Gregory G.
Aldred‡
,
Timothy P. C.
Rooney‡
,
Henriette M. G.
Willems
,
Helen K.
Boffey
,
Christopher
Green§
,
David
Winpenny
,
John
Skidmore
,
Jonathan H.
Clarke
and
Stephen P.
Andrews
*
The ALBORADA Drug Discovery Institute, University of Cambridge, Island Research Building, Cambridge Biomedical Campus, Hills Road, Cambridge, CB2 0AH, UK. E-mail: spa26@cam.ac.uk
First published on 23rd August 2023
The phosphatidylinositol 5-phosphate 4-kinases (PI5P4Ks) are therapeutic targets for diseases such as cancer, neurodegeneration and immunological disorders as they are key components in regulating cell signalling pathways. In an effort to make probe molecules available for further exploring these targets, we have previously reported PI5P4Kα-selective and PI5P4Kγ-selective ligands. Herein we report the rational design of PI5P4Kα/γ dual inhibitors, using knowledge gained during the development of selective inhibitors for these proteins. ARUK2007145 (39) is disclosed as a potent, cell-active probe molecule with ADMET properties amenable to conducting experiments in cells.
Phosphatidylinositol 5-phosphate (PI5P) is converted to PI(4,5)P2 by a family of phosphatidylinositol 5-phosphate 4-kinases (PI5P4Ks) consisting of three distinct isoforms; alpha, beta and gamma. Heterogeneity exists within the family; the alpha isoform, in vitro, is the most catalytically active,7 whereas the gamma isoform has very low intrinsic activity.8 The beta isoform is the only one to contain a recognised nuclear localisation sequence.9 Although all isoforms are generally ubiquitously expressed, tissue specific differences have been observed.8 Diverse cellular functions have also been associated with the different isoforms, for example PI5P4Kβ and PI5P4Kγ may have roles as GTP sensors,10 and both PI5P4Kα and PI5P4Kγ have been shown to associate to autophagic structures.11,12 Hence the different specific roles reported for each of the isoforms could be functions that are tissue or cell-type specific, or spatially or temporally differentiated in response to activation of different metabolic pathways. Structurally there are also differences between the isoforms9,13 but it is clear that each is able to heterodimerise,7,8 which leads to the intriguing possibility that isoforms are able to regulate the localisation, specific function or in vivo activity of each other.
The promiscuity of the phosphoinositides in cellular signalling pathways has implications for disease as enzyme dysfunction or misregulation may be causative in various developmental disorders, inflammation and infection.13–16 As oncology targets, PI5P4Ks have become increasingly associated with a number of diseases.17 Interestingly the involvement of PI5P4Kα alone has been documented in glioblastoma, acute myeloid leukemia (AML) and prostate cancer18–21 but other examples suggest that there may be a combinatorial role for the PI5P4Ks. Both PI5P4Kα and PI5P4Kβ are involved in p53-deficient breast cancer and soft tissue sarcomas22,23 and high expression of both PI5P4Kα and PI5P4Kγ associate with unfavourable clinical outcome in AML.24
Several recent reports detail inhibitors for PI5P4Ks, not only pan-specific25–29 but also isoform specific.30–34 Emerging tools are also being developed for dual-specific inhibitors for PI5P4Kα and PI5P4Kβ22,35 and methods have also been reported for removing protein completely in cells using PROTAC systems, such as JWZ-1-80 for PI5P4Kγ,36 which will be useful to interrogate non-catalytic roles. Here we present a new tool that can be used to investigate the impact of dual inhibition of both the PI5P4Kα and PI5P4Kγ isoforms.
We have previously reported PI5P4Kγ-selective inhibitors including ARUK2001607 (1) which was derived from virtual screening hit 2,30 as well as PI5P4Kα-selective inhibitors including tool compound ARUK2002821 (3) which was derived from virtual screening hit 4 (ref. 33) (Table 1). During the course of those studies, we also became interested in developing dual PI5P4Kα–PI5P4Kγ inhibitors as probe molecules. In particular, we had observed small amounts of PI5P4Kα inhibition in some analogues of 1 which we were able to further optimise through rational design. Herein we describe efforts to further increase the level of PI5P4Kα activity in the 1 chemotype towards the development PI5P4Kα/γ dual inhibitors with IC50 fold-selectivities within approximately 10-fold in either direction.
As the phenyl group of 5 had an increased MW and Xlog
P compared to 1 we also focussed on improving physicochemical properties during iterations of compound design. In particular, we were keen to introduce heteroatoms in this ring to modulate X
log
P. The first synthesised compound of this type, 6, shows a marked decrease in X
log
P compared to 5 and was found to have increased potency for PI5P4Kα and PI5P4Kγ+ (Table 2).37 When the 8BQ4 crystal structure of PI5P4Kγ was later solved with 1 bound in the ATP binding site, docking of 6 suggested that its pyridyl N might interact with Lys216 of PI5P4Kγ (Fig. 1).
Inhibition of PI5P4K | Physicochemical properties | |||||
---|---|---|---|---|---|---|
R 1 | PI5P4Kα pIC50 | PI5P4Kγ+ pIC50 | Fold-selectivity (α IC50/γ IC50) | M W |
X![]() ![]() |
|
a Inconsistent pIC50 measurements between replicates; value shown is upper limit. | ||||||
5 |
![]() |
<5.3 | 7.4 | >130 | 381 | 4.5 |
6 |
![]() |
5.9 | 7.7 | 63 | 382 | 3.1 |
7 |
![]() |
5.9 | 7.8 | 79 | 371 | 3.7 |
8 |
![]() |
5.2 | 7.4 | 160 | 385 | 3.1 |
9 |
![]() |
5.3 | 7.5 | 130 | 382 | 3.1 |
10 |
![]() |
6.2 | 7.9 | 50 | 396 | 3.4 |
11 |
![]() |
5.6a | 7.5a | 79 | 412 | 3.1 |
12 |
![]() |
<4.6 | 7.0 | >250 | 400 | 3.2 |
13 |
![]() |
5.1a | 7.2 | 130 | 417 | 4.2 |
14 |
![]() |
6.0 | 7.9 | 79 | 396 | 3.4 |
15 |
![]() |
6.2 | 7.9 | 50 | 417 | 4.2 |
16 |
![]() |
5.8a | 7.5 | 50 | 400 | 3.7 |
17 |
![]() |
6.0 | 7.7 | 50 | 396 | 3.3 |
![]() | ||
Fig. 1 Compound 6 (green) docked into the 8BQ4 structure of PI5P4Kγ–1 (cyan). |
Indeed, a range of heterocycles was tolerated at the 6-position of the thienylpyrimidine, with PI5P4Kγ+ activity being retained when compared to 5 and many showed some level of PI5P4Kα inhibition as well as a reduction in Xlog
P (Table 2). Pyrazoles 7 and 8 retained good levels of PI5P4Kγ+ inhibition whilst showing some PI5P4Kα inhibition (Table 2). Pyridine 9 showed higher selectivity towards PI5P4Kγ+ than 6 (160 vs. 63-fold, respectively). The relative PI5P4Kα activity of 6 could be increased by methylation of the pyridyl 3-position (10) more effectively than by addition of OMe (11), or F (12) at the same position, or by chlorination at the pyridyl 2-position (13; Table 2). A range of small substituents was also trialled on template 6 and these were generally well tolerated (see 14–17). Compound 17 was a useful lead from this set with a good balance of activity at PI5P4Kα and γ+ as well as one of the lower X
log
Ps in this series. Compound 17 was therefore selected as a template for further exploration around the sulfone group.
Fig. 1 shows that the arylsulfone of dual inhibitor 6 is predicted to overlay well with the corresponding group of the PI5P4Kγ-selective molecule 1 in the PI5P4Kγ structure (8BQ4). A question remained as to whether the corresponding group of PI5P4Kα-selective 3 (i.e. 3,4-dimethylphenyl) would transpose onto this thienylpyrimidine template (Fig. 2). Using 17 as the optimal compound from Table 2, the dimethylphenyl group of 3 was transposed to provide compound 18, which was docked into the 3 structure (8C8C, Fig. 2).
![]() | ||
Fig. 2 A) Compound 18 (orange) docked into the 8C8C structure of PI5P4Kα–3 (gold). The dimethylphenyl group of 3 overlays well with that of 18; B) 2D comparison of 3, 17 and 18. |
The rationally-designed compound 18 was found to have improved PI5P4Kα activity compared to 17 but diminished PI5P4Kγ+ activity (Table 3). An alternative potent PI5P4Kα inhibitor with a 2,4-dimethylphenyl group, “compound 25”,33 was also merged onto the thienylpyrimidine core to give the resulting compound 19. Both 18 and 19 are dual PI5P4Kα/γ+ inhibitors, as predicted. Indeed, the designed modifications increased the PI5P4Kα activity of this chemotype to the point that, relative to PI5P4Kγ-selective starting point 1, both 18 and 19 showed higher levels of PI5P4Kα than PI5P4Kγ+ inhibition (for reference, compound 1 showed no detectable inhibition of PI5P4Kα under the same assay conditions, Table 1).
Lipophilic substituents at R2 have previously been found to be important for PI5P4Kα inhibition.33 As such, a variety of other lipophilic groups was tested at this position (Table 3). Compounds 20 and 21 were modest dual inhibitors and an interesting SAR trend was observed for the series of small aliphatic groups for 22–24, which increased in both PI5P4Kα and PI5P4Kγ+ activity with increasing size of the para substituent. In agreement with the hypothesis that the polar sulfone of 17 was not favourable for PI5P4Kα inhibition, apolar sulphide 25 was synthesised. This molecule showed an expected increase in PI5P4Kα inhibition with a modest reduction in PI5P4Kγ+ inhibition to afford a potent dual inhibitor.
Following on from this, combinations of groups which had been productive at R1 and R2 were explored (Table 4). Owing to the preference for lipophilic groups at R2, it was desirable to use polar heterocyclic groups at position R1, where possible, to help offset Xlog
P. Nonetheless, these compounds were generally high in X
log
P, typically in the range 4–5 (compounds 26–30).
Substitution at the 5-position of the thienopyrimidine core was also investigated (Table 5). In general, this position was tolerant of a wide range of substituents from small lipophilic groups such as methyl (31) and chloro (32), to larger substituents such as a small ring (33) or a much larger group incorporating a polar amide moiety (34). A 5,6-disubstituted thiophene was also tolerated (35). However, these derivatives did not offer an advantage over 6-substituted thienylpyrimidines and were not pursued further.
The thiophene was also ‘reversed’ from thieno[2,3-d]pyrimidines to afford thieno[3,2-d]pyrimidine derivatives such as 36, which is a direct analogue of 2 but with the sulfur atom moved within the thiophene ring (Table 6). Compound 36 shows a higher PI5P4Kγ+ pIC50 than 2 (7.3 vs. 6.5, respectively) and both compounds had undetectable levels of PI5P4Kα inhibition. Matched pairs of molecules in which the regiochemistry of the thienopyridine was alternated while the R groups were kept constant showed that thieno[3,2-d]pyrimidine was generally the more active regioisomer. For example, compound 38 forms a matched pair with 24 with the former being slightly more active at both isoforms (PI5P4Kα 7.0 vs. 6.6, respectively, and PI5P4Kγ+ 7.3 vs. 7.0). Similarly, 39 forms a matched pair with 28, with 39 being slightly more active at both PI5P4K isoforms. The thieno[3,2-d]pyrimidine template was also used to explore further alternative lipophilic groups at R2 (40, 41) and a group from potent PI5P4Kγ inhibitor ‘compound 7’ (42).30
A selection of the more active compounds identified to this point was profiled for PI5P4Kβ activity as well as PI5P4Kγ WT activity and cell penetration in a PI5P4K-WT InCELL Pulse cellular target engagement assay (Table 7). Both assays have previously been described for screening PI5P4Kγ inhibitors.38 All compounds tested in the PI5P4Kβ ADP-Glo assay returned IC50s > 10 μM (Table 7). Furthermore, there is a good correlation between the pIC50s obtained in the PI5P4Kγ+ ADP-Glo assay and the PI5P4K-WT InCELL Pulse assay, confirming that these compounds are bona fide inhibitors of the wild-type form of PI5P4Kγ. In general, there is a slight drop in pIC50 in the cellular assay, possibly owing to the requirement for cell permeability and/or the non-specific binding of the compounds to cellular components; however, these compounds such as 10 and 39 are potent PI5P4Kγ inhibitors in live cells.
ADP-Glo pIC50 | InCELL Pulse pIC50 | |||
---|---|---|---|---|
PI5P4Kα | PI5P4Kβ | PI5P4Kγ+ | PI5P4Kγ-WT | |
5 | <5.3 | <4.6 | 7.4 | 6.6 |
8 | 5.2 | 4.9 | 7.4 | 6.7 |
10 | 6.2 | 4.9 | 7.9 | 7.7 |
15 | 6.2 | 4.9 | 7.9 | 7.1 |
17 | 6.0 | 4.7 | 7.7 | 7.3 |
37 | 6.5 | ND | 7.9 | 7.4 |
39 | 7.3 | ND | 8.1 | 7.3 |
Compounds with interesting profiles were selected for screening in further assays to assess ADMET properties with the aim of identifying potent dual inhibitors that would make suitable probes for applications in cellular experiments (Table 8). As such, cell permeability and efflux were measured in MDCK-MDR1 cells and aqueous solubility was measured at pH 7.4. To give an indication as to whether any of these probes may be suitable for in vivo testing, stability in mouse liver microsomes (MLMs) was also measured. In general, there was a wide range of values in each of the ADMET assays for this series of molecules. Efflux ratios varied from good (0.91) to very high (64). Cell permeabilities were generally good, ranging from low at 1.3 × 10−6 cm s−1 to high at 14.2 × 10−6 cm s−1. Solubilities ranged from low to moderate (1–30 μM). Half-lives in MLMs ranged from very poor to very good (0.94–383 minutes).
Efflux ratioa | P app (A2B, 10−6 cm s−1) | Solubilityb (μM) | MLM t½c (min) | |
---|---|---|---|---|
a Permeability and efflux ratios determined in MDCK-MDR1 cells. b Aqueous solubility at pH 7.4. c Mouse liver microsome stability (half-life). d Recovery low A2B; possible solubility or binding issues; accuracy of result questionable. | ||||
5 | 1.3 | 6.23 | 1 | 54.5 |
8 | 26.4 | 2.88 | 30 | 73.3 |
10 | 25.2 | 3.13 | 30 | 9.23 |
14 | 34.1 | 2.45 | 30 | 8.2 |
15 | 15.7 | 4.73 | 3 | 383 |
17 | 64.3 | 1.33 | 10 | 35.5 |
26 | 1.1d | 3.7d | 3 | 0.94 |
32 | 3.7 | 14.2 | 10 | 51.4 |
37 | 31.8 | 1.33 | 10 | 70.8 |
38 | 0.91 | 9.95 | 3 | 1.94 |
39 | 0.98 | 12.9 | 10 | 0.44 |
42 | 27.4 | 3.15 | 10 | 5.39 |
Compound 39 showed one of the most balanced profiles required for a cell-active probe: good permeability, no efflux and moderate solubility. This molecule shows good activity in the InCELL Pulse target engagement assay (PI5P4Kγ pIC50 = 7.3) and a good dual activity in the ADP-Glo assays showing similar activities for PI5P4Kγ+ and PI5P4Kα (fold-selectivity = 5). However, 39 has a very high turnover in MLMs and is unlikely to make a useful in vivo tool (MLM t½ = 0.44 min). Compound 39 (ARUK2007145) is therefore nominated as a PI5P4Kα/γ dual inhibitor probe molecule for cell experiments.
Other compounds of note are 32, which shows good ADMET properties in general, but is slightly less active than 39, particularly at PI5P4Kα, and the cluster of compounds 8, 15 and 37, which fall into the category of having a high MLM half-life and good PI5P4Kγ+ activity, but lower PI5P4Kα activity. All of these compounds have a 4-(methylsulfonyl)phenyl group at position R2. The polar methyl sulfonyl group is detrimental for PI5P4Kα activity but positively modulates microsomal stability, whereas more lipophilic groups in place of the methylsulfonyl group positively modulate PI5P4Kα activity but are metabolically liable. Nonetheless, compounds 8, 15 and 37, all have higher PI5P4Kα activities than previously reported compounds from this class30 and may make useful complementary probes.
![]() | ||
Fig. 3 A) Compound 14 (teal); B) compound 30 (green) docked to PI5P4Kγ structure (8BQ4). Compound 1 shown for reference in both (cyan). The ψ dihedral angle is indicated in orange, this changes from 4° for 1 and 14 to 170° for 30. |
Later in the program, liganded PI5P4Kα crystal structures became available which allowed further structural assessment. The 8C8C PI5P4Kα structure shows ligand 3 in a binding mode in which, compared to 1 in the 8BQ4 PI5P4Kγ structure, has the aminopyrimidine flipped so that N2 rather than N4 is making an interaction with the hinge (Val199NH in PI5P4Kα, Met206 in PI5P4Kγ; Fig. 4). In the PI5P4Kα structure (8C8C) the aniline NH of 3 is also interacting with the hinge (Asn98OD1), whereas in the PI5P4Kγ structure (8BQ4) the aniline NH of 1 is not near the hinge residues. Retrospective docking suggested that our dual PI5P4Kα/PI5P4Kγ ligands 14 and 18 can interact with PI5P4Kα in two different poses, one is 1-like (Fig. 5A) and the other 3-like (Fig. 5B). For most of the dual inhibitors discussed here, the binding mode for 14 (shown in Fig. 5B) is only accessible in a protein conformation where Leu230 and Pro231 are not near the active site (seen, for example, in the PI5P4Kα structure solved with BAY297 bound, pdb 6YM4),31 as these two residues clash with the R1 extension in the 8C8C protein structure. Docking favours the pose shown in Fig. 5B for thieno[2,3-d] pyrimidines with an R1 3-pyridine and an acceptor atom in R2, e.g.9, 14, 15, 16 and 17. This appears to be driven by interactions with Lys145 and Lys209. In contrast, the thieno[3,2-d]pyrimidines (Table 6) all adopt the pose shown in Fig. 5A in preference. This may stem from a lower conformational energy for the extended conformation shown in Fig. 5A for this core, or perhaps the thiophene C–H offers a better interaction with the side chain of Asn198 than the sulphur.
![]() | ||
Fig. 4 Compound 3 and PI5P4Kα protein (pdb 8C8C; gold) superimposed onto 1 and PI5P4Kγ protein (pdb 8BQ4; cyan), aligned using the hinge residues Thr196–Phe200 (numbering from PI5P4Kα sequence). This shows that the pyrimidine moiety does not bind in the same orientation in the PI5P4Kα and PI5P4Kγ ligands. PI5P4Kα residues are labelled in black and PI5P4Kγ residues in blue. |
![]() | ||
Fig. 5 A) The dimethylphenyl group of 3 bound to PI5P4Kα (pdb 8C8C; gold) overlays well with that of 18 (orange) as positioned by docking to the 8C8C structure; B) 14 (brown) docked to alternative PI5P4Kα structure pdb 6YM4 (grey); 3 (pdb 8C8C; gold) was transposed onto the 6YM4 structure by alignment. |
Interestingly, all of the ligands with good PI5P4Kα activities (pIC50 > 6.5), with the exception of 42, have a much better docking score in the PI5P4Kα–3 structure (8C8C) than the PI5P4Kα–BAY297 (6YM4) structure (see ESI† Table S1).
Binding of compounds to PI5P4Kγ in intact cells was assessed using an InCELL Pulse thermal stabilisation assay (DiscoverX) as described previously.38 PI5P4Kγ was cloned into the pICP vector (DiscoverX) to allow overexpression of the ePL-tagged target. Hek293 cells stably expressing ePL-tagged PI5P4Kγ were incubated with 25 nL of test compound in 100% DMSO in a black skirted PCR plate for 60 minutes at 38 °C. After incubation for 3 minutes at 46 °C, followed by cooling for 3 minutes at room temperature, 12 μL of EA-3 reagent (prepared as per the manufactures guidelines) was added to each well. The plate was then incubated for 60 minutes in the dark prior to luminescence reading on a Pherastar FSX plate reader (BMG Labtech).
ADME | Absorption, distribution, metabolism, excretion and toxicity |
ATP | Adenosine triphosphate |
GTP | Guanosine triphosphate |
MDCK-MDR1 | Madin Darby canine kidney-multidrug resistance mutation 1 |
MLM | Mouse liver microsomes |
M W | Molecular weight |
ND | Not determined |
P app | Apparent permeability coefficient |
PI5P4K | Phosphatidylinositol 5-phosphate 4-kinase |
QM | Quantum mechanics |
SAR | Structure–activity relationship |
WT | Wild-type |
X![]() ![]() | Log partition coefficient |
Footnotes |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3md00355h |
‡ These authors contributed equally. |
§ Present address: UK Dementia Research Institute, University of Cambridge, Island Research Building, Cambridge Biomedical Campus, Hills Road, Cambridge, CB2 0AH, UK. |
This journal is © The Royal Society of Chemistry 2023 |