duncan BERENGIER Claudine CHAOUIYA

Naldi et al (2009) BioSystems 97: 134-9 Chaouiya et al (2012) Methods in Molecular Biology 804: 463-79

Development of dynamical analysis tools

• Decision diagrams
• Identification of attractors [Naldi et al, 2007]
• Analysis of regulatory circuits [Naldi et al, 2007]
• Model reduction [Naldi et al, 2010]
• Compression of state transition graphs

[Berenguier et al, in prep]

• Priority classes
• Mixed a/synchronous simulations [Fauré et al, 2006]
• Petri nets (standard or coloured)
• Model checking
• Verification of dynamical properties (temporal logic)

[Sanchez et al, 2008]

• Constraint programming
• Model identification (regulatory interactions, thresholds, rules)

[Corblin et al, 2001]

Coping with the exponential growth

• f logical state transition graphs
• Model reduction
• Attractor identification
• Temporisation (e.g. priorities, delays, etc.)
• Compaction of state transition graphs

IRF1 IL4 CGC IFNB_e IL12_e STAT3 IL12RB2 IL4R IL17 TBET IL10 IL23R GP130 IL21 STAT6 IL6_e proliferation APC IL15_e CD28 IL2 IL12RB1 IFNGR1 IFNGR STAT4 SMAD3 IL2R IL4_e IFNG IL6RA IL4RA STAT1 IFNGR2 IL15RA IKB TCR IL10_e IL15R TGFB_e IFNG_e IL10RB IL10R IL23_e IL2RA NFKB STAT5 NFAT IL27RA IL27_e IL2_e TGFBR RORGT RUNX3 IFNBR IL10RA IL21R GATA3 IL21_e IL6R TGFB IL23 IL27R IL12R FOXP3 IL2RB

13 input components, 52 internal components, 339 circuits => too large to perform simulations

Logical model of the Th network

Naldi et al (2010) PLoS Comput Biol 6: e1000912.

IFNBR => 1 IFF IFNR_e =1 STAT1 => 1 IFF INFGR=1 OR IL7R=1 OR IFNB_e =1

Reduced Th model

13 input components 21 internal components

Naldi et al (2011) Theoretical Computer Science 412: 2207-18

State transition graph (704 nodes) Pathways? Crucial decisions?

TH0 + (APC, IL4_e, IL6_e, IL12) ON

2 reachable stable states Th2 Th1

Graph of strongly connected components Also 704 nodes!

Hierarchical state transition graph

2 reachable stable states Tbet+ Gata3+ Gata3+ Tbet+

IL2-, IL4-, IL10+, IL21+, IL23+, STAT1+ IFNg-, IL2-, IL10+, IL21+, IL23+, STAT6+

Th2 Th1

5 nodes! On the fly STG compaction using Tarjan algorithm + decision diagrams Schemata

Overview of the simulation results for ≠ micro-environments

Absence of stimulation Pro-Th1 IL2 & IFNg

• r IL12

APC only Pro-Th2 IL4 & IL6 Pro-Treg IL2 & TGFb

• r IL10

Pro-Th17 IL6 & TGFb

GATA3 Tbet Foxp3 RORγt

Naldi et al (2010) PLoS Comput Biol 6: e1000912.

Main biological applications

• Drosophila development
• Segmentation genetic network [Sanchez et al, 2008]
• Mesoderm specification (heart, muscle, with E Furlong)

[Mbodj et al, in prep]

• Heamatopoietic cell differentiation and activation
• TCR signalling & T cell activation/differentiation (with V Soumelis)

[Naldi et al, 2010]

• Mast cell activation (with M Daëron & B Malissen)
• Specification of haematopoietic lineages (with T Graf)

[Collombet et al, in prep]

• Cell proliferation (DIAMONDS, FP6)

[Fauré et al, 2006 & 2009; Sahin et al, 2009]

• Apoptosis (APO-SYS, FP7) [Calzone et al, 2010]
• MAPK network [Grieco et al, in prep]

Computational modelling of FcεRI signalling during mast cell activation.

Anna Niarakis1, Emrah Kamali1, Yacine Bounab2,3, Marc Daëron2,3, Denis Thieffry1

Introduction

Mast cell activation (Figure 1) is a pivotal event in the initiation of inflammatory reactions associated with allergic disorders. It is triggered by the aggregation of high-affinity IgE receptors (FcεRI), on the mast cell surface [1]. FcεRI aggregation is induced by the binding of a multivalent allergen to FcεRI-bound IgE antibodies. Mast cell activation is a complex process relying on multiple layers of tightly controlled intracellular signalling molecules, which form an intricate network [2, 3]. A global and rigorous understanding of the signalling and cross- regulatory processes involved in mast cell activation requires the integration of public and novel data into a comprehensive computational model. Based on a survey of relevant data published in scientific journals or available in public databases, we are currently building and annotating a comprehensive regulatory map using the software CellDesigner [4]. Ultimately, our modelling analysis should contribute to deepen our understanding of how the different functional outcomes of mast cell activation (degranulation, synthesis of lipidic mediators, induction of cytokine transcription) are articulated at the level of the underlying molecular network, and to delineate means to uncouple these functions and control them separately or collectively.

References

[1] Turner et al. (1999). Nature Reviews 402: B24-30. [2] Cao et al. (2007). Journal of Immunology 179: 5864-76. [3] Gilfillan et al.(2009). Immunological Reviews 228: 149-69. [4] Funahashi et al. (2008). Proceedings IEEE 96: 1254-65. [5] Le Bouteiller et al. (1976). European Journal of Immunology 6, 326-32. [6] Naldi et al. (2009). Biosystems 97: 134-9. [7] Naldi et al. (2010). PLoS Computational Biology 6: e100. [8] Naldi et al. (2011).Theoretical Computer Science 412: 2207-18.

Current Status

This regulatory map currently encompasses 60 components and over 300 interactions, along with annotations and links to databases such as PubMED, EntrezGene and UniProt. This mast cell activation map will serve as a scaffold to generate a dynamical model of the underlying network, using a sophisticated logical modelling approach and the software GINsim [6, 7, 8].

Prospects

Novel proteomic data will be used to delineate salient dynamical features of mast cell response under different conditions (e.g. how the FcεRI signalling network operates in the absence or in the presence of negative regulatory signals triggered by the FcγRIIB or by the transmembrane adaptor LAT2). To progressively improve the predictive power of the resulting model, computational results will be systematically confronted with experimental data. Figure 2. Partial view of the molecular interaction map of FcεRI related mast cell activation. This map has been built with the software CellDesigner (version 4.2). Figure 1. Picture of a mast cell releasing granules containing vasoactive amines and proteases. Reprinted from [5].

Anna NIARAKIS Poster 12

MAPK signalling Bibliographical references Logical model Simulation results

Logical modelling of MAPK pathways Logical modelling of MAPK pathways

Mammalian Mitogen-Activated Protein Kinases (MAPKs) can be activated by a wide variety of stimuli, including growth factors and environmental stresses. Activation of MAPK pathways affects diverse cellular activities, including gene expression, cell cycle machinery, survival, apoptosis and differentiation. To date, three groups of MAPKs have been extensively studied: extracellular regulated kinases (ERK1/2), Jun NH2 terminal kinases (JNK1/2/3), and p38 kinases (p38 α/β/γ/δ). Given the wide spectrum of stimuli and the large number of processes regulated, a fundamental and debated question is how signalling specificity is achieved. At least five inter-related mechanisms have been proposed: [1] E. Zehorai, Z. Yao, A. Plotnikov and R. Seger. The subcellular localization of MEK and ERK – a novel nuclear translocation signal (NTS) paves a way to the nucleus. Mol. Cell. Endocrinol. 314: 213-220, 2010. [2] A. Funahashi, Y. Matsuoka, A. Jouraku, M. Morohashi, N. Kikuchi and H. Kitano. CellDesigner 3.5: A Versatile Modeling Tool for Biochemical Networks. Proc. IEEE 96: 1254-1265, 2008. [3] A. Naldi, D. Berenguier, A. Fauré, F. Lopez, D. Thieffry and C. Chaouiya. Logical modelling of regulatory networks with GINsim 2.3. BioSystems 97: 134-139, 2009. [4] A. Naldi, E. Remy, D. Thieffry, C. Chaouiya. Dynamically consistent reduction of logical regulatory graphs. Theor.

• Comput. Sci. 412: 2207-18, 2011.

Luca GRIECO1,2,4, Laurence CALZONE3, Andrei ZINOVYEV3, Brigitte KAHN-PERLES2, Denis THIEFFRY2,4

Using the CellDesigner map as a reference, we derived a comprehensive logical model for the MAPK network, with the aim to reproduce the response of MAPK cascades to different stimuli and better understand their contributions to cell fate decision (between proliferation, apoptosis and growth arrest). The resulting logical model encompasses the three main MAPK cascades in response to four inputs: EGFR, FGFR3, TGFβ, and DNA damage. The model was built using the GINsim software [3], and encompasses 54 Boolean components. The figure below shows the corresponding regulatory graph (nodes represent regulatory components, each associated with a logical rule, while green and red arcs represent activation and negative interactions, respectively).

To analyse the dynamical properties of this large logical model, we took advantage of a novel model reduction function implemented into GINsim [4], which preserves the attractors of the systems. Using this reduction algorithm we obtained a 14-component model, on which we performed an attractor analysis. The table shows the results obtained through our reduced model simulations, according to an asynchronous updating strategy. (SCC stands for strongly connected component; * denotes the values 0 or 1) We simulated various mutant situations (data not shown), either to further check the coherence of the model with known facts or to predict novel mechanisms. Recapitulation of documented phenomena ✔ In the absence of growth factors (EGF and FGF3), Growth_arrest steady states are obtained by DNA_damage (with or without TGFB) (a4,a7). Apoptosis is present only when PI3K is absent (a3,a8). ✔ EGF stimulus alone is able to activate the ERK cascade, and to block p38 and JNK, causing Proliferation (a15). FGF3 stimulus is weaker than EGF, leading to bistability: a Proliferation steady state and a Growth_arrest steady state are both possible, accompanied by active ERK and inactive p38/JNK (a9,a10). Simulations revealing novel tentative regulatory functions ✔ Role of p53: in wild type, DNA_damage (but not TGFB) is sufficient for apoptosis (a3-a6); upon p53 deletion, both DNA_damage and TGFB are needed for apoptosis induction. Conclusion: following loss of p53, TGFB might play an important role in triggering Apoptosis, by sinergysing with DNA_damage stimulus ✔ Role of ERK: upon removal of all ERK feedbacks towards EGFR/FRS2, growth arrest is completely lost in the presence of FGF3 stimulus (a11-a14). Conclusion: ERK might play a fundamental role in determining the phenotype differences observed following EGF versus FGF3 stimuli presence of multiple components with different roles in each level of the cascade; interaction with scaffold proteins that direct each component to distinct upstream regulators and downstream targets; distinct sub-cellular locations of cascade components and their targets; distinct duration and strength of the signal; cross-talks among signalling cascades that are activated simultaneously [1]. To address this question, we have integrated published data on the MAPK network using using the CellDesigner software [2]. Our current annotated map includes 91 components (proteins/genes) and 178 interactions.

Luca GRIECO Poster 13

Macrophage and B cell specification

• We focused on the core network controlling B cell and macrophage diferentiation :
• common Progenitors (LMPP) maintenance necessitates PU1, CEBPb and E2a;
• B-cell genes (blue nodes) are controlled by transcription factors EBF, E2a, Pax5, FoxO1 and PU1;
• macrophage genes (red nodes) are controlled by CEBPa, CEBPb and PU1.

Binary/ternary components are denoted by ellipsoid/rectangular nodes. Red/green arcs denotes activatory/inhibitory regulations. Thick edges represent regulations with converging data from the litterature and ChIP-seq analysis during trans-differentiation; thin edges represent interactions supported by ChIP-seq data only; doted arcs represent interactions supported by data from the litterature only (potentially undirect).

Multiple perturbation and preB reprogramming

• Pax5 and EBF (but not E2a) forced expression (FE) decrease the efficiency of CEBPa induced trans-

differentiation (as shown in [4]).

Lineage differentiation and stability

• Transient stimulation by Il3/Il7 can induced differentiation of LMPP into macrophage/B cell.
• PU.1 KO impairs LMPP and macrophages stability and differentiation, but not B cell stability.
• CEBPa or CEBPb KO impairs macrophages differentiation.
• Each B factor KO impairs B cell differentiation and stability.

Computing stable states

• For the logical rules selected, we found five stable states :

→ 3 corresponding to wild type cell types. → 1 corresponding to PU.1 knock-own preB (found experimentaly in [2]). → 1 stable state with all components off.

Transdifferentiation of proB into macrophages

• Ectopic expression of the transcription factor Cebpa (or Cebpb) in proB can induced their trans-

differentiation into macrophages.

Logical modeling of hematopoietic cell specification

Samuel COLLOMBET 1, Cyrille LEPOIVRE 2,3, Denis PUTHIER 2, Thomas GRAF 3, Denis THIEFFRY 4 IBENS (CNRS UMR8197 / Inserm U1024), Paris, France TAGC (Inserm U1090), Marseille, France CRG, Barcelona, Spain

Introduction

Blood cells are derived from a common set of stem cells, which differentiate into more specific progenitors of erythroid, myeloid and lymphoid lineages, ultimately leading to functional cells such as erythrocytes, macrophages, B and T lymphocytes. This ontogenesis is controlled by a complex regulatory network involving environmental signals, as well as transcriptional and epigenetic factors. These factors regulate each other’s expression in a complex way, with some of them being expressed and required in different cell types [1]. The ectopic expression of some of these factors can induce the reprogramming of one cell type into another. For example, B cells can be reprogrammed into macrophages by forcing the ectopic expression of CEBPa [2]. Using public data from molecular genetic experiments (qPCR, western blot, EMSA) or genome-wide essays (DNA-chip, ChIP-seq), we have built a comprehensive map of the regulatory network of transcription factors and signaling components involved in hematopoietic development (to date encompassing 84 factors and 281 regulations). Based on this map and focusing on macrophage and B cell development, we have developed a dynamical model using the logical modelling software GINsim [3] and reproduced in sillico differentiation and reprogramming experiments.

Logical modelling

• Genes or proteins are associated with discrete variables representing their level of relative activity;

multivalued variable (rectangle on the network) were used only when necessary to minimize complexity.

• The level of a variable is defined by a logical function of the level of its regulators. For example :

PU1 → 1 if [(CEBPa | CEBPb | PU1) & !((CEBPa & CEBPb & PU1) | (PU1 & CEBPb : 2))] PU1 → 2 if [(CEBPa & CEBPb & PU1) | (PU1 & CEBPb : 2)] PU1 → 0 otherwise Where &, | and ! stand for AND, OR, NOT, and A → x mean that the value of variable A tends to x.

• Initially, all activators and none of the inhibitors are considered to be required for activation; these

logical functions are then refined to fit experimental evidences (effects of gain- or loss-of-functions on the expression of other factors).

• Cell types are defined as a vector of variable values, corresponding to qualitative levels of activity for

all factors, based on data from the literature.

PreB trans-differentiation into macrophage

• CEBPa ectopic expression (EE) in preB induces trans-differentiation into macrophages (left).
• Transient expression can be sufficient if long enough.
• Only constant CEBPb ectopic expression can induce preB trans-differentiation.
• When CEBPb expression stops, cells re-differentiate into preB.

Prospects

• Analysis of the functionality of regulatory circuits for differentiation and reprogramming.
• Predictive simulation (multiple perturbations, macrophages reprogramming).
• Refinement of the model (data on chromatin marks, cytokines pathways).
• Extension the network to cover Neutrophil and T-cell development.

References

[1] C. V. Laiosa et al. (2006) Annual Review of Immunology 24:705 [2] H. Xie et al. (2004) Cell 117:663 [3] C. Chaouiya et al. (2012) Methods in Molecular Biology 804:463 [4] L. H. Bussmann et al. (2009) Cell Stem Cell 5:554

Conclusion

• This study shows how high-throughtput data (transcriptome, ChIP-seq) can be used to complete a

predictive (qualitative) dynamical model.

• Our model currently reacpitulate published differentiation and reprogramming experiments.
• Logical rules prove to be sufficient to capture the most essential dynamical properties of this system.

PreB Mac

Samuel COLLOMBET Poster 14

Mesoderm specification map and regulatory network Expression of the main genes uderlying mesoderm specification Recapitulation of published data

Regulatory graph encompassing the main regulatory factors and interactions involved in mesoderm specification ( stages 8-10).

Ellipses denote Boolean nodes, whereas rectangles denote multilevel nodes. Light green nodes denote factors acting from the ectoderm. Green arrows denote activations, whereas red T-arrows denote inhibitions. This model has been defined and analysed using our software GINsim. To each regulatory node is associated a logical rule defining its behaviour depending on regulatory inputs.

Combinations of inputs corresponding to the initial states used to simulate the formation each tissue Simulations of known mutants

Dpp Visceral muscle Heart Somatic muscle Fat body Wg, Slp Hh, Eve, En

Dpp Wg Hh, En Somatic muscle Heart Visceral muscle Fat body Mesoderm

Ectodermal signals Mesodermal genes

Abibatou MBODJ1, Guillaume JUNION2, Duncan BERENGUIER1, Eileen FURLONG2 and Denis THIEFFRY1,3

Logical modelling of mesoderm specification in Drosophila melanogaster

Wg (Slp) lof

Hh + Dpp gof; Wg (Slp) lof

Tin lof

Wg + Dpp gof; Hh (Eve, En) lof

Prediction of the phenotypes of single and double perturbations

Slp (En, Eve, Hh) gof ; Doc gof Doc gof Slp (En, Hh, Eve) gof

338 mutants

Experimental validation of 12 mutants in progress

Map representing the different tissues formed during mesoderm specification ( stages 8-10 of development). Expression patterns for the 55 mesoderm specification network components Key markers genes used to define the identity of each kind of cell or tissue.

Colour code denotes the visual signature of each tissue (blue: visceral muscle (VM) ; red : heart (H) ; orange: somatic muscle (SM) ; green: fat body (FB) ).

Schematisation of wild-type mesoderm specification

Abibatou MBODJ Poster 52

Contributors & supports

★ ENS (Paris)

• Samuel Collombet
• Luca Grieco
• Anna Niarakis

★ CRG (Barcelona)

• Thomas Graf
• Chris van Oeven

★ TAGC (Marseille)

• Cyrille Lepoivre
• Denis Puthier
• Jacques van Helden
• Duncan Berenguier

★ IML (Marseille)

• Elisabeth Rémy

★ IGC (Lisboa)

• Claudine Chaouiya
• Jorge Carneiro

★ UniL (Lausanne)

• Aurélien Naldi

Belgian Inter-university Attraction Pole Bioinformatics and Modelling : from Genomes to Networks

Selected references

• Bérenguier D, Tichit L, Chaouiya C, Naldi A, Graf T, Thieffry D, Remy E (Submitted). Hierarchical transition

graphs, a new compressed, yet instructive, view for the dynamics of logical models.

• Calzone L, Tournier L, Fourquet S, Thieffry D, Zhivotovsky B, Barillot E, Zinovyev A (2010). Mathematical

Modelling of Cell-Fate Decision in Response to Death Receptor Engagement. PLoS Computational Biology 6: e1000702.

• Chaouiya C, Naldi A, Remy E, Thieffry D (2011). Petri net representation of multi-valued logical regulatory
• graphs. Natural Computing 10: 727-50.
• Coolen M, Thieffry D, Drivenes Ø, Becker TS, Bally-Cuif L. (2012). miR-9 controls the timing of

neurogenesis through the direct inhibition of antagonistic factors. Developmental Cell 22:1052-64.

• Fauré A, Naldi A, Lopez F, Chaouiya C, Ciliberto A. Thieffry D (2009). Modular logical modelling of the

budding Yeast cell cycle. Molecular Biosystems 5: 1787–96.

• González AG, Chaouiya C, Thieffry D (2008). Qualitative dynamical modelling of the formation of the

anterior-posterior compartment boundary in the Drosophila wing imaginal disc. Bioinformatics 24: i234-40.

• Naldi A, Thieffry D, Chaouiya C (2007). Decision diagrams for the representation and analysis of logical

models of genetic networks. Lecture Notes in Bioinformatics 4695: 233-47.

• Naldi A, Remy E, Thieffry D, Chaouiya C (2011). Dynamically consistent reduction of logical regulatory
• graphs. Theoretical Computer Science 412: 2207-18.
• Naldi A, Carneiro J, Chaouiya C, Thieffry D (2010). Diversity and plasticity of Th cell types predicted from

regulatory network modelling. PLoS Computational Biology 6: e1000912.

• Sánchez L, Chaouiya C, Thieffry D (2008). Segmenting the fly embryo: logical analysis of the role of the

Segment Polarity cross-regulatory module. International Journal of Developmental Biology 52: 1059-75.

• Sahin O, Fröhlich H, Löbke C, Korf U, Burmester S, Majety M, Mattern J, Schupp I, Chaouiya C, Thieffry

D, Poustka A, Wiemann S, Beissbarth T, Arlt D (2009). Modeling ERBB receptor-regulated G1/S transition to find targets for de novo trastuzumab resistance. BMC Systems Biology 3: 1.

• Thieffry D (2007). Dynamical roles of biological regulatory circuits. Briefings in Bioinformatics 8: 220-5.