Design and application of fragment libraries for protein crystallography
John Badger
March 23, 2009
Library Design, Search Methods and Applications of Fragment-based Drug Design
237th ACS National Meeting, Salt Lake City
Why fragment screening ?
More sophisticated view of potency emphasizes ligand efficiency over IC50 (I.D. Kuntz, K. Chen, K. A. Sharp, and P. A. Kollman, PNAS,1999 96, 9997) with LE > 0.3kcal/atom for the clinic
Feature analysis shows exponentially falling probability of binding with increasing compound complexity (M.M.Hann, A.R.Leach,G.HarperJ.Chem.Inf.Comput.Sci., 2001 41, 856)
Simpler to control compound properties by growing small compounds than by modifying large compounds
Practical methodology for academic and small biotech environments
An operational approach to assess the drugability of a target site
Some disease classes ill-suited to HTS discovery (CNS diseases)
A decade of positive results from early leads to clinic (P.J. Hajduk, J.Greer Nature Reviews Drug Discovery, 2007 6, 211) with 47 compounds at significant development stages, 4 compounds in clinical trials including 2 compounds developed in < 2 years
Some preconceptions
A poster child for modern drug discovery ?
Some preconceptions
But small compounds may have a big physiological impact
Some preconceptions
But small compounds may have a big physiological impact
caffeine
tylenol (acetaminophen)
tegretol(carbamazepine)
Library Design
Special aspects of library design for crystallographic screening
Resource and technology intensive
- small, efficient libraries (typically 300-2000 compounds, assayed in shape-diverse mixtures of 4-10 compounds)
Allows detection of small and very weakly binding compounds- binding constants < 10 mM
Demands high compound solubility- ~200mM DMSO allows mixture experiments
Structure data enables rational, efficient and creative exploitation of small fragment hits- grow compounds by addition of functional groups into accessible space
Fragments for drug discovery
Use small and relatively rigid cores- tabulation of drug-like ring systems and preferred side chains by Hartshorn et al, J.Med.Chem., 48, 403-413, 2005
Combine each selected ring system with preferred side chain(s)Obtain ‘classic’ (small) fragments
Examples of simple ring systems
Examples of preferred side chains
Compounds resemble fragments of drugs
Compound properties
Congreve et al (Drug Discovery Today, 8, 876-877, 2003) suggested ‘rule of 3’ as optimal properties for screening fragments- molecular weight <300Da
- no. H-bond donors/acceptors ≤3
- clogP < 3
- no. rotatable bonds ≤3
- polar surface area <60Å2
Example commercial screening collection containing ~500,000 compounds with ‘rule of 3’ properties
MW < 300Da-> 4,700 compounds
MW < 250Da-> 2,400 compounds
MW < 200Da-> 120 compounds
Commercial screening collections are biased towards larger compoundsand may not meet the design needs of asmall fragment library
Exceptions to the rule
Very low MW compounds might be useful
High counts of H-bond acceptors and high PSA for some unique nitrogen-containing ring systems
Theoretical predictions for logP might eliminate some interesting and usable compounds
Data from www.chemexper.com/tools/propertyExplorer/cLogP.html
MW 110.1
Many hsp90 lead compounds areresorcinol and adenine derivatives !
Design strategy
1. Search available compounds for drug-like core substructures and retain those containing appropriate side chains
2. Check molecular properties of the compounds containing appropriate cores/side chains- filtering pass using the ‘rule of three’ provides
good working set
- filtering pass using relaxed rules may provide useful additional compound types
3. Finalize with visual check - weed out overt toxicities, compounds related to controlled substances, unstable compounds..
4. Experimental determination of solubility
SDsearch:Library design by cherry-picking from compound collections
Molecular weightNumber of H-bond acceptorsNumber of H-bond donorsSolubility (clogP)Polar surface areaNumber of rotatable bonds
Number of non-hydrogen atomsNumber of stereo centersPresence of phosphate atomsBlood-brain barrier permeabilitySpecific structure rejection (vendor ID or similarity check)
SMILES substructuresThree-dimensional pharmacophoresProtein interaction motifs
‘Rule of three’ and associated filters
Useful additional filters
Structure filters
-Supports Maybridgeand Chembridge property tags-Property calculations employOpenBabel 2 (OpenSource)
Zenobia small fragment Library
352 decorated ring compoundsSoluble in 200mM DMSOMean properties- MW 154 Da
- No. H-bond acceptors 2.6
- No. H-bond donors 1.4
- clogP 1.6
- tPSA 52 Å2
Four 96 well plates-100µL/compound
- shape diverse mixtures of 8 compounds organized as plate columns
Chemical images and PDBfiles sorted into mixture groups
SD and Excel files of library
Application to CNS targets
CNS disease targets
Compounds emerging from HTS are often too large for passive diffusion across the blood-brain barrier
- FBDD allows growing by 1-2 functional groups while maintaining desirable compounds properties
Properties that characterize fragments resemble the properties required for leads for effective CNS drugs
Fragments
MW 110-250 Da
HBa, HBd ≤ 3clogP < 3PSA <60Å2
CNS drugs
MW < 450 Da (mean 350 Da)
Hba+Hbd < 8
Ideal logP ~2
PSA < 60Å2
Targets at Zenobia Therapeutics include:- Leucine Rich Repeat Kinase 2 (LRRK2)
- Glycogen Synthase Kinase 3 (GSK3)
Evolution of fragment based drug discovery
Use larger libraries of larger compounds (scaffolds) for initial screen - allows multiple assay methods because larger compounds may have higher binding affinity
- development pathways closer to traditional HTS
Use small libraries of small compounds in initial screen and follow by opportunistic addition of functional groups- x-ray crystallography optimal technology for detecting very weak binder
- computational analog is analyzing PDB data to identify key binding interaction motifs (2-4 atoms) and substructures from known ligands
Early lead discovery pathway
Identify key binding interactions- Experimental small fragment screening (x-ray, SPR)- Computational motif identification by mining protein:ligand structures (PDB), binding database,…
First scaffold screen with focused libraries- Select 80 commercially available scaffold compounds- Perform activity inhibition assays- Perform IC50 measurement on top hits
Second round libraries, expansion around best hits and ‘interesting’ structures- Select 50-100 commercially available small lead compounds per chemotype- Perform activity inhibition assays- Perform IC50 measurement on top hits- Perform cell/PK/BBB analysis on top hit
MW: 253 DaHba: 2.0HBd:1.4clogP:1.9PSA:46.4 Å2
MW: 325 DaHba: 3.0HBd:1.3clogP:2.3PSA:54.6 Å2
Computational pilot study to identify G2019S LRRK2 and GSK3β scaffolds
155
303
7855
450,000 Total
Predictive BBB penetration
Predictive kinasebinding motifs
Steric fit to kinaseactive site
80 compounds chosen for assay to test computational model
LeadModel3D:Viable compound expansion from small fragments or motifs
Input the 3D protein:fragmentstructure and a library of candidate compounds that contain the fragment as a substructureDocks each candidate compound onto the fragment and explodes each compound into conformersPoses that severely clash with protein are not viable; score remaining poses with contact potentialCapture list of viable candidate compounds for synthesis or purchase
Predicted best binding mode (yellow bonds) fits the protein cavity and is close to the experimentally determined structure (green bonds)
Candidate compoundcontaining the fragmentFragment hit
Explode each candidate intoconformers and match fragment atoms
Hits were identified for GSK3β
-60
-40
-20
0
20
40
60
80
100
120
0 20 40 60 80 100
% Inhibition at 200μM
30 hits >80%
4 unique scaffolds chosen for further analysis
IC50, LE, PSA for GSK3β inhibitor scaffolds: Promising Leads
Scaffold 1 Scaffold 2 Scaffold 3 Scaffold 4
%Inh at 200μM 100 100 91 82
IC50(μM) < 2.5 6 25 70
MW 222 237 238 291
HA 14 15 15 14
LE 0.55 0.48 0.42 0.41
PSA (Å2) 42 49 41 41
LRRK2 screen produced hits
-20
0
20
40
60
80
100
120
0 20 40 60 80 100
% Inhibition at 200μM5 unique scaffolds identified and chosen for further analysis
8 hits >80%
IC50 and LE for LRRK2 inhibitor scaffolds: Promising Leads
Scaffold 1 Scaffold 2 Scaffold 3 Scaffold 4 Scaffold 5
%Inh at 200μM
82 76 94 92 87
IC50(μM) 15 66 5 30 70
MW 264 216 237 241 291
HA 18 16 15 16 14
LE 0.36 0.35 0.47 0.38 0.40
PSA (Å2) 59 58 49 58 41
Results from follow-on LRRK2 screen
25
-20
0
20
40
60
80
100
120
0 10 20 30 40 50 60 70 80
% Inhibition at 20uM
IC50’s for follow-on compounds: SARbyCatalogue
Scaffold 3 Cmpnd1 Cmpnd2 Cmpnd3 Cmpnd4 Cmpnd5
IC50 (uM) 5 10 0.8 0.15 3 1
MW 237 323 302 287 302 440
HA 15 21 20 19 18 25
LE 0.49 0.33 0.42 0.49 0.42 0.33
PSA 49 65 62 58 59 71
Chosen to test di-substitution
Acknowledgements
Zenobia Therapeutics
Vicki Nienaber (CEO/CSO)
Ruo Steensma
Barbara Chie-Leon
Vandana Sridhar
Cheyenne Logan
Leslie Hernandez
Kristina Bull
Johns Hopkins
Christopher Ross
Shanshan Zhu