Integrative network analysis with Cytoscape reveals time ... · ACE inhibitors 238 (97%) 193 (97%) 45 (98%) Statins 231 (94%) 189 (95%) 42 (91%) One-year echocardiography follow-up

Integrative network analysis with Cytoscape reveals time-dependent molecular events underlying left ventricular remodeling in

post-myocardial infarction patients Florence Pinet, Inserm UMR1167, Univ Lille,

Institut Pasteur de Lille

[email protected]

Predictive factor of mortality and heart failure post-MI

Frequent process • REVE-1 study (2002-2004) 31% of patients • REVE-2 study (2006-2008) 31% of patients

Left ventricular remodeling and heart failure

Physiopathology Early phase Late phase Myocardial Infarction (MI)

Progressive dilatation of left ventricle (LV)

• Heart failure • Death

Epidemiology

0

2

4

6

No cardiovascular events after 3 years

Death, Heart failure, MI after 3 years

P<0.001

0

2

4

6

LV dilatation 1 yr post-MI

cm2

St John Sutton et al. Circulation 1994; Savoye et al. Am J Cardiol. 2006; Fertin et al, Am J Cardiol. 2010; Bauters et al, PLOS One 2017

Predictive factors

MI size

MI area

duration of ischemia

Permeability of artery

Transmurality

Microvascular obstruction

Difficult prediction in clinical practice

Search of BIOMARKERS

LV remodeling is an important mechanism of ischemic HF

Interesting surrogate endpoint for post-MI HF

Left cardiac remodeling: surrogate marker of heart failure

REVE2 study

Fertin et al, Am J Cardiol 2012

• Heart failure • death

Early phase Late phase Myocardial Infarction (MI)

Progressive dilatation of left ventricle

1 month 3 months 12 months inclusion MI

Plasma Serum

Dunkerque Calais

St Omer

Béthune Douai Valenciennes

Cambrai

Tourcoing Roubaix Boulogne

Arras

Lille

Caen

Lens

246 patients included

Characteristics All men Women

n 246 200 46

Age, yr (mean ± SD) 57 ± 14 56 ± 13 62 ± 15

Diabetes mellitus 51 (21%) 46 (23%) 5 (11%) Initial reperfusion therapy Primary percutaneous coronary intervention

128 (52%) 106 (43%) 22 (48%)

Thombolysis alone 28 (11%) 23 (9%) 5 (11%) Thombolysis and rescue percutaneous intervention

59 (24%) 50 (20%) 9 (20%)

No reperfusion 31 (13%) 21 (11%) 10 (22%) Peak creatitine kinase, IU (mean ± SD)

3018 ± 2376 3099 ± 2483 2666 ± 1824

HF (Killip class ≥ 2) during hospitalization

79 (32%) 65 (33%) 14 (30%)

LVEF, % (mean ± SD) 49 ± 8 49 ± 8 51 ± 9

Medications at discharge Antiplatelet therapy 246 (100%) 200 (100%) 46 (100%) β-blokers 238 (97%) 194 (97%) 44 (96%) ACE inhibitors 238 (97%) 193 (97%) 45 (98%) Statins 231 (94%) 189 (95%) 42 (91%) One-year echocardiography follow-up

Number of patients with follow-up 223 (92%) 182 (91%) 44 (96%) LV remodeling*, n 87 (38.5%#) 67 (36.8%#) 20 (45.5%#)

Characteristics of the patients included in the REVE-2 study

Dubois-Deruy et al, Sci Rep 2017

Fertin et al, Am J Cardiol 2010; Lamblin et al, Eur J Heart Fail 2011; Fertin et al, J Cardiol 2012; Fertin et al, Am J Cardiol 2012; Bauters et al, Int J Cardiol 2013; Fertin et al, PLOS One 2013; Eschallier et al, Circ Heart Fail 2013; Kumarswanny et al, Circ Res 2014; Bauters et al, Cardiology 2016; Dubois-Deruy et al, Sci Rep 2017; Bauters et al, PLOS One 2017; Ferreira et al, submitted

IDM BNP CRP Troponine I MMP8 MMP9 MMPs/TIMPs HGF miR133a miR423-5p sFASL Ratio PIIINP/ICTP

BNP CRP Troponine I MMP8 MMP9 MMPs/TIMPs HGF miR133a miR423-5p

BNP CRP Troponine I MMP8 MMP9 MMPs/TIMPsHGF miR133a miR423-5p

T0 (D5) M3 M1 M12

BNP CRP Troponine I MMP8 MMPs/TIMPs.. HGF miR133a miR423-5p miR21-5p miR23a-3p miR222-3p lncRNA, LIPCAR

BNP

MMP9

Biomarkers associated to LV remodeling in REVE-2 study

Confirmation of BNP and MMP-9 as predictive biomarkers of LV remodeling

REVE-2 is the first study showing that:

a lncRNA, LIPCAR is

A predictive factor of LV remodeling

A predictive factor of cardiovascular events post-MI

A predictive factor of early mortality in HF patients

Interest of a multimarker

Interest of serial blood samples

ST2, galectine-3 miR21-5p miR23a-3p miR222-3p lncRNA, LIPCAR

miR21-5p miR23a-3p miR222-3p lncRNA, LIPCAR

miR21-5p miR23a-3p miR222-3p lncRNA, LIPCAR

• Human pathologies: involvement of several signalling pathways interacting in molecular networks that are interconnected

Interest of system biology for analyzing biomarkers associated to LV remodelling

Dimitrakopoulou et al, BMC Genomics 2015

Layout of 40 consensus modules

Integromics network meta-analysis on cardiac aging

• REVE-2 molecular network

Explore processes associated with LV remodeling at different timepoints after MI

Identification of new biomarkers

Building the REVE-2 molecular network

• Building a molecular network • from 63 clinical variables • from 24 molecular data collected:

•18 proteins, • 5 miRNAs, • 1 long non coding RNA

• measured at 1 to 4 time points

Resource Version URL Type N= nodes,E=edges

ENCODE 2012-09-06 http://encodenets.gersteinlab.org E

EnsemblGenes Release 79 http://www.ensembl.org N

HMDB v3.6 http://www.hmdb.ca N, E

Microcosm v5 http://ebi.ac.uk/enright-srv/microcosm E

miRBase v21 http://mirbase.org N

miRecords v4 http://c1.accurascience.com/miRecords E

miRTarBase v4.5 http://mirtarbase.mbc.nctu.edu.tw E

Reactome v52 http://www.reactome.org E

STRING v9.1 http://string-db.org E

TargetScan v6.2 http://www.targetscan.org E

TFe 2015-04-12 http://www.cisreg.ca/cgi-bin/tfe/home.pI E

WikiPathways 2015-04-12 http://www.wikipathways.org E

Integration of REVE-2 molecular data in 12 public knowledge databases

Building the REVE-2 molecular network

Construction du réseau REVE-2 (en collaboration avec EdgeLeap)

• Molecules included in the network: – all molecule nodes mapping to one or more REVE-2 variables – all molecule nodes that are their direct neighbours – all molecule nodes that are part of the shortest paths until 3 edges/interactions with REVE-2 variables

• The REVE-2 network : 1,310 molecule nodes : 1263 proteins

24 miRNAs 22 metabolites 1 lncRNA 8,639 edges/interactions

• all the molecules in the network are annoted for 33 clusters (Gene Ontology) • 20 REVE-2 variables are part of a cluster • 12 clusters contain at least one REVE-2 variable

Building the REVE-2 molecular network

Pinet et al, BBA 2017

cluster 1

cluster 14

cluster 2

cluster 3

cluster 30

cluster 4

cluster 40

cluster 41

cluster 5

cluster 6

cluster 7

cluster 9

Visualization of the REVE-2 molecular network

Nodes are colored by cluster membership based on the 12 clusters that contain REVE-2 variables

Pinet et al, BBA 2017

Cluster

number

Node

s Edges REVE-2 variables Most significant GO category

1 191 1013 ST2 sequence-specific DNA binding

2 203 302 Mir-21-5p intrinsic component of membrane

3 57 978 ICTP, P1NP, P3NP endoplasmic reticulum lumen

4 65 403 HGF transmembrane receptor protein tyrosine kinase signaling pathway

5 98 123 miR-222-3p RNA binding

6 37 495 TIMP1 platelet alpha granule lumen

7 35 392 troponin muscle filament sliding

8 30 271 nuclear-transcribed mRNA catabolic process

9 47 154 MMP1, MMP2, MMP3, MMP8,

MMP9, TIMP2, TIMP4 extracellular matrix disassembly

10 38 75 ATP binding

12 30 75 mitotic cell cycle phase transition

13 32 33 O-glycan processing

14 22 111 FasL activation of cysteine-type endopeptidase activity involved in apoptotic process

15 17 67 G2/M transition of mitotic cell cycle

16 23 33 G-protein coupled receptor signaling pathway

18 14 61 RNA splicing, via transesterification reactions

19 16 29 positive regulation of intrinsic apoptotic signaling pathway

20 22 23 phosphatase activity

21 14 32 toll-like receptor 2 signaling pathway

22 15 17 steroid hormone receptor activity

24 11 19 regulation of cell differentiation

25 10 18 cell surface

26 11 16 cellular component disassembly involved in execution phase of apoptosis

28 9 14 negative regulation of transforming growth factor beta receptor signaling pathway

29 8 15 small GTPase mediated signal transduction

30 11 12 BNP protein targeting to mitochondrion

32 9 12 DNA repair

33 9 10 nuclear pore

35 7 11 Mitochondrial inner membrane

37 7 8 DNA-directed RNA polymerase II, holoenzyme

39 6 11 posttranscriptional gene silencing

40 7 12 CRP complement activation

41 6 11 Creatine kinase creatine metabolic process

Characteristics of the REVE-2 molecular network

Dynamic regulatory network for adverse left ventricular remodeling following MI

To assess which mechanisms change under LV remodeling and are relevant at each time point after MI, an active module analysis was performed. Active modules were extracted from the REVE-2 network model for each time point

Characteristics of the REVE-2 molecular network

Pinet et al, BBA 2017

Heatmap of cluster coverage of active modules in the REVE-2 network

- Strong decrease of the number of clusters (n=9) in active modules at 1 month - A large number of clusters (n=30) are in active modules at 3 month - Only 3 clusters are in active modules at 1 year Pinet et al, BBA 2017

• The active modules have been extracted in the network by including the REVE-2 variables significantly modulated between patients with high and low LV remodeling, their direct neighbours and nodes with the shortest paths connecting 2 REVE-2variables

• Major changes occur at baseline and 3 months

• Baseline = acute stress post-MI Inflammation (Cluster 40 : 7/7 nodes) Oxydative stress (Cluster 30 : 10/11 nodes, cluster 35 : 3/7 nodes) Lesion of myocytes (Cluster 41 : 6/6 noeuds) Degradation of extracellular matrix (Cluster 9 : 17/47 nodes)

• 3 months = development of LV remodeling myofilament (Cluster 7 : 33/35 nodes) Degradation of extracellular matrix (Cluster 9 : 28/47 nodes)

Baseline 1 month

3 months 1 year

Analysis of active modules in the REVE-2 network

Insights into time-resolved molecular changes associated with LV remodeling progression Pinet et al, BBA 2017

REVE-2 network

Analysis of betweenness centrality

The molecules with the highest centrality in the REVE-2 network are REVE-2 variables SOD2 3 miRNAs

miR-222-3p

miR-21-5p

miR-23a-3p

Centrality : 0.34 miR-222-3p

0.16 miR-21-5p

0.06 miR-423-5p

• The REVE-2 network validate our experimental data • This strategy is helpful for the selection of new biomarkers of left ventricular remodeling post-MI

Betweenness centrality of a molecule : number of shortest paths connecting 2 molecules in the network interacting with this molecule

Proteomic analysis

Strategy for selection of new biomarkers from the REVE-2 network

Dubois-Deruy et al, Sci Rep 2017

Analyse de la centralité des molécules du réseau REVE-2

Molecules Type Centrality Time-point Cluster Best cluster GO Secretion

5 EP300 Transcription factor

0.06 baseline, 3M cluster 1 sequence-specific DNA binding

Not described

6 miR-335-5p miRNA 0.0522 1M, 3M cluster 13 O-glycan processing Yes

7 miR-26b-5p miRNA 0.0448 baseline, 1M, 3M cluster 2 intrinsic component of membrane

Yes

9 CTCF Transcription factor

0.0309 baseline, 1M, 3M cluster 1 sequence-specific DNA binding

Not described

24 ESR1 Transcription factor

0.0096 baseline, 3M cluster 22 steroid hormone receptor activity

Yes

36 miR-375 miRNA 0.0084 baseline, 1M, 3M, 1Y cluster 2 intrinsic component of membrane

Yes

43 miR-17-5p miRNA 0.007 3M cluster 5 RNA binding Yes

• In the REVE-2 network: the betweenness centrality of the 1310 molecules is between 0 and 0.34 • Threehold : > 0.007, corresponding to the first 50 molecules

The molecules only active at baseline were not analysed

• Selection of 3 transcription factors and 4 miRNAs, that could be involved in the pathophysiological processes or could be new biomarkers of LV remodeling

EP300 and ESR1

• Involved in regulation of expression and activity of SOD2 • Can acetylate and be deacylated by Sirtuin 1 (Kuno et al., 2013 ; Kuno et al., 2015)

Baseline 1 month 3 months

• EP300 : Histone acetyl-transferase

• ESR1 : Estrogen receptor α

• Involved in the transcription and activity of SOD2 (Puzianowska-Kuznicka, 2012 ; Luo et al., 2016)

Baseline 3 months

miR-26b-5p and miR-17-5p

Baseline 1 month 3 months

3 months

• miR-26b-5p

• miR-17-5p

• Decreased in the cardiac hypertrophy (Han et al., 2012)

• Significantly decreased in the plasma of patients with acute HF (Ovchinnikova et al., 2016)

• Involved in the apoptosis of cardiomyocytes (Du et al., 2014)

• Increased in patients with cardiac fibrosis following hypertrophic cardiomyopathy (Fang et al., 2015)

miR-335-5p and miR-375

• miR-335-5p

• miR-375

• Involved in angiogenesis and apoptosis

1 month 3 months

Baseline 1 month 3 months 1 year

• Not yet described in the heart

Usefulness of betweenness centrality

• Analysis of centrality of molecules in the REVE-2 network

Selection of 3 transcription factors (EP300, CTCF et ESR1)

Selection of 4 miRNAs (miR-335-5p, 26b-5p, 375 et 17-5p)

Expression of the 4 miRNAS in LV of an experimental model of LVR post-MI Quantification of circulating plasma levels of the 4 miRNAS in the same model Quantification in plasma of REVE-2 patients

• Hypothesis for the regulation of SOD2 Implication of EP300 in its inactivation by acetylation Implication of ESR1 in SOD 2 expression

17

Table 4. Unadjusted and adjusted hazard ratios (HR) for cardiovascular death or hospitalization for heart failure at long-term follow-up according to left ventricular remodeling (LVR) at 1 year after MI .

LVR ≥20%

LVR as a continuous variable

(per 10% increase)

HR 95% CI p HR 95% CI p

Unadjusted

- Cohort 1 2.52 1.45 – 4.36 0.001 1.16 1.06 – 1.27 0.001

- Cohort 2 2.52 1.23 – 5.17 0.012 1.15 1.03 – 1.29 0.012

Adjusted model

- Cohort 1 2.20 1.20 – 4.04 0.011 1.16 1.04 – 1.29 0.008

- Cohort 2 2.57 1.18 – 5.56 0.017 1.15 1.00 – 1.31 0.047

LVR, percent increase in end-diastolic volume from baseline to 1 year.

Adjusted model: age, sex, diabetes mellitus, baseline left ventricular ejection fraction, baseline end-

diastolic volume were entered in the Cox regression.

Follow-up of REVE-2 patients for cardiovascular death

Bauters et al, PLOS One 2017

Follow-up of REVE-2 patients for cardiovascular death

REVE-2 REVE-1

• Incorporate this new clinical endpoint in the REVE-2 network • Characterize the impact of the follow-up in the REVE-2 network • Determine the active modules

Bauters et al, PLOS One 2017

Conclusion

• Building of a molecular network of LV remodeling post-MI from experimental data

The REVE-2 network embeds 1310 molecules/nodes linked by 8639 edges • Analysis of active modules for LV remodeling

Major changes at baseline and 3 months • Analysis of betweenness centrality of molecules in the REVE-2 network :

Selection of 3 transcription factors (EP300, CTCF et ESR1) Selection of 4 miRNAs (miR-335-5p, 26b-5p, 375 et 17-5p)

• Integration of 10 years follow-up of REVE-2 patients

Acknowledgments

Cardiologic hospital C. Bauters N. Lamblin P. De Groote M. Fertin G Lemesle

Inserm U1167 P. Amouyel E. Dubois-Deruy A Turkieh M. Bouvet O. Beseme M. Chwastyniak

M. Radonjic T. Kelder

Utrecht, The Netherland