Combined Inhibition of PI3Kβ and mTOR Inhibits Growth of PTEN-null Tumors

Loss of PTEN protein expression is observed in many tumors and is associated with poorer prognosis (1–5), and resistance to immunotherapy (6, 7). Given the incidence of aberrations in PTEN protein loss, there is a need to find effective therapeutic strategies to treat PTEN-null tumors. The tumor suppressor PTEN regulates PI3K signaling (5, 8). Epithelial cells are normally dependent on the PI3Kα isoform, but when PTEN is deleted, signaling through the PI3Kβ isoform is important for tumor progression (2, 9). Although the mechanism that creates this dependency is unknown, it may be associated with an increase in cellular PIP3 levels (10, 11) via constitutive basal activity of PI3Kβ. PI3Kβ is the only PI3K isoform that can be activated by both G-protein–coupled receptors (GPCR) and transmembrane growth factors receptors (GFR; ref. 12). Unlike the PI3Kα isoform where activating mutations are common (5), the PI3Kβ isoform is rarely mutated (13, 14), hence the activation remains dependent on other proteins.

A number of kinase inhibitors targeting PI3Kβ (GSK2636771, SAR260301, and AZD8186; refs. 15–18) have been progressed to clinical trials with the aim of treating PTEN-null tumors. However, it has become apparent that while PTEN-null cell lines are enriched for a dependency on PI3Kβ, multiple mechanisms have been associated with either feedback-mediated reactivation of signaling, or resistance to PI3Kβ inhibition (19, 20). These mechanisms include the activation of EGFR, IGFR, IR, and activation of PI3Kα or ERK signaling (21–25). All of these pathways have the potential to limit efficacy, therefore identifying combination strategies to enhance or sustain pathway inhibition is important for maximal therapeutic effect. In this study, inhibitors of pathways implicated in feedback reactivation or resistance were combined with the PI3Kβ inhibitor AZD8186 with the aim of identifying an optimal combination for treatment of PI3Kβ-dependent PTEN-null tumors.

Cell lines used for in vitro and in vivo experiments are listed in Supplementary Table S1. Cell lines were cultured at 37°C, 5% carbon dioxide. All cell lines were authenticated at AstraZeneca using DNA fingerprinting short-tandem repeat (STR) assays. Cells were used within 15 passages, and cultured for less than 6 months. Details of the cancer relevant genetics for the cell lines are described in Supplementary Table S1.

Cells were lysed in RIPA buffer (Thermo Fisher Scientific) supplemented with 1× Protease Inhibitor Cocktail (Roche), 1× phosphatase inhibitor (Thermo Fisher Scientific) and 1:5,000 benzonase (Sigma-Aldrich) and equal amounts of protein were loaded and separated by SDS-PAGE. Horseradish peroxidase–linked secondary antibodies (GE Healthcare) and ECL or supersignal (Thermo Fisher Scientific) were used to detect immune complexes. Details of primary antibodies can be found in Supplementary Table S2.

For the combination screen, fresh media/compounds were replaced every 3–4 days over the course of 14 days. Cells were fixed with 3.7% formaldehyde containing 0.01% Triton X-100 (Sigma-Aldrich) for 30 minutes at room temperature. Nuclei were stained with Hoechst (1:5,000, Thermo Fisher Scientific) in PBS for 30 minutes at room temperature. Cells were analyzed on a CellInsight (Thermo Fisher Scientific) using a cell count algorithm. For long-term proliferation assays, fresh media/compounds were replaced every 7 days over the course of 21 days. Every 7 days, cells were counted using the method described for the combination screen proliferation assay. AZD6244 (ARRY-142886), vistusertib, AZD8835, AZD5363, AZD9291, AZD6244, and BMS536924 were all synthesized at AstraZeneca.

All animal experiments were performed to the according to the local regulations Home Office UK. A total of 1 × 10⁶ PC3 cells in Iscove's serum-free medium mixed 50:50 with Matrigel (Becton Dickinson) or 1 × 10⁶ HCC70 cells in RPMI serum-free medium mixed 50:50 with Matrigel were implanted in the flank of female nude mice (nu/nu:Alpk; AstraZeneca) between the ages of 8 and 12 weeks. 786-0 cells (5 × 10⁶ cells in RPMI serum-free medium mixed 50:50 with Matrigel) were implanted into the flank of female SCID mice (AstraZeneca) between the ages of 8 and 12 weeks. MDA-MB-468 cells (ATCC) were implanted into #3 mammary fat pad (10⁷/mouse) in 0.05 mL of medium without serum and Matrigel (Becton Dickinson) at a 1:1 ratio. Once tumors reached approximately 200–500 mm³, animals were randomized into control and treatment groups. Tumor volume was calculated twice weekly from bilateral caliper measurements using the formula (length × width × width) × π/6). AZD8186 was generally formulated once weekly as a suspension in 0.5% HPMC/0.1% Tween 80 and dosed once or twice daily (0 and 6–8 hours). Vistusertib was formulated as a suspension in 0.5% HPMC/0.1% Tween 80. For combination dosing, AZD8186 and vistusertib were coformulated in 0.5% HPMC/0.1% Tween. Growth inhibition from the start of treatment was assessed by comparison of the geometric mean change in tumor volume for the control and treated groups. Further details of all tumor growth studies are included in Supplementary Methods.

For pharmacokinetic analysis, total blood was collected by intracardiac puncture and plasma prepared and immediately frozen at −20°C. For pharmacodynamic protein biomarker or transcript analysis at each time point, a minimum of 4 or 5 tumors were snap frozen in liquid nitrogen. Lysates were generated as follows: lysis buffer (1% Triton X-100, Invitrogen), supplemented with phosphatase inhibitors 2 and 3 (Sigma-Aldrich) and protease inhibitors (Sigma-Aldrich), were added to each tumor in a Fastprep tube (MP Biomedicals). The tumors were homogenized using a MP Biomedicals Fast Prep-24 machine. Samples were sonicated, centrifuged, and protein concentration determined. Protein was separated using SDS-PAGE and immune complexes were detected as described in the In vitro Western blot analysis section. Tumor lysates were added to Meso Scale Discovery plates to measure total and phosphorylated AKT and S6 [pAKT-Ser473 (MSD K15100D) and pS6-Ser235/236 (MSD K150DFD)]. MSD plates were used according to the manufacturer's instructions and developed using SECTOR Imager. The calculated values of the tested biomarker were logged (log₁₀ scale), averaged for animals in the same group, and geomean calculated (10⁁average). Vehicle controls were used for normalizing biomarker signal for the treated samples. A two-sided t test was performed on logged data assuming unequal variance. Pharmacokinetic data (free plasma concentration of each drug) was plotted alongside biomarker data.

For in vivo staining, FFPE tissues were sectioned at 4 μm onto slides, dewaxed, and rehydrated. Antigen retrieval was performed in a RHS microwave vacuum processor (Milestone) at 110°C in EDTA (pH 8; 2 minutes) for Foxo3a (Cell Signaling Technology 2497). Endogenous peroxidise activity was blocked with 0.18% hydrogen for 10 minutes and nonspecific binding sites were blocked with serum-free protein block (Dako X0909) for 20 minutes. Sections were incubated for 1 hour in primary antibodies (0.2 μg/mL) diluted in Tris-buffered saline containing 0.05% Tween (TBS-T). Staining was visualized using rabbit Envision HRP-linked polymer (Dako K4003) followed by incubation for 10 minutes in 3,3′-diaminobenzidine (Dako K3466). Counterstaining was conducted using Carazzi hematoxylin. All washes were performed in TBS-T and all incubations were at room temperature. No staining was observed in samples incubated with appropriate isotype control antibodies. Digital images of stained slides were acquired using an Aperio slide scanner (Leica Biosystems). Slides were annotated manually to exclude areas of poor tissue/staining quality. Cytoplasmic and nuclear FOXO3A was assessed by a pathologist to provide percent tumor cells positive for each localization and data displayed as mean percentage change in percent positive cells relative to controls.

Cell pellets were snap frozen and total RNA was extracted using a mRNeasy kit (Qiagen), with DNAse treatment, following manufacturer's instructions. Targeted gene expression was performed using the BioMark HDTM –Fluidigm Array platform (96.96 dynamic array) and TaqMan primers (human vs. mouse specific when possible) following manufacturer's instructions. Fifty nanograms of total RNA was reverse transcribed and preamplified (Thermo Fisher Scientific: #4374967, #4488593) for 14 cycles, with 96 selected primers selected from previous RNA-sequencing data (26). The 96.96 Fluidigm Dynamic Arrays were primed and loaded on an IFC Controller and qPCR experiments run on the Biomark System, using the standard 96 default protocol. Data were collected and analyzed using the Fluidigm Real-Time PCR Analysis software, generating C_t values. C_t values were normalized to the average of selected housekeeping genes (ΔC_t) and compared with the time-matched DMSO control (−ΔΔC_t). All gene expression calculations and statistical analysis (pairwise Student t test) were performed on gene expression data (−ΔΔC_t) in Jmp12.0.1, and data represented in TIBCOTM Spotfire 6.5.2. Details of primers can be found in Supplementary Table S3.

For static scanning tumor-bearing mice received approximately 15 MBq ¹⁸F-FDG (PETNET Solutions) administered as an intravenous bolus. Following injection, anesthesia was maintained for a 45-minute uptake period followed by a 20-minute emission PET scan (Inveon Multimodality PET scanner from Siemens Medical Solutions; ref. 27). Data were acquired using Inveon Acquisition Workplace (IAW) software (Siemens) version 1.5 and analyzed using Inveon Reconstruction (IRW) Software (Siemens) version 2.2.0. Images were reconstructed using the 2D filtered back projection algorithm. Regions of Interest (ROI) were manually drawn using the 3D visualization package in the IRW software. Data were expressed as maximum standardized uptake value (MaxSUV). MaxSUV was calculated as described by Gambhir and colleagues; where injected dose (ID) is the injected activity (28). For dynamic scanning, tumor-bearing mice received approximately 20 MBq ¹⁸F-FDG administered by tail vein intravenous injection and underwent a 90-minute PET emission scan. Data were acquired as above but analyzed using PMod software version 3.2. After determining the length of the first frame (scan start time to injection time), the list model data were histogrammed using two sequences represented as F:t where F = number of frames and t = time (seconds). Sequence A = 1: (scan start to injection time), (20:1, 2:5, 1:10, 3:30, 3:60, 1:300, 7:600); Sequence B = 1: (scan start to injection time), (6:5, 1:10, 3:30, 3:60, 1:300, 2:600). Images were reconstructed using ordered subset expectation maximization (OSEM)/maximum a posteriori (MAP) algorithm (28 SEM iterations, 18 MAP iterations, β = 0.004278 giving a spatial resolution of 1 mm). Spatial resolution is improved using a lower β value at the expense of a higher image noise. The left ventricle time–activity curve (TAC) was extracted from Sequence A and a hybrid input function was used to correct for myocardium uptake. The tumor TAC was extracted from the imaging sequence [two-compartment five parameter (K1, k, k3, k4, and vb) model was used to fit the tumor TAC for full kinetic analysis]. In all studies to assess biodistribution, blood, muscle, lung, liver, heart, bone, and tail were removed following scanning and weighed. Tissue samples were counted in a gamma counter (Perkin Elmer, 1480, Wizard 3) and after correction for decay converted to kilobecquerels/gram enabling the %ID/g tissue to be determined. All mice in whom the tail activity exceeded 10% of the ID were excluded from analysis.

To assess the dominant complementary drivers following suppression of PI3Kβ signaling, a combination screen was performed in a panel of PTEN-null cell lines from different tumor types. PI3Kβ was inhibited with AZD8186 (18). To inhibit key signaling nodes associated with resistance or feedback, inhibitors targeting the following kinases were selected: mTOR (vistusertib; ref. 29), PI3Kα/δ (AZD8835; ref. 30), AKT (AZD5363; ref. 31), EGFR (AZD9291; ref. 32), MEK (AZD6244; refs. 33, 34), and IGF-1R (BMS536924; ref. 35). Each compound was used at a fixed concentration sufficient to inhibit the primary target, but below that at which other kinases were inhibited. To increase the stringency of the screen, the ability to inhibit growth over 14 days was determined. While many of the combinations showed long-term benefit in individual cell lines the most effective combination across the panel was PI3Kβ and mTOR inhibition. This combination was effective even in cell lines such as PC3 and BT549, which are more resistant to each monotherapy treatment (Fig. 1A). The effects of combined PI3Kβ and mTOR inhibition were confirmed in selected cell lines using a 21-day proliferation assay. Growth of HCC70 and PC3 cell lines was suppressed by the combination at the concentrations tested (Fig. 1B). MDA-MB-468 cells were less sensitive to the combination treatment in the cell line screen, and showed minimal reduction in cell growth in the long-term proliferation assays (Fig. 1B). Collectively, these data suggest that combining AZD8186 and vistusertib has broad combination potential in a number of PTEN-null cell lines.

To assess whether the in vitro combination effects translate in vivo, a panel of PTEN-null human tumor xenograft models were treated with AZD8186 and vistusertib (Fig. 2; Supplementary Fig. S1). Combining AZD8186 (50 mg/kg twice daily) and vistusertib (15 mg/kg once daily) gave increased and durable effects in HCC70 and PC3 models (Fig. 2A and B). The PTEN-null renal tumor xenograft 786-0 is more sensitive to AZD8186, therefore the dose of AZD8186 was reduced to 12.5 mg/kg. Combining AZD8186 with vistusertib resulted in 786-0 tumor regression (Fig. 2C). Increased benefit from the combination was also observed in PTEN-null LNCAP C4-2 tumor xenografts (Supplementary Fig. S1A), a human PTEN-null prostate PDX model LuCAP E86 (Supplementary Fig. S1B), a PTEN-null renal PDX model CTG-824 (Supplementary Fig. S1C) and a PTEN-null glioblastoma xenograft U87-MG (Supplementary Fig. S1D). In contrast, the MDA-MB-468 tumor xenografts failed to show increased antitumor benefit from the combination treatment consistent with the in vitro data (Fig. 1D). Collectively, this suggests that the combination of AZD8186 and vistusertib have potential to be effective across PTEN-null tumors.

Inhibiting PI3Kβ signaling in PTEN-null cell lines reduces PI3K pathway biomarkers such as pAKT and pS6, downregulates enzymes in the cholesterol biosynthesis pathway, and reduces nucleotide synthesis (36). To gain greater insight into the consequence of targeting both mTOR and PI3Kβ, HCC70, PC3, and MDA-MB-468 cells were treated with AZD8186, vistusertib, and the combination (Fig. 3). Twenty-four hours following vistusertib treatment, inhibition of pS6 and p4EBP1 was evident, but pAKT, pNDRG1, and pPRAS40 levels returned to basal levels. Twenty-four hours following AZD8186 treatment, inhibition of pAKT, pNDRG1, and pPRAS40 was observed. However AZD8186 was less effective at inhibiting pS6 and p4EBP1 relative to vistusertib. Combination treatment effectively inhibited both PI3K and mTOR signaling nodes in HCC70 and PC3 cell lines. Consistent with the effects in the long-term proliferation assays, the combination did not effectively inhibit pathway signaling in the MDA-MB-468 cell line. Combination treatment resulted in a small induction of cleaved caspase-3 in HCC70 cells, but not in the PC3 or MDA-MB-468 cell lines. In addition to assessing modulation of protein biomarkers, HCC70 and PC3 cells were profiled for changes in expression of specific transcripts associated with cell metabolism and cell stress. AZD8186 and vistusertib modulated similar transcript profiles; however, the combination resulted in enhanced modulation of a number of genes associated with metabolism and cellular stress (Supplementary Fig. S2). Collectively, these data demonstrate that in cells where the combination of AZD8186 and vistusertib gave long-term growth suppression, there is greater suppression of both PI3K andmTOR signaling nodes beyond which either AZD8186 or vistusertib achieved when used as a monotherapy.

To assess pathway modulation in vivo, HCC70 and PC3 tumor xenografts were treated with AZD8186 and vistusertib and analyzed for changes in the same biomarkers. In both models, biomarker modulation was similar but not identical (Fig. 4; Supplementary Fig. S3). Consistent reduction in pAKT, pS6, pNDRG1 and p4EBP1, and HMGCS1 occurred following both monotherapy and combination treatment. At the doses and schedules assessed, the combination of AZD8186 and vistusertib suppressed similar pathway biomarkers with some evidence of increased suppression of pNDRG1. In both HCC70 and PC3 tumor xenografts, AZD8186 monotherapy treatment was less effective at inhibiting pS6 and p4EBP1 than vistusertib monotherapy or the combination. In contrast, in PC3 tumor xenografts, vistusertib was less effective at reducing pAKT than AZD8186 monotherapy and the combination. AZD8186 has a variable pharmacokinetic profile. While exposure appears higher in the combination treatment groups at these time points, the difference reflects the intrinsic variability in exposure seen across a number of monotherapy and combination experiments analyzed, Evidence of increased biomarker modulation following combination treatment was modest; however, there was an indication that more consistent suppression of pNDRG1 was achieved at 6 hours in the HCC70 and PC3 models. In contrast to the in vitro experiments, it is important to recognize that deconvoluting the contribution of pathway reactivation in vivo and drug pharmacokinetics to pathway recovery is challenging. This is due to the intrinsic variability in biomarker modulation between tumor samples and the pharmacokinetic profile of the compounds changing over time both influence the biomarker signal.

Inhibition of PI3K signaling in PTEN-null tumors results in the translocation of FOXO3A to the nucleus. Conversely, reactivation of the pathway through feedback or as a result of loss of compound-mediated pathway suppression reduces the levels of FOXO3A in the nucleus and increases the cytoplasmic levels (37). Therefore sustained FOXO3A nuclear translocation may serve as a more discriminating measure of sustained pathway suppression in vivo. In HCC70 tumors, nuclear translocation was observed following treatment with AZD8186 and to a lesser extent vistusertib (Fig. 5). Six hours after compound dosing, the levels of nuclear FOXO3A were significantly sustained in mice treated with the combination compared with both monotherapy treatments (Fig. 5B). This supports the conclusion that the combination achieves increased effective pathway suppression at 6 hours.

Inhibition of PI3K-AKT signaling reduces cellular glucose uptake. In PTEN-null tumors, AZD8186 treatment reduced glucose uptake (38, 39) and hence increased reduction of glucose uptake would also serve as an in vivo biomarker of increased pathway suppression. The PTEN-null renal tumor xenograft 786-0 is also ¹⁸F-FDG avid, with uptake modulated by AZD8186 (39). Animals bearing 786-0 tumor were imaged to determine the effect of the combination on ¹⁸F-FDG uptake. Pathway biomarker changes were similar to those observed in the HCC70 and PC3 tumors (Fig. 2). ¹⁸F-FDG uptake was reduced by AZD8186 and vistusertib, and further reduced in combination. AZD8186 gave a 25.6% reduction in ¹⁸F-FDG uptake, (P < 0.05), vistusertib a 23.1% reduction (P < 0.05), and the combination resulted a 37.5% decrease (P < 0.001) in MaxSUV uptake at 2 hours (Fig. 6A). The ¹⁸F-FDG biodistribution in both blood and tumor showed no significant change in systemic glucose levels with any treatment; however, AZD8186 gave a 26.3% reduction in ¹⁸F-FDG uptake (P < 0.05), vistusertib a 25.5% reduction (P < 0.05), and the combination a 42.8% decrease (P < 0.001) in MaxSUV uptake at 2 hours (Fig. 6B). Dynamic compartment analysis confirmed the changes in intracellular ¹⁸F-FDG. ¹⁸F-FDG uptake into the interstitial space across the time course of the 90-minute PET scan did not change, but by the end of the PET scanning procedure, there was significantly less ¹⁸F-FDG trapped in the intracellular space in the combination treated group compared with vehicle (Fig. 6C). Similar effects on ¹⁸F-FDG uptake were observed in HCC70 tumors. A single dose of AZD8186 showed a 29.2% decrease (P = 0.0004) in MaxSUV uptake 2 hours after treatment, while combination treatment gave a 42.7% decrease (P < 0.00001; Supplementary Fig. S4A). Biodistribution analysis revealed a 36.4% decrease in ¹⁸F-FDG uptake (%ID/g) with AZD8186 and a 43.5% decrease in the combination. (Supplementary Fig. S4B). There were no significant changes seen in the blood. In U87-MG tumor xenografts, the combination also gave a 40.1% decrease (P < 0.001) in MaxSUV (Supplementary Fig. S5A). In parallel analyses (previously published in ref. 39), administration of AZD8186 alone resulted in a 26% reduction in ¹⁸F-FDG in this model (39). Biodistribution data also showed a 50.9% decrease in ¹⁸F-FDG uptake (%ID/g) with the combination treatment (Supplementary Fig. S5B). Interestingly, the combination gave significant changes in the vascular delivery of ¹⁸F-FDG uptake in tumors at early time points; however, there was no significant change in the interstitial space at later time points suggesting an early vascular effect of the combination. Combination treatment resulted in a significant decrease in the ¹⁸F-FDG that was trapped in the intracellular space (Supplementary Fig. S5C). Collectively, these data confirm that the combination of AZD8186 and AZD2014 can give increased suppression of PI3K pathway function in PTEN-null tumor cells.

Durable inhibition of PTEN-null tumor cell proliferation in vitro was achieved by combined inhibition of PI3Kβ and mTOR. Pathway biomarker analysis revealed that in sensitive cell lines the combination maintained the suppression of both PI3K and mTOR signaling and prevented recovery of active pS6 and AKT even after long-term exposure. The combination was equally effective in a range of in vivo tumor xenograft models. While it is challenging to show clear differential impact on pathway biomarkers in vivo, analysis of more proximal biomarkers revealed evidence of increased pathway suppression evidenced by increased nuclear translocation of FOXO3A in HCC70 tumor xenografts and suppression of glucose uptake in 786-0 and U87MG. Collectively, the in vitro and in vivo data are consistent with combined inhibition of PI3Kβ and mTOR giving greater pathway suppression. In vitro pathway recovery was observed that was absent in the combination; however, in vivo it is not possible to discriminate whether the increased biomarker modulation is simply additive increased pathway inhibition or prevention of reactivation. Independent of the mechanism, the combination was consistently more effective in multiple in vivo models.

The concept that failure to achieve sufficient inhibition of PI3K signaling or reactivation of signaling may limit efficacy of PI3K inhibitors is well established. This can occur through acute feedback reactivation or through acquired resistance. Inhibition of PI3Kα in PI3Kα-mutant tumors results in reactivation of PI3Kβ (20, 40), while inhibition of PI3Kβ with AZD8186 results in pathway reactivation through activation of PI3Kα (19). This feedback resulting in reactivation of AKT and phosphorylation of S6 kinase occurs through downregulation of PTEN expression or function following PI3Kα inhibition, or through receptor tyrosine kinase activation in PTEN-null tumors. In addition, long-term exposure of PI3Kα-mutant tumors to PI3Kα inhibitors can result in loss of PTEN and activation of PI3Kb (39). Combining AZD8186 with an EGFR inhibitor in the HCC70, MDA-MB-468, and HCC1954 resulted in increased growth suppression consistent with feedback through RTK activation. Moreover, the inhibition of MEK and IGFR gave a modest increase in growth suppression in the HCC70 cells. Across the cell panel, there was a range of intrinsic sensitivity to each monotherapy; however, the combination of vistusertib and AZD8186 was extremely effective even in the more resistant lines such as PC3 and BT549, suggesting this combination had greater potential for broad activity. Combining PI3K isoform inhibitors to achieve maximal pathway inhibition is therefore an attractive strategy to maximize the clinical benefit in tumors with activated PI3K signaling. While it is possible to monitor pathway dynamics in vitro, it is more challenging to demonstrate whether acute feedback is occurring when tumors are treated in vivo. This is because in these experiments, both AZD8186 and AZD2014 compound levels in the blood vary over time as they are cleared from the system. Therefore, biomarker recovery will be influenced by both the exposure of the compound and any pathway reactivation that is occurring. Independent of the mechanism of pathway reactivation, the combination treatment does appear more effective at suppressing signaling.

The combined inhibition of PI3Kβ and mTOR was effective across a broad range of tumor xenograft and explant models confirming the value of targeting these points in the PI3K pathway. The degree of antitumor activity did, however, vary across the models reflecting that PTEN-null tumors vary in sensitivity to inhibition of the PI3K pathway. Modeling of the biomarker changes and pharmacokinetics suggest that this dosing strategy may not achieve full pathway inhibition over a 24-hour dosing period. When tumors are treated with vistusertib, inhibition of pS6 and p4EBP1 is achieved over a 24-hour period, whereas it has only transient effects on the activation of AKT and downstream biomarkers. The combination with AZD8186 in these tumors delivers the suppression of the AKT signaling axis. AZD8186 is less effective at suppressing pS6 activation. Hence, both compounds target complementary nodes in this pivotal signaling axis. There was variability in the response of the tumors tested to the combination therapy ranging from growth inhibition through to tumor stasis and regression. It will be important to explore this in more detail to gain insight into the degree of differential dependency of tumors on mTORC1 and PI3K arms of the pathway, whether this is the sole driver of tumor growth inhibition, or whether other pathways limit efficacy.

Signaling through the broader PI3K pathway is complex with multiple points of regulation, and integration with other networks. There are different downstream effects associated with PI3K inhibition. Recently, PI3K has been shown to control the glycolytic phenotype of tumor cells through the regulation of Aldolase (41). Inhibition of PI3K reduces nucleotide levels regulate nucleotide levels in cells (26, 42). Finally PI3K and AKT inhibition can reduce lipid and cholesterol pathway activity (26, 43, 44), while mTORC2 signaling promoted development of liver cancer through regulation of lipid synthesis (45). Through coordinated targeting of FOXO, GSK3, and TSC2/mTORC1 function, there are significant impacts on pathway critical for tumor cell function (46). A tolerated combination that achieves optimal pathway inhibition over a threshold of inhibition is therefore an attractive approach. Targeting mTOR and PI3Kβ is well-tolerated preclinically and would not be predicted to introduce overlapping toxicities, in contrast to combinations that may target PI3K signaling. This combination is not only active in PTEN-null tumors. Interestingly the combination of vistusertib and AZD8186 also delivers positive benefit in a kras p53-mutant genetic model of pancreatic cancer (KPC model; ref. 47). This suggests that the combination concept may be important beyond PTEN-null tumors.

Activation of both translation-dependent effects and anabolic metabolism as well as cell survival and proliferation pathways is important for all cells. In the PTEN-null cells, inhibition of PI3Kβ reduces cholesterol biosynthesis enzymes, nucleotide levels, and induces cell stress (26). However, the balance of pathway dependency may vary between genotypes, and even between individual cells of a similar genotype. Targeting mTORC1, mTORC2, and PI3Kβ tackles this diversity. We hypothesize that the combination treatment results in more comprehensive pathway modulation achieving a threshold of pathway suppression which the cell is unable to resist. Specific biomarker changes observed support this. The combination of AZD8186 and vistusertib markedly reduced 4EBP1, and pNDRG1 in addition to the classical biomarkers of PI3K–AKT signaling. Moreover, in vivo, there is evidence of sustained FOXO3A nuclear translocation and also reduced glucose uptake. Interestingly, combined inhibition of rapamycin and BKM120 also resulted in increased antitumor benefit in a PTEN-null Her-2–positive brain metastasis model, and was superior to combined PI3K MEK inhibition (48), and is currently being tested clinically (NCT01470209).

In conclusion, AZD8186 and vistusertib can be combined to deliver comprehensive suppression of the PI3K/AKT/mTOR pathway resulting in activity across a broad range PTEN-null tumor cell models. The combination is current being explored in clinical trials.

All authors are current or former AstraZeneca employees and shareholders. No other potential conflicts of interest were disclosed.

Conception and design: J.T. Lynch, U.M. Polanska, U. Hancox, O. Delpuech, J. Maynard, C. Lenaghan, S.E. Critchlow, F. Cruzalegui, S.T. Barry

Development of methodology: J.T. Lynch, U.M. Polanska, O. Delpuech, J. Maynard, C.B. Trigwell, C. Eberlein, C. Lenaghan, A. Avivar-Valderas, M. Cumberbatch

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.T. Lynch, U.M. Polanska, O. Delpuech, J. Maynard, C. Eberlein, R. Polanski, M. Cumberbatch

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.T. Lynch, U.M. Polanska, U. Hancox, O. Delpuech, J. Maynard, C.B. Trigwell, C. Eberlein, A. Avivar-Valderas, F. Cruzalegui

Writing, review, and/or revision of the manuscript: J.T. Lynch, U.M. Polanska, U. Hancox, O. Delpuech, J. Maynard, R. Polanski, T. Klinowska, S.T. Barry

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.T. Lynch, U.M. Polanska, U. Hancox

Study supervision: J.T. Lynch, U. Hancox, S.T. Barry

Other (extensively discussed data with corresponding author as member of project team): F. Cruzalegui

We would like to thank the Alderley Park and the IMED Oncology Bioscience In Vivo group and the Laboratory Animal Science team for support.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.