Dr. Edwin K. Silverman

1. Introduction to Network MedicineIntroduction to Network Medicine
Edwin K. Silverman, M.D., Ph.D.Edwin K. Silverman, M.D., Ph.D.
Channing Division of Network Medicine
Pulmonary and Critical Care Medicine
Brigham and Women’s Hospital
Boston, Massachusetts
BRIGHAMAND
WOMEN’S HOSPITAL
HARVARD
MEDICAL SCHOOL

2. Edwin K. Silverman: Conflicts of Interest
• 1) Personal financial relationships with commercial interests
relevant to medicine, within past 3 years:
• Consulting: GlaxoSmithKline, Merck
• Lecture Fees (Honoraria):
– GlaxoSmithKline
– Merck
– 2) Personal financial support from a non-commercial source
relevant to medicine, within past 3 years:
No relationships to disclose
– 3) Personal relationships with tobacco industry entities:
– No relationships to disclose

3. What Is a Network?
A collection of points (nodes) that are joined in pairs by lines
(edges). A graphical approach to visualizing and analyzing
relationships between variables of interest.
(Adapted from M. Newman, Networks: An Introduction, 2010)
Biological Network Social Network Ecological Network

4. What Is Network Medicine?
The study of cellular, disease, and social networks which
aims to quantify the complex interlinked factors
contributing to individual diseases.
(Adapted from Barabasi, NEJM 2007; 357:404)
Key components of Network Medicine:
--Holistic rather than reductionist approach
--Construction of molecular disease networks
--Non-linear responses of complex systems
--Emergent properties from entire network
--Investigates responses of networks to various types of
perturbation
--Employs systems biology methods

5. Systems Biology and
Systems Pathobiology
•Systems Biology: “The science of integrating
genetic, genomic, biochemical, cellular,
physiological, and clinical data to create a network
that can be used to model predictively a biological
event” (Loscalzo, Circulation 2012; 125: 638)
•Systems Pathobiology: Application of systems
biology approaches to disease

6. Approaches to Complex Diseases in
Channing Division of Network Medicine
Building
Disease
Networks
Defining
Molecular
Pathways
Identifying
Optimal
Disease
Phenotypes
Developing
and Validating
New Disease
Classifications
Developing New
Treatments and
Preventions
Integrating
Multiple
–omics Data
Types

7. What Types of Research Could Be
Considered as “Network Medicine”?
• Integrating SNPs, Gene Expression, and Epigenetics
• Moving from Gene Discovery to Gene Localization to
Functional Validation
• Using Genetic Information to Dissect Disease
Heterogeneity and Reclassify Disease
• Using Network Approaches to Understand Disease

8. Possible Functional Impacts of Genetic Variants on
the Cellular Molecular Interactome Network
“Normal”
Loss/Gain of
interaction
Alter
interaction
strength
None
Loss of many interactions

9. Overview of Complex Disease Genetics
Demonstrate Familial Aggregation
Localize Susceptibility Gene(s)
Identify Disease Susceptibility Gene and Functional Variants
Positional
Cloning
Candidate
Genes
Diagnostics
Determine Fraction of
Explained Variation
Treatment
Genome-Wide
Association
Rare
Variants

10. Statistical Analysis with Genetic
Association Studies
• Linkage Disequilibrium: Statistical association of alleles at two loci
• Association Analysis: Determines if a genetic variant occurs more
frequently in subjects with a phenotype
• Genome-Wide Association Analysis: Uses a panel of polymorphic
markers capturing most of the common linkage disequilibrium
information in the study population

11. Finding the Missing Heritability
of Complex Diseases
(Manolio et al, Nature 2009; 461: 747)
• Potential Explanations for Missing Heritability in GWAS:
– Study Design Problems in GWAS
• Imprecise phenotyping
• Poor control groups
– Much larger numbers of common variants of small effect exist
– Low frequency (0.5% to 5%) and Rare (< 0.5%) variants of larger
effect are important
– Impact of structural variants
– Causal variants have not been found
– Inaccurate heritability estimates
– Contribution of Gene-by-Gene interactions

12. Chronic Obstructive Pulmonary
Disease: Definition
Chronic obstructive pulmonary disease (COPD) is a disease
state characterized by airflow limitation that is not fully
reversible. The airflow limitation is usually both progressive
and associated with an abnormal inflammatory response of
the lungs to noxious particles or gases (GOLD Project).
• COPD includes:
– Emphysema
– Chronic Bronchitis
– Small Airway Disease
Note: COPD is the third leading cause of death in the US

13. COPD: Evidence for Genetic Determinants
• Development of COPD in smokers is highly variable (Burrows 1977).
• Pulmonary function in the general population and COPD cluster in families (Lewitter
1984, Redline 1987, Larson 1970, Silverman 1998).
• Twin study of 22,422 Danish and 27,668 Swedish twin pairs estimated COPD
heritability ~60% (Ingebrigtsen 2010).
• Quantitative Genetic Analysis: COPDGene analysis confirmed significant heritability
for FEV1 (approx. 40%) and emphysema (approx. 30%) (Zhou 2013)
• A small percentage of COPD patients inherit severe alpha-1 antitrypsin deficiency.

14. Potential Impact of Genetics on COPD
Diagnosis and Treatment
• Learning about New Biological Pathways in Disease Pathogenesis:
• Nature’s pertubations of human biological networks
• Identifying targets for new drug development
• Reclassifying Complex Diseases:
• Based on etiology and disease pathophysiology
• Pharmacogenetics:
• Finding patients likely to have excellent treatment response
• Avoiding treatment of individuals at high risk for adverse events

15. GWAS Meta-analysis of Moderate-Severe COPD
(Cho, Lancet Resp Med 2014)
Locus Nearest gene SNP
Meta-analysis
OR (CI) P
4q22 FAM13A rs4416442
1.28
(1.2-1.36)
1.12x10-14
15q25 CHRNA3 rs12914385
1.28
(1.2-1.36)
6.38x10-14
4q31 HHIP rs13141641
1.27
(1.19-1.36)
1.57x10-12
14q32 RIN3 rs754388
1.28
(1.18-1.39)
5.25x10-9
Note: Meta-analysis of COPDGene, GenKOLS, ECLIPSE, and NETT-NAS

16. Moderate-to-Severe COPD and Severe
COPD GWAS (Cho, Lancet Resp 2014)
TGFB2
HHIP
FAM13A
IREB2/CHRNA3/5
RIN3
MMP12
GOLD
2-4
GOLD
3-4

17. Replication of COPD GWAS Regions
Locus First Author (Year) SNP Odds Ratio P-value
HHIP Van Durme (2010) rs13118928 0.80 2 x 10-4
HHIP Zhou (2012) rs13118928 0.68 2 x 10-3
FAM13A Young (2011) rs7671167 0.79 0.01
FAM13A Guo (2011) rs2869967 N/A 0.04
CHRNA3/5 Wilk (2012) rs1051730 1.17 3 x 10-7
CHRNA3/5 Hardin (2012) rs8034191 1.89 7 x 10-7
IREB2 Chappell (2011) rs2568494 1.30 5 x 10-4
MMP12 Hunninghake (2012) rs2276109 N/A 4 x 10-5
RIN3 Hopkins (ATS 2014) rs754388 1.51
(Severe COPD)
0.006

18. • Discovery
– Genetic Association Analysis
• Localization
– Fine Mapping (e.g. More SNPs, Sequencing, More populations)
– Long-range Genetic Interactions
– Regions containing functional activity (e.g. Reporter assays)
• Functional Validation
– In vitro cell-based models
– In vivo murine models (i.e. Knock-out, Transgenic, Knock-in)
Moving from Gene Discovery to Gene
Localization to Functional Validation

19. COPD GWAS Locus Near HHIP
(Wilk 2009)

20. Interactionfrequency
a
3C fragments:
Chr4: 145Mb
7kb SNP region Anchor
*p<0.01
Long-range Interaction Was Detected Between COPD GWAS
Region and HHIP Promoter (Zhou, Hum Mol Genet, 2012)
Chromosome conformation capture

21. **p<0.01
Luciferaseactivity
b
Luciferaseactivity
a **p<0.01
COPD Risk Haplotype Is Associated with Lower HHIP
Promoter Activity (Zhou, Hum Mol Genet 2012; 21: 1325)
Reporter assays
Beas-2B cells

22. Using RNA Interference to Assess HHIP
Biological Effects
(X. Zhou, Genomics 2013)
• Hypothesis: Silencing HHIP would primarily affect
other hedgehog pathway genes
• Assess RNAi effects sequentially:
– Microarray expression analysis in Beas2B cells
– RT-PCR validation in Beas2B cells
– RT-PCR assessment in primary lung epithelial cells

23. Microarray Gene Expression Analysis of
Beas2B Cells Infected with shRNA for HHIP
Decreased
genes
Increased genes
NT
controls
NT
controls
HHIP
shRNAs
HHIP
shRNAs
A B
4 296 146
SV
A
iSVA
Bio3-tech1
Bio3-tech2
Bio3-tech3
Bio1
Bio2
Bio4-tech1
Bio4-tech2
shRNA3-bio1
shRNA1-bio1
shRNA2-bio2
shRNA4-bio4
shRNA2-bio1
shRNA2-bio3-tech1
shRNA2-bio3-tech2
Note: No other Hedgehog pathway genes showed significant
differences in microarray gene expression analysis after HHIP
silencing.

24. Network of Differentially Expressed Genes with HHIP Silencing
(X. Zhou/J. Fah Sathirapongsasuti, Genomics 2013)
Edges:
Brown: Pathway
Commons
Purple: Functional
Interaction
Blue/Teal:
Published
abstract
Red: Manual
literature
curation

25. +/+ AIR(N=12) +/+ CS(N=18) +/- AIR(N=14) +/- CS(N=19)
0
10
20
25
30
35
40
AveolarChordLength
+/+ AIR(N=12)
+/+ CS(N=18)
+/- AIR(N=14)
+/- CS(N=19)
C57BL/6 ---Gill staining Mean Cord Length
Hhip+/-
Mice: Cigarette Smoke Effects
(X. Zhou/T. Lao/C. Owen)
Hhip+/-
Hhip+/+
Air Smoke
**
***
*

26. Evolution of Complex Disease Genetic Studies
(Silverman/Loscalzo, Discovery Med 2012; 14: 143)
Genetic
Variants
Single
-Omics Data
Type
Disease
First Generation Genetic Studies
Genetic
Variants Disease
Second Generation Genetic
Studies
??
Low Power To
Detect Associations
TranscriptomicsGenetic
Variants
Disease
Subtypes
Third Generation Genetic Studies
Metabolomics
Proteomics

27. CT Scans of Two Boston EOCOPD Study Probands
Age 47, FEV1 20%Age 42, FEV1 38%

28. COPD Is Heterogeneous
• COPD Subtypes (Distinct Groups of COPD Subjects)
– Emphysema
– Airway Disease
• COPD-Related Phenotypes (Disease
Manifestations)
– COPD Exacerbations
– Cachexia
– Functional Impairment
– Co-morbid Diseases

29. COPDGene Cluster Analysis
(Castaldi/Cho, Thorax 2014)
• 10,192 enrolled smokers (GOLD 0-4 and
GOLD-U)
• 8,128 with complete data for all potential
clustering variables and outcomes
• split into training and test sample
– training = 4187
– test = 4101
• Clustering based on FEV1, Emphysema,
Emphysema Distribution, and Airway Wall Area

30. K-means, Four Clusters: Training
Set (Castaldi/Cho, Thorax 2014)
C1:Mean C2:Mean C3:Mean C4:Mean
N (#) 1597 623 1123 844
Age 58.9 58 56.8 65.4
Race (% African-American) 30 46 37 19
Gender (%,female) 44 53 52 40
FEV1, percent of predicted 95.3 81.9 74.9 41.2
FEV1/FVC 0.76 0.7 0.71 0.42
BMI 28.7 27.9 31.4 26.7
Pack Years 38 45.8 42.8 56.8
Emphysema at -950HU 2.6 3.3 1.3 20.5
Segmental Airway Thickness 58.8 61.5 64.1 62.7
Upper/Lower Emphysema Ratio 0.69 6.72 0.56 2.19
Upper/Lower Emphysema Difference -1 4.3 -1 7.7
Gas Trapping 12.9 16.5 13.4 52.1
Training Sample
Cluster 1 = Resistant Smokers
Cluster 2 = Mild Upper Lobe Predominant Emphysema
Cluster 3 = Airway Predominant Disease (High BMI, Less Emphysema)
Cluster 4 = Severe Emphysema

31. How Meaningful/Useful Are the Four
Clusters? (Castaldi/Cho, Thorax 2014)
Training Set Only
C2:OR C2:pval C3:OR C3:pval C4:OR C4:pval
Exacerbations 2.27 5.80E-09 3.16 4.60E-23 8.93 1.00E-82
MMRC 2.81 1.30E-30 3.38 1.40E-57 10.87 2.80E-177
BODE 3.36 7.20E-38 4.62 9.00E-80 66.44 <0.00001
Hospitalizations/ER Visits 4.06 1.40E-12 5.04 1.50E-20 11.81 2.30E-48
rs7671167 (FAM13A) 0.94 5.00E-01 0.85 2.40E-02 0.83 9.10E-03
rs1980057 (HHIP) 0.64 3.10E-06 0.91 1.90E-01 0.78 3.90E-04
rs13180 (Chr15q) 0.81 2.30E-02 1.04 5.90E-01 0.82 5.70E-03
rs8034191 (Chr15q) 1.32 2.70E-03 1.02 8.30E-01 1.43 8.90E-07
rs7937 (Chr 19q) 1.28 6.30E-03 1.15 4.30E-02 1.18 1.90E-02

32. Building Phenotypic Networks in COPD
(Jen-Hwa Chu, Submitted)
• Methods:
– Gaussian Graphical Model
– Network based on partial correlations between quantitative phenotypes
– Nodes: Disease-related phenotypes
– Edges: Undirected; placed if “significant” partial correlation coefficient
(p<0.001)
– Compare differences in network connectivity based on permutation testing
• Study Population: 8141 COPDGene Subjects
• Phenotypes:
– Lung Function: FEV1
– Imaging: Emphysema severity and distribution, Gas Trapping, Airway Wall
Area (Pi10)
– Demographics: Age, pack-years of smoking
– Exercise Capacity: Six minute walk distance (6MWD)
– Other Clinical Characteristics: Exacerbations, BMI

33. Building Phenotypic Networks
(Jen-Hwa Chu)

34. Building Phenotypic Networks
(Jen-Hwa Chu)

35. Partial Residual Plots of BMI vs.
Emphysema (Jen Hwa Chu)

36. Development of New COPD Phenotypes:
Emphysema Classification Based on Local Density
Histograms (Castaldi/San Jose/Washko, 2013)
Normal
Moderate CLE
Severe CLE
PLE
Paraseptal
Mild CLE
Classification of
emphysema types based
on the shape of the local
histogram

37. GWAS of Local Histogram Emphysema
Patterns in COPDGene (Castaldi, Submitted)
Phenotype Nearest Gene Locus
Meta-Analysis
P-Value
Normal
CHRNA5 15q25 8.3x10-13
HHIP 4q31 1.7x10-09
MMP12 11q22 1.1x10-08
Moderate
Centrilobular
CHRNA3 15q25 3.1x10-13
CYP2A6 19q13 1.3x10-09
Severe Centrilobular
AGPHD1 15q25 1.8x10-13
MYO1D 17q11 1.5x10-08
Panlobular
AGPHD1 15q25 1.5x10-10
VWA8 13q14 1.1x10-08

38. Potential Approaches to Reclassify
Complex Diseases in Network Medicine
(Silverman/Loscalzo, Discovery Medicine 2012; 14: 143)
Human Subjects
Biological
Samples
Phenotyping
Whole Genome
Sequencing
Physiology
Disease
Determinants
Clinical
Disease
Subtypes
Proteomics
Imaging Clinical
Metabolomics
Transcriptomics
New Disease
Classification
Integrated Network
Analysis
Machine Learning
Approaches

39. • ECLIPSE Genetics Study: Michael Cho, DK Kim, Wayne Anderson, Sreekumar Pillai, Xiangyang Kong, David
Lomas, ECLIPSE Steering/Scientific Committees
• Norway Case-Control Study: Inga-Cecilie Soerheim, Per Bakke, Amund Gulsvik, Sreekumar Pillai, Craig Hersh,
Dawn DeMeo, Michael Cho
• International COPD Genetics Network: David Lomas, Harvey Coxson, Steve Rennard, Barry Make, Peter Pare,
Peter Calverley, Emiel Wouters, Alvar Agusti, Claudio Donner, Jorgen Vestbo, Sreekumar Pillai, Wayne
Anderson, David Goldstein
• Functional Genetics of COPD: Xiaobo Zhou, Augustine Choi, Dawn DeMeo, Craig Hersh, John Quackenbush,
Kimberly Glass, John Platig, Amitabh Sharma, Yang-Yu Liu, Caroline Owen, Mark Perrella, Bart Celli, Miguel
Divo, Zhiqiang Jiang, Taotao Lao
• COPDGene: James Crapo, Barry Make, John Hokanson, Doug Everett, Terri Beaty, Michael Cho, David Lynch,
George Washko, Raul San Jose Estepar, James Ross, Merry-Lynn McDonald, Jen-Hwa Chu, and 21 Clinical
Centers
• Funding: NIH R01 R01 HL075478 (ICGN), R01 HL089856 and R01 HL089897 (COPDGene), R01 111759 (COPD
Networks), P01 HL105339 (COPD Functional Genetics PPG) and GlaxoSmithKline (ICGN and ECLIPSE)
Collaborators