Content uploaded by Henriette Willems
Author content
All content in this area was uploaded by Henriette Willems on Jun 26, 2023
Content may be subject to copyright.
The discovery of orthosteric and
allosteric PI5P4K inhibitors
Henriëtte Willems
21 June 2023
ALBORADA Drug Discovery Institute
ALBORADA DDI, Cambridge ARUK DDI, Oxford ARUK DDI, UCL
•3 Institutes with 30 people each
•Part of university
• Core funded by Alzheimer’s Research
UK
•Target validation, compound
screening, chemistry, computational
chemistry and some in vivo work in
house
•DMPK, ADMET, X-ray and some
synthesis outsourced
2
Focus: Enhancing cellular protein clearance pathways
upregulates
autophagy
PI PI(5)P PI(4,5)P2
PIKfyve PI5P4K
(,or )
Inhibitor
•Autophagy target: PI5P4K lipid kinases
•Inhibiting of PI5P4K lipid kinases may
upregulate autophagy
3
Lundquist et al 2018, Mol.Cell,70(3), 531−544
De Leo et al 2016 Nat.Cell Biol.2016,18(8),839−850
PI5P4K lipid kinases
The 3 PI5P4K isoforms superposed
PI5P4Kα
PI5P4K
PI5P4K
•Lipid kinases have low homology with other
kinase families
•PI5P4K has 3 isoforms with high homology
4
DiscoverX TreeSpot
kinome map
•Active site homology of (from top):
•PI5P4Kγ
•PI5P4Kβ
•PI5P4Kα
Active site residues are
highlighted: green=identical and
shades of blue = similar.
Horizontal bars indicate secondary
structure elements
PI5P4K virtual screen
•No crystal structures with inhibitors in the PDB at the time
•No potent known ligands in ChEMBL at the time
osome where reported as part of kinome scans, but none were active in our assay
oATP is weak substrate
•Part of G-loop near active site not visible in crystal structure
PDB: 3X01: PI5P4Kbeta with AMP PDB 2YBX: proposed key interactions
5
MOE pharmacophore with excluded volumes
BioAscent compound library
Purchase and screen
960 compounds
Top 1000
by ASP Score
6148
compounds
from GOLD
+ MOE VS
GOLD
VS
Glide
VS
Top 1000
by Glide Score
Top 1000
by ChemScore
1. Redocking with more
‘expensive’ protocol
Selecting compounds for purchase
•Compounds were screened in a biochemical ADP Glo assay (single point, 83 µM)
•Actives (> mean +3x robust SD) were followed up with IC50’s
•9 hits < 10 μM for PI5P4Kα(0.94% hit rate)
•5 hits < 10 μM for PI5P4Kγ(0.52% hit rate)
2057
unique
com-
pounds
6
logD< 5
PSA <100
ChemAxon
1743
CNS-like
com-
pounds
2. Re-evaluating
properties
k-medoid
clustering 960
compounds
selected by
cluster and
docking rank
Knime
3. Clustering
Retrospective analysis
Conformers
from
Pharma-
cophore
Comment
Hits retrieved
Alpha Actives
present (max 9)
Gamma Actives
present (max 5)
GOLD 2 feat Rescore 4158 8 5
GOLD 2 of 4 feat Rescore 5009 9 5
MOE 2 feat 250 MOE confs 15570 8 5
MOE 2 of 4 feat 250 MOE confs 20766 9 5
MOE 3 of 4 feat 250 MOE confs 4584 2 3
GOLD none CS Fitness > 30 3723 5 3
•Retrospective analysis shows that more restrictive pharmacophore-only strategy did not find all hits
•Using the exact docked position provided by GOLD to filter with MOE pharmacophore worked best
•Gold without additional pharmacophore scoring missed the best hits
• Analysis based on the 960 compounds actually tested, so it’s possible there were more actives
7
PI5P4K virtual screening hits
PI5P4Kαhits with pIC50sPI5P4Kγhits with pIC50s
Rooney, T. P. C. et al (2023) J. Med. Chem. 66, 804-821.
Willems, H. M. G. et al (2023) RSC Med. Chem. 14, 934-946. 8
PI5P4K SAR by catalogue Best follow up hits
9
PI5P4K
γHit
Hit
pIC
50
#
bought
#
active*
Best
pIC
50
6.5
20
10
7.4
6.1
17
1
5.3
5.6
6
0
NA
5.3
37
0
NA
PI5P4K
αHit
Hit
pIC
50
#
bought
#
active*
Best
pIC
50
6.4
39
13
7.1
5.7
60
2
5.0
5.4
42
2
5.4
5.6
24
0
NA
5.5
0
0
NA
5.3
0
0
NA
*Active: pIC50 > 5
Rooney, T. P. C. et al (2023) J. Med. Chem. 66, 804-821.
Willems, H. M. G. et al (2023) RSC Med. Chem. 14, 934-946.
NIH-12848
PI5P4Kγ+ pIC50 = 6.1
Chrom logD = 6.3
Solubility < 5 μM
Hit from kinome profiling study
Clarke et al 2015, Biochem. J., p 359, DOI: 10.1042/BJ20141333
Liang et al 2014 Nat. Chem. Biol. P 298 DOI: 10.1038/nchembio.1455
HDX-MS areas in blue
projected onto 2GK9 (apo)
•Selective: Not active at PI5P4Kαor β
•Looks like a kinase hinge binder, but could not be
docked in ATP site
•HDX-MS studies suggest that the protein areas in
blue are involved in binding the ligand
• However, there isn’t a clear pocket in this area
10
Site finder results as spheres
Alternative starting point for PI5P4Kγ
Profile of allosteric and orthosteric key compounds
2821
1607
40
PI5P4K
pIC50
8.0
<4.3
<4.3
PI5P4K
4%@100
uM
<4.6
<4.6
PI5P4K
+ pIC50
<4.3
7.0
6.2
PI5P4K
αWT InCell pIC50
6.6
PI5P4K
WT InCell pIC50
6.6
6.0
Human
Liver Mics t½ (min)
62
50
47
Mouse Liver
Mics t½ (min)
11
42
12
Kinetic Solubility (µM)
3
38
7
P
app A>B (10-6 cm/s)
0.2
20
122
Efflux ratio
0.8
3.7
0.4
Plasma protein binding (m,%)
91.1
99
brain protein binding (m, %)
94.6
Kp,uu
0.35
11
Panel
2821
1607
40
140
protein
kinases
no
hits
2 hits with <50%
residual activity
@ 10
M
1 hit with <50%
residual activity @
10
M
19 lipid
kinases
PI5P4K
α
only
PI5P4K
and
one other IC
50
<1
M
Plotted with CORAL: https://doi.org/10.1016/j.cels.2018.07.001
PI5P4Kγ crystal structure with ‘compound 40’
Two mutually exclusive binding sites:
•Green: ligand in ATP site (pdb:7QIE, chain B)
•Orange: ligand in allosteric site (pdb:7QIE, chain A)
Allosteric binding mode (above) is dominant
12
Compound 40
PI5P4Kγ+ pIC50 = 6.2
Chrom logD = 4.5
Solubility = 7 μM
Boffey, H. K. et al (2022) J. Med. Chem. 65, 3359–3370.
PI5P4Kγbinding sites only open upon ligand binding
•The binding site for 40 in the allosteric pocket is
occluded in the apo crystal structure (2GK9; pink)
Pink: 2GK9 apo PI5P4Kγ
Light orange: location of compound 40
in chain A of 7QIE PI5P4Kγ 13
•The apo crystal structure is very similar to chain B where
compound 40 is bound to the ATP site, but the Phe141, Met206
and Lys216 orientation prevent docking
Pink: 2GK9 apo PI5P4Kγ
Green: 7QIE chainB
1607 binding mode to PI5P4Kγ differs from AMP and compound 40
1607 binding mode
(cyan; pdb: 8BQ4);
PI5P4Kγ ;
crystal structure
chain B with
compound 40 in
green (pdb:7QIE;
PI5P4Kγ )
•The binding mode of 1607 (pdb: 8BQ4) was unexpected,
•Only 1 typical hinge interaction was made
•1607 and 40 chain B binding sites are very similar
14
1607 binding mode
(cyan; pdb: 8BQ4;
PI5P4Kγ );
AMP from PI5P4Kβ
structure pdb: 3x01
superposed in grey
PI5P4Kalpha crystal structure with 2821
Comparison with apo structure pdb:2ybx in blue Comparison with probe structure pdb:6ym5 in magenta
•Loop Lys228-Thr232 has re-arranged
•Phe134 and Lys209 side chains re-oriented
•Water network re-arranged and through water binding
•No loop re-arrangement
•Only Phe134 re-oriented
•Water network and through
water binding as for 2821
PDB:8C8C
with 2821 in
gold
15
Comparison of alpha and gamma orthosteric binding modes
•Through water
interaction with
Thr196 results in
different placement of
pyrimidine ring in the
alpha structure
•Resulting water
network pushes the
dimethylphenyl ring
forward
PDB:8C8C
PI5P4Kαwith
2821 in gold
PDB:8BQ4
PI5P4Kγwith
1607 in blue
16
Docking of PI5P4K ligands
• Docking with ‘hinge’ constraints in place resulted in incorrect binding mode for gamma ligand
•However, constraint needed for docking to apo alpha structure
10 binding mode in
cyan based on
pdb:8BQ4 of
analogue 1607;
10 docking pose
(hinge constraints
used) in teal
17
2821 binding mode
to PI5P4Kαin
yellow (pdb:8C8C);
VS hit 1docking
pose in orange
PI5P4KαPI5P4Kγ
Conformation of PI5P4Kαligands
A
B
C
Origin/
angle
(
°)
PDB:
8C8C
2821
CSD:
ETUWAU
CSD:
OFALEQ
21
χ
56
6
79
ψ
15
7
171
ω
45
61
22
Flipping Ψangle of 2821 (A) 180 degrees results in conformation only
0.2 kcals/mol higher in energy*
*Jaguar optimization, CC-pVTZ-F, M06-X2-D3, PBF solv
Willems, H. M. G. et al (2023) RSC Med. Chem. 14, 934-946. 18
Interesting alpha SAR
R1
R2
PI5P4K
α
pIC
50
18
7.1
19
6.9
20
<4.3
21
6.6
22
7.0
23
5.9
24
<4.3
PDB:8C8C
PI5P4Kαwith
2821 in gold
19
Willems, H. M. G. et al (2023) RSC Med. Chem. 14, 934-946.
Summary
20
•Three different series of PI5P4K inhibitors were identified
oSeries 1 and 2, identified from virtual screening, were found to bind in the ATP site
oThe third series, developed from a literature compound, were found to be allosteric inhibitors.
•The amino-pyrimidine privileged motif can act as hinge binder (PI5P4Kα), but doesn’t in all cases (e.g.
orthosteric or the allosteric PI5P4Kγligands)
•The activation loop can obstruct (PI5P4Kγ) or from part of the binding site (PI5P4Kα).
•A single subtype-agnostic virtual screen yielded selective PI5P4Kαand selective PI5P4Kγinhibitors
•The best results from virtual screening were obtained by docking, followed by a MOE pharmacophore
screen using the docked position as input, rather than just relying on docking scores
•The dockings produced by the virtual screening turned out to be incorrect, because side chain re-
arrangements, loop movements and water networks changed the binding pocket
•Conformational preferences are important and often not recognized in docking scoring functions
Acknowledgements
ALBORADA DDI
Greg Aldred
Steve Andrews
Helen Boffey
Stephen Chawner
Jon Clarke
Simon Edwards
Chris Green
Tamara Romero
Tim Rooney
John Skidmore
David Winpenny
PEAK PROTEINS
Steven Harborne
Tina Howard
Derek Ogg
21