Chemical Interrogation of FOXO3a Nuclear Translocation Identifies Potent and Selective Inhibitors of Phosphoinositide 3-Kinases*

Activation of the phosphoinositide 3-kinase (PI3K)/Akt signaling pathway is one the most frequent genetic events in human cancer. A cell-based imaging assay that monitored the translocation of the Akt effector protein, Forkhead box O (FOXO), from the cytoplasm to the nucleus was employed to screen a collection of 33,992 small molecules. The positive compounds were used to screen kinases known to be involved in FOXO translocation. Pyrazolopyrimidine derivatives were found to be potent FOXO relocators as well as biochemical inhibitors of PI3Kα. A combination of virtual screening and molecular modeling led to the development of a structure-activity relationship, which indicated the preferred substituents on the pyrazolopyrimidine scaffold. This leads to the synthesis of ETP-45658, which is a potent and selective inhibitor of phosphoinositide 3-kinases and demonstrates mechanism of action in tumor cell lines and in vivo in treated mice.

The phosphoinositide 3-kinase (PI3K) 4 /Akt pathway is activated in a variety of solid and non-solid tumors (1) and therefore is considered as a potential intervention point for anticancer therapeutics. Activation of the pathway is frequently caused by mutations in PI3K␣ that enhance its catalytic activity, leading to the generation of phosphatidyl 3,4,5-trisphosphate (PIP3) (2) or by mutations or deletions in the tumor suppressor PTEN (phosphatase and tensin homolog) that result in its loss of function. PTEN antagonizes the activity of PI3K␣ through the dephosphorylation PIP3 (3). In addition, PI3K␣ can be activated by mutations in certain receptor-tyrosine kinases as well as by mutations in the oncogene KRAS (4,5).
The PIP3 generated by activation of PI3K␣ or sustained by the inactivation of PTEN binds to a subset of lipid-binding domains in downstream targets such as the pleckstrin homology (PH) domain of the oncogene Akt (6,7); thereby, recruiting it to the plasma membrane. Once at the plasma membrane, Akt can be activated (8,9). When active, Akt phosphorylates several effector molecules including the Forkhead box O (FOXO) transcription factors (10,11). FOXO proteins are a family of conserved polypeptides that bind to DNA as a monomer and activate the transcription of genes that are involved in numerous biologically relevant processes such as metabolism, differentiation, proliferation, longevity, and apoptosis (12,13). Akt phosphorylates FOXO proteins at three conserved consensus sites, which leads to conformational changes that facilitate CRM-1mediated nuclear export (14,15). Nuclear FOXO proteins function as regulators of transcription, whereas cytoplasmic FOXO proteins are considered inactive. It is well established that FOXO is negatively regulated by various proliferative and antiapoptotic signaling pathways that activate the PI3K/Akt signaling cascade (11). Therefore, we chose to employ a high content imaging approach to monitor the nucleocytoplasmic translocation of a GFP-FOXO3a fusion protein in U2OS cells (U2foxRELOC) (16,17) as the readout for biological inhibition of PI3K/Akt signaling. The rapid kinetics of the assay allowed us to reduce the incubation time and minimize possible toxic effects that might interfere with the analysis. Furthermore, this image-based high-throughput strategy provides a filter for adequate solubility, permeability, and stability in a cellular context and enables compounds that produce artifacts or cytotoxicity to be identified on a single cell basis. Often a key limitation of cell-based screening approaches is the identification of the molecular target of the compound. For this reason, the cellbased screen was followed by a focused screen of kinases thought to be involved in the regulation of the intracellular localization of FOXO proteins. This screen identified pyrazolopyrimidine derivatives as inhibitors of PI3K␣. Finally, a combination of computational and synthetic medicinal chemistry was used to optimize the chemical features required for activity. Here we report the discovery of a novel series of PI3K inhibitors discovered by cellular high content screening that are potent, selective, and demonstrate mechanism of action in vivo.
Different but complementary criteria were considered for compound selection and purchase: (a) chemistry perspective: drug-like compounds and (b) biological point of view: structures focused on kinases (kinases libraries) as well as a diverse set of compounds exploring different chemical space with a potential biological activity.
Compounds were purchased from ChemDiv and Life Chemicals. Their structure and purity (Ͼ90%) have been validated by NMR and/or LCMS. The compounds that have been synthesized in-house are at least 95% pure by LCMS and NMR (see supplemental information for details).
Plasmids-The GFP-FOXO3a fusion protein was kindly provided by T. Finkel. The U2nesRELOC assay uses the reporter construct pRev MAPKKnes GFP, which has been described earlier (18,19). pRev MAPKKnes GFP carries the nuclear export signal from MAPKK cloned between the BamHI and AgeI sites of pRev(1.4)-GFP, sandwiched between the Rev and the GFP-coding sequences.
U2foxRELOC and U2nesRELOC Assay-The U2nesRELOC assay and the U2foxRELOC assay have been described previously (16,19). Briefly, cells were seeded at a density of 1.0 ϫ 10 5 cells/ml into black wall clear bottom 96-well microplates (BD Biosciences) after 12 h of incubation at 37°C with 5% CO 2 , 2 l of each test compound were transferred from the mother plates to the assay plates. Cells were incubated in the presence of the compounds for 1 h. Then cells were fixed, and the nucleus stained with DAPI (Invitrogen). Finally the plates were washed with 1ϫ phosphate-buffered saline twice and stored at 4°C before analysis.
Kinase Assays-Detailed methods are described in supplemental information.
Docking-The Gold 3.1 program (21) was used to carry out docking of LY294002, PI-103, and ETP-45658 5 to p110␥ with the crystal structure of p110␥ complexed with staurosporine Protein Data Bank (PDB) entry 1e8z (20). The binding site was defined using the available experimental information; thus, the docking region used was a 15-Å sphere around the nitrogen NH of Val-882. The GoldScore scoring function was used to rank docking poses, and protein hydrogen bond constraints for binding to NH of Val-882 were imposed on the ligand. Per each ligand, the top five best docked structures out of 40 independent genetic algorithm runs were retrieved. For validation purposes, we compared the data for LY294002 and PI-103 docked to p110␥ with the corresponding crystal structure of p110␥ for LY294002 PDB entry 1e7v (20) and the proposed binding mode for PI-103 as described by Knight et al. (21). Besides qualitative assessment and key interactions properly identified, root meansquare deviation (r.m.s.d.) is utilized as a quantitative validation criteria for the docking algorithm set-up; thus, r.m.s.d. between experimentally reported LY294002 and the best predicted pose are 2.3 Å (1e8z, cross-docked structure) and 1.5 Å when the crystal for LY294002 (1e7v) is used. For PI-103, only docking using 1e8z yielded a reliable binding mode similar to the one described by Knight et al. (21) in which the morpholine oxygen was seen to form a hydrogen bond to Val-882, and the phenol group is oriented into the hydrophobic cavity, by qualitative comparison. By superimposing the 1e7v and 1e8z structures, we observed that the side chain of Lys-833 is more extended than in 1e8z, entering into the cavity occupied by the phenol group of PI-103 in the binding mode obtained with 1e8z. We have hypothesized that this extended conformation of the side chain of Lys-833 underlies the docking failure of PI-103 in 1e7v crystal. Based on these validation results, the 1e8z crystal was finally selected as the protein structure for docking of ETP-45658.

Identification of FOXO Relocators-
The strategy used to identify biologically active inhibitors of PI3K/Akt signaling is outlined in Fig. 1A. As the primary filter, we used a cellular imaging assay that follows the intracellular location of FOXO proteins (16). U2OS osteosarcoma cells that stably express a green fluorescent protein (GFP)-tagged version of FOXO3a were seeded in 96-well assay plates and treated with compounds at a final concentration of 10 M for 1 h. A total of 33,992 compounds were screened for their ability to induce the translocation the GFP-FOXO3a reporter protein from the cytoplasm into the nucleus. The nuclear accumulation of fluorescence triggered by the pan-PI3K inhibitor, LY294002, was defined as 100% activity and used as a reference. Primary hits were defined as those compounds that have an activity greater than 60%. After retesting, 242 compounds were confirmed as positive yielding a hit rate of 0.71%. A sample image of an active compound is shown in Fig. 1B (upper right). To exclude compounds that could be general inhibitors of the nuclear export machinery rather than selective inducers of FOXO nuclear translocation, the ability of the compounds to influence the localization of a nuclear export signal (NES)-containing fluorescent reporter protein (19) was determined. Briefly, the nuclear export assay uses U2OS cells that stably express a GFPlabeled Rev protein, which contains a strong heterologous NES sequence. The fluorescent signal of untreated cells localizes exclusively to the cytoplasm. Upon treatment with a nuclear export inhibitor such as leptomycin B, the GFP-labeled reporter protein accumulates rapidly in the cell nucleus (Fig.  1B, middle right). 14 of the original 242 confirmed hits induced the nuclear localization of Rev protein and were classified as nonspecific inhibitors as inhibitors of the Crml-dependent nuclear export, an example of a positive compound is shown in the lower right panel of Fig. 1B. The remaining 228 confirmed positive in the FOXO translocation assay and that were negative (Fig. 1B, lower left panel) in the nuclear export assay were classified as FOXO relocators.
The FOXO relocators were then screened against protein kinases that are known to phosphorylate FOXO3a at sites involved in the regulation of its intracellular localization either directly, such as Akt1-3, Sgk 1-3, or indirectly by activating Akt, such as PI3K␣ and PDK1 (10,22). No inhibitors of PDK1, Akt1-3, or Sgk1-3 were identified. However, 54 compounds representing 24 distinct chemical classes, six of which had two or more representatives, were identified as biochemical inhibitors of PI3K␣.
Of the remaining 174 FOXO relocators that were not inhibitors of PI3K␣, 155 were screened against an extended panel of protein kinases that have been implicated in signaling networks that regulate FOXO proteins (supplemental Fig. S1). Five compounds nonspecifically inhibited several receptor-tyrosine kinases (RTKs) whereas no other significant inhibitory activity was observed. In summary, a total of the 242 FOXO relocators identified, 6% were nonspecific inhibitors of nuclear export, 21% were inhibitors of PI3K␣, 2% were nonspecific inhibitors of RTKs, and 71% were classified as inhibitors of an unknown target(s).
Pyrazolopyrimidines as Inhibitors of PI3K-Pyrazolopyrimidine derivatives were the most potent biochemical inhibitors of PI3K␣ identified from the primary screen (Fig. 2). The pyrazolopyrimidines had IC 50 values for inhibition of PI3K␣ ranging from 0.4 to 12 M ( Fig. 2A). In addition they induced clear nuclear FOXO translocation EC 50 values of between 0.54 and 26.6 M ( Fig. 2A). As expected, the compounds inhibited Akt phosphorylation at serine 473 in treated cells (Fig. 2B).
The pyrazolopyrimidine, ETP-7382, was the most potent biochemical inhibitor of PI3K␣ and the most potent inducer of FOXO nuclear translocation obtained from the primary screen ( Fig. 2A). Therefore, this compound was selected for further mechanism of action studies. Inhibition of signaling downstream of Akt was observed in cells treated with ETP-7382; both the Akt-dependent phosphorylation of FOXO3a on threonine 32 and mTOR-dependent phosphorylation of p70 S6K on threonine 389 was reduced (Fig. 2C). These observations confirmed the mechanism of action of ETP-7382 as an inhibitor of PI3K␣dependent signaling. Based on these data, the pyrazolopyrimidines were selected for further investigation.
Optimization of Pyrazolopyrimidines-To gain insights into the structural features of pyrazolopyrimidines required to inhibit the PI3K␣ and maintain cellular activity, we pursued a combination of virtual screening and synthetic medicinal chemistry. In the first step, a Bayesian model (supplemental Materials and Methods) was built based on two-dimensional descriptors of structures of the confirmed positives and negatives using the data obtained from the primary FOXO relocation screen and from PI3K␣ biochemical assay. Then, a virtual screen against our electronic, proprietary compound collection, data base of ϳ1.5 million commercially available compounds was performed using the model. This was followed by additional substructural searches, using as references the chemotypes represented by the initial hits with PI3K␣ inhibitory activity. These data lead to the acquisition of 216 pyrazolopyrimidines. 20 of these compounds were confirmed as inducers of FOXO relocation, and 18 of these also were biochemical inhibitors of PI3K␣ yielding a hit rate of 8.33% (supplemental Table SI). Analysis of these data indicated that an aromatic ring played an important role as R1 or R4.
Molecular docking studies supported the hypothesis that morpholine could play a key role in either the R1 or the R4 position. The crystal structure of PI3K␥ (23) in complex with the reference compounds LY294002 (20) and PI-103 (21) (Fig.  3A) indicates that the morpholine oxygen appears to act as a driving force to maintain a key interaction with Val-882 in the hinge region (21). Therefore, the synthetic chemistry plan used morpholine as one of the key substituents for the diversity points R1 and R4, as described in Table 1. The data presented in Table 1 suggested that an aryl ring with hydrogen bond donor/ acceptor moieties in the meta positions were preferred, and in particular meta-phenol, as suggested by the structural similarity to the PI3K inhibitor PI-103 (Fig. 3A). This structure-activity relationship (SAR)-driven rational design was further confirmed by data from Folkes et al. (24).
ETP-45658, which contains both a morpholine and metaphenol moiety (Table 1) was found to be a potent inhibitor of PI3K␣ with an IC 50 of 22 nM (Fig. 3B). The compound was predicted to interact with PI3K␥ in a manner similar to LY294002 and PI-103 where the main driving forces for binding in the ATP pocket are interactions with Val-882 and Tyr-867 (Fig. 3A). The mutant PI3K␣ proteins, H1047K and K545E, which are found in human tumors, showed similar sensitivity toward ETP-45658 as the wild-type PI3K␣ protein (Fig. 3B).
ETP-45658 was also tested for its ability to inhibit PI3K␤, PI3K␦, and PI3K␥. ETP-45658 was essentially equipotent toward PI3K␦, IC 50 30 nM, and slightly less potent toward PI3K␤, IC 50 129 nM. However it had significantly less inhibitory  OCTOBER 9, 2009 • VOLUME 284 • NUMBER 41 activity toward PI3K␥ with an IC 50 of 710 nM. ETP-45658 had modest, but significant activity toward the class IV PI3-related kinase mTOR and DNA PK with IC 50 values of 152 nM and 70.6 nM, respectively (Fig. 3B). ETP-45658 was further profiled against a panel of 72 protein kinases (supplemental Table SII); of these kinases only focal adhesion kinase (FAK) and B-Raf were inhibited more than 35% at a concentration of 10 M (supplemental Table SII). Thus it appears that ETP-45658 is relatively selective toward phosphoinositide 3-kinases.

Selective Inhibitors of PI3 Kinases
Mechanism of Action of ETP-45658-ETP-45658 was a potent inducer of GFP-FOXO nuclear translocation with an EC 50 of 45 nM ( Fig.  4A and supplemental Fig. S2A). The compound provoked a dosedependent decrease of Akt phosphorylation on serine 473 after a 1-h treatment of U2OS cells (supplemental Fig. S2B). The ability of ETP-45658 to impact the phosphorylation of the Akt substrates FOXO3a and Gsk3-␤ was also examined. Consistent with the decrease in Akt phosphorylation observed above, ETP-45658 caused a 95% decrease in the phosphorylation of FOXO3a at threonine 32 and a 55% reduction on Gsk3-␤ at serine 9 (Fig. 4B), both Akt substrates. It is not clear why the inhibition of the phosphorylation of FOXO3a on threonine 32 was greater than of GSK3-␤ on serine 9 with both LY294002 and ETP-45658; perhaps the half-life of the phosphate on serine 9 of GSK3-␤ is longer and therefore longer incubation times with the inhibitor would be required to see complete inhibition. In addition, the phosphorylation of the mTOR substrate p70 S6K was similarly reduced (Fig. 4B). Whether this was due to the inhibition of PI3K by ETP-45658 or by ETP-45658 activity toward mTOR was not be determined. ETP-45658 treatment of cells did not alter the phosphorylation status of p42/45 ERK kinases (Fig. 4B) indicative of cellular selectivity for the PI3K pathway. ETP-45658 also decreased the expression of cyclin D1 after 1 h of treatment (supplemental Fig.  S2C). Consistent with this observation, the cell cycle analysis of PC3 cells treated for 24 h with ETP-45658 revealed a clear G1 arrest with no significant evidence of apoptosis as indicated by the lack of a   OCTOBER 9, 2009 • VOLUME 284 • NUMBER 41

JOURNAL OF BIOLOGICAL CHEMISTRY 28397
sub-G1 population of cells (supplemental Fig. S2D). These data confirm that ETP-45658 efficiently inhibits PI3K signaling in treated tumor cells.
The anti-proliferative activity of ETP-45658 was examined using a small panel of human cancer cell lines with either activating mutations in PI3K␣ or loss of function of the negative

Selective Inhibitors of PI3 Kinases
regulator of PI3K, PTEN. The cellular EC 50 values for inhibition of proliferation were found to be between 0.4 and 5.5 M (supplemental Table SIII). There was no obvious difference in potency toward cell lines with activating mutations in PI3K or inactivating mutations in PTEN.
Because our ultimate aim is to explore the feasibility of PI3K/ Akt pathway inhibitors as a cancer therapy, the pharmacological activity of ETP-45658 in vivo was examined. ETP-45658 was injected in MMTV-myr-p110␣ transgenic females. These transgenic lines express myristylated p110␣ under control of the MMTV promoter, which leads to strong PI3K␣ activation in the mammary ducts of female mice (25). The female mice were injected with ETP-45658 or vehicle alone into transgenic littermates. As a positive control, a third cohort of mice was injected with the reference compound PI-103. Sixteen hours after treatment, the mice were sacrificed, and mammary glands were processed for immunohistochemistry. The epithelial-specific signal was quantified, and values of intensity of the immunohistochemical images of compound-treated mammary glands were compared with the values of the respective vehicle control images. As shown in Fig. 4C, the level of phosphorylated Akt on serine 473 was noticeably lower in the mammary ducts of transgenic mice injected with ETP-45658 as compared with the vehicle-treated mice. The phosphorylation of Akt was decreased to 50% of the control level (p ϭ 0.001). The phosphorylation status of p70S6K on theonine 389 and eIF4G on serine 1108 were also examined. In all cases, a clear reduction of the levels of phosphorylation was observed (Fig. 4C). Treatment with ETP-45658 induced a significant reduction in the phosphorylation in the markers tested (Fig. 4C). These results confirm that ETP-45658 inhibits PI3K signaling in vivo.

DISCUSSION
Because of the high frequency of genetic alterations in the PI3K/Akt signaling cascade in many types of human cancer there is intense interest to discover inhibitors of this pathway to evaluate them as a therapeutic treatment. The research presented in this report describes the use of a combination of high content cell-based screening and focused biochemical profiling of selected protein kinases to identify the molecular target of compounds with the desired mechanism of action. This was followed by molecular modeling and traditional medicinal chemistry that led to the discovery of ETP-45658 a potent and selective inhibitor of PI3 kinases. The primary screen of 33,992 compounds followed the intracellular location of the Akt effector, FOXO3a, in tumor cells (16). Compounds found that induced the nuclear translocation of GFP-FOXO yet did not affect general nuclear export were profiled against a small panel of protein kinases involved in the regulation of FOXO nuclear localization. This identified pyrazolopyrimidine derivatives as inhibitors of PI3K␣. To our surprise, we did not find any inhibitors of PDK1, Akt 1-3, or SGK1-3. This is perhaps due to an unintended bias of our compound collection against targeting the AGC family of protein kinases or the possibility that the inhibitors of AGC kinases that are present in our compound collection were unable to penetrate the cell membrane and therefore could not be selected as actives in the primary cellbased screen. As we have not performed any biochemical screens versus AGC kinase using our compound collection, we cannot be sure which explanation is more accurate. A total of 174 FOXO relocators were identified that apparently had no inhibitory activity toward protein kinases that directly phosphorylates FOXO or selected upstream kinases that regulate these kinases, suggesting that these compounds perhaps target proteins outside of the canonical PI3K/Akt phosphorylation cascade. Several compounds were found to be nonselective inhibitors of several receptor-tyrosine kinases, which could account for their activity in the FOXO assay.
Kau et al. (26) reported results in a similar screen for inhibitors of FOXO1a nuclear export. No common compounds were obtained between the two screens; however, three common chemical classes were found (supplemental Fig. S3A), none of which were active toward PI3K␣. Chemotype 1 (C1) were phenothiazines, which have been reported as ligands for monoaminerergic GPCR, dopamine receptor antagonists (27). The second chemotype (C2) is based on four fused rings and has been reported as reversible inhibitors of monoamine oxidase (MAO) type A (28). Chemotype 3, aminoquinazolines, has been reported as protein kinase inhibitors (29). The most frequent chemotype, C1, was present in eleven of the FOXO relocators where key chemical and pharmacophoric features were clearly identified. An additional seven FOXO relocators were found to have the same pharmacophore yet were different chemotypes (C2 and C3).
Using a high-throughput, cell-based assay to screen for modulators of the mTOR signaling network Yang et al. (30) identified quinostatin as an inhibitor of S6 phosphorylation, which they subsequently found to be a weak inhibitor of PI3K␣. Interestingly a derivative of quinostatin was identified in the imagebased screen used in the present work and biochemically inhibited PI3K␣ (supplemental Fig. S3B).
A SAR was developed for the pyrazolopyrimidines using the FOXO nuclear localization data coupled with the PI3K␣ biochemical inhibitory data. The SAR analysis indicates that mor- The inset in the left lower corner of each panel shows a representative region of the epithelial layer of a duct in higher magnification (ϫ400). All immunohistochemistry staining was performed in parallel. The phosphorylation sites detected are as follows: P-Akt (serine 473), P-eIF4G (serine 1108), and P-p70/S6K (threonine 389). The epithelial-specific signal was quantified, and values of intensity of the immunohistochemical images of compound-treated mammary glands were compared with the values of the respective vehicle control images. Legend: ***, p Ͻ 0.001. pholine and substituted aryls with donor/acceptor functionalities are the preferred substituents in positions R1 and R4 of the pyrazolopyrimidine scaffold. The most active analog synthesized, ETP-45658, combines morpholine in the R1 position and meta-phenol in the R4 position of the pyrazolopyrimidine scaffold. It is an ATP competitive inhibitor of PI3K␣ with an IC 50 of 22 nM. It is equally potent toward PI3K␤ and PI3K␦ as well as toward two mutant forms of PI3K␣ that are commonly found in human tumors. The compound also inhibits PI3K␥, mTOR, and DNA PK at submicromolar concentrations.
ETP-45658 elicits the expected mechanism of action in vitro and in vivo; there is a decrease in the phosphorylation of Akt at serine 473, of FOXO at theonine 32 and of p70S6K at threonine 389 in cells after treatment with the compound. The EC 50 for induction of the nuclear translocation of GFP FOXO of 45 nM consistent with the IC 50 of 22 nM obtained for biochemical inhibition of PI3K␣. ETP-45658 induces a G1 arrest concomitant with cyclin D1 decrease in treated cells. The compound has antiproliferative activity in a variety of tumor-derived cell lines, and this activity appears to be independent of PI3K␣ and PTEN mutational status since we obtained similar results in cells with mutant PI3K␣ and wild-type PTEN (MCF7 or HCT116) or cells with wild type PI3K␣ and mutant PTEN (786-0, PC3, or U251). Similar observations have been reported for other PI3K inhibitors (34). ETP-45658 is pharmacologically active in vivo. The treatment of mice that express a myristylated p110␣ PI3K subunit in mammary ductal cells with the compound caused the expected changes in selected molecular markers of pathway inhibition, including Akt, eIF4G, and p70 S6K. Work continues to further optimize the pyrazolopyrimidines to propose a compound suitable for clinical exploration.