Intermittent High-Dose Scheduling of AZD8835, a Novel Selective Inhibitor of PI3Kα and PI3Kδ, Demonstrates Treatment Strategies for PIK3CA-Dependent Breast Cancers

The PI3K-AKT-mTOR pathway is a critical regulator of many cellular processes including proliferation, survival, and transformation, and is one of the most frequently deregulated signaling networks in human cancer (1). The PI3K family of lipid kinases are key components of this pathway, in particular, the Class I PI3K subfamily which comprise PI3Kα, PI3Kβ, and PI3Kδ (Class IA) and PI3Kγ (Class IB).

Different PI3K isoforms have been shown to play key roles in distinct tumor types, often linked with genetic alterations. In particular, PI3Kα is a commonly deregulated oncoprotein in human cancer (1). Activating somatic point mutations in the PIK3CA gene (encoding p110α catalytic subunit of PI3Kα) have been widely reported across many tumor types and these mutations often display a transforming gain-of-function activity (2–6). Additional reported mechanisms of deregulation of PI3Kα include amplification of the PIK3CA gene (7, 8), mutations in a PI3K regulatory subunit (9) or activating events elsewhere in the signaling pathway. Hence PI3Kα is a target subject to intensive efforts in identifying inhibitors and evaluating their therapeutic potential. Thus far, clinical activity observed has been moderate (10, 11), although more promising data have recently been reported with more selective PI3K-inhibitors such as BYL719 (alpelisib; refs. 12, 13) and GDC0032 (taselisib; ref. 14).

In this report, we describe preclinical studies with AZD8835, a novel dual inhibitor of PI3Kα and PI3Kδ, currently in phase I clinical evaluation. A particular focus of our preclinical studies was to evaluate intermittent high-dosing schedules (IHDS) as an alternative to a commonly applied default of continuous daily dosing. Several considerations motivated us to explore this path. First, continuous dosing may achieve suboptimal pathway inhibition in tumors, because PI3K-inhibitor clinical dose and exposure on continuous dosing schedules are capped by normal tissue toxicities such as hyperglycemia, diarrhea, and rash (12, 13, 15). Also, additional dose reduction may be required when PI3K-inhibitors are dosed continually in combination with other therapies (16). So, an intermittent schedule that allows higher doses may permit greater target inhibition and also allow for a beneficial recovery period from dose-limiting physiological side effects such as rash. Second, it is reported widely across biologic systems, including the PI3K-pathway, that continuous exposure and pathway inhibition lead to pathway reactivation via perturbation of pathway feedback mechanisms, leading to reduced effect over time (17, 18). In contrast, intermittent dosing may allow a “reset” of pathway signaling, allowing repeat apoptotic response on re-exposure to compound.

Here, reporting on preclinical studies in estrogen receptor positive (ER+), mutant PIK3CA (mPIK3CA) breast models, we illustrate that monotherapy IHDS with AZD8835 achieved good antitumor efficacy. In particular, we exemplify how this schedule was used in combination with other agents to deliver profound antitumor regression.

Cells lines were grown in RPMI-1640 media, 10% FCS, 2 mmol/L glutamine at 37°C/5% CO₂ unless indicated otherwise. Cell lines used for core in vitro and in vivo experiments are listed, along with source, dates of acquisition, relevant molecular profiling details and AZD8835 antiproliferative potencies, in Supplementary Table S1. Cell line panel details are shown in Supplementary Table S2. All cell lines were authenticated via the AstraZeneca (AZ) Cell Bank using DNA fingerprinting short tandem repeat (STR) assays. All revived cells were used within 20 passages, and cultured for less than 6 months.

Compounds AZD8835, fulvestrant, AZD9496, palbociclib, and AZD2014 were all synthesized at AZ.

For in vitro studies, compounds were prepared as 10 mmol/L stocks in DMSO, stored under nitrogen, and dispensed resulting in final DMSO concentration less than 0.5%.

For in vivo studies, AZD8835 was formulated as a suspension in HPMC/Tween [0.5% hydroxypropyl methocellulose (Methocel (Colocorn))/0.1% Polysorbate 80] and dosed per oral once (QD) or twice (BID) daily at a dose volume of 0.1 mL/10 g mouse. For BID dosing, AZD8835 was administered 6 to 8 hours apart (except the glucose/insulin timecourse study in CD1 mice were 12 hours apart). Fulvestrant was formulated in peanut oil (Sigma) and dosed three times weekly subcutaneously (s.c.) at 5 mg/mouse at dose volume of 0.1 mL/mouse. AZD9496, was formulated in 40% Polyethylene glycol (PEG)/30% Captisol and protected from light. It was dosed per oral QD at a dose volume of 0.1 mL/10 g mouse. Where dosed in combination, it was dosed 1 hour after the morning dose of AZD8835. Palbociclib and AZD2014 were formulated and dosed as for AZD8835, and coformulated when dosed with AZD8835. QD dosing was always a morning dose.

Caspase inhibitor QVD was from Sigma (SML0063).

See elsewhere for details (19). Inhibition of recombinant PI3Kα, PI3Kβ, PI3Kδ, and PI3Kγ was evaluated in ADP Kinase Glo-based enzyme assays at AZ. Broader kinase selectivity profile was determined via kinase panels at University of Dundee (United Kingdom) and DiscoveRx (KinomeScan). Cellular inhibitory activity against each of the Class I PI3K isoforms was determined by measuring phosphorylation of AKT protein in four different cell lines under assay conditions where signaling was dependent on each of the individual isoforms; cell lines were BT474 (PI3Kα), MDA-MB-468 (PI3Kβ), JeKo-1 (PI3Kδ), and RAW264 (PI3Kγ). In addition, selectivity against additional PI3K-related kinase (PIKK) family members, mTOR, DNAPK, ATR, and ATM, was demonstrated via bespoke assays at AZ.

Cell or ex vivo lysates were generated in ice-cold lysis buffer containing phosphatase and protease inhibitors. Tumors were homogenized and sonicated in ice-cold lysis buffer. Samples were prepared in a reducing loading buffer, separated on 4% to 12% Bis-Tris Novex gels, transferred onto nitrocellulose membranes, and probed with primary antibodies overnight at 4°C (Supplementary Table S3). After a washing-step, membranes were incubated with HRP-tagged secondary antibodies (Supplementary Table S3) for 1 to 2 hours at RT and visualized on a Syngene ChemiGenius Imager using Super-Signal West Dura Chemiluminescence Substrate. More details are provided in Supplementary File S1.

AZD8835 effect on cell-cycle progression was determined using a Cytomics FC500 (Beckman Coulter) flow cytometer. Cells were grown in 10 cm plates. After AZD8835 treatment, cells were washed twice with PBS, detached with Versene (Invitrogen) and centrifuged at 1,200 rpm for 5 minutes. Cell pellets were washed with PBS and centrifuged at 1,200 rpm for 5 minutes before fixation with 70% ethanol O/N at 4°C. Ethanol was removed by washing the cells twice with PBS. DNA staining was performed using FxCycle (Life Technologies) following the manufacturer's instructions. Analysis was performed using CXP analysis software (Beckman Coulter).

Pharmacology data measuring cell growth inhibition by AZD8835 were generated for a collection of 209 cancer cell lines (Supplementary Table S2). Methods were a build on those previously described (20). Additional detail is provided in Supplementary File S1.

BT474, MCF7, or T47D cells were seeded in 384-well plates at a density of 500 to 2,000 cells per well and incubated overnight. Cells were dosed with compound(s) and cell confluency was measured at 4-hour intervals over several days using the Incucyte platform (Essen Bioscience) following the manufacturer's protocol.

RNA was isolated from frozen cell pellets using RNeasy MiniKit (Qiagen-RLT Buffer), with an additional DNAse treatment step, following the manufacturer's protocol. Reverse transcription was performed using 50 ng of RNA with a Reverse Transcription kit and cDNA was then preamplified (14 cycles) using a pool of TaqMan primers, following the manufacturer's instructions (Life Technologies). E2F, ER, or FOXO modulated genes were selected for analysis using literature (21–23) and internal data. Sample and assay preparation of the 96.96 Fluidigm Dynamic arrays were carried out according to the manufacturer's instructions. Data were collected and analyzed using Fluidigm Real-Time PCR Analysis 2.1.1 software followed by normalization and statistical analysis as described in Supplementary File S1.

Studies in BT474 and MCF7 xenograft models were performed at AZ and according to local regulations (Home Office UK), as previously described (20). Female Swiss athymic nude mice (swiss nu/nu– AZ UK) were transplanted s.c. with human breast tumor cell line BT474c [derived in AZ from BT474 (ATCC HTB-20) tumors passaged in mice]; mice were implanted with 0.36 mg 60-day estrogen pellet (Innovative Research of America, #SE-121) 24 hours before cell implantation. Male SCID mice (AZ UK) were transplanted s.c. with human breast tumor cell line MCF7 (ICRF London); mice were implanted with a 0.5-mg 21-day estrogen pellet 24 hours before cell implantation. On day zero, 5 × 10⁶ cells (MCF7 or BT474) in 50% Matrigel (BD Bioscience #354234) were injected s.c. on the left flank of the animals.

Studies in T47D xenograft model were performed at Molecular Imaging Inc. and according to local regulations (NIH). Female SCID Beige mice (Harlan) were transplanted s.c. with human breast tumor cell line T47D (ATCC HTB-133); mice were implanted with 0.36 mg 60-day estrogen pellet 24 hours before cell implantation. On day zero, 1 × 10⁷ cells in 50% Matrigel were injected s.c. on the right flank of the animals.

For efficacy studies, mice were randomized into groups of 8 to 15 when average tumor volume reached approximately 200 to 500 mm³. Mice were dosed for 1 to 4 weeks at defined doses and schedules. Tumors were measured two to three times weekly by caliper and volume calculated using elliptical formula (pi/6 × width × width × length). Tumor growth inhibition (%TGI) from the start of treatment was assessed by comparison of the geometric mean change in tumor volume for the control and treated groups. Tumor regression was calculated as the percentage reduction in tumor volume from baseline (pretreatment) value: % Regression = (1 − RTV) ×100 % where RTV = Geometric Mean Relative Tumor Volume. Statistical significance was evaluated using a one-tailed t test. Detailed endpoint data for individual studies are captured in Supplementary Tables S4 and S5.

For pharmacokinetics (PK), blood samples were collected by intracardiac puncture (terminal) or by capillary micro (5 μL) sampling from tail vein (in life). Plasma samples were prepared and stored for bioanalysis at −20°C. Plasma samples were extracted by protein precipitation in acetonitrile. Following centrifugation, the supernatants were mixed with water 1:6 (v/v). Extracts were analyzed by high-performance liquid chromatography/mass spectrometry (MS) using a reverse phase C18 column and a gradient mobile phase containing water/methanol/formic acid. Compounds were quantified by MS/MS.

For tumor pharmacodynamics (PD) studies, animals were randomized into groups when tumors reached a volume of approximately 200 to 500 mm³. Tumor establishment and dosing are described above. Tumors were harvested at the defined time points using a randomized process. They were flash frozen in liquid nitrogen and stored at −80°C, for subsequent analysis of proteins by Western blotting, MSD or ELISA, or gene expression profiling using Fluidigm. Alternatively the tumor was first split and part fixed in 10% formalin buffer for 24 hours and then embedded in paraffin for IHC staining. Three or more tumors were analyzed per timepoint. Formalin-fixed paraffin-embedded samples were sectioned, subjected to antigen retrieval, and stained for cleaved caspase-3 (CC3) or for phospho-H2AX (γH2AX) using primary detection antibodies #32042 (Abcam) and #2577 (CST), respectively. More details in Supplementary File S1 and Table S3.

Mathematical PK/PD models quantifying the magnitude and time course of the effect of AZD8835 on tumor pAKT-T308, CC3, and volume in mice BT474 xenograft studies were developed. Full description of the models and data fitting approaches used are presented in Supplementary File S2.

AZD8835 (Fig. 1A) is an isoform selective small-molecule inhibitor of PI3Kα and PI3Kδ. The discovery and assay profiling of AZD8835 are described in detail elsewhere (19). In brief, enzyme assays (Table 1) demonstrated potent inhibition of PI3Kα and PI3Kδ with relative sparing of PI3Kγ, and particularly PI3Kβ. Potency against common hotspot mutant variants of PI3Kα (H1047R and E545K) and wtPI3Kα was equivalent (Table 1). Cellular assays measuring inhibitory activity of AZD8835 against each of the Class I PI3K isoforms demonstrated rank order sensitivity consistent with the data from isolated enzyme assays (Table 1). Additional selectivity profiling confirmed the broader kinase selectivity of AZD8835 (19).

Western blot analyses (Fig. 1B) demonstrated inhibition of PI3K-pathway signaling by AZD8835 across ER⁺ breast cancer cell lines, MCF7, BT474, and T47D; these all possess a mPIK3CA gene (E545K, K111N, and H1047R, respectively) and displayed sub-μmol/L growth inhibition sensitivity (GI₅₀s: 0.31 μmol/L, 0.53 μmol/L, and 0.2 μmol/L, respectively) to AZD8835 (Supplementary Table S1), consistent with PI3Kα dependency. In contrast, AZD8835 was relatively ineffective in inhibiting PI3K-pathway signaling (Fig. 1B) or cell growth (GI₅₀ 9.33 μmol/L; Supplementary Table S1) in MDA-MB-468 cells (wtPIK3CA/PTEN null), consistent with the previously reported PI3Kβ dependency of this cell line (24).

Next, we measured antiproliferative activity of AZD8835 across a pan-tissue human cancer cell line panel comprising 209 cell lines. Unbiased analyses of the genomic markers associated with AZD8835 sensitivity identified mutations in PIK3CA gene (mPIK3CA) as the most highly associated predictor of positive response. Only 12 of 177 PIK3CA WT cell lines (7.7%) were sensitive to AZD8835, whereas 15 of 32 mPIK3CA cell lines (46.9%) were sensitive, corresponding to an OR of 12.1 and a P value of 1.2 × 10⁻⁷ (Fig. 1C). In addition, KRAS mutation was demonstrated as a marker of resistance to AZD8835 (Fig. 1C and Supplementary File S1). AZD8835 was also evaluated in a larger cell panel comprising >900 cell lines (25) and again identified mPIK3CA as the strongest biomarker of response (MANOVA analysis, P = 2.7 × 10⁻⁸, data not shown). Therefore, mPIK3CA is a predictive response biomarker for AZD8835 and may potentially be used in patient selection. Overall, our data are consistent with previous studies demonstrating that models with mPIK3CA and/or other markers of PI3Kα deregulation are preferentially sensitive to PI3Kα inhibitors (26). These data also provided further evidence for PI3Kα inhibition as primary pharmacology for AZD8835. Given that AZD8835 preferentially displayed activity in mPIK3CA background, including ER⁺ breast cancer models, such models were used in our subsequent cell culture and in vivo studies.

We studied the mechanism behind AZD8835 growth inhibition of MCF7, BT474, and T47D cells by analyzing cell-cycle profiles using flow cytometry. Using a moderate concentration of 250 nmol/L AZD8835, corresponding to the GI₅₀s for proliferation (Supplementary Table S1), we observed an elevated G₀–G₁ population in BT474 and MCF7 cells, consistent with cell-cycle arrest in G₁ (Fig. 1D). This was less evident in T47D where instead there was a small increase in sub-G₀–G₁ population consistent with induction of cell death. Notably, the increase in the sub-G₀–G₁ population was evident in all three cell lines at a higher concentration (2.5 μmol/L) of AZD8835. As anticipated, higher concentration of AZD8835 also resulted in stronger inhibition of PI3K-pathway signaling (Fig. 1B). A timecourse performed in BT474 cells illustrated that the strongest pathway inhibition was observed at early timepoints (1–2 hours) with subsequent partial recovery of signaling (Fig. 1E). Cleaved PARP was also observed as a similarly early onset event (Fig. 1E), suggesting that induction of apoptosis contributes to the cell death phenotype, as reported for other PI3K-inhibitors (27). This transient pathway inhibition is consistent with pathway feedback and reactivation (17, 18) and was similarly observed in a timecourse study in MCF7 cells (Supplementary Fig. S1).

Our observation of pathway feedback, early onset apoptosis, and dose-dependent cell death made us question whether continuous dosing of PI3K-pathway inhibitors provides an optimal in vivo dosing schedule. Therefore, we were motivated to also evaluate intermittent high-dose scheduling (IHDS) of AZD8835.

We initially evaluated AZD8835 when dosed continuously (every day). Using a maximum well-tolerated dose (MWTD) of 25 mg/kg BID, we observed good antitumor efficacy in both BT474 (93% TGI) and MCF7 (25% regression) breast xenograft models in mice (Fig. 2A and B). In subsequent efficacy studies, IHDS was evaluated. Figure 2C illustrates a head to head study in the BT474 model, where similar efficacy was achieved using AZD8835 continuous dosing at 25 mg/kg BID (11% regression), a 2 days on/5 off schedule at the MWTD of 50 mg/kg BID (91% TGI), or a day 1 and 4 schedule at 100 mg/kg (92% TGI; in this initial study dosed QD). Greater tolerability of the day 1 and 4 dosing schedule in nude mice allowed a more intense MWTD of 100 mg/kg BID to be explored in the BT474 model. This schedule resulted in greater efficacy; 40% regression (Fig. 2D). IHDS studies in SCID mice strains permitted a lower MWTD of 50 mg/kg BID AZD8835 but nevertheless delivered tumor growth inhibition (see Discussion). We therefore investigated the day 1 and 4 BID AZD8835 schedule in our subsequent in vivo efficacy studies in the combination setting (see below).

To understand the efficacy observed with IHDS, we evaluated pharmacokinetic/pharmacodynamic (PK/PD) relationships in both MCF7 and BT474 xenografts.

Initially, analyzing both proximal (pAKT) and downstream (pPRAS40, pS6) PI3K-pathway biomarkers in tumor tissue, we demonstrated that pathway inhibition was both time and dose/exposure dependent (Fig. 3A). In parallel studies, we observed glucose and insulin elevation as a transient pharmacologic response to AZD8835 in mice (Supplementary Fig. S2A and S2B), as also observed for other PI3K-inhibitors (10, 28, 29).

Figure 3B shows an extended AZD8835 dose response in BT474 xenografts at 2-hour timepoint. This demonstrates a relationship between exposure and pAKT reduction in tumor tissue. Furthermore, AZD8835 induced rapid-onset apoptosis (2 hours) as measured by cleavage of caspase-3 (CC3), consistent with our cell studies (Fig. 1B, D, and E). This CC3 elevation was confirmed using an IHC endpoint (Fig. 3C). We also observed that in addition to CC3 induction on day 1, CC3 was also induced when AZD8835 was redosed on day 4 (Fig. 3D).

Interestingly, in BT474 xenografts, we observed that AZD8835 increased γH2AX (pSer139-H2AX), which tracked with an increase in CC3 signal (Supplementary Fig. S3A). Similarly, AZD8835 increased γH2AX (Supplementary Fig. S3B) in MCF7 xenografts (which lack caspase-3 expression; ref. 30). We suspected that the increase in γH2AX was a consequence of caspase activation, possibly as a response to endonuclease/caspase-mediated DNA fragmentation during apoptosis (31), thereby providing an alternative apoptosis readout in MCF7. Supportive of this, we demonstrated that the AZD8835-induced γH2AX signal in MCF7 cells was caspase dependent by using the caspase inhibitor, QVD (Supplementary Fig. S3C). In addition, we measured PARP cleavage as another measure of apoptosis in MCF7 xenografts; the response tracked that for γH2AX (Supplementary Fig. S3B and S3C).

The pAKT/CC3 PD and efficacy data from multiple AZD8835 monotherapy studies in the BT474 xenograft were compiled to build a PK/PD/efficacy model (see Supplementary File S2). Data generated across studies using different doses and schedules (continuous and IHDS) were a good fit and could be explained using this model. In the model, AZD8835 is assumed to have a dual action comprising antiproliferative and proapoptotic effects. The antiproliferative effects are reflected by tumor pAKT-T308 levels (PI3K-pathway signaling) and proapoptotic effects by CC3 levels. A direct response Emax-type model was used in the PK/PD analysis of tumor pAKT-T308 versus concentration of AZD8835 in plasma. An indirect response model with tumor cells divided into sensitive and insensitive cells was used to explain the CC3 effects. Results of the model-fits to tumor pAKT, CC3, and Volume are shown in Supplementary File S2. The model captures a desensitization to apoptosis (CC3) on continuous dosing, contrasting with IHDS which produces repeat waves of strong apoptosis induction (Fig. 3E and Supplementary File S2).

Our initial data demonstrated some potential for AZD8835 antitumor activity in the clinic when used as monotherapy. However, it is more likely that optimal patient benefit will be achieved when using AZD8835 as a combination treatment, particularly where combination cotherapies and dosing schedule are optimized for therapeutic index. Therefore, continuing with our focus on mPIK3CA, ER⁺ breast cancer models, our next aim was to investigate potential combinations in vitro which could then be evaluated in in vivo efficacy studies using a backbone of AZD8835 IHDS. With a focus on mechanisms that have demonstrated benefit in ER⁺ breast cancer patients (see Discussion), we evaluated two broad classes of combination partner for AZD8835: “inter-pathway,” combining with agents targeting parallel but interconnected driver pathways (ER, CDK4/6); or “intra-pathway,” combining with an agent targeting downstream in the PI3K-pathway.

First, we evaluated the effect of AZD8835 combinations with these classes of agents in cell growth studies using an Incucyte Imager. Anti-estrogens were selective estrogen receptor downregulators (SERDs), fulvestrant (32) or AZD9496 (33), CDK4/6 inhibitor was palbociclib (PD-0332991; ref. 34) and AZD2014 (35) provided a dual mTORC1/2 kinase inhibitor.

Consistent with our expectations, stronger growth inhibition was observed with all combinations compared with monotherapies. This is illustrated in MCF7 cells where the combination of 0.3 μmol/L AZD8835 plus 0.1 μmol/L fulvestrant or 0.1 μmol/L palbociclib prevented cell growth (Fig. 4A). Similar results were observed with the alternative SERD, AZD9496 (Supplementary Fig. S4A–S4C), and also with fulvestrant in other mPIK3CA cell lines, BT474 and T47D (Supplementary Fig. S4A and S4B). In addition, we confirmed that AZD8835/fulvestrant combination activity was retained (Supplementary Fig. S4C) in two fulvestrant-refractory MCF7-derived cell lines that were generated by continuous incubation with fulvestrant (MCF7-F100-16; ref. 36) or by long-term estrogen deprivation (MCF7-GHPED, generated at AZ).

To provide mechanistic understanding of these combination effects, we analyzed pathway biomarkers across MCF7, BT474, and T47D cells. Several consistent findings were observed, although (consistent with heterogeneity across cancer cell lines) there were also cell line-specific differences. One notable general theme was stronger pathway-biomarker suppression with combinations compared with monotherapies. Furthermore, in some cases monotherapy induced pathway cross-talk/activation, consistent with a potential resistance mechanism, which was reversed by the combination partner.

Some of the specific mechanistic observations for combinations of AZD8835 with SERDs, fulvestrant or AZD9496, were observed across more than one cell line (Fig. 4B). Across all cell lines, the combinations achieved increased suppression of pP70S6K and/or pS6. Also, particularly in T47D and BT474 cells, the SERDs increased pAKT which was reversed by the addition of AZD8835. In addition, particularly in T47D cells and to a small extent in MCF7 cells, but not in BT474 cells, these combinations increased ERα downregulation. Finally, in T47D the SERDs reversed AZD8835-mediated induction of progesterone receptor (PR), an ERα transcriptional target. Collectively, these findings are consistent with and build on previous reports with related agents (23, 37–40).

AZD8835, when combined with palbociclib, consistently enhanced suppression of pRb (and total Rb) and the E2F transcriptional targets, CDC6 and E2F (E2F-1), consistent with stronger inhibition of the Rb-pathway (Fig. 4C). AZD8835 also suppressed palbociclib-induced increase in Cyclin-D1. Also, particularly in T47D, PI3K-pathway biomarkers (pAKT and pS6) were increased by palbociclib treatment but reversed by AZD8835 (Fig. 4C). In addition, AZD8835 plus palbociclib-treated cells were subject to transcriptional profiling, using the Fluidigm platform. We observed that the expression of E2F transcriptional markers were suppressed more with the combination, compared with single agents, and more consistently than for selected ER and FOXO transcriptional targets (Fig. 4D and Supplementary Fig. S4D).

The combination of AZD8835 plus the dual mTORC1/2 inhibitor, AZD2014, also enhanced inhibition of cell proliferation (Supplementary Fig. S4E). Alongside, we observed enhanced suppression of PI3K-pathway biomarkers, including pS6 (Supplementary Fig. S4F), as reported previously for PI3K/mTOR inhibitor combinations (41). In addition, the combination enhanced suppression of pRb (and total Rb; Supplementary Fig. S4F).

Collectively the above data supported progression of each of these classes of combinations into in vivo efficacy studies.

Next, we performed combination in vivo efficacy studies, applying AZD8835 IHDS (day 1 and 4 BID). As previously demonstrated (Fig. 2D), AZD8835 dosed at 100 mg/kg day 1 and 4 BID in the BT474/nude-mice xenograft model again induced tumor regression (31 and 36%: Fig. 5A). Notably, AZD8835 combined as a doublet with fulvestrant or palbociclib increased tumor regression (59 and 54%, respectively) compared with single agents (Fig. 5A and Supplementary Table S5). We also performed similar studies in the MCF7 xenograft model, grown in SCID mice, where the tolerated dose (MWTD) of AZD8835 was lower (50 mg/kg BID day 1 and 4). Again, a combination benefit with either fulvestrant or palbociclib was seen (Fig. 5B and Supplementary Table S5). Moreover, when the agents were used in a triplet combination, thereby simultaneously targeting three driver pathways, a strikingly profound tumor regression (92%) was observed (Fig. 5B and Supplementary Table S5). We also observed stronger in vivo biomarker modulation with combinations relative to monotherapies for a subset of PD endpoints (Fig. 5C), consistent with cell studies (Fig. 4B and C). Superior efficacy with combinations was also observed when fulvestrant was replaced with an alternative SERD, AZD9496 (Supplementary Fig. S5A and Table S5), or with fulvestrant in an additional xenograft model, T47D (Supplementary Fig. S5B and Supplementary Table S5). Using an intra-pathway approach, the combination of AZD8835 with the dual mTORC1/2 inhibitor, AZD2014, also resulted in improved efficacy compared with monotherapy. This was particularly evident in the BT474 model (in nude mice) where tumor regression was achieved with a triplet combination incorporating fulvestrant, despite a tolerability requirement for AZD8835 dose reduction (Supplementary Fig. S6A and Supplementary Table S5). In the MCF7 model (SCID mice), this combination was less well tolerated so the overall efficacy achieved with the lower MWTD was relatively moderate; nevertheless a positive combination effect was still observed (Supplementary Fig. S6B and S6C and Supplementary Table S5).

Here, we describe pharmacologic studies with a novel PI3Kα/δ inhibitor, AZD8835, in mPIK3CA, ER⁺ breast cancer models.

In particular, we have explored the potential of intermittent high-dose scheduling, here termed IHDS, as an alternative to continuous dosing. We were motivated to explore this path given that clinical dose and exposure of PI3K inhibitors is capped by normal tissue toxicities (12, 13, 15) which may result in suboptimal pathway inhibition in tumors. Indeed tolerability issues observed on continuous dosing of PI3K pathway inhibitors are driving a shift towards more intermittent scheduling, albeit to date these are less radical shifts than illustrated in this manuscript.

We demonstrated that we could achieve higher dose in in vivo studies when applying IHDS, compared with continuous dosing. This was particularly the case in nude mice (BT474 model), where greater tolerability allowed the IHDS concept to be tested maximally (100 mg/kg AZD8835 day 1 and 4 BID dosing schedule) and here outperformed continuous dosing in efficacy studies (see Fig. 2A vs. Figs. 2D/5A). In contrast, in SCID strains (MCF7, T47D efficacy studies) intermittent scheduling allowed a relatively modest doubling of dose (to 50 mg/kg AZD8835) compared with that achievable using continuous dosing; even so, despite near halving of accumulative weekly dose relative to continuous dosing, monotherapy AZD8835 IHDS nevertheless delivered tumor growth inhibition (MCF7 - Fig. 5B and Supplementary Fig. S5A; T47D - Supplementary Fig. S5B) so this encouraged us to progress into combination studies. Another notable feature of BT474 xenografts is a relatively slow growth rate compared with some xenograft models, although arguably this is more representative of growth rates of clinical tumors. Growth of such tumors is less likely to “take off” during the nondosing days characteristic of intermittent schedules.

A PK/PD/efficacy model was built using monotherapy BT474 data and resulting simulations fitted well to our overall observations, as detailed in Supplementary File S2. The model captures our observations that on continuous dosing schedule (25 mg/kg BID) there is a reduced apoptosis signal, where the signal is stronger following the first dose compared with the subsequent doses which induced relatively little apoptosis. In contrast, IHDS schedules, which incorporate a break in dosing, produce repeat waves of apoptosis induction on repeat dosing (see Fig. 3D and E). We speculate that there may be a subpopulation of cells sensitive to induction of apoptosis and days are required to replenish such population. Alternatively, the break in dosing may allow “reset” of the pathway signaling from pathway feedback and reactivation which overcomes the desensitization to apoptosis induction.

We then prioritized our combination therapy strategies for AZD8835. We considered evaluation of combinations with anti-estrogen/anti-estrogen-receptor (ER) therapy to be of particular interest. Indeed, PIK3CA mutations are most prevalent (29%–45%) in the ER⁺ (luminal) subset of breast tumors (6). In addition, there is literature evidence for bidirectional PI3K/ER signaling pathway interactions, coupled with reports that combining PI3K-inhibitor and ER directed agents can combat resistance (37–40). Such PI3K/ER pathway directed combination therapies have recently progressed into the clinic (15, 42, 43). A second attractive combination opportunity for PI3K-inhibitors was combination with CDK4/6-inhibitors, which target the Rb-pathway, since in mPIK3CA tumors this combination has previously been reported as synergistic (44). Also CDK4/6 inhibitors may combine well with anti-estrogens and help to combat acquired resistance to ER antagonists (37, 45, 46), and this approach has generated positive clinical data (47). Therefore triplet combinations may also have potential and initial clinical studies are already underway, for example, combining CDK4/6 inhibitor (LEE011) with PI3K inhibitor (BYL719) and anti-estrogen letrozole (48). Regarding intra-PI3K-pathway combinations, an interesting combination opportunity is with mTOR inhibitors. Again there is some precedent in this area, for example, Elkabets and colleagues (41) reported synergy through combining a rapalogue mTORC1-inhibitor, everolimus, with PI3K inhibitor BYL719; also there is some precedent for dual PI3K/mTOR inhibition using mixed profile inhibitors such as GDC-0980, dosed intermittently (49).

We initially combined AZD8835 with ER, CDK4/6, and mTOR directed agents [SERDs (fulvestrant or AZD9496), palbociclib, AZD2014, respectively] in in vitro studies, observing enhanced combination activity and coupled with mechanistic data. We then evaluated these combinations in in vivo efficacy studies using a foundation of AZD8835 IHDS. Compelling efficacy, consistently observed as a tumor regression response (Fig. 5A and B and Supplementary Fig. S5A and S5B), was observed with the inter-pathway combinations where AZD8835 was combined with SERDs and/or CDK4/6 inhibitors. Also notable with all these inter-pathway combinations was that the MWTD of each agent was the same as used in monotherapy studies, indicating minimal combined toxicities, as illustrated in Supplementary Fig. S7.

Combination benefit was also illustrated with an “intra-pathway” combination where AZD8835 was combined with AZD2014. Despite a requirement for significant dose reductions (Supplementary Fig. S6), particularly in the MCF7/SCID model, combination benefit was again observed. We anticipate that this combination may be better tolerated in the clinic where hyperglycemia, the suspected tolerability issue in mice, could be better managed.

We conclude that AZD8835 IHDS provides flexibility and a promising alternative to continuous dosing with potential for an improved therapeutic index. Such schedules and combinations merit clinical evaluation.

All authors are current or former employees of AstraZeneca. K. Hudson, U.M. Polanska and C. Trigwell have ownership interest (including patents) and are AstraZeneca shareholders. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K. Hudson, U.J. Hancox, U.M. Polanska, P. Morentin Gutierrez, S.C. Cosulich, L. Ward, F. Cruzalegui, S. Green

Development of methodology: K. Hudson, C. Trigwell, U.M. Polanska, P. Morentin Gutierrez, M. Cumberbatch

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): U.M. Polanska, A. Avivar-Valderas, O. Delpuech, P. Dudley, M. Cumberbatch

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Hudson, U.J. Hancox, C. Trigwell, R. McEwen, U.M. Polanska, M. Nikolaou, P. Morentin Gutierrez, A. Avivar-Valderas, O. Delpuech, L. Hanson, R. Ellston, A. Jones, M. Cumberbatch, S.C. Cosulich

Writing, review, and/or revision of the manuscript: K. Hudson, U.J. Hancox, C. Trigwell, R. McEwen, U.M. Polanska, M. Nikolaou, P. Morentin Gutierrez, O. Delpuech, A. Jones, M. Cumberbatch, S.C. Cosulich, L. Ward, F. Cruzalegui, S. Green

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): U.J. Hancox, R. McEwen, U.M. Polanska, L. Hanson

Study supervision: K. Hudson, L. Ward

Other (project leader): S. Green

The authors thank Mike Dymond for advice on statistical analysis of data and to Amar Rahi, Emily Lawrie, and the AZ Lab Animal Sciences Group in support of PD/efficacy studies.

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.