Computational Systems Medicine: Drug Development at the Interface of Roger L. Chang

advertisement
Computational Systems Medicine:
Drug Development at the Interface of
Structural and Systems Biology
Roger L. Chang
November 23, 2011
Path of Drug
Action
×
Secretion
Uptake
Metabolic
network
Systemic
response
Affect protein
Target
binding
function
×
×
×××
×
×
Protein
×
Drug molecules
inhibition
Active
site
Computational Evaluation of Drug Target Effects
Proteome
Drug binding
site alignments
SMAP
Predicted drug targets
Drug and endogenous
substrate binding site analysis
Competitively inhibitable targets
Inhibition simulations in
context-specific model
COBRA Toolbox
Predicted causal targets
and genetic risk factors
(Chang et al PLoS Comput Biol 2010)
Polypharmacology
• Drug promiscuity
predominates.
• Comprehensive
experimental
detection of drug
targets currently
impractical.
• Computational
prediction
valuable for
defining
experimental
targets.
(Paolini et al Nat Biotech 2006)
Edge connects proteins if ≥1 drug binds both.
SMAP for Prediction of Drug Off-Targets
Identify drug binding
site of known target
Identify off-targets by
binding site similarity
(SOIPPA)
Scoring based on:
• Geometrical fit
• Residue conservation
• Physiochemical similarity
Dock drug
to off-targets
(Ren et al Nucleic Acids Res 2010)
Drug
Endogenous
substrate
Binding Site Analysis for Inhibitability
• Overlap between drug-binding sites (SMAP) and native active sites
(PDB, Catalytic Site Atlas) suggests inhibitability.
• Estimation of relative binding affinities (docking) suggests strength
of competitive inhibition.
Constraint-based Metabolic Modeling
Metabolic network reactions
Flux space
Steady-state
assumption
S·v=0
Perturbation
constraint
Flux
Matrix representation of network
HEX1
?
PGI
?
PFK
?
FBA
?
TPI
?
GAPD
?
PGK
?
PGM
?
ENO
?
PYK
?
Change in system capacity
Recon1: A Human Metabolic Network
Global Metabolic Map
Comprehensively represents
known reactions in human cells
Pathways
(98)
Reactions
(3,311)
Compounds
(2,712)
Genes (1,496)
Transcripts (1,905)
Proteins (2,004)
Compartments (7)
http://bigg.ucsd.edu
(Duarte et al Proc Natl Acad Sci USA 2007)
Context-specific Modeling Pipeline
metabolomic
biofluid & tissue
localization data
metabolic
network
gene
expression
data
constrain
exchange
fluxes
preliminary
model
refine
based on
capabilities
model
set flux
constraints
objective
function
literature
normalize &
set threshold
GIMME
metabolic
influx
set minimum
objective flux
metabolic
efflux
Case Study: Torcetrapib Side Effect
• Cholesteryl ester transfer protein (CETP) inhibitors treat
cardiovascular disease by raising HDL and lowering LDL
cholesterol (Torcetrapib, Anacetrapib, JTT-705).
• Torcetrapib withdrawn due to occasional lethal
side effect, severe hypertension.
• Cause of
suggested.
hypertension
undetermined;
off-target
effects
• Predicted off-targets include metabolic enzymes. Renal function
is strong determinant of blood pressure. Causal off-targets may be
found through the modeling kidney metabolism.
Kidney Model Subsystem Distribution
• Kidney model: 228 genes, 448 reactions
• Largest subsystem is membrane transporters, expected for renal filtration.
Remainder involved in indirect reabsorption and secretion synthesis.
Renal objectives
Exchange
Metabolite
Prostaglandin I2
Prostaglandin D2
Calcitriol
secretion
Urea
Cyclic AMP
Urate
Tryptamine
Water
Phosphate
Sodium
Calcium
Chloride
Protium
Potassium
absorption
Bicarbonate
Acetate
Citrate
Oxalate
D-Glucose
amino acids
L-Carnosine
Glutathione
Perturbed Phenotype Simulation
Secretions
Absorptions
Objective
flux
×
×
Perturbation
constraint
Max perturbed flux
Max unperturbed flux
=
Degree of phenotype =
System
boundary
constraint
Max flux
# exchanges
Gene-deficient Renal Phenotypes
• Disorders caused by
20 out of 118
simulated gene
deficiencies clinically
validated in literature.
100
Gene deficiencies
• Predicted genedeficient renal
disorders also
constitute potential
risk factors for
treatment.
0
Cryptic genetic risk factors phenotype only under
combined gene-deficient,
drug-treated perturbation.
Deficiency and target could be
isozymes or in parallel
pathways.
Drug off-targets
Torcetrapib Renal Response Phenotypes
Severity of phenotype increases when
combined with cryptic gene-deficiency.
Predicted Hypertension Causal Drug
Off-Targets
Official
Symbol
PTGIS
Protein
Prostacyclin
synthase
Impacts
Functional Reactions Renal
Off-Target Site
Limited by Function in
Prediction Overlap
Expression Simulation
Stronger
Drug
Binding
Affinity
Cryptic Genetic Risk Factors
x
x
x
x
x
ACOX1 Acyl CoA oxidase
x
x
x
x
x
AK3L1
Adenylate kinase 4
x
x
x
x
HAO2
Hydroxyacid oxidase 2
x
x
x
x
SLC3A1; SLC7A9; SLC7A10;
ABCC1
x
x
x
CYP27B1; ABCC1
x
x
x
CYP27B1; ABCC1
Mitochondrial
cytochrome c oxidase I
Ubiquinol-cytochrome c
UQCRC1
reductase core protein I
MT-COI
*Clinically linked to hypertension.
Conclusions
• Torcetrapib hypertension side effect may result from
renal metabolic off-target effects.
• Framework for perturbation phenotype simulation
capable of predicting metabolic disorders, causal
drug targets, and genetic risk factors for drug
treatment (including cryptic risk factors).
• Pipeline established for in silico prediction of
systemic drug response.
Structural Reconstruction: a Resource for Structural
Systems Biology
Structure-enabled Network
Metabolic Network
1st structure-enabled network for
T. maritima central metabolism
(Zhang et al Science 2009)
Protein Structures
Genome
Structure-enabled Metabolic Network
• Enables structure analysis in network context:
– Protein-ligand interactions
– Post-translational modifications
– Protein stability with environmental shifts
– Fold and pathway evolution
• E. coli chosen over human because:
– More PDB enzyme structures (2975 vs 1609)
– More complete metabolic network (iJO1366)
– Simpler physiology for simulations
– Ease of experimental validation
E. coli Metabolic Network (iJO1366)
• E. coli K-12 MG1655 reconstruction
• Genes: 1366
• Unique Proteins: 1254
• 858 in multimeric complexes
• Unique metabolites: 1136
• Reactions: 2251
(Orth et al Mol Syst Biol 2011)
Scope of E. coli Structural Reconstruction
Single-chain Phase
Complex Phase
Best PDB
structures
•
•
Native WT structure coverage
Metabolite substrate(s) bound
•
Physiological assemblies
(asymmetric unit)
Model
structures
•
Template with max sequence
coverage/similarity
Template with metabolite
substrate(s) bound
•
Physiological assemblies
(symmetry operations)
•
Residue-level •
functional
•
annotation
.
•
.
•
.
.
Asymmetric Unit
Cell
Unit
Full Crystal
E. coli Structural Reconstruction: Single-chain Phase
• Structures = 2892
– PDB = 581
– Models = 2311
• Proteins = 1366
• Prot-met pairs UB = 8467
443
98
1553
462
4299
806
PDB
Model
No rep.
2172
PDB
Model
No rep. definite
No rep. indefinite
E. coli Structural Reconstruction: Complex Phase
• Structures (best rep.) = 523
– PDB AU = 421 (3369 total)
– PDB BU = 82
– PISA assembly = 20
• Complexes = 1106
• Oligomeric states
150 5
434
189
587
85
Complete
Partial
None
510
252
Assumed monomer
Monomer
Homodimer
Homomultimer (>2)
Heteromultimer
Applications Under Way
• PTMs & allosteric regulation
frequency
allosteric
p = 2.3E-8
competitive
p = 3.0E-3
number of regulated reactions
• Protein-antibiotic interactions
• Protein stability and pathway
usage
Acknowledgements
Bernhard Palsson
Teddy O’Brien Josh Lerman
Phil Bourne
Nate Lewis
Lei Xie
Li Xie
Adam Godzik
Zhanwen Li
Download