A drug targeting only p110α can block phosphoinositide 3-kinase signalling and tumour growth in certain cell types

The three class-Ia PI3Ks (phosphoinositide 3-kinases; p110α/p110β/p110δ) and the sole class-Ib PI3K (p110γ) couple growth factor receptors and G-protein-coupled receptors to a wide range of downstream pathways [1–3]. These enzymes have different tissue distributions, difference in methods of activation and different kinetic properties [4–6], but they all use PtdIns(4,5)P₂ to produce PtdIns(3,4,5)P₃. The cellular levels of PtdIns(3,4,5)P₃ are tightly controlled by phosphatases, including PTEN (phosphatase and tensin homologue deleted on chromosome 10) which dephosphorylates PtdIns(3,4,5)P₃ back to PtdIns(4,5)P₂ [7,8]. The importance of this pathway in cancer is highlighted by the fact that defects in both the kinase and phosphatase activities are commonly observed in tumours. PTEN is a tumour suppressor gene whose function is commonly lost in tumours [7,8], whereas the PIK3CA gene, which codes for p110α, appears to be the most important form of PI3K involved in solid tumours as it is commonly mutated [9,10] or amplified [11] in such cancers. The mutations in PIK3CA mainly occur in two distinct regions of the gene. It is not fully understood how these mutations contribute to the development of tumours, but they do confer a modest increase in catalytic activity [12,13], are capable of inducing transformation of cultured cells [14–16] and are capable of inducing tumours in vivo [17,18]. However, evidence is emerging that the main two different hot spot mutations in PIK3CA represent functionally distinct oncogenic activities [12,13,19–23]. The full implications of PIK3CA gene amplification are not fully understood, but presumably act by increasing overall PI3K activity levels.

The identification of oncogenic mutations and amplifications in PIK3CA has spurred the development of a wide range of small molecule inhibitors targeting PI3K, with many of these currently in clinical trials [2,24,25]. Most of the compounds developed to date target multiple PI3K isoforms and related kinases such as mTOR (mammalian target of rapamycin). Compounds in this class show efficacy in inhibiting growth of cells in culture and xenograft models [2,24,25]. However, a question that remains to be answered is whether selectively targeting p110α might achieve similar results given that this seems to be the predominant oncogenic form of class-I PI3Ks.

The potential importance of targeting p110α is shown by studies showing specific genetic knockdown of PIK3CA does block cell signalling and cell growth in a range of tumour lines [26–28]. To date the lack of suitable small-molecule inhibitors has meant that it has not been possible to properly evaluate whether pharmacological inhibition of p110α can achieve similar effects. Only one series of small molecules has been described that has a high degree of selectivity for p110α compared with other PI3K isoforms [29]. One member of this family, PIK-75, has been used to study the role of p110α, but was found to have significant off-target activity [30], meaning it is difficult to know whether any actions of this drug are in fact due to its activity against PI3K. Despite these limitations, this drug has been used in some studies to infer that blocking p110α is sufficient to block signalling to Akt/PKB (protein kinase B) in some cell types but not others [28,31,32]. Furthermore, compounds related to PIK-75 have shown antitumour activity in vivo, hinting that p110α inhibition might be a useful therapeutic strategy [29,33]. However these findings cannot be confirmed until a suitably clean p110α-selective inhibitor is available.

In the present paper, we report the properties of A66, a compound that was recently found to be a potent p110α inhibitor [34]. We show that this compound is highly selective for p110α over other PI3Ks and has a high degree of specificity as it does not target other protein kinases tested. We use this to demonstrate that inhibition of p110α attenuates signalling in a subset of cell types that are characterized by having kinase domain mutations in PIK3CA, high p110α levels and high total class-1a PI3K activity. We go on to show that A66 has efficacy in retarding growth of tumours in in vivo xenograft models that use cell lines that were responsive in culture. These results show that inhibition of p110α alone has the potential to block growth factor signalling and reduce growth in a subset of tumours.

The S-enantiomer of A66 (Figure 1) was prepared as described in Patent WO 2009/080705 [35], except that 2-(tert-butyl)-4′-methyl-[4,5′-bithiazol]-2′-amine was converted into A66 in one-pot by addition of L-prolinamide directly to the intermediate imidazolide solution. Aqueous work-up followed by recrystallization from aqueous methanol gave A66 as a white solid with a 81% yield. ¹H NMR (400 MHz, CDCl₃) δ 8.56 (br s, 1H), 7.03 (s, 1H), 6.76 (br s, 1H), 5.65 (br s, 1H), 4.62 (d, J 8.0 Hz, 1H), 3.62 (m, 1H), 3.49 (m, 1H), 2.52 (s, 3H), 2.43 (m, 1H), 2.03–2.20 (m, 3H), 1.45 (s, 9H). LC-MS (APCI⁺) 394 (MH⁺, 100%). Analysis calculated for C₁₇H₂₃N₅O₂S₂: C, 51.89; H, 5.89: N, 17.80. Found C, 51.85; H, 5.84; N, 17.81. The R-enantiomer of A66 was synthesized in the same way, except that D-prolinamide was used. Compound SN34452 was prepared similarly using pyrrolidine. ¹H NMR (400 MHz, CDCl₃) δ 7.78 (br s, 1H), 7.02 (s, 1H), 3.48 (m, 4H), 2.54 (s, 3H), 2.00 (m, 4H), 1.45 (s, 9H). LC-MS (APCI⁺) 351 (MH⁺, 100%). Analysis calculated for C₁₆H₂₂N₄OS₂: C, 54.83; H, 6.33; N, 15.98. Found C, 55.10; H, 6.47; N, 15.94.

NVP-BEZ-235 was synthesized as described previously [31]. PIK-75, TGX-221 and IC87114 were obtained from Symansis. LY294002 and wortmannin were obtained from Sigma–Aldrich.

An energy-minimized model of A66 was generated using SKETCHER (SYBYL8.2, Tripos) and minimized using MAXMIN2 with the MMFF94s forcefield and MMFF94 charges. Minimization was performed using 1000 steps of step descents followed by conjugate gradients until convergence at the 0.05 kcal/(mol·A) level. A distance-dependent dielectric function was used with a dielectric constant of 80. The major tautomer at pH 7.4 was generated using ChemAxon software. Docking was performed using GOLDv5.0. The apo p110α structure (PDB code 2RD0) was prepared by stripping all water molecules and the addition of protons using SYBYL8.2, and side-chain orientations were modified according to the results of MolProbity [36]. The docking site was defined as an 18 Å (1 Å=0.1 nm) cavity centred on the Ile⁸⁰⁰ CD1 atom. The Chemscore fitness function with kinase modification was used as the scoring function and 20 Genetic Algorithm runs were performed using a search efficiency of 200% with all poses were kept. Atom types for both protein and ligand were generated automatically and all ligand flexibility terms were turned on, although Ring-NH₂ and Ring-NR1R2 were set to flip, other settings were kept at default. All docking poses were minimized and rescored using the kinase-modified Chemscore with receptor depth scaling implemented. X-ray crystal structures for p110γ (PDB code 2CHZ) and p110δ (PDB code 2WXR) were superimposed on to p110α using PyMOL (http://www.pymol.org) and docking was performed under the same conditions with the 18 Å cavity centred on the CD1 of Ile⁷⁴⁴ and Ile⁷⁷⁷ respectively.

IC₅₀ values were evaluated using the PI3K (human) HTRF Assay (Millipore, #33-016). p85α/p110δ was obtained from Invitrogen. All other isoforms were produced in-house by co-expressing full-length human p85α with the indicated human full-length catalytic subunit containing a histidine tag at the N-terminus to allow purification. The PI3Ks were titrated and used at a concentration between their EC₆₅–EC₈₀ values. PI3K activity in immunoprecipitates was assayed as described previously [5] using an antibody to the N-SH2 (N-Src homology 2) domain of p85α (Symansis). Assays for other lipid kinases and protein kinases were performed by the National Centre for Protein Kinase Profiling (Dundee, U.K.) and Invitrogen Drug Discovery Services (Madison, WI, U.S.A.).

All animal experiments followed protocols approved by the Animal Ethics Committee of The University of Auckland. Age-matched specific pathogen-free male CD-1 mice were administered a single dose of A66 (10 mg/kg of body weight) in 20% 2-hydroxypropyl-β-cyclodextrin in water or BEZ-235 in 15% (v/v) DMSO, 20% (v/v) 0.1 M HCl, 0.7% Tween 20 and 64.3% (v/v) saline. Mice were killed at five or six time points after dosing (n=3/time point) and blood was removed by cardiac puncture into EDTA collection tubes (Becton Dickinson). Blood samples were centrifuged for 10 min at 6000 rev./min at 20 °C and the plasma supernatant was retained. Methanol was added to the plasma for protein extraction. Quantitative analysis was performed on an Agilent 6460 triple quadrupole LC-MS/MS (tandem MS) (Agilent Technologies) using multiple reaction monitoring and electrospray ionization. For chromatographic separation, an Agilent Zorbax SB-C₁₈ column (2.1 mm×50 mm; 5 μm) was used with a mobile phase gradient of 20–100% methanol in 0.1% formic acid and 5 mM ammonium formate at a flow rate of 0.4 ml/min. Plasma drug concentrations were quantified against a calibration curve of known drug concentrations ranging from 10 to 10000 nM, with quality controls included at 65, 650 and 6500 nM. To prevent contamination from previous samples, a methanol slug was run between each plasma sample. Pharmacokinetic parameters were determined by noncompartmental analysis using WinNonlin 5.3 software (Pharsight).

Treatment of cells with drugs and Western blotting was performed as described previously [31]. All antibodies for Western blotting were from Cell Signaling Technologies. Melanoma cell cultures were established and genotyped in-house. Established cell lines were obtained from A.T.C.C. and genotypes for cell lines were assigned on the basis of data from the COSMIC database (http://www.sanger.ac.uk/genetics/CGP/CellLines/).

Age-matched specific pathogen-free Rag1^−/− or NIH-III mice were subcutaneously inoculated on the right flank with 5×10⁶ U87MG, SK-OV-3 or HCT-116 cells in PBS. Tumour diameter as measured by electronic calipers was used to calculate tumour volume (mm³) based on the formula (L×w²)×π/6 (where L=longest tumour diameter and w=perpendicular diameter). A66 was administered in 20% 2-hydroxypropyl-β-cyclodextrin (Sigma–Aldrich) in water, whereas BEZ-235 was administered in 10% ethanol. Control mice were administered the A66 dosing vehicle alone. The drugs were dosed by intraperitoneal injection as the free base equivalent at a dosing volume of 10 ml/kg of body weight. For tumour pharmacodynamic studies, mice were administered a single dose of A66 or the control vehicle when tumours reached approximately 8–9 mm in diameter. Animals were killed 1 or 6 h after dosing and the tumours were removed, biopulverized and assayed for protein concentration (BCA protein assay; Sigma–Aldrich). For antitumour efficacy studies, dosing began when tumours were well established, averaging approximately 7 mm in diameter. Doses were administered once daily (QD) or twice daily (BID) with injections separated by a minimum of approximately 8 h. Different dosing schedules were used for the three xenograft models depending on the rate of tumour growth and the body weight tolerance of control mice. Animals were dosed daily for 21 days or twice daily for 16 days (SK-OV-3), daily for 14 days (U87MG) and daily for 7 days (HCT-116). Animals were monitored daily for any signs of emerging toxicity and body weight was recorded. Mice were killed if they developed moderate signs of toxicity or if body weight loss exceeded 20% of starting weight. TGI (tumour growth inhibition) was calculated on the final day of dosing by determining the relative tumour size of drug-treated mice as a percentage of the average relative tumour size of control mice. The statistical significance of TGI values was determined by one-way ANOVA with Bonferroni multiple comparison analysis using GraphPad Prism 5.02.

We first characterized A66 and confirmed it was a potent inhibitor of the wild-type and oncogenic forms of p110α but not other class-I PI3K isoforms (Table 1). We found A66 has a much greater degree of selectivity for p110α than PIK-75. Given the important roles of class-II PI3Ks [37], class-III PI3K [38] and PI4Ks (phosphoinositide 4-kinases) [39] in growth factor signalling, we also assessed the activity of A66 towards these and found some limited cross-reactivity with the class-II PI3K PI3K-C2β and the PI4Kβ isoform of PI4K (Table 2). There was no inhibition of other lipid kinases or the related kinases DNA-PK and mTOR (Table 2). We also tested the inhibitory effects of 10 μM A66 effects on two large panels of 110 protein kinases (Supplementary Figure S1 at http://www.BiochemJ.org/bj/438/bj4380053add.htm) and 318 kinases (Supplementary Figure S2 at http://www.BiochemJ.org/bj/438/bj4380053add.htm). These show A66 is a very specific inhibitor of p110α, whereas PIK-75, the compound described previously as a p110α-selective inhibitor, inhibited a large number of protein kinases at this concentration (Supplementary Figure S1). Our data for TGX-221 and IC87114 generated using the HTRF assay agreed with previous studies using other assay methods and confirmed these are highly selective inhibitors of p110β and p110δ respectively (Table 1), although TGX-221 will cross-react with p110δ at higher concentrations [30,31,40]. We report further that these inhibitors do not have any major effects on a panel of 110 protein kinases (Supplementary Fig 1).

A66 shares its central aminothiazole scaffold with the known PI3K inhibitor PIK-93, and the X-ray crystal structure of PIK-93 bound to the related p110γ isoform (PDB code 2CHZ) shows that the embedded hydrogen-bond donor acceptor group in this core interacts with the kinase domain through backbone amide and carbonyl groups of the inter-lobe linker region amino acid Val⁸⁸² [30]. The aminothiazole unit in A66 may also influence its interaction with p110α by binding similarly, this is in part supported by the inhibition of PI4K by both these compounds [41]. The availability of the p110α X-ray crystal structure [42] allowed modelling of A66 in the p110α kinase domain and the likely mechanisms for its selectivity towards this compound were identified. The top-ranked binding mode for the A66 S form docked into the p110α ATP-binding site (PDB code 2RD0), after minimization and rescoring with the kinase modified Chemscore scoring function using receptor depth scaling, is shown in Figure 2. Critically in this predicted binding mode, the ligand forms an interaction with Val⁸⁵¹ of the inter-lobe linker region. Both the backbone amide and carbonyl of Val⁸⁵¹ interact with the hydrogen bond donor and acceptor nitrogen atoms embedded in the central aminothiazole core, consistent with the binding mode observed for PIK-93 bound to p110γ [30]. The tertiary butyl-thiazole moiety extends from the amino-thioazole core into the lipophilic affinity pocket, whereas the pyrrolidine carboxamide group extends in the opposite direction towards a region of the binding site wall defined by the C-terminal lobe that contains p110α-specific residues, known to affect ligand binding [43]. In this predicted binding pose, the carboxamide amine moiety forms hydrogen bonds with the side-chain carbonyl group of Gln⁸⁵⁹ and possibly the backbone carbonyl group of Ser⁸⁵⁴ (Figure 2). Notably, the unminimized pose predicted a hydrogen bond interaction between both the carboxamide amide and carbonyl groups of the ligand and those in the Gln⁸⁵⁹ side chain. These residues were predicted previously to be involved in inhibitor interactions in the p110α active site [44].

We also investigated possible binding modes for the A66 R form, and observed that a pose similar to that of the S form was not found, and it failed to form a hydrogen bond interaction with the backbone amide of Val⁸⁵¹ as well. In the top ranked pose, the R pyrrolidine carboxamide amino group was predicted to form a hydrogen bond with the Val⁸⁵¹ backbone carbonyl. In this orientation, the ligand's central urea carbonyl was predicted to interact with the side-chain amino group of Gln⁸⁵⁹ and also the affinity pocket was not occupied. Interestingly, a few clashes between the protein and ligand were observed with the S form, whereas more were present for that of the R form. These results, taken together along with the higher Chemscore fitness value (S, 17.46 compared with R, 13.26), indicate that the A66 S form appears to complement better the p110α active site. In agreement with this, we find that the A66 R form lost inhibitory activity and did not inhibit p110α at 10 μM (Table 1).

Superimposition of the p110γ and p110δ kinase domains on to that of p110α illustrates that in p110γ Lys⁸⁹⁰ replaces the p110α Gln⁸⁵⁹, whereas in p110δ Asn⁸³⁶ is the equivalent residue. As both these amino acid side-chains have hydrogen bond donor and acceptor groups that may interact with the ligand's pyrrolidine carboxamide, A66 was docked into the p110γ structure with the aminothiazole containing PIK-93 bound (PDB code 2CHZ) and the p110δ apo enzyme structure (PDB code 2WXR). The top ranked pose predicted for A66 binding to p110γ had a similar orientation to that predicted with p110α; however, the Chemscore fitness value was much lower, indicating a worse fit (Chemscore 7.29 compared with 17.46). An interaction with the p110γ Val⁸⁸² backbone amide (the equivalent residue to p110α Val⁸⁵¹) was also not predicted, even though the PIK-93 aminothiazole forms this interaction. No interaction was observed with Lys⁸⁹⁰ and, moreover, the pyrrolidine group clashed with Trp⁸¹² on the N-terminal lobe wall of the active site, further supporting poor complementarity of this compound with p110γ. A low Chemscore value was also recorded for the top ranked pose in the p110δ active site (8.44 compared with 17.46), indicating a poor fit in this isoform. In the present study, no pose was found that was similar to those predicted in either the p110α apo structure or p110γ PIK93 structure, and an interaction with the backbone amide of the p110α Val⁸⁵¹ equivalent, Val⁸²⁸, was not observed. Neither was an interaction with Asn⁸³⁶. The lack of similarity between the binding mode predicted for p110α and those for p110γ and p110δ suggest that other active site features, more than residue substitutions at Gln⁸⁵⁹, may influence A66 binding. On the basis of the preferred binding model of A66 in p110α, we characterized the role of the carboxamide by docking an unsubstituted pyrrolidine derivative SN34552 (Figure 1). The binding mode was similar to that of A66, although the Chemscore was much lower in the absence of the predicted carboxamide-mediated hydrogen bonds (8.44 compared with 17.46), suggesting reduced potency. This was supported by biochemical data which showed that SN34452 has a much lower potency against p110α (Table 1) and clearly indicates that the pyrrolindine carboxamide group makes p110α-specific contacts that are accessible in both the wild-type and oncogenic forms. Interestingly, SN34452 largely retains its potency against PI4K TypeIIIβ (Table 2), which indicates the carboxamide is not critical for binding to this enzyme.

To investigate the role of p110α in regulating proximal elements of PI3K-dependent signalling pathways, we determined the ability of various concentrations of the A66 S form to acutely block the activation of Akt/PKB in a range of cell lines as assessed by both phosphorylation of Ser⁴⁷³ and Thr³⁰⁸ (Figure 3). Loading was controlled for by reprobing for total PKB (see Supplementary Figure S3 at http://www.BiochemJ.org/bj/438/bj4380053add.htm). We found that phosphorylation of both Ser⁴⁷³ and Thr³⁰⁸ is sensitive to LY294002 in all cell lines tested, implying that class-I PI3K activity is required for activation of Akt/PKB. However, we found the amount of the A66 S form required to inhibit phosphorylation of Ser⁴⁷³ and Thr³⁰⁸ followed two distinct patterns, being either sensitive to inhibition by the A66 S form at concentrations consistent with it acting through p110α or being resistant. The most obvious feature of the sensitive cell lines was that they harboured H1047R mutations in PIK3CA, whereas all other cell lines were resistant. As a control we tested the effect of the A66 R form and found it was not able to inhibit the phosphorylation of Akt/PKB (see Supplementary Figure S4 at http://www.BiochemJ.org/bj/438/bj4380053add.htm).

We then went on to investigate this in more detail in a larger panel of cell lines. The broader panel included early passage melanoma cultures that were genotyped in-house (NZM lines) [45]. The latter are likely to have characteristics more representative of real tumour tissue. In these studies we used PIK-75 as an alternative p110α inhibitor and we found that a low concentration of PIK-75 also blocks the insulin-stimulated phosphorylation of Thr³⁰⁸ and Ser⁴⁷³ on Akt/PKB in all lines harbouring PIK3CA H1047R mutations (Figure 4A). TGX-221 and IC87114 had no effect in these cells.

PIK-75 was less effective in cell lines that lack the H1047R mutation (Figure 4B), with the main exception being MCF7 cells, where both PIK-75 and TGX-221 had a partial inhibitory effect. In other cells, the activation of Akt/PKB was not inhibited by TGX-221 or IC87114 at concentrations at which they would be specifically inhibiting p110β or p110δ respectively. However, in these cells, the combination of PIK-75, TGX-221 and IC87114 together did block activation of Akt/PKB, which was consistent with the finding that wortmannin and LY294002 were also effective.

To further understand why certain cell lines are sensitive to p110α inhibitors, we compared total levels of class-Ia PI3K activity in the eight cell lines used in Figure 3. The cell lines that were responsive to the p110α inhibitors have significantly higher total levels of PI3K (Figures 5A and 5B). We next compared total levels of p110α and p110β protein in the cell lines used (Figure 5C). The levels of p110α were highest in the cell lines that were responsive to A66 and PIK-75. These cells also had levels of p110β that were higher than the other cell lines, with the exception of MCF7 cells which also had high levels of p110β. It is of note that the MCF7 cells were the only cell line that had a partial response to TGX-221 (Figure 4B) and this may relate to the ratio of p110β/p110α in these cells.

To investigate whether the inhibitory effects of A66 S on activation of Akt/PKB signalling translated into the ability to block cell growth in vivo, we performed xenograft studies alongside the well-established pan-PI3K inhibitor BEZ-235 in U87MG cells, which are PTEN-null, and HCT-116 and SK-OV-3 cells, both of which contain H1047R mutations. First, we determined the optimal dosing strategy for xenograft studies by investigating the drug pharmacokinetics after a dose of 10 mg/kg of body weight by intraperitoneal injection in CD-1 mice. Despite a short half-life of only 0.42 h, the large C_max (8247 nM) of A66 S that was reached 30 min after dosing ensured that the AUC_0-inf (area under the curve from zero time to infinity) (6809 nM·h) was similar to that of BEZ-235 (7333 nM·h), which has a longer half-life of 2.73 h (Table 3). Furthermore, we tested the effect of the A66 S form on SK-OV-3 tumour tissue in vivo using a single dose of 100 mg/kg of body weight to determine whether a long-lasting effect of the drug could be achieved on target tissues (Figure 6). These studies show that A66 S causes a profound reduction in the phosphorylation of Akt/PKB and p70 S6 kinase, but not of ERK (extracellular-signal-regulated kinase), at both 1 and 6 h after dosing (Figure 6). This is consistent with A66 S having a full inhibitory effect on PI3K signalling in the tumours during this time. In the present study, levels of A66 S in plasma were determined to be 21.1±1.2 μM and 9.1±1.1 μM at 1 and 6 h after drug injection, whereas levels of A66 S in the tumour were 22.7±2.1 μM and 16.0±1.3 μM at the same time points. Thus, the retention of drug in the tumour is likely to explain the persistence of the inhibitory effect.

On the basis of the pharmacokinetic and pharmacodynamic findings, A66 S was dosed QD at 100 mg/kg of body weight for up to 21 days or BID at 75 mg/kg of body weight for 16 days in tumour efficacy studies. Both dosing strategies induced a significant delay in growth of SK-OV-3 xenografted tumours, which was even greater than that induced by the well-established pan-PI3K inhibitor BEZ-235 (Figure 7A). At the final day of dosing, the average TGI for A66 S form was 45.9% of control (QD; P<0.05) and 29.9% of control (BID; P<0.01) (Table 4). QD A66 S was well tolerated in this xenograft model with minimum body weight loss; however BID treatment was associated with moderate body weight loss and two deaths, although it is not clear whether the deaths were due to drug toxicity or other causes since these mice did not show significant body weight loss (Figure 7B). In comparison, BEZ-235 induced a non-significant reduction in tumour growth and was even less tolerated, with moderate body weight loss and four deaths. QD dosing of A66 S in an HCT-116 xenograft model also induced a significant reduction in tumour volume with a TGI of 77.2% of control (P<0.01) at the end of dosing, but caused a non-significant reduction in tumour volume in the U87MG xenograft model (Table 4 and Supplementary Figure S5 at http://www.BiochemJ.org/bj/438/bj4380053add.htm). In contrast, BEZ-235 significantly reduced U87MG tumour growth (TGI=61.1% of control; P<0.05), but had no effect on HCT-116 tumours. The drugs were well tolerated in both the U87MG model, despite the toxicity with the same dose level of BEZ-235 in the SK-OV-3 study, and in the HCT-116 model, where a lower dose (10 mg/kg of body weight) of BEZ-235 was used due to the moderate body weight loss of control-treated mice.

The present study demonstrates that A66 S is a highly specific and selective inhibitor of p110α that is suitable for in vitro and in vivo studies. The contacts made by the carboxamide group give A66 S its potency and selectivity for p110α but, interestingly, it does inhibit PI4K IIIβ at concentrations approximately one order of magnitude higher. This is not surprising given the degree of homology between these enzymes in the catalytic sites [41]. However, SN34452 retains this activity against PI4K IIIβ when the carboxamide is removed (Table 2), which makes this one of the more selective PI4K IIIβ inhibitors described to date [41]. The other is PIK-93, which is structurally quite different from A66 apart from sharing an amino thiazole core, but it also inhibits both p110α and PI4K IIIβ, again highlighting the similarities in the catalytic site of these two enzymes [41].

Our results confirm previous studies that highlight the limitations of using PIK-75 and related compounds [30]. However, PIK-75 can still play a useful role as a backup for confirmatory experiments and it is worth noting that PIK-75 complements A66 in that it does not inhibit the two non-p110α lipid kinases that A66 targets, i.e. PI4K IIIβ and PI3K-C2β. Our studies also add extra weight to the case for TGX-221 and IC87114 being considered as highly selective inhibitors of p110β and p110δ if used at suitable concentrations.

The finding that A66 S potently blocks phosphorylation of Akt/PKB in a subgroup of the cell lines tested demonstrates that some cell types are highly dependent on p110α activity. This is consistent with genetic studies which show that knockdown of p110α blocked signalling to Akt/PKB in cell lines harbouring mutations in PI3K [26,27]. It also supports previous studies using PIK-75 [28,31,33] and A66 [34] and suggests at least some cell types are more sensitive to p110α inhibitors. The finding that TGX-221 and IC87114 alone do not inhibit the phosphorylation of Akt/PKB at Ser⁴⁷³ or Thr³⁰⁸ in any of the cell lines tested, with the exception of a partial effect of TGX-221 in MCF7 cells, indicates that this pathway is not reliant on the catalytic activities of p110β and p110δ in most cells. The findings with regard to p110δ are not unexpected, but the findings with TGX-221 are somewhat at odds with some previous studies. Although no oncogenic mutations have been found in p110β, overexpression of p110β is capable of inducing transformation [46]. Knockdown of PIK3CB has been shown to block the ability of PTEN-deficient cell lines to form foci in in vitro transformation assays [26,47,48] and in in vivo tumour models. The knockdown of PIK3CB has also been reported to result in a small reduction in Akt/PKB phosphorylation in PTEN-deficient cells [26,47,48]. Although some functions of p110β appear to be independent of its lipid kinase activity [49], the finding that TGX-221 blocks signalling to Akt/PKB in PTEN-deficient cells has been taken as evidence that the catalytic activity of p110β is required in this context [47,50]. However, it should be pointed out that those studies used a 20- to 100-fold higher concentration of TGX-221 than those used in the present study, which would provide for a significant opportunity for cross-reactivity with other PI3K isoforms. In support of this, Wee et al. [26] found that 2 μM TGX-221 was required to induce reduction in Akt/PKB activation in PTEN-deficient cell lines, but that at these concentrations also partially reduced activation of Akt/PKB in the DLD1 cell line that harbours a PIK3CA mutation. This would be consistent with our results from the present study which demonstrate that binary combinations of A66 S, TGX-221 and IC87114 induce varying degrees of partial inhibition of activation of Akt/PKB, whereas the combination of all three drugs induced maximal inhibition. This indicates that the three class-Ia PI3K isoforms are functionally redundant to some extent and can substitute each other in signalling to Akt/PKB in these PTEN-null cells, as has been observed previously in other cell types [31,51,52].

In the present study, activation of Akt/PKB was sensitive to p110α inhibitors in H1047R cells but not in PTEN-null cell lines and those harbouring E545K mutations, which is in agreement with the studies of Torbett et al. [32] who used PIK-75. It would be tempting to conclude that the sensitivity to p110α inhibitors is a direct consequence of the presence of the H1047R mutation, since this isoform has increased catalytic activity [13]. However, the PIK3CA mutants are not intrinsically sensitive to A66 or PIK-75 (Table 1), and gene knockout studies have shown that sensitivity of HCT-116 cells to p110α-selective PIK-75 analogues is not changed by deletion of the H1047R allele of PIK3CA [33]. Furthermore, the study by Torbett et al. [32] showed that MCF10A cells and Hs578t cells were also sensitive to PIK-75. The latter can be explained by the fact that this line was subsequently found to have a mutation in PIK3R1 (Cosmic database) and such mutations have been shown to be sensitive to p110α inhibitors [34]. Although MCF10A cells have no reported mutations in PI3K signalling pathways, a certain sub-population of these cells has been reported to have high PI3K activity [53]. This is consistent with another study which observed PI3K is not mutated in medulloblastoma, but that p110α is overexpressed and that such cells are very sensitive to PIK-75 [22]. Furthermore, we have observed previously in other cells that the degree of PIK-75 sensitivity is proportional to the relative amount of the total PI3K activity that is attributable to p110α [31]. Our results from the present study also show that the cells with high overall class-Ia PI3K and p110α protein levels are the ones that are sensitive to p110α inhibitors. Therefore the increased catalytic activity of the H1047 mutant may not be sufficient on its own to confer sensitivity to p110α inhibitors, but rather it may be the total levels of p110α (mutant or wild-type) in the cells that is most important. In this regard it is worth noting that evidence has recently been presented to indicate that at least part of the effect of the H1047R mutant might be to stabilize p110α levels in the cell [53]. The relative levels of class-Ia PI3K isoforms is also likely to be important and it is in this regard it is noteworthy that MCF7 cells are partially responsive to TGX-221, suggesting a dependence on p110β, and this cell line is the only one where we found high p110β and low p110α levels. Further studies will be required to clarify these issues.

The reason for the difference in characteristics between the H1047R and E545K cell lines is not clear. However, a number of studies have indicated that these two main oncogenic forms of p110α are likely to function differently in vitro and in vivo [12,13,19–23]. In particular, the helical domain mutants appear to signal independently of the p85 adapter subunit, and hence of activation by receptor tyrosine kinases, but require Ras [54]. The kinase domain mutants, on the other hand, require p85 but are independent of Ras [54]. Again it will require further studies to clarify this issue.

The finding that A66 S is more effective at inducing growth delay in the HCT-116 and SK-OV-3 xenograft models than the pan-PI3K/mTOR inhibitor BEZ-235 [55] demonstrates that a p110α-selective inhibitor can be effective at slowing cell growth in the absence of mTOR inhibition in certain cell types. In addition, although A66 S did not induce tumour regression in xenograft models, the ability to induce growth delay indicates p110α-selective inhibitors have to ability to be effective as cytostatic agents in some tumour types. Further studies will be required to determine whether A66 might contribute to tumour regression as part of a combination drug treatment strategy.

BID

twice daily dosing

ERK

extracellular-signal-regulated kinase

mTOR

mammalian target of rapamycin

PI3K

phosphoinositide 3-kinase

PI4K

phosphoinositide 4-kinase

PKB

protein kinase B

PTEN

phosphatase and tensin homologue deleted on chromosome 10

QD

once daily dosing

TGI

tumour growth inhibition

Stephen Jamieson, Jack Flanagan, Bruce Baguley and Peter Shepherd developed concepts, and designed and supervised experiments. Jackie Kendall, Gordon Rewcastle and William Denny designed and performed the chemical synthesis. Stephen Jamieson, Jack Flanagan, Sharada Kolekar, Christina Buchanan, Jackie Kendall, Woo-Jeong Lee, Gordon Rewcastle, Ripudaman Singh and James Dickson performed experiments. Peter Shepherd, Jack Flanagan, Gordon Rewcastle and Stephen Jamieson were the main contributors in the writing of the paper. Sharada Kolekar, Christina Buchanan and William Denny made minor contributions in the writing of the paper.

We thank Aaron Thompson and Phil Kestell for their help with tumour xenograft and pharmacokinetic studies, and Dr Graham Atwell for synthesizing the A66 R form.

This work was funded by the Health Research Council of New Zealand [grant number 09-388], the Maurice Wilkins Centre for Molecular Biodiscovery and the Auckland Cancer Society.