GPCR fragment-based design and a novel structurebased perspective on druggability
John Christopher & Jonathan Mason
4th RSC/SCI Symposium on GPCRs in Medicinal Chemistry.
Windlesham September 2012
Heptares Therapeutics

Breakthrough medicines targeting previously undruggable GPCRs

$40M from leading venture investors since 2009

Major R&D partnerships with multiple large pharma companies
 Structure-Based Drug Design for GPCRs, enabled for the first time by StaRs
(Stabilised Receptors)

Leading GPCR capability in industry, integrating chemistry & structural biology

Engine creating both small molecule NCEs and antibody therapeutics

Exceptional pipeline of investigational medicines for serious diseases
1
Non-Confidential
Heptares Product Pipeline
GPCR
Discovery
Preclinical
Phase 1
Phase 2
Muscarinic M1
Indication(s)
Alzheimer’s Disease, Schizophrenia
Orexin 1
Binge Eating, Nicotine Addiction
Dual Orexin 1/2
Chronic Insomnia
GLP-1
Type 2 Diabetes (First Oral NCE)
GPR39
Type 2 Diabetes (Disease Modifying)
mGlu5
Autism, Dyskinesia, Depression
CGRP
Migraine Treatment & Prophylaxis
MedImmune
Adenosine A2A / Progress Confidential
CNS Disorders
Progress Confidential
CNS Disorders
Progress Confidential
Undisclosed
Progress Confidential
Antibody Therapeutics
Progress Confidential
Undisclosed
Non-Confidential
Outline of Presentation
 Fragment
 Fragment

Screening Methods
H2L examples
A Water perspective...
– Druggability
– SAR & Kinetics
3
Non-Confidential
StaRs enable Fragment Screening
Primary screening
validated with
• SPR
• NMR
• HCS
• CE
Hits triaged by
• SPR kinetics
• SPR stoichiometry
• Binding assays
• Thermal shift
Hits validated by
• SAR / analogues
• X-ray / modelling
• BPM / SDM
SPR
NMR
CE
HCS
SPR
kinetics /
stoichiometry
SAR
Primary
Screening
Radioligand
binding
X-ray
Structural
model
4
Thermal shift
BPM
Hit
Confirmation
Hit
Validation
Non-Confidential
SPR Fragment Screening POC
A2A StaR
riboflavin
More potent (mM)
fragments
theophylline
DPCPX
Highly stable surface
– DPCPX controls
caffeine
1-methylxanthine
FAD
7-methylxanthine
xanthine
3-methylxanthine
allopurinol
hypoxanthine
Weak (mM) fragments
easily detected
Inactive fragments

Weakly binding fragments hits easily discriminated from inactives

Xanthines added to library as likely binders

Chip stable for days
Congreve, M. et al, Methods Enzymol. 2011, 493, 115
5
Non-Confidential
Vs. plots
Adenosine A2A and b1 Adrenergic Receptor (b1 AR)
SPR screen
b1 selective
hits
Standard
Non-selective
hits
70
b1AR
50
30
A2A selective
hits
10
-10
-10
10
30
50
70
A2A
Similar results for screening Heptares
fragment library against A2A and b1AR
6
Non-Confidential
NMR Screening: b1AR (with ZoBio)
 Immobilized protein – only small amounts needed (~1mg)
 Very sensitive method: mM hits identified (not found by SPR)
TINS = Target Immobilized NMR Screening:
Congreve, M. et al, Methods Enzymol. 2011, 493, 115
7
Non-Confidential
High Concentration Screening
Lipid Receptor Agonist StaR
StaR
Wild-type
Bound ligand (cpm)
Bound ligand (cpm)
25000
20000
15000
10000
5000
Log [Agonist] (M)
assay radioligand usage significantly reduced).
10% DMSO has a minimal effect on ligand-binding
to the StaR in membranes, unlike wild-type.
Enables high concentration (100 mM) fragment
screening which is not feasible with wild-type.

1000
100
% inhibition of specific binding (n = 1)
Agonist StaR binds known agonists with higher
affinity compared to wild-type receptor (binding

1%
5% [DMSO]
10%
2000
-12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2
Log [Agonist] (M)
-11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1

3000
0
0

4000
Approx 6% hit rate (84 / 1419 fragments inhibit
80
60
40
20
0
0
binding > 30%, n = 2).
8
20
40
60
80
% inhibition of specific binding (n = 2)
100
Non-Confidential
Outline of Presentation
 Fragment
 Fragment

Screening Methods
H2L examples
A Water perspective...
– Druggability
– SAR & Kinetics
9
Non-Confidential
Hit to lead example 1:
CXCR4 Chemokine Receptor Antagonist






39 hits (4.8%) from 808 molecules
(fragments + targeted screening set)
N
HN
NH
NH
Follow up gives clear SAR
HN
N
H
Rapid identification of potent hit
series with considerably better
profile than gold standard
Example fragment hit
Ki = 150 mM (LE = 0.47)
Good solubility
MWT 144, cLogP 1.3, PSA 39
N
N
H
Plerixafor
Dosed s.c.
MWT = 502
cLogP = -0.2



Hit Series Exemplar
Ki = 10 nM (LE = 0.34)
Good solubility
MWT 300, cLogP 1.3, PSA 76
10
Non-Confidential
b1AR
Hit to lead example 2:
b1AR antagonists

b1
Control
SPR screening
with A2A as
counter screen
b1 Hits
Caffeine

Several related hits
A2A
Control
A2A Hits
A2A
Typical results for „well behaved‟ hits
H
N
N
F
F
F
KD = 16 mM
11
Non-Confidential
b1AR: Structure-guided hit to lead progression
O
NH
OH
Warne, T. et al.
Nature 2011, 469, 241
O
N
H
OH
H
N
SPR hit
KD = 16 mM
LE = 0.41
N
F
F
Analogue
purchasing
IC50 = 7 mM
Binding Assay
LE = 0.50
Analogue
purchasing
IC50 = 58 nM
Binding Assay
LE = 0.66
MW <250, Highly soluble
F
Non-Confidential
b1AR: Fragment X-ray crystallography
• Protein-Fragment crystal co-crystal complexes solved to high resolution.
• Novel chemical matter, breaking the mould of existing chemistry.
- Lack aminoalcohol motif
Non-Confidential
Comparison of Heptares GPCR Fragment
Hits with Enzyme Targets
Target
Method
Hit LE
Target
Method
Hit LE
Adenosine A2A
TINS NMR
0.56
Protein kinase B
X-ray soak
0.47
Adenosine A2A
SPR
0.53
DPPIV
HCS
0.46
Family A aminergic
SPR
0.41
Thrombin
X-ray soak
0.40
Family A peptidergic
SPR
0.31
BACE
SPR
~0.30
Family A lipid
HCS
0.55*
HSP90
NMR
0.53
CXCR4
HCS
0.47
PDE4
HCS
0.46
11 GPCRs (Family A, B & C) screened in 14 assays
GPCRs are highly comparable to enzyme targets in terms of quality of hits,
when StaR proteins are used for screening.
* StaR assay
14
Non-Confidential
Outline of Presentation
 Fragment
Screening Methods
 Fragment
H2L examples

A Water perspective...
– Druggability
– SAR & Kinetics
15
Non-Confidential
GPCR structural information can be revolutionary – previously we were
biased by ligand + analogs data
Beware of biases in how we see and “force-fit” data
Non-Confidential
Courtesy of Arthur Doweyko,
Why Water?
 Water molecules play an essential role in the structure and function of
biological systems
-appear to play a key role in the GPCR structures available to date, perhaps due
to the deep pockets present in the GPCR transmembrane binding sites?
 Displacement of waters from a binding site is a key component of ligand
binding, with significant binding energy, and thus potency, often from the
entropic gain of the displacement
 But all waters are not equal…
- Burying an ”unhappy” water [i.e. entropically and/or enthalpically worse
than bulk sovent] may affect both potency and kinetics
- Pertubation of the remaining waters will also affect binding ±
 Opportunity to provide new insights into druggability, to drive SAR and find
solutions to many SAR issues (predicition of ”magic methyl”, SAR not explainable
by direct ligand-protein interactions...)
17
Non-Confidential
Water: The new wave...
 New computational approaches now available that can create water
networks (with or without a ligand present)
+ energetically differentiate ”happy” and ”unhappy” waters
 Favourable binding sites contain multiple ”unhappy” waters, with a cluster
being particularly favourable ( higher affinity/LE fragment hits etc...)
- in an environment compatible with a drug-like molecule (mixture of
hydrophobic & hydrophilic hotspots rule of 5 etc properties)
OH
... e.g. not all serines are the same
Ser
OH
OH
All pockets are
equal, but some
pockets are more
equal than others
Ser
Ser
18
Non-Confidential
The New Wave: WaterMap (Schrödinger) + GRID (Mol. Discovery)
DG
WaterMap coded:
Vacuum
bubble
Factor Xa
very ‘unhappy’
water
‘unhappy’
water
S4
Bulk
solvent
bulk-like
water
‘happy’
water
S1
19
1.
2.
GRID Map contoured at:
Aromatic CH (lipophilic) probe: Yellow
-2.5 kcal/mol
Water probe: Green -6.0 kcal/mol
Surface defined by CH3 probe: Grey
1.0 kcal/mol
MD simulation and clustering technique to build a map of water occupancy in the factor Xa active site
Chemical potentials assigned to the water sites using the inhomogenous solvation theory
Non-Confidential
Abel et al.
A New Structure-Based Perspective on Druggability
– 3D GRID physicochemical + Water energetics

 GRID (Molecular Discovery) used to identify binding site complementary
properties – energetic analysis of binding site with ligand functional groups:
focus on hotspots for lipophilic probe (C1=) + H-bonding (H2O probe)
- new combined hydrophobic (DRY) and lipophilic (C1=) probe : CRY
 WaterMap (Schrödinger) & SZMAP (Open Eye) used to generate a full
network of waters in binding site with free energy estimation (vs bulk solvent –
neutral atom) including a breakdown into entropic & enthalpic contributions

 Combined analysis enables a new druggability assessment:
- fragment/ligand potency, properties, shape requirements…
20
Non-Confidential
A New Structure-Based Perspective on Druggability

– 3D GRID physicochemical + Water energetics
 Waters that are calculated to have a significant positive free
energy (e.g. relative to being in bulk solvent) are termed „unhappy‟.
- There should be a particularly good free energy gain from
displacing these waters; they are only in the site as creating a
vacuum is even more energetically unfavourable.
- The number and the relative position of the predicted „unhappy‟
water clusters reveal hotspots for small molecule ligand
binding
21
Non-Confidential
Water Networks: Validation
 Can we generate a water network in a protein-ligand complex similar
to that experimentally determined in high resolution X-ray structures ?
- e.g. A2A at 1.8Å (4EIY)
 Excellent correlation
+ we also know the relative energies
of the waters, and can re-evaluate
with different ligands
- a significant advance
 Can we now explain previously
unexplained SAR (in terms of direct
ligand-protein interactions) etc
22
Non-Confidential
Binding Site Analysis for Druggability

 Druggability – ligandability with properties of an oral drug
 Use a combined analysis of „unhappy‟ waters & binding site
preference (lipophilic/hydrophobic)
 Analyse numbers, connectivity & distribution of „unhappy‟ waters
 „Unhappy‟ water = >2.5 kcal vs bulk solvent (vacuum usually worse)
The link with protein active site druggability is not only to the total number of
„unhappy‟ waters in each binding pocket but also to their arrangement and
proximity (connectivity) to other „unhappy‟ waters.
-Two representative protease enzyme
targets factor Xa and BACE1 have key
„unhappy‟ waters displaced by the ligand
that are not connected
- extreme case in factor Xa where they are
in subpockets almost 10 Å apart, giving a
lower value of 0–1 connections;
- In GPCRs such as A2A the predicted
„unhappy‟ waters have a favourable
more globular arrangement with an
average 2-3 other adjacent „unhappy‟
waters at < 3 Å
23
Non-Confidential
β1- (from StaR) & β2- (from T4L insertion) Adrenergic Receptors
β1 antag
β1 agonist
Inactive
Active
β2 antag
β2 agonist
24
Non-Confidential
Druggability : Dopamine, Histamine & Muscarinic GPCRs
Dopamine D3
Muscarinic M2
Histamine H1
+
HN
O
O
OH
Druggable regions with clusters of „unhappy‟ waters (WaterMap) with GRID showing concurrent
lipophilic & H-bond hotspots
25
Non-Confidential
New insights from structural biology into the druggability of G protein-coupled receptors. Mason, Bortolato, Congreve, Marshall. Trends Pharmacol Sci. 2012;33(5):249-60.
What about more recent structures?
O
H
+
H N
H
O
-O
P
-
OH
O
NH
O HN
+
N
H
NH
S1P1 Lipid GPCR
Opioid kappa Receptor
26
Non-Confidential
OH
Kinases are more like GPCRs for druggability in terms
of regions of connected „unhappy‟ waters
c-Abl DFG-in
c-Abl DFG-out
New insights from structural biology into the druggability of G protein-coupled receptors. Mason, Bortolato, Congreve, Marshall. Trends Pharmacol Sci. 2012;33(5):249-60.
27
Non-Confidential
Less druggable enzyme targets do not have large clusters
of „unhappy‟ waters
The myth:
Basic S1 for serine
proteases (factor Xa)
The structure that
broke the myth
The clinical candidate
These sparsely distributed ‘unhappy’ waters are distributed in different subpockets, meaning
that fairly ‘angular’ ligands are needed, with a requirement for a precise shape and
conformation; this could explain why random screening (HTS) often fails.  ranked lower in
overall druggability
28
Non-Confidential
Less Druggable Sites Still Tractable to SBDD
GPCR Compared to Protease
CXCR4 has a more challenging site:
Structure shows the “hot-spot” for binding to be
less deep / druggable (lipophilic, less buried
ionic interaction)
BACE has delocalised druggable site
– hot-spots linked by SBDD
29
Non-Confidential
New insights from structural biology into the druggability of G protein-coupled receptors. Mason, Bortolato, Congreve, Marshall. Trends Pharmacol Sci. 2012;33(5):249-60.
Conserved mechanism suggests orthosteric agonists
should be attainable for any GPCR
Inactive
Active
Druggable A2A Agonist site
Druggable A2A Antagonist site
30
Non-Confidential
First GPCR Candidate Wholly Derived from SBDD
Treatment of multiple neuro disorders

Radical binding mode with highly
optimised receptor interactions

Candidate series (blue) binds very efficiently to receptor
Features
Effect of istradefylline, preladenant or HTL
(1 mg/kg, PO) on haloperidol-induced catalepsy in rats

% 0.8 mg/kg haloperidol response
– Non-furan, non-xanthine
– Very low molecular weight
– Relatively polar
Benefits
1h post-dose
2h post-dose
80
60
40
20
0
31
fy
l li
ad
e
tr
H
Langmead, C. et al. (2012) J. Med. Chem. Mar 8;55(5):1904-9
Congreve, M. et al. (2012) J. Med. Chem. Mar 8;55(5):1898-903
Is
al
op
er
id
ol
– Attractive safety profile
– Improved oral bioavailability and PK
– Excellent in vivo efficacy
100
TL

H
Candidate licensed to Shire
t

ad
en
an
Entirely novel chemotypes
el

Pr
Superior adenosine A2A antagonists
ne

Non-Confidential
Using GRID hotspots enabled the first full GPCR SBDD
to give a clinical candidate for the Adenosine A2a
ZM24138
A2A antag
 The highly ligand efficient designed structures displaces a cluster
of unhappy waters deep in the pocket, missed by previous ZM-like
ligands (from HTS etc)
32
Non-Confidential
New insights from structural biology into the druggability of G protein-coupled receptors. Mason, Bortolato, Congreve, Marshall. Trends Pharmacol Sci. 2012;33(5):249-60.
Biophysical MappingTM uses StaRs to map the binding
sites of GPCRs & conf
Site directed
mutagenesis
on StaR®
10-30 Mutations to the
Binding site region
Mutant StaRs screened on
Biacore chips
Biophysical Map
of binding site
Ligand refined
homology model and
prediction of proteinligand binding modes
Correlating binding data from
multiple ligands with multiple
StaR proteins
Detection of binding of
different ligands to each
mutant StaR
33
Zhukov, A. et al., J. Med. Chem. 2011, 54, 4312.
Non-Confidential
Experimentally Enhanced Homology Models
Existing structures
Knowledge of conformations
Reported variants
Further X-ray
structures
First
homology
model
StaR mutagenesis
Multiple generations of
Experimentally Enhanced
Homology Models
Fragment screening
Biophysical
Mapping
Virtual screening
34
Non-Confidential
Structure-Based Lead Optimisation
• Models of A2A, refined using the biophysical mapping, & of A1 used dockings prioritized from large virtual libraries (1,2,4-triazines)
– Used GRID hotspots to design compounds that efficiently displaced the
“unhappy “ waters deep in the site
– Used GRID surfaces of A2 vs A1 to probe for subltle differences to efficiently
design improved selectivity
BP
- Structure can enable small ligand efficient
changes to modulate selectivity & potency
35
Non-Confidential
Waters & GPCR SBDD / Druggability
• New perspective on druggability possible by analysis of binding site using
water free energies & 3D physicochemical properties (GRID)
• Looking at pertubation of all waters, including those remaining, is looking
promising
• Multiple approaches possible, with combined approaches often useful
- for apo structure generate water network with WaterMap or GRID
- for complexes generate initial complete water network with SZMAP
& optimize with WaterMap
- binding site preferences from GRID analysis with functional groups
- scoring (displacement and perturbation) by all 3 methods:
(WaterMap, SZMAP, GRID/CRY probe)
36
Non-Confidential
Acknowledgements
Heptares Therapeutics Ltd.
Computational Chemistry
Protein Engineering
Structural Sciences
Pharmacology
Medicinal Chemistry
Management
U. of Leiden / ZoBio
Gregg Siegal
Francis Figaroa
Johan Hollander
Eiso AB
MRC LMB, Cambridge
U. of Utah / Biosensor Tools
Richard Henderson
Chris Tate
Guillaume Lebon
Tony Warne
Jessica Li
Jenny Miller
David Myszka
Rebecca Rich
Biofocus
Chris Richardson
Paul Scherrer Institute
Gebhard Schertler
37
Non-Confidential