NetBioSIG2013-Talk Thomas Kelder

Network signatures link hepatic effects of anti-diabetic interventions with systemic

disease parameters

Thomas KelderMicrobiology and Systems Biology, TNO, The Netherlands

Network Biology SIG, ISMB 2013, Berlin

2

4

Anti-Diabetic Treatment (ADT) study

DISEASE PARAMETERS• Plasma glucose, insulin• Body and organ weights• Atherosclerosis lesion area• Plasma cholesterol• Plasma & liver triglycerides

+

Dietary Lifestyle Intervention (DLI)

Fenofibrate, T0901317

Improves all disease parameters

Improves glycemiaDeteriorates dyslipidemia

Radonjic, et al., PLoS ONE, 2012

?

5

Intervention – hepatic mechanisms – disease parameters

TRIGLYCERIDES

ATHEROSCLEROSIS

GLUCOSEINTERVENTION

6

Which paths?

TRIGLYCERIDES

ATHEROSCLEROSIS

GLUCOSEINTERVENTION

7

Network analysis workflow

8

Link to disease parameters

9

WGCNA

• Weighted Gene Co-expression Analysis*• Identify co-expressed network modules• Correlate modules to disease parameters based on their “eigengene” (1st

Principal Component)

*Langfelder et al. BMC Bioinformatics, 2008

Disease parameter

Disease parameter

Disease parameter

?

?

?

10

Modules to disease parameters

• 14 coherent co-expression modules• 10 modules with GO annotation• 4 modules correlated with disease parameter(s)• All correlating endpoints related to dyslipidemia rather than dysglycemia

despite improvement of dysglycemia by all interventions

11

Link to intervention targets

Prior knowledge-based networks

• Curated pathways• Protein-protein interactions• Transcription factor targets

• Drug targets• DLI “targets”

– Ingenuity Upstream Regulator Analysis– Enrichment of known TF targets with DEGs for DLI

(p<0.001)– 25 transcription factors– Some overlap with drug targets (e.g. PPARA)

Total network has >12,000 (gene) nodes and >75,000 edges

13

Intervention specific networks

Filter by differential expression for intervention vs HFD

Network Total DEGs in dataset (p < 0.05) Connected nodes Edges

DLI 1,287 497 5,975

Fenofibrate 2,149 828 21,598

T0901317 2,924 1245 38,472

14

Network signatures

15

Random walks algorithm

[1] Dupont, et al. "Relevant subgraph extraction from random walks in a graph." Machine Learning (2006)[2] Faust, et al. "Pathway discovery in metabolic networks by subgraph extraction." Bioinformatics (2010)

Random w

alks

Intervention

Nodes and edges scored by probability of being visited by the random walker

Intervention

Diseaseparameter

Diseaseparameter

16

Network signaturesDLI signaturesTemplate for successful intervention

Drug signaturesCircumvent this response

17

Network signatures

• DLI vs drug, distinct response:• Small overlap• Opposite regulation

• Potential drug targets:• key nodes unique for DLI• cross-talk between processes

18

Performance assessmentImproved ability to prioritize genes by relevance to disease parameters

8.78 fold enrichmentp = 3.44E-10

3.09 fold enrichmentp = 0.023

Cholesterol

Atherosclerosis

Liver weight

Cholesterol

Atherosclerosis

Liver weight

1TFIndirect links

5 TFsDirect links

20

Conclusions

Network signatures underlying effects of interventions on dyslipidemia-related disease parameters

– Template for successful intervention or response to circumvent– Improves selection of genes relevant to disease parameters– Underlying interaction help interpretation

21

Acknowledgements

• Marijana Radonjic• Lars Verschuren• Alain van Gool• Ben van Ommen• Ivana Bobeldijk

Check out our poster at ISMB on SundayNetwork Biology of Systems Flexibility

R scripts and data for this analysis available at:https://github.com/thomaskelder/ADT-liver-network

igraph

22

23Study setup to investigate differential effects of anti-diabetic drug and dietary lifestyle interventions [1].

High fat diet “diseased” control group

Chow diet “healthy” control group

High fat diet DLI (switch to chow)

Fenofibrate

T0901317

0wk 9wk 16wk

LDLR-/-MICE

HEPA

TIC

TR

AN

SC

RIP

TO

ME

24

DLI Fenofibrate T0901317

Hepatic transcriptome dataset: - Chow control- Dietary lifestyle intervention (DLI)- Fenofibrate- T0901317Compared to high fat diet (HFD) at 16wk.

Co-expression network modules identified by Weighted Gene Co-expression Network Analysis (WGCNA) [2]. Provides high-level overview of relevant processes.

WGCNA

25

DLI Fenofibrate T0901317

WGCNA

DLI

T0901317

FENOFIBRATE

EXTEND WITH PRIOR KNOWLEDGE

FILTER FOR REGULATED GENES

26

Out of ten modules that could be annotated to a biological process, three modules correlated significantly with disease parameters. All significant correlations were with dyslipidemia related disease parameters, despite the evident improvement of glycemic status by the interventions.

MODULE NO. GENES GO TERMS SIGNIFICANT CORRELATIONS

YELLOW 198Lipid biosynthetic process,Oxidoreductase activity

Liver weight (-0.91), Triglycerides (-0.90), Atherosclerosis (-0.79), Cholesterol (-0.79)

RED 161Cell activation, Immune system process, Inflammatory response

Atherosclerosis (0.80), Cholesterol (0.78), Liver weight (0.75)

BLACK 142Lipid metabolic process,Oxidation-reduction process

Liver weight (0.88); Cholesterol (0.83)

WGCNA• Weighted co-expression network analysis*• Correlate modules to other measurements (clinical, plasma proteins,

microbiome)

*Langfelder et al. BMC Bioinformatics, 2008

glucose

Chow

HF 16

weeks

Lifest

yle

Rosiglit

azone

T0901

317

0

5

10

15

20

** ***

glu

cose

(m

M)

Omics, genetics, physiological data, prior knowledge

Molecular signatures of metabolic health and disease

Mechanistic insight: Biological context ofmolecular signatures

Prognostic / diagnostics molecular signatures

Coexpression networks (WGCNA)Prior-knowledge networksCausality networksVariable selection methodsSubgraph ID/ (K-walks)topology/ network clustering

Network signatures for improved diagnostics & interventions

Link to pathological endpoint

Subgroup-specific molecular signatures prioritization and refinement