Transcript Mike Hann - UK-QSAR
“Making Lead Discovey less Complex?”
Mike Hann, Andrew Leach & Gavin Harper.
Discovery Research GlaxoSmithKline Medicines Research Centre Gunnels Wood Rd Stevenage SG1 2NY email [email protected]
Introduction A simple model of molecular recognition and it’s implications Experimental data An extreme example Conclusions
HTS & Libraries - have they been successful at revolutionising the drug discovery business?
Despite some successes, it is clear that the high throughput synthesis of libraries and the resulting HTS screening paradigms have not delivered the results that were initially anticipated.
– immaturity of the technology, – lack of understanding of what the right types of molecule to make actually are .
– the inability to make the right types of molecules with the technology .
The Right Type of Molecules?
Drug likeness – Lipinski for oral absorption – Models (eg Mike Abrahams) for BBB penetration – But all these address the properties required for the final candidate drug Lead Likeness – What should we be seeking as good molecules as starting points for drug discovery programs?
– A theoretical analysis of why they need to be different to drug like molecules – Some practical data
A very simple model of Molecular Recognition
Define a linear pattern of +’s and -’s to represent the recognition features of a binding site – these are generic descriptors of recognition (shape, charge, etc) Vary the Length (= Complexity) of this linear Binding site as +’s and -’s Vary the Length (= Complexity) of this linear Ligand up to that of the Binding site Calculate
varies. of number of matches as ligand complexity Example for binding site of 12 features and ligand of 4 features:
Feature Position Binding site features 1 2 3 4 5 6 7 8 9 10 11 12 - - + - + - - + - + + + Ligand mode 1 Ligand mode 2 + + - + + + - +
Probabilities of ligands of varying complexity (i.e. number of features) matching a binding site of complexity 12 0.4
0 1 0.9
As the ligand/receptor match becomes more complex the probability of any given molecule matching falls to zero. i.e. there are many more ways of getting it wrong than right!
2 3 4 5 6 7 8 9 10 11 Complexity of Ligand (I.e. number of ligand features) 12
Example from last slide
Match any 1 matches 2 matches 3 matches 4 matches 5 matches 6 matches 7 matches 8 matches 9 matches 10 matches 11 matches
The effect of potency
(binding site 12; ligand complexity =12)
P (useful event) = P(measure binding) x P(ligand matches) Probability of matching just one way
Low probability of even if there is a 0.8
unique mode 0.7
2 3 4 Optimal.
But where is it for any given system?
5 6 7 8
Low probability of finding lead even if it has high affinity
10 11 12
Probability of useful event (unique mode)
Limitations of the model
Linear representation of complex events No chance for mismatches - ie harsh model No flexibility only + and - considered But the characteristics of any model will be the same
2 3 4 5 9 10 11 12 6 7 8
Ligand Complexity P (useful event) = P(measure binding) x P(ligand matches)
Real data to support this hypothesis!!
Leads vs Drugs
Data taken from W. Sneader’s book “Drug Prototypes and their exploitation” Converted to Daylight Database and then profiled with ADEPT 480 drug case histories in the following plots
Sneader Lead Sneader Drug WDI
Change in MW on going from Lead to Drug for 470 drugs 4 0 0 3 0 0 2 0 0 1 0 0 0 0 -1 0 0 -2 0 0 -3 0 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 Average MW increase = 42 7 0 0 8 0 0 M W o f S n e a d e r D ru g s
ADEPT plots for WDI & a variety of GW libraries
WDI WDI WDI WDI WDI
Molecules in libraries are still even more complex than WDI drugs, let alone Sneader Leads
In terms of numbers
Average property values for the Sneader lead set, average change on going to Sneader drug set and percentage change.
Av # arom 1.3
Av # HBA 2.2
arom 0.2** HBA .3** % 15 % Av ClogP ClogP 1.9
0.5** Av # HBD .85
+ % 26 % Av CMR CMR 7.6
1.0** Av # heavy heavy 19.
3.0** % 14.5
% 14 (4) 16 Av MW MW % Av MV MV % Av # Rot B Rot B % 272 42.0** 15 289 38.0** 13 3.5
.9** 23 Astra Zeneca data similar using hand picked data from literature AZ increases typically even larger RSC/SCI Medchem conference Cambridge 2001. MW increase ca. 70-90 depending on starting definitions
Catch 22 problem
2 3 4 5 9 10 11 12 6 7 8
We are dealing with probabilities so increasing the number of samples assayed will increase the number of hits (=HTS).
We have been increasing the number of samples by making big libraries (=combichem) And to make big libraries you have to have many points of diversity Which leads to greater complexity Which decreases the probability of a given molecule being a hit
Concentration as the escape route
Screen less complex molecules to find more hits – Less potent but higher chance of getting on to the success landscape – Opportunity for medicinal chemists to then optimise by adding back complexity and properties Need for it to be appropriate assay and ligands – e.g the extreme Mulbits (Mul tiple Bits) approach – Mulbits are molecules of MW < 150 and highly soluble.
– Screen at up to 1mM An example indicating how far this can be taken – from 5 years ago - Thrombin: – Screen preselected (in silico) basic Mulbits in a Proflavin displacement assay specific – known to be be specific for P1 pocket.
Thrombin Mulbit to “drug”
N NH NH 2 2-Amino Imidazole (5mM), as the sulphate, showed 30% displacement of Proflavin (18µM) from Thrombin (10µM) (cf Benzamidine (at 5mM) shows 70% displacement) under similar conditions Absorbance at 466nM relative to that at 444nM was used as the measure of amount of proflavin displaced N O N NH 2 S O N H Thrombin IC50 = 4µM (15 min pre-incubation; for assay conditions see reference 23) O N H O N
Related Literature examples of Mulbits type methods
Needles method in use at Roche .
Boehm, H-J.; et al Novel Inhibitors of DNA Gyrase: 3D Structure Based Biased Needle Screening, Hit Validation by Biophysical Methods, and 3D Guided Optimization. A Promising Alternative to Random Screening. J. Med. Chem., 2000, 43 (14), 2664 -2674.
NMR by SAR method in use at Abbott Hajduk, P. J.; Meadows, R. P.; Fesik, S. W.. Discovering high-affinity ligands for proteins.
, 278(5337), 497-499.
Ellman method at Sunesis Maly, D. J.; Choong, I. C.; Ellman, J. A.. Combinatorial target-guided ligand assembly: identification of potent subtype-selective c-Src inhibitors.
Proc. Natl. Acad. Sci. U. S. A.
, 97(6), 2419-2424.
Enzyme target - bangs per bucks
Plot of Log Enzyme activity vs MW for “Interesting monomer” containing inhibitors
-1 1 0 0 m M 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750
-4 -5 -2 -3 nM
Most interesting lead MW of inhibitor
Mwt <200 350
Drug-like H2L problems ?
Lipinski Data zone HTS screening Non-HTS Shapes (Vertex ) Needles(Roche) MULBITS(GSK) Crystallead(Abbott) SARbyNMR(Abbott) Slide adapted from Andy Davis @ AZ
Leads for Drug Discovery
Lipinski etc does not go far enough in directing us to leads.
Michael M. Hann,* Andrew R. Leach, and Gavin Harper
We have provided a model which explains why.
J. Chem. Inf. Comput. Sci., 41 (3), 856 -864, 2001.
“Everything should be made as simple as possible but no simpler.” Einstein where is that optimal peak in the plot for each target?
– Simple does not mean easy!!
Andrew Leach, Gavin Harper.
Darren Green, Craig Jamieson , Rich Green, Giampa Bravi, Andy Brewster, Robin Carr, Miles Congreve,Brian Evans, Albert Jaxa-Chamiec, Duncan Judd, Xiao Lewell, Mika Lindvall, Steve McKeown, Adrian Pipe, Nigel Ramsden, Derek Reynolds, Barry Ross, Nigel Watson, Steve Watson, Malcolm Weir, John Bradshaw, Colin Grey, Vipal Patel, Sue Bethell, Charlie Nichols, Chun-wa Chun and Terry Haley.
Andy Davis and Tudor Oprea at AZ