Phosphoinositide 3-Kinase Inhibitors Combined with Imatinib in Patient-Derived Xenograft Models of Gastrointestinal Stromal Tumors: Rationale and Efficacy

Gastrointestinal stromal tumor (GIST) is the most common mesenchymal tumor of the digestive tract and is the most common sarcoma in some regions of Europe (1, 2). In the vast majority (85%–90%) of patients, GIST is driven by oncogenic KIT or PDGFRA mutations, which lead to constitutive activation of the tyrosine kinase activity of the encoded proteins (3). The majority of GISTs (75%–80%) harbor KIT mutations, which are most frequently found in exon 11, encoding the juxtamembrane region of the protein. These mutations are typically in-frame deletions (e.g., p.W557_K558del), insertions, missense mutations, or combinations of the above (4). Approximately 6% of GISTs carry mutations in KIT exon 9, which encodes the extracellular domain of the receptor. Less frequently (∼2%), GISTs harbor primary KIT exon 13 or exon 17 mutations, respectively encoding the ATP-binding site and activation-loop of the two receptor kinase domains (4). Around 20% to 25% of GISTs lack primary KIT mutations, of these GISTs, about one third have PDGFRA mutations, resulting in PDGFRA kinase activation (4). PDGFRA mutations are mainly found in exon 18, most frequently presenting a p.D842V substitution in the activation loop of the kinase domain. PDGFRA mutations in exons 12 and 14, which respectively induce changes in the juxtamembrane domain and ATP-binding kinase domain I, are far less common (5). Surgical resection is the mainstay of treatment for nonmetastatic, resectable GISTs, but unfortunately is not feasible in up to 30% of patients due to anatomic site, metastatic disease, or tumor extent. Moreover, about 40% of patients will eventually relapse even after surgical resection (6). The approval of imatinib (IMA), a tyrosine kinase inhibitor (TKI), for the treatment of unresectable or metastatic GIST, together with sunitinib (approved for the treatment of patients failing imatinib) and regorafenib (approved for the treatment of patients failing imatinib and sunitinib), has revolutionized the therapeutic approach in GIST and dramatically improved the outcome of patients with advanced, metastatic disease. The response of patients with GIST to TKIs depends on the underlying specific KIT/PDGFRA mutations in the tumor (7, 8). The emergence of heterogeneous resistance to TKIs, mainly through acquisition of secondary KIT/PDGFRA mutations, is one of the main problems in the treatment of GIST today (9, 10). After failure of all three approved lines of treatment, patients with advanced GIST are currently left without any registered treatment options. Therefore, novel therapeutic strategies are being developed and tested. One of the most promising approaches to combat resistance is inhibition of the PI3K pathway, which is downstream of KIT/PDGFRA. This pathway is deregulated in many human cancers, including GIST and other sarcomas (11–13). Results obtained in vitro by Bauer and colleagues (14) support the hypothesis that PI3K inhibition might be a valuable option in the clinical management of patients with GIST. Recently, our group showed antitumor effects of the combination of imatinib with GDC-0941, an oral pan PI3K inhibitor (PI3Ki), in a panel of GIST xenograft models in vivo (15). In this study, we evaluated the efficacy of three other oral PI3Kis with different pharmacologic properties in a variety of human GIST xenografts with variable molecular background. Buparlisib (BKM120; BKM) is a pan-PI3Ki, BEZ235 (BEZ) is a dual pan PI3K/mTOR (mammalian target of rapamycin) inhibitor, and BYL719 (BYL) is a selective inhibitor of the PI3K catalytic p110α subunit (16–18). These inhibitors have been shown to potently inhibit tumor cell growth and induce apoptosis both in vitro and in vivo (19–21) in a number of preclinical studies.

GIST xenografts were established by bilateral, subcutaneous transplantation of human GIST tissues in female adult nu/nu NMRI (Janvier Laboratories) as described previously (15, 22–24). All but GIST48 xenografts were derived from biopsies obtained from patients treated in the Department of General Medical Oncology, University Hospitals Leuven (Leuven, Belgium). The GIST48 model was derived from injection of the GIST48 cell line (gift from Dr. J.A. Fletcher, Brigham and Women's Hospital, Boston, MA) in mice. All of these models have been extensively characterized and used for in vivo experiments (15, 22–24), and a summary of their most important characteristics can be found in Table 1. The mutational analysis for KIT and PIK3CA, as well as the fluorescence in situ hybridization (FISH) to assess PTEN (phosphatase and tensin homolog) copy number, was performed as described previously (15). The xenografting of patient-derived mesenchymal tumor material has been approved by the Medical Ethics Committee, University Hospitals Leuven (S53483). All animal experiments were conducted in accordance with Belgian law and approved by the Ethics Committee for Laboratory Animals, KU Leuven (Leuven, Belgium).

Imatinib, BEZ, BKM, and BYL were provided by Novartis Pharmaceuticals Corporation. Imatinib was dissolved in sterile water, BEZ, BKM, and BYL were dissolved in 0.5% methylcellulose (Sigma-Aldrich, for BYL), supplemented with 0.05% (for BEZ) or 0.5% (for BKM) Tween 80 (Sigma-Aldrich), followed by 30 minutes of sonication at 4°C. In addition, sonication was applied to all solutions for 5 to 10 minutes prior the administration.

Western blotting and immunohistochemistry (IHC) were performed using the following primary antibodies: pKIT Y719, pAKT S473, AKT, α-tubulin, pMAPK T202/Y204, MAPK, PI3K p110α, phospho-histone H3 (pHH3; all from Cell Signaling Technology); pKIT Y703 (Life Technologies); KIT (Dako); β-actin and tubulin (Sigma-Aldrich), cleaved PARP (Abcam), and Ki67 (Thermo Scientific). For Western blotting, anti-rabbit or anti-mouse secondary antibodies, conjugated with horseradish peroxidase (HRP; DAKO) were applied, and specific bands were visualized using Western Lightning Plus-ECL (PerkinElmer). For IHC, Signalstain boost IHC detection reagent (Cell Signaling Technology) and anti-rabbit or anti-mouse Envision+ system and 3′-diaminobenzidine-tetrahydrochloride (DAB; both from DAKO) were used.

For in vivo experiments, 168 engrafted mice (314 tumors) were used. The average tumor volume was approximately 560 mm³ at the start of the experiment. Animals were grouped according to tumor size into different treatment groups: a control group (untreated), a group receiving standard treatment (imatinib 50 mg/kg/BID, p.o.), groups receiving single-agent PI3Ki, and groups treated with a combination of imatinib and PI3Ki. A detailed summary on treatment doses, schedules, and number of animals in each experiment is given in Table 2.

Tumor volume and mouse body weight were assessed regularly as previously described (15, 22–24). Experiments were divided in two phases: two weeks treatment followed by 3 weeks of the observation to assess the effect of treatment discontinuation (regrowth experiment) in half of the mice; remaining animals were euthanized after 2 weeks of the treatment (Table 2). Tumor specimens in both phases were collected and fixed in 4% buffered formaldehyde or snap-frozen in liquid nitrogen for further histologic and molecular analysis. For the analysis of the efficacy assessment, control tumors collected after 2 weeks of treatment and after the regrowth experiment were combined.

In a first stage of the study, BEZ and BYL were tested in two imatinib-sensitive (UZLX-GIST3 and UZLX-GIST4) and in one imatinib-resistant (UZLX-GIST2, dose-dependent imatinib resistance through KIT^exon9 mutation) models. In the second stage, BKM was tested in UZLX-GIST2 and GIST48 (imatinib resistance through a secondary KIT exon 17 mutation). A detailed description can be found in Supplementary Table S1.

Fixed tumor specimens embedded in paraffin were cut in 4-μm sections for hematoxylin and eosin (H&E) and immunostainings. Histologic response was graded by assessing the magnitude of necrosis, myxoid degeneration, and/or fibrosis on H&E staining: grade 1 (0%–10%), grade 2 (>10% and ≤50%), grade 3 (>50% and ≤90%), and grade 4 (>90%) as described previously (25, 26). Mitotic and apoptotic activity was assessed by counting mitotic figures and apoptotic cells in 10 high-power fields (HPF; 400-fold magnification; 0.45-mm field diameter) on H&E-stained slides. IHC for pHH3 and cleaved PARP as a measure for proliferative and apoptotic activity, respectively, were assessed by counting positive cells in 10 HPF. Digital microscopic pictures were taken to calculate the Ki67 labeling index by calculating the average percentage of Ki67-stained tumor cells in five images taken at 400-fold magnification. Microscopy was done with an Olympus CH30 microscope equipped with Color View digital camera and images were analyzed using Cell D imaging software (both from Olympus).

For Western blotting, tumor lysates were prepared from snap-frozen tumor specimens as described previously (23). Levels of chemiluminescence were captured with the FUJI-LAS mini 3000 system (Fujifilm) and densitometry was performed using the AIDA software (Raytest) to semiquantify the protein levels as previously described (15).

The Wilcoxon matched paired test was used for tumor volume comparison between day 0 and at the end of every in vivo experiment. The Mann–Whitney U test was used to compare histopathologic results and tumor volume between different treatment groups. A value of P < 0.05 was considered as statistically significant. STATISTICA 12.0 (StatSoft) was used for all calculations.

As expected, tumor volume reduction was seen in both KIT exon 11 mutants (KIT^exon11; Table 1) under imatinib treatment after 2 weeks of treatment (26% of the starting volume in UZLX-GIST3 and 27% in UZLX-GIST4; P < 0.005 for both, Wilcoxon matched paired; Table 3 and Supplementary Fig. S1). Single-agent PI3Ki treatment generally resulted in tumor growth delay in both models, and BYL even induced tumor volume stabilization in UZLX-GIST3. IMA+BEZ and IMA+BYL resulted in a superior tumor volume reduction compared with single-agent imatinib treatment in the UZLX-GIST4 model.

In the KIT exon 9 (KIT^exon9) mutant UZLX-GIST2 model, imatinib yielded tumor volume stabilization (115% after 2 weeks; P > 0.05, Wilcoxon matched paired). Similar to imatinib, BYL stabilized tumor volume (112% after 2 weeks), whereas both BEZ and BKM reduced tumor volume to respectively 77% and 74% (P < 0.05, Wilcoxon matched paired). IMA+BKM and IMA+BYL yielded a decrease in tumor burden to 58% and 66%, respectively (P < 0.05; Wilcoxon matched paired); while IMA+BEZ was not superior to single-agent BEZ treatment in reducing tumor volume. It has to be noted that for the UZLX-GIST2 model a dose of 40 mg/kg/QD was used for the single BEZ treatment, while 10 mg/kg/QD was used in combination with imatinib. This was due to the observation of severe adverse events (see below) under combination of 20 mg/kg/QD of BEZ with 50 mg/kg/BID imatinib (results not shown).

Interestingly, in GIST48 (KIT^exon11+17 mutant), the tumor burden was decreased by imatinib to 52% as compared with baseline. Single-agent treatment with BKM led to a comparable tumor volume decrease (reduction to 66%). Importantly, the IMA+BKM combination yielded a significantly better tumor volume reduction (to 14%) than either single treatment (P < 0.005, Mann–Whitney U test).

Of note, in the BEZ and BYL single-agent treatment groups, we observed mild body weight loss and dry-skin desquamation. In the BYL group, we also observed polyuria. While testing the IMA+BEZ combination in the first group of UZLX-GIST2–bearing mice, some animals showed severe body weight loss and dry-skin desquamation accompanied with fatigue, prompting us to euthanize them for ethical reasons on day 8 (results not included in this study). In addition, IMA+BYL–treated mice showed substantial body weight loss, moderate dry-skin desquamation, fatigue, and more frequent urination. The onset of such undesirable effects in the IMA+BYL group was delayed as compared with the IMA+BEZ cohort, and symptoms disappeared after treatment discontinuation. For these reasons, we reduced the BEZ and BYL doses, both as single agents as well as in combination regimens in subsequent experiments and we retested IMA+BEZ with lowered BEZ dose in the UZLX-GIST2 model during the second stage of the study (Table 2).

All models retained the morphologic features and showed the same KIT mutations as previously found in patients' biopsies used to establish the xenograft models or in the original cell line.

Of all models tested, UZLX-GIST3 was most susceptible to induction of histologic response under the chosen experimental conditions. In this model, histologic response most often involved a replacement of viable tumor tissue by myxoid degeneration, a low cellular amorphous matrix, which resembles the typical response pattern of GIST to imatinib in the clinic. Imatinib induced grade 2 or 3 histologic response in 87.5% of UZLX-GIST3 grafts. BEZ and BYL yielded grade 2 or 3 histologic response in 25% of tumors. Importantly, IMA+BEZ and IMA+BYL combinations induced even a higher level of response than any single-agent treatment, yielding grade 4 in at least 50% of tumors (Fig. 1A).

In UZLX-GIST4 and UZLX-GIST2 models, histologic response was characterized mainly by necrosis and was less pronounced than in UZLX-GIST3 (Fig. 1B and 1C). In UZLX-GIST2, histologic response in the majority of tumors was limited to grade 1 or 2 responses, while the interpretation of histologic response in the UZLX-GIST4 model was more uncertain as about 20% of control tumors showed spontaneous necrosis (Fig. 1B and 1C). In GIST48, imatinib and BKM induced grade 2 histologic response in 50% and 25% of tumors, respectively (Fig. 1D). IMA+BKM induced grade 2 or higher histologic response in 40% of tumors.

In addition to histologic response, the mitotic and apoptotic activity was assessed in all untreated control tumors and treated tumors collected at the time of treatment discontinuation. When all xenograft models were considered, control tumors showed brisk mitotic activity with an average of 25 (in UZLX-GIST2), 26 (UZLX-GIST3), 37 (UZLX-GIST4), and 17 (GIST48) mitotic figures per 10 HPF.

As expected, mitotic activity was virtually absent under imatinib treatment in both KIT^exon11 mutants as compared with control (P < 0.005, Mann–Whitney U test). Under imatinib, the apoptotic activity was significantly induced only in the UZLX-GIST3 model (P < 0.005, Mann–Whitney U test; Table 4). BEZ and BYL showed only limited effects on the mitotic and apoptotic activity in UZLX-GIST3 and -GIST4. Combination treatment with imatinib, however, virtually silenced proliferative activity, similar to imatinib alone. In addition, we found that IMA+BEZ and IMA+BYL were significantly better in inducing apoptosis as compared with either single-agent treatment.

In the UZLX-GIST2 model (KIT^exon9 mutant), imatinib yielded a 1.5-fold reduction of mitotic activity (P < 0.05, Mann–Whitney U test). Remarkably, BEZ (11.7-fold) and BYL (2.2-fold) significantly reduced mitotic activity (P < 0.005, Mann–Whitney U test), while BKM did not induce a substantial decrease in mitotic activity. IMA+BYL (3.7-fold) and IMA+BKM (2.2-fold) induced a slightly more potent decrease in proliferative activity compared with either single treatment. Interestingly, BEZ as a single agent was more potent in decreasing mitotic activity than IMA+BEZ combination. This observation is most likely related to the higher dose (40 mg/kg/QD) administered in the BEZ single-agent experiment in the UZLX-GIST2 model as compared with the dose (10 mg/kg QD) given in all other models as well as in the IMA+BEZ combination study in UZLX-GIST2. In UZLX-GIST2, neither treatment was able to induce an obvious proapoptotic affect.

In the GIST48 model, imatinib and BKM produced a similar decrease in mitotic activity (14.8- and 15.3-fold, respectively) when compared with control (P < 0.005, Mann–Whitney U test). Importantly, IMA+BKM induced a more potent inhibition of mitotic activity than either imatinib or BKM alone. Similar as in the KIT^exon11 mutants, combination treatment showed the best proapoptotic activity (3.3-fold increase as compared with the control, P < 0.005, Mann–Whitney U test). The results of mitotic and apoptotic activity, assessed on H&E, were confirmed using IHC markers in all models (Table 4).

Western blotting was performed to assess the effects on KIT activation and signaling. The analysis showed expression and activation of KIT and known crucial signaling intermediates in untreated tumors from all GIST xenografts (Supplementary Fig. S2). Furthermore, the PI3K p110α subunit was shown to be expressed in all our models, proving the presence of the main target for all PI3Ki (Supplementary Fig. S2).

As expected, imatinib inhibited (>65%) pKIT Y719 and pKIT Y703 in both KIT^exon11 mutants (Fig. 2A and B and Supplementary Fig. S2). Notably, imatinib treatment inhibited pAKT more in UZLX-GIST3 than in UZLX-GIST4 model (92% and 32% reduction, respectively). Similarly, in the former model, BEZ and BYL decreased AKT activation by 95% and 72%, whereas there was no decrease in AKT activation under BEZ in UZLX-GIST4, and under BYL AKT activation was even increased as compared with controls. In UZLX-GIST3, the combinations of IMA+BYL and IMA+BEZ were similar to single imatinib with respect to effects on AKT activation. In contrast, the combination of imatinib with PI3Ki in UZLX-GIST4 led to improved efficacy in terms of inhibition of AKT activation. IMA+BEZ and IMA+BYL induced a decrease in AKT activation by 90% and 75%, respectively, whereas imatinib as a single agent yielded only a 32% reduction in AKT activation (Fig. 2 and Supplementary Fig. S2).

As expected, imatinib induced a less pronounced inhibition of pKIT and pAKT in UZLX-GIST2 than in the KIT^exon11 mutants (37% and 35% reduction, respectively for both proteins). However, a substantial reduction in MAPK (84%) activation was observed (Fig. 2C and Supplementary Fig. S2). BEZ and BYL induced a more impressive decrease in pAKT (76% and 84% reduction) as compared with single imatinib treatment, and BKM induced a complete inactivation of AKT (>99%reduction). Combination treatment did not result in a remarkable improved inhibition of AKT as compared with single agents.

Despite the secondary KIT^exon17 mutation, imatinib induced a strong inhibition of both pKIT and its downstream intermediates in GIST48 (Fig. 2D and Supplementary Fig. S2). Importantly, under BKM and IMA+BKM activation of AKT was virtually absent; similar to what was observed in the UZLX-GIST2 model.

Of note, although the PI3Kis tested have no known direct activity against KIT and MAPK, the activation of these proteins showed inconsistencies and/or hyperactivation in our GIST xenograft models under PI3Kis and combination treatments. The most striking observation was a hyperactivation of MAPK in UZLX-GIST4 under PI3Kis, and even more prominent under combination treatments. These findings could be explained by complex cross-talk networks and release of negative feedback loops associated with the inhibition of PI3K signaling.

Tumor regrowth after treatment discontinuation was observed in all models after treatment discontinuation irrespective of treatment. However, tumor regrowth rate as compared with imatinib seemed delayed under IMA+BEZ and IMA+BYL in UZLX-GIST3 and under IMA+BYL in UZLX-GIST4 (Supplementary Table S2 and Supplementary Fig. S2). However, we did not observe any significant, long-lasting inhibition of mitosis or induction of apoptosis in either model irrespective of treatment regimen (Supplementary Table S2). Therefore, the delay in tumor regrowth is most likely related to the higher induction of histologic response under the combination regimens as compared with single agent observed after 2 weeks of treatment (Fig. 1). In the UZLX-GIST2 model, we observed an immediate tumor regrowth under IMA+BEZ 8 days after treatment discontinuation; therefore mice entered in xenograft regrowth in this stage of the study were euthanized at that time (data not shown).

The PI3K–AKT signaling pathway is a crucial regulator of cell proliferation and survival in GIST (13, 14). For this reason, we evaluated the efficacy of PI3K signaling inhibitors in animal models of this malignancy. In this study, we have shown antitumor effects of three PI3Kis (namely BEZ, BYL, and BKM) with different pharmacologic profiles. In combination with standard TKI treatment, we observed a remarkably improved efficacy as compared with single-agent TKI treatment in GIST xenograft models with diverse KIT genotype and PTEN genomic status and a different sensitivity to the established TKIs.

In this study, BEZ, a dual PI3K/mTOR inhibitor, caused tumor stabilization in both KIT^exon11 mutants. In KIT^exon9 mutants, it induced a more pronounced tumor growth delay and significant reduction in mitotic activity than imatinib alone. AKT activation was strongly inhibited in KIT^exon9 mutants, whereas in the KIT^exon11 mutants, AKT inhibition was only observed in the UZLX-GIST3 model, but not in the UZLX-GIST4 model. This observation could be explained by homozygous PTEN loss in the latter xenograft. Indeed, PTEN acts as a negative regulator of the PI3K–AKT signaling pathway (27). This hypothesis and our results are consistent with a study conducted by Quattrone and colleagues (28), in which the effect of PTEN silencing was evaluated in the imatinib-sensitive GIST-T1 (KIT^exon11 p.V560_Y579del) and imatinib-resistant GIST430 (KIT^exon11+13 p.V560_L576del+V654A) cell lines. Importantly, silencing of PTEN resulted in over-activation of AKT in both models in vitro. Moreover, AKT inhibition under BEZ treatment was less pronounced in PTEN silenced cells than in nonsilenced. In addition, in vivo studies performed by Floris and colleagues (15) showed a more pronounced inhibitory effect of the pan PI3Ki GDC-0941 on AKT activation in KIT^exon11 mutant xenografts without PTEN loss, as compared with those with this specific genomic change. Of note, GDC-0941 is a pan PI3Ki, whereas BEZ235 is a dual PI3K-mTOR inhibitor. In addition, the UZLX-GIST4 model also harbors a mutation in exon 6 of the PI3KCA gene (c.1093 G>A; p.E365K). This mutation has been described before in endometrial carcinoma (COSM86044; refs. 29, 30). Furthermore, Oda and colleagues (31) have provided experimental evidence that this mutation can cause a PI3K gain-of-function phenotype.

In our study, we also tested BKM, another pan PI3Ki, in the two imatinib-resistant models, UZLX-GIST2 (dose-dependent imatinib resistance through KIT^exon9 mutation) and GIST48 (imatinib resistance due to secondary KIT^exon17 mutation). BKM caused moderate tumor volume reduction and total AKT inhibition in both models, but the proliferative activity was inhibited only in the GIST48 model. This observation could be related to differences (e.g., molecular characteristics, drug clearance, etc.) between both models. Our findings are consistent with in vivo experiments with GDC-0941 mentioned above (15). In KIT^exon9 mutants, GDC-0941 induced tumor volume stabilization, a mild decrease in mitotic activity and more pronounced AKT inactivation than single-agent imatinib. In GIST48, effects of GDC-0941 were similar to our findings with BKM, although single-agent BKM was more effective in terms of reduction in tumor volume and mitotic activity. In addition, our results are also in line with work performed by Bauer and colleagues (14), who have observed inhibition of mitotic activity and AKT activation in vitro with the PI3Ki LY294002, a pan PI3Ki, in GIST48. Despite the secondary KIT^exon17 mutation, in our experiments GIST48 showed a strong response to imatinib in terms of reduction of tumor burden and mitotic activity. Moderate to strong responses to imatinib have been previously described in GIST48, both in vitro and in vivo, and might be due to the heterozygous nature of the secondary KIT^exon17 mutation in GIST48 (14,15,23,32). Because of the lack of imatinib resistance in the GIST48 model, we decided not to further test PI3Kis in this model.

Besides BEZ and BKM, we also tested the efficacy of BYL, a specific inhibitor blocking the p110α catalytic domain of PI3K. BYL used alone induced a statistically significant tumor volume decrease only in UZLX-GIST3, whereas tumor burden increased in the UZLX-GIST2 and UZLX-GIST4 models. BYL induced a strong AKT inhibition in UZLX-GIST3 and UZLX-GIST2 models, in contrast, to UZLX-GIST4, where an elevation of AKT activation was observed.

Interestingly, as demonstrated by differential AKT inhibition, we have observed intra- and inter-model variability in the response of our xenograft panel to different PI3Kis. This variability is most likely explained by the distinct molecular characteristics of different GIST models and the complex nature of the PI3K signaling pathway. The PTEN–PI3K–AKT pathway is known to be associated with complex cross-talk networks (e.g., with the RAS and c-JUN N-terminal kinases pathways), and elaborate negative feedback loops modulating upstream signaling mediators, including receptor tyrosine kinases (RTK; refs. 33–35). Hence, the inconsistencies in MAPK and KIT activation under PI3Ki treatment observed in our studies and in other published experiments are most likely related to these complex networks of cross-talk and feedback loops. This is further illustrated by the differential effects on AKT activation by the diverse PI3Ki in our GIST models tested. Indeed, PI3K inhibition resulted in a strong inhibition of AKT phosphorylation in all models but UZLX-GIST4, which is characterized by homozygous PTEN loss and an additional PIK3CA mutation. As mentioned above, these molecular defects could be involved in the lack of AKT inhibition observed under BEZ and BYL in UZLX-GIST4. In addition, the difference in response could be explained by the different specificity of the tested inhibitors. Over-activation of AKT observed in xenografts treated with BYL, a specific PI3K p100α inhibitor, could be explained through release of negative feedback loops due to PI3K inhibition (33–35). In addition, class I PI3K are known to function as heterodimers of a catalytic subunit (p110α, β, δ, and γ) and a regulatory subunit (p85α, p85β, p55γ, p101, or p84), and it has been shown that PI3K signaling in PTEN-deficient tumors seems to depend mainly on the p110β PI3K catalytic subunit for activation (36, 37). BEZ, as opposed to BYL, has activity against all PI3K p110 subunits and simultaneously inhibits mTOR, which could potentially inhibit the release of negative feedback loops (33–35). Certain PI3Ki characteristics, such as off-target effects, could also influence the response of GIST models with different molecular backgrounds. BKM has been shown to induce changes in expression of genes involved in mitosis, and inhibition of microtubule dynamics (38). BEZ is also known to induce a number of off-target effects as was demonstrated by Kong and colleagues (39). They showed that BEZ does not solely target class I PI3K, but also potently inhibits class II and III PI3K, whose functions are complex and not completely elucidated (36). These off-target effects could potentially affect the sensitivity to BEZ through reactivation loops.

It has been suggested that the combination of PI3Kis with RTK inhibitors or inhibitors of related active signaling pathways might prevent cross-activation of signaling pathways and release of negative feedback loops (33–35). Indeed, combinations of PI3Ki with inhibitors of other pathways (e.g., RAF–MEK–MAPK) or RTK inhibitors have shown some synergistic effects in preclinical models (35, 40–42).

Our results provide strong evidence for the hypothesis that the combination of RTK inhibitors and PI3Kis could lead to improved efficacy over single-agent PI3Ki and imatinib. The most striking observation in our study was the prominent improved antitumor activity of combining IMA+PI3Ki over either single treatment. Certainly, combination treatment was more potent in reducing tumor volume than the administration of single agents, which was most pronounced in both ULZX-GIST4 (KIT^exon11 mutant) and in GIST48 (KIT^{exon11+exon17} mutant). Moreover, the induction of apoptosis was much stronger under combination treatment than under single-agent treatment regimens in all but the KIT^exon9 mutant model. In addition, a remarkable induction of histologic response characterized by myxoid degeneration was observed in UZLX-GIST3, which was more potent than under single-agent imatinib (grade 2 or higher in all treated tumors).

Our results are consistent with in vitro studies, which have demonstrated improved antiproliferative effects of combining PI3Ki with imatinib in both imatinib-sensitive and -resistant cell lines (14, 43). Our findings are also in line with experiments reported by our group (15), which have proven the superior efficacy of IMA+GDC-0941, over either single treatment. Similarly to our study, Floris and colleagues observed a more potent reduction in tumor volume under combination treatment than under either single agent in general. GDC-0941+IMA also induced a high degree of histologic response (grade 3 or 4) in 65% of tumors. In this study, the amount of grade 3–4 histologic response was limited to 38%, 29%, and 9%, respectively, for IMA+BYL, IMA+BEZ, and IMA+BKM. We believe that this discordance in induction of histologic response is caused by a shorter duration of treatment (2 weeks) in our experiments as compared with those performed by Floris and colleagues (4 weeks; 15). In addition, because no statistically significant long-lasting effects on cell survival or proliferation were reported in our study or the one published by Floris and colleagues, we believe that the lack of long-lasting effect on tumor regrowth in our experiments is caused by a lower degree of histologic response due to shorter treatment duration.

In both studies, combination treatment was far more potent in inducing apoptosis than either single-agent treatment in all but the KIT^exon9 mutants. The lack of remarkable improved efficacy in KIT^exon9 mutants could be attributed to the differential activation of signaling molecules, discordances in gene, and protein expression and imatinib sensitivity between different KIT mutants. Duensing and colleagues (13) have reported that specific KIT^exon9 mutants show lower levels of AKT and S6K activation than KIT^exon11 mutants. In addition, it has been shown that there are significant differences in gene and protein expression between KIT^exon9 and KIT^exon11 mutants (44, 45). These differences in signaling activation and gene and protein expression could explain differences in response to treatment (e.g., effect on sensitivity to PI3K inhibition, reduced sensitivity to proapoptotic signals, etc.). Interestingly, the absence of enhanced efficacy upon combination of imatinib and PI3Kis in KIT^exon9 could be due to the known reduced sensitivity to imatinib in this subset of GIST. This hypothesis is supported by the fact that the two imatinib-sensitive KIT^exon11 and the GIST48 model all showed enhanced efficacy upon combination of imatinib and PI3Kis. Therefore, we hypothesize that the response to a combination of a PI3Ki with imatinib depends on the KIT genotype and sensitivity of the model to imatinib-related KIT inhibition. Hence, similar to clinical observations where dose-escalation improved prognosis in KIT^exon9 mutants, dose-finding studies to optimize combination of imatinib and PI3Ki could lead to an enhancement of efficacy (8).

Besides, the importance of KIT genotype in the response to combinations of imatinib and PI3Kis, our findings support the importance of PTEN status for the sensitivity to PI3Ki. Indeed, the absence of AKT inactivation under single-agent PI3Ki in UZLX-GIST4 was counteracted by combining IMA+PI3Ki, which led to improved efficacy over either single treatment. The same evidence comes from Quattrone and colleagues (28) who have shown that PTEN silencing reduced the inhibitory effect of BEZ in GIST cell lines, which were reverted substantially by IMA+BEZ combination. Floris and colleagues (15) showed that GDC-0941 had a more pronounced inhibitory effect on AKT in KIT^exon11 mutants without PTEN loss as compared with PTEN-deficient KIT^exon11 mutants. However, to confirm this hypothesis, it would need to be validated in a clinical trial. Nevertheless, PTEN status will most likely not be the only important determinant of the response of GIST to PI3Ki. Several studies have described divergent gene expression profiles and activation of certain signaling pathways in different subsets of patients with GIST (13, 44). Hence, variability in oncogenic signaling connected to KIT genotype and most likely altered after progression under standard treatment will further influence the response to PI3Ki. Our data provide justification for a more in-depth molecular/genetic characterization of refractory GIST tissue compared with what is currently done in clinical routine. Interestingly, in two currently on-going clinical trials, testing the combination of imatinib with either BYL or BKM in imatinib and sunitinib refractory patients, KIT/PDGFRA/PIK3CA/PTEN mutational status and changes in downstream PI3K signaling markers will be evaluated in archival, pretreatment or fresh tumor biopsies (when available; refs. 46, 47). These clinical studies are better suited and will have more statistical power to provide us with a better understanding of the importance of PIK3CA, PTEN, and KIT genotype in the response to PI3Kis.

Side effects observed in our study are most likely related to the concomitant inhibition of different signaling pathways and possible off-target effects of PI3Ki tested. Besides aforementioned off-target effects of BKM (mitotic index) and BEZ (class II and II PI3K), other off-target effects have been observed (38, 39). BEZ is known to inhibit DNA-dependent protein kinase, which is an important member of the DNA repair mechanism and could explain some of the adverse events observed under IMA+BEZ combination (39). Polyuria observed in the BYL- and IMA+BYL–treated groups could be explained through an effect of BYL on insulin-mediated glucose regulation (48).

In conclusion, we were able to show in vivo antitumor activity of three PI3Ki in GIST, providing supportive evidence of the use of this class of targeted agents in this malignancy. However, complex networks of cross-talk pathways and release of negative feedback loops are known to be associated with the use of PI3Ki. Overall, the combination of IMA+PI3Kis led to significantly enhanced efficacy over either single-agent treatment, as observed by more pronounced tumor volume reduction and greater proapoptotic effects than imatinib single treatment. However, molecular categorization of patients with GIST and dose-optimization should be considered in the further clinical development of such experimental combinations. Our results provide a convincing preclinical rationale for two on-going clinical trials in patients with refractory GIST combining IMA+BYL or IMA+BKM (46, 47).

A. Wozniak reports receiving speakers bureau honoraria from Novartis. J.A. Fletcher reports receiving speaker's bureau honoraria from Novartis and is a consultant/advisory board member for Ariad, Bayer, and Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Van Looy, A. Wozniak, G. Floris, P.W. Manley, M. Debiec-Rychter, P. Schöffski

Development of methodology: T. Van Looy, A. Wozniak, G. Floris, M. Debiec-Rychter, P. Schöffski

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Van Looy, A. Wozniak, H. Li, J. Wellens, J.A. Fletcher, P. Schöffski

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Van Looy, A. Wozniak, R. Sciot, M. Debiec-Rychter, P. Schöffski

Writing, review, and/or revision of the manuscript: T. Van Looy, A. Wozniak, G. Floris, R. Sciot, H. Li, J. Wellens, J.A. Fletcher, P.W. Manley, M. Debiec-Rychter, P. Schöffski

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Van Looy, A. Wozniak, J. Wellens, U. Vanleeuw, P. Schöffski

Study supervision: G. Floris, R. Sciot, M. Debiec-Rychter, P. Schöffski

Others (drug supply): P.W. Manley

The authors thank Lorna Omodho for her excellent technical assistance.

Novartis Pharmaceuticals Corporation (Basel, Switzerland) provided drugs and financial support for the described 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.