PI3K Inhibitors Synergize with FGFR Inhibitors to Enhance Antitumor Responses in FGFR2^mutant Endometrial Cancers

Endometrial cancer is the most common gynecologic malignancy in developed countries, and its incidence is increasing in postmenopausal women (1). In 2016, the American Cancer Society estimated that about 10,500 U.S. women will die of endometrial cancer (2). Treatment options for patients with recurrent or persistent endometrial cancer are limited to radiation and chemotherapy, which offer limited clinical benefit. As a result, the average survival of patients with metastatic or recurrent endometrial cancer is only 7 to 12 months (3). Thus, there is a need for more effective therapies with reduced side effects, as well as predictive biomarkers to identify patients most likely to respond to these treatment options.

Our group and others have identified activating somatic mutations in FGFR2 in about 10% of patients presenting with primary endometrioid endometrial cancer (4–9). With regard to the endometrial cancer subtypes identified by The Cancer Genome Atlas (TCGA), FGFR2 mutations occur at a similar frequency in the microsatellite instability (MSI) hypermutated subtype as well as the copy number–low subtype, which has also been described as those tumors with no specific molecular aberration (NSMP) (7, 10). More recently, mutational analysis in a large multi-institute cohort has revealed that FGFR2 mutations are more common in the tumors of patients who present with late-stage disease (17%) as well as those who progress (progressed, recurred, or died from disease; 26%) (11). In multivariate analysis where age, grade, and stage were also taken into account, the presence of an FGFR2 mutation was associated with decreased progression-free survival and decreased endometrial cancer–specific survival (11).

Preclinical studies by our group and others have shown that FGFR2^mutant endometrial cancer cells are highly sensitive to a range of FGFR inhibitors including PD173074 (5, 12) ponatinib (13, 14), BGJ398, dovitinib (15), and AZD4547 (16). The majority (93%) of FGFR2^mutant endometrial cancers also harbor mutations in the PI3K pathway (PIK3CA, PIK3CB, PIK3R1, PIK3R2, PTEN, AKT1) (7). Western blot analyses of FGFR2^mutant endometrial cancer cell lines show that FGFR inhibitors fail to completely block PI3K pathway activation (12, 15, 16). Although these in vitro studies had shown classic oncogene addiction in FGFR2^mutant endometrial cancer cell lines, in vivo studies with 30 mg/kg BGJ398 showed that FGFR inhibition alone in the FGFR2^mutant AN3CA cell line led to a delay in tumor growth but not tumor regression (15). More recently, a similar study evaluating 30 mg/kg AZD4547 in AN3CA xenografts in vivo did show tumor regression (16), consistent with earlier studies performed in our laboratory using twice daily dosing of PD173074 (data not shown).

The PI3K pathway regulates proteins involved in cell cycle, survival, and metabolism. It is thought to be the most commonly activated signaling pathway in human cancer, and endometrioid endometrial cancers have the highest frequency (80%–90%) of somatic mutations affecting this pathway (7, 8). In most tumor types, loss of PTEN and activation of PIK3CA are mutually exclusive events; however, endometrial cancer is unusual in that many tumors carry aberrations in multiple members of this signaling pathway (7).

There are several different classes of PI3K pathway inhibitors designed to target this pathway at one or more nodes and these include pan-PI3K, isoform-specific PI3K, mTOR, AKT, dual PI3K/mTOR, and dual mTORC1/mTORC2 inhibitors (reviewed in 17). Unfortunately, many inhibitors targeting this pathway have shown disappointing results in phase II/III clinical trials, and this has been attributed to a small therapeutic window accompanied by on-target toxicity from inhibiting this pathway in normal tissues, as well as a lack of predictive biomarkers to better identify the patients who will respond (18, 19).

In this study, we chose to evaluate BGJ398 (infigratinib), an orally bioavailable selective pan-FGFR inhibitor currently being evaluated in phase II trials as a single agent in several FGFR-dependent malignancies (NCT02160041, NCT02150967) as well as the pan-PI3K inhibitor (BKM120) and the p110α-selective inhibitor BYL719 (alpelisib), all developed by Novartis. Of direct relevance to this project, there is currently a phase Ib expansion trial evaluating the efficacy of BGJ398 + BYL719 in breast and lung cancers (NCT01928459). As BKM120 has been shown to possess off-target effects at concentrations above 1 μmol/L (20), we also assessed BGJ398 in combination with the class I pan-PI3K inhibitor GDC-0941 (pictilisib). This research shows that partial abrogation of signaling through the PI3K pathway enhances the efficacy of BGJ398 in FGFR2^mutant endometrial cancer models in vitro and in vivo.

AN3CA, MFE296, and JHUEM2 were obtained from ATCC (2005), ECACC (2007), and Riken Cell Bank (2012), respectively. AN3CA, JHUEM2, and MFE296 were authenticated by short tandem repeat (STR) profiling at the sequencing facility of The QIMR Berghofer Medical Research Institute in 2013 and 2016 and passaged less than 20 times since authentication. AN3CA and MFE296 cells were grown in MEM-α and JHUEM2 cells in 1:1 DMEM:HamF12, supplemented with 10% FBS, 1% penicillin/streptomycin, and 0.1 mmol/L nonessential amino acids. According to the Cancer Cell Line Encyclopedia, the cell lines harbor the following mutations: AN3CA expresses FGFR2 N550K and K310R, PIK3R1 p.557_561REIDK>Q, and PTEN p.R130fs; JHUEM2 expresses FGFR2 C383R, PIK3CA p.V344G, p.E978K, PIK3R1 p.N707del, and PTEN p.N212_splice; and MFE296 cells harbor FGFR2 N550K, PIK3CA p.I20M, p.P539R, and PTEN p.R130Q, and p.T321fs*23. Kinase inhibitors (BGJ398, GDC-0941, BKM120, and BYL719) were purchased from Selleck Chemicals for in vitro experiments and from Synkinase for in vivo studies. Structures of compounds are shown in Supplementary Fig. S1B.

Cell viability was assessed by sulforhodamine B (SRB) staining. Briefly, 3,000 cells were seeded in a 96-well plate. The following day, cells were treated with half-log dilutions of drug (1 nmol/L to 10 μmol/L). After 96 hours, cells were fixed in methanol, stained with SRB, solubilized with 10 mmol/L Tris and absorbance read at 492 nm. Values were normalized to DMSO control. IC₅₀ values are the mean of 3 independent experiments and were calculated using nonlinear regression analysis with variable slope in GraphPad Prism v6.0.

Synergy between BGJ398 and the PI3K inhibitors was assessed using the methodology proposed by Chou and Talalay (21). Drug concentrations were in a series of 2-fold dilutions above and below the IC₅₀ of each drug. One day after seeding, cells were treated with BGJ398, GDC-0941, BYL719, BKM120 alone or in combination for 96 hours. All experiments were repeated 3 independent times. The combination index and fraction affected was calculated by CalcuSyn, v2.0 (Biosoft).

Cells (600–1,000) were seeded in 6-well plates and the following day treated with DMSO or inhibitors for 72 hours. Cells were washed 3 times in PBS and grown in full-growth medium for 10 to 16 days, fixed with methanol, and stained with crystal violet (0.1% in 25% methanol). Colonies were counted, and the mean of 2 (JHUEM2) or 3 (AN3CA, MFE296) independent experiments (each performed in triplicate) was plotted as a percentage of the DMSO control.

Cells (4 × 10⁵) were seeded in 6-well plates overnight. On the second day, cells were treated with the indicated drugs or DMSO for 72 hours. Floating and attached cells were collected and analyzed for Annexin V and propidium iodide staining according to the manufacturer's instructions (FITC Annexin V Apoptosis Detection Kit II, BD Biosciences) using BD LSR II and FlowJo, v10.7.

AN3CA and JHUEM2 cells (3.5 × 10⁵) were reversed-transfected with 10 nmol Dharmacon ON-TARGETplus siRNA pools targeting p110α/PIK3CA and p110β/PIK3CB or a nontargeting control (D-001810-10) using Lipofectamine RNAiMAX in serum-free media in a 6-well plate. Full-growth media was added after 24 hours. Forty-eight hours posttransfection, cells were treated with DMSO, 0.3 μmol/L GDC-0941, or 0.6 μmol/L BYL719 for 6 hours and then lysed and subjected to Western blot analysis. Quantification of band intensities from duplicate (JHUEM2) and triplicate (AN3CA) experiments (normalized to tubulin) was performed using ImageJ.

Proteins were harvested using RIPA buffer [50 mmol/L Tris, pH 7.4, 150 mmol/L NaCl, 1 mmol/L EDTA, 1% IGEPAL, 0.1% SDS, 0.5% sodium deoxycholate, 1 mmol/L sodium orthovanadate, 1 mmol/L NaF, 1 mmol/L phenylmethylsulfonylfluoride (PMSF), 10 μg/mL aprotinin, and leupeptin]. For immunoprecipitation, 1 mg of protein lysate was precleared with Protein A SureBeads beads (Bio-Rad) for 30 minutes before incubating the lysate with anti-FGFR2 (C-17) antibody (Santa Cruz) and Protein A SureBeads beads overnight at 4°C. Western blotting was performed using standard protocols. The following antibodies were used: p110α (#4249), p110β (#3011), PARP (#9542), pFRS2α(Tyr436) (#3861), pAKT (Ser473) (#4060), AKT (#2920), pERK(Thr202/Tyr204) (#4695), ERK (#9107), pS6 (Ser240/244) (#2215), S6 (#2317), p-Tyr-100 (#9411) from Cell Signaling Technology; pERK (Thr202/Tyr204) (mouse), tubulin (T9026) from Sigma; FGFR2 Bek-C17 (sc-122), ERK2 (sc-154), FRS2 (sc-8318), and GAPDH (sc-32233) antibodies from Santa Cruz Biotechnology, Inc. Secondary antibodies IRDye 680LT Donkey anti-Rabbit IgG (C31024-04) and IRDye 800CW Donkey anti-mouse IgG (#C30904-02) were from LI-COR Corporate.

Six-week-old female NOD/SCID mice (15–18 g) were purchased from the Animal Resources Centre (Canningvale, WA, Australia) and hosted in the pathogen-free Biological Resource Facility (BRF) of the Translational Research Institute (Brisbane, Australia). Mice were maintained and handled under aseptic conditions and were allowed access to food and water ad libitum. In vivo animal studies were performed according to institution-approved protocols (TRI/160/14/AUC) and guidelines for maintenance of animals and endpoint of tumor studies were followed. Xenografts of AN3CA and JHUEM2 endometrial cancer cell lines were established by subcutaneously injecting 1 × 10⁶ viable cells in growth factor–reduced Matrigel (#354230, BD Biosciences) 1:1 with PBS into the flank of the mice. Perpendicular tumor diameters were measured by a single observer using Vernier scale calipers, and tumor volumes were calculated using the formula [(x × y2)/2]. JHUEM2 and AN3CA xenografts were allowed to grow for 10 and 14 days, respectively (to allow formation of tumors with mean xenograft volume ∼ 150 mm³). Mice were then stratified into treatment groups with one tumor per mouse on the basis of their weight and tumor volume. Mice (8/group) were treated for 3 weeks via oral gavage, 5 days on/1 day off, of (i) vehicle control [100 mmol/L acetic acid/sodium acetate buffer, pH 4.6/PEG300 (1:1)]; (ii) BGJ398, 20 mg/kg; (iii) GDC-0941, 75 mg/kg; (iv) BYL719, 12.5 mg/kg; (v) BGJ398 + GDC-0941; and (vi) BGJ398 + BYL719. Body weight was recorded for each animal every other day to monitor potential toxicities. Additional animals (4/group) were treated for 4 days, with their final treatment 6 hours prior to tumor collection. Part of the tumor was snap-frozen and then lysed in RIPA lysis buffer (2.5 μL/mg) for Western blot analysis and the other part fixed in 4% paraformaldehyde.

Tumors were fixed in 4% paraformaldehyde solution overnight, paraffin-embedded, and cut into 5-μm-thick sections. Sections were deparaffinized, rehydrated, followed by antigen retrieval with CC1 buffer at 100°C for 64 minutes using the Ventana Discovery Ultra. Slides were blocked with Discovery Inhibitor for 8 minutes, incubated with Anti-Rabbit Cleaved Caspase-3 antibody (#9661; Cell Signaling Technology) for 1 hour at 37°C, followed by secondary anti-Rabbit HQ and anti-HQ HRP. The signal was detected with DAB substrate (Discovery ChomoMap kit) followed by a hematoxylin counterstain. All images were taken with Olympus IX73 inverted Fluorescence microscope fitted with XM10 monochrome camera. Histopathologic scoring of cleaved caspase-3 was performed on 5 fields (×400 magnification) for each of the 4 tumors treated with the different drug/s avoiding areas of marked necrosis. Identification of positive cells was performed blinded and independently on a multiheader microscope by M.C. Cummings, V.F. Bonazzi, and P.M. Pollock and averaged for each sample and condition. Data for caspase positivity for each drug treatment are presented as a ratio over vehicle control.

The in vitro data were analyzed using one-way ANOVA with Tukey multiple comparison to test all treatment combinations. Differences in xenograft volume between groups were assessed for significance using a repeated 2-way ANOVA. P-values, calculated with Prism (GraphPad), are coded by asterisks: <0.05 (*), <0.01 (**), <0.001 (***), P < 0.0001 (****).

We first determined the effect of increasing concentrations of BGJ398, GDC-0941, BKM120, and BYL719 on phospho-ERK (pERK) and phospho-AKT (pAKT) downstream signaling in FGFR2^mutant endometrial cancer cell lines AN3CA and JHUEM2. Both AN3CA and JHUEM2 cells showed complete inhibition of ERK activity at 0.1 μmol/L BGJ398 (Supplementary Fig. S1A). While phosphorylated ERK1/2 is often used to indicate FGFR inhibition, we confirmed inhibition of FGFR2 by BGJ398 by immunoprecipitating FGFR2 and staining for phosphorylated tyrosine (Supplementary Fig. S1C). As expected, BGJ398 caused loss of phosphorylated FGFR2 in all 3 cell lines. AKT (Ser473) phosphorylation was unchanged even at higher concentrations of BGJ398. This is consistent with previous data we have published using the FGFR inhibitor PD173074 (12). The pan-PI3K inhibitor GDC-0941 showed significant inhibition of AKT activity in JHUEM2 and AN3CA at 0.3 μmol/L and complete inhibition at 1 μmol/L, whereas higher concentrations of BKM120 were required to obtain similar pathway inhibition (Supplementary Fig. S1). JHUEM2 cells were more sensitive to the p110α/PIK3CA-specific inhibitor BYL719, consistent with this cell line carrying activating mutations in PIK3CA.

To assess synergy, the half maximal inhibitory concentration (IC₅₀) for each single agent (BGJ398, GDC-0941, BKM120, and BYL719) was determined (Table 1). AN3CA and JHUEM2 were slightly more sensitive to BGJ398 (IC₅₀, 20 and 30 nmol/L, respectively) than MFE296 (IC₅₀, 80 nmol/L). AN3CA cells showed greater sensitivity to GDC-0941 than JHUEM2 and MFE296 cells, with an IC₅₀ value of 140 nmol/L compared with 355 and 630 nmol/L, respectively. JHUEM2 cells were more sensitive to BYL719, with an IC₅₀ value of 530 nmol/L compared with 1,730 nmol/L for AN3CA and 4,050 nmol/L for MFE296. AN3CA and JHUEM2 cells responded in a similar manner to BKM120, with IC₅₀ values of 320 and 260 nmol/L, respectively.

These IC₅₀ values were used to assess potential synergy using the Chou–Talalay equipotent fixed ratio method (Fig. 1A–I). Combination treatment of AN3CA, JHUEM2, and MFE296 cells with BGJ398 and either GDC-0941 or BYL719 resulted in enhanced inhibition of cell proliferation compared with BGJ398 alone. The combination of BGJ398 with GDC-0941 or BYL719 was synergistic (combination index values < 0.7) at all concentrations in JHUEM2 and MFE296 and in all but the lower 2 concentrations in AN3CA (Fig. 1C, F, I). The BGJ398 and BKM120 combination had a more subtle effect, with synergy seen only at the highest concentrations (Supplementary Fig. S1D–S1F). These results suggest that dual treatment with BGJ398 and either GDC-0941 or BYL719 is more synergistic at inhibiting proliferation of FGFR2^mutant endometrial cancer cells than BGJ398 combined with BKM120.

Clonogenic assays were performed to further examine the effect of the combination treatments on long-term cell survival and to determine whether synergy could be seen at clinically relevant doses (Fig. 2). Plasma concentrations of BGJ398 in phase I trial patients were found to have a C_min − C_max range of approximately 100 to 450 nmol/L (22), and as such, 100 and 300 nmol/L concentrations were assessed, as these represented a low- and mid-range concentration, respectively. The reported C_max of GDC-0941, BYL719, and BKM120 were 2.07 μmol/L (23), 2.3 μmol/L (24), and 1.8 μmol/L (25), respectively; therefore 0.3, 0.6, and 1 μmol/L were initially assessed. When combined with BGJ398, even low concentrations of the PI3K inhibitors caused a substantial reduction in colony formation (Supplementary Fig. S2A and S2B), despite the small reduction in AKT phosphorylation seen with the lower drug concentrations (Supplementary Fig. S1). For subsequent analysis, we used 0.3 μmol/L GDC-0941, 0.6 μmol/L BYL719, and 0.6 μmol/L BKM120. These concentrations were well below the plasma C_max values (often close to the C_min) such that evidence of synergism might open new avenues for using these drugs at subtherapeutic doses.

Treatment with 0.3 μmol/L BGJ398 significantly reduced colony formation by about 70% (P < 0.0001) in AN3CA cells (Fig. 2A and B). Long-term survival of AN3CA cells was slightly reduced by single-agent PI3K inhibitor treatment, suggesting that this pathway contributes to the survival of these cells. GDC-0941 alone significantly reduced colony formation by about 50% (P < 0.0001), whereas BYL719 and BKM120 only reduced colony formation by about 20% and 35% (P < 0.01), respectively. In AN3CA cells, combining BGJ398 with GDC-0941, BYL719, or BKM120 reduced clonogenic survival by a further 20% to 30% compared with BGJ398 single-agent treatment (BGJ + GDC P < 0.01, BGJ + BYL P < 0.05, one-way ANOVA). In JHUEM2 cells, BGJ398 treatment led to an approximately 50% reduction (P < 0.001) in colony formation, whereas single-agent PI3K inhibition had no significant effect (Fig. 2C and D). Nevertheless, these cells were particularly sensitive to dual inhibition of FGFR2 and PI3K, with a further reduction of about 40% to 50% (P < 0.01) by all 3 PI3K inhibitors compared with BGJ398 treatment alone. The colony formation in MFE296 cells was significantly reduced by BGJ398 (∼75% reduction; P > 0.0001), but not by any of the PI3K inhibitors alone. The combination of GDC-0941 or BKM120 with BGJ398 further reduced the colony formation by 15% (nonsignificant) and 10% (nonsignificant), respectively, but no additional benefit was seen when BYL719 was combined with BGJ398 (Fig. 2E and F). It should be noted that for this assay, cells were treated with single agents or combinations for 72 hours after which cells were washed to remove residual drugs before plating, so this assay underrepresents the cell death that would be seen following continuous drug exposure.

To test the hypothesis that BGJ398 combined with a PI3K inhibitor enhances apoptosis, Annexin V positivity was assessed following 72-hour drug treatment. Treatment with BGJ398 alone induced approximately 30% to 40% cell death in AN3CA (P < 0.05), JHUEM2 (P < 0.0001), and MFE296 (P < 0.05), compared with 10% in the vehicle control (Fig. 2G–I). The single-agent PI3K inhibitors had little effect on cell death in any cell line at the low concentrations chosen. The combination of BGJ398 with any of the PI3K inhibitors induced significantly more cell death than BGJ398 alone. These data demonstrate that combining the FGFR inhibitor BGJ398 with a pan-PI3K or p110α/PIK3CA-selective inhibitor not only reduces cell proliferation and long-term survival but also enhances the effect of BGJ398 in inducing cell death in BGJ398- sensitive endometrial cancer cells.

To understand the molecular basis of the synergistic cell death induced by the combination of BGJ398 and PI3K inhibitors, we measured the response of key downstream targets to the individual inhibitors and combinatorial treatments after 1, 8, and 24 hours (Fig. 3A and B; Supplementary Fig. S3). Phosphorylation of ERK, a downstream marker of FGFR activity, is totally abrogated by 0.3 μmol/L BGJ398 at all time points in all 3 cell lines. BGJ398 slightly blocks AKT phosphorylation in JHUEM2 and MFE296 and shows little effect on AKT phosphorylation in AN3CA cells.

In JHUEM2, all 3 PI3K inhibitors inhibit AKT activity to a similar extent, leading to dephosphorylation of S6 at 8 and 24 hours. In AN3CA and MFE296 cells, GDC-0941 is the most effective inhibitor of AKT and S6 with strongest inhibition seen at the earlier 2 time points. BYL719 has little effect on AKT or S6 activity in AN3CA cells. This may be explained by the lack of a PIK3CA mutation in AN3CA cells, the presence of which likely sensitizes JHUEM2 cells to this isoform-selective inhibitor. The inhibition of the PI3K pathway by single agents is short-lived, with phospho-S6 levels returning to almost normal by 24 hours in AN3CA and JHUEM2. The dephosphorylation of S6 in response to dual targeting of the FGFR and PI3K pathways was greatest at 8 hours and still evident at 24 hours in all 3 cell lines.

To determine whether AN3CA and JHUEM2 cells show preferential signaling through p110α/PIK3CA or p110β/PIK3CB, siRNA knockdown of each gene was performed (Fig. 3C). As predicted, knockdown of p110α in JHUEM2 (PIK3CA^mutant) resulted in almost complete inhibition of AKT and S6, suggesting that p110α regulates PI3K pathway activity in these cells. In contrast, while p110α knockdown in AN3CA (PIK3CA^WT) cells inhibited AKT (though to a lesser extent than JHUEM2), this did not translate into an equivalent inhibition of the downstream effector S6, suggesting that AN3CA does not solely rely on p110α/PIK3CA for PI3K pathway activation.

Knockdown of p110β resulted in an unexpected increase in the expression of p110α in both cell line models (Fig. 3D), which resulted in an increase in the activation of AKT and S6 (Fig. 3C), unlike knockdown of p110α, which did not alter p110β levels (Fig. 3D). Knockdown of both p110α and p110β abrogated this activation of AKT signaling. The pan-PI3K inhibitor GDC-0941 almost completely inhibits AKT/S6 signaling in AN3CA, more so than the combined siRNAs (which elicit only partial knockdown), suggesting that AN3CA relies on both p110α and p110β for PI3K pathway activation.

We then studied the antitumor activity of the BGJ398 + GDC-0941 and BGJ398 + BYL719 combinations in AN3CA- and JHUEM2-derived murine xenografts. BGJ398, GDC-0941, and BYL719 alone and in combination were well-tolerated with no significant weight loss observed throughout the course of treatment (Supplementary Fig. S4A and S4B). While BGJ398 has been used at concentrations ranging from 5 to 45 mg/kg in vivo (26), we utilized a concentration of 20 mg/kg to detect increased efficacy in our combination studies. As expected, 20 mg/kg BGJ398 resulted in significantly delayed tumor growth in both cell line models compared with the control group (P < 0.001, Fig. 4A–C). GDC-0941 (75 mg/kg) and BYL719 (12.5 mg/kg) administered as single agents had surprising opposite effects. GDC-0941 inhibited tumor growth to a similar extent as BGJ398, consistent with increased pathway inhibition by 75 mg/kg GDC-0941 in vivo compared with the lower concentrations utilized for the in vitro studies (Fig. 4D). In contrast, BYL719 monotherapy unexpectedly enhanced the growth of AN3CA-derived tumors (P < 0.0001).

Combinatory treatment of BGJ398 + GDC-0941 and BGJ398 + BYL719 resulted in a marked inhibition of tumor growth in both AN3CA (P < 0.001, P < 0.05, respectively) and JHUEM2 xenograft models, compared with the BGJ398-treated groups. Indeed, these combinations caused complete or partial tumor regression, with no palpable tumor present in 5 of 8 and 4 of 8 mice with AN3CA and 2 of 8 and 1 of 8 mice with JHUEM2 xenografts treated with BGJ398 + GDC-0941 and BGJ398 + BYL719, respectively (Supplementary Fig. S4B).

Biochemically, BGJ398-treated xenografts show partial inhibition of AKT and almost complete inhibition of ERK at 4 days (Fig. 4D). AN3CA xenografts treated with GDC-0941 for 4 days show marked reduction in pAKT, pS6, and p4EBP-1 levels, confirming inhibition of the PI3K pathway. BYL719 treatment also resulted in a reduction in pAKT and pS6 levels, albeit not to the extent seen with GDC-0941. The combination of BGJ398 + GDC-0941 caused a stronger reduction in pAKT, pS6, and p4EBP1 than GDC-0941 alone. These results are consistent with the cell death and clonogenic data, which confirm that a synergistic effect occurs when blocking FGFR and PI3K pathways.

Histologic analysis of the AN3CA tumors, which had been treated for 4 days, revealed very frequent mitoses in both the vehicle control and in the BYL719 tumors (data not shown). Only occasional apoptotic bodies were observed in the GDC-0941–treated tumors, whereas a high number of apoptotic bodies were seen in the tumors treated with BGJ398 + GDC-0941 (data not shown). Both combination treatments showed broad, confluent areas of necrosis.

The single-agent and combination treatments were assessed for their effect on the apoptotic marker PARP and cleaved caspase-3 (Figs. 3A and B and 4E). In AN3CA cells, cleavage of PARP resulting in 25- and 89-kDa fragments, is more pronounced following treatment with BGJ398 + GDC-0941 at 8 and 24 hours (Fig. 3A). This is consistent with a significant increase in caspase-3 cleavage by the BGJ398 + GDC-0941 combination in AN3CA-derived xenografts and the presence of numerous apoptotic bodies compared with BGJ398 and GDC-0941 single treatments (Fig. 4E). This high level of caspase cleavage is not observed with the BGJ398 + BYL719 combination in the AN3CA xenografts, despite large areas of tumor necrosis evident histologically, suggesting different mechanisms of cell death may be occurring.

Similar Annexin positivity was seen with all 3 combinations (Fig. 2G), suggesting that BYL719 and BKM120 might elicit a different cell death mechanism. The inhibition of AKT caused by BYL719 alone or in combination with BGJ398 is much less than that caused by GDC-0941. It is possible that only minor abrogation of AKT signaling is required to kill AN3CA cells when FGFR is also inhibited. In contrast to the results observed in AN3CA cells, a strong induction of PARP cleavage in response to BGJ398 was seen in JHUEM2 cells, which was enhanced equally in all 3 PI3K inhibitor combinations treatments. This PARP cleavage coincides with similar levels of Annexin V positivity seen at 72 hours in all 3 combinations (Fig. 2H).

Small-molecule inhibitors that target oncogenic drivers of tumorigenesis are becoming standard therapies in many cancer types. Given the high frequency of PI3K aberrations in endometrial cancer, several phase II trials evaluating PI3K inhibitors as single agents have been undertaken, with overall disappointing results (27). A number of FGFR inhibitors have shown remarkable clinical responses in a subset of other FGFR-dependent malignancies (28, 29), but to date, only multi-kinase inhibitors such as dovitinib have been tested in endometrial cancer (30). No complete responses were documented in the latter; however; no hyperphosphatemia was reported, bringing into question whether sufficient inhibition of the FGFR receptors was obtained. Although FGFR inhibitors have been shown to induce cell death in FGFR2^mutant endometrial cancer cell lines with concomitant PI3K pathway activation (12), it is reasonable to assume that the co-occurrence of activating PI3K pathway mutations may limit the extent and durability of tumor responses to single-agent FGFR inhibitors.

Here we show that multiple inhibitors targeting the PI3K pathway enhance the efficacy of the FGFR inhibitor BGJ398 in FGFR2^mutant endometrial cancers. Notably, we show that low doses of PI3K inhibitors, correlating with only partial inhibition of AKT phosphorylation, synergize with FGFR inhibition to achieve cell death and tumor shrinkage in vivo. Our data suggest isoform-specific inhibition of p110α by BYL719 has different effects in AN3CA and JHUEM2 cells. In JHUEM2 cells, carrying an activating PIK3CA mutation, BYL719 resulted in partial AKT inhibition, a small decrease in pS6 phosphorylation (Fig. 3B), and a reduction in tumor growth in vivo (Fig. 4B). In AN3CA cells, BYL719 had less of an effect on pAKT (Figs. 3A and 4D), with little to no reduction of phosphorylated pS6, suggesting that inhibition of the PI3K/AKT/mTORC1 pathway by this p110α-specific inhibitor is incomplete. Unexpectedly, BYL719 single-agent treatment increased growth of AN3CA-derived xenografts (Fig. 3A), which is blocked by the addition of BGJ398. This suggests that whatever prosurvival pathway is being activated by BYL719 in AN3CA tumors, it is blocked by pan-FGFR inhibition. Activation of parallel pathways has previously been implicated in resistance to PI3K inhibitors (reviewed in 31), leading to the belief that combination therapies are required to overcome such feedback loops. Further studies are required to understand the molecular basis of BYL719-induced tumor growth in AN3CA xenografts.

The differential reliance of AN3CA and JHUEM2 on p110α/PIK3CA may explain their differential response to BYL719. BYL719 inhibits AKT/S6 signaling to a greater extent in JHUEM2 cells, resulting in PARP cleavage when combined with BGJ398 in JHUEM2 cells (Fig. 3B), which are more reliant on p110α than AN3CA (Fig. 3C). In contrast, PI3K signaling is only partially inhibited by BYL719 in AN3CA cells, which fails to cause PARP cleavage (Fig. 3A) or caspase-dependent cell death in combination with BGJ398 (Fig. 4E). These results suggest that complete PI3K inhibition is required to induce caspase-dependent cell death. The fact that BGJ398 + BYL719 induces tumor regression to the same extent as BGJ398 + GDC-0941 suggests that only partial inhibition of the PI3K pathway is needed to lower the apoptotic threshold of FGFR inhibitors. Furthermore, the results suggest that cell death induced by BYL719 is caspase-independent and may also be PI3K-independent.

Our results show that in the context of endometrial cancer S6 is regulated by both the PI3K and FGFR2 pathways, with the combination treatments reducing levels of phosphorylated S6 more than the individual treatments (Fig. 3). The sustained inhibition of S6 by combination treatment is likely the result of inhibiting both the PI3K and FGFR2 pathways. Together with previous studies targeting the PI3K pathway alone or in combination with MEK inhibition, these results indicate that levels of phosphorylated S6 may be an effective biomarker of response to targeting these key survival pathways (32–34).

Our data in endometrial cancer are supported by similar combination studies in endometrial cancer and other cancers. Specifically in endometrial cancer, Gozgit and colleagues reported synergy between the multi-kinase inhibitor ponatinib and the mTOR inhibitor ridaforolimus (35). In liver cancer, the addition of the mTOR inhibitor RAD001 to dovitinib also showed an increase in growth inhibition of Hep3B xenografts (36), and the addition of the mTOR inhibitor rapamycin to BGJ398 resulted in increased tumor growth inhibition in a subcutaneous and a syngeneic orthotopic model (37). In FGFR1-amplified lung cancer, the combination of BGJ398 and GDC-0941 led to enhanced growth suppression in vivo (38). Furthermore, a recent study in FGFR1-amplified lung and bladder cancer cells found that the FGFR inhibitor AZD4547 caused synergistic induction of cell death in vitro when combined with an mTOR inhibitor (either AZD8055 or KU0063794) or a pan-PI3K inhibitor (GDC-0941) (29).

Early trials using pan-PI3K inhibitors were associated with on-target toxicity. Thus, considerable effort has gone into developing PI3K isoform-selective inhibitors and the identification of predictive biomarkers for these isoform-selective PI3K inhibitors. Initial studies utilizing both genetically engineered mouse models and panels of cell lines from multiple cancer types showed that PTEN-deficient breast and prostate cancer cells preferentially signal through p110β (39–41). Moreover, clinical resistance to BYL719 in a metastatic breast cancer lesion carrying an activating PIK3CA mutation was attributed to loss of PTEN in this specific lesion (42). More recently, there have been reports that PI3K isoform usage in the context of PTEN loss is dependent on the genetic context by which the PI3K pathway is activated. Tumors where the PI3K pathway was activated by either activated KRAS (43) or the polyoma middle T antigen (44) showed a reliance on p110α even in the presence of concurrent PTEN loss, in contrast to earlier studies where the reliance on p110β occurred only in models with PTEN loss.

Our attempt to determine whether preferential signaling through p110α or p110β was taking place in AN3CA or JHUEM2 cells was somewhat hindered by the unexpected activation of AKT following p110β depletion. While unexpected, our results were in line with those of Utermark and colleagues (45), showing that increased AKT activation in a conditional knockout of p110β in murine mammary epithelium in transgenic models of breast cancer driven by the polyoma middle T antigen or Her2/Neu resulted in enhanced tumor growth. The authors presented data indicating that there was competition between p110α and p110β for binding to the oncogenic receptor and removal of p110β allows increased binding of the more active p110α, leading to increased pathway activation (45). In the endometrial cancer lines examined, the increased AKT activation could be due to both an increase in p110α expression (Fig. 3C and D) as well as higher activation of AKT elicited by p110α. Whether the ablation of p110β resulted in an increase in p110α expression was not assessed in the murine models, but the data presented here suggest a level of compensatory crosstalk that has not been previously reported.

As endometrial tumors are unique in that they often carry aberrations in multiple members of the PI3K pathway, identifying biomarkers of response to isoform-specific PI3K inhibitors in endometrial cancer has proven difficult. The study by Weigelt and colleagues investigating a number of drugs targeting different nodes of the PI3K pathway in a large panel of endometrial cancer cell lines showed that PTEN-null endometrial cancers require inhibition of both p110α and p110β to reduce cell viability and suggest that the preferential use of specific catalytic subunits of PI3K may also depend on tissue context (17). This is in line with clinical data using mTOR inhibitors in endometrial cancer where no association between specific mutations and clinical responses was observed (46). Our in vitro data are consistent with that of Weigelt and colleagues (17), with single-agent pan-PI3K inhibitors GDC-0941 and BKM120 showing greater activity than BYL719 in a variety of assays. However, perhaps surprisingly, when combined with FGFR inhibition, p110α-specific inhibition by BYL719 induced similar tumor growth inhibition to pan-PI3K inhibition with GDC-0941.

A variety of studies have demonstrated that crosstalk between the MAPK and PI3K signaling pathways are associated with resistance to targeted therapies (47–49). Thus, we hypothesize that dual inhibition of both the FGFR/MAPK axis and PI3K signaling pathways will not only induce more tumor cell death but also result in a decrease in intrinsic and acquired resistance. In a similar manner, we would also hypothesize that pan-PI3K inhibition may well prevent the emergence of resistance via altered p110 isoform usage.

Although many companies are developing inhibitors against these pathways, only a few companies have inhibitors against both FGFR and the PI3K in clinical development. ArQule, Inc. has a specific FGFR inhibitor (ARQ087) and a pan-AKT inhibitor (ARQ092/ARQ751) and Astra Zeneca also has a pan-FGFR inhibitor (AZD4547) and a pan-AKT inhibitor (AKT5363). At this time, Novartis has the only combination in clinical trials (BYL719 + BGJ398) with enrolment focused on those patients with an activating PIK3CA mutation. Our data, and that of Weigelt and colleagues (17), would suggest that pan-PI3K inhibition is more efficacious than isoform-specific PIK3CA inhibition in endometrial cancer. As with all combination therapies, there remains a need to determine whether the toxicities seen with inhibition of either the PI3K or FGFR pathways are additive or synergistic when blocked together. Our in vitro data suggest that subtherapeutic inhibition of the PI3K pathway may be effective in combination, allowing lower doses and ideally less toxicity.

P.M. Pollock has ownership interest in a patent on the detection of FGFR2 mutations in endometrial cancer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L.M. Packer, X. Geng, V.F. Bonazzi, P.M. Pollock

Development of methodology: L.M. Packer, X. Geng, V.F. Bonazzi, P.M. Pollock

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.M. Packer, X. Geng, V.F. Bonazzi, R. Ju, C. Mahon, M.C. Cummings

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.M. Packer, X. Geng, V.F. Bonazzi, C. Mahon, S.-A. Stephenson, P.M. Pollock

Writing, review, and/or revision of the manuscript: L.M. Packer, X. Geng, V.F. Bonazzi, M.C. Cummings, S.-A. Stephenson, P.M. Pollock

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.M. Packer, X. Geng, V.F. Bonazzi, P.M. Pollock

Study supervision: L.M. Packer, V.F. Bonazzi, S.-A. Stephenson, P.M. Pollock

L.M. Packer has been supported by NHMRC CJ Martin Fellowship (443038). P.M. Pollock has been supported by an NHMRC CDF2 Fellowship (#1032851). The study is supported by a Cancer Australia Grant (#1087165).

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.