A Systems Approach to Understanding the Mechanisms of Sudden Cardiac Death in Heart Failure Raimond L. Winslow Center for Cardiovascular Bioinformatics

1
A Systems Approach to Understanding the
Mechanisms of Sudden Cardiac Death in Heart
FailureRaimond L. WinslowCenter for
Cardiovascular Bioinformatics ModelingJohns
Hopkins University Whiting School of Engineering
andSchool of Medicine(www.ccbm.jhu.edu)
2
Center for Cardiovascular Bioinformatics
Modeling(www.ccbm.jhu.edu)
Bioinformatics Modeling Core
D. W. Reynolds Cardiovascular Clinical Research
Center (www.reynolds.jhmi.edu) JHU-NHLBI
Proteomics Center (www.proteomics.jhu.edu) NHLBI
Microarray Resource NHLBI SCOR in SCD
To develop new methods for the representation,
storage, analysis and modeling of biological
data, and to apply these methods to better
understand cardiovascular function in both health
and disease
3
Understanding the Cause, Risk Prediction and
Treatment of Sudden Cardiac Death
Sudden Cardiac Death due to

• Heart Failure
• The primary U.S. hospital discharge diagnosis
• Incidence 400,000/year
• Prevalence 4.5 million
• 15 mortality at 1 Yr, 80 mortality at 6 Yr
• Coronary Artery Disease

4
The Heart Failure Phenotype
MR Imaging of Canine Heart Pre- and Post- Failure
Chamber Dilation Wall Thinning
5
Approaches
Experimental Techniques Human and Animal Models
Cell Membrane Transporter Function
Cell/Tissue Electro- Physiology
Gene/Protein Expression
Cardiac Imaging
Ventricular Conduction
Patient Cohorts
Microarrays 2D PAGE Mass Spec (MALDI-TOF,
TOF-TOF, SELDI)
MR Diffusion Tensor Imaging
Recombinant Expression Systems Whole Cell
Patch-Clamp Recording
Ca2, Na V NADH, FADH, Vmito, Ca2mito
Electrode Arrays
Clinical Outcomes
Modeling Data Analysis
Topics Data Representation Storage Class
Prediction Cellular Modeling Integrative
Whole-Heart Modeling
6
Data Representation and Storage
HTML
SOAP
IBM WebsphereTM
SQL
SOAP
SOAP
Database Federation Software (DB2 Information
Integrator)
Data Analysis Visualization
Models
MAGE-DB2
Protein-DB2
IMAGING
Clinical
7
Data Representation and Storage - Genomics
Experiments Completed or Underway

• Altered Gene Expression in End-Stage Human Heart
  Failure
• 5 normals, 6 failing, two replicate measurements
  each, HG-U133A arrays, Yung et al. (2004)
  Genomics 83(2) 281
• Meta-analysis of gene expression in human
  end-stage heart failure by the International
  Heart Failure Consortium
• 300 failing tissue samples, 50 controls
• Time-evolving changes in gene expression in the
  canine tachycardia pacing-induced model of heart
  failure
• Tissue samples at 0 days, 3 days, 1 week, 3 weeks
  and end-stage heart failure
• Triplicate measurements, Affy canine arrays,
  custom canine cDNA arrays

8
Data Representation and Organization MAGE-DB2
IBM Tousley Dubbles Murphy

• Goal - provide researchers with an integrated
  web-based environment for the analysis,
  visualization and storage of gene expression
  data.
• MAGE-OM MicroArray and Gene Expression Object
  Model is a standard for the representation of
  microarray data.
• MAGE-ML is a XML format based on MAGE-OM.
• MAGE-DB2 database is a relational mapping of
  MAGE-OM optimized for IBM/DB2
• Operational

9
Data Representation and OrganizationProteomics
Experiments Completed or Underway

• Altered protein expression/post-translational
  modifications in selected sub-proteomes in
  end-stage human and canine tachycardia
  pacing-induced heart failure
• Meta-analysis of protein expression in human
  end-stage heart failure by the International
  Heart Failure Consortium
• Blood serum protein profiling in patients
  diagnosed with IDCM and ICM
• As a function of NYHA stage
• With/without cardiac events

10
Data Representation and OrganizationProtein-DB2
Store Experiment Metadata And ProGenesis Gel
Spot Data
Protein-DB2
Models
Query and Analyze Data Sets
11
Data Representation and Organization Protein-DB2
(cont.)

• Schema based on emerging standards for
  representation of proteomics experiments and data
  (PEDRO/MIAPE - Taylor et al (2003) Nature Biotech
  21(3) pp.247-254)
• Full relational mapping of this schema

12
Disease Classification/Risk Prediction Based on
Genomic and Proteomic Profiles

• Challenges
• Small sample-size
• Use of complex classifiers (SVMs, NNs,) can
  easily lead to over-fitting and inflated
  estimates of classification rates
• Simple is better
• Interpretability
• Rank-Based Classification

13
Rank-Based Classification

• Define a set C 1,2 of classes (e.g., high
  versus low risk of SCD)
• Let Yi be expression level of the ith gene
• Define for all unique gene pairs (i,j)
• Identify marker gene pairs (i,j) for which the
  score ?ij is large
• Choose the top scoring gene pairs ( Top-Scoring
  Pair Classifier)
• Apply a classification decision rule based on
  pairwise comparison of relative expression values
  for the marker gene pairs (i,j) (e.g., majority
  voting)
• Method invariant to any data normalization method
  which is monotonic in expression level
• No parameters

14
Performance
Prediction Problem Sample Size TSP Previous Results
Cardiac 22 100 NA
Survival 60 83 47-77 1
Leukemia 72 94 85-95 2
Prostate 102 95 86-92 3
Geman, DAvignon, Naiman and Winslow (2004).
Classifying gene expression profiles from
pairwise mRNA comparison, submitted
1Dudoit, S. Fridlyand, J. in Statistical
Analysis of Gene Expression Microarray Data (ed.
Speed, T.) (2003) 2Golub, T. R. et al. Science
286, 531-537 (1999) 3Singh, D. et al. Cancer
Cell 1, 203-209 (2002)
15
Integrative Modeling of the Cardiac Ventricular
Myocyte

• Human and canine ventricular myocyte models

• Ion channels membrane transporters

INaK
INab
INa
Winslow et al Circ. Res. 84 571-586
16
Altered Expression of EC Coupling Proteins and
the Cellular Phenotype of Heart Failure
Altered Gene Expression in End-Stage Canine and
Human Heart Failure
Yung et al (2004). Genomics. 83 281-297
Genes Encoding K Currents
Genes Encoding EC Coupling Proteins
KCND3 (Ito1) 66 ATP2A2 (62)
KCNJ12 (IK1) 32 NCX1
(75)


Effects on the Action Potential?
Greenstein Winslow (2002). Biophys. J. 83(6)
2918
17
Computational Model of Heart FailureAP Duration
and Ca2 Transients
Model
Normal
CHF
Experiment
Winslow et al (1999). Circ. Res. 84 571
18
Tight Negative-Feedback Control of ICa,L by JSR
Ca2 Release
10 nm
10 nm
Bers (2002) Nature 415 198-205
19
Integrating from Cell to Ventricular
FunctionDTMR Imaging of Ventricular Anatomic
Structure
DTMRI vs HISTO Fiber Angles
DTMRI Reconstruction of Canine Ventricles
Scollan et al (1998). Am. J. Physiol. 275
H2308 Holmes, A. et al (2000). Magn. Res. Med.,
44157
Scollan et al (2000). Ann. BME. 28(8) 934 Helm
et al (2004). Mag. Res. Imaging, in review Beg
et al (2004). Mag. Res. Med., in review
20
Data Representation and OrganizationImaging
Experiments Completed or Underway

• High-resolution diffusion tensor MR imaging of
  normal and failing canine hearts
• 300 mm in-plane resolution
• 10 normal and 10 failing canine hearts
• Reconstruction of 1 normal human heart, more
  normal and end-stage heart failure
  reconstructions planned

21
Cardiac Imaging Database
22
Finite Element Models of Cardiac Ventricular
Anatomy

• User selects number of volume elements/nodes
• Matlab GUI for visual control of the fitting
  process
• All imaging datasets, FE models, and FEM software
  are available at www.ccmb.jhu.edu

Endocardial Fibers FEM Model
Epicardial Fibers FEM Model
23
Modeling Electrical Conduction in the Cardiac
VentriclesEADs Can Trigger Ventricular
Arrhythmias
Reaction-Diffusion Equation
EADs Trigger Reentry and Polymorpic VT
Winslow et al (2000). Ann. Rev. Biomed. Eng., 2
119-155
24
Closing the Loop on Whole-Heart Experimentsand
Models
256 Epicardial Electrode Array
MR Image and Model Ventricular Anatomy
Measure Electrode Positions
25
Closing the Loop on Whole-Heart Experiments and
Models (cont.)

• Electrically mapped and DTMR imaged 4 normal and
  3 failing canine hearts
• 256-electrode sock array, 5mm electrode spacing
• Complete anatomical and electrical reconstruction
  performed on one normal canine heart

Winslow et al. (2002). Novartis Foundation
Symposium 247 In Silico Simulation of Biological
Processes, pgs. 129-150, John Wiley Sons, Ltd.
2002.
26
Acknowledgements
Modeling Analysis
Experiments
Databases/Software Eng.
Christina Yung Kass Lab Marban Lab McVeigh Lab
(NIH) ORourke Lab Tomaselli Lab Van Eyk Lab Yue
Lab
Tabish Almas Don Geman Joseph Greenstein Pat
Helm Robert Hinch Vivek Iyer Saleet Jafri Reza
Mazhari Dan Naiman Jeremy Rice David
Scollan Antti Tanskanen Lei Xu
William Baumgartner Jr. Eric Chen Lin
Chen Stephen Granite Donita Robinson Keith
Stevens Joseph Xiang Joel Dubbels (IBM) Kelly
Murphy (IBM) Bob Szabo) Brian Tousley (IBM)
Imaging
Pat Helm Alex Holmes David Scollan Jiangyang Zhang
Supported by the NIH (HL60133, HL70894, HL61711,
HL72488, P50 HL52307, NO1-HV-28180, ), the Falk
Medical Trust, the Whitaker Foundation, the D. W
Reynolds Foundation and IBM Corporation