Parallel PI3K, AKT and mTOR inhibition is required to control feedback loops that limit tumor therapy

Targeting the PI3K pathway has achieved limited success in cancer therapy. One reason for the disappointing activity of drugs that interfere with molecules that are important player in this pathway is the induction of multiple feedback loops that have been only partially understood. To understand these limitations and develop improved treatment strategies, we comprehensively characterized molecular mechanisms of PI3K pathway signaling in bladder cancer cell lines upon using small molecule inhibitors and RNAi technologies against all key molecules and protein complexes within the pathway and analyzed functional and molecular consequences. When targeting either mTORC1, mTOR, AKT or PI3K, only S6K1 phosphorylation was affected in most cell lines examined. Dephosphorylation of 4E-BP1 required combined inhibition of PI3K and mTORC1, independent from AKT, and resulted in a robust reduction in cell viability. Long-term inhibition of PI3K however resulted in a PDK1-dependent, PIP3 and mTORC2 independent rephosphorylation of AKT. AKT rephosphorylation could also be induced by mTOR or PDK1 inhibition. Combining PI3K/mTOR inhibitors with AKT or PDK1 inhibitors suppressed this rephosphorylation, induced apoptosis, decreased colony formation, cell viability and growth of tumor xenografts. Our findings reveal novel molecular mechanisms that explain the requirement for simultaneous targeting of PI3K, AKT and mTORC1 to achieve effective tumor growth inhibition.


Introduction
Frequent hyperactivation and deregulation of the phosphoinositide 3-kinase (PI3K)/ AKT/ mammalian target of rapamycin (mTOR) pathway in cancer has made it one of the most investigated therapeutic targets in tumor therapy. Class IA PI3Ks, consisting of a p85 regulatory subunit and a p110 catalytic subunit, with the isoforms p110α, p110β, p110γ and p110δ, phosphorylate phosphatidylinositol-4, 5 bisphosphate (PI-4,5-P2) to phosphatidylinositol-3,4,5-trisphosphate (PIP3). This reaction is reversed by the protein phosphatase and tensin homolog (PTEN) [1,2]). PIP3 initiates further signaling cascades by recruiting molecules such a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 as AKT and PDK1 via their pleckstrin homology domains. AKT, a serine-threonine kinase, is functionally activated by phosphorylation at two distinct amino acid residues, threonine 308 and serine 473, by PDK1 and mTORC2, respectively. mTORC2 is a protein complex including the kinase mTOR and rapamycin-insensitive companion of mTOR (Rictor) [3]. Phosphorylated AKT in turn has the potential to regulate multiple downstream effectors and signaling pathways that are involved for example in cell proliferation, apoptosis, migration, and metabolism [4].
The activity of mTORC1 has multiple levels of contro [7]. Raptor recruits substrates, including S6K1 and 4E-BP1, via their Tor signaling (TOS) motifs. It thus acts as a scaffolding molecule and directs the catalytic activy of mTORC1 [8]. AKT can stimulate mTORC1 activity by GTP-bound Rheb by regulating its GTPase activating protein (GAP) activity via phosphorylation of tuberous sclerosis complex 2 (TSC2) [7]. Activation of mTORC1 is further regulated by PRAS40 by competitive binding of its own TOS motif to Raptor. This inhibition can be reversed by PRAS40 phosphorylation at distinct sites by AKT and mTOR.
The first agents to target the PI3K pathway were rapamycin analogues (rapalogs), which bind to the protein FKBP-12 that complexes with mTOR, and thus allosterically inhibit mTORC1 activity [9]. These drugs have shown potential for the treatment of renal cell carcinoma, mantle cell lymphoma and neuroendocrine tumors which has fueled the development of additional classes of PI3K pathway inhibitors targeting all or specific PI3K isoforms, AKT, mTOR, or both PI3K and mTOR [10,11]. However, success in clinical trials has been lacking so far, with FDA approval granted only for the use of a PI3K δ inhibitor in chronic lymphocytic leukemia (CLL) [12]. Preclinical studies have demonstrated that inhibitors of the PI3K pathway can induce signaling feedback loops limiting their anti-tumor effects. For instance, rapalogs lead to increased AKT and ERK phosphorylation whereas dual PI3K/mTOR inhibitors lead to overexpression of different receptor tyrosine kinases [13]. Also, adaptive signaling responses after PI3Kα inhibition that increase PIP3 synthesis and AKT phosphorylation or enable SGK1-mediated mTORC1 activation have been recently described [14,15]. Successful therapeutic targeting of PI3K signaling thus requires a thorough understanding of the biochemical effects of PI3K pathway inhibition as well as effective drug combination strategies to overcome feedback loops limiting efficacy.
The PI3K pathway is overactive in around 72% of metastatic urothelial bladder cancer patients, making it an attractive target for therapy [16]. Currently, the average survival of these patients is only 12-14 months due to limited progress in therapy development since more than three decades, with PD1/PD-L1 immunotherapy only recently approved as second line therapy [17,18]. Clinical trials with rapalogs and other target therapies in bladder cancer have also demonstrated limited efficacy [19][20][21][22]. Our previous investigations have revealed that inhibition of both S6K1 and 4E-BP1 is necessary to efficiently suppress bladder cancer cell growth and proliferation. Rapalogs and AKT inhibitors, however, result in only S6K1 but not 4E-BP1 dephosphorylation, with a limited anti-proliferative effect. On the other hand, dual PI3K/ mTOR inhibitors achieve dephosphorylation of both S6K1 and 4E-BP1, prohibiting cell cycle progression, but induce rephosphorylation of AKT after long term treatment [23,24]. In this study, by using different types of PI3K pathway inhibitors and knockdown of selected molecules in the pathway we demonstrate that parallel inhibition of mTORC1 and PI3K is crucial for 4E-BP1 dephosphorlyation, independently of AKT signaling. Furthermore, we describe a novel PI3K/PDK-1 dependent but PIP3 independent feedback loop resulting in AKT rephosphorylation upon long term PI3K inhibition that might limit therapy efficacy. This feedback loop can be overcome by parallel inhibition of PI3K/AKT and mTORC1 to efficiently block PI3K signaling, induction of apoptosis and tumor growth in bladder cancer.

Cell lines and cell culture
Cell lines were maintained as early passages of subconfluent cultures in RPMI or DMEM (Biochrom AG, Berlin, Germany) at 5% or 10% CO2, respectively, supplemented with 10% FBS (Biochrom AG) and 1% NEAA (Biochrom AG). RT112 and 647V from the Leibniz institute German collection of microorganisms and cell cultures (Braunschweig, Germany), UMUC3, J82 and T24 were obtained from the American type culture collection (Manassas, VA, USA), 639 V and VmCUB1 were a kind gift from Professor Dr WA Schulz (Heinrich-Heine-University, Düsseldorf, Germany) and 253J were kindly provided by Professor Dr G. Unteregger (University of Saarland, Homburg/Saar, Germany). Cell lines were authenticated by short tandem repeat profiling and tested for mycoplasma using PCR as described previously [24,25].

Cell viability, apoptosis and cell cycle assays
CellTiter-Blue cell viability assay, caspase-Glo 3/7 assay (Promega Corporation) and click-it EdU Alexa Fluor 488 flow cytometry cell cycle analysis (Life Technologies) were performed as described [25].

Clonogenic assay
150 RT112 cells were seeded in a 6-well format and treated with indicated inhibitors every 48 or 72 hours over a period of 12 days. Colonies were then fixed using 6% v/v glutaraldehyde and stained with 0.5% w/v crystal violet as described previously [26].

Chorioallantoic membrane (CAM) assay
The CAM assay was performed as described previously with seeding of 2 million cells on embryonic day (ED) 9, topical treatments on ED 11 and 13 and tumor harvesting on ED 15. Concentrations were calculated for the total embryo blood volume [24].

Statistical analysis
Control or treatment conditions were compared using unpaired Student's T test in Microsoft Excel. Determination of synergy was performed using Compusyn software (ComboSyn, Inc., Parasmus, NJ, USA) [27].

PI3K/mTOR inhibitors control S6K1 and 4E-BP1 dephosphorylation and cell viability
Understanding regulation of 4E-BP1 is essential for targeting the PI3K signaling pathway for tumor therapy [23]. We extended previous data, showing that the mTORC1 inhibitor Everolimus (RAD001) robustly reduced S6K1 phosphorylation level. However, 4E-BP1 phosphorylation at Thr37/46 was not or only partially affected at high concentration and Ser65 phosphorylation level of 4E-BP1 was only downregulated at high concentration in the nonresponding cell lines 253J and 647V S1A Fig [23,28]. Cell viability was affected by 20 to 50% Fig 1E. We next tested the effects of PI3K inhibition on 4E-BP1 phosphorylation using the pan-PI3K inhibitors PIK-90 and NVP-BKM120 [29,30]. These inhibitors dose-dependently diminished AKT phosphorylation at concentrations above 50-100 nM, accompanied by a parallel reduction in S6K1 phosphorylation and 10-50% effects on cell viability Fig   tested Fig 1E. Possible off-target effects of NVP-BEZ235 resulting in 4E-BP1 dephosphorylation were excluded by using specific inhibitors against ATM (KU60019) and DNA-PK (NU7441), as also PDK-1 (GSK2334470) S3C Fig [32]. mTOR silencing using an shRNA revealed that mTOR by itself only induced AKT and S6K1 dephosphorylation Fig 1D. INK128 treatment of cells with silenced mTOR expression continued to result in 4E-BP1 dephosphorylation. This indicates that either INK128 has an additional effect on PI3K activity or that remaining very low expression level of mTOR are sufficient to maintain 4E-BP1 phosphorylation.

Combined inhibition of PI3K and mTORC1 results in 4E-BP1 dephosphorylation
To examine the underlying molecular mechanisms of 4E-BP1 regulation, we combined RAD001 with increasing concentrations of PIK-90. This treatment decreased 4E-BP1 phosphorylation at specific concentrations in all the cell lines examined Fig 2A and S4 Fig. Cell

Long-term PI3K/mTOR inhibition results in AKT rephosphorylation
NVP-BEZ235 treatment caused dephosphorylation of AKT at Thr 308 that recovered at 4-6 hours and was hyperphosphorylated at 24-48 hours after treatment Fig 3A and S5A Fig [23]. AKT at Ser 473, S6K1 and 4E-BP1, remained persistently dephosphorylated. Re-treatment of cells one hour before harvesting had no major effect on the rephosphorylation indicating no loss of inhibitor activity.
We then combined NVP-BEZ235 with the PDK1 inhibitor GSK2334470 to examine whether PDK-1 is responsible for the AKT rephosphorylation Fig 3B and S5B Fig. GSK2334470 monotherapy resulted in AKT dephosphorylation at Thr 308 within one hour of treatment but a hyperphosphorylation at both Thr 308 and Ser 473 was observed as described before [33]. However, combining GSK2334470 with NVP-BEZ235 successfully suppressed this rephosphorylation of AKT.
Recruitment of PDK1 to the cell membrane depends on generation of PIP3. In order to test PIP3 level, we used a previously validated ELISA based quantification of PIP3 [15]. The PIP3/ PIP2 ratio decreased by 95% after 1 hour of and was not restored 24 hours after NVP-BEZ235 treatment Fig 3C indicating that PIP3 is not essential for the PDK-1 dependent rephosphorylation of AKT.
To further examine the role of PI3K or mTOR on inducing rephosphorylation of AKT Thr308, we first treated cells with a combination of PIK-90 and GSK233470 that successfully suppressed AKT rephosphorylation Fig 3D and S5C Fig. Next   473 within 1 hour. However, at 24 hours AKT was rephosphorylated at both Ser and Thr residues. The Thr308 rephosphorylation could only be partially suppressed by a combination of PDK1 and mTOR inhibitors. When using NVP-BEZ235, no dephosphorylation of AKT but only of S6K1 was observed at 1 hour. Interestingly, AKT became hyperphosphorylated after long-term treatment, which could not be prevented by additional PDK1 inhibition.

Combination of dual PI3K/mTOR and AKT inhibitors reduces tumor growth
In order to translate these results to long-term survival we used colony formation assays. Colony size decreased dramatically after NVP-BEZ235 treatment but 65% colonies continued to survive Fig 4A and 4B.
No significant effects were observed using MK-2206 or GSK2334470, whereas combining NVP-BEZ235 with either GSK2334470 or MK-2206 almost completely suppressed colony formation.
The combination of MK-2206 and NVP-BEZ235 suppressed AKT hyperphosphorylation 24 hours after treatment and induced apoptosis measured by dephosphorylated Bad and caspase 3/7 activity Fig 4C and 4D. Compared to NVP-BEZ235 monotherapy, cell cycle progression was not further affected S6A Fig Table. Dose reduction indices (DRI) for NVP-BEZ235 were greater than 1, indicating that a comparable anti-tumor effect requires lower concentrations of NVP-BEZ235 when combined with either PDK1 or AKT inhibitors.

Combination treatment of BEZ-235 and MK2206 strongly reduces tumor growth
For translation the in vitro observations into a three-dimensional tumor xenograft system we applied the chicken chorioallantoic membrane (CAM) mode [34]. Treatment of tumors with either PIK-90 or RAD001 resulted in decreases in tumor weight of 35% or 20% respectively with no effect of MK-2206 Fig 5A. Combining PIK-90 with RAD001 or using NVP-BEZ235 resulted in a 60% decrease. However, combination of NVP-BEZ235 with MK-2206 resulted in an additional 46% reduction in tumor weight as compared to treatment with NVP-BEZ235 alone. The level of Ki-67 expression was not reduced by MK-2206 but decreased by 13% with NVP-BEZ23 and 30% when combining NVP-BEZ235 and MK-2206 Fig 5B and 5C [24].

Discussion
Understanding the complexity of the PI3K signaling pathway is a crucial prerequisite for designing therapies that overcome the present limitations in clinical application. One of the challenges is to understand the cellular response to a specific drug in a tumor entity in order to understand what kind of feedback loops as cellular rescue mechanisms are induced after treatment.
We have previously demonstrated that various classes of inhibitors against molecules in the PI3K pathway differ in their effects on S6K1 and 4E-BP1 phosphorylation level. Understanding the regulation of 4E-BP1 phosphorylation is imperative since its inhibition is required to mediate an effective reduction in cell growth and proliferation [35]. Our previous results and the present study demonstrate that the inhibition of either PI3K, AKT, mTORC1 or mTOR alone is insufficient to reduce 4E-BP1 phosphorylation in most cell lines examined, whereas dual PI3K/mTOR inhibitors successfully suppress 4E-BP1 phosphorylation [23,24]. Regulation of 4E-BP1 phosphorylation by parallel inactivation of both PI3K and mTOR is supported by Comprehensive analysis of PI3K inhibition in bladder cancer the concentration dependency of this effect with PI3K inhibitors, PIK-90 and BKM-120, the dual PI3K/mTOR inhibitor NVP-BEZ235 as well as a combination of PIK90 and the mTORC1 inhibitor RAD001. We extended these data by combining the PI3K inhibitor PIK-90 with either mTOR shRNA, Raptor or Rictor siRNA. These data indicate that concomitant inhibition of PI3K and mTORC1, but not mTORC2, results in 4E-BP1 dephosphorylation. Furthermore, AKT inhibition failed to affect 4E-BP1 phosphorylation, alone or in combination with Raptor silencing.
It has been proposed that the lack of 4E-BP1 dephosphorylation observed with allosteric mTORC1 inhibitors like rapalogs can be overcome by using ATP-competitive mTOR inhibitors [36]. Using INK128 as an example of this class of mTOR inhibitors, we show here that only at non-specific doses that besides mTORC1 additionally inhibit PI3K-AKT induce inhibitory effects on 4E-BP1. Raptor regulates the catalytic activity of mTOR by controlling the recognition and recruitment of substrates via their TOS motifs [8,37,38]. Moreover, Raptor-free Comprehensive analysis of PI3K inhibition in bladder cancer mTOR is capable of phosphorylating S6K1 but not 4E-BP1, providing a basis for the differential regulation of these two substrates by mTORC1 [39]. Our results support this notion and indicate that the regulation of 4E-BP1 by mTOR relies on the gatekeeper function of Raptor. However, silencing either Raptor or mTOR expression alone did not reduce 4E-BP1 phosphorylation and required additional PI3K inhibition. Hence, PI3K may influence Raptor function, directly or indirectly. Of note, the combined inhibition of Raptor and PI3K dephosphorylated 4E-BP1 more completely than inhibition of mTOR and PI3K. Taken together, we postulate that formation of the mTORC1 protein complex rather than regulation of mTOR kinase activity might be controlled by PI3K activity by a mechanism that is independent from the PI3K/AKT axis and which is a crucial step in the control of 4E-BP1 activation. Interestingly, dephosphorylation of S6K1 was observed using inhibitors of PI3K, AKT, PDK1 and also mTOR or mTORC1. This indicates that S6K1 is regulated by mTORC1 and that in the context of bladder cancer, inhibitors directed against upstream elements of the pathway do not activate feedback elements such as GSK1 that can rescue mTORC1 activity as described in other tumor entities [14]. Our results demonstrate a novel mechanism for the regulation of 4E-BP1 phosphorylation in bladder cancer, which is mediated by PI3K and mTORC1 but is not dependent on activity of AKT Fig 6A. From a translational perspective, we have also demonstrated that dual PI3K/mTOR inhibitors result in a greater reduction in cell viability as compared to either PI3K, AKT or mTORC1 inhibitors alone Fig 1E and S4B Fig [24]. Previous reports investigating PI3K pathway inhibition in bladder cancer have demonstrated that cells possessing activating mutations in PIK3CA are selectively sensitive to PI3K or AKT inhibition, while TSC1 or mTOR mutations contribute to sensitivity to rapalogs [40][41][42]. Thus, monotherapy with those inhibitors is therefore only suitable for a highly selected patient population and final parameter for stratification have not been validated to date. We therefore investigated combination therapies to improve the efficacy in patients. The panel of cell lines included in our study possesses diverse molecular alterations in FGFR3, RAS, PIK3CA, PTEN, TSC1 and mTOR that are capable of influencing response to PI3K pathway inhibition and are very frequent in bladder cancer specimen [24,43,44]. Our results indicate that bladder cancer cells are sensitive to dual PI3K/mTOR inhibitors regardless of their genetic background. Furthermore, this response is greater than that achieved with PI3K, AKT, mTOR or PDK1 inhibitors alone.
Despite the promising effects of dual PI3K/mTOR inhibitors in cell lines, a recent clinical trial failed to demonstrate clinical benefit with NVP-BEZ235 as second line therapy in bladder cancer patients with locally advanced or metastatic disease [45]. In this trial, stable disease or partial response were observed in 3 out of 20 patients. NVP-BEZ235 also demonstrated an unfavorable toxicity profile with 10 patients experiencing grade 3-4 adverse effects. A similar pattern of clinical efficacy and toxicity has been demonstrated in clinical trials with other tumor entities [12,46]. This failure may be explained in part by our observation that the efficacy of dual PI3K/ mTOR inhibitors is limited by a feedback activation of AKT over time. It has previously been described that the p110β isoform results in a PIP3 dependent compensatory AKT phosphorylation following selective p110α inhibition [15]. We demonstrate here a novel feedback activation of AKT that occurs despite suppression of all PI3K isoforms, requires additional suppression of PDK1 activity and occurs independently of PIP3 synthesis, revealing an additional level of adaptive signaling after PI3K inhibition. Similar to previously published data, PDK1 inhibition alone by GSK2334470 was insufficient to control AKT dephosphorylation [14]. Our results indicate that this requires additional inhibition of PI3K Fig 3 and S4 and S5 Figs. We could also observe previously described AKT rephosphorylation following treatment with mTOR kinase inhibitors that is proposed to depend on feedback activation of receptor tyrosine kinases and formation of PIP3 [47]. In agreement with published results, this rephosphorylation could be partially suppressed by additional PDK1 inhibition for one hour [33]. However, at 24 hours this combination was insufficient to suppress AKT phosphorylation.
Our data indicate that this is due to persistence of PI3K activity and demonstrate a tight interplay between mTOR, PI3K and PDK1 that can induce AKT rephosphorylation surprisingly also independently of PIP3. The key molecule for maintaining phosphorylated AKT despite concurrent mTOR and PDK1 inhibition seems to be PI3K. This suggests a novel link between PI3K and AKT, independent of PIP3 dependent PDK1 regulation of AKT phosphorylation Fig 6B. Previously described PIP3 independent mechanisms of PDK1 mediated regulation of AKT might play a role in this process and raise the intriguing possibility that PI3K controls these processes by regulating protein complex formation [33,48]. Hence, if AKT should not be targeted directly, for sustained dephosphorylation of AKT both PI3K and PDK1 must be simultaneously inhibited. These compensatory signaling mechanisms might operate to ensure the persistence of activated AKT to enable or maintain cell proliferation in the presence of inhibitory signals.
AKT hyperphosphorylation resulting from dual PI3K/mTOR inhibitors was suppressed by a combination treatment with AKT or PDK1 inhibitors. This combination was synergistic and resulted in apoptosis induction, reduced colony formation ability and decreased tumor growth, as compared to dual PI3K/mTOR inhibition alone Fig 6C. The combination therapy also demonstrated DRI values greater than 1, which might permit a reduction in the dose of NVP-BEZ235 to avoid its dose-dependent toxicities. The combined inhibition of PI3K, AKT and mTOR ensures the dephosphorylation of the essential downstream targets S6K1, 4E-BP1 and AKT and is required for an effective and sustained reduction in tumor growth. This preclinical study provides an explanation for the limitations of the efficacy of targeting the PI3K pathway in the clinic. We also present a novel treatment regimen that might benefit a broad range of patients and could limit dose-dependent toxicities.

Conclusion
The PI3K signaling pathway is frequently altered in cancer and thus provides an attractive target for cancer therapy. However, inhibitors against this pathway showed very disappointing results in clinical trials. We present a comprehensive analysis of functional and biochemical effects using different classes of inhibitors to understand limitations and improve therapy designs used to date. Our results demonstrate that PI3K/mTORC1 regulate 4E-BP1 phosphorylation independent from AKT. Long-term inhibition of PI3K results in PDK1 mediated rephosphorylation of AKT. Simultaneous targeting of PI3K, AKT and mTOR is required for effective tumor suppression by promoting sustained AKT, S6K1 and 4E-BP1 dephosphorylation and induction of apoptosis. Since this therapy acts synergistic it might reduce adverse effects as well as improve patient response. . Cells that were treated for 24 and 48 hours were additionally retreated with the same concentration for 1 hour (indicated by +, no retreatment indicated by -) and immunoblotting was performed with the respective antibodies (B) NVP-BEZ235 or control treatment was combined with 500 nM of GSK2334470 for the indicated duration or with control (ctrl). Immunoblotting was performed with the indicated antibodies. Results are representative of at least three independent experiments. (C to E) Cells were treated with control (ctrl), 500 nM GSK2334470, 500 nM PIK-90, 25 nM INK128 or 10 nM NVP-BEZ235 for the indicated duration and immunoblotting was performed on lysates with the indicated antibodies. Results are representative of at least three independent experiments.