Transcript In silico Modeling of Membrane Proteins: Application to CFTR

Computer Aided Drug Design
Hanoch Senderowitz
Department of Chemistry
Bar Ilan University
BIU-Valencia Workshop
April 2010
Computer Aided Drug Discovery
Structure/
Sequence
Structure-based
Modeling
Ligand-based
Modeling
Virtual Library
Screening
Virtual Hits
Scoring
Binding Assays
3D Optimization
Leads
Known Ligands
In silico
Chemistry
Biology
Drug Candidate
2
Homology (Comparative) Modeling
• Given a sequence of amino acids, predict the 3D structure of the protein
Template selection
• Multiple sequence alignment
• Multiple structure alignment
Model generation
• External servers
• In-house tools
Model refinement
• Energy minimization
• Molecular dynamics
• Virtual co-crystallization
Model validation
• Model “health”
• Agreement with available data
• Enrichment experiments
3
In Silico Screening
Library Generation
Docking
BMA
Scoring
Selection
• Start: 2D representation of commercially available
compounds
• Filtration: Ligands and/or binding site characteristics
• End: Multiple 3D conformations of ~100K compounds
• Multiple docking tools
• Selection of the most plausible binding mode
• Multiple scoring functions
• Consensus scoring algorithms
• Selection of virtual hits
• Biological assays
4
Ligand-Based Screening
• Pharnacophore: A 3D arrangement of function groups which is responsible for
the biological activity
• Obtained by the superposition of active (and inactive) compounds
• A Database can be screened against pharmacophore
Donor
Aromatic ring
Excluded
volume
Aromatic ring
Donor
Shape based on largest
active compound
Acceptor
5
In Silico Screening Track Record
#
Target
Indication
Comp.
Tested
Hits Hit Rate
Best Hit
(Ki)
#
Target
Indication
1
5-HT1A
Anxiety,
Depression
78
16
21%
1 nM
9
NK1
Depression
2
5-HT4
Alzheimer's
Disease
93
19
21%
21 nM
10
NK2
Asthma
3
5-HT2B
Pulmonary
Hypertension
13 nM
11
V2
4
5-HT6
Obesity
5
D2
* From the 5HT4 screen
(1)
30
42
P2Y2
15%
56 nM
452
47
10%
54 nM
Cirrhosis, CHF
223(4)
7
3%
1.5 uM
(3)
(5)
14 nM
12
B1
Inflamation, Pain
75
3
4%
~6 uM
7
17%
58 nM
13
B2
Inflamation, Pain
136
6
4%
~8 uM(5)
100
129
28
14%
496 nM
20
20%
100 nM (2)
14
CB1
Obesity
Chemokine GPCR
MS, RA
140
7
5%
0.7 uM (2)
Purinergic GPCR
8
8
Cannabinoid GPCR
Kv1.5/4.3
A-fib; Brugada
(Ikur/Ito)
S1P1
53
17%
Lipid GPCR
7
Best Hit
(Ki)
5
Ion Channels
6
Hits Hit Rate
Peptide GPCR
Biogenic amine GPCR
CNS
Comp.
Tested
Cystic Fibrosis
167
16
10%
15
CCR3
Inflammation
43
5
12%
~10 uM(5)
16
CCR2
Inflammation, RA
158
12
8%
399 nM(2)
17
CXCR2
COPD, RA
130
8
6%
82 nM(4)
10 nM
(1) Conformational analysis; (2) IC50 from functionality assay; (3) Pharma collaboration; (4) Pharmacophore screening;
(5) Ki estimated from IC50
OM Becker et al, PNAS 101 (2004), 11304-11309
6
The Cystic Fibrosis Disease
•
•
•
•
CF is the most common lethal genetic disease
among Caucasians
The number of CF patients is estimated at 70,000
worldwide, about 30,000 of which are in the US
In 2008, the median survival age of was ~37 years
CF results in pathologies in multiple organs

•
Depressed lung function, lung infection,
inflammation and advanced lung disease
Airways
Liver
Pancreas
Intestine
Reproductive
Tract
Currently, there is no cure for CF and the only
treatment is symptomatic
Skin
7
The Molecular Basis of CF
• CF is caused by mutations to the Cystic Fibrosis Transmembrane
Conductance Regulator (CFTR) which is the largest Cl- channel in
the body
Normal lung
• Most common disease causing mutation is DF508
• DF508-CFTR does not fold properly: Most channels does not
reach the cell surface; those that do have impaired Clconductance
• In absence of proper Cl- conductance the salt/water balance in the
airways is disrupted leading to dehydration of the mucus layer
lining the airways.
CF lung
• The dehydrated mucus layer becomes colonized by bacteria
leading to chronic lung disease, lung failure and death
• CFTR is a relevant target for developing CF therapeutics but
its structure is unknown
8
Model of Full Length Structure of CFTR
wt-CFTR
• Site is mostly linear
and aligned by
hydrophilic and
aromatic moieties
• Site sufficiently large
for drug like
compounds
• Site supports specific
interactions
DF508-CFTR
• Site small and linear and aligned mainly
by hydrophilic groups
• Site sufficiently large for drug like
compounds
• Site supports specific interactions
9
In vitro Screening
• Compounds tested in vitro in functional, electro-physiology assays in two cell lines

Assays measure channel conductance
~300 compounds from in silico screen
YFP Fluorescence Quenching
• A549 Cells: 12 structure-based hits at 10
FRT DF508 (rat) and A549 DF508 (human)
FRT DF508 Ussing chamber
mM corresponds to a hit rate of 3.9%
pSAR
In-house or at ChanTest
pSAR = Purchased SAR, i.e., purchasable analogs
•
•
•
•
• FRT Cells: 21 structure-based hits at 10 mM
corresponds to a hit rate of 6.6%
• Similar screening campaigns reported
in the literature yielded hit rates of 0.041.1%
Hits represent multiple scaffolds
In these assays, hits activity is similar to the best known CFTR corrector (Corr-4a)
Several hits show dual mechanism acting as both correctors and potentiators
Most promising hits entered lead optimization
10
Lead Optimization: The Art of Balance
BBB
Permeability
hERG
CYP
Efficacy
Solubility
Binding
11
Binding
• MM-GBSA simulations on a model system
(Urokinase-type plasminogen activator (uPA))
• Good correlating when simulation initiated
from crystal structure (R2 = 0.75)
• Poorer correlation when the binding mode
could only be approximated (R2 = 0.60)
• Poor correlation observed when only a
model of the protein is available and /or
when the binding mode is obtained
through docking simulations
• Challenges
• Improved docking and scoring methods
• Improved treatment of entropy
12
When Binding is Improved…
• The hERG gene encodes a potassium channel conducting the repolarizing IKr current of
the cardiac action potential.
• Drug related hERG inhibition could lead to a sudden cardiac death
“Classic” hERG
pharmacophore
Astemizole (potent Privileged structures
hERG binder)
for e.g., GPCRs
N+
Binding to primary target often goes hand in hand with
hERG binding
Solution: hERG model
13
When hERG is Reduced…
• Due to the hydrophobic nature of the hERG binding site, increased polarity may
reduce hERG binding.
• Increased polarity will also lead to:
• Increased solubility
• Decreased permeation through biological membranes
• Decreased permeation through the Blood Brain Barrier
Permeability
Hydrophobicity
Affinity
hERG binding
14
Last But (Certainly) not Least
• Cyp inhibition may lead to toxicity via drug-drug interactions
• Cyp binding sites are large and promiscuous but are otherwise similar to “regular”
binding sites
CYP450-3A4 (PDB code 2v0m)
Cavity size: 950 Å3 to 2000 Å3
CYP450-2D6 (PDB code 2f9q)
Cavity size: 540 Å3
15
Optimization in Chemoinformatics and Drug Design
• Drug Discovery is a multi-objective optimization problem
•
Successful drug candidates necessarily represent a compromise between numerous,
sometimes competing objectives
• Many other problems in chemoinformatics and drug design could be casted
into the form of an optimization problem
Synthesis design
Docking & scoring
QSAR/QSPR
Multiobjective QSAR
Optimization Engine
Classification Models
Conformational search
Consensus scoring
Diversity analysis
16
The Target Function and Variables
• Define a target function (f) and corresponding variables f = f(x1,x2,x3…xn)
Target function and variables related to the scientific problem

Target function and variables define a multi-dimensional surface
Energy

Cartesian/internal
coordinate 1
17
Monte Carlo/Simulated Annealing (MC/SA) Based
Optimization Engine
Random Move
“Trial”
DE
Metropolis
Test
NO
DE < 0
or
exp(-DE/RT) > X[0,1]
YES
?
X[0,1] is a random number in the range 0 to 1
Tmin
100
80
70
60
50
40
30
20
10
0
0
10
20
30
40
50
60
70
MC Steps
80
90
100
100
100
90
80
70
60
50
40
30
20
10
0
0
10
20
30
40
50
60
70
MC Steps
80
90
100
Temperature
90
Temperature
Temperature
MC
Temperature
Tmax
100
90
80
70
60
50
40
30
20
10
0
1000
2000
3000
MC Steps
4000
5000
90
80
70
60
50
40
30
20
10
0
0
10
20
30
40
50
60
70
80
90
100
MC Steps
18
Quantitative Structure Activity Relationship (QSAR)
Quantitative Structure Property Relationship (QSPR)
• Correlate specific biological activity for a set of compounds with their
structure-derived molecular descriptors by means of a mathematical model
• The nature of correlation, activity and descriptors are unlimited

BBB permeability = f (hydrophobicity, H-bonding potential)

Metabolic stability = f (presence/absence of specific fragments)

Protein crystallizability = f (amino acid composition, secondary structure)
19
QSAREngine
1. Descriptors selection
2. Outliers removal
Dataset
3. Generation of multiple models
4. Model(s) validation and selection
Descriptors Calculation
5. Consensus model
6. Validation
7. Predictions
Descriptors Selection
Outlier Removal
Internal Set
External Set
Multiple Divisions
Y-Scrambling
• Avoid chance correlation
Training Set
Test Set
Model Derivation
• Linear (MLR)
• Non-linear (kNN)
Model Selection
Consensus Prediction
• Average
• SD
20
QSAR Model for Metabolic Stability in Human Liver
Microsomes (HLM)
•
Metabolism alters chemicals to speed their removal from the body and is performed
primarily in the liver by the Cytochromes
•
HLM experiments measure compounds resistance to metabolism
•
Compounds incubated with HLM (vesicles containing drug-metabolizing enzymes)
and their t1/2 half life determined
Pred HLM t1/2 (min)
1000
R2 = 0.8127
100
10
1
1
10
100
Obs HLM t1/2 (min)
1000
Dataset
290 in-house compounds and 58
literature compounds
Descriptors
41 descriptors including
fragment counts
Outliers removal
50 outliers removed
External test set
102 compounds
Algorithm
kNN, MLR
Consensus model 190 kNN model
21
The Grand Challenge
• How can we reliably and consistently predict the pharmacological profile of
bio-active compounds?

Basic scientific research

Practical applications in drug design
• How can we make better drugs?
CYP
hERG
Solubility
22
Acknowledgments
• EPIX Pharmaceuticals
• Lab members
• Dr. Efrat Noy
• Dr. Merav Fichman
• Gal Fradin
• Yocheved Beim
• Funding
• CFFT
23