anna linusson - biomedbridges · pharmacophore-based screening execution • fit molecules (i.e....
TRANSCRIPT
Anna Linusson
Department of Chemistry
Umeå University
Sweden
Virtual Screening
Virtual screening
N
NH
O
O
OH
Cl
O
O
S N
N
S
O
O
OH
N N
NS O
HN
O
NN
OSNH
O
NOO
N
N N
NH2
Free energy of binding
++
000 ln STHKRTG A
)( 000
''
0
LaqRaqaqLRG
Experimentally determined inhibition constants =10-2 to 10-12 M, i.e. 10 pM to10
mM; pKi= 2 to 8 (at T=298 K)
A one order of magnitude change corresponds to 5.7 kJ/mol or 1.36 kcal/mol
(1kcal = 4.18 kJ)
Virtual screening
Development Validation
O
O
HO
OH
6 (Embelin)
Preparation Execution Evaluation
Outline
• Methodology
– Ligand-based
– Structure-based
• What can we expect?
• Virtual screening vs. HTS
• Case study
• Take home message
Virtual screening
• Enthalpy
– Ionic interactions
– Hydrogen bonds
– Dipole-dipole moments
– Aromatic interactions
– van der Waal interactions
• Binding entropy
– loss of conformational flexibility
Solvation/desolvation
Flexibility (degrees of freedom)
++
Virtual screening
• Similarity principle
• Receptor-based approach
A plethora of methods/approaches to
choose from
O
O
HO
OH
6 (Embelin)
Case dependent
Don’t trust retrospective studies
What do I know?
Years 1 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15
No. of compounds Up to 10,000 10-15 1-8 1-3 1
First patent application
Clinical trial application
Product licence application
Drug Discovery Drug Development Target and lead identification
Lead optimisation Concept testing
Development for launch Launch
Clinical Development Phase I 50-150 people
Phase II 100-200 people
Phase III 500-5,000 people
Phase IV studies continue
Product life cycle support
Toxicology and pharmacokinetic studies (absorption, distribution, metabolism, excretion)
Pharmaceutical and analytical development
Process chemistry and manufacturing
Registration and regulatory affairs
Sales and marketing (preparation, promotion, advertising and selling)
Cost $3 million $225 million $450 million
Drug Discovery and Development
Virtual screening - methodology
Ligand-based
• Demands a (set of) known
ligands/active molecule/s
• These molecules may be
structurally similar or different
• Validated inactive molecules may
also be of use
• Demands a 3D-structure of the target
receptor (protein)
• These may be crystal structures,
NMR structures or homology models
• Several structures of the same target
bound to different ligands is an
advantage
Structure-based
Read more: Taboureau, O. et al. Chemistry & Biology, 2012, 19, 29-41.
Klebe, DDT, 2006, 11: 580-594
Shoichet, Nature, 2004, 432: 862-865
O
O
HO
OH
6 (Embelin)
Virtual screening – generic procedures
Development Validation
Preparation Execution Evaluation
• Source of chemicals (e.g. commercial)
• Selection of known ligand/s and/or
target/s
• Preparation of chemical structures – Chemicals and ligands: 2D, 2D->3D
– Macromolecules: X-ray, NMR, homology models
– Tautomers
– Protonation states
– Stereochemistry
– Conformations
• Experimental evaluation
• Confirmation of hits – Biologically
– Chemical structure
• Follow up
Ligand-based virtual screening
Screening methods
• Descriptor-based
• Shape-based
• Pharmacophore-based
• ...
Prerequisite
One or a set
of known
active
molecules
• Unsupervised (similarity search)
• Supervised (regression models)
O
O
HO
OH
6 (Embelin)
Descriptor-based screening
Preparation
• Calculation of descriptors
• Fingerprints
• Physical properties
• 3D-GRID descriptors
• Selection of similarity measure
(Tanimoto)
• Creation of a quantitative structure-
activity relationship
• Linear regression model
• Non-linear regression model
• Rule-based model
• Assessment of model quality
• Applicability domain
Based on the assumption that a similarity in compound physicochemical descriptors
(i.e. physicochemical properties) will lead to a similar biological action
Execution
• Estimation of similarity
• Prediction of the biological
activity of the compound library
structures using the QSAR model
Evaluation
• Use the ranking list from execution
• Cut offs
• Selection criteria (diversity)
• Additional filtering
• Clustering
Shape-based screening
The goal is to find compounds similar in 3D-shape to the query molecule(s)
Preparation
• Generation of 3D structures
• (Alignment of structures)
• Calculating volumes
• Creation of shape-query
• (Combine with electrostatics)
• Selection of similarity measure
Execution
• Screening of the compound
database (3D-structures) based on
overlapping volumes
• Multiple shape queries
• A good overlap will be ranked high
Evaluation
• Prioritization of hit compounds
• Cut offs
• Chemical features
• Combination of shape queries
Read more: Kirchmair et al, J. Chem. Inf.
Model. 2009, 49, 678–692
Pharmacophore-based screening
Preparation
“ A pharmacophore is an abstract (3D) description of chemical features in a
molecule, necessary for biological recognition ”
• Generation of low-energy 3D
structures of the active
molecules
• Alignment of structures
• Identification of 2-4 common
features
• Creation of 3D-pharmacophore
• Assign pharmacophore features
to a search chemicals
• Select measure
Align analogues
Create
pharmacophore
Pseudo-receptor
Pharmacophore-based screening
Execution
• Fit molecules (i.e. the
pharmacophore features) in the
database to the generated
pharmacophore
• A good match is ranked high
Evaluation
• Prioritization of hit compounds
• What about diversity among
the hits?
• Many different pharmacophores
Read more: Kim et al., Expert Opin. Drug
Discov., 2010, 5(3),205-222
Structure-based virtual screening
Molecular Docking
• Rigid receptor - flexible ligands
• “Soft” receptor - flexible ligands
• Multiple receptors
• Induced-fit docking
Prerequisite
Structure of known target
Docking-based virtual screening
Preparation
• Preparation of protein structure(s)
• Addition of hydrogens
• Optimization of H-bond networks
• Treatment of water
• Protonation state and tautomers
• Definition of binding site
• Optional settings depending on docking
software
• Inclusion of water
• Movement of side-chains
• Energy minimization of resulting
complexes
Structure-based virtual screening
Re-docking (and cross-docking) of the crystal structure ligand may be guide the choice of docking software
Docking-based virtual screening
Structure-based virtual screening
Execution
• Generation of molecule conformations
• Generation of docking-poses for each
molecule
• Judgment of clashes and interactions
between ligand and protein
• Depending on docking software
• addition and removal of water
molecules in the binding site
• docking filtering (removal unwanted
molecules based on e.g. absence of
specific interactions with the protein)
Docking-based virtual screening
Evaluation
• Scoring of docking poses using one
scoring function
• Scoring using several scoring functions
(consensus scoring or fused)
• Selection of a subset of molecules using
a simple scoring function and re-scoring
with a computational intensive function
• Biological evaluation
Choice of scoring function
may be guided by the scoring
of known active and inactive
molecules
Example of a “simple” scoring function that
estimates the free energy of binding (∆Gbind ):
∆Gbind= ∆Gsolv +∆Gint +∆Gconf +∆Ghbond +∆Gelect
Account for (some) flexibility
Multiple receptors
• Crystal structures of the same protein
with different conformations induced
by different ligands
• Conformation ensembles of a protein
generated by MD-simulations
• “Manual” alterations of proteins
structures (e.g. conformations of
flexible loops)
Soft receptors
• Mimics movement by varying the
Lennard-Jones potential of the
receptor atoms (i.e. the distance at
which atoms clash)
• Allowing for more “clashes”
• Variable binding site size
Ligated complexes usually performs better than apo-structures
The complex with the largest ligand is usually a good choice
Account for (some) flexibility
• Movements or position of hydrogen
atoms attached to hetero atoms (O, N
or S)
• Flips of histidine, aspargine and
glutamine side chains (i.e. tautomers)
• Allowing for movements of user-
specified receptor side chain and/or
backbone atoms.
• The entire complex is optimized (e.g.
the protein is forced to adopt to the
ligand)
Further reading: Kokh et al.,, Comput. Mol. Sci.,
2011, 1, 298–314
Demands a scoring function that can account for protein internal energy changes
Induced-fit docking
Water
• Tools to predict important waters
• Docking tools to account for waters
on/off
Crystal structure can guide for important waters Highly conserved waters can be included The benefit of included waters is questionable If any – then case dependentc
To include or not to include?
NB: It is not questionable that
water is important
Using multiple methods
Compound Library
106-108
104-105
102-103
Fast: Rigid/shape based filtering,
Fast: Pharmacophore/substructure searches,
Fast: High-throughput docking
Slow: Flexible docking
Fast: Rescoring,
Slow: Visual Inspection
ex. ROCS, FRED
ex. GOLD, Glide,
FlexX, ...
Biological evaluation
Which method/approach should I use?
• How to best use existing information about ligands and targets?
• What is your goal?
• Combining methods is often beneficial
Nicholls, JCAMD, 2008,
22:239-255
Warren et al., JCM, 2006,
49: 5912-5931
Case dependent
Don’t trust retrospective studies
What can we expect?
Marsden et al. OBC, 2004, 2:3267-3273
Enyedy and Egan, JCAMD, 2008, 22:161-168 Scoring functions cannot
predict binding affinities
What can we expect?
++
000 ln STHKRTG A
31 pitfalls: Agrafiotis and co-workers,
JCIM, 2012, 52:867-881
Pitfalls and traps: Hawkin et al., JCAMD,
2008, 22:170-179
Shoichet, Nature, 2004, 432: 862-865
Klebe, DDT, 2006, 11: 580-594
• Limitations in fundamental knowledge
• Assumptions made (flexibility, water, simplifications)
• Bad science (pre-treatments, statistics, parameters, validations)
Sources of errors in virtual screening
Knowledge-based driven approach
What can we expect?
Enrichment. Increase the likelihood of finding promising hits.
To what good? • Recover false negatives from HTS
and disclose false positives
• When only a smaller library can be
screened
Hit rates:
Virtual screening and HTS are complementary methods
Ferreria et al., JCM, 2010, 53: 4893-4905
Polgár et al. Comb. Chem. High Throughput Screening, 2011,
14, 889-897.
Outline
• Methodology
– Ligand-based
– Structure-based
• What can we expect?
• Virtual screening vs. HTS
• Case study
• Take home message
ADP-ribosyltransferases
• Post-translational
modification
• NAD+ as cosubstrate
• Mono- or Poly-
ribosylation
• Writers in epigenetics
The nomenclature... ARTD: diphteria toxin-like ADP-ribosyltransferase
also known as PARP: poly ADP-ribose polymerase
Docking-based virtual screening
...to find new small molecules that bind to ADP-ribosyltransferases
...to develop selective ligands for different ADP-ribosyltransferases
...to use the molecules as chemical tools in the elucidation of biological
functions
Andersson et al., JCM,
2012, 55: 7706-7718
Virtual screening
PARP15 PARP14
What do we know?
And what do we want?
Virtual screening methodology
Preparation • Re-docking and cross-docking of
crystal structures
• Evaluation of docking tools and
settings
Chemicals
• In house compound collection
• Protonation states, tautomers and
stereoisomers
• Determination of features and property
criteria
• Filtering of compound collection
Proteins
• Investigation of D-loop conformations of
all crystalized PARPs
• Selection of protein structures
• Selection of conformers
• Protein preparation (hydrogens,
tautomers, protonation states)
• Definition of binding site
Known binders
• Collection of list known binders based on
crystal structures and/or experimental
assays
• Protonation states, tautomers and
stereoisomers
Compound Database Target/s
Virtual screening methodology
Execution Evaluation
Docking score resemblance
Previously known binders
Compounds selected for testing
Compounds were selected based on scoring
similarities to known binders
Selection of compounds for wet lab
• 64 were chosen by our scoring method and 64 (47) were selected
solely based on Glidescore
• 111 molecules were selected for biological testing
• No enzymatic assay available!
Senisterra et al. Mol. BioSyst., 2009, 5, 217-223
Thermal shift assay
Hits for PARP14 and/or PARP15
ΔTm of 1.5-3.9 degrees
Identified by scoring resemblance Identified by Glidescore
Multiple protein structures
• Hits were identified from screening against all
protein structures
• 63% of the hits were identified from PARPs with
unaltered D-loop
• Three hits were exclusively identified from
structures with truncated loop (including the two
best hits)
Selectivity profile
PARP15 PARP14 PARP1
All 111 compounds experimentally tested for binding to PARP1
Hit validation
OH
O
O
NH
O
NH2
O
NHO
H2N
O OH
A16(E)
A16(Z)
Hit validation
A16(Z) A16(E)
Kd = 7.6 μM Kd = 11.2 μM
O
NHO
H2N
O OH
OH
O
O
NH
O
NH2
Isothermal titration calorimetry
Crystallization of A16 and PARP14
O
NHO
H2N
O OH
Resolution 1.9 Å Resolution 2.8 Å
Z-isomer A16
OH
O
O
NH
O
NH2
E-isomer A16
Binding mode of A16 to PARP14
Crystal structure of A16 (E and Z)
1
2
Docking pose of A16(E)
Case summary
• 16 binders (8 different structural classes)
=> hit rate of 14 %
• Our scoring method found structurally diverse
compounds
• Both isomers of A16 binds to PARP14 (~10μM)
• Crystal structures of the A16 isomers confirmed
docking poses
• The virtual screen provided good starting points to
develop selective inhibitors
Take home message
• Virtual screening increases the likelihood of finding
hits in a set
• Virtual screening is a knowledge-based method –
use available information
• Don’t trust retrospective studies
• Virtual screening is a complementary method to
HTS
Acknowledgements
Department of Chemisty, UmU
Linusson’s Lab
David Andersson
Lotta Berg
Brijesh Kumar Mishra
Cecilia Lindgren
Urszula Uciechowska
Cecilia Engdahl
Elofsson’s Lab
Sara Spjut
Anders Lindgren
Rémi Caraballo
Wittung-Stafshede’s Lab
Pernilla Wittung-Stafshede
Moritz Niemiec
Laboratories for Chemical Biology
Umeå (LCBU)
Karolinska institutet
Schüler’s Lab
Tobias Karlberg
Torun Ekblad
Ann-Gerd Thorsell
Johan Weigelt
Financial Support
The Swedish Research Council
Umeå University
Swedish Foundation for Strategic Research