Genome scale network modelling, from observations to predictions

Fabien Jourdan

Fabien.Jourdan@toulouse.inra.fr

INRA Toulouse

MetaboHub-Metatoul

From metabolome to interpretation/prediction

“This urine wheel was published in 1506 by Ullrich Pinder, in his book Epiphanie Medicorum. It describes the possible colours, smells and tastes of urine, and uses them to diagnose disease.” THE ROYAL LIBRARY, COPENHAGEN

Pearson H. Nature 2007; 446:8

Nicholson JK, Lindon JC. Nature 2008; 455:1054–1056

Interpreting metabolomic profiles

Sample preparation

Analytical chemistry

Metabolite identification

Statistics Interpretation in the metabolic network

Biochemical context : genome-scale metabolic networks

list of reactions

GLCt4 Nae + gluce Nac + glucc

HEX1 glucc + ATPc ADPc + g6pc

PGI g6pc f6pc

G6PDH2r g6pc + nadpc 6pglc + nadphc

GLCter glucc glucr

Biochemical reactions known to take place in a target organism & associated genes

Reconstruction based on genome annotation1

list of genes

(6524) or (6526)

(3098)

(2821) (2539)

mathematically structured knowledge base

1. Thiele I, Palsson BØ. A protocol for generating a high-quality genome-scale metabolic reconstruction. Nat. Protoc. 2010; 5:93–121 2. Thiele I, Swainston N, Fleming RMT, et al. A community-driven global reconstruction of human metabolism. Nat. Biotechnol. 2013; 31:419–25

Recon2 human metabolic network2 7440 reactions 2626 metabolites 1733 genes

Genome sequencing & annotation

Model reconstruction & refinement

Making sense of metabolomics profiles

Structural / topological analysis

→ understand the organisation of the system, study some structural

properties such as connectivity, possible pathways between several metabolites

Graph based analysis1

1-Frainay C. & Jourdan F. Computational methods to identify metabolic sub-networks based on metabolomic profiles. 2016. Briefings in Bioinformatics. 2-Kell DB. Metabolomics and systems biology: making sense of the soup. Curr. Opin. Microbiol. 2004; 7:296–307

« Making sense of the soup » 2 How can we connect metabolites?

MetExplore: server to perform this data interpretation

www.metexplore.fr

L Cottret, D Wildridge, F Vinson, M. P. Barrett, H Charles, MF Sagot et Fabien Jourdan. MetExplore: a web server to link metabolomic experiments and genome-scale metabolic networks. (2010) Nucleic Acids Research. 1;38 Suppl:W132-7

Functionalities : • Import/export networks • Collaborative curation of networks • Pathway enrichment • Import omics data • Chemical library mapping • Visualization • Flux computations • Graph algorithms

etExplore

Database

www.metexplor

e.fr

Visualisation Computation

User Curation

0 € to use it, 10 developers, 227 persons trained, 351 registered users, 937 networks, 15441 visitors

In silico global anlaysis of metabolism

Network modelling to predict metabolic behaviors

Dynamic / semi-quantitative analysis

→ understand the behavior of the system under specific conditions

Predict metabolic fluxes, cell growth, drug targets…

Extracting phenotypic sub-networks

aim = computing an active sub-network

flux ≠ 0 « active » reactions flux = 0 « inactive » reactions

1 flux distribution

1 sub-network of active reactions

1 phenotype

flux distribution 1

flux distribution 2

A global network model = an infinity of possible phenotypes

Deciphering metabolic shift during differentiation

comparison of metabolic network of HepaRG cells at 2 differentiation stages

3 days (non differentiated

progenitor cells)

30 days (differentatiated

hepatocyte-like cells)

Study of metabolic shifts during the differentiation process

Bipotent progenitor cells

differentiated cells

Hepatocyte-like cells

Biliary-like cells

Proliferation & differentiation

d. 3

d. 30

Poupin et al. In prep

Building a specific network model for HepaRG cells

generic human genome-scale metabolic network

reconstruction Recon21

(7440 reactions, 2140 genes)

sub-networks of predicted active reactions at d3

sub-networks of predicted active reactions at d30

Comparisons of differentially active reactions and pathways

Algorithm for identification of context-specific metabolic networks

(iMat2)

Transcriptomic data (≈ 6200 genes)

& Metabolomic data

(NMR 34 metabolites)

1 Thiele I., et al. Nat Biotechnol. 2013

2 Shlomi T., et al. Nat Biotechnol. 2008 Poupin et al. In prep

Evaluating set up of hepatic functions

FastCore

1st SOL

115th SOL

UNION

A B

155 defined metabolic functions were simulated GENERIC FUNCTIONS: 111 of these functions are expected to be achieved by any type cells HEPATHIC FUNCTIONS: 44 functions that are known to specifically take place in liver cells, such as ammonia detoxification through ureogenesis, ketogenesis and bile formation.

Poupin et al. In prep

+ highly expressed gene - not expressed gene

predicted active reaction predicted inactive reaction

Recon ID

CPS1 = carbamoyl-phosphate synthase CBPSam & r0034

OTC = ornithine carbamoyltransferase OCBTm

ASS = argininosuccinate synthase ARGSS

ASL = argininosuccinate lyase ARGSL

ARG1 = arginase ARGN

NOS = L-Arginine,NADPH:oxygen oxidoreductase (NO-forming) r0145

ornithine mitochondrial transport exchange with citruline ORNt4m

d03 d30

Network analysis highlight set up of urea cycle

Poupin et al. In prep

PREDICTING POTENTIAL DRUG TARGETS

Predict system behaviour: gene deletion analysis

Example: Targeting of synthetic lethal gene for the treatment of HLRCC patients (Frezza C. et al., Nature, 2011)

Fumarate hydratase (FH) is an enzyme of the TCA cycle that catalyses the hydration of fumarate into malate. Renal-cancer cells (HLRCC) are deficient for FH and display a truncated TCA cycle. As TCA is a major source for mitochondrial NADH, its truncature may have severe bioenergetic outcomes. But cancer cells are able to survive despite a non functional TCA, and no mechanism to explain this particularity had been provided.

in silico modelling approach to identify genes that are synthetic lethal with Fh1 (predict genes KO, that together with Fh1 mutation, would selectively affect the growth ability of cancer cells without affecting the wild-type cells).

GC/MS

Construction of Fh1-/- and Fh1fl/fl specific networks

Human genome scale network (Recon1)

Fh1-/- specific networks Fh1fl/fl specific networks

Metabolic genes that are highly expressed acrossmany cancer cell lines. Lee, J. K. et Proc. Natl Acad. Sci. USA 104, 13086–13091 (2007).

Fh1-/- transcriptomics data Fh1fl/fl transcriptomics data

Metabolic genes that are highly expressed acrossmany cancer cell lines. Lee, J. K. et Proc. Natl Acad. Sci. USA 104, 13086–13091 (2007).

Predicting synthetic lethal genes

• 24 reactions were predicted to be synthetic lethal with Fh1 • 18 belong to a linear pathway of haem metabolism • haem metabolism was also predicted by the model to have increased flux in Fh1-/- cells.

Fh1-/- specific networks

Gene 1 KO

Flux Balance Analysis

Produce enough biomass?

Gene 1 is not a synthetic lethal gene with Fh1

Gene 1 is a synthetic lethal gene with Fh1

YES NO

Predict system behaviour: gene deletion analysis

Example: Targeting of synthetic lethal gene for the treatment of HLRCC patients (Frezza C. et al., Nature, 2011)

The inhibition of the haem biosynthesis/degradation pathway, and in particular Hmox, is synthetically lethal with Fh1 valid therapeutic window for the treatment of HLRCC patients.

in wild-type cells in cancer cells

Predict system behavior: gene deletion analysis

Example: Targeting of synthetic lethal gene for the treatment of HLRCC patients (Frezza C. et al., Nature, 2011)

Experimental validation for the implication of the Haem metabolic pathway

1. the levels of excreted bilirubin were higher in Fh1-deficient cells.

2. in both wild-type and Fh1-deficient cells bilirubin excretion was completely blocked by ZnPP (zinc protoporphyrin), a Hmox inhibitor.

acute treatment with ZnPP had no profound effect on wild-type cells BUT it decreased the growth of cancer cells

Conclusions

• Genome scale metabolic networks provide a good context for metabolomics data analysis

• Extracing tissue/cell specific sub-networks is essential for the quality of predictions

• Global metabolic modeling of cancer cells allows to: – understand the metabolic characteristics of cancer cells

– identify selective drug targets

– predict specific biomarkers

• Flux predictions can be validated/improved with experimental fluxomics

French-speaking Metabolomics and Fluxomics network

Created in 2005, affiliated to metabolomics society since 2013

Currently ≈300 membres

Aims:

– to make an inventory and promote French skills in the fields of metabolomics and fluxomics

– to provide and support scientific meetings or workshops in metabolomics and fluxomics

– to facilitate knowledge transfer to students and newcomers in the field and help students to promote their work

President: Fabien Jourdan, Toulouse, France Vice president: Patrick Giraudeau, Nantes, France Secretary: Floriant Bellvert, Toulouse, France Treasurer: Anne-Emmanuelle Hay- de Bettignies, Lyon, France Communication: Frédérique Courant, Montpellier, France

Julien Boccard, Genève, Suisse Alain Bouchereau, Rennes, France Pascal de Tullio, Liège, Belgique Yann Guitton, Nantes, France Jean-Charles Martin, Marseille, France Etienne Thévenot, Gif-sur-Yvette, France

Yearly scientific conference

0

50

100

150

200

250

300

2005 2006 2008 2010 2011 2012 2013 2014 2015 2016

27 39 44

56

79 70 75

84 98

134

86

107 109 100

142

209

162

260

227

260

Laboratoires Participants

- Community is growing. -Important involvement of young scientists - Next year, joined conference with French Mass Spec. and Proteomics societies.

INRA Toxalim: MeX team (D. Zalko) Metatoul AXIOM

University of Glasgow: Michael P. Barrett David Wildridge Karl Burgess

Bioinformatics facility: Didier Laborie Christine Gaspin University of Sao Paulo: Ricardo Silva Ricardo Vencio

INRIA Bamboo: Marie-France Sagot Vincent Lacroix Vicente Acuña Paulo Milreu Hubert Charles

LISBP Jean-Charles Portais Stéphanie Heux Fabien Letisse MetaSys team RFMF and METBOHUB people

Fabien JOURDAN INRA-MTH

Yoann GLOAGUEN

Univ of Glasgow

Florence VINSON

INRA-MTH

MANY THANKS!

Sanu SHAMEER

INRA

Clément FRAINAY

INRA

Benjamin MERLET

INRA-MTH

CEA: Christophe Junot

Nathalie POUPIN

CR2 INRA

Maxime CHAZALVIEL

M2 UPS

Floréal CABANETTES

M2 UPS

Florence MAURIER INRA-MTH

Ludovic COTTRET IR INRA

3262

4178

29

78

44

62

Reactions predicted to be

ACTIVATED

Reactions predicted to be INACTIVATED

484

200

Comparison of specific HepaRG-cell reaction activity predictions between d3 & d30 Predicted modulated reactions

Pathway P-value % R° with

transc. data

Biotin metabolism° 1.31e-11 33%

Fatty acid oxidation*° 6.80e-07 42-37%

Bile acid synthesis 5.13e-05 54-61%

Tryptophan metabolism*° 2.44e-04 50-60%

Cytochrome metabolism 3.76e-04 79-71%

Blood group synthesis*° 1.59e-03 9-13%

Lysine metabolism*° 4.44e-03 43%

Limonene and pinene degradation*° 3.45e-02 100%

P-value Pathway

5.60e-31 *Transport, extracellular

1.30e-15 *Fatty acid synthesis

*do not emerge from transcriptomic data (rxns) °do not emerge from transcriptomic data (genes)

Poupin et al. In prep