The PI3K/Akt Pathway Regulates Oxygen Metabolism via Pyruvate Dehydrogenase (PDH)-E1α Phosphorylation

The PI3K/Akt/mTOR pathway is commonly activated in human cancers and plays a critical role in the development and maintenance of tumors (1). It has been implicated in multiple cellular processes involved in cell survival and growth, including proliferation, adhesion, migration, invasion, and metabolism (2). For this reason, pharmacologic companies have developed multiple drugs targeting this pathway (3–5). Some of these compounds have shown tolerable toxicity profiles in early-stage trials and are being further investigated as single agents or in combination with other modalities.

The PI3K pathway has been found to have an important role in metabolism by increasing glucose uptake (6, 7). However, its effects on oxygen consumption have been less well studied. In the current study, we investigated the effects of PI3K/mTOR inhibition on oxygen utilization and tumor hypoxia. Hypoxia is present in most solid tumors (8) and has been associated with resistance to therapy, including radiation and chemotherapeutic agents (9–13). A number of strategies have been used to reverse tumor hypoxia, such as increasing oxygen delivery to tumors using hyperbaric oxygen or carbogen (14, 15); however, these have met with limited success, partly due to the abnormal and leaky tumor vasculature and the consumption of oxygen by the tumor cells limiting its diffusion to regions distal to tumor vessels (16). An alternate means of decreasing hypoxia in tumors would be to decrease oxygen consumption by tumor cells. We and others have previously reported that tumor hypoxia can be reversed in vivo by agents that affect the PI3K/mTOR pathway (17–19). In investigating the molecular mechanism underlying this effect, we identified a novel link between PI3K/mTOR activation and phosphorylation (and inactivation) of pyruvate dehydrogenase (PDH), which catalyzes the conversion of pyruvate to acetyl CoA, thereby regulating mitochondrial respiration. Consequently, inhibition of the PI3K pathway would be predicted to lead to decreased oxygen consumption and concomitantly increased tumor pO₂. Our findings shed further light as to how the PI3K/mTOR pathway regulates cellular metabolism. They have important potential clinical implications in terms of using PI3K/mTOR inhibitors in combination with radiation to treat human cancers.

NVP-BEZ235 (referred to as BEZ235), NVP-BGT226 (referred to as BGT226), GDC-0068, and GDC-0980 were obtained from Selleck Pharmaceuticals. These drugs were dissolved in DMSO at a stock concentration of 100 μmol/L.

SQ20B and FaDu cells were obtained from ATCC. SQ20B and FaDu head and neck squamous cell carcinoma cells were cultured in DMEM (4,500 mg/L glucose; Invitrogen) containing 10% FBS (Atlanta Biologicals), penicillin (100 U/mL), and streptomycin (100 mg/mL; Life Technologies, Inc.) at 37°C in humidified 5% CO₂–95% air. U251-PTEN and U251-C124S cells were obtained from Dr. Georgescu at MD Anderson Cancer Center (Houston, TX; ref. 20). All four cells lines were authenticated by IDEXX RADIL.

Cells were transfected with ON-TARGET plus SMART pool siRNA (GE Dharmacon) against Akt-1 or PDH-E1α. Briefly, cells were harvested and plated at a density of 200,000 cells per well in a 6-well plate and allowed to attach overnight. The next day media were removed and cells were washed twice with PBS and re-fed with 1 mL of OPTI-MEM from Gibco. The 6-well plate was returned to the incubator for 1 hour before they were transfected. siRNA was mixed with Oligofectamine reagent (Invitrogen) for 20 minutes before being added to the dishes.

Protein isolation and quantitation and Western blotting were performed as described previously (21). Antibodies directed against the following proteins were obtained from Cell Signaling Technology: phospho-Akt (Ser473), Akt1, phospho-4E-BP1 (Ser 65), phospho-S6, pyruvate dehydrogenase (C54G1), β-actin, and PTEN. The following antibodies were obtained from Abcam: pyruvate dehydrogenase E1α subunit (phospho-S293), pyruvate dehydrogenase E1α subunit (phospho-S232), pyruvate dehydrogenase E1α subunit (phospho-S300), pyruvate dehydrogenase E2 subunit, pyruvate dehydrogenase E1β subunit, pyruvate dehydrogenase E2/E3 subunit. The secondary antibody used for these blots was either a goat anti-mouse and goat anti-rabbit antibody from Thermo Scientific. Antibody binding was detected using an enhanced chemiluminescence kit (GE Healthcare).

Cells were treated with drug for 16 hours before being trypsinized and suspended in media (DMEM with 1% FBS, 1 mmol/L pyruvate, 1 mmol/L glutamate, and 25 mmol/L HEPES) and kept on ice until added to sealed chambers. An aliquot of the cell suspension was added to 3 mL of media in the glass chamber of the YSI magnetic stirring apparatus. Oxygen consumption was measured using the YSI 5300A Biological Oxygen Monitor, which is a polarographic Clark-style oxygen electrode, as previously described (22).

Cells were seeded (60,000 cells/well) in 24-well plates from Seahorse Biosciences. The following day, they were treated with drug for 16 hours before measuring their oxygen consumption rate (OCR). One hour before the assay, culture medium was replaced with modified DMEM supplemented with 1 mmol/L sodium pyruvate, 1 mmol/L glutamate, and 5 mmol/L glucose (pH 7.4). The rate of oxygen consumption (OCR) was measured at 37°C using an XF24 Extracellular Flux Analyzer from Seahorse Bioscience. The baseline (basal) OCR was measured three times before and three times after each sequential injection of oligomycin (1 μmol/L), FCCP (0.8 μmol/L), and rotenone (both 1 μmol/L). At the end of the assay, protein concentration was determined for individual wells as described previously (21). To account for variations in cell number brought about by drug-induced effects on proliferation or cell death, all raw OCR values were normalized to total protein content.

Pathogen-free female Ncr-nu/nu mice were obtained from the National Cancer Institute (stock # 01B74) Taconic Industries and housed in the animal facilities of University Laboratory Animal Resources and the Institute for Human Gene Therapy of the University of Pennsylvania (Philadelphia, PA). All experiments were carried out in accordance with University Institutional Animal Care and Use Committee guidelines.

The OxyLab pO₂ single chamber oxygen monitor (Oxford Optronix Ltd.) was used to monitor tissue oxygen levels in mice bearing subcutaneous flank tumors. This technique has been described previously (23). Before the start of drug treatment, baseline oxygen levels were determined for each mouse in the control as well as the BEZ-treated group. Mice were anesthetized with isoflurane before inserting the probe longitudinally through the tumor. The probe was then retracted through the tumor stopping several times to record the pO₂ along the longitudinal axis. An average was calculated from 4 to 6 readings through one track.

The hypoxia marker EF3 [(2-(2-nitroimidazol-1 [H]-yl)-N-(3, 3, 3-trifluoropropyl) acetamide)] which forms longlasting covalent bonds with hypoxic cells was used to label hypoxic regions in tumors. EF3 dissolved in saline (20 mmol/L) was injected into tumor-bearing mice 3 hours before tumor removal via the tail vein at 0.01 ml/g. Two hours before tumor removal, a second injection of EF3 (0.03 mL/g) was given by intraperitoneal injection. Hoechst (3 mg/mL in saline) was injected (0.01 mL/g) 1 minute before tumors were removed.

Cryosectioning, IHC, and fluorescence microscopy for EF3 were performed as described previously (13). Briefly, sections (20-um thickness) were cut and they were fixed with 4% PF, rinsed in Dulbecco's PBS (Sigma), and blocked in PBS containing 0.3% Tween 20 and 1.5% albumin, plus 20% nonfat milk and 5% normal mouse serum. Antibody staining for EF3 was performed for 4.5 to 5 hours using a monoclonal antibody (ELK5-A8) conjugated to the fluorochrome Cy5 (Amersham Life Sciences). Slides were rinsed in PBS containing 0.3% Tween 20 and PBS without Tween 20, and then stored in 1% PF. Images were taken on a Zeiss Axio Observer Z1 microscope using Zen 2011 software. A total of 82 tiles were taken for each tumor and the intensity was determined using ImageJ software. All images were photographed the same day using the same exposure time.

pCMV6 plasmid containing Flag-tagged PDHA1 cDNA sequence was purchased from Origene. QuickChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies) was used to substitute serine with alanine. Briefly, the entire plasmid was amplified using PCR with primers containing the desired mutation using PfuUltra DNA polymerase. Following amplification, template plasmids were digested with Dpn 1 and mutated plasmids were used to transform competent cells. Mutations were confirmed by sequencing plasmids. The wild-type and mutant PDHA1 cDNAs were excised and placed into the pBABE-Neo vector.

A total of 5 × 10⁵ 293T cells were plated per 100 mm plate. Transfections of pBABE plasmids containing wild-type or mutant PDHA1 were performed with Lipofectamine 2000 according to the manufactures protocol (Invitrogen).

SQ20B cells to be infected were seeded 1 × 10⁶ cells per 100 mm dish 24 hours before infection. On the day of infection, virus-containing media were collected and centrifuged to remove floating cells and debris (1,000 rpm for 5 minutes). Polybrene was added to the virus containing media to a final concentration of 8 μg/mL before placing on the target cells. The 293T cells were fed with fresh media, which was used to infect cells as above a second time. Twenty-four hours following the first day infections, the cells were infected a third time and allowed to grow for 48 hours before placing them under selection with G418.

We previously demonstrated that the anti-HIV agent nelfinavir, which happens to inhibit PI3K signaling, can decrease hypoxia in human tumor xenografts (18). Others have shown that more specific drugs such as NVP-BEZ235 (henceforth referred to as BEZ235), which is a dual inhibitor of the p110 subunit of PI3K and mTOR (24) can also decrease hypoxia (25). We set out to understand how this occurred. One potential explanation for this reduced tumor hypoxia is that the tumor cells decreased their consumption of O₂ in response to drug treatment. To test this, we measured O₂ consumption in vitro using the YSI 5300A Biological Oxygen Monitor. Because the cells are suspended in a sealed chamber, the decrease in O₂ measure as a function of time is a direct measure of cellular OCR. We used BEZ235 at a concentration of 50 nmol/L, which decreases phosphorylation of Akt as well as the mTOR targets S6 and 4E-BP1 (Supplementary Fig. S1A). Treatment of SQ20B cells with BEZ235 resulted in a significant decrease (37%) in OCR compared with vehicle only-treated cells (Fig. 1A). Trypan blue exclusion assay of SQ20B cells incubated with BEZ235 for 16 hours showed no significant difference in cell viability compared with vehicle treatment (data not shown), indicating that the decrease in OCR by BEZ235 was not caused by drug-induced cell death. We found no evidence that treatment of these cells with BEZ235 led to apoptosis, as there was no increased in cleaved caspase-3 (Supplementary Fig. S1B), consistent with our published results showing no effect of the drug on PARP cleavage (26).

To further confirm our observation about OCR, we used the XF Extracellular Flux Analyzer (27), which uses fluorescence quenching to measure dissolved O₂. Figure 1B shows the tracings from an experiment in which we started the measurement 16 hours after BEZ235 treatment (corresponding to T = 0, 8, 16 minutes). BEZ235 treatment decreased OCR by 44% compared with control cells, similar to that seen using the Clark electrode. To determine whether the decrease of OCR in SQ20B cells was specific for BEZ235, we treated cells with an alternate dual PI3K/mTOR inhibitor BGT226 (28) at a dose that inhibits phosphorylation of Akt, S6, and 4E-BP1 (50 nmol/L; Supplementary Fig. S1C). Treatment with BGT226 showed a similar decrease in basal OCR which was comparable with that observed following treatment with BEZ235 (Fig. 1B). Thus taken together, these results indicate that pharmacologic inhibition of PI3K/mTOR pathway decreases basal OCR.

Both electron transport chain (ETC) activity and the rate of ATP production in the mitochondria determine the OCR. These two processes are coupled via the proton gradient in normally functioning cells. To assess the specific contributions of ATP synthesis and ETC activity to altered OCR, we performed measurements in real-time following sequential addition of oligomycin (an ATP synthase inhibitor) and FCCP (an ionophore that dissipates the proton gradient thereby uncoupling proton pumping from ATP synthesis; Fig. 1B). As expected, following the addition of oligomycin (line A), OCR was decreased in all treatment groups, although it did not go to zero (likely due to proton leak). After the addition of FCCP (line B), there was a robust increase in OCR in the control group (OCR at T = 48, 56 64 minutes), which represents the maximal mitochondrial O₂ consumption. However, treatment with BEZ235 or BGT226 reduced uncoupled respiration compared with controls. Thus, both coupled and uncoupled respirations are inhibited by PI3K/mTOR inhibition. As expected, addition of the irreversible complex I inhibitor rotenone (line C) led to inhibition of O₂ consumption in control and drug-treated groups.

To determine the kinetics at which BEZ235 or BGT226 exerts their inhibitory effect on O₂ consumption, cells were incubated in either drug for varying lengths of time (1–16 hours) before measuring OCR in the flux analyzer. Treatment of either drug for 1 or 2 hours before measurement had little effect on OCR. However, a statistically significant reduction in OCR was observed after 4, 8, and 16 hours of BEZ235 treatment (Supplementary Fig. S2).

Both BEZ23 and BGT226 inhibit PI3K and mTOR; however, we have not ruled out the possibility that their effect on OCR may be the result of some off-target effect. To test this, we used siRNA to knock down Akt1, which is directly downstream of PI3K. As shown in Fig. 1C, inhibition of Akt1 expression led to a corresponding decrease in P-Akt and a transient decrease in P-S6 (at 48 hours). Using the flux analyzer, we measured OCR in cells transfected with Akt1 siRNA or control siRNA. The baseline OCR readings (at T = 0 hours) as shown in Fig. 1D indicate that knockdown of Akt1 resulted in a 37% decrease in OCR. We also used KU-0063794, which is a specific inhibitor of mTOR (29), leading to decreased phosphorylation of both P-S6 and P-4EBP1 (Fig. 1E). Treatment of cells for 16 hours with this drug did not reduce OCR (Fig. 1F).

To further determine the mechanism by which these drugs decreased O₂ consumption, we investigated the possibility of some structural or functional change induced in the mitochondria. Using Mito-Tracker Green (Invitrogen M7514), which localizes to mitochondria regardless of membrane potential, we found no difference in signal intensity between the BEZ235-treated and vehicle only-treated groups (Supplementary Fig. S3A); hence, drug treatment did not substantially alter mitochondrial mass. We also examined mitochondrial membrane potential (MMP) using Mito-Tracker Red (Invitrogen M22425), a stain that accumulates in live cells in a manner dependent upon MMP, and found no difference in signal intensity between the BEZ235-treated and vehicle only-treated groups (Supplementary Fig. S3B). Finally, analysis of the DNA levels by primers specific to the mitochondrial gene COX1 by RT-PCR (Supplementary Fig. S4 and Supplementary Table S1) did not change significantly in response to BEZ235 treatment.

We subsequently examined whether drug treatment might alter mitochondrial respiratory chain activity. A major regulator of mitochondrial oxidative function is the pyruvate dehydrogenase complex (PDC) which catalyzes the irreversible decarboxylation of pyruvate to form acetyl-coA, which can then enter the citric acid cycle and be used as a substrate for oxidative phosphorylation (30). PDC activity is inhibited by phosphorylation of its pyruvate dehydrogenase (PDH) E1α subunit (30). Hence, phosphorylation of PDH-E1α reduces entry of pyruvate into the citric acid cycle and consequently decreases O₂ consumption. We found that incubation of SQ20B cells with either BEZ235 or BGT226 led to a time-dependent increase in of PDH-E1α phosphorylation, which was detectable within 2 hours and continued to increase for up to 16 hours (Fig. 2A). No change in the level of PDH-E1α, E1β, E2, or E2/E3bp was seen (Fig. 2B and Supplementary Fig. S5). For the remainder of this manuscript for simplicity's sake, we will refer to PDH-E1α, as simply PDH.

To test whether inhibition of the PI3K/Akt pathway was necessary for the drug-induced effect on PDH phosphorylation, we used siRNA to knock down Akt1. Inhibition of Akt1 resulted in a 4.5-fold increase in Ser293 phosphorylation of PDH (Fig. 2B), similar to that seen with either BEZ235 or BGT226. In this particular experiment, in which samples were harvested at 72 hours following transfection, Akt1 siRNA had no effect on decreasing S6 phosphorylation (which is directly downstream of mTOR). However, this still resulted in increased PDH-E1α phosphorylation, suggesting that Akt itself rather than mTOR might be responsible for this effect. As a complementary approach, we used the drug GDC-0068, an Akt inhibitor currently being tested in clinical trials (31), and GDC-0980, another dual PI3K/mTOR inhibitor (32). Incubation with GDC-0068 actually increased Akt phosphorylation as has been reported previously for this drug and other ATP-competitive Akt inhibitors (31). However, as expected, there was decreased phosphorylation of the downstream targets S6 and 4E-BP1 (Fig. 2C). There was a concomitant increase in PDH-E1α phosphorylation with GDC-0068 treatment, similar to what we observed with BEZ235, BGT226, and Akt1 siRNA. Flux analyzer measurements showed that the OCR was also reduced with GDC-0068 treatment by 64% (Fig. 2D and Supplementary Fig. S6A). Treatment with GDC-0980 showed similar effects on PDH-E1α phosphorylation and OCR (Fig. 2C and D and Supplementary Fig. S6B). Finally, we used MK-2206, which, unlike GDC-0068, is a non-ATP competitive allosteric Akt inhibitor (33). In contrast with GDC-0068, MK-2206 does decrease Akt phosphorylation (Fig. 2E). It also led to decreased OCR (Fig. 2F)

To generalize these results, we used another head and neck cancer cell line, FaDu. There was a time-dependent increase in PDH phosphorylation between 2 and 16 hours of exposure to either BEZ235 or BGT226 (Fig. 3A). Using the flux analyzer, we found that 8 hours of incubation with BEZ235 or BGT226 resulted in significant decreases in OCR (Fig. 3B and Supplementary Fig. S 6C).

To further test the relationship between Akt and PDH phosphorylation using a genetic approach, we used U251 glioblastoma cells engineered to express wild-type PTEN under the control of a tetracycline-inducible promoter (34). These cells express constitutively high levels of P-Akt due to their mutant PTEN status. Addition of doxycycline caused a substantial decrease in P-Akt, and an increase in PDH phosphorylation (Fig. 3C). Importantly, this did not occur in a control cell line that expresses a mutant form of PTEN that cannot decrease Akt phosphorylation, indicating that this was not simply a response to doxycycline (Fig. 3C). We also found that addition of doxycycline led to a 46% decrease in the basal OCR in the cells induced to express wild-type PTEN (Fig. 3D and Supplementary Fig. S7A), further supporting the link between Akt and O₂ consumption. Notably, this decrease in the basal OCR did not occur when mutant PTEN was induced in U251-C124S cells (Supplementary Fig. S7B). We determined whether PDH was phosphorylated in this cell line and found Ser293 phosphorylation occurred with similar kinetics as noted in SQ20B and FaDu cells (Fig. 3E). We also found that PDH was phosphorylated at both Ser232 and Ser300 in response to treatment with either drug.

To test the importance of PDH on O₂ consumption in SQ20B cells, we knocked down the PDH-E1α subunit using siRNA (Fig. 4A; compare lanes 4 and 2). Reduced PDH resulted in a 32% decrease in basal O₂ consumption (Fig. 4B and Supplementary Fig. S8A). Treatment of cells with BEZ235 or BGT226 led to a similar decrease in OCR.

As a means of examining the importance of PDH-E1α phosphorylation in the ability of BEZ235 to regulate O₂ consumption, we treated cells with DCA (dicholoroacetate). DCA treatment decreased Ser293 phosphorylation (Fig. 4C; compare lanes 1 and 2), through inhibition of pyruvate dehydrogenase kinases (PDK) as previously reported (35). Treatment of cells with DCA blunted the increase in PDH phosphorylation seen in response to BEZ235 (Fig. 4C: lane 3 vs. 4) or to BGT226 (lane 5 versus 6). Treatment of cells with DCA along with BEZ235 reversed the effect of BEZ235 on decreasing OCR (Fig. 4D and Supplementary Fig. S 8B). These figures also show that DCA had a similar abrogating effect on the decrease in OCR seen with BGT226 treatment.

We also used a genetic means to establish that the phosphorylation of PDH E1α was essential for the ability of BEZ235 to decrease O₂ consumption. We infected SQ20B cells with retrovirus expressing FLAG-tagged wild-type PDH-E1α or PDH-E1α containing a single S1 (S232A) or triple S3 (S232A, S292A, S300A) serine to alanine mutation(s) (Fig. 5A). Figure 5B shows that an anti-FLAG antibody recognized a FLAG-tagged protein in the cells infected with a PDH-E1α (wild-type or mutant) virus (lanes 2–4). Using an antibody recognizing P-PDH (S232), we found that in wild-type PDH-E1α–infected cells, there were two separate proteins (marked by the two arrows, second row). The bottom band corresponds to the endogenous protein and upper band FLAG-tagged exogenous protein. This upper band is not visible in lanes 3 and 4 because these cells express mutant PDH-E1α in which the S2332 has been altered to alanine, hence the exogenous protein is not recognized by this antibody. Supplementary Figure S9 shows another immunoblot analysis of lysates from cells infected with the 1S or 3S PDH-mutant probed using three different antibodies. The P-PDH (S232) antibody recognizes both the endogenous and the exogenous proteins. However, the P-PDH (S293) and P-PDH(S300) antibodies appear to only express the exogenous protein. Of note, the 3S mutant does suppress phosphorylation of the endogenous PDH at S293 as well as S300 (compare lane 10 with 8 or 6).

We treated these cells with BEZ235, and then performed OCR measurements. Figure 5C shows that the cells expressing FLAG-tagged wild-type PDH-E1α exhibited a similar decrease in OCR in response to BEZ235 treatment as did control cells. This is more readily appreciated in Fig. 5D in which the same data are plotted to show the % decrease in OCR in response to BEZ235. The % decrease in OCR in cells expressing wild-type PDH-E1α was not different than in control cells (P = ns). However, the% decrease in OCR in cells expressing either the single (or triple) mutant PDH was significantly different than in control cells (P ≤ 0.004). Hence, expression of mutant PDH-E1α that cannot be phosphorylated on Ser232 blunts the effect of BEZ235 on decreasing OCR.

SQ20B human squamous head and neck cells were grown as tumor xenografts in nude mice and were then injected with EF3, a nitroimidazole that forms adducts with proteins in hypoxic regions (36). Treatment of mice with 50 nmol/L BEZ235 led to a significant decrease in EF3 binding (Fig. 6A and B) when compared with vector only treated tumors (P = 0.05), indicating a decrease in overall tumor hypoxic fraction. As an alternate method of assessing the effect of BEZ235 within tumor xenografts, we used the OxyLab pO₂ probe. SQ20B xenografts were grown in nude mice. Drug treatment was not started until tumors were at least 400 mm³ in size (400–1,800 mm³). pO₂ measurements were made before start of drug (day 0), at day 1 and at day 3. Each tumor served as its internal control, so the pO₂ was compared with the day 0 readings for that particular tumor. We calculated the fold change in pO₂ at day 1 or day 3 relative to the day 0 reading (Fig. 6D). In BEZ235-treated mice, there was nearly a 5-fold increase in pO₂ both at day 1 and at day 3 relative to day 0. We also had a set of control tumors in mice not treated drug. In these tumors, there was no appreciable change in pO₂ at day 1 or day 3 relative to day 0 (Fig. 6D). The fold difference in pO₂ in BEZ-235–treated tumors was statistically significant compared with the fold difference seen in control (non drug-treated) mice at both day 1 (P = 0.026) and day 3 (P = 0.007). The mean pO₂ for control (non-drug treated) tumors was 0.95 mm Hg at day 1 and 0.65 mm Hg at day 3. For BEZ235-treated tumors, it was 2.1 and 3.0 mm Hg, respectively.

Following pO₂ measurement on day 3, mice were sacrificed and tumors were removed to measure the in vivo level of PDH 293 phosphorylation by immunoblot analysis (Fig. 6E). Most BEZ-235–treated tumors showed a decrease in P-Akt and an increase in P-PDH (S293). The mean level of phosphorylation BEZ-treated mice was 2.08 compared with 1.04 for control, nondrug-treated mice. This difference was found to be significantly different (P = 0.009).

The presence of hypoxia within human tumors has been associated with resistance to therapy. For decades, this has been appreciated in the case of radiotherapy due to the fact that O₂ must be present for optimal fixation of DNA damage-induced by ionizing radiation (37). There is particularly strong evidence in head and neck cancers that hypoxia plays an important role in resistance to radiotherapy (38–40). There are also reports indicating that hypoxic cells are more resistant to killing following exposure to commonly used cytotoxic agents such as cisplatin, etoposide, and doxorubicin (9–11, 13). Tumor hypoxia has typically been attributed to impaired blood flow due to the disorganized vasculature often found in tumors (16). However, our results suggest that genetic mutations that activate the PI3K/mTOR pathway may lead to increased hypoxia by increasing tumor oxygen consumption.

There are currently numerous ongoing clinical trials PI3K inhibitors in patients with cancer. In the current study, we show that treatment of mice bearing tumor xenografts with the dual PI3K/mTOR inhibitor BEZ235 decreased in vivo tumor hypoxia. To demonstrate this, we used both the nitroimidazole EF3 and OxyLab pO2 probe measurements. We measured the effects of two different PI3K/mTOR inhibitors and two different Akt inhibitors on in vitro oxygen consumption using the Clark electrode and/or the flux analyzer. As these drugs could have off-target effects, we used genetic approaches, including RNAi and cells with inducible PTEN to confirm that it was indeed inhibition of this pathway that was specifically responsible for the decrease in oxygen consumption. Similar findings on tumor O₂ consumption have been made by others using PI3K inhibitors, although these studies did not report any potential mechanism and were focused on pharmacologic inhibition of the pathway (41). However, in further investigating potential molecular mechanisms underlying this effect, we found that PI3K/Akt/mTOR inhibition led to increased PDH-E1α phosphorylation (i.e., decreased activity of the enzyme). As PDH is the critical determinant as to whether pyruvate is converted to acetylCoA, which can then participate in the tricarboxylic acid cycle, inhibiting activity of this enzyme should reduce O₂ consumption. Hence, PDH phosphorylation offers an explanation as to how PI3K/mTOR inhibition can decrease O₂ consumption and reduce tumor hypoxia. siRNA directed against PDH-E1α decreased OCR, similar in extent to that seen with either BEZ235 or BGT226 (Fig. 4B). Dichloroactetate treatment, which prevents the PDH phosphorylation in response to PI3K/mTOR inhibition, also reversed the effect on OCR (Fig. 4D). Ectopic expression of PDH-E1α that was mutated so that the serine sites could not be phosphorylated resulted in a blunting of the ability of BEZ235 to decrease OCR. Hence, our evidence supports a causal relationship between PDH-E1α phosphorylation in response to PI3K/mTOR inhibition and reduced O₂ consumption.

We started our studies using BEZ235, which inhibits both PI3K and mTOR; however, we also we found that Akt1 siRNA increased PDH-E1α phosphorylation. This occurred despite the fact that S6 phosphorylation was not affected (Fig. 2B), which lead us to believe that it is Akt rather than mTOR that regulates this phosphorylation. Our results do not preclude the possibility that the PI3KmTOR pathway may regulate other molecular changes that could contribute to mitochondrial respiration. mTOR itself has been implicated in the regulation of oxygen consumption (42–44). Interestingly, mTOR has been shown to regulate the translation of certain mitochondria-related mRNAs (45). However, our results uncover a novel link between the PI3K/Akt pathway and PDH phosphorylation, which play an essential role in the regulation of O₂ consumption. PDH is one of the key players in the regulation of oxidative metabolism. Papandreou and colleagues showed that PDH phosphorylation is increased under hypoxia (46), and PDH has recently been implicated in cellular senescence (47).

An unanswered question is why the PI3K/Akt pathway should result in greater oxidative metabolism. During oncogenic transformation, this pathway is coopted; however, its original function is in normal growth and development. This pathway is activated by growth factor receptor signaling and appears to be important during proliferation, which is an energy requiring process. Hence, increased oxidative phosphorylation could play a teleologic role as a means of generating more ATP from each molecule of glucose.

In addition to identifying a new mechanism by which the PI3K/Akt pathway regulates metabolism, our results also have potential clinical implications. As mentioned previously, the presence of hypoxia is a negative prognostic factor in cancers treated definitively with radiation, including head and neck cancer (38, 40) and prostate cancer (48). A recent review and meta-analysis of clinical trials using manipulations that target the hypoxic fraction in head and neck SCC concluded that hypoxic modification of radiotherapy led to increased locoregional control, disease-specific survival, and overall survival (14, 15). Another randomized trial showed that the addition of carbogen breathing and nicotinamide to decrease tumor hypoxia improved outcome in patients with laryngeal cancer treated with radiation (49). However, despite these suggestive results, these manipulations that aim to address the problem by increasing the supply of oxygen have shown marginal benefit and have not gained widespread adoption.

An alternative strategy to reduce tumor hypoxia is to attack the problem on the demand side, that is, reducing cancer cell O₂ consumption. By decreasing the O₂ consumption of tumors, there should be more O₂ available to better oxygenate the tumor both intracellularly and extracellularly, which is what we have found. In fact, mathematical modeling from Secomb and colleagues showed that reducing O₂ consumption rate may be more effective than elevating blood flow or oxygen content as a method to reduce tumor hypoxia. (50). These authors found that hypoxia (<3 mm Hg) was abolished by a reduction in consumption rate of at least 30%. Recently, the diabetes medication metformin was reported to inhibit O₂ consumption, and the authors proposed that it might be effective in combination with radiotherapy by reducing tumor hypoxia (51). Our findings indicate that PI3K inhibitors currently in clinical trials may also be useful in this regard. In future trials, this hypothesis could be tested by using noninvasive hypoxia imaging. Such imaging is available with a number of agents, including ¹⁸F-misonidazole, ¹⁸F -EF5, and ¹⁸F-IAZA (reviewed in ref. 52).

No potential conflicts of interest were disclosed.

Conception and design: S. Tuttle, N. Denko, A. Maity

Development of methodology: G.J. Cerniglia, S. Dey, N.A. Daurio, S. Tuttle, S.A. Vinogradov, N. Denko, C. Koumenis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.J. Cerniglia, S.M. Gallagher-Colombo, N.A. Daurio, T.M. Busch, R. Sun, C. Koumenis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Dey, N.A. Daurio, S. Tuttle, T.M. Busch, A. Lin, S.A. Vinogradov, N. Denko

Writing, review, and/or revision of the manuscript: S. Dey, S.M. Gallagher-Colombo, S. Tuttle, T.M. Busch, A. Lin, R. Sun, C. Koumenis, A. Maity

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G.J. Cerniglia, C. Koumenis

Study supervision: C. Koumenis, A. Maity

Other (synthesis of oxygen sensor): T.V. Esipova

The authors thank Dr. Cameron Koch (Department of Radiation Oncology) for developing and providing EF3.

This work was supported in part by NIH RO1 grant CA174976 (to A. Maity and A. Lin) and grants CA163581 and CA67166 (to N. Denko) and CA094214 (to C. Koumenis).

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.